Fix wording

Each inode has: image name, parent inode number & pool, size and readonly flag

Snapshots are created by switching image name to a different inode number
while using the older inode as parent.

.dockerignore

@@ -0,0 +1,19 @@
+.git
+build
+packages
+mon/node_modules
+*.o
+*.so
+osd
+stub_osd
+stub_uring_osd
+stub_bench
+osd_test
+dump_journal
+nbd_proxy
+rm_inode
+fio
+qemu
+rpm/*.Dockerfile
+debian/*.Dockerfile
+Dockerfile

.gitignore

@@ -0,0 +1,18 @@
+*.o
+*.so
+package-lock.json
+fio
+qemu
+osd
+stub_osd
+stub_uring_osd
+stub_bench
+osd_test
+osd_peering_pg_test
+dump_journal
+nbd_proxy
+rm_inode
+test_allocator
+test_blockstore
+test_shit
+osd_rmw_test

.gitmodules

@@ -0,0 +1,6 @@
+[submodule "cpp-btree"]
+	path = cpp-btree
+	url = ../cpp-btree.git
+[submodule "json11"]
+	path = json11
+	url = ../json11.git

CMakeLists.txt

@@ -0,0 +1,5 @@
+cmake_minimum_required(VERSION 2.8)
+
+project(vitastor)
+
+add_subdirectory(src)

GPL-2.0.txt

@@ -0,0 +1,339 @@
+                    GNU GENERAL PUBLIC LICENSE
+                       Version 2, June 1991
+
+ Copyright (C) 1989, 1991 Free Software Foundation, Inc.,
+ 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+                            Preamble
+
+  The licenses for most software are designed to take away your
+freedom to share and change it.  By contrast, the GNU General Public
+License is intended to guarantee your freedom to share and change free
+software--to make sure the software is free for all its users.  This
+General Public License applies to most of the Free Software
+Foundation's software and to any other program whose authors commit to
+using it.  (Some other Free Software Foundation software is covered by
+the GNU Lesser General Public License instead.)  You can apply it to
+your programs, too.
+
+  When we speak of free software, we are referring to freedom, not
+price.  Our General Public Licenses are designed to make sure that you
+have the freedom to distribute copies of free software (and charge for
+this service if you wish), that you receive source code or can get it
+if you want it, that you can change the software or use pieces of it
+in new free programs; and that you know you can do these things.
+
+  To protect your rights, we need to make restrictions that forbid
+anyone to deny you these rights or to ask you to surrender the rights.
+These restrictions translate to certain responsibilities for you if you
+distribute copies of the software, or if you modify it.
+
+  For example, if you distribute copies of such a program, whether
+gratis or for a fee, you must give the recipients all the rights that
+you have.  You must make sure that they, too, receive or can get the
+source code.  And you must show them these terms so they know their
+rights.
+
+  We protect your rights with two steps: (1) copyright the software, and
+(2) offer you this license which gives you legal permission to copy,
+distribute and/or modify the software.
+
+  Also, for each author's protection and ours, we want to make certain
+that everyone understands that there is no warranty for this free
+software.  If the software is modified by someone else and passed on, we
+want its recipients to know that what they have is not the original, so
+that any problems introduced by others will not reflect on the original
+authors' reputations.
+
+  Finally, any free program is threatened constantly by software
+patents.  We wish to avoid the danger that redistributors of a free
+program will individually obtain patent licenses, in effect making the
+program proprietary.  To prevent this, we have made it clear that any
+patent must be licensed for everyone's free use or not licensed at all.
+
+  The precise terms and conditions for copying, distribution and
+modification follow.
+
+                    GNU GENERAL PUBLIC LICENSE
+   TERMS AND CONDITIONS FOR COPYING, DISTRIBUTION AND MODIFICATION
+
+  0. This License applies to any program or other work which contains
+a notice placed by the copyright holder saying it may be distributed
+under the terms of this General Public License.  The "Program", below,
+refers to any such program or work, and a "work based on the Program"
+means either the Program or any derivative work under copyright law:
+that is to say, a work containing the Program or a portion of it,
+either verbatim or with modifications and/or translated into another
+language.  (Hereinafter, translation is included without limitation in
+the term "modification".)  Each licensee is addressed as "you".
+
+Activities other than copying, distribution and modification are not
+covered by this License; they are outside its scope.  The act of
+running the Program is not restricted, and the output from the Program
+is covered only if its contents constitute a work based on the
+Program (independent of having been made by running the Program).
+Whether that is true depends on what the Program does.
+
+  1. You may copy and distribute verbatim copies of the Program's
+source code as you receive it, in any medium, provided that you
+conspicuously and appropriately publish on each copy an appropriate
+copyright notice and disclaimer of warranty; keep intact all the
+notices that refer to this License and to the absence of any warranty;
+and give any other recipients of the Program a copy of this License
+along with the Program.
+
+You may charge a fee for the physical act of transferring a copy, and
+you may at your option offer warranty protection in exchange for a fee.
+
+  2. You may modify your copy or copies of the Program or any portion
+of it, thus forming a work based on the Program, and copy and
+distribute such modifications or work under the terms of Section 1
+above, provided that you also meet all of these conditions:
+
+    a) You must cause the modified files to carry prominent notices
+    stating that you changed the files and the date of any change.
+
+    b) You must cause any work that you distribute or publish, that in
+    whole or in part contains or is derived from the Program or any
+    part thereof, to be licensed as a whole at no charge to all third
+    parties under the terms of this License.
+
+    c) If the modified program normally reads commands interactively
+    when run, you must cause it, when started running for such
+    interactive use in the most ordinary way, to print or display an
+    announcement including an appropriate copyright notice and a
+    notice that there is no warranty (or else, saying that you provide
+    a warranty) and that users may redistribute the program under
+    these conditions, and telling the user how to view a copy of this
+    License.  (Exception: if the Program itself is interactive but
+    does not normally print such an announcement, your work based on
+    the Program is not required to print an announcement.)
+
+These requirements apply to the modified work as a whole.  If
+identifiable sections of that work are not derived from the Program,
+and can be reasonably considered independent and separate works in
+themselves, then this License, and its terms, do not apply to those
+sections when you distribute them as separate works.  But when you
+distribute the same sections as part of a whole which is a work based
+on the Program, the distribution of the whole must be on the terms of
+this License, whose permissions for other licensees extend to the
+entire whole, and thus to each and every part regardless of who wrote it.
+
+Thus, it is not the intent of this section to claim rights or contest
+your rights to work written entirely by you; rather, the intent is to
+exercise the right to control the distribution of derivative or
+collective works based on the Program.
+
+In addition, mere aggregation of another work not based on the Program
+with the Program (or with a work based on the Program) on a volume of
+a storage or distribution medium does not bring the other work under
+the scope of this License.
+
+  3. You may copy and distribute the Program (or a work based on it,
+under Section 2) in object code or executable form under the terms of
+Sections 1 and 2 above provided that you also do one of the following:
+
+    a) Accompany it with the complete corresponding machine-readable
+    source code, which must be distributed under the terms of Sections
+    1 and 2 above on a medium customarily used for software interchange; or,
+
+    b) Accompany it with a written offer, valid for at least three
+    years, to give any third party, for a charge no more than your
+    cost of physically performing source distribution, a complete
+    machine-readable copy of the corresponding source code, to be
+    distributed under the terms of Sections 1 and 2 above on a medium
+    customarily used for software interchange; or,
+
+    c) Accompany it with the information you received as to the offer
+    to distribute corresponding source code.  (This alternative is
+    allowed only for noncommercial distribution and only if you
+    received the program in object code or executable form with such
+    an offer, in accord with Subsection b above.)
+
+The source code for a work means the preferred form of the work for
+making modifications to it.  For an executable work, complete source
+code means all the source code for all modules it contains, plus any
+associated interface definition files, plus the scripts used to
+control compilation and installation of the executable.  However, as a
+special exception, the source code distributed need not include
+anything that is normally distributed (in either source or binary
+form) with the major components (compiler, kernel, and so on) of the
+operating system on which the executable runs, unless that component
+itself accompanies the executable.
+
+If distribution of executable or object code is made by offering
+access to copy from a designated place, then offering equivalent
+access to copy the source code from the same place counts as
+distribution of the source code, even though third parties are not
+compelled to copy the source along with the object code.
+
+  4. You may not copy, modify, sublicense, or distribute the Program
+except as expressly provided under this License.  Any attempt
+otherwise to copy, modify, sublicense or distribute the Program is
+void, and will automatically terminate your rights under this License.
+However, parties who have received copies, or rights, from you under
+this License will not have their licenses terminated so long as such
+parties remain in full compliance.
+
+  5. You are not required to accept this License, since you have not
+signed it.  However, nothing else grants you permission to modify or
+distribute the Program or its derivative works.  These actions are
+prohibited by law if you do not accept this License.  Therefore, by
+modifying or distributing the Program (or any work based on the
+Program), you indicate your acceptance of this License to do so, and
+all its terms and conditions for copying, distributing or modifying
+the Program or works based on it.
+
+  6. Each time you redistribute the Program (or any work based on the
+Program), the recipient automatically receives a license from the
+original licensor to copy, distribute or modify the Program subject to
+these terms and conditions.  You may not impose any further
+restrictions on the recipients' exercise of the rights granted herein.
+You are not responsible for enforcing compliance by third parties to
+this License.
+
+  7. If, as a consequence of a court judgment or allegation of patent
+infringement or for any other reason (not limited to patent issues),
+conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License.  If you cannot
+distribute so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you
+may not distribute the Program at all.  For example, if a patent
+license would not permit royalty-free redistribution of the Program by
+all those who receive copies directly or indirectly through you, then
+the only way you could satisfy both it and this License would be to
+refrain entirely from distribution of the Program.
+
+If any portion of this section is held invalid or unenforceable under
+any particular circumstance, the balance of the section is intended to
+apply and the section as a whole is intended to apply in other
+circumstances.
+
+It is not the purpose of this section to induce you to infringe any
+patents or other property right claims or to contest validity of any
+such claims; this section has the sole purpose of protecting the
+integrity of the free software distribution system, which is
+implemented by public license practices.  Many people have made
+generous contributions to the wide range of software distributed
+through that system in reliance on consistent application of that
+system; it is up to the author/donor to decide if he or she is willing
+to distribute software through any other system and a licensee cannot
+impose that choice.
+
+This section is intended to make thoroughly clear what is believed to
+be a consequence of the rest of this License.
+
+  8. If the distribution and/or use of the Program is restricted in
+certain countries either by patents or by copyrighted interfaces, the
+original copyright holder who places the Program under this License
+may add an explicit geographical distribution limitation excluding
+those countries, so that distribution is permitted only in or among
+countries not thus excluded.  In such case, this License incorporates
+the limitation as if written in the body of this License.
+
+  9. The Free Software Foundation may publish revised and/or new versions
+of the General Public License from time to time.  Such new versions will
+be similar in spirit to the present version, but may differ in detail to
+address new problems or concerns.
+
+Each version is given a distinguishing version number.  If the Program
+specifies a version number of this License which applies to it and "any
+later version", you have the option of following the terms and conditions
+either of that version or of any later version published by the Free
+Software Foundation.  If the Program does not specify a version number of
+this License, you may choose any version ever published by the Free Software
+Foundation.
+
+  10. If you wish to incorporate parts of the Program into other free
+programs whose distribution conditions are different, write to the author
+to ask for permission.  For software which is copyrighted by the Free
+Software Foundation, write to the Free Software Foundation; we sometimes
+make exceptions for this.  Our decision will be guided by the two goals
+of preserving the free status of all derivatives of our free software and
+of promoting the sharing and reuse of software generally.
+
+                            NO WARRANTY
+
+  11. BECAUSE THE PROGRAM IS LICENSED FREE OF CHARGE, THERE IS NO WARRANTY
+FOR THE PROGRAM, TO THE EXTENT PERMITTED BY APPLICABLE LAW.  EXCEPT WHEN
+OTHERWISE STATED IN WRITING THE COPYRIGHT HOLDERS AND/OR OTHER PARTIES
+PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED
+OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.  THE ENTIRE RISK AS
+TO THE QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU.  SHOULD THE
+PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY SERVICING,
+REPAIR OR CORRECTION.
+
+  12. IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
+WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MAY MODIFY AND/OR
+REDISTRIBUTE THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES,
+INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING
+OUT OF THE USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED
+TO LOSS OF DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY
+YOU OR THIRD PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER
+PROGRAMS), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE
+POSSIBILITY OF SUCH DAMAGES.
+
+                     END OF TERMS AND CONDITIONS
+
+            How to Apply These Terms to Your New Programs
+
+  If you develop a new program, and you want it to be of the greatest
+possible use to the public, the best way to achieve this is to make it
+free software which everyone can redistribute and change under these terms.
+
+  To do so, attach the following notices to the program.  It is safest
+to attach them to the start of each source file to most effectively
+convey the exclusion of warranty; and each file should have at least
+the "copyright" line and a pointer to where the full notice is found.
+
+    <one line to give the program's name and a brief idea of what it does.>
+    Copyright (C) <year>  <name of author>
+
+    This program is free software; you can redistribute it and/or modify
+    it under the terms of the GNU General Public License as published by
+    the Free Software Foundation; either version 2 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU General Public License for more details.
+
+    You should have received a copy of the GNU General Public License along
+    with this program; if not, write to the Free Software Foundation, Inc.,
+    51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+
+Also add information on how to contact you by electronic and paper mail.
+
+If the program is interactive, make it output a short notice like this
+when it starts in an interactive mode:
+
+    Gnomovision version 69, Copyright (C) year name of author
+    Gnomovision comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
+    This is free software, and you are welcome to redistribute it
+    under certain conditions; type `show c' for details.
+
+The hypothetical commands `show w' and `show c' should show the appropriate
+parts of the General Public License.  Of course, the commands you use may
+be called something other than `show w' and `show c'; they could even be
+mouse-clicks or menu items--whatever suits your program.
+
+You should also get your employer (if you work as a programmer) or your
+school, if any, to sign a "copyright disclaimer" for the program, if
+necessary.  Here is a sample; alter the names:
+
+  Yoyodyne, Inc., hereby disclaims all copyright interest in the program
+  `Gnomovision' (which makes passes at compilers) written by James Hacker.
+
+  <signature of Ty Coon>, 1 April 1989
+  Ty Coon, President of Vice
+
+This General Public License does not permit incorporating your program into
+proprietary programs.  If your program is a subroutine library, you may
+consider it more useful to permit linking proprietary applications with the
+library.  If this is what you want to do, use the GNU Lesser General
+Public License instead of this License.

LICENSE

@@ -0,0 +1,27 @@
+Copyright (c) Vitaliy Filippov (vitalif [at] yourcmc.ru), 2019+
+
+All server-side code (OSD, Monitor and so on) is licensed under the terms of
+Vitastor Network Public License 1.1 (VNPL 1.1), a copyleft license based on
+GNU GPLv3.0 with the additional "Network Interaction" clause which requires
+opensourcing all programs directly or indirectly interacting with Vitastor
+through a computer network and expressly designed to be used in conjunction
+with it ("Proxy Programs"). Proxy Programs may be made public not only under
+the terms of the same license, but also under the terms of any GPL-Compatible
+Free Software License, as listed by the Free Software Foundation.
+This is a stricter copyleft license than the Affero GPL.
+
+Please note that VNPL doesn't require you to open the code of proprietary
+software running inside a VM if it's not specially designed to be used with
+Vitastor.
+
+Basically, you can't use the software in a proprietary environment to provide
+its functionality to users without opensourcing all intermediary components
+standing between the user and Vitastor or purchasing a commercial license
+from the author 😀.
+
+Client libraries (cluster_client and so on) are dual-licensed under the same
+VNPL 1.1 and also GNU GPL 2.0 or later to allow for compatibility with GPLed
+software like QEMU and fio.
+
+You can find the full text of VNPL-1.1 in the file [VNPL-1.1.txt](VNPL-1.1.txt).
+GPL 2.0 is also included in this repository as [GPL-2.0.txt](GPL-2.0.txt).

Makefile

@@ -1,66 +0,0 @@
-BLOCKSTORE_OBJS := allocator.o blockstore.o blockstore_impl.o blockstore_init.o blockstore_open.o blockstore_journal.o blockstore_read.o \
-	blockstore_write.o blockstore_sync.o blockstore_stable.o blockstore_rollback.o blockstore_flush.o crc32c.o ringloop.o timerfd_interval.o
-# -fsanitize=address
-CXXFLAGS := -g -O3 -Wall -Wno-sign-compare -Wno-comment -Wno-parentheses -Wno-pointer-arith -fPIC -fdiagnostics-color=always
-all: $(BLOCKSTORE_OBJS) libfio_blockstore.so osd libfio_sec_osd.so test_blockstore stub_osd stub_bench osd_test
-clean:
-	rm -f *.o
-
-crc32c.o: crc32c.c
-	g++ $(CXXFLAGS) -c -o $@ $<
-json11.o: json11/json11.cpp
-	g++ $(CXXFLAGS) -c -o json11.o json11/json11.cpp
-allocator.o: allocator.cpp allocator.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-ringloop.o: ringloop.cpp ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-timerfd_interval.o: timerfd_interval.cpp timerfd_interval.h ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-
-%.o: %.cpp allocator.h blockstore_flush.h blockstore.h blockstore_impl.h blockstore_init.h blockstore_journal.h crc32c.h ringloop.h timerfd_interval.h object_id.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-
-libblockstore.so: $(BLOCKSTORE_OBJS)
-	g++ $(CXXFLAGS) -o libblockstore.so -shared $(BLOCKSTORE_OBJS) -ltcmalloc_minimal -luring
-libfio_blockstore.so: ./libblockstore.so fio_engine.cpp json11.o
-	g++ $(CXXFLAGS) -shared -o libfio_blockstore.so fio_engine.cpp json11.o ./libblockstore.so -ltcmalloc_minimal -luring
-
-OSD_OBJS := osd.o osd_secondary.o osd_receive.o osd_send.o osd_peering.o osd_peering_pg.o osd_primary.o osd_rmw.o json11.o timerfd_interval.o
-osd_secondary.o: osd_secondary.cpp osd.h osd_ops.h ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-osd_receive.o: osd_receive.cpp osd.h osd_ops.h ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-osd_send.o: osd_send.cpp osd.h osd_ops.h ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-osd_peering.o: osd_peering.cpp osd.h osd_ops.h osd_peering_pg.h ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-osd_peering_pg.o: osd_peering_pg.cpp object_id.h osd_peering_pg.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-osd_rmw.o: osd_rmw.cpp osd_rmw.h xor.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-osd_rmw_test: osd_rmw_test.cpp osd_rmw.cpp osd_rmw.h xor.h
-	g++ $(CXXFLAGS) -o $@ $<
-osd_primary.o: osd_primary.cpp osd.h osd_ops.h osd_peering_pg.h xor.h ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-osd.o: osd.cpp osd.h osd_ops.h osd_peering_pg.h ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-osd: ./libblockstore.so osd_main.cpp osd.h osd_ops.h $(OSD_OBJS)
-	g++ $(CXXFLAGS) -o osd osd_main.cpp $(OSD_OBJS) ./libblockstore.so -ltcmalloc_minimal -luring
-stub_osd: stub_osd.cpp osd_ops.h rw_blocking.o
-	g++ $(CXXFLAGS) -o stub_osd stub_osd.cpp rw_blocking.o -ltcmalloc_minimal
-stub_bench: stub_bench.cpp osd_ops.h rw_blocking.o
-	g++ $(CXXFLAGS) -o stub_bench stub_bench.cpp rw_blocking.o -ltcmalloc_minimal
-rw_blocking.o: rw_blocking.cpp rw_blocking.h
-	g++ $(CXXFLAGS) -c -o $@ $<
-osd_test: osd_test.cpp osd_ops.h rw_blocking.o
-	g++ $(CXXFLAGS) -o osd_test osd_test.cpp rw_blocking.o -ltcmalloc_minimal
-
-libfio_sec_osd.so: fio_sec_osd.cpp osd_ops.h rw_blocking.o
-	g++ $(CXXFLAGS) -ltcmalloc_minimal -shared -o libfio_sec_osd.so fio_sec_osd.cpp rw_blocking.o -luring
-
-test_blockstore: ./libblockstore.so test_blockstore.cpp
-	g++ $(CXXFLAGS) -o test_blockstore test_blockstore.cpp ./libblockstore.so -ltcmalloc_minimal -luring
-test: test.cpp osd_peering_pg.o
-	g++ $(CXXFLAGS) -o test test.cpp osd_peering_pg.o -luring
-test_allocator: test_allocator.cpp allocator.o
-	g++ $(CXXFLAGS) -o test_allocator test_allocator.cpp allocator.o

README-ru.md

@@ -0,0 +1,491 @@
+## Vitastor
+
+[Read English version](README.md)
+
+## Идея
+
+Я всего лишь хочу сделать качественную блочную SDS!
+
+Vitastor - распределённая блочная SDS, прямой аналог Ceph RBD и внутренних СХД популярных
+облачных провайдеров. Однако, в отличие от них, Vitastor быстрый и при этом простой.
+Только пока маленький :-).
+
+Архитектурная схожесть с Ceph означает заложенную на уровне алгоритмов записи строгую консистентность,
+репликацию через первичный OSD, симметричную кластеризацию без единой точки отказа
+и автоматическое распределение данных по любому числу дисков любого размера с настраиваемыми схемами
+избыточности - репликацией или с произвольными кодами коррекции ошибок.
+
+## Возможности
+
+Vitastor на данный момент находится в статусе предварительного выпуска, расширенные
+возможности пока отсутствуют, а в будущих версиях вероятны "ломающие" изменения.
+
+Однако следующее уже реализовано:
+
+- Базовая часть - надёжное кластерное блочное хранилище без единой точки отказа
+- Производительность ;-D
+- Несколько схем отказоустойчивости: репликация, XOR n+1 (1 диск чётности), коды коррекции ошибок
+  Рида-Соломона на основе библиотеки jerasure с любым числом дисков данных и чётности в группе
+- Конфигурация через простые человекочитаемые JSON-структуры в etcd
+- Автоматическое распределение данных по OSD, с поддержкой:
+  - Математической оптимизации для лучшей равномерности распределения и минимизации перемещений данных
+  - Нескольких пулов с разными схемами избыточности
+  - Дерева распределения, выбора OSD по тегам / классам устройств (только SSD, только HDD) и по поддереву
+  - Настраиваемых доменов отказа (диск/сервер/стойка и т.п.)
+- Восстановление деградированных блоков
+- Ребаланс, то есть перемещение данных между OSD (дисками)
+- Поддержка "ленивого" fsync (fsync не на каждую операцию)
+- Сбор статистики ввода/вывода в etcd
+- Клиентская библиотека режима пользователя для ввода/вывода
+- Драйвер диска для QEMU (собирается вне дерева исходников QEMU)
+- Драйвер диска для утилиты тестирования производительности fio (также собирается вне дерева исходников fio)
+- NBD-прокси для монтирования образов ядром ("блочное устройство в режиме пользователя")
+- Утилита удаления образов/инодов (vitastor-rm)
+- Пакеты для Debian и CentOS
+- Статистика операций ввода/вывода и занятого места в разрезе инодов
+- Именование инодов через хранение их метаданных в etcd
+- Снапшоты и copy-on-write клоны
+
+## Планы разработки
+
+- Более корректные скрипты разметки дисков и автоматического запуска OSD
+- Другие инструменты администрирования
+- Плагины для OpenStack, Kubernetes, OpenNebula, Proxmox и других облачных систем
+- iSCSI-прокси
+- Таймауты операций и более быстрое выявление отказов
+- Фоновая проверка целостности без контрольных сумм (сверка реплик)
+- Контрольные суммы
+- Оптимизации для гибридных SSD+HDD хранилищ
+- Поддержка RDMA и NVDIMM
+- Web-интерфейс
+- Возможно, сжатие
+- Возможно, поддержка кэширования данных через системный page cache
+
+## Архитектура
+
+Так же, как и в Ceph, в Vitastor:
+
+- Есть пулы (pools), PG, OSD, мониторы, домены отказа, дерево распределения (аналог crush-дерева).
+- Образы делятся на блоки фиксированного размера (объекты), и эти объекты распределяются по OSD.
+- У OSD есть журнал и метаданные и они тоже могут размещаться на отдельных быстрых дисках.
+- Все операции записи тоже транзакционны. В Vitastor, правда, есть режим отложенного/ленивого fsync
+  (коммита), в котором fsync не вызывается на каждую операцию записи, что делает его более
+  пригодным для использования на "плохих" (десктопных) SSD. Однако все операции записи
+  в любом случае атомарны.
+- Клиентская библиотека тоже старается ждать восстановления после любого отказа кластера, то есть,
+  вы тоже можете перезагрузить хоть весь кластер разом, и клиенты только на время зависнут,
+  но не отключатся.
+
+Некоторые базовые термины для тех, кто не знаком с Ceph:
+
+- OSD (Object Storage Daemon) - процесс, который хранит данные на одном диске и обрабатывает
+  запросы чтения/записи от клиентов.
+- Пул (Pool) - контейнер для данных, имеющих одну и ту же схему избыточности и правила распределения по OSD.
+- PG (Placement Group) - группа объектов, хранимых на одном и том же наборе реплик (OSD).
+  Несколько PG могут храниться на одном и том же наборе реплик, но объекты одной PG
+  в норме не хранятся на разных наборах OSD.
+- Монитор - демон, хранящий состояние кластера.
+- Домен отказа (Failure Domain) - группа OSD, которым вы разрешаете "упасть" всем вместе.
+  Иными словами, это группа OSD, в которые СХД не помещает разные копии одного и того же
+  блока данных. Например, если домен отказа - сервер, то на двух дисках одного сервера
+  никогда не окажется 2 и более копий одного и того же блока данных, а значит, даже
+  если в этом сервере откажут все диски, это будет равносильно потере только 1 копии
+  любого блока данных.
+- Дерево распределения (Placement Tree / CRUSH Tree) - иерархическая группировка OSD
+  в узлы, которые далее можно использовать как домены отказа. То есть, диск (OSD) входит в
+  сервер, сервер входит в стойку, стойка входит в ряд, ряд в датацентр и т.п.
+
+Чем Vitastor отличается от Ceph:
+
+- Vitastor в первую очередь сфокусирован на SSD. Также Vitastor, вероятно, должен неплохо работать
+  с комбинацией SSD и HDD через bcache, а в будущем, возможно, будут добавлены и нативные способы
+  оптимизации под SSD+HDD. Однако хранилище на основе одних лишь жёстких дисков, вообще без SSD,
+  не в приоритете, поэтому оптимизации под этот кейс могут вообще не состояться.
+- OSD Vitastor однопоточный и всегда таким останется, так как это самый оптимальный способ работы.
+  Если вам не хватает 1 ядра на 1 диск, просто делите диск на разделы и запускайте на нём несколько OSD.
+  Но, скорее всего, вам хватит и 1 ядра - Vitastor не так прожорлив к ресурсам CPU, как Ceph.
+- Журнал и метаданные всегда размещаются в памяти, благодаря чему никогда не тратится лишнее время
+  на чтение метаданных с диска. Размер метаданных линейно зависит от размера диска и блока данных,
+  который задаётся в конфигурации кластера и по умолчанию составляет 128 КБ. С блоком 128 КБ метаданные
+  занимают примерно 512 МБ памяти на 1 ТБ дискового пространства (и это всё равно меньше, чем нужно Ceph-у).
+  Журнал вообще не должен быть большим, например, тесты производительности в данном документе проводились
+  с журналом размером всего 16 МБ. Большой журнал, вероятно, даже вреден, т.к. "грязные" записи (записи,
+  не сброшенные из журнала) тоже занимают память и могут немного замедлять работу.
+- В Vitastor нет внутреннего copy-on-write. Я считаю, что реализация CoW-хранилища гораздо сложнее,
+  поэтому сложнее добиться устойчиво хороших результатов. Возможно, в один прекрасный день
+  я придумаю красивый алгоритм для CoW-хранилища, но пока нет - внутреннего CoW в Vitastor не будет.
+  Всё это не относится к "внешнему" CoW (снапшотам и клонам).
+- Базовый слой Vitastor - простое блочное хранилище с блоками фиксированного размера, а не сложное
+  объектное хранилище с расширенными возможностями, как в Ceph (RADOS).
+- В Vitastor есть режим "ленивых fsync", в котором OSD группирует запросы записи перед сбросом их
+  на диск, что позволяет получить лучшую производительность с дешёвыми настольными SSD без конденсаторов
+  ("Advanced Power Loss Protection" / "Capacitor-Based Power Loss Protection").
+  Тем не менее, такой режим всё равно медленнее использования нормальных серверных SSD и мгновенного
+  fsync, так как приводит к дополнительным операциям передачи данных по сети, поэтому рекомендуется
+  всё-таки использовать хорошие серверные диски, тем более, стоят они почти так же, как десктопные.
+- PG эфемерны. Это означает, что они не хранятся на дисках и существуют только в памяти работающих OSD.
+- Процессы восстановления оперируют отдельными объектами, а не целыми PG.
+- PGLOG-ов нет.
+- "Мониторы" не хранят данные. Конфигурация и состояние кластера хранятся в etcd в простых человекочитаемых
+  JSON-структурах. Мониторы Vitastor только следят за состоянием кластера и управляют перемещением данных.
+  В этом смысле монитор Vitastor не является критичным компонентом системы и больше похож на Ceph-овский
+  менеджер (MGR). Монитор Vitastor написан на node.js.
+- Распределение PG не основано на консистентных хешах. Вместо этого все маппинги PG хранятся прямо в etcd
+  (ибо нет никакой проблемы сохранить несколько сотен-тысяч записей в памяти, а не считать каждый раз хеши).
+  Перераспределение PG по OSD выполняется через математическую оптимизацию,
+  а конкретно, сведение задачи к ЛП (задаче линейного программирования) и решение оной с помощью утилиты
+  lp_solve. Такой подход позволяет обычно выравнивать распределение места почти идеально - равномерность
+  обычно составляет 96-99%, в отличие от Ceph, где на голом CRUSH-е без балансировщика обычно выходит 80-90%.
+  Также это позволяет минимизировать объём перемещения данных и случайность связей между OSD, а также менять
+  распределение вручную, не боясь сломать логику перебалансировки. В таком подходе есть и потенциальный
+  недостаток - есть предположение, что в очень большом кластере он может сломаться - однако вплоть до
+  нескольких сотен OSD подход точно работает нормально. Ну и, собственно, при необходимости легко
+  реализовать и консистентные хеши.
+- Отдельный слой, подобный слою "CRUSH-правил", отсутствует. Вы настраиваете схемы отказоустойчивости,
+  домены отказа и правила выбора OSD напрямую в конфигурации пулов.
+
+## Понимание сути производительности систем хранения
+
+Вкратце: для быстрой хранилки задержки важнее, чем пиковые iops-ы.
+
+Лучшая возможная задержка достигается при тестировании в 1 поток с глубиной очереди 1,
+что приблизительно означает минимально нагруженное состояние кластера. В данном случае
+IOPS = 1/задержка. Ни числом серверов, ни дисков, ни серверных процессов/потоков
+задержка не масштабируется... Она зависит только от того, насколько быстро один
+серверный процесс (и клиент) обрабатывают одну операцию.
+
+Почему задержки важны? Потому, что некоторые приложения *не могут* использовать глубину
+очереди больше 1, ибо их задача не параллелизуется. Важный пример - это все СУБД
+с поддержкой консистентности (ACID), потому что все они обеспечивают её через
+журналирование, а журналы пишутся последовательно и с fsync() после каждой операции.
+
+fsync, кстати - это ещё одна очень важная вещь, про которую почти всегда забывают в тестах.
+Смысл в том, что все современные диски имеют кэши/буферы записи и не гарантируют, что
+данные реально физически записываются на носитель до того, как вы делаете fsync(),
+который транслируется в команду сброса кэша операционной системой.
+
+Дешёвые SSD для настольных ПК и ноутбуков очень быстрые без fsync - NVMe диски, например,
+могут обработать порядка 80000 операций записи в секунду с глубиной очереди 1 без fsync.
+Однако с fsync, когда они реально вынуждены писать каждый блок данных во флеш-память,
+они выжимают лишь 1000-2000 операций записи в секунду (число практически постоянное
+для всех моделей SSD).
+
+Серверные SSD часто имеют суперконденсаторы, работающие как встроенный источник
+бесперебойного питания и дающие дискам успеть сбросить их DRAM-кэш в постоянную
+флеш-память при отключении питания. Благодаря этому диски с чистой совестью
+*игнорируют fsync*, так как точно знают, что данные из кэша доедут до постоянной
+памяти.
+
+Все наиболее известные программные СХД, например, Ceph и внутренние СХД, используемые
+такими облачными провайдерами, как Amazon, Google, Яндекс, медленные в смысле задержки.
+В лучшем случае они дают задержки от 0.3мс на чтение и 0.6мс на запись 4 КБ блоками
+даже при условии использования наилучшего возможного железа.
+
+И это в эпоху SSD, когда вы можете пойти на рынок и купить там SSD, задержка которого
+на чтение будет 0.1мс, а на запись - 0.04мс, за 100$ или даже дешевле.
+
+Когда мне нужно быстро протестировать производительность дисковой подсистемы, я
+использую следующие 6 команд, с небольшими вариациями:
+
+- Линейная запись:
+  `fio -ioengine=libaio -direct=1 -invalidate=1 -name=test -bs=4M -iodepth=32 -rw=write -runtime=60 -filename=/dev/sdX`
+- Линейное чтение:
+  `fio -ioengine=libaio -direct=1 -invalidate=1 -name=test -bs=4M -iodepth=32 -rw=read -runtime=60 -filename=/dev/sdX`
+- Запись в 1 поток (T1Q1):
+  `fio -ioengine=libaio -direct=1 -invalidate=1 -name=test -bs=4k -iodepth=1 -fsync=1 -rw=randwrite -runtime=60 -filename=/dev/sdX`
+- Чтение в 1 поток (T1Q1):
+  `fio -ioengine=libaio -direct=1 -invalidate=1 -name=test -bs=4k -iodepth=1 -rw=randread -runtime=60 -filename=/dev/sdX`
+- Параллельная запись (numjobs используется, когда 1 ядро CPU не может насытить диск):
+  `fio -ioengine=libaio -direct=1 -invalidate=1 -name=test -bs=4k -iodepth=128 [-numjobs=4 -group_reporting] -rw=randwrite -runtime=60 -filename=/dev/sdX`
+- Параллельное чтение (numjobs - аналогично):
+  `fio -ioengine=libaio -direct=1 -invalidate=1 -name=test -bs=4k -iodepth=128 [-numjobs=4 -group_reporting] -rw=randread -runtime=60 -filename=/dev/sdX`
+
+## Теоретическая максимальная производительность Vitastor
+
+При использовании репликации:
+- Задержка чтения в 1 поток (T1Q1): 1 сетевой RTT + 1 чтение с диска.
+- Запись+fsync в 1 поток:
+  - С мгновенным сбросом: 2 RTT + 1 запись.
+  - С отложенным ("ленивым") сбросом: 4 RTT + 1 запись + 1 fsync.
+- Параллельное чтение: сумма IOPS всех дисков либо производительность сети, если в сеть упрётся раньше.
+- Параллельная запись: сумма IOPS всех дисков / число реплик / WA либо производительность сети, если в сеть упрётся раньше.
+
+При использовании кодов коррекции ошибок (EC):
+- Задержка чтения в 1 поток (T1Q1): 1.5 RTT + 1 чтение.
+- Запись+fsync в 1 поток:
+  - С мгновенным сбросом: 3.5 RTT + 1 чтение + 2 записи.
+  - С отложенным ("ленивым") сбросом: 5.5 RTT + 1 чтение + 2 записи + 2 fsync.
+- Под 0.5 на самом деле подразумевается (k-1)/k, где k - число дисков данных,
+  что означает, что дополнительное обращение по сети не нужно, когда операция
+  чтения обслуживается локально.
+- Параллельное чтение: сумма IOPS всех дисков либо производительность сети, если в сеть упрётся раньше.
+- Параллельная запись: сумма IOPS всех дисков / общее число дисков данных и чётности / WA либо производительность сети, если в сеть упрётся раньше.
+  Примечание: IOPS дисков в данном случае надо брать в смешанном режиме чтения/записи в пропорции, аналогичной формулам выше.
+
+WA (мультипликатор записи) для 4 КБ блоков в Vitastor обычно составляет 3-5:
+1. Запись метаданных в журнал
+2. Запись блока данных в журнал
+3. Запись метаданных в БД
+4. Ещё одна запись метаданных в журнал при использовании EC
+5. Запись блока данных на диск данных
+
+Если вы найдёте SSD, хорошо работающий с 512-байтными блоками данных (Optane?),
+то 1, 3 и 4 можно снизить до 512 байт (1/8 от размера данных) и получить WA всего 2.375.
+
+Кроме того, WA снижается при использовании отложенного/ленивого сброса при параллельной
+нагрузке, т.к. блоки журнала записываются на диск только когда они заполняются или явным
+образом запрашивается fsync.
+
+## Пример сравнения с Ceph
+
+Железо - 4 сервера, в каждом:
+- 6x SATA SSD Intel D3-4510 3.84 TB
+- 2x Xeon Gold 6242 (16 cores @ 2.8 GHz)
+- 384 GB RAM
+- 1x 25 GbE сетевая карта (Mellanox ConnectX-4 LX), подключённая к свитчу Juniper QFX5200
+
+Экономия энергии CPU отключена. В тестах и Vitastor, и Ceph развёрнуто по 2 OSD на 1 SSD.
+
+Все результаты ниже относятся к случайной нагрузке 4 КБ блоками (если явно не указано обратное).
+
+Производительность голых дисков:
+- T1Q1 запись ~27000 iops (задержка ~0.037ms)
+- T1Q1 чтение ~9800 iops (задержка ~0.101ms)
+- T1Q32 запись ~60000 iops
+- T1Q32 чтение ~81700 iops
+
+Ceph 15.2.4 (Bluestore):
+- T1Q1 запись ~1000 iops (задержка ~1ms)
+- T1Q1 чтение ~1750 iops (задержка ~0.57ms)
+- T8Q64 запись ~100000 iops, потребление CPU процессами OSD около 40 ядер на каждом сервере
+- T8Q64 чтение ~480000 iops, потребление CPU процессами OSD около 40 ядер на каждом сервере
+
+Тесты в 8 потоков проводились на 8 400GB RBD образах со всех хостов (с каждого хоста запускалось 2 процесса fio).
+Это нужно потому, что в Ceph несколько RBD-клиентов, пишущих в 1 образ, очень сильно замедляются.
+
+Настройки RocksDB и Bluestore в Ceph не менялись, единственным изменением было отключение cephx_sign_messages.
+
+На самом деле, результаты теста не такие уж и плохие для Ceph (могло быть хуже).
+Собственно говоря, эти серверы как раз хорошо сбалансированы для Ceph - 6 SATA SSD как раз
+утилизируют 25-гигабитную сеть, а без 2 мощных процессоров Ceph-у бы не хватило ядер,
+чтобы выдать пристойный результат. Собственно, что и показывает жор 40 ядер в процессе
+параллельного теста.
+
+Vitastor:
+- T1Q1 запись: 7087 iops (задержка 0.14ms)
+- T1Q1 чтение: 6838 iops (задержка 0.145ms)
+- T2Q64 запись: 162000 iops, потребление CPU - 3 ядра на каждом сервере
+- T8Q64 чтение: 895000 iops, потребление CPU - 4 ядра на каждом сервере
+- Линейная запись (4M T1Q32): 2800 МБ/с
+- Линейное чтение (4M T1Q32): 1500 МБ/с
+
+Тест на чтение в 8 потоков проводился на 1 большом образе (3.2 ТБ) со всех хостов (опять же, по 2 fio с каждого).
+В Vitastor никакой разницы между 1 образом и 8-ю нет. Естественно, примерно 1/4 запросов чтения
+в такой конфигурации, как и в тестах Ceph выше, обслуживалась с локальной машины. Если проводить
+тест так, чтобы все операции всегда обращались к первичным OSD по сети - тест сильнее упирался
+в сеть и результат составлял примерно 689000 iops.
+
+Настройки Vitastor: `--disable_data_fsync true --immediate_commit all --flusher_count 8
+  --disk_alignment 4096 --journal_block_size 4096 --meta_block_size 4096
+  --journal_no_same_sector_overwrites true --journal_sector_buffer_count 1024
+  --journal_size 16777216`.
+
+### EC/XOR 2+1
+
+Vitastor:
+- T1Q1 запись: 2808 iops (задержка ~0.355ms)
+- T1Q1 чтение: 6190 iops (задержка ~0.16ms)
+- T2Q64 запись: 85500 iops, потребление CPU - 3.4 ядра на каждом сервере
+- T8Q64 чтение: 812000 iops, потребление CPU - 4.7 ядра на каждом сервере
+- Линейная запись (4M T1Q32): 3200 МБ/с
+- Линейное чтение (4M T1Q32): 1800 МБ/с
+
+Ceph:
+- T1Q1 запись: 730 iops (задержка ~1.37ms latency)
+- T1Q1 чтение: 1500 iops с холодным кэшем метаданных (задержка ~0.66ms), 2300 iops через 2 минуты прогрева (задержка ~0.435ms)
+- T4Q128 запись (4 RBD images): 45300 iops, потребление CPU - 30 ядер на каждом сервере
+- T8Q64 чтение (4 RBD images): 278600 iops, потребление CPU - 40 ядер на каждом сервере
+- Линейная запись (4M T1Q32): 1950 МБ/с в пустой образ, 2500 МБ/с в заполненный образ
+- Линейное чтение (4M T1Q32): 2400 МБ/с
+
+### NBD
+
+NBD - на данный момент единственный способ монтировать Vitastor ядром Linux, но он
+приводит к дополнительным копированиям данных, поэтому немного ухудшает производительность,
+правда, в основном - линейную, а случайная затрагивается слабо.
+
+NBD расшифровывается как "сетевое блочное устройство", но на самом деле оно также
+работает просто как аналог FUSE для блочных устройств, то есть, представляет собой
+"блочное устройство в пространстве пользователя".
+
+Vitastor с однопоточной NBD прокси на том же стенде:
+- T1Q1 запись: 6000 iops (задержка 0.166ms)
+- T1Q1 чтение: 5518 iops (задержка 0.18ms)
+- T1Q128 запись: 94400 iops
+- T1Q128 чтение: 103000 iops
+- Линейная запись (4M T1Q128): 1266 МБ/с (в сравнении с 2800 МБ/с через fio)
+- Линейное чтение (4M T1Q128): 975 МБ/с (в сравнении с 1500 МБ/с через fio)
+
+## Установка
+
+### Debian
+
+- Добавьте ключ репозитория Vitastor:
+  `wget -q -O - https://vitastor.io/debian/pubkey | sudo apt-key add -`
+- Добавьте репозиторий Vitastor в /etc/apt/sources.list:
+  - Debian 11 (Bullseye/Sid): `deb https://vitastor.io/debian bullseye main`
+  - Debian 10 (Buster): `deb https://vitastor.io/debian buster main`
+- Для Debian 10 (Buster) также включите репозиторий backports:
+  `deb http://deb.debian.org/debian buster-backports main`
+- Установите пакеты: `apt update; apt install vitastor lp-solve etcd linux-image-amd64 qemu`
+
+### CentOS
+
+- Добавьте в систему репозиторий Vitastor:
+  - CentOS 7: `yum install https://vitastor.io/rpms/centos/7/vitastor-release-1.0-1.el7.noarch.rpm`
+  - CentOS 8: `dnf install https://vitastor.io/rpms/centos/8/vitastor-release-1.0-1.el8.noarch.rpm`
+- Включите EPEL: `yum/dnf install epel-release`
+- Включите дополнительные репозитории CentOS:
+  - CentOS 7: `yum install centos-release-scl`
+  - CentOS 8: `dnf install centos-release-advanced-virtualization`
+- Включите elrepo-kernel:
+  - CentOS 7: `yum install https://www.elrepo.org/elrepo-release-7.el7.elrepo.noarch.rpm`
+  - CentOS 8: `dnf install https://www.elrepo.org/elrepo-release-8.el8.elrepo.noarch.rpm`
+- Установите пакеты: `yum/dnf install vitastor lpsolve etcd kernel-ml qemu-kvm`
+
+### Установка из исходников
+
+- Установите ядро 5.4 или более новое, для поддержки io_uring. Желательно 5.8 или даже новее,
+  так как в 5.4 есть как минимум 1 известный баг, ведущий к зависанию с io_uring и контроллером HP SmartArray.
+- Установите liburing 0.4 или более новый и его заголовки.
+- Установите lp_solve.
+- Установите etcd. Внимание: вам нужна версия с исправлением отсюда: https://github.com/vitalif/etcd/,
+  из ветки release-3.4, так как в etcd есть баг, который [будет](https://github.com/etcd-io/etcd/pull/12402)
+  исправлен только в 3.4.15. Баг приводит к неспособности Vitastor запустить PG, когда их хотя бы 500 штук.
+- Установите node.js 10 или новее.
+- Установите gcc и g++ 8.x или новее.
+- Склонируйте данный репозиторий с подмодулями: `git clone https://yourcmc.ru/git/vitalif/vitastor/`.
+- Желательно пересобрать QEMU с патчем, который делает необязательным запуск через LD_PRELOAD.
+  См `qemu-*.*-vitastor.patch` - выберите версию, наиболее близкую вашей версии QEMU.
+- Установите QEMU 3.0 или новее, возьмите исходные коды установленного пакета, начните его пересборку,
+  через некоторое время остановите её и скопируйте следующие заголовки:
+   - `<qemu>/include` &rarr; `<vitastor>/qemu/include`
+   - Debian:
+      * Берите qemu из основного репозитория
+      * `<qemu>/b/qemu/config-host.h` &rarr; `<vitastor>/qemu/b/qemu/config-host.h`
+      * `<qemu>/b/qemu/qapi` &rarr; `<vitastor>/qemu/b/qemu/qapi`
+   - CentOS 8:
+      * Берите qemu из репозитория Advanced-Virtualization. Чтобы включить его, запустите
+        `yum install centos-release-advanced-virtualization.noarch` и далее `yum install qemu`
+      * `<qemu>/config-host.h` &rarr; `<vitastor>/qemu/b/qemu/config-host.h`
+      * Для QEMU 3.0+: `<qemu>/qapi` &rarr; `<vitastor>/qemu/b/qemu/qapi`
+      * Для QEMU 2.0+: `<qemu>/qapi-types.h` &rarr; `<vitastor>/qemu/b/qemu/qapi-types.h`
+   - `config-host.h` и `qapi` нужны, т.к. в них содержатся автогенерируемые заголовки
+- Установите fio 3.7 или новее, возьмите исходники пакета и сделайте на них симлинк с `<vitastor>/fio`.
+- Соберите и установите Vitastor командой `mkdir build && cd build && cmake .. && make -j8 && make install`.
+  Обратите внимание на переменную cmake `QEMU_PLUGINDIR` - под RHEL её нужно установить равной `qemu-kvm`.
+
+## Запуск
+
+Внимание: процедура пока что достаточно нетривиальная, задавать конфигурацию и смещения
+на диске нужно почти вручную. Это будет исправлено в ближайшем будущем.
+
+- Желательны SATA SSD или NVMe диски с конденсаторами (серверные SSD). Можно использовать и
+  десктопные SSD, включив режим отложенного fsync, но производительность однопоточной записи
+  в этом случае пострадает.
+- Быстрая сеть, минимум 10 гбит/с
+- Для наилучшей производительности нужно отключить энергосбережение CPU: `cpupower idle-set -D 0 && cpupower frequency-set -g performance`.
+- Пропишите нужные вам значения вверху файлов `/usr/lib/vitastor/mon/make-units.sh` и `/usr/lib/vitastor/mon/make-osd.sh`.
+- Создайте юниты systemd для etcd и мониторов: `/usr/lib/vitastor/mon/make-units.sh`
+- Создайте юниты для OSD: `/usr/lib/vitastor/mon/make-osd.sh /dev/disk/by-partuuid/XXX [/dev/disk/by-partuuid/YYY ...]`
+- Вы можете поменять параметры OSD в юнитах systemd. Смысл некоторых параметров:
+  - `disable_data_fsync 1` - отключает fsync, используется с SSD с конденсаторами.
+  - `immediate_commit all` - используется с SSD с конденсаторами.
+  - `disable_device_lock 1` - отключает блокировку файла устройства, нужно, только если вы запускаете
+    несколько OSD на одном блочном устройстве.
+  - `flusher_count 256` - "flusher" - микропоток, удаляющий старые данные из журнала.
+    Не волнуйтесь об этой настройке, 256 теперь достаточно практически всегда.
+  - `disk_alignment`, `journal_block_size`, `meta_block_size` следует установить равными размеру
+    внутреннего блока SSD. Это почти всегда 4096.
+  - `journal_no_same_sector_overwrites true` запрещает перезапись одного и того же сектора журнала подряд
+    много раз в процессе записи. Большинство (99%) SSD не нуждаются в данной опции. Однако выяснилось, что
+    диски, используемые на одном из тестовых стендов - Intel D3-S4510 - очень сильно не любят такую
+    перезапись, и для них была добавлена эта опция. Когда данный режим включён, также нужно поднимать
+    значение `journal_sector_buffer_count`, так как иначе Vitastor не хватит буферов для записи в журнал.
+- Запустите все etcd: `systemctl start etcd`
+- Создайте глобальную конфигурацию в etcd: `etcdctl --endpoints=... put /vitastor/config/global '{"immediate_commit":"all"}'`
+  (если все ваши диски - серверные с конденсаторами).
+- Создайте пулы: `etcdctl --endpoints=... put /vitastor/config/pools '{"1":{"name":"testpool","scheme":"replicated","pg_size":2,"pg_minsize":1,"pg_count":256,"failure_domain":"host"}}'`.
+  Для jerasure EC-пулов конфигурация должна выглядеть так: `2:{"name":"ecpool","scheme":"jerasure","pg_size":4,"parity_chunks":2,"pg_minsize":2,"pg_count":256,"failure_domain":"host"}`.
+- Запустите все OSD: `systemctl start vitastor.target`
+- Ваш кластер должен быть готов - один из мониторов должен уже сконфигурировать PG, а OSD должны запустить их.
+- Вы можете проверить состояние PG прямо в etcd: `etcdctl --endpoints=... get --prefix /vitastor/pg/state`. Все PG должны быть 'active'.
+- Пример команды для запуска тестов: `fio -thread -ioengine=libfio_vitastor.so -name=test -bs=4M -direct=1 -iodepth=16 -rw=write -etcd=10.115.0.10:2379/v3 -pool=1 -inode=1 -size=400G`.
+- Пример команды для заливки образа ВМ в vitastor через qemu-img:
+  ```
+  qemu-img convert -f qcow2 debian10.qcow2 -p -O raw 'vitastor:etcd_host=10.115.0.10\:2379/v3:pool=1:inode=1:size=2147483648'
+  ```
+  Если вы используете немодифицированный QEMU, данной команде потребуется переменная окружения `LD_PRELOAD=/usr/lib/x86_64-linux-gnu/qemu/block-vitastor.so`.
+- Пример команды запуска QEMU:
+  ```
+  qemu-system-x86_64 -enable-kvm -m 1024
+    -drive 'file=vitastor:etcd_host=10.115.0.10\:2379/v3:pool=1:inode=1:size=2147483648',format=raw,if=none,id=drive-virtio-disk0,cache=none
+    -device virtio-blk-pci,scsi=off,bus=pci.0,addr=0x5,drive=drive-virtio-disk0,id=virtio-disk0,bootindex=1,write-cache=off,physical_block_size=4096,logical_block_size=512
+    -vnc 0.0.0.0:0
+  ```
+- Пример команды удаления образа (инода) из Vitastor:
+  ```
+  vitastor-rm --etcd_address 10.115.0.10:2379/v3 --pool 1 --inode 1 --parallel_osds 16 --iodepth 32
+  ```
+
+## Известные проблемы
+
+- Запросы удаления объектов могут в данный момент приводить к "неполным" объектам в EC-пулах,
+  если в процессе удаления произойдут отказы OSD или серверов, потому что правильная обработка
+  запросов удаления в кластере должна быть "трёхфазной", а это пока не реализовано. Если вы
+  столкнётесь с такой ситуацией, просто повторите запрос удаления.
+
+## Принципы реализации
+
+- Я люблю архитектурно простые решения. Vitastor проектируется именно так и я намерен
+  и далее следовать данному принципу.
+- Если вы пришли сюда за идеальным кодом на C++, вы, вероятно, не по адресу. "Общепринятые"
+  практики написания C++ кода меня не очень волнуют, так как зачастую, опять-таки, ведут к
+  излишним усложнениям и код получается красивый... но медленный.
+- По той же причине в коде иногда можно встретить велосипеды типа собственного упрощённого
+  HTTP-клиента для работы с etcd. Зато эти велосипеды маленькие и компактные и не требуют
+  использования десятка внешних библиотек.
+- node.js для монитора - не случайный выбор. Он очень быстрый, имеет встроенную событийную
+  машину, приятный нейтральный C-подобный язык программирования и развитую инфраструктуру.
+
+## Автор и лицензия
+
+Автор: Виталий Филиппов (vitalif [at] yourcmc.ru), 2019+
+
+Заходите в Telegram-чат Vitastor: https://t.me/vitastor
+
+Лицензия: VNPL 1.1 на серверный код и двойная VNPL 1.1 + GPL 2.0+ на клиентский.
+
+VNPL - "сетевой копилефт", собственная свободная копилефт-лицензия
+Vitastor Network Public License 1.1, основанная на GNU GPL 3.0 с дополнительным
+условием "Сетевого взаимодействия", требующим распространять все программы,
+специально разработанные для использования вместе с Vitastor и взаимодействующие
+с ним по сети, под лицензией VNPL или под любой другой свободной лицензией.
+
+Идея VNPL - расширение действия копилефта не только на модули, явным образом
+связываемые с кодом Vitastor, но также на модули, оформленные в виде микросервисов
+и взаимодействующие с ним по сети.
+
+Таким образом, если вы хотите построить на основе Vitastor сервис, содержаший
+компоненты с закрытым кодом, взаимодействующие с Vitastor, вам нужна коммерческая
+лицензия от автора 😀.
+
+На Windows и любое другое ПО, не разработанное *специально* для использования
+вместе с Vitastor, никакие ограничения не накладываются.
+
+Клиентские библиотеки распространяются на условиях двойной лицензии VNPL 1.0
+и также на условиях GNU GPL 2.0 или более поздней версии. Так сделано в целях
+совместимости с таким ПО, как QEMU и fio.
+
+Вы можете найти полный текст VNPL 1.1 в файле [VNPL-1.1.txt](VNPL-1.1.txt),
+а GPL 2.0 в файле [GPL-2.0.txt](GPL-2.0.txt).

README.md

@@ -0,0 +1,449 @@
+## Vitastor
+
+[Читать на русском](README-ru.md)
+
+## The Idea
+
+Make Software-Defined Block Storage Great Again.
+
+Vitastor is a small, simple and fast clustered block storage (storage for VM drives),
+architecturally similar to Ceph which means strong consistency, primary-replication, symmetric
+clustering and automatic data distribution over any number of drives of any size
+with configurable redundancy (replication or erasure codes/XOR).
+
+## Features
+
+Vitastor is currently a pre-release, a lot of features are missing and you can still expect
+breaking changes in the future. However, the following is implemented:
+
+- Basic part: highly-available block storage with symmetric clustering and no SPOF
+- Performance ;-D
+- Multiple redundancy schemes: Replication, XOR n+1, Reed-Solomon erasure codes
+  based on jerasure library with any number of data and parity drives in a group
+- Configuration via simple JSON data structures in etcd
+- Automatic data distribution over OSDs, with support for:
+  - Mathematical optimization for better uniformity and less data movement
+  - Multiple pools
+  - Placement tree, OSD selection by tags (device classes) and placement root
+  - Configurable failure domains
+- Recovery of degraded blocks
+- Rebalancing (data movement between OSDs)
+- Lazy fsync support
+- I/O statistics reporting to etcd
+- Generic user-space client library
+- QEMU driver (built out-of-tree)
+- Loadable fio engine for benchmarks (also built out-of-tree)
+- NBD proxy for kernel mounts
+- Inode removal tool (vitastor-rm)
+- Packaging for Debian and CentOS
+- Per-inode I/O and space usage statistics
+- Inode metadata storage in etcd
+- Snapshots and copy-on-write image clones
+
+## Roadmap
+
+- Better OSD creation and auto-start tools
+- Other administrative tools
+- Plugins for OpenStack, Kubernetes, OpenNebula, Proxmox and other cloud systems
+- iSCSI proxy
+- Operation timeouts and better failure detection
+- Scrubbing without checksums (verification of replicas)
+- Checksums
+- SSD+HDD optimizations, possibly including tiered storage and soft journal flushes
+- RDMA and NVDIMM support
+- Web GUI
+- Compression (possibly)
+- Read caching using system page cache (possibly)
+
+## Architecture
+
+Similarities:
+
+- Just like Ceph, Vitastor has Pools, PGs, OSDs, Monitors, Failure Domains, Placement Tree.
+- Just like Ceph, Vitastor is transactional (even though there's a "lazy fsync mode" which
+  doesn't implicitly flush every operation to disks).
+- OSDs also have journal and metadata and they can also be put on separate drives.
+- Just like in Ceph, client library attempts to recover from any cluster failure so
+  you can basically reboot the whole cluster and only pause, but not crash, your clients
+  (I consider this a bug if the client crashes in that case).
+
+Some basic terms for people not familiar with Ceph:
+
+- OSD (Object Storage Daemon) is a process that stores data and serves read/write requests.
+- PG (Placement Group) is a container for data that (normally) shares the same replicas.
+- Pool is a container for data that has the same redundancy scheme and placement rules.
+- Monitor is a separate daemon that watches cluster state and handles failures.
+- Failure Domain is a group of OSDs that you allow to fail. It's "host" by default.
+- Placement Tree groups OSDs in a hierarchy to later split them into Failure Domains.
+
+Architectural differences from Ceph:
+
+- Vitastor's primary focus is on SSDs. Proper SSD+HDD optimizations may be added in the future, though.
+- Vitastor OSD is (and will always be) single-threaded. If you want to dedicate more than 1 core
+  per drive you should run multiple OSDs each on a different partition of the drive.
+  Vitastor isn't CPU-hungry though (as opposed to Ceph), so 1 core is sufficient in a lot of cases.
+- Metadata and journal are always kept in memory. Metadata size depends linearly on drive capacity
+  and data store block size which is 128 KB by default. With 128 KB blocks metadata should occupy
+  around 512 MB per 1 TB (which is still less than Ceph wants). Journal doesn't have to be big,
+  the example test below was conducted with only 16 MB journal. A big journal is probably even
+  harmful as dirty write metadata also take some memory.
+- Vitastor storage layer doesn't have internal copy-on-write or redirect-write. I know that maybe
+  it's possible to create a good copy-on-write storage, but it's much harder and makes performance
+  less deterministic, so CoW isn't used in Vitastor.
+- The basic layer of Vitastor is block storage with fixed-size blocks, not object storage with
+  rich semantics like in Ceph (RADOS).
+- There's a "lazy fsync" mode which allows to batch writes before flushing them to the disk.
+  This allows to use Vitastor with desktop SSDs, but still lowers performance due to additional
+  network roundtrips, so use server SSDs with capacitor-based power loss protection
+  ("Advanced Power Loss Protection") for best performance.
+- PGs are ephemeral. This means that they aren't stored on data disks and only exist in memory
+  while OSDs are running.
+- Recovery process is per-object (per-block), not per-PG. Also there are no PGLOGs.
+- Monitors don't store data. Cluster configuration and state is stored in etcd in simple human-readable
+  JSON structures. Monitors only watch cluster state and handle data movement.
+  Thus Vitastor's Monitor isn't a critical component of the system and is more similar to Ceph's Manager.
+  Vitastor's Monitor is implemented in node.js.
+- PG distribution isn't based on consistent hashes. All PG mappings are stored in etcd.
+  Rebalancing PGs between OSDs is done by mathematical optimization - data distribution problem
+  is reduced to a linear programming problem and solved by lp_solve. This allows for almost
+  perfect (96-99% uniformity compared to Ceph's 80-90%) data distribution in most cases, ability
+  to map PGs by hand without breaking rebalancing logic, reduced OSD peer-to-peer communication
+  (on average, OSDs have fewer peers) and less data movement. It also probably has a drawback -
+  this method may fail in very large clusters, but up to several hundreds of OSDs it's perfectly fine.
+  It's also easy to add consistent hashes in the future if something proves their necessity.
+- There's no separate CRUSH layer. You select pool redundancy scheme, placement root, failure domain
+  and so on directly in pool configuration.
+
+## Understanding Storage Performance
+
+The most important thing for fast storage is latency, not parallel iops.
+
+The best possible latency is achieved with one thread and queue depth of 1 which basically means
+"client load as low as possible". In this case IOPS = 1/latency, and this number doesn't
+scale with number of servers, drives, server processes or threads and so on.
+Single-threaded IOPS and latency numbers only depend on *how fast a single daemon is*.
+
+Why is it important? It's important because some of the applications *can't* use
+queue depth greater than 1 because their task isn't parallelizable. A notable example
+is any ACID DBMS because all of them write their WALs sequentially with fsync()s.
+
+fsync, by the way, is another important thing often missing in benchmarks. The point is
+that drives have cache buffers and don't guarantee that your data is actually persisted
+until you call fsync() which is translated to a FLUSH CACHE command by the OS.
+
+Desktop SSDs are very fast without fsync - NVMes, for example, can process ~80000 write
+operations per second with queue depth of 1 without fsync - but they're really slow with
+fsync because they have to actually write data to flash chips when you call fsync. Typical
+number is around 1000-2000 iops with fsync.
+
+Server SSDs often have supercapacitors that act as a built-in UPS and allow the drive
+to flush its DRAM cache to the persistent flash storage when a power loss occurs.
+This makes them perform equally well with and without fsync. This feature is called
+"Advanced Power Loss Protection" by Intel; other vendors either call it similarly
+or directly as "Full Capacitor-Based Power Loss Protection".
+
+All software-defined storages that I currently know are slow in terms of latency.
+Notable examples are Ceph and internal SDSes used by cloud providers like Amazon, Google,
+Yandex and so on. They're all slow and can only reach ~0.3ms read and ~0.6ms 4 KB write latency
+with best-in-slot hardware.
+
+And that's in the SSD era when you can buy an SSD that has ~0.04ms latency for 100 $.
+
+I use the following 6 commands with small variations to benchmark any storage:
+
+- Linear write:
+  `fio -ioengine=libaio -direct=1 -invalidate=1 -name=test -bs=4M -iodepth=32 -rw=write -runtime=60 -filename=/dev/sdX`
+- Linear read:
+  `fio -ioengine=libaio -direct=1 -invalidate=1 -name=test -bs=4M -iodepth=32 -rw=read -runtime=60 -filename=/dev/sdX`
+- Random write latency (T1Q1, this hurts storages the most):
+  `fio -ioengine=libaio -direct=1 -invalidate=1 -name=test -bs=4k -iodepth=1 -fsync=1 -rw=randwrite -runtime=60 -filename=/dev/sdX`
+- Random read latency (T1Q1):
+  `fio -ioengine=libaio -direct=1 -invalidate=1 -name=test -bs=4k -iodepth=1 -rw=randread -runtime=60 -filename=/dev/sdX`
+- Parallel write iops (use numjobs if a single CPU core is insufficient to saturate the load):
+  `fio -ioengine=libaio -direct=1 -invalidate=1 -name=test -bs=4k -iodepth=128 [-numjobs=4 -group_reporting] -rw=randwrite -runtime=60 -filename=/dev/sdX`
+- Parallel read iops (use numjobs if a single CPU core is insufficient to saturate the load):
+  `fio -ioengine=libaio -direct=1 -invalidate=1 -name=test -bs=4k -iodepth=128 [-numjobs=4 -group_reporting] -rw=randread -runtime=60 -filename=/dev/sdX`
+
+## Vitastor's Theoretical Maximum Random Access Performance
+
+Replicated setups:
+- Single-threaded (T1Q1) read latency: 1 network roundtrip + 1 disk read.
+- Single-threaded write+fsync latency:
+  - With immediate commit: 2 network roundtrips + 1 disk write.
+  - With lazy commit: 4 network roundtrips + 1 disk write + 1 disk flush.
+- Saturated parallel read iops: min(network bandwidth, sum(disk read iops)).
+- Saturated parallel write iops: min(network bandwidth, sum(disk write iops / number of replicas / write amplification)).
+
+EC/XOR setups:
+- Single-threaded (T1Q1) read latency: 1.5 network roundtrips + 1 disk read.
+- Single-threaded write+fsync latency:
+  - With immediate commit: 3.5 network roundtrips + 1 disk read + 2 disk writes.
+  - With lazy commit: 5.5 network roundtrips + 1 disk read + 2 disk writes + 2 disk fsyncs.
+  - 0.5 in actually (k-1)/k which means that an additional roundtrip doesn't happen when
+    the read sub-operation can be served locally.
+- Saturated parallel read iops: min(network bandwidth, sum(disk read iops)).
+- Saturated parallel write iops: min(network bandwidth, sum(disk write iops * number of data drives / (number of data + parity drives) / write amplification)).
+  In fact, you should put disk write iops under the condition of ~10% reads / ~90% writes in this formula.
+
+Write amplification for 4 KB blocks is usually 3-5 in Vitastor:
+1. Journal block write
+2. Journal data write
+3. Metadata block write
+4. Another journal block write for EC/XOR setups
+5. Data block write
+
+If you manage to get an SSD which handles 512 byte blocks well (Optane?) you may
+lower 1, 3 and 4 to 512 bytes (1/8 of data size) and get WA as low as 2.375.
+
+Lazy fsync also reduces WA for parallel workloads because journal blocks are only
+written when they fill up or fsync is requested.
+
+## Example Comparison with Ceph
+
+Hardware configuration: 4 nodes, each with:
+- 6x SATA SSD Intel D3-4510 3.84 TB
+- 2x Xeon Gold 6242 (16 cores @ 2.8 GHz)
+- 384 GB RAM
+- 1x 25 GbE network interface (Mellanox ConnectX-4 LX), connected to a Juniper QFX5200 switch
+
+CPU powersaving was disabled. Both Vitastor and Ceph were configured with 2 OSDs per 1 SSD.
+
+All of the results below apply to 4 KB blocks and random access (unless indicated otherwise).
+
+Raw drive performance:
+- T1Q1 write ~27000 iops (~0.037ms latency)
+- T1Q1 read ~9800 iops (~0.101ms latency)
+- T1Q32 write ~60000 iops
+- T1Q32 read ~81700 iops
+
+Ceph 15.2.4 (Bluestore):
+- T1Q1 write ~1000 iops (~1ms latency)
+- T1Q1 read ~1750 iops (~0.57ms latency)
+- T8Q64 write ~100000 iops, total CPU usage by OSDs about 40 virtual cores on each node
+- T8Q64 read ~480000 iops, total CPU usage by OSDs about 40 virtual cores on each node
+
+T8Q64 tests were conducted over 8 400GB RBD images from all hosts (every host was running 2 instances of fio).
+This is because Ceph has performance penalties related to running multiple clients over a single RBD image.
+
+cephx_sign_messages was set to false during tests, RocksDB and Bluestore settings were left at defaults.
+
+In fact, not that bad for Ceph. These servers are an example of well-balanced Ceph nodes.
+However, CPU usage and I/O latency were through the roof, as usual.
+
+Vitastor:
+- T1Q1 write: 7087 iops (0.14ms latency)
+- T1Q1 read: 6838 iops (0.145ms latency)
+- T2Q64 write: 162000 iops, total CPU usage by OSDs about 3 virtual cores on each node
+- T8Q64 read: 895000 iops, total CPU usage by OSDs about 4 virtual cores on each node
+- Linear write (4M T1Q32): 2800 MB/s
+- Linear read (4M T1Q32): 1500 MB/s
+
+T8Q64 read test was conducted over 1 larger inode (3.2T) from all hosts (every host was running 2 instances of fio).
+Vitastor has no performance penalties related to running multiple clients over a single inode.
+If conducted from one node with all primary OSDs moved to other nodes the result was slightly lower (689000 iops),
+this is because all operations resulted in network roundtrips between the client and the primary OSD.
+When fio was colocated with OSDs (like in Ceph benchmarks above), 1/4 of the read workload actually
+used the loopback network.
+
+Vitastor was configured with: `--disable_data_fsync true --immediate_commit all --flusher_count 8
+  --disk_alignment 4096 --journal_block_size 4096 --meta_block_size 4096
+  --journal_no_same_sector_overwrites true --journal_sector_buffer_count 1024
+  --journal_size 16777216`.
+
+### EC/XOR 2+1
+
+Vitastor:
+- T1Q1 write: 2808 iops (~0.355ms latency)
+- T1Q1 read: 6190 iops (~0.16ms latency)
+- T2Q64 write: 85500 iops, total CPU usage by OSDs about 3.4 virtual cores on each node
+- T8Q64 read: 812000 iops, total CPU usage by OSDs about 4.7 virtual cores on each node
+- Linear write (4M T1Q32): 3200 MB/s
+- Linear read (4M T1Q32): 1800 MB/s
+
+Ceph:
+- T1Q1 write: 730 iops (~1.37ms latency)
+- T1Q1 read: 1500 iops with cold cache (~0.66ms latency), 2300 iops after 2 minute metadata cache warmup (~0.435ms latency)
+- T4Q128 write (4 RBD images): 45300 iops, total CPU usage by OSDs about 30 virtual cores on each node
+- T8Q64 read (4 RBD images): 278600 iops, total CPU usage by OSDs about 40 virtual cores on each node
+- Linear write (4M T1Q32): 1950 MB/s before preallocation, 2500 MB/s after preallocation
+- Linear read (4M T1Q32): 2400 MB/s
+
+### NBD
+
+NBD is currently required to mount Vitastor via kernel, but it imposes additional overhead
+due to additional copying between the kernel and userspace. This mostly hurts linear
+bandwidth, not iops.
+
+Vitastor with single-thread NBD on the same hardware:
+- T1Q1 write: 6000 iops (0.166ms latency)
+- T1Q1 read: 5518 iops (0.18ms latency)
+- T1Q128 write: 94400 iops
+- T1Q128 read: 103000 iops
+- Linear write (4M T1Q128): 1266 MB/s (compared to 2800 MB/s via fio)
+- Linear read (4M T1Q128): 975 MB/s (compared to 1500 MB/s via fio)
+
+## Installation
+
+### Debian
+
+- Trust Vitastor package signing key:
+  `wget -q -O - https://vitastor.io/debian/pubkey | sudo apt-key add -`
+- Add Vitastor package repository to your /etc/apt/sources.list:
+  - Debian 11 (Bullseye/Sid): `deb https://vitastor.io/debian bullseye main`
+  - Debian 10 (Buster): `deb https://vitastor.io/debian buster main`
+- For Debian 10 (Buster) also enable backports repository:
+  `deb http://deb.debian.org/debian buster-backports main`
+- Install packages: `apt update; apt install vitastor lp-solve etcd linux-image-amd64 qemu`
+
+### CentOS
+
+- Add Vitastor package repository:
+  - CentOS 7: `yum install https://vitastor.io/rpms/centos/7/vitastor-release-1.0-1.el7.noarch.rpm`
+  - CentOS 8: `dnf install https://vitastor.io/rpms/centos/8/vitastor-release-1.0-1.el8.noarch.rpm`
+- Enable EPEL: `yum/dnf install epel-release`
+- Enable additional CentOS repositories:
+  - CentOS 7: `yum install centos-release-scl`
+  - CentOS 8: `dnf install centos-release-advanced-virtualization`
+- Enable elrepo-kernel:
+  - CentOS 7: `yum install https://www.elrepo.org/elrepo-release-7.el7.elrepo.noarch.rpm`
+  - CentOS 8: `dnf install https://www.elrepo.org/elrepo-release-8.el8.elrepo.noarch.rpm`
+- Install packages: `yum/dnf install vitastor lpsolve etcd kernel-ml qemu-kvm`
+
+### Building from Source
+
+- Install Linux kernel 5.4 or newer, for io_uring support. 5.8 or later is highly recommended because
+  there is at least one known io_uring hang with 5.4 and an HP SmartArray controller.
+- Install liburing 0.4 or newer and its headers.
+- Install lp_solve.
+- Install etcd. Attention: you need a fixed version from here: https://github.com/vitalif/etcd/,
+  branch release-3.4, because there is a bug in upstream etcd which makes Vitastor OSDs fail to
+  move PGs out of "starting" state if you have at least around ~500 PGs or so. The custom build
+  will be unnecessary when etcd merges the fix: https://github.com/etcd-io/etcd/pull/12402.
+- Install node.js 10 or newer.
+- Install gcc and g++ 8.x or newer.
+- Clone https://yourcmc.ru/git/vitalif/vitastor/ with submodules.
+- Install QEMU 3.0+, get its source, begin to build it, stop the build and copy headers:
+   - `<qemu>/include` &rarr; `<vitastor>/qemu/include`
+   - Debian:
+      * Use qemu packages from the main repository
+      * `<qemu>/b/qemu/config-host.h` &rarr; `<vitastor>/qemu/b/qemu/config-host.h`
+      * `<qemu>/b/qemu/qapi` &rarr; `<vitastor>/qemu/b/qemu/qapi`
+   - CentOS 8:
+      * Use qemu packages from the Advanced-Virtualization repository. To enable it, run
+        `yum install centos-release-advanced-virtualization.noarch` and then `yum install qemu`
+      * `<qemu>/config-host.h` &rarr; `<vitastor>/qemu/b/qemu/config-host.h`
+      * For QEMU 3.0+: `<qemu>/qapi` &rarr; `<vitastor>/qemu/b/qemu/qapi`
+      * For QEMU 2.0+: `<qemu>/qapi-types.h` &rarr; `<vitastor>/qemu/b/qemu/qapi-types.h`
+   - `config-host.h` and `qapi` are required because they contain generated headers
+- You can also rebuild QEMU with a patch that makes LD_PRELOAD unnecessary to load vitastor driver.
+  See `qemu-*.*-vitastor.patch`.
+- Install fio 3.7 or later, get its source and symlink it into `<vitastor>/fio`.
+- Build & install Vitastor with `mkdir build && cd build && cmake .. && make -j8 && make install`.
+  Pay attention to the `QEMU_PLUGINDIR` cmake option - it must be set to `qemu-kvm` on RHEL.
+
+## Running
+
+Please note that startup procedure isn't currently simple - you specify configuration
+and calculate disk offsets almost by hand. This will be fixed in near future.
+
+- Get some SATA or NVMe SSDs with capacitors (server-grade drives). You can use desktop SSDs
+  with lazy fsync, but prepare for inferior single-thread latency.
+- Get a fast network (at least 10 Gbit/s).
+- Disable CPU powersaving: `cpupower idle-set -D 0 && cpupower frequency-set -g performance`.
+- Check `/usr/lib/vitastor/mon/make-units.sh` and `/usr/lib/vitastor/mon/make-osd.sh` and
+  put desired values into the variables at the top of these files.
+- Create systemd units for the monitor and etcd: `/usr/lib/vitastor/mon/make-units.sh`
+- Create systemd units for your OSDs: `/usr/lib/vitastor/mon/make-osd.sh /dev/disk/by-partuuid/XXX [/dev/disk/by-partuuid/YYY ...]`
+- You can edit the units and change OSD configuration. Notable configuration variables:
+  - `disable_data_fsync 1` - only safe with server-grade drives with capacitors.
+  - `immediate_commit all` - use this if all your drives are server-grade.
+  - `disable_device_lock 1` - only required if you run multiple OSDs on one block device.
+  - `flusher_count 256` - flusher is a micro-thread that removes old data from the journal.
+    You don't have to worry about this parameter anymore, 256 is enough.
+  - `disk_alignment`, `journal_block_size`, `meta_block_size` should be set to the internal
+    block size of your SSDs which is 4096 on most drives.
+  - `journal_no_same_sector_overwrites true` prevents multiple overwrites of the same journal sector.
+    Most (99%) SSDs don't need this option. But Intel D3-4510 does because it doesn't like when you
+    overwrite the same sector twice in a short period of time. The setting forces Vitastor to never
+    overwrite the same journal sector twice in a row which makes D3-4510 almost happy. Not totally
+    happy, because overwrites of the same block can still happen in the metadata area... When this
+    setting is set, it is also required to raise `journal_sector_buffer_count` setting, which is the
+    number of dirty journal sectors that may be written to at the same time.
+- `systemctl start vitastor.target` everywhere.
+- Create global configuration in etcd: `etcdctl --endpoints=... put /vitastor/config/global '{"immediate_commit":"all"}'`
+  (if all your drives have capacitors).
+- Create pool configuration in etcd: `etcdctl --endpoints=... put /vitastor/config/pools '{"1":{"name":"testpool","scheme":"replicated","pg_size":2,"pg_minsize":1,"pg_count":256,"failure_domain":"host"}}'`.
+  For jerasure pools the configuration should look like the following: `2:{"name":"ecpool","scheme":"jerasure","pg_size":4,"parity_chunks":2,"pg_minsize":2,"pg_count":256,"failure_domain":"host"}`.
+- At this point, one of the monitors will configure PGs and OSDs will start them.
+- You can check PG states with `etcdctl --endpoints=... get --prefix /vitastor/pg/state`. All PGs should become 'active'.
+- Run tests with (for example): `fio -thread -ioengine=libfio_vitastor.so -name=test -bs=4M -direct=1 -iodepth=16 -rw=write -etcd=10.115.0.10:2379/v3 -pool=1 -inode=1 -size=400G`.
+- Upload VM disk image with qemu-img (for example):
+  ```
+  qemu-img convert -f qcow2 debian10.qcow2 -p -O raw 'vitastor:etcd_host=10.115.0.10\:2379/v3:pool=1:inode=1:size=2147483648'
+  ```
+  Note that the command requires to be run with `LD_PRELOAD=/usr/lib/x86_64-linux-gnu/qemu/block-vitastor.so qemu-img ...`
+  if you use unmodified QEMU.
+- Run QEMU with (for example):
+  ```
+  qemu-system-x86_64 -enable-kvm -m 1024
+    -drive 'file=vitastor:etcd_host=10.115.0.10\:2379/v3:pool=1:inode=1:size=2147483648',format=raw,if=none,id=drive-virtio-disk0,cache=none
+    -device virtio-blk-pci,scsi=off,bus=pci.0,addr=0x5,drive=drive-virtio-disk0,id=virtio-disk0,bootindex=1,write-cache=off,physical_block_size=4096,logical_block_size=512
+    -vnc 0.0.0.0:0
+  ```
+- Remove inode with (for example):
+  ```
+  vitastor-rm --etcd_address 10.115.0.10:2379/v3 --pool 1 --inode 1 --parallel_osds 16 --iodepth 32
+  ```
+
+## Known Problems
+
+- Object deletion requests may currently lead to 'incomplete' objects in EC pools
+  if your OSDs crash during deletion because proper handling of object cleanup
+  in a cluster should be "three-phase" and it's currently not implemented.
+  Just repeat the removal request again in this case.
+
+## Implementation Principles
+
+- I like architecturally simple solutions. Vitastor is and will always be designed
+  exactly like that.
+- I also like reinventing the wheel to some extent, like writing my own HTTP client
+  for etcd interaction instead of using prebuilt libraries, because in this case
+  I'm confident about what my code does and what it doesn't do.
+- I don't care about C++ "best practices" like RAII or proper inheritance or usage of
+  smart pointers or whatever and I don't intend to change my mind, so if you're here
+  looking for ideal reference C++ code, this probably isn't the right place.
+- I like node.js better than any other dynamically-typed language interpreter
+  because it's faster than any other interpreter in the world, has neutral C-like
+  syntax and built-in event loop. That's why Monitor is implemented in node.js.
+
+## Author and License
+
+Copyright (c) Vitaliy Filippov (vitalif [at] yourcmc.ru), 2019+
+
+Join Vitastor Telegram Chat: https://t.me/vitastor
+
+All server-side code (OSD, Monitor and so on) is licensed under the terms of
+Vitastor Network Public License 1.1 (VNPL 1.1), a copyleft license based on
+GNU GPLv3.0 with the additional "Network Interaction" clause which requires
+opensourcing all programs directly or indirectly interacting with Vitastor
+through a computer network and expressly designed to be used in conjunction
+with it ("Proxy Programs"). Proxy Programs may be made public not only under
+the terms of the same license, but also under the terms of any GPL-Compatible
+Free Software License, as listed by the Free Software Foundation.
+This is a stricter copyleft license than the Affero GPL.
+
+Please note that VNPL doesn't require you to open the code of proprietary
+software running inside a VM if it's not specially designed to be used with
+Vitastor.
+
+Basically, you can't use the software in a proprietary environment to provide
+its functionality to users without opensourcing all intermediary components
+standing between the user and Vitastor or purchasing a commercial license
+from the author 😀.
+
+Client libraries (cluster_client and so on) are dual-licensed under the same
+VNPL 1.1 and also GNU GPL 2.0 or later to allow for compatibility with GPLed
+software like QEMU and fio.
+
+You can find the full text of VNPL-1.1 in the file [VNPL-1.1.txt](VNPL-1.1.txt).
+GPL 2.0 is also included in this repository as [GPL-2.0.txt](GPL-2.0.txt).

VNPL-1.1.txt

@@ -0,0 +1,648 @@
+                     VITASTOR NETWORK PUBLIC LICENSE
+                     Version 1.1,  6 February 2021
+
+ Copyright (C) 2021 Vitaliy Filippov <vitalif@yourcmc.ru>
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+                            Preamble
+
+  The Vitastor Network Public License is a free, copyleft license for
+software and other kinds of works, specifically designed to ensure
+cooperation with the community in the case of network server software.
+
+  The licenses for most software and other practical works are designed
+to take away your freedom to share and change the works.  By contrast,
+GNU General Public Licenses and Vitastor Network Public License are
+intended to guarantee your freedom to share and change all versions
+of a program--to make sure it remains free software for all its users.
+
+  When we speak of free software, we are referring to freedom, not
+price.  GNU General Public Licenses and Vitastor Network Public License
+are designed to make sure that you have the freedom to distribute copies
+of free software (and charge for them if you wish), that you receive
+source code or can get it if you want it, that you can change the software
+or use pieces of it in new free programs, and that you know you can do these
+things.
+
+  Developers that use GNU General Public Licenses and Vitastor
+Network Public License protect your rights with two steps:
+(1) assert copyright on the software, and (2) offer
+you this License which gives you legal permission to copy, distribute
+and/or modify the software.
+
+  A secondary benefit of defending all users' freedom is that
+improvements made in alternate versions of the program, if they
+receive widespread use, become available for other developers to
+incorporate.  Many developers of free software are heartened and
+encouraged by the resulting cooperation.  However, in the case of
+software used on network servers, this result may fail to come about.
+The GNU General Public License permits making a modified version and
+letting the public access it on a server without ever releasing its
+source code to the public. Even the GNU Affero General Public License
+permits running a modified version in a closed environment where
+public users only interact with it through a closed-source proxy, again,
+without making the program and the proxy available to the public
+for free.
+
+  The Vitastor Network Public License is designed specifically to
+ensure that, in such cases, the modified program and the proxy stays
+available to the community. It requires the operator of a network server to
+provide the source code of the original program and all other programs
+communicating with it running there to the users of that server.
+Therefore, public use of a modified version, on a server accessible
+directly or indirectly to the public, gives the public access to the source
+code of the modified version.
+
+  The precise terms and conditions for copying, distribution and
+modification follow.
+
+                       TERMS AND CONDITIONS
+
+  0. Definitions.
+
+  "This License" refers to version 1 of the Vitastor Network Public License.
+
+  "Copyright" also means copyright-like laws that apply to other kinds of
+works, such as semiconductor masks.
+
+  "The Program" refers to any copyrightable work licensed under this
+License.  Each licensee is addressed as "you".  "Licensees" and
+"recipients" may be individuals or organizations.
+
+  To "modify" a work means to copy from or adapt all or part of the work
+in a fashion requiring copyright permission, other than the making of an
+exact copy.  The resulting work is called a "modified version" of the
+earlier work or a work "based on" the earlier work.
+
+  A "covered work" means either the unmodified Program or a work based
+on the Program.
+
+  To "propagate" a work means to do anything with it that, without
+permission, would make you directly or secondarily liable for
+infringement under applicable copyright law, except executing it on a
+computer or modifying a private copy.  Propagation includes copying,
+distribution (with or without modification), making available to the
+public, and in some countries other activities as well.
+
+  To "convey" a work means any kind of propagation that enables other
+parties to make or receive copies.  Mere interaction with a user through
+a computer network, with no transfer of a copy, is not conveying.
+
+  An interactive user interface displays "Appropriate Legal Notices"
+to the extent that it includes a convenient and prominently visible
+feature that (1) displays an appropriate copyright notice, and (2)
+tells the user that there is no warranty for the work (except to the
+extent that warranties are provided), that licensees may convey the
+work under this License, and how to view a copy of this License.  If
+the interface presents a list of user commands or options, such as a
+menu, a prominent item in the list meets this criterion.
+
+  1. Source Code.
+
+  The "source code" for a work means the preferred form of the work
+for making modifications to it.  "Object code" means any non-source
+form of a work.
+
+  A "Standard Interface" means an interface that either is an official
+standard defined by a recognized standards body, or, in the case of
+interfaces specified for a particular programming language, one that
+is widely used among developers working in that language.
+
+  The "System Libraries" of an executable work include anything, other
+than the work as a whole, that (a) is included in the normal form of
+packaging a Major Component, but which is not part of that Major
+Component, and (b) serves only to enable use of the work with that
+Major Component, or to implement a Standard Interface for which an
+implementation is available to the public in source code form.  A
+"Major Component", in this context, means a major essential component
+(kernel, window system, and so on) of the specific operating system
+(if any) on which the executable work runs, or a compiler used to
+produce the work, or an object code interpreter used to run it.
+
+  The "Corresponding Source" for a work in object code form means all
+the source code needed to generate, install, and (for an executable
+work) run the object code and to modify the work, including scripts to
+control those activities.  However, it does not include the work's
+System Libraries, or general-purpose tools or generally available free
+programs which are used unmodified in performing those activities but
+which are not part of the work.  For example, Corresponding Source
+includes interface definition files associated with source files for
+the work, and the source code for shared libraries and dynamically
+linked subprograms that the work is specifically designed to require,
+such as by intimate data communication or control flow between those
+subprograms and other parts of the work.
+
+  The Corresponding Source need not include anything that users
+can regenerate automatically from other parts of the Corresponding
+Source.
+
+  The Corresponding Source for a work in source code form is that
+same work.
+
+  2. Basic Permissions.
+
+  All rights granted under this License are granted for the term of
+copyright on the Program, and are irrevocable provided the stated
+conditions are met.  This License explicitly affirms your unlimited
+permission to run the unmodified Program.  The output from running a
+covered work is covered by this License only if the output, given its
+content, constitutes a covered work.  This License acknowledges your
+rights of fair use or other equivalent, as provided by copyright law.
+
+  You may make, run and propagate covered works that you do not
+convey, without conditions so long as your license otherwise remains
+in force.  You may convey covered works to others for the sole purpose
+of having them make modifications exclusively for you, or provide you
+with facilities for running those works, provided that you comply with
+the terms of this License in conveying all material for which you do
+not control copyright.  Those thus making or running the covered works
+for you must do so exclusively on your behalf, under your direction
+and control, on terms that prohibit them from making any copies of
+your copyrighted material outside their relationship with you.
+
+  Conveying under any other circumstances is permitted solely under
+the conditions stated below.  Sublicensing is not allowed; section 10
+makes it unnecessary.
+
+  3. Protecting Users' Legal Rights From Anti-Circumvention Law.
+
+  No covered work shall be deemed part of an effective technological
+measure under any applicable law fulfilling obligations under article
+11 of the WIPO copyright treaty adopted on 20 December 1996, or
+similar laws prohibiting or restricting circumvention of such
+measures.
+
+  When you convey a covered work, you waive any legal power to forbid
+circumvention of technological measures to the extent such circumvention
+is effected by exercising rights under this License with respect to
+the covered work, and you disclaim any intention to limit operation or
+modification of the work as a means of enforcing, against the work's
+users, your or third parties' legal rights to forbid circumvention of
+technological measures.
+
+  4. Conveying Verbatim Copies.
+
+  You may convey verbatim copies of the Program's source code as you
+receive it, in any medium, provided that you conspicuously and
+appropriately publish on each copy an appropriate copyright notice;
+keep intact all notices stating that this License and any
+non-permissive terms added in accord with section 7 apply to the code;
+keep intact all notices of the absence of any warranty; and give all
+recipients a copy of this License along with the Program.
+
+  You may charge any price or no price for each copy that you convey,
+and you may offer support or warranty protection for a fee.
+
+  5. Conveying Modified Source Versions.
+
+  You may convey a work based on the Program, or the modifications to
+produce it from the Program, in the form of source code under the
+terms of section 4, provided that you also meet all of these conditions:
+
+    a) The work must carry prominent notices stating that you modified
+    it, and giving a relevant date.
+
+    b) The work must carry prominent notices stating that it is
+    released under this License and any conditions added under section
+    7.  This requirement modifies the requirement in section 4 to
+    "keep intact all notices".
+
+    c) You must license the entire work, as a whole, under this
+    License to anyone who comes into possession of a copy.  This
+    License will therefore apply, along with any applicable section 7
+    additional terms, to the whole of the work, and all its parts,
+    regardless of how they are packaged.  This License gives no
+    permission to license the work in any other way, but it does not
+    invalidate such permission if you have separately received it.
+
+    d) If the work has interactive user interfaces, each must display
+    Appropriate Legal Notices; however, if the Program has interactive
+    interfaces that do not display Appropriate Legal Notices, your
+    work need not make them do so.
+
+  A compilation of a covered work with other separate and independent
+works, which are not by their nature extensions of the covered work,
+and which are not combined with it such as to form a larger program,
+in or on a volume of a storage or distribution medium, is called an
+"aggregate" if the compilation and its resulting copyright are not
+used to limit the access or legal rights of the compilation's users
+beyond what the individual works permit.  Inclusion of a covered work
+in an aggregate does not cause this License to apply to the other
+parts of the aggregate.
+
+  6. Conveying Non-Source Forms.
+
+  You may convey a covered work in object code form under the terms
+of sections 4 and 5, provided that you also convey the
+machine-readable Corresponding Source under the terms of this License,
+in one of these ways:
+
+    a) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by the
+    Corresponding Source fixed on a durable physical medium
+    customarily used for software interchange.
+
+    b) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by a
+    written offer, valid for at least three years and valid for as
+    long as you offer spare parts or customer support for that product
+    model, to give anyone who possesses the object code either (1) a
+    copy of the Corresponding Source for all the software in the
+    product that is covered by this License, on a durable physical
+    medium customarily used for software interchange, for a price no
+    more than your reasonable cost of physically performing this
+    conveying of source, or (2) access to copy the
+    Corresponding Source from a network server at no charge.
+
+    c) Convey individual copies of the object code with a copy of the
+    written offer to provide the Corresponding Source.  This
+    alternative is allowed only occasionally and noncommercially, and
+    only if you received the object code with such an offer, in accord
+    with subsection 6b.
+
+    d) Convey the object code by offering access from a designated
+    place (gratis or for a charge), and offer equivalent access to the
+    Corresponding Source in the same way through the same place at no
+    further charge.  You need not require recipients to copy the
+    Corresponding Source along with the object code.  If the place to
+    copy the object code is a network server, the Corresponding Source
+    may be on a different server (operated by you or a third party)
+    that supports equivalent copying facilities, provided you maintain
+    clear directions next to the object code saying where to find the
+    Corresponding Source.  Regardless of what server hosts the
+    Corresponding Source, you remain obligated to ensure that it is
+    available for as long as needed to satisfy these requirements.
+
+    e) Convey the object code using peer-to-peer transmission, provided
+    you inform other peers where the object code and Corresponding
+    Source of the work are being offered to the general public at no
+    charge under subsection 6d.
+
+  A separable portion of the object code, whose source code is excluded
+from the Corresponding Source as a System Library, need not be
+included in conveying the object code work.
+
+  A "User Product" is either (1) a "consumer product", which means any
+tangible personal property which is normally used for personal, family,
+or household purposes, or (2) anything designed or sold for incorporation
+into a dwelling.  In determining whether a product is a consumer product,
+doubtful cases shall be resolved in favor of coverage.  For a particular
+product received by a particular user, "normally used" refers to a
+typical or common use of that class of product, regardless of the status
+of the particular user or of the way in which the particular user
+actually uses, or expects or is expected to use, the product.  A product
+is a consumer product regardless of whether the product has substantial
+commercial, industrial or non-consumer uses, unless such uses represent
+the only significant mode of use of the product.
+
+  "Installation Information" for a User Product means any methods,
+procedures, authorization keys, or other information required to install
+and execute modified versions of a covered work in that User Product from
+a modified version of its Corresponding Source.  The information must
+suffice to ensure that the continued functioning of the modified object
+code is in no case prevented or interfered with solely because
+modification has been made.
+
+  If you convey an object code work under this section in, or with, or
+specifically for use in, a User Product, and the conveying occurs as
+part of a transaction in which the right of possession and use of the
+User Product is transferred to the recipient in perpetuity or for a
+fixed term (regardless of how the transaction is characterized), the
+Corresponding Source conveyed under this section must be accompanied
+by the Installation Information.  But this requirement does not apply
+if neither you nor any third party retains the ability to install
+modified object code on the User Product (for example, the work has
+been installed in ROM).
+
+  The requirement to provide Installation Information does not include a
+requirement to continue to provide support service, warranty, or updates
+for a work that has been modified or installed by the recipient, or for
+the User Product in which it has been modified or installed.  Access to a
+network may be denied when the modification itself materially and
+adversely affects the operation of the network or violates the rules and
+protocols for communication across the network.
+
+  Corresponding Source conveyed, and Installation Information provided,
+in accord with this section must be in a format that is publicly
+documented (and with an implementation available to the public in
+source code form), and must require no special password or key for
+unpacking, reading or copying.
+
+  7. Additional Terms.
+
+  "Additional permissions" are terms that supplement the terms of this
+License by making exceptions from one or more of its conditions.
+Additional permissions that are applicable to the entire Program shall
+be treated as though they were included in this License, to the extent
+that they are valid under applicable law.  If additional permissions
+apply only to part of the Program, that part may be used separately
+under those permissions, but the entire Program remains governed by
+this License without regard to the additional permissions.
+
+  When you convey a copy of a covered work, you may at your option
+remove any additional permissions from that copy, or from any part of
+it.  (Additional permissions may be written to require their own
+removal in certain cases when you modify the work.)  You may place
+additional permissions on material, added by you to a covered work,
+for which you have or can give appropriate copyright permission.
+
+  Notwithstanding any other provision of this License, for material you
+add to a covered work, you may (if authorized by the copyright holders of
+that material) supplement the terms of this License with terms:
+
+    a) Disclaiming warranty or limiting liability differently from the
+    terms of sections 15 and 16 of this License; or
+
+    b) Requiring preservation of specified reasonable legal notices or
+    author attributions in that material or in the Appropriate Legal
+    Notices displayed by works containing it; or
+
+    c) Prohibiting misrepresentation of the origin of that material, or
+    requiring that modified versions of such material be marked in
+    reasonable ways as different from the original version; or
+
+    d) Limiting the use for publicity purposes of names of licensors or
+    authors of the material; or
+
+    e) Declining to grant rights under trademark law for use of some
+    trade names, trademarks, or service marks; or
+
+    f) Requiring indemnification of licensors and authors of that
+    material by anyone who conveys the material (or modified versions of
+    it) with contractual assumptions of liability to the recipient, for
+    any liability that these contractual assumptions directly impose on
+    those licensors and authors.
+
+  All other non-permissive additional terms are considered "further
+restrictions" within the meaning of section 10.  If the Program as you
+received it, or any part of it, contains a notice stating that it is
+governed by this License along with a term that is a further
+restriction, you may remove that term.  If a license document contains
+a further restriction but permits relicensing or conveying under this
+License, you may add to a covered work material governed by the terms
+of that license document, provided that the further restriction does
+not survive such relicensing or conveying.
+
+  If you add terms to a covered work in accord with this section, you
+must place, in the relevant source files, a statement of the
+additional terms that apply to those files, or a notice indicating
+where to find the applicable terms.
+
+  Additional terms, permissive or non-permissive, may be stated in the
+form of a separately written license, or stated as exceptions;
+the above requirements apply either way.
+
+  8. Termination.
+
+  You may not propagate or modify a covered work except as expressly
+provided under this License.  Any attempt otherwise to propagate or
+modify it is void, and will automatically terminate your rights under
+this License (including any patent licenses granted under the third
+paragraph of section 11).
+
+  However, if you cease all violation of this License, then your
+license from a particular copyright holder is reinstated (a)
+provisionally, unless and until the copyright holder explicitly and
+finally terminates your license, and (b) permanently, if the copyright
+holder fails to notify you of the violation by some reasonable means
+prior to 60 days after the cessation.
+
+  Moreover, your license from a particular copyright holder is
+reinstated permanently if the copyright holder notifies you of the
+violation by some reasonable means, this is the first time you have
+received notice of violation of this License (for any work) from that
+copyright holder, and you cure the violation prior to 30 days after
+your receipt of the notice.
+
+  Termination of your rights under this section does not terminate the
+licenses of parties who have received copies or rights from you under
+this License.  If your rights have been terminated and not permanently
+reinstated, you do not qualify to receive new licenses for the same
+material under section 10.
+
+  9. Acceptance Not Required for Having Copies.
+
+  You are not required to accept this License in order to receive or
+run a copy of the Program.  Ancillary propagation of a covered work
+occurring solely as a consequence of using peer-to-peer transmission
+to receive a copy likewise does not require acceptance.  However,
+nothing other than this License grants you permission to propagate or
+modify any covered work.  These actions infringe copyright if you do
+not accept this License.  Therefore, by modifying or propagating a
+covered work, you indicate your acceptance of this License to do so.
+
+  10. Automatic Licensing of Downstream Recipients.
+
+  Each time you convey a covered work, the recipient automatically
+receives a license from the original licensors, to run, modify and
+propagate that work, subject to this License.  You are not responsible
+for enforcing compliance by third parties with this License.
+
+  An "entity transaction" is a transaction transferring control of an
+organization, or substantially all assets of one, or subdividing an
+organization, or merging organizations.  If propagation of a covered
+work results from an entity transaction, each party to that
+transaction who receives a copy of the work also receives whatever
+licenses to the work the party's predecessor in interest had or could
+give under the previous paragraph, plus a right to possession of the
+Corresponding Source of the work from the predecessor in interest, if
+the predecessor has it or can get it with reasonable efforts.
+
+  You may not impose any further restrictions on the exercise of the
+rights granted or affirmed under this License.  For example, you may
+not impose a license fee, royalty, or other charge for exercise of
+rights granted under this License, and you may not initiate litigation
+(including a cross-claim or counterclaim in a lawsuit) alleging that
+any patent claim is infringed by making, using, selling, offering for
+sale, or importing the Program or any portion of it.
+
+  11. Patents.
+
+  A "contributor" is a copyright holder who authorizes use under this
+License of the Program or a work on which the Program is based.  The
+work thus licensed is called the contributor's "contributor version".
+
+  A contributor's "essential patent claims" are all patent claims
+owned or controlled by the contributor, whether already acquired or
+hereafter acquired, that would be infringed by some manner, permitted
+by this License, of making, using, or selling its contributor version,
+but do not include claims that would be infringed only as a
+consequence of further modification of the contributor version.  For
+purposes of this definition, "control" includes the right to grant
+patent sublicenses in a manner consistent with the requirements of
+this License.
+
+  Each contributor grants you a non-exclusive, worldwide, royalty-free
+patent license under the contributor's essential patent claims, to
+make, use, sell, offer for sale, import and otherwise run, modify and
+propagate the contents of its contributor version.
+
+  In the following three paragraphs, a "patent license" is any express
+agreement or commitment, however denominated, not to enforce a patent
+(such as an express permission to practice a patent or covenant not to
+sue for patent infringement).  To "grant" such a patent license to a
+party means to make such an agreement or commitment not to enforce a
+patent against the party.
+
+  If you convey a covered work, knowingly relying on a patent license,
+and the Corresponding Source of the work is not available for anyone
+to copy, free of charge and under the terms of this License, through a
+publicly available network server or other readily accessible means,
+then you must either (1) cause the Corresponding Source to be so
+available, or (2) arrange to deprive yourself of the benefit of the
+patent license for this particular work, or (3) arrange, in a manner
+consistent with the requirements of this License, to extend the patent
+license to downstream recipients.  "Knowingly relying" means you have
+actual knowledge that, but for the patent license, your conveying the
+covered work in a country, or your recipient's use of the covered work
+in a country, would infringe one or more identifiable patents in that
+country that you have reason to believe are valid.
+
+  If, pursuant to or in connection with a single transaction or
+arrangement, you convey, or propagate by procuring conveyance of, a
+covered work, and grant a patent license to some of the parties
+receiving the covered work authorizing them to use, propagate, modify
+or convey a specific copy of the covered work, then the patent license
+you grant is automatically extended to all recipients of the covered
+work and works based on it.
+
+  A patent license is "discriminatory" if it does not include within
+the scope of its coverage, prohibits the exercise of, or is
+conditioned on the non-exercise of one or more of the rights that are
+specifically granted under this License.  You may not convey a covered
+work if you are a party to an arrangement with a third party that is
+in the business of distributing software, under which you make payment
+to the third party based on the extent of your activity of conveying
+the work, and under which the third party grants, to any of the
+parties who would receive the covered work from you, a discriminatory
+patent license (a) in connection with copies of the covered work
+conveyed by you (or copies made from those copies), or (b) primarily
+for and in connection with specific products or compilations that
+contain the covered work, unless you entered into that arrangement,
+or that patent license was granted, prior to 28 March 2007.
+
+  Nothing in this License shall be construed as excluding or limiting
+any implied license or other defenses to infringement that may
+otherwise be available to you under applicable patent law.
+
+  12. No Surrender of Others' Freedom.
+
+  If conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License.  If you cannot convey a
+covered work so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you may
+not convey it at all.  For example, if you agree to terms that obligate you
+to collect a royalty for further conveying from those to whom you convey
+the Program, the only way you could satisfy both those terms and this
+License would be to refrain entirely from conveying the Program.
+
+  13. Remote Network Interaction.
+
+  A "Proxy Program" means a separate program which is specially designed to
+be used in conjunction with the covered work and interacts with it directly
+or indirectly through any kind of API (application programming interfaces),
+a computer network, an imitation of such network, or another Proxy Program
+itself.
+
+  Notwithstanding any other provision of this License, if you provide any user
+with an opportunity to interact with the covered work through a computer
+network, an imitation of such network, or any number of "Proxy Programs",
+you must prominently offer that user an opportunity to receive the
+Corresponding Source of the covered work and all Proxy Programs from a
+network server at no charge, through some standard or customary means of
+facilitating copying of software. The Corresponding Source for the covered
+work must be made available under the conditions of this License, and
+the Corresponding Source for all Proxy Programs must be made available
+under the conditions of either this License or any GPL-Compatible
+Free Software License, as described by the Free Software Foundation
+in their "GPL-Compatible License List".
+
+  14. Revised Versions of this License.
+
+  Vitastor Author may publish revised and/or new versions of
+the Vitastor Network Public License from time to time.  Such new versions
+will be similar in spirit to the present version, but may differ in detail to
+address new problems or concerns.
+
+  Each version is given a distinguishing version number.  If the
+Program specifies that a certain numbered version of the Vitastor Network
+Public License "or any later version" applies to it, you have the
+option of following the terms and conditions either of that numbered
+version or of any later version. If the Program does not specify a version
+number of the Vitastor Network Public License, you may choose any version
+ever published.
+
+  Later license versions may give you additional or different
+permissions.  However, no additional obligations are imposed on any
+author or copyright holder as a result of your choosing to follow a
+later version.
+
+  15. Disclaimer of Warranty.
+
+  THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
+APPLICABLE LAW.  EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
+HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
+OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
+THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
+IS WITH YOU.  SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
+ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
+
+  16. Limitation of Liability.
+
+  IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
+WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
+THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
+GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
+USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
+DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
+PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
+EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
+SUCH DAMAGES.
+
+  17. Interpretation of Sections 15 and 16.
+
+  If the disclaimer of warranty and limitation of liability provided
+above cannot be given local legal effect according to their terms,
+reviewing courts shall apply local law that most closely approximates
+an absolute waiver of all civil liability in connection with the
+Program, unless a warranty or assumption of liability accompanies a
+copy of the Program in return for a fee.
+
+                     END OF TERMS AND CONDITIONS
+
+            How to Apply These Terms to Your New Programs
+
+  If you develop a new program, and you want it to be of the greatest
+possible use to the public, the best way to achieve this is to make it
+free software which everyone can redistribute and change under these terms.
+
+  To do so, attach the following notices to the program.  It is safest
+to attach them to the start of each source file to most effectively
+state the exclusion of warranty; and each file should have at least
+the "copyright" line and a pointer to where the full notice is found.
+
+    <one line to give the program's name and a brief idea of what it does.>
+    Copyright (C) <year>  <name of author>
+
+    This program is free software: you can redistribute it and/or modify
+    it under the terms of the Vitastor Network Public License as published by
+    the Vitastor Author, either version 1 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    Vitastor Network Public License for more details.
+
+Also add information on how to contact you by electronic and paper mail.
+
+  If your software can interact with users remotely through a computer
+network, you should also make sure that it provides a way for users to
+get its source.  For example, if your program is a web application, its
+interface could display a "Source" link that leads users to an archive
+of the code.  There are many ways you could offer source, and different
+solutions will be better for different programs; see section 13 for the
+specific requirements.

blockstore_impl.cpp

@@ -1,458 +0,0 @@
-#include "blockstore_impl.h"
-
-blockstore_impl_t::blockstore_impl_t(blockstore_config_t & config, ring_loop_t *ringloop)
-{
-    assert(sizeof(blockstore_op_private_t) <= BS_OP_PRIVATE_DATA_SIZE);
-    this->ringloop = ringloop;
-    ring_consumer.loop = [this]() { loop(); };
-    ringloop->register_consumer(ring_consumer);
-    initialized = 0;
-    zero_object = (uint8_t*)memalign(MEM_ALIGNMENT, block_size);
-    data_fd = meta_fd = journal.fd = -1;
-    parse_config(config);
-    try
-    {
-        open_data();
-        open_meta();
-        open_journal();
-        calc_lengths();
-        data_alloc = new allocator(block_count);
-    }
-    catch (std::exception & e)
-    {
-        if (data_fd >= 0)
-            close(data_fd);
-        if (meta_fd >= 0 && meta_fd != data_fd)
-            close(meta_fd);
-        if (journal.fd >= 0 && journal.fd != meta_fd)
-            close(journal.fd);
-        throw;
-    }
-    flusher = new journal_flusher_t(flusher_count, this);
-}
-
-blockstore_impl_t::~blockstore_impl_t()
-{
-    delete data_alloc;
-    delete flusher;
-    free(zero_object);
-    ringloop->unregister_consumer(ring_consumer);
-    if (data_fd >= 0)
-        close(data_fd);
-    if (meta_fd >= 0 && meta_fd != data_fd)
-        close(meta_fd);
-    if (journal.fd >= 0 && journal.fd != meta_fd)
-        close(journal.fd);
-    if (metadata_buffer)
-        free(metadata_buffer);
-    if (clean_bitmap)
-        free(clean_bitmap);
-}
-
-bool blockstore_impl_t::is_started()
-{
-    return initialized == 10;
-}
-
-bool blockstore_impl_t::is_stalled()
-{
-    return queue_stall;
-}
-
-// main event loop - produce requests
-void blockstore_impl_t::loop()
-{
-    // FIXME: initialized == 10 is ugly
-    if (initialized != 10)
-    {
-        // read metadata, then journal
-        if (initialized == 0)
-        {
-            metadata_init_reader = new blockstore_init_meta(this);
-            initialized = 1;
-        }
-        if (initialized == 1)
-        {
-            int res = metadata_init_reader->loop();
-            if (!res)
-            {
-                delete metadata_init_reader;
-                metadata_init_reader = NULL;
-                journal_init_reader = new blockstore_init_journal(this);
-                initialized = 2;
-            }
-        }
-        if (initialized == 2)
-        {
-            int res = journal_init_reader->loop();
-            if (!res)
-            {
-                delete journal_init_reader;
-                journal_init_reader = NULL;
-                initialized = 10;
-                ringloop->wakeup();
-            }
-        }
-    }
-    else
-    {
-        // try to submit ops
-        unsigned initial_ring_space = ringloop->space_left();
-        auto cur_sync = in_progress_syncs.begin();
-        while (cur_sync != in_progress_syncs.end())
-        {
-            continue_sync(*cur_sync++);
-        }
-        auto cur = submit_queue.begin();
-        int has_writes = 0;
-        while (cur != submit_queue.end())
-        {
-            auto op_ptr = cur;
-            auto op = *(cur++);
-            // FIXME: This needs some simplification
-            // Writes should not block reads if the ring is not full and reads don't depend on them
-            // In all other cases we should stop submission
-            if (PRIV(op)->wait_for)
-            {
-                check_wait(op);
-#ifdef BLOCKSTORE_DEBUG
-                if (PRIV(op)->wait_for)
-                {
-                    printf("still waiting for %d\n", PRIV(op)->wait_for);
-                }
-#endif
-                if (PRIV(op)->wait_for == WAIT_SQE)
-                {
-                    break;
-                }
-                else if (PRIV(op)->wait_for)
-                {
-                    if (op->opcode == BS_OP_WRITE || op->opcode == BS_OP_DELETE)
-                    {
-                        has_writes = 2;
-                    }
-                    continue;
-                }
-            }
-            unsigned ring_space = ringloop->space_left();
-            unsigned prev_sqe_pos = ringloop->save();
-            int dequeue_op = 0;
-            if (op->opcode == BS_OP_READ)
-            {
-                dequeue_op = dequeue_read(op);
-            }
-            else if (op->opcode == BS_OP_WRITE || op->opcode == BS_OP_DELETE)
-            {
-                if (has_writes == 2)
-                {
-                    // Some writes could not be submitted
-                    break;
-                }
-                dequeue_op = dequeue_write(op);
-                has_writes = dequeue_op ? 1 : 2;
-            }
-            else if (op->opcode == BS_OP_SYNC)
-            {
-                // wait for all small writes to be submitted
-                // wait for all big writes to complete, submit data device fsync
-                // wait for the data device fsync to complete, then submit journal writes for big writes
-                // then submit an fsync operation
-                if (has_writes)
-                {
-                    // Can't submit SYNC before previous writes
-                    continue;
-                }
-                dequeue_op = dequeue_sync(op);
-            }
-            else if (op->opcode == BS_OP_STABLE)
-            {
-                dequeue_op = dequeue_stable(op);
-            }
-            else if (op->opcode == BS_OP_ROLLBACK)
-            {
-                dequeue_op = dequeue_rollback(op);
-            }
-            else if (op->opcode == BS_OP_LIST)
-            {
-                process_list(op);
-                dequeue_op = true;
-            }
-            if (dequeue_op)
-            {
-                submit_queue.erase(op_ptr);
-            }
-            else
-            {
-                ringloop->restore(prev_sqe_pos);
-                if (PRIV(op)->wait_for == WAIT_SQE)
-                {
-                    PRIV(op)->wait_detail = 1 + ring_space;
-                    // ring is full, stop submission
-                    break;
-                }
-            }
-        }
-        if (!readonly)
-        {
-            flusher->loop();
-        }
-        int ret = ringloop->submit();
-        if (ret < 0)
-        {
-            throw std::runtime_error(std::string("io_uring_submit: ") + strerror(-ret));
-        }
-        if ((initial_ring_space - ringloop->space_left()) > 0)
-        {
-            live = true;
-        }
-        queue_stall = !live && !ringloop->get_loop_again();
-        live = false;
-    }
-}
-
-bool blockstore_impl_t::is_safe_to_stop()
-{
-    // It's safe to stop blockstore when there are no in-flight operations,
-    // no in-progress syncs and flusher isn't doing anything
-    if (submit_queue.size() > 0 || in_progress_syncs.size() > 0 || !readonly && flusher->is_active())
-    {
-        return false;
-    }
-    if (unsynced_big_writes.size() > 0 || unsynced_small_writes.size() > 0)
-    {
-        if (!readonly && !stop_sync_submitted)
-        {
-            // We should sync the blockstore before unmounting
-            blockstore_op_t *op = new blockstore_op_t;
-            op->opcode = BS_OP_SYNC;
-            op->buf = NULL;
-            op->callback = [](blockstore_op_t *op)
-            {
-                delete op;
-            };
-            enqueue_op(op);
-            stop_sync_submitted = true;
-        }
-        return false;
-    }
-    return true;
-}
-
-void blockstore_impl_t::check_wait(blockstore_op_t *op)
-{
-    if (PRIV(op)->wait_for == WAIT_SQE)
-    {
-        if (ringloop->space_left() < PRIV(op)->wait_detail)
-        {
-            // stop submission if there's still no free space
-            return;
-        }
-        PRIV(op)->wait_for = 0;
-    }
-    else if (PRIV(op)->wait_for == WAIT_IN_FLIGHT)
-    {
-        auto dirty_it = dirty_db.find((obj_ver_id){
-            .oid = op->oid,
-            .version = PRIV(op)->wait_detail,
-        });
-        if (dirty_it != dirty_db.end() && IS_IN_FLIGHT(dirty_it->second.state))
-        {
-            // do not submit
-            return;
-        }
-        PRIV(op)->wait_for = 0;
-    }
-    else if (PRIV(op)->wait_for == WAIT_JOURNAL)
-    {
-        if (journal.used_start == PRIV(op)->wait_detail)
-        {
-            // do not submit
-            return;
-        }
-        PRIV(op)->wait_for = 0;
-    }
-    else if (PRIV(op)->wait_for == WAIT_JOURNAL_BUFFER)
-    {
-        int next = ((journal.cur_sector + 1) % journal.sector_count);
-        if (journal.sector_info[next].usage_count > 0 ||
-            journal.sector_info[next].dirty)
-        {
-            // do not submit
-            return;
-        }
-        PRIV(op)->wait_for = 0;
-    }
-    else if (PRIV(op)->wait_for == WAIT_FREE)
-    {
-        if (!data_alloc->get_free_count() && !flusher->is_active())
-        {
-            return;
-        }
-        PRIV(op)->wait_for = 0;
-    }
-    else
-    {
-        throw std::runtime_error("BUG: op->wait_for value is unexpected");
-    }
-}
-
-void blockstore_impl_t::enqueue_op(blockstore_op_t *op, bool first)
-{
-    if (op->opcode < BS_OP_MIN || op->opcode > BS_OP_MAX ||
-        ((op->opcode == BS_OP_READ || op->opcode == BS_OP_WRITE) && (
-            op->offset >= block_size ||
-            op->len > block_size-op->offset ||
-            (op->len % disk_alignment)
-        )) ||
-        readonly && op->opcode != BS_OP_READ ||
-        first && op->opcode == BS_OP_WRITE)
-    {
-        // Basic verification not passed
-        op->retval = -EINVAL;
-        op->callback(op);
-        return;
-    }
-    if (op->opcode == BS_OP_SYNC_STAB_ALL)
-    {
-        std::function<void(blockstore_op_t*)> *old_callback = new std::function<void(blockstore_op_t*)>(op->callback);
-        op->opcode = BS_OP_SYNC;
-        op->callback = [this, old_callback](blockstore_op_t *op)
-        {
-            if (op->retval >= 0 && unstable_writes.size() > 0)
-            {
-                op->opcode = BS_OP_STABLE;
-                op->len = unstable_writes.size();
-                obj_ver_id *vers = new obj_ver_id[op->len];
-                op->buf = vers;
-                int i = 0;
-                for (auto it = unstable_writes.begin(); it != unstable_writes.end(); it++, i++)
-                {
-                    vers[i] = {
-                        .oid = it->first,
-                        .version = it->second,
-                    };
-                }
-                unstable_writes.clear();
-                op->callback = [this, old_callback](blockstore_op_t *op)
-                {
-                    obj_ver_id *vers = (obj_ver_id*)op->buf;
-                    delete[] vers;
-                    op->buf = NULL;
-                    (*old_callback)(op);
-                    delete old_callback;
-                };
-                this->enqueue_op(op);
-            }
-            else
-            {
-                (*old_callback)(op);
-                delete old_callback;
-            }
-        };
-    }
-    if (op->opcode == BS_OP_WRITE && !enqueue_write(op))
-    {
-        op->callback(op);
-        return;
-    }
-    if (0 && op->opcode == BS_OP_SYNC && immediate_commit)
-    {
-        op->retval = 0;
-        op->callback(op);
-        return;
-    }
-    // Call constructor without allocating memory. We'll call destructor before returning op back
-    new ((void*)op->private_data) blockstore_op_private_t;
-    PRIV(op)->wait_for = 0;
-    PRIV(op)->sync_state = 0;
-    PRIV(op)->pending_ops = 0;
-    if (!first)
-    {
-        submit_queue.push_back(op);
-    }
-    else
-    {
-        submit_queue.push_front(op);
-    }
-    ringloop->wakeup();
-}
-
-void blockstore_impl_t::process_list(blockstore_op_t *op)
-{
-    // Count objects
-    uint32_t list_pg = op->offset;
-    uint32_t pg_count = op->len;
-    uint64_t parity_block_size = op->oid.stripe;
-    if (pg_count != 0 && (parity_block_size < MIN_BLOCK_SIZE || list_pg >= pg_count))
-    {
-        op->retval = -EINVAL;
-        FINISH_OP(op);
-        return;
-    }
-    uint64_t stable_count = 0;
-    if (pg_count > 0)
-    {
-        for (auto it = clean_db.begin(); it != clean_db.end(); it++)
-        {
-            uint32_t pg = (it->first.inode + it->first.stripe / parity_block_size) % pg_count;
-            if (pg == list_pg)
-            {
-                stable_count++;
-            }
-        }
-    }
-    else
-    {
-        stable_count = clean_db.size();
-    }
-    uint64_t total_count = stable_count;
-    for (auto it = dirty_db.begin(); it != dirty_db.end(); it++)
-    {
-        if (!pg_count || ((it->first.oid.inode + it->first.oid.stripe / parity_block_size) % pg_count) == list_pg)
-        {
-            if (IS_STABLE(it->second.state))
-            {
-                stable_count++;
-            }
-            total_count++;
-        }
-    }
-    // Allocate memory
-    op->version = stable_count;
-    op->retval = total_count;
-    op->buf = malloc(sizeof(obj_ver_id) * total_count);
-    if (!op->buf)
-    {
-        op->retval = -ENOMEM;
-        FINISH_OP(op);
-        return;
-    }
-    obj_ver_id *vers = (obj_ver_id*)op->buf;
-    int i = 0;
-    for (auto it = clean_db.begin(); it != clean_db.end(); it++)
-    {
-        if (!pg_count || ((it->first.inode + it->first.stripe / parity_block_size) % pg_count) == list_pg)
-        {
-            vers[i++] = {
-                .oid = it->first,
-                .version = it->second.version,
-            };
-        }
-    }
-    int j = stable_count;
-    for (auto it = dirty_db.begin(); it != dirty_db.end(); it++)
-    {
-        if (!pg_count || ((it->first.oid.inode + it->first.oid.stripe / parity_block_size) % pg_count) == list_pg)
-        {
-            if (IS_STABLE(it->second.state))
-            {
-                vers[i++] = it->first;
-            }
-            else
-            {
-                vers[j++] = it->first;
-            }
-        }
-    }
-    FINISH_OP(op);
-}

blockstore_rollback.cpp

@@ -1,187 +0,0 @@
-#include "blockstore_impl.h"
-
-int blockstore_impl_t::dequeue_rollback(blockstore_op_t *op)
-{
-    obj_ver_id* v;
-    int i, todo = op->len;
-    for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
-    {
-        // Check that there are some versions greater than v->version (which may be zero),
-        // check that they're unstable, synced, and not currently written to
-        auto dirty_it = dirty_db.lower_bound((obj_ver_id){
-            .oid = v->oid,
-            .version = UINT64_MAX,
-        });
-        if (dirty_it == dirty_db.begin())
-        {
-        bad_op:
-            op->retval = -EINVAL;
-            FINISH_OP(op);
-            return 1;
-        }
-        else
-        {
-            dirty_it--;
-            if (dirty_it->first.oid != v->oid || dirty_it->first.version < v->version)
-            {
-                goto bad_op;
-            }
-            while (dirty_it->first.oid == v->oid && dirty_it->first.version > v->version)
-            {
-                if (!IS_SYNCED(dirty_it->second.state) ||
-                    IS_STABLE(dirty_it->second.state))
-                {
-                    goto bad_op;
-                }
-                if (dirty_it == dirty_db.begin())
-                {
-                    break;
-                }
-                dirty_it--;
-            }
-        }
-    }
-    // Check journal space
-    blockstore_journal_check_t space_check(this);
-    if (!space_check.check_available(op, todo, sizeof(journal_entry_rollback), 0))
-    {
-        return 0;
-    }
-    // There is sufficient space. Get SQEs
-    struct io_uring_sqe *sqe[space_check.sectors_required];
-    for (i = 0; i < space_check.sectors_required; i++)
-    {
-        BS_SUBMIT_GET_SQE_DECL(sqe[i]);
-    }
-    // Prepare and submit journal entries
-    auto cb = [this, op](ring_data_t *data) { handle_rollback_event(data, op); };
-    int s = 0, cur_sector = -1;
-    if ((journal_block_size - journal.in_sector_pos) < sizeof(journal_entry_rollback) &&
-        journal.sector_info[journal.cur_sector].dirty)
-    {
-        if (cur_sector == -1)
-            PRIV(op)->min_used_journal_sector = 1 + journal.cur_sector;
-        cur_sector = journal.cur_sector;
-        prepare_journal_sector_write(journal, cur_sector, sqe[s++], cb);
-    }
-    for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
-    {
-        // FIXME This is here only for the purpose of tracking unstable_writes. Remove if not required
-        // FIXME ...aaaand this is similar to blockstore_init.cpp - maybe dedup it?
-        auto dirty_it = dirty_db.lower_bound((obj_ver_id){
-            .oid = v->oid,
-            .version = UINT64_MAX,
-        });
-        uint64_t max_unstable = 0;
-        while (dirty_it != dirty_db.begin())
-        {
-            dirty_it--;
-            if (dirty_it->first.oid != v->oid)
-                break;
-            else if (dirty_it->first.version <= v->version)
-            {
-                if (!IS_STABLE(dirty_it->second.state))
-                    max_unstable = dirty_it->first.version;
-                break;
-            }
-        }
-        auto unstab_it = unstable_writes.find(v->oid);
-        if (unstab_it != unstable_writes.end())
-        {
-            if (max_unstable == 0)
-                unstable_writes.erase(unstab_it);
-            else
-                unstab_it->second = max_unstable;
-        }
-        journal_entry_rollback *je = (journal_entry_rollback*)
-            prefill_single_journal_entry(journal, JE_ROLLBACK, sizeof(journal_entry_rollback));
-        journal.sector_info[journal.cur_sector].dirty = false;
-        je->oid = v->oid;
-        je->version = v->version;
-        je->crc32 = je_crc32((journal_entry*)je);
-        journal.crc32_last = je->crc32;
-        if (cur_sector != journal.cur_sector)
-        {
-            if (cur_sector == -1)
-                PRIV(op)->min_used_journal_sector = 1 + journal.cur_sector;
-            cur_sector = journal.cur_sector;
-            prepare_journal_sector_write(journal, cur_sector, sqe[s++], cb);
-        }
-    }
-    PRIV(op)->max_used_journal_sector = 1 + journal.cur_sector;
-    PRIV(op)->pending_ops = s;
-    return 1;
-}
-
-void blockstore_impl_t::handle_rollback_event(ring_data_t *data, blockstore_op_t *op)
-{
-    live = true;
-    if (data->res != data->iov.iov_len)
-    {
-        throw std::runtime_error(
-            "write operation failed ("+std::to_string(data->res)+" != "+std::to_string(data->iov.iov_len)+
-            "). in-memory state is corrupted. AAAAAAAaaaaaaaaa!!!111"
-        );
-    }
-    PRIV(op)->pending_ops--;
-    if (PRIV(op)->pending_ops == 0)
-    {
-        // Release used journal sectors
-        release_journal_sectors(op);
-        obj_ver_id* v;
-        int i;
-        for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
-        {
-            // Erase dirty_db entries
-            auto rm_end = dirty_db.lower_bound((obj_ver_id){
-                .oid = v->oid,
-                .version = UINT64_MAX,
-            });
-            rm_end--;
-            auto rm_start = rm_end;
-            while (1)
-            {
-                if (rm_end->first.oid != v->oid)
-                    break;
-                else if (rm_end->first.version <= v->version)
-                    break;
-                rm_start = rm_end;
-                if (rm_end == dirty_db.begin())
-                    break;
-                rm_end--;
-            }
-            if (rm_end != rm_start)
-                erase_dirty(rm_start, rm_end, UINT64_MAX);
-        }
-        journal.trim();
-        // Acknowledge op
-        op->retval = 0;
-        FINISH_OP(op);
-    }
-}
-
-void blockstore_impl_t::erase_dirty(blockstore_dirty_db_t::iterator dirty_start, blockstore_dirty_db_t::iterator dirty_end, uint64_t clean_loc)
-{
-    auto dirty_it = dirty_end;
-    while (dirty_it != dirty_start)
-    {
-        dirty_it--;
-        if (IS_BIG_WRITE(dirty_it->second.state) && dirty_it->second.location != clean_loc)
-        {
-#ifdef BLOCKSTORE_DEBUG
-            printf("Free block %lu\n", dirty_it->second.location >> block_order);
-#endif
-            data_alloc->set(dirty_it->second.location >> block_order, false);
-        }
-#ifdef BLOCKSTORE_DEBUG
-        printf("remove usage of journal offset %lu by %lu:%lu v%lu\n", dirty_it->second.journal_sector,
-            dirty_it->first.oid.inode, dirty_it->first.oid.stripe, dirty_it->first.version);
-#endif
-        int used = --journal.used_sectors[dirty_it->second.journal_sector];
-        if (used == 0)
-        {
-            journal.used_sectors.erase(dirty_it->second.journal_sector);
-        }
-    }
-    dirty_db.erase(dirty_start, dirty_end);
-}

blockstore_sync.cpp

@@ -1,269 +0,0 @@
-#include "blockstore_impl.h"
-
-#define SYNC_HAS_SMALL 1
-#define SYNC_HAS_BIG 2
-#define SYNC_DATA_SYNC_SENT 3
-#define SYNC_DATA_SYNC_DONE 4
-#define SYNC_JOURNAL_WRITE_SENT 5
-#define SYNC_JOURNAL_WRITE_DONE 6
-#define SYNC_JOURNAL_SYNC_SENT 7
-#define SYNC_DONE 8
-
-int blockstore_impl_t::dequeue_sync(blockstore_op_t *op)
-{
-    if (PRIV(op)->sync_state == 0)
-    {
-        stop_sync_submitted = false;
-        PRIV(op)->sync_big_writes.swap(unsynced_big_writes);
-        PRIV(op)->sync_small_writes.swap(unsynced_small_writes);
-        PRIV(op)->sync_small_checked = 0;
-        PRIV(op)->sync_big_checked = 0;
-        unsynced_big_writes.clear();
-        unsynced_small_writes.clear();
-        if (PRIV(op)->sync_big_writes.size() > 0)
-            PRIV(op)->sync_state = SYNC_HAS_BIG;
-        else if (PRIV(op)->sync_small_writes.size() > 0)
-            PRIV(op)->sync_state = SYNC_HAS_SMALL;
-        else
-            PRIV(op)->sync_state = SYNC_DONE;
-        // Always add sync to in_progress_syncs because we clear unsynced_big_writes and unsynced_small_writes
-        PRIV(op)->prev_sync_count = in_progress_syncs.size();
-        PRIV(op)->in_progress_ptr = in_progress_syncs.insert(in_progress_syncs.end(), op);
-    }
-    continue_sync(op);
-    // Always dequeue because we always add syncs to in_progress_syncs
-    return 1;
-}
-
-int blockstore_impl_t::continue_sync(blockstore_op_t *op)
-{
-    auto cb = [this, op](ring_data_t *data) { handle_sync_event(data, op); };
-    if (PRIV(op)->sync_state == SYNC_HAS_SMALL)
-    {
-        // No big writes, just fsync the journal
-        for (; PRIV(op)->sync_small_checked < PRIV(op)->sync_small_writes.size(); PRIV(op)->sync_small_checked++)
-        {
-            if (IS_IN_FLIGHT(dirty_db[PRIV(op)->sync_small_writes[PRIV(op)->sync_small_checked]].state))
-            {
-                // Wait for small inflight writes to complete
-                return 0;
-            }
-        }
-        if (journal.sector_info[journal.cur_sector].dirty)
-        {
-            // Write out the last journal sector if it happens to be dirty
-            BS_SUBMIT_GET_ONLY_SQE(sqe);
-            prepare_journal_sector_write(journal, journal.cur_sector, sqe, cb);
-            PRIV(op)->min_used_journal_sector = PRIV(op)->max_used_journal_sector = 1 + journal.cur_sector;
-            PRIV(op)->pending_ops = 1;
-            PRIV(op)->sync_state = SYNC_JOURNAL_WRITE_SENT;
-            return 1;
-        }
-        else
-        {
-            PRIV(op)->sync_state = SYNC_JOURNAL_WRITE_DONE;
-        }
-    }
-    if (PRIV(op)->sync_state == SYNC_HAS_BIG)
-    {
-        for (; PRIV(op)->sync_big_checked < PRIV(op)->sync_big_writes.size(); PRIV(op)->sync_big_checked++)
-        {
-            if (IS_IN_FLIGHT(dirty_db[PRIV(op)->sync_big_writes[PRIV(op)->sync_big_checked]].state))
-            {
-                // Wait for big inflight writes to complete
-                return 0;
-            }
-        }
-        // 1st step: fsync data
-        if (!disable_data_fsync)
-        {
-            BS_SUBMIT_GET_SQE(sqe, data);
-            my_uring_prep_fsync(sqe, data_fd, IORING_FSYNC_DATASYNC);
-            data->iov = { 0 };
-            data->callback = cb;
-            PRIV(op)->min_used_journal_sector = PRIV(op)->max_used_journal_sector = 0;
-            PRIV(op)->pending_ops = 1;
-            PRIV(op)->sync_state = SYNC_DATA_SYNC_SENT;
-            return 1;
-        }
-        else
-        {
-            PRIV(op)->sync_state = SYNC_DATA_SYNC_DONE;
-        }
-    }
-    if (PRIV(op)->sync_state == SYNC_DATA_SYNC_DONE)
-    {
-        for (; PRIV(op)->sync_small_checked < PRIV(op)->sync_small_writes.size(); PRIV(op)->sync_small_checked++)
-        {
-            if (IS_IN_FLIGHT(dirty_db[PRIV(op)->sync_small_writes[PRIV(op)->sync_small_checked]].state))
-            {
-                // Wait for small inflight writes to complete
-                return 0;
-            }
-        }
-        // 2nd step: Data device is synced, prepare & write journal entries
-        // Check space in the journal and journal memory buffers
-        blockstore_journal_check_t space_check(this);
-        if (!space_check.check_available(op, PRIV(op)->sync_big_writes.size(), sizeof(journal_entry_big_write), 0))
-        {
-            return 0;
-        }
-        // Get SQEs. Don't bother about merging, submit each journal sector as a separate request
-        struct io_uring_sqe *sqe[space_check.sectors_required];
-        for (int i = 0; i < space_check.sectors_required; i++)
-        {
-            BS_SUBMIT_GET_SQE_DECL(sqe[i]);
-        }
-        // Prepare and submit journal entries
-        auto it = PRIV(op)->sync_big_writes.begin();
-        int s = 0, cur_sector = -1;
-        if ((journal_block_size - journal.in_sector_pos) < sizeof(journal_entry_big_write) &&
-            journal.sector_info[journal.cur_sector].dirty)
-        {
-            if (cur_sector == -1)
-                PRIV(op)->min_used_journal_sector = 1 + journal.cur_sector;
-            cur_sector = journal.cur_sector;
-            prepare_journal_sector_write(journal, cur_sector, sqe[s++], cb);
-        }
-        while (it != PRIV(op)->sync_big_writes.end())
-        {
-            journal_entry_big_write *je = (journal_entry_big_write*)
-                prefill_single_journal_entry(journal, JE_BIG_WRITE, sizeof(journal_entry_big_write));
-            dirty_db[*it].journal_sector = journal.sector_info[journal.cur_sector].offset;
-            journal.sector_info[journal.cur_sector].dirty = false;
-            journal.used_sectors[journal.sector_info[journal.cur_sector].offset]++;
-#ifdef BLOCKSTORE_DEBUG
-            printf("journal offset %lu is used by %lu:%lu v%lu\n", dirty_db[*it].journal_sector, it->oid.inode, it->oid.stripe, it->version);
-#endif
-            je->oid = it->oid;
-            je->version = it->version;
-            je->offset = dirty_db[*it].offset;
-            je->len = dirty_db[*it].len;
-            je->location = dirty_db[*it].location;
-            je->crc32 = je_crc32((journal_entry*)je);
-            journal.crc32_last = je->crc32;
-            it++;
-            if (cur_sector != journal.cur_sector)
-            {
-                if (cur_sector == -1)
-                    PRIV(op)->min_used_journal_sector = 1 + journal.cur_sector;
-                cur_sector = journal.cur_sector;
-                prepare_journal_sector_write(journal, cur_sector, sqe[s++], cb);
-            }
-        }
-        PRIV(op)->max_used_journal_sector = 1 + journal.cur_sector;
-        PRIV(op)->pending_ops = s;
-        PRIV(op)->sync_state = SYNC_JOURNAL_WRITE_SENT;
-        return 1;
-    }
-    if (PRIV(op)->sync_state == SYNC_JOURNAL_WRITE_DONE)
-    {
-        if (!disable_journal_fsync)
-        {
-            BS_SUBMIT_GET_SQE(sqe, data);
-            my_uring_prep_fsync(sqe, journal.fd, IORING_FSYNC_DATASYNC);
-            data->iov = { 0 };
-            data->callback = cb;
-            PRIV(op)->pending_ops = 1;
-            PRIV(op)->sync_state = SYNC_JOURNAL_SYNC_SENT;
-            return 1;
-        }
-        else
-        {
-            PRIV(op)->sync_state = SYNC_DONE;
-        }
-    }
-    if (PRIV(op)->sync_state == SYNC_DONE)
-    {
-        ack_sync(op);
-    }
-    return 1;
-}
-
-void blockstore_impl_t::handle_sync_event(ring_data_t *data, blockstore_op_t *op)
-{
-    live = true;
-    if (data->res != data->iov.iov_len)
-    {
-        throw std::runtime_error(
-            "write operation failed ("+std::to_string(data->res)+" != "+std::to_string(data->iov.iov_len)+
-            "). in-memory state is corrupted. AAAAAAAaaaaaaaaa!!!111"
-        );
-    }
-    PRIV(op)->pending_ops--;
-    if (PRIV(op)->pending_ops == 0)
-    {
-        // Release used journal sectors
-        release_journal_sectors(op);
-        // Handle states
-        if (PRIV(op)->sync_state == SYNC_DATA_SYNC_SENT)
-        {
-            PRIV(op)->sync_state = SYNC_DATA_SYNC_DONE;
-        }
-        else if (PRIV(op)->sync_state == SYNC_JOURNAL_WRITE_SENT)
-        {
-            PRIV(op)->sync_state = SYNC_JOURNAL_WRITE_DONE;
-        }
-        else if (PRIV(op)->sync_state == SYNC_JOURNAL_SYNC_SENT)
-        {
-            PRIV(op)->sync_state = SYNC_DONE;
-            ack_sync(op);
-        }
-        else
-        {
-            throw std::runtime_error("BUG: unexpected sync op state");
-        }
-    }
-}
-
-int blockstore_impl_t::ack_sync(blockstore_op_t *op)
-{
-    if (PRIV(op)->sync_state == SYNC_DONE && PRIV(op)->prev_sync_count == 0)
-    {
-        // Remove dependency of subsequent syncs
-        auto it = PRIV(op)->in_progress_ptr;
-        int done_syncs = 1;
-        ++it;
-        // Acknowledge sync
-        ack_one_sync(op);
-        while (it != in_progress_syncs.end())
-        {
-            auto & next_sync = *it++;
-            PRIV(next_sync)->prev_sync_count -= done_syncs;
-            if (PRIV(next_sync)->prev_sync_count == 0 && PRIV(next_sync)->sync_state == SYNC_DONE)
-            {
-                done_syncs++;
-                // Acknowledge next_sync
-                ack_one_sync(next_sync);
-            }
-        }
-        return 1;
-    }
-    return 0;
-}
-
-void blockstore_impl_t::ack_one_sync(blockstore_op_t *op)
-{
-    // Handle states
-    for (auto it = PRIV(op)->sync_big_writes.begin(); it != PRIV(op)->sync_big_writes.end(); it++)
-    {
-#ifdef BLOCKSTORE_DEBUG
-        printf("Ack sync big %lu:%lu v%lu\n", it->oid.inode, it->oid.stripe, it->version);
-#endif
-        auto & unstab = unstable_writes[it->oid];
-        unstab = unstab < it->version ? it->version : unstab;
-        dirty_db[*it].state = ST_D_META_SYNCED;
-    }
-    for (auto it = PRIV(op)->sync_small_writes.begin(); it != PRIV(op)->sync_small_writes.end(); it++)
-    {
-#ifdef BLOCKSTORE_DEBUG
-        printf("Ack sync small %lu:%lu v%lu\n", it->oid.inode, it->oid.stripe, it->version);
-#endif
-        auto & unstab = unstable_writes[it->oid];
-        unstab = unstab < it->version ? it->version : unstab;
-        dirty_db[*it].state = dirty_db[*it].state == ST_DEL_WRITTEN ? ST_DEL_SYNCED : ST_J_SYNCED;
-    }
-    in_progress_syncs.erase(PRIV(op)->in_progress_ptr);
-    op->retval = 0;
-    FINISH_OP(op);
-}

blockstore_write.cpp

@@ -1,328 +0,0 @@
-#include "blockstore_impl.h"
-
-bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
-{
-    // Check or assign version number
-    bool found = false, deleted = false, is_del = (op->opcode == BS_OP_DELETE);
-    uint64_t version = 1;
-    if (dirty_db.size() > 0)
-    {
-        auto dirty_it = dirty_db.upper_bound((obj_ver_id){
-            .oid = op->oid,
-            .version = UINT64_MAX,
-        });
-        dirty_it--; // segfaults when dirty_db is empty
-        if (dirty_it != dirty_db.end() && dirty_it->first.oid == op->oid)
-        {
-            found = true;
-            version = dirty_it->first.version + 1;
-            deleted = IS_DELETE(dirty_it->second.state);
-        }
-    }
-    if (!found)
-    {
-        auto clean_it = clean_db.find(op->oid);
-        if (clean_it != clean_db.end())
-        {
-            version = clean_it->second.version + 1;
-        }
-        else
-        {
-            deleted = true;
-        }
-    }
-    if (op->version == 0)
-    {
-        op->version = version;
-    }
-    else if (op->version < version)
-    {
-        // Invalid version requested
-        op->retval = -EINVAL;
-        return false;
-    }
-    if (deleted && is_del)
-    {
-        // Already deleted
-        op->retval = 0;
-        return false;
-    }
-    // Immediately add the operation into dirty_db, so subsequent reads could see it
-#ifdef BLOCKSTORE_DEBUG
-    printf("%s %lu:%lu v%lu\n", is_del ? "Delete" : "Write", op->oid.inode, op->oid.stripe, op->version);
-#endif
-    dirty_db.emplace((obj_ver_id){
-        .oid = op->oid,
-        .version = op->version,
-    }, (dirty_entry){
-        .state = (uint32_t)(
-            is_del
-                ? ST_DEL_IN_FLIGHT
-                : (op->len == block_size || deleted ? ST_D_IN_FLIGHT : ST_J_IN_FLIGHT)
-        ),
-        .flags = 0,
-        .location = 0,
-        .offset = is_del ? 0 : op->offset,
-        .len = is_del ? 0 : op->len,
-        .journal_sector = 0,
-    });
-    return true;
-}
-
-// First step of the write algorithm: dequeue operation and submit initial write(s)
-int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
-{
-    auto dirty_it = dirty_db.find((obj_ver_id){
-        .oid = op->oid,
-        .version = op->version,
-    });
-    if (dirty_it->second.state == ST_D_IN_FLIGHT)
-    {
-        blockstore_journal_check_t space_check(this);
-        if (!space_check.check_available(op, unsynced_big_writes.size() + 1, sizeof(journal_entry_big_write), JOURNAL_STABILIZE_RESERVATION))
-        {
-            return 0;
-        }
-        // Big (redirect) write
-        uint64_t loc = data_alloc->find_free();
-        if (loc == UINT64_MAX)
-        {
-            // no space
-            if (flusher->is_active())
-            {
-                // hope that some space will be available after flush
-                PRIV(op)->wait_for = WAIT_FREE;
-                return 0;
-            }
-            op->retval = -ENOSPC;
-            FINISH_OP(op);
-            return 1;
-        }
-        BS_SUBMIT_GET_SQE(sqe, data);
-        dirty_it->second.location = loc << block_order;
-        dirty_it->second.state = ST_D_SUBMITTED;
-#ifdef BLOCKSTORE_DEBUG
-        printf("Allocate block %lu\n", loc);
-#endif
-        data_alloc->set(loc, true);
-        uint64_t stripe_offset = (op->offset % bitmap_granularity);
-        uint64_t stripe_end = (op->offset + op->len) % bitmap_granularity;
-        // Zero fill up to bitmap_granularity
-        int vcnt = 0;
-        if (stripe_offset)
-        {
-            PRIV(op)->iov_zerofill[vcnt++] = (struct iovec){ zero_object, stripe_offset };
-        }
-        PRIV(op)->iov_zerofill[vcnt++] = (struct iovec){ op->buf, op->len };
-        if (stripe_end)
-        {
-            stripe_end = bitmap_granularity - stripe_end;
-            PRIV(op)->iov_zerofill[vcnt++] = (struct iovec){ zero_object, stripe_end };
-        }
-        data->iov.iov_len = op->len + stripe_offset + stripe_end; // to check it in the callback
-        data->callback = [this, op](ring_data_t *data) { handle_write_event(data, op); };
-        my_uring_prep_writev(
-            sqe, data_fd, PRIV(op)->iov_zerofill, vcnt, data_offset + (loc << block_order) + op->offset - stripe_offset
-        );
-        PRIV(op)->pending_ops = 1;
-        PRIV(op)->min_used_journal_sector = PRIV(op)->max_used_journal_sector = 0;
-        // Remember big write as unsynced
-        unsynced_big_writes.push_back((obj_ver_id){
-            .oid = op->oid,
-            .version = op->version,
-        });
-    }
-    else
-    {
-        // Small (journaled) write
-        // First check if the journal has sufficient space
-        blockstore_journal_check_t space_check(this);
-        if (unsynced_big_writes.size() && !space_check.check_available(op, unsynced_big_writes.size(), sizeof(journal_entry_big_write), 0)
-            || !space_check.check_available(op, 1, sizeof(journal_entry_small_write), op->len + JOURNAL_STABILIZE_RESERVATION))
-        {
-            return 0;
-        }
-        // There is sufficient space. Get SQE(s)
-        struct io_uring_sqe *sqe1 = NULL;
-        if ((journal_block_size - journal.in_sector_pos) < sizeof(journal_entry_small_write) &&
-            journal.sector_info[journal.cur_sector].dirty)
-        {
-            // Write current journal sector only if it's dirty and full
-            BS_SUBMIT_GET_SQE_DECL(sqe1);
-        }
-        struct io_uring_sqe *sqe2 = NULL;
-        if (op->len > 0)
-        {
-            BS_SUBMIT_GET_SQE_DECL(sqe2);
-        }
-        // Got SQEs. Prepare previous journal sector write if required
-        auto cb = [this, op](ring_data_t *data) { handle_write_event(data, op); };
-        if (sqe1)
-        {
-            prepare_journal_sector_write(journal, journal.cur_sector, sqe1, cb);
-            // FIXME rename to min/max _flushing
-            PRIV(op)->min_used_journal_sector = PRIV(op)->max_used_journal_sector = 1 + journal.cur_sector;
-            PRIV(op)->pending_ops++;
-        }
-        else
-        {
-            PRIV(op)->min_used_journal_sector = PRIV(op)->max_used_journal_sector = 0;
-        }
-        // Then pre-fill journal entry
-        journal_entry_small_write *je = (journal_entry_small_write*)
-            prefill_single_journal_entry(journal, JE_SMALL_WRITE, sizeof(journal_entry_small_write));
-        dirty_it->second.journal_sector = journal.sector_info[journal.cur_sector].offset;
-        journal.used_sectors[journal.sector_info[journal.cur_sector].offset]++;
-#ifdef BLOCKSTORE_DEBUG
-        printf("journal offset %lu is used by %lu:%lu v%lu\n", dirty_it->second.journal_sector, dirty_it->first.oid.inode, dirty_it->first.oid.stripe, dirty_it->first.version);
-#endif
-        // Figure out where data will be
-        journal.next_free = (journal.next_free + op->len) <= journal.len ? journal.next_free : journal_block_size;
-        je->oid = op->oid;
-        je->version = op->version;
-        je->offset = op->offset;
-        je->len = op->len;
-        je->data_offset = journal.next_free;
-        je->crc32_data = crc32c(0, op->buf, op->len);
-        je->crc32 = je_crc32((journal_entry*)je);
-        journal.crc32_last = je->crc32;
-        if (op->len > 0)
-        {
-            // Prepare journal data write
-            if (journal.inmemory)
-            {
-                // Copy data
-                memcpy(journal.buffer + journal.next_free, op->buf, op->len);
-            }
-            ring_data_t *data2 = ((ring_data_t*)sqe2->user_data);
-            data2->iov = (struct iovec){ op->buf, op->len };
-            data2->callback = cb;
-            my_uring_prep_writev(
-                sqe2, journal.fd, &data2->iov, 1, journal.offset + journal.next_free
-            );
-            PRIV(op)->pending_ops++;
-        }
-        else
-        {
-            // Zero-length overwrite. Allowed to bump object version in EC placement groups without actually writing data
-        }
-        dirty_it->second.location = journal.next_free;
-        dirty_it->second.state = ST_J_SUBMITTED;
-        journal.next_free += op->len;
-        if (journal.next_free >= journal.len)
-        {
-            journal.next_free = journal_block_size;
-        }
-        // Remember small write as unsynced
-        unsynced_small_writes.push_back((obj_ver_id){
-            .oid = op->oid,
-            .version = op->version,
-        });
-        if (!PRIV(op)->pending_ops)
-        {
-            ack_write(op);
-        }
-    }
-    return 1;
-}
-
-void blockstore_impl_t::handle_write_event(ring_data_t *data, blockstore_op_t *op)
-{
-    live = true;
-    if (data->res != data->iov.iov_len)
-    {
-        // FIXME: our state becomes corrupted after a write error. maybe do something better than just die
-        throw std::runtime_error(
-            "write operation failed ("+std::to_string(data->res)+" != "+std::to_string(data->iov.iov_len)+
-            "). in-memory state is corrupted. AAAAAAAaaaaaaaaa!!!111"
-        );
-    }
-    PRIV(op)->pending_ops--;
-    if (PRIV(op)->pending_ops == 0)
-    {
-        release_journal_sectors(op);
-        ack_write(op);
-    }
-}
-
-void blockstore_impl_t::release_journal_sectors(blockstore_op_t *op)
-{
-    // Release used journal sectors
-    if (PRIV(op)->min_used_journal_sector > 0 &&
-        PRIV(op)->max_used_journal_sector > 0)
-    {
-        uint64_t s = PRIV(op)->min_used_journal_sector;
-        while (1)
-        {
-            journal.sector_info[s-1].usage_count--;
-            if (s == PRIV(op)->max_used_journal_sector)
-                break;
-            s = 1 + s % journal.sector_count;
-        }
-        PRIV(op)->min_used_journal_sector = PRIV(op)->max_used_journal_sector = 0;
-    }
-}
-
-void blockstore_impl_t::ack_write(blockstore_op_t *op)
-{
-    // Switch object state
-    auto & dirty_entry = dirty_db[(obj_ver_id){
-        .oid = op->oid,
-        .version = op->version,
-    }];
-#ifdef BLOCKSTORE_DEBUG
-    printf("Ack write %lu:%lu v%lu = %d\n", op->oid.inode, op->oid.stripe, op->version, dirty_entry.state);
-#endif
-    if (dirty_entry.state == ST_J_SUBMITTED)
-    {
-        dirty_entry.state = ST_J_WRITTEN;
-    }
-    else if (dirty_entry.state == ST_D_SUBMITTED)
-    {
-        dirty_entry.state = ST_D_WRITTEN;
-    }
-    else if (dirty_entry.state == ST_DEL_SUBMITTED)
-    {
-        dirty_entry.state = ST_DEL_WRITTEN;
-    }
-    // Acknowledge write without sync
-    op->retval = op->len;
-    FINISH_OP(op);
-}
-
-int blockstore_impl_t::dequeue_del(blockstore_op_t *op)
-{
-    auto dirty_it = dirty_db.find((obj_ver_id){
-        .oid = op->oid,
-        .version = op->version,
-    });
-    blockstore_journal_check_t space_check(this);
-    if (!space_check.check_available(op, 1, sizeof(journal_entry_del), 0))
-    {
-        return 0;
-    }
-    BS_SUBMIT_GET_ONLY_SQE(sqe);
-    // Prepare journal sector write
-    journal_entry_del *je = (journal_entry_del*)
-        prefill_single_journal_entry(journal, JE_DELETE, sizeof(struct journal_entry_del));
-    dirty_it->second.journal_sector = journal.sector_info[journal.cur_sector].offset;
-    journal.used_sectors[journal.sector_info[journal.cur_sector].offset]++;
-#ifdef BLOCKSTORE_DEBUG
-    printf("journal offset %lu is used by %lu:%lu v%lu\n", dirty_it->second.journal_sector, dirty_it->first.oid.inode, dirty_it->first.oid.stripe, dirty_it->first.version);
-#endif
-    je->oid = op->oid;
-    je->version = op->version;
-    je->crc32 = je_crc32((journal_entry*)je);
-    journal.crc32_last = je->crc32;
-    auto cb = [this, op](ring_data_t *data) { handle_write_event(data, op); };
-    prepare_journal_sector_write(journal, journal.cur_sector, sqe, cb);
-    PRIV(op)->min_used_journal_sector = PRIV(op)->max_used_journal_sector = 1 + journal.cur_sector;
-    PRIV(op)->pending_ops = 1;
-    dirty_it->second.state = ST_DEL_SUBMITTED;
-    // Remember small write as unsynced
-    unsynced_small_writes.push_back((obj_ver_id){
-        .oid = op->oid,
-        .version = op->version,
-    });
-    return 1;
-}

copy-fio-includes.sh

@@ -0,0 +1,13 @@
+#!/bin/bash
+
+gcc -I. -E -o fio_headers.i src/fio_headers.h
+
+rm -rf fio-copy
+for i in `grep -Po 'fio/[^"]+' fio_headers.i | sort | uniq`; do
+    j=${i##fio/}
+    p=$(dirname $j)
+    mkdir -p fio-copy/$p
+    cp $i fio-copy/$j
+done
+
+rm fio_headers.i

copy-qemu-includes.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+
+#cd qemu
+#debian/rules b/configure-stamp
+#cd b/qemu; make qapi
+
+gcc -I qemu/b/qemu `pkg-config glib-2.0 --cflags` \
+    -I qemu/include -E -o qemu_driver.i src/qemu_driver.c
+
+rm -rf qemu-copy
+for i in `grep -Po 'qemu/[^"]+' qemu_driver.i | sort | uniq`; do
+    j=${i##qemu/}
+    p=$(dirname $j)
+    mkdir -p qemu-copy/$p
+    cp $i qemu-copy/$j
+done
+
+rm qemu_driver.i

cpp-btree

@@ -0,0 +1 @@
+Subproject commit 5dc108754ad40d3b1d024f9bd7cca0595ef1a1db

debian/build-vitastor-bullseye.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+
+sed 's/$REL/bullseye/g' < vitastor.Dockerfile > ../Dockerfile
+cd ..
+mkdir -p packages
+sudo podman build -v `pwd`/packages:/root/packages -f Dockerfile .
+rm Dockerfile

debian/build-vitastor-buster.sh

@@ -0,0 +1,7 @@
+#!/bin/bash
+
+sed 's/$REL/buster/g' < vitastor.Dockerfile > ../Dockerfile
+cd ..
+mkdir -p packages
+sudo podman build -v `pwd`/packages:/root/packages -f Dockerfile .
+rm Dockerfile

debian/changelog

@@ -0,0 +1,17 @@
+vitastor (0.5.10-1) unstable; urgency=medium
+
+  * Bugfixes
+
+ -- Vitaliy Filippov <vitalif@yourcmc.ru>  Tue, 02 Feb 2021 23:01:24 +0300
+
+vitastor (0.5.1-1) unstable; urgency=medium
+
+  * Add jerasure support
+
+ -- Vitaliy Filippov <vitalif@yourcmc.ru>  Sat, 05 Dec 2020 17:02:26 +0300
+
+vitastor (0.5-1) unstable; urgency=medium
+
+  * First packaging for Debian
+
+ -- Vitaliy Filippov <vitalif@yourcmc.ru>  Thu, 05 Nov 2020 02:20:59 +0300

debian/compat

@@ -0,0 +1 @@
+13

debian/control

@@ -0,0 +1,17 @@
+Source: vitastor
+Section: admin
+Priority: optional
+Maintainer: Vitaliy Filippov <vitalif@yourcmc.ru>
+Build-Depends: debhelper, liburing-dev (>= 0.6), g++ (>= 8), libstdc++6 (>= 8), linux-libc-dev, libgoogle-perftools-dev, libjerasure-dev, libgf-complete-dev
+Standards-Version: 4.5.0
+Homepage: https://vitastor.io/
+Rules-Requires-Root: no
+
+Package: vitastor
+Architecture: amd64
+Depends: ${shlibs:Depends}, ${misc:Depends}, fio (= ${dep:fio}), qemu (= ${dep:qemu}), nodejs (>= 10), node-sprintf-js, node-ws (>= 7), libjerasure2, lp-solve
+Description: Vitastor, a fast software-defined clustered block storage
+ Vitastor is a small, simple and fast clustered block storage (storage for VM drives),
+ architecturally similar to Ceph which means strong consistency, primary-replication,
+ symmetric clustering and automatic data distribution over any number of drives of any
+ size with configurable redundancy (replication or erasure codes/XOR).

debian/copyright

@@ -0,0 +1,21 @@
+Format: https://www.debian.org/doc/packaging-manuals/copyright-format/1.0/
+Upstream-Name: vitastor
+Upstream-Contact: Vitaliy Filippov <vitalif@yourcmc.ru>
+Source: https://vitastor.io
+
+Files: *
+Copyright: 2019+ Vitaliy Filippov <vitalif@yourcmc.ru>
+License: Multiple licenses VNPL-1.1 and/or GPL-2.0+
+ All server-side code (OSD, Monitor and so on) is licensed under the terms of
+ Vitastor Network Public License 1.1 (VNPL 1.1), a copyleft license based on
+ GNU GPLv3.0 with the additional "Network Interaction" clause which requires
+ opensourcing all programs directly or indirectly interacting with Vitastor
+ through a computer network and expressly designed to be used in conjunction
+ with it ("Proxy Programs"). Proxy Programs may be made public not only under
+ the terms of the same license, but also under the terms of any GPL-Compatible
+ Free Software License, as listed by the Free Software Foundation.
+ This is a stricter copyleft license than the Affero GPL.
+ .
+ Client libraries (cluster_client and so on) are dual-licensed under the same
+ VNPL 1.1 and also GNU GPL 2.0 or later to allow for compatibility with GPLed
+ software like QEMU and fio.

debian/install

@@ -0,0 +1,3 @@
+VNPL-1.1.txt usr/share/doc/vitastor
+GPL-2.0.txt usr/share/doc/vitastor
+mon usr/lib/vitastor

debian/patched-qemu.Dockerfile

@@ -0,0 +1,44 @@
+# Build patched QEMU for Debian Buster or Bullseye/Sid inside a container
+# cd ..; podman build --build-arg REL=bullseye -v `pwd`/packages:/root/packages -f debian/patched-qemu.Dockerfile .
+
+FROM debian:$REL
+
+WORKDIR /root
+
+RUN if [ "$REL" = "buster" ]; then \
+        echo 'deb http://deb.debian.org/debian buster-backports main' >> /etc/apt/sources.list; \
+        echo >> /etc/apt/preferences; \
+        echo 'Package: *' >> /etc/apt/preferences; \
+        echo 'Pin: release a=buster-backports' >> /etc/apt/preferences; \
+        echo 'Pin-Priority: 500' >> /etc/apt/preferences; \
+    fi; \
+    grep '^deb ' /etc/apt/sources.list | perl -pe 's/^deb/deb-src/' >> /etc/apt/sources.list; \
+    echo 'APT::Install-Recommends false;' >> /etc/apt/apt.conf; \
+    echo 'APT::Install-Suggests false;' >> /etc/apt/apt.conf
+
+RUN apt-get update
+RUN apt-get -y install qemu fio liburing1 liburing-dev libgoogle-perftools-dev devscripts
+RUN apt-get -y build-dep qemu
+RUN apt-get -y build-dep fio
+RUN apt-get --download-only source qemu
+RUN apt-get --download-only source fio
+
+ADD qemu-5.0-vitastor.patch qemu-5.1-vitastor.patch /root/vitastor/
+RUN set -e; \
+    mkdir -p /root/packages/qemu-$REL; \
+    rm -rf /root/packages/qemu-$REL/*; \
+    cd /root/packages/qemu-$REL; \
+    dpkg-source -x /root/qemu*.dsc; \
+    if [ -d /root/packages/qemu-$REL/qemu-5.0 ]; then \
+        cp /root/vitastor/qemu-5.0-vitastor.patch /root/packages/qemu-$REL/qemu-5.0/debian/patches; \
+        echo qemu-5.0-vitastor.patch >> /root/packages/qemu-$REL/qemu-5.0/debian/patches/series; \
+    else \
+        cp /root/vitastor/qemu-5.1-vitastor.patch /root/packages/qemu-$REL/qemu-*/debian/patches; \
+        P=`ls -d /root/packages/qemu-$REL/qemu-*/debian/patches`; \
+        echo qemu-5.1-vitastor.patch >> $P/series; \
+    fi; \
+    cd /root/packages/qemu-$REL/qemu-*/; \
+    V=$(head -n1 debian/changelog | perl -pe 's/^.*\((.*?)(~bpo[\d\+]*)?\).*$/$1/')+vitastor1; \
+    DEBFULLNAME="Vitaliy Filippov <vitalif@yourcmc.ru>" dch -D $REL -v $V 'Plug Vitastor block driver'; \
+    DEB_BUILD_OPTIONS=nocheck dpkg-buildpackage --jobs=auto -sa; \
+    rm -rf /root/packages/qemu-$REL/qemu-*/

debian/rules

@@ -0,0 +1,9 @@
+#!/usr/bin/make -f
+export DH_VERBOSE = 1
+
+%:
+	dh $@
+
+override_dh_installdeb:
+	cat debian/substvars >> debian/vitastor.substvars
+	dh_installdeb

debian/source/format

@@ -0,0 +1 @@
+3.0 (quilt)

debian/substvars

@@ -0,0 +1,2 @@
+dep:fio=3.16-1
+dep:qemu=1:5.1+dfsg-4+vitastor1

debian/vitastor.Dockerfile

@@ -0,0 +1,67 @@
+# Build Vitastor packages for Debian Buster or Bullseye/Sid inside a container
+# cd ..; podman build --build-arg REL=bullseye -v `pwd`/packages:/root/packages -f debian/vitastor.Dockerfile .
+
+FROM debian:$REL
+
+WORKDIR /root
+
+RUN if [ "$REL" = "buster" ]; then \
+        echo 'deb http://deb.debian.org/debian buster-backports main' >> /etc/apt/sources.list; \
+        echo >> /etc/apt/preferences; \
+        echo 'Package: *' >> /etc/apt/preferences; \
+        echo 'Pin: release a=buster-backports' >> /etc/apt/preferences; \
+        echo 'Pin-Priority: 500' >> /etc/apt/preferences; \
+    fi; \
+    grep '^deb ' /etc/apt/sources.list | perl -pe 's/^deb/deb-src/' >> /etc/apt/sources.list; \
+    echo 'APT::Install-Recommends false;' >> /etc/apt/apt.conf; \
+    echo 'APT::Install-Suggests false;' >> /etc/apt/apt.conf
+
+RUN apt-get update
+RUN apt-get -y install qemu fio liburing1 liburing-dev libgoogle-perftools-dev devscripts
+RUN apt-get -y build-dep qemu
+RUN apt-get -y build-dep fio
+RUN apt-get --download-only source qemu
+RUN apt-get --download-only source fio
+RUN apt-get -y install libjerasure-dev cmake
+
+ADD . /root/vitastor
+RUN set -e -x; \
+    mkdir -p /root/fio-build/; \
+    cd /root/fio-build/; \
+    rm -rf /root/fio-build/*; \
+    dpkg-source -x /root/fio*.dsc; \
+    cd /root/packages/qemu-$REL/; \
+    rm -rf qemu*/; \
+    dpkg-source -x qemu*.dsc; \
+    cd /root/packages/qemu-$REL/qemu*/; \
+    debian/rules b/configure-stamp; \
+    cd b/qemu; \
+    make -j8 qapi/qapi-builtin-types.h; \
+    mkdir -p /root/packages/vitastor-$REL; \
+    rm -rf /root/packages/vitastor-$REL/*; \
+    cd /root/packages/vitastor-$REL; \
+    cp -r /root/vitastor vitastor-0.5.10; \
+    ln -s /root/packages/qemu-$REL/qemu-*/ vitastor-0.5.10/qemu; \
+    ln -s /root/fio-build/fio-*/ vitastor-0.5.10/fio; \
+    cd vitastor-0.5.10; \
+    FIO=$(head -n1 fio/debian/changelog | perl -pe 's/^.*\((.*?)\).*$/$1/'); \
+    QEMU=$(head -n1 qemu/debian/changelog | perl -pe 's/^.*\((.*?)\).*$/$1/'); \
+    sh copy-qemu-includes.sh; \
+    sh copy-fio-includes.sh; \
+    rm qemu fio; \
+    mkdir -p a b debian/patches; \
+    mv qemu-copy b/qemu; \
+    mv fio-copy b/fio; \
+    diff -NaurpbB a b > debian/patches/qemu-fio-headers.patch || true; \
+    echo qemu-fio-headers.patch >> debian/patches/series; \
+    rm -rf a b; \
+    rm -rf /root/packages/qemu-$REL/qemu*/; \
+    echo "dep:fio=$FIO" > debian/substvars; \
+    echo "dep:qemu=$QEMU" >> debian/substvars; \
+    cd /root/packages/vitastor-$REL; \
+    tar --sort=name --mtime='2020-01-01' --owner=0 --group=0 --exclude=debian -cJf vitastor_0.5.10.orig.tar.xz vitastor-0.5.10; \
+    cd vitastor-0.5.10; \
+    V=$(head -n1 debian/changelog | perl -pe 's/^.*\((.*?)\).*$/$1/'); \
+    DEBFULLNAME="Vitaliy Filippov <vitalif@yourcmc.ru>" dch -D $REL -v "$V""$REL" "Rebuild for $REL"; \
+    DEB_BUILD_OPTIONS=nocheck dpkg-buildpackage --jobs=auto -sa; \
+    rm -rf /root/packages/vitastor-$REL/vitastor-*/

json11

@@ -0,0 +1 @@
+Subproject commit 97f06cb20c1e136fd37d58fb40f57dd8f8a3a4a7

lambda_size.cpp

@@ -1,48 +0,0 @@
-#include <iostream>
-#include <functional>
-#include <array>
-#include <cstdlib> // for malloc() and free()
-using namespace std;
-
-// replace operator new and delete to log allocations
-void* operator new(std::size_t n)
-{
-    cout << "Allocating " << n << " bytes" << endl;
-    return malloc(n);
-}
-
-void operator delete(void* p) throw()
-{
-    free(p);
-}
-
-class test
-{
-public:
-    std::string s;
-    void a(std::function<void()> & f, const char *str)
-    {
-        auto l = [this, str]() { cout << str << " ? " << s << " from this\n"; };
-        cout << "Assigning lambda3 of size " << sizeof(l) << endl;
-        f = l;
-    }
-};
-
-int main()
-{
-    std::array<char, 16> arr1;
-    auto lambda1 = [arr1](){};
-    cout << "Assigning lambda1 of size " << sizeof(lambda1) << endl;
-    std::function<void()> f1 = lambda1;
-
-    std::array<char, 17> arr2;
-    auto lambda2 = [arr2](){};
-    cout << "Assigning lambda2 of size " << sizeof(lambda2) << endl;
-    std::function<void()> f2 = lambda2;
-
-    test t;
-    std::function<void()> f3;
-    t.s = "str";
-    t.a(f3, "huyambda");
-    f3();
-}

mon/PGUtil.js

@@ -0,0 +1,104 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+module.exports = {
+    scale_pg_count,
+};
+
+function add_pg_history(new_pg_history, new_pg, prev_pgs, prev_pg_history, old_pg)
+{
+    if (!new_pg_history[new_pg])
+    {
+        new_pg_history[new_pg] = {
+            osd_sets: {},
+            all_peers: {},
+            epoch: 0,
+        };
+    }
+    const nh = new_pg_history[new_pg], oh = prev_pg_history[old_pg];
+    nh.osd_sets[prev_pgs[old_pg].join(' ')] = prev_pgs[old_pg];
+    if (oh && oh.osd_sets && oh.osd_sets.length)
+    {
+        for (const pg of oh.osd_sets)
+        {
+            nh.osd_sets[pg.join(' ')] = pg;
+        }
+    }
+    if (oh && oh.all_peers && oh.all_peers.length)
+    {
+        for (const osd_num of oh.all_peers)
+        {
+            nh.all_peers[osd_num] = Number(osd_num);
+        }
+    }
+    if (oh && oh.epoch)
+    {
+        nh.epoch = nh.epoch < oh.epoch ? oh.epoch : nh.epoch;
+    }
+}
+
+function finish_pg_history(merged_history)
+{
+    merged_history.osd_sets = Object.values(merged_history.osd_sets);
+    merged_history.all_peers = Object.values(merged_history.all_peers);
+}
+
+function scale_pg_count(prev_pgs, prev_pg_history, new_pg_history, new_pg_count)
+{
+    const old_pg_count = prev_pgs.length;
+    // Add all possibly intersecting PGs to the history of new PGs
+    if (!(new_pg_count % old_pg_count))
+    {
+        // New PG count is a multiple of old PG count
+        for (let i = 0; i < new_pg_count; i++)
+        {
+            add_pg_history(new_pg_history, i, prev_pgs, prev_pg_history, i % old_pg_count);
+            finish_pg_history(new_pg_history[i]);
+        }
+    }
+    else if (!(old_pg_count % new_pg_count))
+    {
+        // Old PG count is a multiple of the new PG count
+        const mul = (old_pg_count / new_pg_count);
+        for (let i = 0; i < new_pg_count; i++)
+        {
+            for (let j = 0; j < mul; j++)
+            {
+                add_pg_history(new_pg_history, i, prev_pgs, prev_pg_history, i+j*new_pg_count);
+            }
+            finish_pg_history(new_pg_history[i]);
+        }
+    }
+    else
+    {
+        // Any PG may intersect with any PG after non-multiple PG count change
+        // So, merge ALL PGs history
+        let merged_history = {};
+        for (let i = 0; i < old_pg_count; i++)
+        {
+            add_pg_history(merged_history, 1, prev_pgs, prev_pg_history, i);
+        }
+        finish_pg_history(merged_history[1]);
+        for (let i = 0; i < new_pg_count; i++)
+        {
+            new_pg_history[i] = { ...merged_history[1] };
+        }
+    }
+    // Mark history keys for removed PGs as removed
+    for (let i = new_pg_count; i < old_pg_count; i++)
+    {
+        new_pg_history[i] = null;
+    }
+    // Just for the lp_solve optimizer - pick a "previous" PG for each "new" one
+    if (old_pg_count < new_pg_count)
+    {
+        for (let i = old_pg_count; i < new_pg_count; i++)
+        {
+            prev_pgs[i] = prev_pgs[i % old_pg_count];
+        }
+    }
+    else if (old_pg_count > new_pg_count)
+    {
+        prev_pgs.splice(new_pg_count, old_pg_count-new_pg_count);
+    }
+}

mon/afr.js

@@ -0,0 +1,89 @@
+// Functions to calculate Annualized Failure Rate of your cluster
+// if you know AFR of your drives, number of drives, expected rebalance time
+// and replication factor
+// License: VNPL-1.1 (see https://yourcmc.ru/git/vitalif/vitastor/src/branch/master/README.md for details) or AGPL-3.0
+// Author: Vitaliy Filippov, 2020+
+
+module.exports = {
+    cluster_afr_fullmesh,
+    failure_rate_fullmesh,
+    cluster_afr,
+    c_n_k,
+};
+
+/******** "FULL MESH": ASSUME EACH OSD COMMUNICATES WITH ALL OTHER OSDS ********/
+
+// Estimate AFR of the cluster
+// n - number of drives
+// afr - annualized failure rate of a single drive
+// l - expected rebalance time in days after a single drive failure
+// k - replication factor / number of drives that must fail at the same time for the cluster to fail
+function cluster_afr_fullmesh(n, afr, l, k)
+{
+    return 1 - (1 - afr * failure_rate_fullmesh(n-(k-1), afr*l/365, k-1)) ** (n-(k-1));
+}
+
+// Probability of at least <f> failures in a cluster with <n> drives with AFR=<a>
+function failure_rate_fullmesh(n, a, f)
+{
+    if (f <= 0)
+    {
+        return (1-a)**n;
+    }
+    let p = 1;
+    for (let i = 0; i < f; i++)
+    {
+        p -= c_n_k(n, i) * (1-a)**(n-i) * a**i;
+    }
+    return p;
+}
+
+/******** PGS: EACH OSD ONLY COMMUNICATES WITH <pgs> OTHER OSDs ********/
+
+// <n> hosts of <m> drives of <capacity> GB, each able to backfill at <speed> GB/s,
+// <k> replicas, <pgs> unique peer PGs per OSD (~50 for 100 PG-per-OSD in a big cluster)
+//
+// For each of n*m drives: P(drive fails in a year) * P(any of its peers fail in <l*365> next days).
+// More peers per OSD increase rebalance speed (more drives work together to resilver) if you
+// let them finish rebalance BEFORE replacing the failed drive (degraded_replacement=false).
+// At the same time, more peers per OSD increase probability of any of them to fail!
+// osd_rm=true means that failed OSDs' data is rebalanced over all other hosts,
+// not over the same host as it's in Ceph by default (dead OSDs are marked 'out').
+//
+// Probability of all except one drives in a replica group to fail is (AFR^(k-1)).
+// So with <x> PGs it becomes ~ (x * (AFR*L/365)^(k-1)). Interesting but reasonable consequence
+// is that, with k=2, total failure rate doesn't depend on number of peers per OSD,
+// because it gets increased linearly by increased number of peers to fail
+// and decreased linearly by reduced rebalance time.
+function cluster_afr({ n_hosts, n_drives, afr_drive, afr_host, capacity, speed, ec, ec_data, ec_parity, replicas, pgs = 1, osd_rm, degraded_replacement, down_out_interval = 600 })
+{
+    const pg_size = (ec ? ec_data+ec_parity : replicas);
+    pgs = Math.min(pgs, (n_hosts-1)*n_drives/(pg_size-1));
+    const host_pgs = Math.min(pgs*n_drives, (n_hosts-1)*n_drives/(pg_size-1));
+    const resilver_disk = n_drives == 1 || osd_rm ? pgs : (n_drives-1);
+    const disk_heal_time = (down_out_interval + capacity/(degraded_replacement ? 1 : resilver_disk)/speed)/86400/365;
+    const host_heal_time = (down_out_interval + n_drives*capacity/pgs/speed)/86400/365;
+    const disk_heal_fail = ((afr_drive+afr_host/n_drives)*disk_heal_time);
+    const host_heal_fail = ((afr_drive+afr_host/n_drives)*host_heal_time);
+    const disk_pg_fail = ec
+        ? failure_rate_fullmesh(ec_data+ec_parity-1, disk_heal_fail, ec_parity)
+        : disk_heal_fail**(replicas-1);
+    const host_pg_fail = ec
+        ? failure_rate_fullmesh(ec_data+ec_parity-1, host_heal_fail, ec_parity)
+        : host_heal_fail**(replicas-1);
+    return 1 - ((1 - afr_drive * (1-(1-disk_pg_fail)**pgs)) ** (n_hosts*n_drives))
+        * ((1 - afr_host * (1-(1-host_pg_fail)**host_pgs)) ** n_hosts);
+}
+
+/******** UTILITY ********/
+
+// Combination count
+function c_n_k(n, k)
+{
+    let r = 1;
+    for (let i = 0; i < k; i++)
+    {
+        r *= (n-i) / (i+1);
+    }
+    return r;
+}

mon/afr_test.js

@@ -0,0 +1,28 @@
+const { sprintf } = require('sprintf-js');
+const { cluster_afr } = require('./afr.js');
+
+print_cluster_afr({ n_hosts: 4, n_drives: 6, afr_drive: 0.03, afr_host: 0.05, capacity: 4000, speed: 0.1, replicas: 2 });
+print_cluster_afr({ n_hosts: 4, n_drives: 3, afr_drive: 0.03, afr_host: 0, capacity: 4000, speed: 0.1, replicas: 2 });
+print_cluster_afr({ n_hosts: 4, n_drives: 3, afr_drive: 0.03, afr_host: 0.05, capacity: 4000, speed: 0.1, replicas: 2 });
+print_cluster_afr({ n_hosts: 4, n_drives: 3, afr_drive: 0.03, afr_host: 0, capacity: 4000, speed: 0.1, ec: true, ec_data: 2, ec_parity: 1 });
+print_cluster_afr({ n_hosts: 4, n_drives: 3, afr_drive: 0.03, afr_host: 0.05, capacity: 4000, speed: 0.1, ec: true, ec_data: 2, ec_parity: 1 });
+print_cluster_afr({ n_hosts: 10, n_drives: 10, afr_drive: 0.1, afr_host: 0, capacity: 8000, speed: 0.02, replicas: 2 });
+print_cluster_afr({ n_hosts: 10, n_drives: 10, afr_drive: 0.1, afr_host: 0.05, capacity: 8000, speed: 0.02, replicas: 2 });
+print_cluster_afr({ n_hosts: 10, n_drives: 10, afr_drive: 0.1, afr_host: 0, capacity: 8000, speed: 0.02, replicas: 3 });
+print_cluster_afr({ n_hosts: 10, n_drives: 10, afr_drive: 0.1, afr_host: 0.05, capacity: 8000, speed: 0.02, replicas: 3 });
+print_cluster_afr({ n_hosts: 10, n_drives: 10, afr_drive: 0.1, afr_host: 0, capacity: 8000, speed: 0.02, replicas: 3, pgs: 100 });
+print_cluster_afr({ n_hosts: 10, n_drives: 10, afr_drive: 0.1, afr_host: 0.05, capacity: 8000, speed: 0.02, replicas: 3, pgs: 100 });
+print_cluster_afr({ n_hosts: 10, n_drives: 10, afr_drive: 0.1, afr_host: 0.05, capacity: 8000, speed: 0.02, replicas: 3, pgs: 100, degraded_replacement: 1 });
+
+function print_cluster_afr(config)
+{
+    console.log(
+        `${config.n_hosts} nodes with ${config.n_drives} ${sprintf("%.1f", config.capacity/1000)}TB drives`+
+        `, capable to backfill at ${sprintf("%.1f", config.speed*1000)} MB/s, drive AFR ${sprintf("%.1f", config.afr_drive*100)}%`+
+        (config.afr_host ? `, host AFR ${sprintf("%.1f", config.afr_host*100)}%` : '')+
+        (config.ec ? `, EC ${config.ec_data}+${config.ec_parity}` : `, ${config.replicas} replicas`)+
+        `, ${config.pgs||1} PG per OSD`+
+        (config.degraded_replacement ? `\n...and you don't let the rebalance finish before replacing drives` : '')
+    );
+    console.log('-> '+sprintf("%.7f%%", 100*cluster_afr(config))+'\n');
+}

mon/lp-optimizer.js

@@ -0,0 +1,716 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+// Data distribution optimizer using linear programming (lp_solve)
+
+const child_process = require('child_process');
+
+const NO_OSD = 'Z';
+
+async function lp_solve(text)
+{
+    const cp = child_process.spawn('lp_solve');
+    let stdout = '', stderr = '', finish_cb;
+    cp.stdout.on('data', buf => stdout += buf.toString());
+    cp.stderr.on('data', buf => stderr += buf.toString());
+    cp.on('exit', () => finish_cb && finish_cb());
+    cp.stdin.write(text);
+    cp.stdin.end();
+    if (cp.exitCode == null)
+    {
+        await new Promise(ok => finish_cb = ok);
+    }
+    if (!stdout.trim())
+    {
+        return null;
+    }
+    let score = 0;
+    let vars = {};
+    for (const line of stdout.split(/\n/))
+    {
+        let m = /^(^Value of objective function: (-?[\d\.]+)|Actual values of the variables:)\s*$/.exec(line);
+        if (m)
+        {
+            if (m[2])
+            {
+                score = m[2];
+            }
+            continue;
+        }
+        else if (/This problem is (infeasible|unbounded)/.exec(line))
+        {
+            return null;
+        }
+        let [ k, v ] = line.trim().split(/\s+/, 2);
+        if (v)
+        {
+            vars[k] = v;
+        }
+    }
+    return { score, vars };
+}
+
+async function optimize_initial({ osd_tree, pg_count, pg_size = 3, pg_minsize = 2, max_combinations = 10000, parity_space = 1 })
+{
+    if (!pg_count || !osd_tree)
+    {
+        return null;
+    }
+    const all_weights = Object.assign({}, ...Object.values(osd_tree));
+    const total_weight = Object.values(all_weights).reduce((a, c) => Number(a) + Number(c), 0);
+    const all_pgs = Object.values(random_combinations(osd_tree, pg_size, max_combinations, parity_space > 1));
+    const pg_per_osd = {};
+    for (const pg of all_pgs)
+    {
+        for (let i = 0; i < pg.length; i++)
+        {
+            const osd = pg[i];
+            pg_per_osd[osd] = pg_per_osd[osd] || [];
+            pg_per_osd[osd].push((i >= pg_minsize ? parity_space+'*' : '')+"pg_"+pg.join("_"));
+        }
+    }
+    const pg_effsize = Math.min(pg_minsize, Object.keys(osd_tree).length)
+        + Math.max(0, Math.min(pg_size, Object.keys(osd_tree).length) - pg_minsize) * parity_space;
+    let lp = '';
+    lp += "max: "+all_pgs.map(pg => 'pg_'+pg.join('_')).join(' + ')+";\n";
+    for (const osd in pg_per_osd)
+    {
+        if (osd !== NO_OSD)
+        {
+            let osd_pg_count = all_weights[osd]/total_weight*pg_effsize*pg_count;
+            lp += pg_per_osd[osd].join(' + ')+' <= '+osd_pg_count+';\n';
+        }
+    }
+    for (const pg of all_pgs)
+    {
+        lp += 'pg_'+pg.join('_')+" >= 0;\n";
+    }
+    lp += "sec "+all_pgs.map(pg => 'pg_'+pg.join('_')).join(', ')+";\n";
+    const lp_result = await lp_solve(lp);
+    if (!lp_result)
+    {
+        console.log(lp);
+        throw new Error('Problem is infeasible or unbounded - is it a bug?');
+    }
+    const int_pgs = make_int_pgs(lp_result.vars, pg_count);
+    const eff = pg_list_space_efficiency(int_pgs, all_weights, pg_minsize, parity_space);
+    const res = {
+        score: lp_result.score,
+        weights: lp_result.vars,
+        int_pgs,
+        space: eff * pg_effsize,
+        total_space: total_weight,
+    };
+    return res;
+}
+
+function make_int_pgs(weights, pg_count)
+{
+    const total_weight = Object.values(weights).reduce((a, c) => Number(a) + Number(c), 0);
+    let int_pgs = [];
+    let pg_left = pg_count;
+    let weight_left = total_weight;
+    for (const pg_name in weights)
+    {
+        let n = Math.round(weights[pg_name] / weight_left * pg_left);
+        for (let i = 0; i < n; i++)
+        {
+            int_pgs.push(pg_name.substr(3).split('_'));
+        }
+        weight_left -= weights[pg_name];
+        pg_left -= n;
+    }
+    return int_pgs;
+}
+
+function calc_intersect_weights(pg_size, pg_count, prev_weights, all_pgs)
+{
+    const move_weights = {};
+    if ((1 << pg_size) < pg_count)
+    {
+        const intersect = {};
+        for (const pg_name in prev_weights)
+        {
+            const pg = pg_name.substr(3).split(/_/);
+            for (let omit = 1; omit < (1 << pg_size); omit++)
+            {
+                let pg_omit = [ ...pg ];
+                let intersect_count = pg_size;
+                for (let i = 0; i < pg_size; i++)
+                {
+                    if (omit & (1 << i))
+                    {
+                        pg_omit[i] = '';
+                        intersect_count--;
+                    }
+                }
+                pg_omit = pg_omit.join(':');
+                intersect[pg_omit] = Math.max(intersect[pg_omit] || 0, intersect_count);
+            }
+        }
+        for (const pg of all_pgs)
+        {
+            let max_int = 0;
+            for (let omit = 1; omit < (1 << pg_size); omit++)
+            {
+                let pg_omit = [ ...pg ];
+                for (let i = 0; i < pg_size; i++)
+                {
+                    if (omit & (1 << i))
+                    {
+                        pg_omit[i] = '';
+                    }
+                }
+                pg_omit = pg_omit.join(':');
+                max_int = Math.max(max_int, intersect[pg_omit] || 0);
+            }
+            move_weights['pg_'+pg.join('_')] = pg_size-max_int;
+        }
+    }
+    else
+    {
+        const prev_pg_hashed = Object.keys(prev_weights).map(pg_name => pg_name.substr(3).split(/_/).reduce((a, c) => { a[c] = 1; return a; }, {}));
+        for (const pg of all_pgs)
+        {
+            if (!prev_weights['pg_'+pg.join('_')])
+            {
+                let max_int = 0;
+                for (const prev_hash in prev_pg_hashed)
+                {
+                    const intersect_count = pg.reduce((a, osd) => a + (prev_hash[osd] ? 1 : 0), 0);
+                    if (max_int < intersect_count)
+                    {
+                        max_int = intersect_count;
+                        if (max_int >= pg_size)
+                        {
+                            break;
+                        }
+                    }
+                }
+                move_weights['pg_'+pg.join('_')] = pg_size-max_int;
+            }
+        }
+    }
+    return move_weights;
+}
+
+function add_valid_previous(osd_tree, prev_weights, all_pgs)
+{
+    // Add previous combinations that are still valid
+    const hosts = Object.keys(osd_tree).sort();
+    const host_per_osd = {};
+    for (const host in osd_tree)
+    {
+        for (const osd in osd_tree[host])
+        {
+            host_per_osd[osd] = host;
+        }
+    }
+    skip_pg: for (const pg_name in prev_weights)
+    {
+        const seen_hosts = {};
+        const pg = pg_name.substr(3).split(/_/);
+        for (const osd of pg)
+        {
+            if (!host_per_osd[osd] || seen_hosts[host_per_osd[osd]])
+            {
+                continue skip_pg;
+            }
+            seen_hosts[host_per_osd[osd]] = true;
+        }
+        if (!all_pgs[pg_name])
+        {
+            all_pgs[pg_name] = pg;
+        }
+    }
+}
+
+// Try to minimize data movement
+async function optimize_change({ prev_pgs: prev_int_pgs, osd_tree, pg_size = 3, pg_minsize = 2, max_combinations = 10000, parity_space = 1 })
+{
+    if (!osd_tree)
+    {
+        return null;
+    }
+    const pg_effsize = Math.min(pg_minsize, Object.keys(osd_tree).length)
+        + Math.max(0, Math.min(pg_size, Object.keys(osd_tree).length) - pg_minsize) * parity_space;
+    const pg_count = prev_int_pgs.length;
+    const prev_weights = {};
+    const prev_pg_per_osd = {};
+    for (const pg of prev_int_pgs)
+    {
+        const pg_name = 'pg_'+pg.join('_');
+        prev_weights[pg_name] = (prev_weights[pg_name]||0) + 1;
+        for (let i = 0; i < pg.length; i++)
+        {
+            const osd = pg[i];
+            prev_pg_per_osd[osd] = prev_pg_per_osd[osd] || [];
+            prev_pg_per_osd[osd].push([ pg_name, (i >= pg_minsize ? parity_space : 1) ]);
+        }
+    }
+    // Get all combinations
+    let all_pgs = random_combinations(osd_tree, pg_size, max_combinations, parity_space > 1);
+    add_valid_previous(osd_tree, prev_weights, all_pgs);
+    all_pgs = Object.values(all_pgs);
+    const pg_per_osd = {};
+    for (const pg of all_pgs)
+    {
+        const pg_name = 'pg_'+pg.join('_');
+        for (let i = 0; i < pg.length; i++)
+        {
+            const osd = pg[i];
+            pg_per_osd[osd] = pg_per_osd[osd] || [];
+            pg_per_osd[osd].push([ pg_name, (i >= pg_minsize ? parity_space : 1) ]);
+        }
+    }
+    // Penalize PGs based on their similarity to old PGs
+    const move_weights = calc_intersect_weights(pg_size, pg_count, prev_weights, all_pgs);
+    // Calculate total weight - old PG weights
+    const all_pg_names = all_pgs.map(pg => 'pg_'+pg.join('_'));
+    const all_pgs_hash = all_pg_names.reduce((a, c) => { a[c] = true; return a; }, {});
+    const all_weights = Object.assign({}, ...Object.values(osd_tree));
+    const total_weight = Object.values(all_weights).reduce((a, c) => Number(a) + Number(c), 0);
+    // Generate the LP problem
+    let lp = '';
+    lp += 'max: '+all_pg_names.map(pg_name => (
+        prev_weights[pg_name] ? `${pg_size+1}*add_${pg_name} - ${pg_size+1}*del_${pg_name}` : `${pg_size+1-move_weights[pg_name]}*${pg_name}`
+    )).join(' + ')+';\n';
+    lp += all_pg_names
+        .map(pg_name => (prev_weights[pg_name] ? `add_${pg_name} - del_${pg_name}` : `${pg_name}`))
+        .join(' + ')+' = '+(pg_count
+            - Object.keys(prev_weights).reduce((a, old_pg_name) => (a + (all_pgs_hash[old_pg_name] ? prev_weights[old_pg_name] : 0)), 0)
+        )+';\n';
+    for (const osd in pg_per_osd)
+    {
+        if (osd !== NO_OSD)
+        {
+            const osd_sum = (pg_per_osd[osd]||[]).map(([ pg_name, space ]) => (
+                prev_weights[pg_name] ? `${space} * add_${pg_name} - ${space} * del_${pg_name}` : `${space} * ${pg_name}`
+            )).join(' + ');
+            const rm_osd_pg_count = (prev_pg_per_osd[osd]||[])
+                .reduce((a, [ old_pg_name, space ]) => (a + (all_pgs_hash[old_pg_name] ? space : 0)), 0);
+            const osd_pg_count = all_weights[osd]*pg_effsize/total_weight*pg_count - rm_osd_pg_count;
+            lp += osd_sum + ' <= ' + osd_pg_count + ';\n';
+        }
+    }
+    let pg_vars = [];
+    for (const pg_name of all_pg_names)
+    {
+        if (prev_weights[pg_name])
+        {
+            pg_vars.push(`add_${pg_name}`, `del_${pg_name}`);
+            // Can't add or remove less than zero
+            lp += `add_${pg_name} >= 0;\n`;
+            lp += `del_${pg_name} >= 0;\n`;
+            // Can't remove more than the PG already has
+            lp += `add_${pg_name} - del_${pg_name} >= -${prev_weights[pg_name]};\n`;
+        }
+        else
+        {
+            pg_vars.push(pg_name);
+            lp += `${pg_name} >= 0;\n`;
+        }
+    }
+    lp += 'sec '+pg_vars.join(', ')+';\n';
+    // Solve it
+    const lp_result = await lp_solve(lp);
+    if (!lp_result)
+    {
+        console.log(lp);
+        throw new Error('Problem is infeasible or unbounded - is it a bug?');
+    }
+    // Generate the new distribution
+    const weights = { ...prev_weights };
+    for (const k in prev_weights)
+    {
+        if (!all_pgs_hash[k])
+        {
+            delete weights[k];
+        }
+    }
+    for (const k in lp_result.vars)
+    {
+        if (k.substr(0, 4) === 'add_')
+        {
+            weights[k.substr(4)] = (weights[k.substr(4)] || 0) + Number(lp_result.vars[k]);
+        }
+        else if (k.substr(0, 4) === 'del_')
+        {
+            weights[k.substr(4)] = (weights[k.substr(4)] || 0) - Number(lp_result.vars[k]);
+        }
+        else if (k.substr(0, 3) === 'pg_')
+        {
+            weights[k] = Number(lp_result.vars[k]);
+        }
+    }
+    for (const k in weights)
+    {
+        if (!weights[k])
+        {
+            delete weights[k];
+        }
+    }
+    const int_pgs = make_int_pgs(weights, pg_count);
+    // Align them with most similar previous PGs
+    const new_pgs = align_pgs(prev_int_pgs, int_pgs);
+    let differs = 0, osd_differs = 0;
+    for (let i = 0; i < pg_count; i++)
+    {
+        if (new_pgs[i].join('_') != prev_int_pgs[i].join('_'))
+        {
+            differs++;
+        }
+        for (let j = 0; j < pg_size; j++)
+        {
+            if (new_pgs[i][j] != prev_int_pgs[i][j])
+            {
+                osd_differs++;
+            }
+        }
+    }
+    return {
+        prev_pgs: prev_int_pgs,
+        score: lp_result.score,
+        weights,
+        int_pgs: new_pgs,
+        differs,
+        osd_differs,
+        space: pg_effsize * pg_list_space_efficiency(new_pgs, all_weights, pg_minsize, parity_space),
+        total_space: total_weight,
+    };
+}
+
+function print_change_stats(retval, detailed)
+{
+    const new_pgs = retval.int_pgs;
+    const prev_int_pgs = retval.prev_pgs;
+    if (prev_int_pgs)
+    {
+        if (detailed)
+        {
+            for (let i = 0; i < new_pgs.length; i++)
+            {
+                if (new_pgs[i].join('_') != prev_int_pgs[i].join('_'))
+                {
+                    console.log("pg "+i+": "+prev_int_pgs[i].join(' ')+" -> "+new_pgs[i].join(' '));
+                }
+            }
+        }
+        console.log(
+            "Data movement: "+retval.differs+" pgs, "+
+            retval.osd_differs+" pg*osds = "+Math.round(retval.osd_differs / prev_int_pgs.length / 3 * 10000)/100+" %"
+        );
+    }
+    console.log(
+        "Total space (raw): "+Math.round(retval.space*100)/100+" TB, space efficiency: "+
+        Math.round(retval.space/(retval.total_space||1)*10000)/100+" %"
+    );
+}
+
+function align_pgs(prev_int_pgs, int_pgs)
+{
+    const aligned_pgs = [];
+    put_aligned_pgs(aligned_pgs, int_pgs, prev_int_pgs, (pg) => [ pg.join(':') ]);
+    put_aligned_pgs(aligned_pgs, int_pgs, prev_int_pgs, (pg) => [ pg[0]+'::'+pg[2], ':'+pg[1]+':'+pg[2], pg[0]+':'+pg[1]+':' ]);
+    put_aligned_pgs(aligned_pgs, int_pgs, prev_int_pgs, (pg) => [ pg[0]+'::', ':'+pg[1]+':', '::'+pg[2] ]);
+    const free_slots = prev_int_pgs.map((pg, i) => !aligned_pgs[i] ? i : null).filter(i => i != null);
+    for (const pg of int_pgs)
+    {
+        if (!free_slots.length)
+        {
+            throw new Error("Can't place unaligned PG");
+        }
+        aligned_pgs[free_slots.shift()] = pg;
+    }
+    return aligned_pgs;
+}
+
+function put_aligned_pgs(aligned_pgs, int_pgs, prev_int_pgs, keygen)
+{
+    let prev_indexes = {};
+    for (let i = 0; i < prev_int_pgs.length; i++)
+    {
+        for (let k of keygen(prev_int_pgs[i]))
+        {
+            prev_indexes[k] = prev_indexes[k] || [];
+            prev_indexes[k].push(i);
+        }
+    }
+    PG: for (let i = int_pgs.length-1; i >= 0; i--)
+    {
+        let pg = int_pgs[i];
+        let keys = keygen(int_pgs[i]);
+        for (let k of keys)
+        {
+            while (prev_indexes[k] && prev_indexes[k].length)
+            {
+                let idx = prev_indexes[k].shift();
+                if (!aligned_pgs[idx])
+                {
+                    aligned_pgs[idx] = pg;
+                    int_pgs.splice(i, 1);
+                    continue PG;
+                }
+            }
+        }
+    }
+}
+
+// Convert multi-level osd_tree = { level: number|string, id?: string, size?: number, children?: osd_tree }[]
+// levels = { string: number }
+// to a two-level osd_tree suitable for all_combinations()
+function flatten_tree(osd_tree, levels, failure_domain_level, osd_level, domains = {}, i = { i: 1 })
+{
+    osd_level = levels[osd_level] || osd_level;
+    failure_domain_level = levels[failure_domain_level] || failure_domain_level;
+    for (const node of osd_tree)
+    {
+        if ((levels[node.level] || node.level) < failure_domain_level)
+        {
+            flatten_tree(node.children||[], levels, failure_domain_level, osd_level, domains, i);
+        }
+        else
+        {
+            domains['dom'+(i.i++)] = extract_osds([ node ], levels, osd_level);
+        }
+    }
+    return domains;
+}
+
+function extract_osds(osd_tree, levels, osd_level, osds = {})
+{
+    for (const node of osd_tree)
+    {
+        if ((levels[node.level] || node.level) >= osd_level)
+        {
+            osds[node.id] = node.size;
+        }
+        else
+        {
+            extract_osds(node.children||[], levels, osd_level, osds);
+        }
+    }
+    return osds;
+}
+
+// ordered = don't treat (x,y) and (y,x) as equal
+function random_combinations(osd_tree, pg_size, count, ordered)
+{
+    let seed = 0x5f020e43;
+    let rng = () =>
+    {
+        seed ^= seed << 13;
+        seed ^= seed >> 17;
+        seed ^= seed << 5;
+        return seed + 2147483648;
+    };
+    const hosts = Object.keys(osd_tree).sort();
+    const osds = Object.keys(osd_tree).reduce((a, c) => { a[c] = Object.keys(osd_tree[c]).sort(); return a; }, {});
+    const r = {};
+    // Generate random combinations including each OSD at least once
+    for (let h = 0; h < hosts.length; h++)
+    {
+        for (let o = 0; o < osds[hosts[h]].length; o++)
+        {
+            const pg = [ osds[hosts[h]][o] ];
+            const cur_hosts = [ ...hosts ];
+            cur_hosts.splice(h, 1);
+            for (let i = 1; i < pg_size && i < hosts.length; i++)
+            {
+                const next_host = rng() % cur_hosts.length;
+                const next_osd = rng() % osds[cur_hosts[next_host]].length;
+                pg.push(osds[cur_hosts[next_host]][next_osd]);
+                cur_hosts.splice(next_host, 1);
+            }
+            const cyclic_pgs = [ pg ];
+            if (ordered)
+            {
+                for (let i = 1; i < pg.size; i++)
+                {
+                    cyclic_pgs.push([ ...pg.slice(i), ...pg.slice(0, i) ]);
+                }
+            }
+            for (const pg of cyclic_pgs)
+            {
+                while (pg.length < pg_size)
+                {
+                    pg.push(NO_OSD);
+                }
+                r['pg_'+pg.join('_')] = pg;
+            }
+        }
+    }
+    // Generate purely random combinations
+    while (count > 0)
+    {
+        let host_idx = [];
+        const cur_hosts = [ ...hosts.map((h, i) => i) ];
+        const max_hosts = pg_size < hosts.length ? pg_size : hosts.length;
+        if (ordered)
+        {
+            for (let i = 0; i < max_hosts; i++)
+            {
+                const r = rng() % cur_hosts.length;
+                host_idx[i] = cur_hosts[r];
+                cur_hosts.splice(r, 1);
+            }
+        }
+        else
+        {
+            for (let i = 0; i < max_hosts; i++)
+            {
+                const r = rng() % (cur_hosts.length - (max_hosts - i - 1));
+                host_idx[i] = cur_hosts[r];
+                cur_hosts.splice(0, r+1);
+            }
+        }
+        let pg = host_idx.map(h => osds[hosts[h]][rng() % osds[hosts[h]].length]);
+        while (pg.length < pg_size)
+        {
+            pg.push(NO_OSD);
+        }
+        r['pg_'+pg.join('_')] = pg;
+        count--;
+    }
+    return r;
+}
+
+// Super-stupid algorithm. Given the current OSD tree, generate all possible OSD combinations
+// osd_tree = { failure_domain1: { osd1: size1, ... }, ... }
+// ordered = return combinations without duplicates having different order
+function all_combinations(osd_tree, pg_size, ordered, count)
+{
+    const hosts = Object.keys(osd_tree).sort();
+    const osds = Object.keys(osd_tree).reduce((a, c) => { a[c] = Object.keys(osd_tree[c]).sort(); return a; }, {});
+    while (hosts.length < pg_size)
+    {
+        osds[NO_OSD] = [ NO_OSD ];
+        hosts.push(NO_OSD);
+    }
+    let host_idx = [];
+    let osd_idx = [];
+    for (let i = 0; i < pg_size; i++)
+    {
+        host_idx.push(i);
+        osd_idx.push(0);
+    }
+    const r = [];
+    while (!count || count < 0 || r.length < count)
+    {
+        r.push(host_idx.map((hi, i) => osds[hosts[hi]][osd_idx[i]]));
+        let inc = pg_size-1;
+        while (inc >= 0)
+        {
+            osd_idx[inc]++;
+            if (osd_idx[inc] >= osds[hosts[host_idx[inc]]].length)
+            {
+                osd_idx[inc] = 0;
+                inc--;
+            }
+            else
+            {
+                break;
+            }
+        }
+        if (inc < 0)
+        {
+            // no osds left in the current host combination, select the next one
+            inc = pg_size-1;
+            same_again: while (inc >= 0)
+            {
+                host_idx[inc]++;
+                for (let prev_host = 0; prev_host < inc; prev_host++)
+                {
+                    if (host_idx[prev_host] == host_idx[inc])
+                    {
+                        continue same_again;
+                    }
+                }
+                if (host_idx[inc] < (ordered ? hosts.length-(pg_size-1-inc) : hosts.length))
+                {
+                    while ((++inc) < pg_size)
+                    {
+                        host_idx[inc] = (ordered ? host_idx[inc-1]+1 : 0);
+                    }
+                    break;
+                }
+                else
+                {
+                    inc--;
+                }
+            }
+            if (inc < 0)
+            {
+                break;
+            }
+        }
+    }
+    return r;
+}
+
+function pg_weights_space_efficiency(weights, pg_count, osd_sizes)
+{
+    const per_osd = {};
+    for (const pg_name in weights)
+    {
+        for (const osd of pg_name.substr(3).split(/_/))
+        {
+            per_osd[osd] = (per_osd[osd]||0) + weights[pg_name];
+        }
+    }
+    return pg_per_osd_space_efficiency(per_osd, pg_count, osd_sizes);
+}
+
+function pg_list_space_efficiency(pgs, osd_sizes, pg_minsize, parity_space)
+{
+    const per_osd = {};
+    for (const pg of pgs)
+    {
+        for (let i = 0; i < pg.length; i++)
+        {
+            const osd = pg[i];
+            per_osd[osd] = (per_osd[osd]||0) + (i >= pg_minsize ? (parity_space||1) : 1);
+        }
+    }
+    return pg_per_osd_space_efficiency(per_osd, pgs.length, osd_sizes);
+}
+
+function pg_per_osd_space_efficiency(per_osd, pg_count, osd_sizes)
+{
+    // each PG gets randomly selected in 1/N cases
+    // & there are x PGs per OSD
+    // => an OSD is selected in x/N cases
+    // => total space * x/N <= OSD size
+    // => total space <= OSD size * N/x
+    let space;
+    for (let osd in per_osd)
+    {
+        if (osd in osd_sizes)
+        {
+            const space_estimate = osd_sizes[osd] * pg_count / per_osd[osd];
+            if (space == null || space > space_estimate)
+            {
+                space = space_estimate;
+            }
+        }
+    }
+    return space == null ? 0 : space;
+}
+
+module.exports = {
+    NO_OSD,
+
+    optimize_initial,
+    optimize_change,
+    print_change_stats,
+    pg_weights_space_efficiency,
+    pg_list_space_efficiency,
+    pg_per_osd_space_efficiency,
+    flatten_tree,
+
+    lp_solve,
+    make_int_pgs,
+    align_pgs,
+    random_combinations,
+    all_combinations,
+};

mon/make-osd.sh

@@ -0,0 +1,76 @@
+#!/bin/bash
+# Very simple systemd unit generator for vitastor-osd services
+# Not the final solution yet, mostly for tests
+# Copyright (c) Vitaliy Filippov, 2019+
+# License: MIT
+
+# USAGE: ./make-osd.sh /dev/disk/by-partuuid/xxx [ /dev/disk/by-partuuid/yyy]...
+
+IP_SUBSTR="10.200.1."
+ETCD_HOSTS="etcd0=http://10.200.1.10:2380,etcd1=http://10.200.1.11:2380,etcd2=http://10.200.1.12:2380"
+
+set -e -x
+
+IP=`ip -json a s | jq -r '.[].addr_info[] | select(.local | startswith("'$IP_SUBSTR'")) | .local'`
+[ "$IP" != "" ] || exit 1
+ETCD_MON=$(echo $ETCD_HOSTS | perl -pe 's/:2380/:2379/g; s/etcd\d*=//g;')
+D=`dirname $0`
+
+# Create OSDs on all passed devices
+OSD_NUM=1
+for DEV in $*; do
+
+# Ugly :) -> node.js rework pending
+while true; do
+    ST=$(etcdctl --endpoints="$ETCD_MON" get --print-value-only /vitastor/osd/stats/$OSD_NUM)
+    if [ "$ST" = "" ]; then
+        break
+    fi
+    OSD_NUM=$((OSD_NUM+1))
+done
+etcdctl --endpoints="$ETCD_MON" put /vitastor/osd/stats/$OSD_NUM '{}'
+
+echo Creating OSD $OSD_NUM on $DEV
+
+OPT=`node $D/simple-offsets.js --device $DEV --format options | tr '\n' ' '`
+META=`echo $OPT | grep -Po '(?<=data_offset )\d+'`
+dd if=/dev/zero of=$DEV bs=1048576 count=$(((META+1048575)/1048576)) oflag=direct
+
+cat >/etc/systemd/system/vitastor-osd$OSD_NUM.service <<EOF
+[Unit]
+Description=Vitastor object storage daemon osd.$OSD_NUM
+After=network-online.target local-fs.target time-sync.target
+Wants=network-online.target local-fs.target time-sync.target
+PartOf=vitastor.target
+
+[Service]
+LimitNOFILE=1048576
+LimitNPROC=1048576
+LimitMEMLOCK=infinity
+ExecStart=/usr/bin/vitastor-osd \\
+    --etcd_address $IP:2379/v3 \\
+    --bind_address $IP \\
+    --osd_num $OSD_NUM \\
+    --disable_data_fsync 1 \\
+    --immediate_commit all \\
+    --flusher_count 256 \\
+    --disk_alignment 4096 --journal_block_size 4096 --meta_block_size 4096 \\
+    --journal_no_same_sector_overwrites true \\
+    --journal_sector_buffer_count 1024 \\
+    $OPT
+WorkingDirectory=/
+ExecStartPre=+chown vitastor:vitastor $DEV
+User=vitastor
+PrivateTmp=false
+TasksMax=infinity
+Restart=always
+StartLimitInterval=0
+RestartSec=10
+
+[Install]
+WantedBy=vitastor.target
+EOF
+
+systemctl enable vitastor-osd$OSD_NUM
+
+done

mon/make-units.sh

@@ -0,0 +1,85 @@
+#!/bin/bash
+# Very simple systemd unit generator for etcd & vitastor-mon services
+# Not the final solution yet, mostly for tests
+# Copyright (c) Vitaliy Filippov, 2019+
+# License: MIT
+
+# USAGE: ./make-units.sh
+
+IP_SUBSTR="10.200.1."
+ETCD_HOSTS="etcd0=http://10.200.1.10:2380,etcd1=http://10.200.1.11:2380,etcd2=http://10.200.1.12:2380"
+
+# determine IP
+IP=`ip -json a s | jq -r '.[].addr_info[] | select(.local | startswith("'$IP_SUBSTR'")) | .local'`
+[ "$IP" != "" ] || exit 1
+ETCD_NUM=${ETCD_HOSTS/$IP*/}
+[ "$ETCD_NUM" != "$ETCD_HOSTS" ] || exit 1
+ETCD_NUM=$(echo $ETCD_NUM | tr -d -c , | wc -c)
+
+# etcd
+useradd etcd
+
+mkdir -p /var/lib/etcd$ETCD_NUM.etcd
+cat >/etc/systemd/system/etcd.service <<EOF
+[Unit]
+Description=etcd for vitastor
+After=network-online.target local-fs.target time-sync.target
+Wants=network-online.target local-fs.target time-sync.target
+
+[Service]
+Restart=always
+ExecStart=/usr/local/bin/etcd -name etcd$ETCD_NUM --data-dir /var/lib/etcd$ETCD_NUM.etcd \\
+    --advertise-client-urls http://$IP:2379 --listen-client-urls http://$IP:2379 \\
+    --initial-advertise-peer-urls http://$IP:2380 --listen-peer-urls http://$IP:2380 \\
+    --initial-cluster-token vitastor-etcd-1 --initial-cluster $ETCD_HOSTS \\
+    --initial-cluster-state new --max-txn-ops=100000 --auto-compaction-retention=10 --auto-compaction-mode=revision
+WorkingDirectory=/var/lib/etcd$ETCD_NUM.etcd
+ExecStartPre=+chown -R etcd /var/lib/etcd$ETCD_NUM.etcd
+User=etcd
+PrivateTmp=false
+TasksMax=infinity
+Restart=always
+StartLimitInterval=0
+RestartSec=10
+
+[Install]
+WantedBy=local.target
+EOF
+
+systemctl daemon-reload
+systemctl enable etcd
+systemctl start etcd
+
+useradd vitastor
+chmod 755 /root
+
+# Vitastor target
+cat >/etc/systemd/system/vitastor.target <<EOF
+[Unit]
+Description=vitastor target
+[Install]
+WantedBy=multi-user.target
+EOF
+
+# Monitor unit
+ETCD_MON=$(echo $ETCD_HOSTS | perl -pe 's/:2380/:2379/g; s/etcd\d*=//g;')
+cat >/etc/systemd/system/vitastor-mon.service <<EOF
+[Unit]
+Description=Vitastor monitor
+After=network-online.target local-fs.target time-sync.target
+Wants=network-online.target local-fs.target time-sync.target
+
+[Service]
+Restart=always
+ExecStart=node /usr/lib/vitastor/mon/mon-main.js --etcd_url '$ETCD_MON' --etcd_prefix '/vitastor' --etcd_start_timeout 5
+WorkingDirectory=/
+User=vitastor
+PrivateTmp=false
+TasksMax=infinity
+Restart=always
+StartLimitInterval=0
+RestartSec=10
+
+[Install]
+WantedBy=vitastor.target
+EOF

mon/merge.js

@@ -0,0 +1,23 @@
+const fsp = require('fs').promises;
+
+async function merge(file1, file2, out)
+{
+    if (!out)
+    {
+        console.error('USAGE: nodejs merge.js layer1 layer2 output');
+        process.exit();
+    }
+    const layer1 = await fsp.readFile(file1);
+    const layer2 = await fsp.readFile(file2);
+    const zero = Buffer.alloc(4096);
+    for (let i = 0; i < layer2.length; i += 4096)
+    {
+        if (zero.compare(layer2, i, i+4096) != 0)
+        {
+            layer2.copy(layer1, i, i, i+4096);
+        }
+    }
+    await fsp.writeFile(out, layer1);
+}
+
+merge(process.argv[2], process.argv[3], process.argv[4]);

mon/mon-main.js

@@ -0,0 +1,25 @@
+#!/usr/bin/node
+
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+const Mon = require('./mon.js');
+
+const options = {};
+
+for (let i = 2; i < process.argv.length; i++)
+{
+    if (process.argv[i].substr(0, 2) == '--')
+    {
+        options[process.argv[i].substr(2)] = process.argv[i+1];
+        i++;
+    }
+}
+
+if (!options.etcd_url)
+{
+    console.error('USAGE: '+process.argv[0]+' '+process.argv[1]+' --etcd_url "http://127.0.0.1:2379,..." --etcd_prefix "/vitastor" --etcd_start_timeout 5 [--verbose 1]');
+    process.exit();
+}
+
+new Mon(options).start().catch(e => { console.error(e); process.exit(); });

mon/package.json

@@ -0,0 +1,15 @@
+{
+  "name": "vitastor-mon",
+  "version": "1.0.0",
+  "description": "Vitastor SDS monitor service",
+  "main": "mon-main.js",
+  "scripts": {
+    "test": "echo \"Error: no test specified\" && exit 1"
+  },
+  "author": "Vitaliy Filippov",
+  "license": "UNLICENSED",
+  "dependencies": {
+    "sprintf-js": "^1.1.2",
+    "ws": "^7.2.5"
+  }
+}

mon/simple-offsets.js

@@ -0,0 +1,93 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: MIT
+
+// Simple tool to calculate journal and metadata offsets for a single device
+// Will be replaced by smarter tools in the future
+
+const fs = require('fs').promises;
+const child_process = require('child_process');
+
+async function run()
+{
+    const options = {
+        object_size: 128*1024,
+        bitmap_granularity: 4096,
+        journal_size: 16*1024*1024,
+        device_block_size: 4096,
+        journal_offset: 0,
+        device_size: 0,
+        format: 'text',
+    };
+    for (let i = 2; i < process.argv.length; i++)
+    {
+        if (process.argv[i].substr(0, 2) == '--')
+        {
+            options[process.argv[i].substr(2)] = process.argv[i+1];
+            i++;
+        }
+    }
+    if (!options.device)
+    {
+        process.stderr.write('USAGE: nodejs '+process.argv[1]+' --device /dev/sdXXX\n');
+        process.exit(1);
+    }
+    options.device_size = Number(options.device_size);
+    let device_size = options.device_size;
+    if (!device_size)
+    {
+        const st = await fs.stat(options.device);
+        options.device_block_size = st.blksize;
+        if (st.isBlockDevice())
+            device_size = Number(await system("/sbin/blockdev --getsize64 "+options.device))
+        else
+            device_size = st.size;
+    }
+    if (!device_size)
+    {
+        process.stderr.write('Failed to get device size\n');
+        process.exit(1);
+    }
+    options.journal_offset = Math.ceil(options.journal_offset/options.device_block_size)*options.device_block_size;
+    const meta_offset = options.journal_offset + Math.ceil(options.journal_size/options.device_block_size)*options.device_block_size;
+    const entries_per_block = Math.floor(options.device_block_size / (24 + 2*options.object_size/options.bitmap_granularity/8));
+    const object_count = Math.floor((device_size-meta_offset)/options.object_size);
+    const meta_size = Math.ceil(object_count / entries_per_block) * options.device_block_size;
+    const data_offset = meta_offset + meta_size;
+    const meta_size_fmt = (meta_size > 1024*1024*1024 ? Math.round(meta_size/1024/1024/1024*100)/100+" GB"
+        : Math.round(meta_size/1024/1024*100)/100+" MB");
+    if (options.format == 'text' || options.format == 'options')
+    {
+        if (options.format == 'text')
+        {
+            process.stderr.write(
+                `Metadata size: ${meta_size_fmt}\n`+
+                `Options for the OSD:\n`
+            );
+        }
+        process.stdout.write(
+            `    --data_device ${options.device}\n`+
+            `    --journal_offset ${options.journal_offset}\n`+
+            `    --meta_offset ${meta_offset}\n`+
+            `    --data_offset ${data_offset}\n`+
+            (options.device_size ? `    --data_size ${device_size-data_offset}\n` : '')
+        );
+    }
+    else if (options.format == 'env')
+    {
+        process.stdout.write(
+            `journal_offset=${options.journal_offset}\n`+
+            `meta_offset=${meta_offset}\n`+
+            `data_offset=${data_offset}\n`+
+            `data_size=${device_size-data_offset}\n`
+        );
+    }
+    else
+        process.stdout.write('Unknown format: '+options.format);
+}
+
+function system(cmd)
+{
+    return new Promise((ok, no) => child_process.exec(cmd, { maxBuffer: 64*1024*1024 }, (err, stdout, stderr) => (err ? no(err.message) : ok(stdout))));
+}
+
+run().catch(err => { console.error(err); process.exit(1); });

mon/stable-stringify.js

@@ -0,0 +1,78 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: MIT
+
+function stableStringify(obj, opts)
+{
+    if (!opts)
+        opts = {};
+    if (typeof opts === 'function')
+        opts = { cmp: opts };
+    let space = opts.space || '';
+    if (typeof space === 'number')
+        space = Array(space+1).join(' ');
+    const cycles = (typeof opts.cycles === 'boolean') ? opts.cycles : false;
+    const cmp = opts.cmp && (function (f)
+    {
+        return function (node)
+        {
+            return function (a, b)
+            {
+                let aobj = { key: a, value: node[a] };
+                let bobj = { key: b, value: node[b] };
+                return f(aobj, bobj);
+            };
+        };
+    })(opts.cmp);
+    const seen = new Map();
+    return (function stringify (parent, key, node, level)
+    {
+        const indent = space ? ('\n' + new Array(level + 1).join(space)) : '';
+        const colonSeparator = space ? ': ' : ':';
+        if (node === undefined)
+        {
+            return;
+        }
+        if (typeof node !== 'object' || node === null)
+        {
+            return JSON.stringify(node);
+        }
+        if (node instanceof Array)
+        {
+            const out = [];
+            for (let i = 0; i < node.length; i++)
+            {
+                const item = stringify(node, i, node[i], level+1) || JSON.stringify(null);
+                out.push(indent + space + item);
+            }
+            return '[' + out.join(',') + indent + ']';
+        }
+        else
+        {
+            if (seen.has(node))
+            {
+                if (cycles)
+                    return JSON.stringify('__cycle__');
+                throw new TypeError('Converting circular structure to JSON');
+            }
+            else
+                seen.set(node, true);
+            const keys = Object.keys(node).sort(cmp && cmp(node));
+            const out = [];
+            for (let i = 0; i < keys.length; i++)
+            {
+                const key = keys[i];
+                const value = stringify(node, key, node[key], level+1);
+                if (!value)
+                    continue;
+                const keyValue = JSON.stringify(key)
+                    + colonSeparator
+                    + value;
+                out.push(indent + space + keyValue);
+            }
+            seen.delete(node);
+            return '{' + out.join(',') + indent + '}';
+        }
+    })({ '': obj }, '', obj, 0);
+}
+
+module.exports = stableStringify;

mon/test-nonuniform.js

@@ -0,0 +1,130 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+// Interesting real-world example coming from Ceph with EC and compression enabled.
+// EC parity chunks can't be compressed as efficiently as data chunks,
+// thus they occupy more space (2.26x more space) in OSD object stores.
+// This leads to really uneven OSD fill ratio in Ceph even when PGs are perfectly balanced.
+// But we support this case with the "parity_space" parameter in optimize_initial()/optimize_change().
+
+const LPOptimizer = require('./lp-optimizer.js');
+
+const osd_tree = {
+    ripper5: {
+        osd0: 3.493144989013672,
+        osd1: 3.493144989013672,
+        osd2: 3.454082489013672,
+        osd12: 3.461894989013672,
+    },
+    ripper7: {
+        osd4: 3.638690948486328,
+        osd5: 3.638690948486328,
+        osd6: 3.638690948486328,
+    },
+    ripper4: {
+        osd9: 3.4609375,
+        osd10: 3.4609375,
+        osd11: 3.4609375,
+    },
+    ripper6: {
+        osd3: 3.5849609375,
+        osd7: 3.5859336853027344,
+        osd8: 3.638690948486328,
+        osd13: 3.461894989013672
+    },
+};
+
+const prev_pgs = [[12,7,5],[6,11,12],[3,6,9],[10,0,5],[2,5,13],[9,8,6],[3,4,12],[7,4,12],[12,11,13],[13,6,0],[4,13,10],[9,7,6],[7,10,0],[10,8,0],[3,10,2],[3,0,4],[6,13,0],[13,10,0],[13,10,5],[8,11,6],[3,9,2],[2,8,5],[8,9,5],[3,12,11],[0,7,4],[13,11,1],[11,3,12],[12,8,10],[7,5,12],[2,13,5],[7,11,0],[13,2,6],[0,6,8],[13,1,6],[0,13,4],[0,8,10],[4,10,0],[8,12,4],[8,12,9],[12,7,4],[13,9,5],[3,2,11],[1,9,7],[1,8,5],[5,12,9],[3,5,12],[2,8,10],[0,8,4],[1,4,11],[7,10,2],[12,13,5],[3,1,11],[7,1,4],[4,12,8],[7,0,9],[11,1,8],[3,0,5],[11,13,0],[1,13,5],[12,7,10],[12,8,4],[11,13,5],[0,11,6],[2,11,3],[13,1,11],[2,7,10],[7,10,12],[7,12,10],[12,11,5],[13,12,10],[2,3,9],[4,3,9],[13,2,5],[7,12,6],[12,10,13],[9,8,1],[13,1,5],[9,5,12],[5,11,7],[6,2,9],[8,11,6],[12,5,8],[6,13,1],[7,6,11],[2,3,6],[8,5,9],[1,13,6],[9,3,2],[7,11,1],[3,10,1],[0,11,7],[3,0,5],[1,3,6],[6,0,9],[3,11,4],[8,10,2],[13,1,9],[12,6,9],[3,12,9],[12,8,9],[7,5,0],[8,12,5],[0,11,3],[12,11,13],[0,7,11],[0,3,10],[1,3,11],[2,7,11],[13,2,6],[9,12,13],[8,2,4],[0,7,4],[5,13,0],[13,12,9],[1,9,8],[0,10,3],[3,5,10],[7,12,9],[2,13,4],[12,7,5],[9,2,7],[3,2,9],[6,2,7],[3,1,9],[4,3,2],[5,3,11],[0,7,6],[1,6,13],[7,10,2],[12,4,8],[13,12,6],[7,5,11],[6,2,3],[2,7,6],[2,3,10],[2,7,10],[11,12,6],[0,13,5],[10,2,4],[13,0,11],[7,0,6],[8,9,4],[8,4,11],[7,11,2],[3,4,2],[6,1,3],[7,2,11],[8,9,4],[11,4,8],[10,3,1],[2,10,13],[1,7,11],[13,11,12],[2,6,9],[10,0,13],[7,10,4],[0,11,13],[13,10,1],[7,5,0],[7,12,10],[3,1,4],[7,1,5],[3,11,5],[7,5,0],[1,3,5],[10,5,12],[0,3,9],[7,1,11],[11,8,12],[3,6,2],[7,12,9],[7,11,12],[4,11,3],[0,11,13],[13,2,5],[1,5,8],[0,11,8],[3,5,1],[11,0,6],[3,11,2],[11,8,12],[4,1,3],[10,13,4],[13,9,6],[2,3,10],[12,7,9],[10,0,4],[10,13,2],[3,11,1],[7,2,9],[1,7,4],[13,1,4],[7,0,6],[5,3,9],[10,0,7],[0,7,10],[3,6,10],[13,0,5],[8,4,1],[3,1,10],[2,10,13],[13,0,5],[13,10,2],[12,7,9],[6,8,10],[6,1,8],[10,8,1],[13,5,0],[5,11,3],[7,6,1],[8,5,9],[2,13,11],[10,12,4],[13,4,1],[2,13,4],[11,7,0],[2,9,7],[1,7,6],[8,0,4],[8,1,9],[7,10,12],[13,9,6],[7,6,11],[13,0,4],[1,8,4],[3,12,5],[10,3,1],[10,2,13],[2,4,8],[6,2,3],[3,0,10],[6,7,12],[8,12,5],[3,0,6],[13,12,10],[11,3,6],[9,0,13],[10,0,6],[7,5,2],[1,3,11],[7,10,2],[2,9,8],[11,13,12],[0,8,4],[8,12,11],[6,0,3],[1,13,4],[11,8,2],[12,3,6],[4,7,1],[7,6,12],[3,10,6],[0,10,7],[8,9,1],[0,10,6],[8,10,1]]
+    .map(pg => pg.map(n => 'osd'+n));
+
+const by_osd = {};
+
+for (let i = 0; i < prev_pgs.length; i++)
+{
+    for (let j = 0; j < prev_pgs[i].length; j++)
+    {
+        by_osd[prev_pgs[i][j]] = by_osd[prev_pgs[i][j]] || [];
+        by_osd[prev_pgs[i][j]][j] = (by_osd[prev_pgs[i][j]][j] || 0) + 1;
+    }
+}
+
+/*
+
+This set of PGs was balanced by hand, by heavily tuning OSD weights in Ceph:
+
+{
+  osd0: 4.2,
+  osd1: 3.5,
+  osd2: 3.45409,
+  osd3: 4.5,
+  osd4: 1.4,
+  osd5: 1.4,
+  osd6: 1.75,
+  osd7: 4.5,
+  osd8: 4.4,
+  osd9: 2.2,
+  osd10: 2.7,
+  osd11: 2,
+  osd12: 3.4,
+  osd13: 3.4,
+}
+
+EC+compression is a nightmare in Ceph, yeah :))
+
+To calculate the average ratio between data chunks and parity chunks we
+calculate the number of PG chunks for each chunk role for each OSD:
+
+{
+  osd12: [ 18, 22, 17 ],
+  osd7: [ 35, 22, 8 ],
+  osd5: [ 6, 17, 27 ],
+  osd6: [ 13, 12, 28 ],
+  osd11: [ 13, 26, 20 ],
+  osd3: [ 30, 20, 10 ],
+  osd9: [ 8, 12, 26 ],
+  osd10: [ 15, 23, 20 ],
+  osd0: [ 22, 22, 14 ],
+  osd2: [ 22, 16, 16 ],
+  osd13: [ 29, 19, 13 ],
+  osd8: [ 20, 18, 12 ],
+  osd4: [ 8, 10, 28 ],
+  osd1: [ 17, 17, 17 ]
+}
+
+And now we can pick a pair of OSDs and determine the ratio by solving the following:
+
+osd5 = 23*X + 27*Y = 3249728140
+osd13 = 48*X + 13*Y = 2991675992
+
+=>
+
+osd5 - 27/13*osd13 = 23*X - 27/13*48*X = -76.6923076923077*X = -2963752766.46154
+
+=>
+
+X = 38644720.1243731
+Y = (osd5-23*X)/27 = 87440725.0792377
+Y/X = 2.26268232239284 ~= 2.26
+
+Which means that parity chunks are compressed ~2.26 times worse than data chunks.
+
+Fine, let's try to optimize for it.
+
+*/
+
+async function run()
+{
+    const all_weights = Object.assign({}, ...Object.values(osd_tree));
+    const total_weight = Object.values(all_weights).reduce((a, c) => Number(a) + Number(c), 0);
+    const eff = LPOptimizer.pg_list_space_efficiency(prev_pgs, all_weights, 2, 2.26);
+    const orig = eff*4.26 / total_weight;
+    console.log('Original efficiency was: '+Math.round(orig*10000)/100+' %');
+
+    let prev = await LPOptimizer.optimize_initial({ osd_tree, pg_size: 3, pg_count: 256, parity_space: 2.26 });
+    LPOptimizer.print_change_stats(prev);
+
+    let next = await LPOptimizer.optimize_change({ prev_pgs, osd_tree, pg_size: 3, max_combinations: 10000, parity_space: 2.26 });
+    LPOptimizer.print_change_stats(next);
+}
+
+run().catch(console.error);

mon/test-optimize-simple.js

@@ -0,0 +1,25 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+const LPOptimizer = require('./lp-optimizer.js');
+
+async function run()
+{
+    const osd_tree = { a: { 1: 1 }, b: { 2: 1 }, c: { 3: 1 } };
+    let res;
+
+    console.log('16 PGs, size=3');
+    res = await LPOptimizer.optimize_initial({ osd_tree, pg_size: 3, pg_count: 16 });
+    LPOptimizer.print_change_stats(res, false);
+
+    console.log('\nReduce PG size to 2');
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs.map(pg => pg.slice(0, 2)), osd_tree, pg_size: 2 });
+    LPOptimizer.print_change_stats(res, false);
+
+    console.log('\nRemove OSD 3');
+    delete osd_tree['c'];
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree, pg_size: 2 });
+    LPOptimizer.print_change_stats(res, false);
+}
+
+run().catch(console.error);

mon/test-optimize-undersized.js

@@ -0,0 +1,74 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+const LPOptimizer = require('./lp-optimizer.js');
+
+const crush_tree = [
+    { level: 1, children: [
+        { level: 2, children: [
+            { level: 3, id: 1, size: 3 },
+            { level: 3, id: 2, size: 3 },
+        ] },
+        { level: 2, children: [
+            { level: 3, id: 3, size: 3 },
+            { level: 3, id: 4, size: 3 },
+        ] },
+    ] },
+    { level: 1, children: [
+        { level: 2, children: [
+            { level: 3, id: 5, size: 3 },
+            { level: 3, id: 6, size: 3 },
+        ] },
+        { level: 2, children: [
+            { level: 3, id: 7, size: 3 },
+            { level: 3, id: 8, size: 3 },
+        ] },
+    ] },
+    { level: 1, children: [
+        { level: 2, children: [
+            { level: 3, id: 9, size: 3 },
+            { level: 3, id: 10, size: 3 },
+        ] },
+        { level: 2, children: [
+            { level: 3, id: 11, size: 3 },
+            { level: 3, id: 12, size: 3 },
+        ] },
+    ] },
+];
+
+const osd_tree = LPOptimizer.flatten_tree(crush_tree, {}, 1, 3);
+console.log(osd_tree);
+
+async function run()
+{
+    const cur_tree = {};
+    console.log('Empty tree:');
+    let res = await LPOptimizer.optimize_initial({ osd_tree: cur_tree, pg_size: 3, pg_count: 256 });
+    LPOptimizer.print_change_stats(res, false);
+    console.log('\nAdding 1st failure domain:');
+    cur_tree['dom1'] = osd_tree['dom1'];
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree: cur_tree, pg_size: 3 });
+    LPOptimizer.print_change_stats(res, false);
+    console.log('\nAdding 2nd failure domain:');
+    cur_tree['dom2'] = osd_tree['dom2'];
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree: cur_tree, pg_size: 3 });
+    LPOptimizer.print_change_stats(res, false);
+    console.log('\nAdding 3rd failure domain:');
+    cur_tree['dom3'] = osd_tree['dom3'];
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree: cur_tree, pg_size: 3 });
+    LPOptimizer.print_change_stats(res, false);
+    console.log('\nRemoving 3rd failure domain:');
+    delete cur_tree['dom3'];
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree: cur_tree, pg_size: 3 });
+    LPOptimizer.print_change_stats(res, false);
+    console.log('\nRemoving 2nd failure domain:');
+    delete cur_tree['dom2'];
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree: cur_tree, pg_size: 3 });
+    LPOptimizer.print_change_stats(res, false);
+    console.log('\nRemoving 1st failure domain:');
+    delete cur_tree['dom1'];
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree: cur_tree, pg_size: 3 });
+    LPOptimizer.print_change_stats(res, false);
+}
+
+run().catch(console.error);

mon/test-optimize.js

@@ -0,0 +1,115 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+const LPOptimizer = require('./lp-optimizer.js');
+
+const osd_tree = {
+    100: {
+        7: 3.63869,
+    },
+    300: {
+        10: 3.46089,
+        11: 3.46089,
+        12: 3.46089,
+    },
+    400: {
+        1: 3.49309,
+        2: 3.49309,
+        3: 3.49309,
+    },
+    500: {
+        4: 3.58498,
+//        8: 3.58589,
+        9: 3.63869,
+    },
+    600: {
+        5: 3.63869,
+        6: 3.63869,
+    },
+/*    100: {
+        1: 2.72800,
+    },
+    200: {
+        2: 2.72900,
+    },
+    300: {
+        3: 1.87000,
+    },
+    400: {
+        4: 1.87000,
+    },
+    500: {
+        5: 3.63869,
+    },*/
+};
+
+const crush_tree = [
+    { level: 1, children: [
+        { level: 2, children: [
+            { level: 3, id: 1, size: 3 },
+            { level: 3, id: 2, size: 2 },
+        ] },
+        { level: 2, children: [
+            { level: 3, id: 3, size: 4 },
+            { level: 3, id: 4, size: 4 },
+        ] },
+    ] },
+    { level: 1, children: [
+        { level: 2, children: [
+            { level: 3, id: 5, size: 4 },
+            { level: 3, id: 6, size: 1 },
+        ] },
+        { level: 2, children: [
+            { level: 3, id: 7, size: 3 },
+            { level: 3, id: 8, size: 5 },
+        ] },
+    ] },
+    { level: 1, children: [
+        { level: 2, children: [
+            { level: 3, id: 9, size: 5 },
+            { level: 3, id: 10, size: 2 },
+        ] },
+        { level: 2, children: [
+            { level: 3, id: 11, size: 3 },
+            { level: 3, id: 12, size: 3 },
+        ] },
+    ] },
+];
+
+async function run()
+{
+    let res;
+
+    // Test: add 1 OSD of almost the same size. Ideal data movement could be 1/12 = 8.33%. Actual is ~13%
+    // Space efficiency is ~99% in all cases.
+
+    console.log('256 PGs, size=2');
+    res = await LPOptimizer.optimize_initial({ osd_tree, pg_size: 2, pg_count: 256 });
+    LPOptimizer.print_change_stats(res, false);
+    console.log('\nAdding osd.8');
+    osd_tree[500][8] = 3.58589;
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree, pg_size: 2 });
+    LPOptimizer.print_change_stats(res, false);
+    console.log('\nRemoving osd.8');
+    delete osd_tree[500][8];
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree, pg_size: 2 });
+    LPOptimizer.print_change_stats(res, false);
+
+    console.log('\n256 PGs, size=3');
+    res = await LPOptimizer.optimize_initial({ osd_tree, pg_size: 3, pg_count: 256 });
+    LPOptimizer.print_change_stats(res, false);
+    console.log('\nAdding osd.8');
+    osd_tree[500][8] = 3.58589;
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree, pg_size: 3 });
+    LPOptimizer.print_change_stats(res, false);
+    console.log('\nRemoving osd.8');
+    delete osd_tree[500][8];
+    res = await LPOptimizer.optimize_change({ prev_pgs: res.int_pgs, osd_tree, pg_size: 3 });
+    LPOptimizer.print_change_stats(res, false);
+
+    console.log('\n256 PGs, size=3, failure domain=rack');
+    res = await LPOptimizer.optimize_initial({ osd_tree: LPOptimizer.flatten_tree(crush_tree, {}, 1, 3), pg_size: 3, pg_count: 256 });
+    LPOptimizer.print_change_stats(res, false);
+}
+
+run().catch(console.error);

osd.cpp

@@ -1,401 +0,0 @@
-#include <sys/socket.h>
-#include <sys/epoll.h>
-#include <sys/poll.h>
-#include <netinet/in.h>
-#include <netinet/tcp.h>
-#include <arpa/inet.h>
-
-#include "osd.h"
-
-static const char* osd_op_names[] = {
-    "",
-    "read",
-    "write",
-    "sync",
-    "stabilize",
-    "rollback",
-    "delete",
-    "sync_stab_all",
-    "list",
-    "show_config",
-    "primary_read",
-    "primary_write",
-    "primary_sync",
-};
-
-osd_t::osd_t(blockstore_config_t & config, blockstore_t *bs, ring_loop_t *ringloop)
-{
-    this->config = config;
-    this->bs = bs;
-    this->ringloop = ringloop;
-    this->tick_tfd = new timerfd_interval(ringloop, 3, [this]()
-    {
-        for (int i = 0; i <= OSD_OP_MAX; i++)
-        {
-            if (op_stat_count[i] != 0)
-            {
-                printf("avg latency for op %d (%s): %ld us\n", i, osd_op_names[i], op_stat_sum[i]/op_stat_count[i]);
-                op_stat_count[i] = 0;
-                op_stat_sum[i] = 0;
-            }
-        }
-        for (int i = 0; i <= OSD_OP_MAX; i++)
-        {
-            if (subop_stat_count[i] != 0)
-            {
-                printf("avg latency for subop %d (%s): %ld us\n", i, osd_op_names[i], subop_stat_sum[i]/subop_stat_count[i]);
-                subop_stat_count[i] = 0;
-                subop_stat_sum[i] = 0;
-            }
-        }
-        if (send_stat_count != 0)
-        {
-            printf("avg latency to send stabilize subop: %ld us\n", send_stat_sum/send_stat_count);
-            send_stat_count = 0;
-            send_stat_sum = 0;
-        }
-    });
-    this->bs_block_size = bs->get_block_size();
-    // FIXME: use bitmap granularity instead
-    this->bs_disk_alignment = bs->get_disk_alignment();
-
-    bind_address = config["bind_address"];
-    if (bind_address == "")
-        bind_address = "0.0.0.0";
-    bind_port = strtoull(config["bind_port"].c_str(), NULL, 10);
-    if (!bind_port || bind_port > 65535)
-        bind_port = 11203;
-    osd_num = strtoull(config["osd_num"].c_str(), NULL, 10);
-    if (!osd_num)
-        throw std::runtime_error("osd_num is required in the configuration");
-    run_primary = config["run_primary"] == "true" || config["run_primary"] == "1" || config["run_primary"] == "yes";
-    if (run_primary)
-        init_primary();
-
-    listen_fd = socket(AF_INET, SOCK_STREAM, 0);
-    if (listen_fd < 0)
-    {
-        throw std::runtime_error(std::string("socket: ") + strerror(errno));
-    }
-    int enable = 1;
-    setsockopt(listen_fd, SOL_SOCKET, SO_REUSEADDR, &enable, sizeof(enable));
-
-    sockaddr_in addr;
-    int r;
-    if ((r = inet_pton(AF_INET, bind_address.c_str(), &addr.sin_addr)) != 1)
-    {
-        close(listen_fd);
-        throw std::runtime_error("bind address "+bind_address+(r == 0 ? " is not valid" : ": no ipv4 support"));
-    }
-    addr.sin_family = AF_INET;
-    addr.sin_port = htons(bind_port);
-
-    if (bind(listen_fd, (sockaddr*)&addr, sizeof(addr)) < 0)
-    {
-        close(listen_fd);
-        throw std::runtime_error(std::string("bind: ") + strerror(errno));
-    }
-
-    if (listen(listen_fd, listen_backlog) < 0)
-    {
-        close(listen_fd);
-        throw std::runtime_error(std::string("listen: ") + strerror(errno));
-    }
-
-    fcntl(listen_fd, F_SETFL, fcntl(listen_fd, F_GETFL, 0) | O_NONBLOCK);
-
-    epoll_fd = epoll_create(1);
-    if (epoll_fd < 0)
-    {
-        close(listen_fd);
-        throw std::runtime_error(std::string("epoll_create: ") + strerror(errno));
-    }
-
-    epoll_event ev;
-    ev.data.fd = listen_fd;
-    ev.events = EPOLLIN | EPOLLET;
-    if (epoll_ctl(epoll_fd, EPOLL_CTL_ADD, listen_fd, &ev) < 0)
-    {
-        close(listen_fd);
-        close(epoll_fd);
-        throw std::runtime_error(std::string("epoll_ctl: ") + strerror(errno));
-    }
-
-    consumer.loop = [this]() { loop(); };
-    ringloop->register_consumer(consumer);
-}
-
-osd_t::~osd_t()
-{
-    delete tick_tfd;
-    ringloop->unregister_consumer(consumer);
-    close(epoll_fd);
-    close(listen_fd);
-}
-
-osd_op_t::~osd_op_t()
-{
-    if (bs_op)
-    {
-        delete bs_op;
-    }
-    if (op_data)
-    {
-        free(op_data);
-    }
-    if (rmw_buf)
-    {
-        free(rmw_buf);
-    }
-    if (buf)
-    {
-        // Note: reusing osd_op_t WILL currently lead to memory leaks
-        // So we don't reuse it, but free it every time
-        free(buf);
-    }
-}
-
-bool osd_t::shutdown()
-{
-    stopping = true;
-    if (inflight_ops > 0)
-    {
-        return false;
-    }
-    return bs->is_safe_to_stop();
-}
-
-void osd_t::loop()
-{
-    if (!wait_state)
-    {
-        handle_epoll_events();
-        wait_state = 1;
-    }
-    handle_peers();
-    read_requests();
-    send_replies();
-    ringloop->submit();
-}
-
-void osd_t::handle_epoll_events()
-{
-    io_uring_sqe *sqe = ringloop->get_sqe();
-    if (!sqe)
-    {
-        throw std::runtime_error("can't get SQE, will fall out of sync with EPOLLET");
-    }
-    ring_data_t *data = ((ring_data_t*)sqe->user_data);
-    my_uring_prep_poll_add(sqe, epoll_fd, POLLIN);
-    data->callback = [this](ring_data_t *data)
-    {
-        if (data->res < 0)
-        {
-            throw std::runtime_error(std::string("epoll failed: ") + strerror(-data->res));
-        }
-        handle_epoll_events();
-    };
-    ringloop->submit();
-    int nfds;
-    epoll_event events[MAX_EPOLL_EVENTS];
-restart:
-    nfds = epoll_wait(epoll_fd, events, MAX_EPOLL_EVENTS, 0);
-    for (int i = 0; i < nfds; i++)
-    {
-        if (events[i].data.fd == listen_fd)
-        {
-            // Accept new connections
-            sockaddr_in addr;
-            socklen_t peer_addr_size = sizeof(addr);
-            int peer_fd;
-            while ((peer_fd = accept(listen_fd, (sockaddr*)&addr, &peer_addr_size)) >= 0)
-            {
-                char peer_str[256];
-                printf("osd: new client %d: connection from %s port %d\n", peer_fd, inet_ntop(AF_INET, &addr.sin_addr, peer_str, 256), ntohs(addr.sin_port));
-                fcntl(peer_fd, F_SETFL, fcntl(listen_fd, F_GETFL, 0) | O_NONBLOCK);
-                int one = 1;
-                setsockopt(peer_fd, SOL_TCP, TCP_NODELAY, &one, sizeof(one));
-                clients[peer_fd] = {
-                    .peer_addr = addr,
-                    .peer_port = ntohs(addr.sin_port),
-                    .peer_fd = peer_fd,
-                    .peer_state = PEER_CONNECTED,
-                };
-                // Add FD to epoll
-                epoll_event ev;
-                ev.data.fd = peer_fd;
-                ev.events = EPOLLIN | EPOLLET | EPOLLRDHUP;
-                if (epoll_ctl(epoll_fd, EPOLL_CTL_ADD, peer_fd, &ev) < 0)
-                {
-                    throw std::runtime_error(std::string("epoll_ctl: ") + strerror(errno));
-                }
-                // Try to accept next connection
-                peer_addr_size = sizeof(addr);
-            }
-            if (peer_fd == -1 && errno != EAGAIN)
-            {
-                throw std::runtime_error(std::string("accept: ") + strerror(errno));
-            }
-        }
-        else
-        {
-            auto & cl = clients[events[i].data.fd];
-            if (cl.peer_state == PEER_CONNECTING)
-            {
-                // Either OUT (connected) or HUP
-                handle_connect_result(cl.peer_fd);
-            }
-            else if (events[i].events & EPOLLRDHUP)
-            {
-                // Stop client
-                printf("osd: client %d disconnected\n", cl.peer_fd);
-                stop_client(cl.peer_fd);
-            }
-            else
-            {
-                // Mark client as ready (i.e. some data is available)
-                cl.read_ready++;
-                if (cl.read_ready == 1)
-                {
-                    read_ready_clients.push_back(cl.peer_fd);
-                    ringloop->wakeup();
-                }
-            }
-        }
-    }
-    if (nfds == MAX_EPOLL_EVENTS)
-    {
-        goto restart;
-    }
-}
-
-void osd_t::cancel_osd_ops(osd_client_t & cl)
-{
-    for (auto p: cl.sent_ops)
-    {
-        cancel_op(p.second);
-    }
-    cl.sent_ops.clear();
-    for (auto op: cl.outbox)
-    {
-        cancel_op(op);
-    }
-    cl.outbox.clear();
-    if (cl.write_op)
-    {
-        cancel_op(cl.write_op);
-        cl.write_op = NULL;
-    }
-}
-
-void osd_t::cancel_op(osd_op_t *op)
-{
-    if (op->op_type == OSD_OP_OUT)
-    {
-        op->reply.hdr.magic = SECONDARY_OSD_REPLY_MAGIC;
-        op->reply.hdr.id = op->req.hdr.id;
-        op->reply.hdr.opcode = op->req.hdr.opcode;
-        op->reply.hdr.retval = -EPIPE;
-        op->callback(op);
-    }
-    else
-    {
-        delete op;
-    }
-}
-
-void osd_t::stop_client(int peer_fd)
-{
-    auto it = clients.find(peer_fd);
-    if (it == clients.end())
-    {
-        return;
-    }
-    auto & cl = it->second;
-    if (epoll_ctl(epoll_fd, EPOLL_CTL_DEL, peer_fd, NULL) < 0)
-    {
-        throw std::runtime_error(std::string("epoll_ctl: ") + strerror(errno));
-    }
-    if (cl.osd_num)
-    {
-        // Cancel outbound operations
-        cancel_osd_ops(cl);
-        osd_peer_fds.erase(cl.osd_num);
-        repeer_pgs(cl.osd_num, false);
-        peering_state |= OSD_PEERING_PEERS;
-    }
-    if (cl.read_op)
-    {
-        delete cl.read_op;
-    }
-    for (auto rit = read_ready_clients.begin(); rit != read_ready_clients.end(); rit++)
-    {
-        if (*rit == peer_fd)
-        {
-            read_ready_clients.erase(rit);
-            break;
-        }
-    }
-    for (auto wit = write_ready_clients.begin(); wit != write_ready_clients.end(); wit++)
-    {
-        if (*wit == peer_fd)
-        {
-            write_ready_clients.erase(wit);
-            break;
-        }
-    }
-    clients.erase(it);
-    close(peer_fd);
-}
-
-void osd_t::exec_op(osd_op_t *cur_op)
-{
-    clock_gettime(CLOCK_REALTIME, &cur_op->tv_begin);
-    if (stopping)
-    {
-        // Throw operation away
-        delete cur_op;
-        return;
-    }
-    cur_op->send_list.push_back(cur_op->reply.buf, OSD_PACKET_SIZE);
-    if (cur_op->req.hdr.magic != SECONDARY_OSD_OP_MAGIC ||
-        cur_op->req.hdr.opcode < OSD_OP_MIN || cur_op->req.hdr.opcode > OSD_OP_MAX ||
-        (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_READ || cur_op->req.hdr.opcode == OSD_OP_SECONDARY_WRITE) &&
-        (cur_op->req.sec_rw.len > OSD_RW_MAX || cur_op->req.sec_rw.len % OSD_RW_ALIGN || cur_op->req.sec_rw.offset % OSD_RW_ALIGN) ||
-        (cur_op->req.hdr.opcode == OSD_OP_READ || cur_op->req.hdr.opcode == OSD_OP_WRITE) &&
-        (cur_op->req.rw.len > OSD_RW_MAX || cur_op->req.rw.len % OSD_RW_ALIGN || cur_op->req.rw.offset % OSD_RW_ALIGN))
-    {
-        // Bad command
-        cur_op->reply.hdr.magic = SECONDARY_OSD_REPLY_MAGIC;
-        cur_op->reply.hdr.id = cur_op->req.hdr.id;
-        cur_op->reply.hdr.opcode = cur_op->req.hdr.opcode;
-        cur_op->reply.hdr.retval = -EINVAL;
-        outbox_push(this->clients[cur_op->peer_fd], cur_op);
-        return;
-    }
-    inflight_ops++;
-    if (cur_op->req.hdr.opcode == OSD_OP_TEST_SYNC_STAB_ALL)
-    {
-        exec_sync_stab_all(cur_op);
-    }
-    else if (cur_op->req.hdr.opcode == OSD_OP_SHOW_CONFIG)
-    {
-        exec_show_config(cur_op);
-    }
-    else if (cur_op->req.hdr.opcode == OSD_OP_READ)
-    {
-        continue_primary_read(cur_op);
-    }
-    else if (cur_op->req.hdr.opcode == OSD_OP_WRITE)
-    {
-        continue_primary_write(cur_op);
-    }
-    else if (cur_op->req.hdr.opcode == OSD_OP_SYNC)
-    {
-        continue_primary_sync(cur_op);
-    }
-    else
-    {
-        exec_secondary(cur_op);
-    }
-}

osd.h

@@ -1,274 +0,0 @@
-#pragma once
-
-#include <sys/types.h>
-#include <time.h>
-#include <sys/ioctl.h>
-#include <sys/stat.h>
-#include <fcntl.h>
-#include <unistd.h>
-#include <malloc.h>
-#include <arpa/inet.h>
-#include <malloc.h>
-
-#include <set>
-#include <deque>
-
-#include "blockstore.h"
-#include "ringloop.h"
-#include "timerfd_interval.h"
-#include "osd_ops.h"
-#include "osd_peering_pg.h"
-
-#include "sparsepp/sparsepp/spp.h"
-
-#define OSD_OP_IN 0
-#define OSD_OP_OUT 1
-
-#define CL_READ_OP 1
-#define CL_READ_DATA 2
-#define CL_READ_REPLY_DATA 3
-#define CL_WRITE_READY 1
-#define CL_WRITE_REPLY 2
-#define MAX_EPOLL_EVENTS 64
-#define OSD_OP_INLINE_BUF_COUNT 16
-
-#define PEER_CONNECTING 1
-#define PEER_CONNECTED 2
-#define OSD_PEERING_PEERS 1
-#define OSD_PEERING_PGS 2
-
-//#define OSD_STUB
-
-struct osd_op_buf_list_t
-{
-    int count = 0, alloc = 0, sent = 0;
-    iovec *buf = NULL;
-    iovec inline_buf[OSD_OP_INLINE_BUF_COUNT];
-
-    ~osd_op_buf_list_t()
-    {
-        if (buf && buf != inline_buf)
-        {
-            free(buf);
-        }
-    }
-
-    inline iovec* get_iovec()
-    {
-        return (buf ? buf : inline_buf) + sent;
-    }
-
-    inline int get_size()
-    {
-        return count - sent;
-    }
-
-    inline void push_back(void *nbuf, size_t len)
-    {
-        if (count >= alloc)
-        {
-            if (!alloc)
-            {
-                alloc = OSD_OP_INLINE_BUF_COUNT;
-                buf = inline_buf;
-            }
-            else if (buf == inline_buf)
-            {
-                int old = alloc;
-                alloc = ((alloc/16)*16 + 1);
-                buf = (iovec*)malloc(sizeof(iovec) * alloc);
-                memcpy(buf, inline_buf, sizeof(iovec)*old);
-            }
-            else
-            {
-                alloc = ((alloc/16)*16 + 1);
-                buf = (iovec*)realloc(buf, sizeof(iovec) * alloc);
-            }
-        }
-        buf[count++] = { .iov_base = nbuf, .iov_len = len };
-    }
-};
-
-struct osd_primary_op_data_t;
-
-struct osd_op_t
-{
-    timespec tv_begin;
-    timespec tv_send;
-    int op_type = OSD_OP_IN;
-    int peer_fd;
-    osd_any_op_t req;
-    osd_any_reply_t reply;
-    blockstore_op_t *bs_op = NULL;
-    void *buf = NULL;
-    void *rmw_buf = NULL;
-    osd_primary_op_data_t* op_data = NULL;
-    std::function<void(osd_op_t*)> callback;
-
-    osd_op_buf_list_t send_list;
-
-    ~osd_op_t();
-};
-
-struct osd_peer_def_t
-{
-    osd_num_t osd_num = 0;
-    std::string addr;
-    int port = 0;
-    time_t last_connect_attempt = 0;
-};
-
-struct osd_client_t
-{
-    sockaddr_in peer_addr;
-    int peer_port;
-    int peer_fd;
-    int peer_state;
-    std::function<void(osd_num_t, int)> connect_callback;
-    osd_num_t osd_num = 0;
-
-    // Read state
-    int read_ready = 0;
-    osd_op_t *read_op = NULL;
-    int read_reply_id = 0;
-    iovec read_iov;
-    msghdr read_msg;
-    void *read_buf = NULL;
-    int read_remaining = 0;
-    int read_state = 0;
-
-    // Outbound operations sent to this client (which is probably an OSD peer)
-    std::map<int, osd_op_t*> sent_ops;
-
-    // Outbound messages (replies or requests)
-    std::deque<osd_op_t*> outbox;
-
-    // PGs dirtied by this client's primary-writes
-    std::set<pg_num_t> dirty_pgs;
-
-    // Write state
-    osd_op_t *write_op = NULL;
-    msghdr write_msg;
-    int write_state = 0;
-};
-
-struct osd_rmw_stripe_t;
-
-struct osd_object_id_t
-{
-    osd_num_t osd_num;
-    object_id oid;
-};
-
-class osd_t
-{
-    // config
-
-    osd_num_t osd_num = 1; // OSD numbers start with 1
-    bool run_primary = false;
-    std::vector<osd_peer_def_t> peers;
-    blockstore_config_t config;
-    std::string bind_address;
-    int bind_port, listen_backlog;
-    int client_queue_depth = 128;
-    bool allow_test_ops = true;
-
-    // peer OSDs
-
-    std::map<uint64_t, int> osd_peer_fds;
-    std::vector<pg_t> pgs;
-    int peering_state = 0;
-    unsigned pg_count = 0;
-    uint64_t next_subop_id = 1;
-
-    // Unstable writes
-    std::map<osd_object_id_t, uint64_t> unstable_writes;
-    std::deque<osd_op_t*> syncs_in_progress;
-
-    // client & peer I/O
-
-    bool stopping = false;
-    int inflight_ops = 0;
-    blockstore_t *bs;
-    uint32_t bs_block_size, bs_disk_alignment;
-    uint64_t parity_block_size = 4*1024*1024; // 4 MB by default
-    ring_loop_t *ringloop;
-    timerfd_interval *tick_tfd;
-
-    int wait_state = 0;
-    int epoll_fd = 0;
-    int listen_fd = 0;
-    ring_consumer_t consumer;
-
-    std::unordered_map<int,osd_client_t> clients;
-    std::vector<int> read_ready_clients;
-    std::vector<int> write_ready_clients;
-    uint64_t op_stat_sum[OSD_OP_MAX+1] = { 0 };
-    uint64_t op_stat_count[OSD_OP_MAX+1] = { 0 };
-    uint64_t subop_stat_sum[OSD_OP_MAX+1] = { 0 };
-    uint64_t subop_stat_count[OSD_OP_MAX+1] = { 0 };
-    uint64_t send_stat_sum = 0;
-    uint64_t send_stat_count = 0;
-
-    // methods
-
-    // event loop, socket read/write
-    void loop();
-    void handle_epoll_events();
-    void read_requests();
-    void handle_read(ring_data_t *data, int peer_fd);
-    void handle_op_hdr(osd_client_t *cl);
-    void handle_reply_hdr(osd_client_t *cl);
-    bool try_send(osd_client_t & cl);
-    void send_replies();
-    void handle_send(ring_data_t *data, int peer_fd);
-    void outbox_push(osd_client_t & cl, osd_op_t *op);
-
-    // peer handling (primary OSD logic)
-    void connect_peer(osd_num_t osd_num, const char *peer_host, int peer_port, std::function<void(osd_num_t, int)> callback);
-    void handle_connect_result(int peer_fd);
-    void cancel_osd_ops(osd_client_t & cl);
-    void cancel_op(osd_op_t *op);
-    void stop_client(int peer_fd);
-    osd_peer_def_t parse_peer(std::string peer);
-    void init_primary();
-    void handle_peers();
-    void repeer_pgs(osd_num_t osd_num, bool is_connected);
-    void start_pg_peering(int i);
-
-    // op execution
-    void exec_op(osd_op_t *cur_op);
-
-    // secondary ops
-    void exec_sync_stab_all(osd_op_t *cur_op);
-    void exec_show_config(osd_op_t *cur_op);
-    void exec_secondary(osd_op_t *cur_op);
-    void secondary_op_callback(osd_op_t *cur_op);
-
-    // primary ops
-    bool prepare_primary_rw(osd_op_t *cur_op);
-    void continue_primary_read(osd_op_t *cur_op);
-    void continue_primary_write(osd_op_t *cur_op);
-    void continue_primary_sync(osd_op_t *cur_op);
-    void finish_primary_op(osd_op_t *cur_op, int retval);
-    void handle_primary_subop(osd_op_t *cur_op, int ok, uint64_t version);
-    void submit_primary_subops(int submit_type, int read_pg_size, const uint64_t* osd_set, osd_op_t *cur_op);
-    void submit_primary_sync_subops(osd_op_t *cur_op);
-    void submit_primary_stab_subops(osd_op_t *cur_op);
-public:
-    osd_t(blockstore_config_t & config, blockstore_t *bs, ring_loop_t *ringloop);
-    ~osd_t();
-    bool shutdown();
-};
-
-inline bool operator == (const osd_object_id_t & a, const osd_object_id_t & b)
-{
-    return a.osd_num == b.osd_num && a.oid.inode == b.oid.inode && a.oid.stripe == b.oid.stripe;
-}
-
-inline bool operator < (const osd_object_id_t & a, const osd_object_id_t & b)
-{
-    return a.osd_num < b.osd_num || a.osd_num == b.osd_num && (
-        a.oid.inode < b.oid.inode || a.oid.inode == b.oid.inode && a.oid.stripe < b.oid.stripe
-    );
-}

osd_client.cpp

@@ -1,40 +0,0 @@
-void slice()
-{
-    // Slice the request into blockstore requests to individual objects
-    // Primary OSD still operates individual stripes, except they're twice the size of the blockstore's stripe.
-    std::vector read_parts;
-    int block = bs->get_block_size();
-    uint64_t stripe1 = cur_op->req.rw.offset / block / 2;
-    uint64_t stripe2 = (cur_op->req.rw.offset + cur_op->req.rw.len + block*2 - 1) / block / 2 - 1;
-    for (uint64_t s = stripe1; s <= stripe2; s++)
-    {
-        uint64_t start = s == stripe1 ? cur_op->req.rw.offset - stripe1*block*2 : 0;
-        uint64_t end = s == stripe2 ? cur_op->req.rw.offset + cur_op->req.rw.len - stripe2*block*2 : block*2;
-        if (start < block)
-        {
-            read_parts.push_back({
-                .role = 1,
-                .oid = {
-                    .inode = cur_op->req.rw.inode,
-                    .stripe = (s << STRIPE_ROLE_BITS) | 1,
-                },
-                .version = UINT64_MAX,
-                .offset = start,
-                .len = (block < end ? block : end) - start,
-            });
-        }
-        if (end > block)
-        {
-            read_parts.push_back({
-                .role = 2,
-                .oid = {
-                    .inode = cur_op->req.rw.inode,
-                    .stripe = (s << STRIPE_ROLE_BITS) | 2,
-                },
-                .version = UINT64_MAX,
-                .offset = (start > block ? start-block : 0),
-                .len = end - (start > block ? start-block : 0),
-            });
-        }
-    }
-}

osd_id.h

@@ -1,4 +0,0 @@
-#pragma once
-
-typedef uint64_t osd_num_t;
-typedef uint32_t pg_num_t;

osd_peering.cpp

@@ -1,406 +0,0 @@
-#include <netinet/tcp.h>
-#include <sys/epoll.h>
-
-#include <algorithm>
-
-#include "osd.h"
-
-void osd_t::init_primary()
-{
-    // Initial test version of clustering code requires exactly 2 peers
-    // FIXME Hardcode
-    std::string peerstr = config["peers"];
-    while (peerstr.size())
-    {
-        int pos = peerstr.find(',');
-        peers.push_back(parse_peer(pos < 0 ? peerstr : peerstr.substr(0, pos)));
-        peerstr = pos < 0 ? std::string("") : peerstr.substr(pos+1);
-        for (int i = 0; i < peers.size()-1; i++)
-            if (peers[i].osd_num == peers[peers.size()-1].osd_num)
-                throw std::runtime_error("same osd number "+std::to_string(peers[i].osd_num)+" specified twice in peers");
-    }
-    if (peers.size() < 2)
-        throw std::runtime_error("run_primary requires at least 2 peers");
-    pgs.push_back((pg_t){
-        .state = PG_OFFLINE,
-        .pg_cursize = 0,
-        .pg_num = 1,
-        .target_set = { 1, 2, 3 },
-        .cur_set = { 1, 0, 0 },
-    });
-    pg_count = 1;
-    peering_state = OSD_PEERING_PEERS;
-}
-
-osd_peer_def_t osd_t::parse_peer(std::string peer)
-{
-    // OSD_NUM:IP:PORT
-    int pos1 = peer.find(':');
-    int pos2 = peer.find(':', pos1+1);
-    if (pos1 < 0 || pos2 < 0)
-        throw new std::runtime_error("OSD peer string must be in the form OSD_NUM:IP:PORT");
-    osd_peer_def_t r;
-    r.addr = peer.substr(pos1+1, pos2-pos1-1);
-    std::string osd_num_str = peer.substr(0, pos1);
-    std::string port_str = peer.substr(pos2+1);
-    r.osd_num = strtoull(osd_num_str.c_str(), NULL, 10);
-    if (!r.osd_num)
-        throw new std::runtime_error("Could not parse OSD peer osd_num");
-    r.port = strtoull(port_str.c_str(), NULL, 10);
-    if (!r.port)
-        throw new std::runtime_error("Could not parse OSD peer port");
-    return r;
-}
-
-void osd_t::connect_peer(osd_num_t osd_num, const char *peer_host, int peer_port, std::function<void(osd_num_t, int)> callback)
-{
-    struct sockaddr_in addr;
-    int r;
-    if ((r = inet_pton(AF_INET, peer_host, &addr.sin_addr)) != 1)
-    {
-        callback(osd_num, -EINVAL);
-        return;
-    }
-    addr.sin_family = AF_INET;
-    addr.sin_port = htons(peer_port ? peer_port : 11203);
-    int peer_fd = socket(AF_INET, SOCK_STREAM, 0);
-    if (peer_fd < 0)
-    {
-        callback(osd_num, -errno);
-        return;
-    }
-    fcntl(peer_fd, F_SETFL, fcntl(peer_fd, F_GETFL, 0) | O_NONBLOCK);
-    r = connect(peer_fd, (sockaddr*)&addr, sizeof(addr));
-    if (r < 0 && errno != EINPROGRESS)
-    {
-        close(peer_fd);
-        callback(osd_num, -errno);
-        return;
-    }
-    clients[peer_fd] = (osd_client_t){
-        .peer_addr = addr,
-        .peer_port = peer_port,
-        .peer_fd = peer_fd,
-        .peer_state = PEER_CONNECTING,
-        .connect_callback = callback,
-        .osd_num = osd_num,
-    };
-    osd_peer_fds[osd_num] = peer_fd;
-    // Add FD to epoll (EPOLLOUT for tracking connect() result)
-    epoll_event ev;
-    ev.data.fd = peer_fd;
-    ev.events = EPOLLOUT | EPOLLIN | EPOLLRDHUP | EPOLLET;
-    if (epoll_ctl(epoll_fd, EPOLL_CTL_ADD, peer_fd, &ev) < 0)
-    {
-        throw std::runtime_error(std::string("epoll_ctl: ") + strerror(errno));
-    }
-}
-
-void osd_t::handle_connect_result(int peer_fd)
-{
-    auto & cl = clients[peer_fd];
-    osd_num_t osd_num = cl.osd_num;
-    auto callback = cl.connect_callback;
-    int result = 0;
-    socklen_t result_len = sizeof(result);
-    if (getsockopt(peer_fd, SOL_SOCKET, SO_ERROR, &result, &result_len) < 0)
-    {
-        result = errno;
-    }
-    if (result != 0)
-    {
-        stop_client(peer_fd);
-        callback(osd_num, -result);
-        return;
-    }
-    int one = 1;
-    setsockopt(peer_fd, SOL_TCP, TCP_NODELAY, &one, sizeof(one));
-    // Disable EPOLLOUT on this fd
-    cl.connect_callback = NULL;
-    cl.peer_state = PEER_CONNECTED;
-    epoll_event ev;
-    ev.data.fd = peer_fd;
-    ev.events = EPOLLIN | EPOLLRDHUP | EPOLLET;
-    if (epoll_ctl(epoll_fd, EPOLL_CTL_MOD, peer_fd, &ev) < 0)
-    {
-        throw std::runtime_error(std::string("epoll_ctl: ") + strerror(errno));
-    }
-    callback(osd_num, peer_fd);
-}
-
-// Peering loop
-void osd_t::handle_peers()
-{
-    if (peering_state & OSD_PEERING_PEERS)
-    {
-        for (int i = 0; i < peers.size(); i++)
-        {
-            if (osd_peer_fds.find(peers[i].osd_num) == osd_peer_fds.end() &&
-                time(NULL) - peers[i].last_connect_attempt > 5) // FIXME hardcode 5
-            {
-                peers[i].last_connect_attempt = time(NULL);
-                connect_peer(peers[i].osd_num, peers[i].addr.c_str(), peers[i].port, [this](osd_num_t osd_num, int peer_fd)
-                {
-                    // FIXME: Check peer config after connecting
-                    if (peer_fd < 0)
-                    {
-                        printf("Failed to connect to peer OSD %lu: %s\n", osd_num, strerror(-peer_fd));
-                        return;
-                    }
-                    printf("Connected with peer OSD %lu (fd %d)\n", clients[peer_fd].osd_num, peer_fd);
-                    int i;
-                    for (i = 0; i < peers.size(); i++)
-                    {
-                        if (osd_peer_fds.find(peers[i].osd_num) == osd_peer_fds.end())
-                            break;
-                    }
-                    if (i >= peers.size())
-                    {
-                        // Connected to all peers
-                        peering_state = peering_state & ~OSD_PEERING_PEERS;
-                    }
-                    repeer_pgs(osd_num, true);
-                });
-            }
-        }
-    }
-    if (peering_state & OSD_PEERING_PGS)
-    {
-        bool still_doing_pgs = false;
-        for (int i = 0; i < pgs.size(); i++)
-        {
-            if (pgs[i].state == PG_PEERING)
-            {
-                if (!pgs[i].peering_state->list_ops.size())
-                {
-                    pgs[i].calc_object_states();
-                }
-                else
-                {
-                    still_doing_pgs = true;
-                }
-            }
-        }
-        if (!still_doing_pgs)
-        {
-            // Done all PGs
-            peering_state = peering_state & ~OSD_PEERING_PGS;
-        }
-    }
-}
-
-void osd_t::repeer_pgs(osd_num_t osd_num, bool is_connected)
-{
-    // Re-peer affected PGs
-    // FIXME: We shouldn't rely just on target_set. Other OSDs may also contain PG data.
-    osd_num_t real_osd = (is_connected ? osd_num : 0);
-    for (int i = 0; i < pgs.size(); i++)
-    {
-        bool repeer = false;
-        for (int r = 0; r < pgs[i].target_set.size(); r++)
-        {
-            if (pgs[i].target_set[r] == osd_num &&
-                pgs[i].cur_set[r] != real_osd)
-            {
-                pgs[i].cur_set[r] = real_osd;
-                repeer = true;
-                break;
-            }
-        }
-        if (repeer)
-        {
-            // Repeer this pg
-            printf("Repeer PG %d because of OSD %lu\n", i, osd_num);
-            start_pg_peering(i);
-            peering_state |= OSD_PEERING_PGS;
-        }
-    }
-}
-
-// Repeer on each connect/disconnect peer event
-void osd_t::start_pg_peering(int pg_idx)
-{
-    auto & pg = pgs[pg_idx];
-    pg.state = PG_PEERING;
-    pg.state_dict.clear();
-    pg.obj_states.clear();
-    pg.ver_override.clear();
-    pg.pg_cursize = 0;
-    for (int role = 0; role < pg.cur_set.size(); role++)
-    {
-        if (pg.cur_set[role] != 0)
-        {
-            pg.pg_cursize++;
-        }
-    }
-    if (pg.pg_cursize < pg.pg_minsize)
-    {
-        pg.state = PG_INCOMPLETE;
-    }
-    if (pg.peering_state)
-    {
-        // Adjust the peering operation that's still in progress
-        for (auto it = pg.peering_state->list_ops.begin(); it != pg.peering_state->list_ops.end(); it++)
-        {
-            int role;
-            for (role = 0; role < pg.cur_set.size(); role++)
-            {
-                if (pg.cur_set[role] == it->first)
-                    break;
-            }
-            if (pg.state == PG_INCOMPLETE || role >= pg.cur_set.size())
-            {
-                // Discard the result after completion, which, chances are, will be unsuccessful
-                auto list_op = it->second;
-                if (list_op->peer_fd == 0)
-                {
-                    // Self
-                    list_op->bs_op->callback = [list_op](blockstore_op_t *bs_op)
-                    {
-                        if (list_op->bs_op->buf)
-                            free(list_op->bs_op->buf);
-                        delete list_op;
-                    };
-                }
-                else
-                {
-                    // Peer
-                    list_op->callback = [](osd_op_t *list_op)
-                    {
-                        delete list_op;
-                    };
-                }
-                pg.peering_state->list_ops.erase(it);
-                it = pg.peering_state->list_ops.begin();
-            }
-        }
-        for (auto it = pg.peering_state->list_results.begin(); it != pg.peering_state->list_results.end(); it++)
-        {
-            int role;
-            for (role = 0; role < pg.cur_set.size(); role++)
-            {
-                if (pg.cur_set[role] == it->first)
-                    break;
-            }
-            if (pg.state == PG_INCOMPLETE || role >= pg.cur_set.size())
-            {
-                if (it->second.buf)
-                {
-                    free(it->second.buf);
-                }
-                pg.peering_state->list_results.erase(it);
-                it = pg.peering_state->list_results.begin();
-            }
-        }
-    }
-    if (pg.state == PG_INCOMPLETE)
-    {
-        if (pg.peering_state)
-        {
-            delete pg.peering_state;
-            pg.peering_state = NULL;
-        }
-        printf("PG %d is incomplete\n", pg.pg_num);
-        return;
-    }
-    if (!pg.peering_state)
-    {
-        pg.peering_state = new pg_peering_state_t();
-    }
-    auto ps = pg.peering_state;
-    for (int role = 0; role < pg.cur_set.size(); role++)
-    {
-        osd_num_t role_osd = pg.cur_set[role];
-        if (!role_osd)
-        {
-            continue;
-        }
-        if (ps->list_ops.find(role_osd) != ps->list_ops.end() ||
-            ps->list_results.find(role_osd) != ps->list_results.end())
-        {
-            continue;
-        }
-        if (role_osd == this->osd_num)
-        {
-            // Self
-            osd_op_t *op = new osd_op_t();
-            op->op_type = 0;
-            op->peer_fd = 0;
-            op->bs_op = new blockstore_op_t();
-            op->bs_op->opcode = BS_OP_LIST;
-            op->bs_op->oid.stripe = parity_block_size;
-            op->bs_op->len = pg_count,
-            op->bs_op->offset = pg.pg_num-1,
-            op->bs_op->callback = [ps, op, role_osd](blockstore_op_t *bs_op)
-            {
-                if (op->bs_op->retval < 0)
-                {
-                    throw std::runtime_error("local OP_LIST failed");
-                }
-                printf(
-                    "Got object list from OSD %lu (local): %d object versions (%lu of them stable)\n",
-                    role_osd, bs_op->retval, bs_op->version
-                );
-                ps->list_results[role_osd] = {
-                    .buf = (obj_ver_id*)op->bs_op->buf,
-                    .total_count = (uint64_t)op->bs_op->retval,
-                    .stable_count = op->bs_op->version,
-                };
-                ps->list_done++;
-                ps->list_ops.erase(role_osd);
-                delete op;
-            };
-            bs->enqueue_op(op->bs_op);
-            ps->list_ops[role_osd] = op;
-        }
-        else
-        {
-            // Peer
-            auto & cl = clients[osd_peer_fds[role_osd]];
-            osd_op_t *op = new osd_op_t();
-            op->op_type = OSD_OP_OUT;
-            op->send_list.push_back(op->req.buf, OSD_PACKET_SIZE);
-            op->peer_fd = cl.peer_fd;
-            op->req = {
-                .sec_list = {
-                    .header = {
-                        .magic = SECONDARY_OSD_OP_MAGIC,
-                        .id = this->next_subop_id++,
-                        .opcode = OSD_OP_SECONDARY_LIST,
-                    },
-                    .list_pg = pg.pg_num,
-                    .pg_count = pg_count,
-                    .parity_block_size = parity_block_size,
-                },
-            };
-            op->callback = [this, ps, role_osd](osd_op_t *op)
-            {
-                if (op->reply.hdr.retval < 0)
-                {
-                    printf("Failed to get object list from OSD %lu (retval=%ld), disconnecting peer\n", role_osd, op->reply.hdr.retval);
-                    ps->list_ops.erase(role_osd);
-                    stop_client(op->peer_fd);
-                    delete op;
-                    return;
-                }
-                printf(
-                    "Got object list from OSD %lu: %ld object versions (%lu of them stable)\n",
-                    role_osd, op->reply.hdr.retval, op->reply.sec_list.stable_count
-                );
-                ps->list_results[role_osd] = {
-                    .buf = (obj_ver_id*)op->buf,
-                    .total_count = (uint64_t)op->reply.hdr.retval,
-                    .stable_count = op->reply.sec_list.stable_count,
-                };
-                // set op->buf to NULL so it doesn't get freed
-                op->buf = NULL;
-                ps->list_done++;
-                ps->list_ops.erase(role_osd);
-                delete op;
-            };
-            outbox_push(cl, op);
-            ps->list_ops[role_osd] = op;
-        }
-    }
-    ringloop->wakeup();
-}

osd_peering_pg.cpp

@@ -1,265 +0,0 @@
-#include "osd_peering_pg.h"
-
-void pg_t::remember_object(pg_obj_state_check_t &st, std::vector<obj_ver_role> &all)
-{
-    auto & pg = *this;
-    // Remember the decision
-    uint64_t state = 0;
-    if (st.n_roles == pg.pg_cursize)
-    {
-        if (st.n_matched == pg.pg_cursize)
-            state = OBJ_CLEAN;
-        else
-        {
-            state = OBJ_MISPLACED;
-            pg.state = pg.state | PG_HAS_MISPLACED;
-        }
-    }
-    else if (st.n_roles < pg.pg_minsize)
-    {
-        printf("Object is unfound: inode=%lu stripe=%lu version=%lu/%lu\n", st.oid.inode, st.oid.stripe, st.target_ver, st.max_ver);
-        state = OBJ_INCOMPLETE;
-        pg.state = pg.state | PG_HAS_UNFOUND;
-    }
-    else
-    {
-        printf("Object is degraded: inode=%lu stripe=%lu version=%lu/%lu\n", st.oid.inode, st.oid.stripe, st.target_ver, st.max_ver);
-        state = OBJ_DEGRADED;
-        pg.state = pg.state | PG_HAS_DEGRADED;
-    }
-    if (st.n_copies > pg.pg_size)
-    {
-        state |= OBJ_OVERCOPIED;
-        pg.state = pg.state | PG_HAS_UNCLEAN;
-    }
-    if (st.n_stable < st.n_copies)
-    {
-        state |= OBJ_NEEDS_STABLE;
-        pg.state = pg.state | PG_HAS_UNCLEAN;
-    }
-    if (st.target_ver < st.max_ver || st.has_old_unstable)
-    {
-        state |= OBJ_NEEDS_ROLLBACK;
-        pg.state = pg.state | PG_HAS_UNCLEAN;
-        pg.ver_override[st.oid] = st.target_ver;
-    }
-    if (st.is_buggy)
-    {
-        state |= OBJ_BUGGY;
-        // FIXME: bring pg offline
-        throw std::runtime_error("buggy object state");
-    }
-    if (state != OBJ_CLEAN)
-    {
-        st.osd_set.clear();
-        for (int i = st.ver_start; i < st.ver_end; i++)
-        {
-            st.osd_set.push_back((pg_obj_loc_t){
-                .role = (all[i].oid.stripe & STRIPE_MASK),
-                .osd_num = all[i].osd_num,
-                .stable = all[i].is_stable,
-            });
-        }
-        std::sort(st.osd_set.begin(), st.osd_set.end());
-        auto it = pg.state_dict.find(st.osd_set);
-        if (it == pg.state_dict.end())
-        {
-            std::vector<uint64_t> read_target;
-            read_target.resize(pg.pg_size);
-            for (int i = 0; i < pg.pg_size; i++)
-            {
-                read_target[i] = 0;
-            }
-            for (auto & o: st.osd_set)
-            {
-                read_target[o.role] = o.osd_num;
-            }
-            pg.state_dict[st.osd_set] = {
-                .read_target = read_target,
-                .osd_set = st.osd_set,
-                .state = state,
-                .object_count = 1,
-            };
-            it = pg.state_dict.find(st.osd_set);
-        }
-        else
-        {
-            it->second.object_count++;
-        }
-        pg.obj_states[st.oid] = &it->second;
-        if (st.target_ver < st.max_ver)
-        {
-            pg.ver_override[st.oid] = st.target_ver;
-        }
-        if (state & (OBJ_NEEDS_ROLLBACK | OBJ_NEEDS_STABLE))
-        {
-            spp::sparse_hash_map<obj_piece_id_t, obj_piece_ver_t> pieces;
-            for (int i = st.obj_start; i < st.obj_end; i++)
-            {
-                auto & pcs = pieces[(obj_piece_id_t){ .oid = all[i].oid, .osd_num = all[i].osd_num }];
-                if (!pcs.max_ver)
-                {
-                    pcs.max_ver = all[i].version;
-                }
-                if (all[i].is_stable && !pcs.stable_ver)
-                {
-                    pcs.stable_ver = all[i].version;
-                }
-            }
-            for (auto pp: pieces)
-            {
-                auto & pcs = pp.second;
-                if (pcs.stable_ver < pcs.max_ver)
-                {
-                    auto & act = obj_stab_actions[pp.first];
-                    if (pcs.max_ver > st.target_ver)
-                    {
-                        act.rollback = true;
-                        act.rollback_to = st.target_ver;
-                    }
-                    else if (pcs.max_ver < st.target_ver && pcs.stable_ver < pcs.max_ver)
-                    {
-                        act.rollback = true;
-                        act.rollback_to = pcs.stable_ver;
-                    }
-                    if (pcs.max_ver >= st.target_ver && pcs.stable_ver < st.target_ver)
-                    {
-                        act.make_stable = true;
-                        act.stable_to = st.target_ver;
-                    }
-                }
-            }
-        }
-    }
-    else
-        pg.clean_count++;
-    pg.total_count++;
-}
-
-// FIXME: Write at least some tests for this function
-void pg_t::calc_object_states()
-{
-    auto & pg = *this;
-    // Copy all object lists into one array
-    std::vector<obj_ver_role> all;
-    auto ps = pg.peering_state;
-    for (auto it: ps->list_results)
-    {
-        auto nstab = it.second.stable_count;
-        auto n = it.second.total_count;
-        auto osd_num = it.first;
-        uint64_t start = all.size();
-        all.resize(start + n);
-        obj_ver_id *ov = it.second.buf;
-        for (uint64_t i = 0; i < n; i++, ov++)
-        {
-            all[start+i] = {
-                .oid = ov->oid,
-                .version = ov->version,
-                .osd_num = osd_num,
-                .is_stable = i < nstab,
-            };
-        }
-        free(it.second.buf);
-        it.second.buf = NULL;
-    }
-    ps->list_results.clear();
-    // Sort
-    std::sort(all.begin(), all.end());
-    // Walk over it and check object states
-    pg.clean_count = 0;
-    pg.total_count = 0;
-    pg.state = 0;
-    int replica = 0;
-    pg_obj_state_check_t st;
-    for (int i = 0; i < all.size(); i++)
-    {
-        if (st.oid.inode != all[i].oid.inode ||
-            st.oid.stripe != (all[i].oid.stripe & ~STRIPE_MASK))
-        {
-            if (st.oid.inode != 0)
-            {
-                // Remember object state
-                st.obj_end = st.ver_end = i;
-                remember_object(st, all);
-            }
-            st.obj_start = st.ver_start = i;
-            st.oid = { .inode = all[i].oid.inode, .stripe = all[i].oid.stripe & ~STRIPE_MASK };
-            st.max_ver = st.target_ver = all[i].version;
-            st.has_roles = st.n_copies = st.n_roles = st.n_stable = st.n_matched = 0;
-            st.is_buggy = st.has_old_unstable = false;
-        }
-        else if (st.target_ver != all[i].version)
-        {
-            if (st.n_stable > 0 || st.n_roles >= pg.pg_minsize)
-            {
-                // Last processed version is either recoverable or stable, choose it as target and skip previous versions
-                st.ver_end = i;
-                i++;
-                while (i < all.size() && st.oid.inode == all[i].oid.inode &&
-                    st.oid.stripe == (all[i].oid.stripe & ~STRIPE_MASK))
-                {
-                    if (!all[i].is_stable)
-                    {
-                        st.has_old_unstable = true;
-                    }
-                    i++;
-                }
-                st.obj_end = i;
-                i--;
-                continue;
-            }
-            else
-            {
-                // Last processed version is unstable and unrecoverable
-                // We'll know that because target_ver < max_ver
-                st.ver_start = i;
-                st.target_ver = all[i].version;
-                st.has_roles = st.n_copies = st.n_roles = st.n_stable = st.n_matched = 0;
-            }
-        }
-        replica = (all[i].oid.stripe & STRIPE_MASK);
-        st.n_copies++;
-        if (replica >= pg.pg_size)
-        {
-            // FIXME In the future, check it against the PG epoch number to handle replication factor/scheme changes
-            st.is_buggy = true;
-        }
-        else
-        {
-            if (all[i].is_stable)
-            {
-                st.n_stable++;
-            }
-            if (pg.cur_set[replica] == all[i].osd_num)
-            {
-                st.n_matched++;
-            }
-            if (!(st.has_roles & (1 << replica)))
-            {
-                st.has_roles = st.has_roles | (1 << replica);
-                st.n_roles++;
-            }
-        }
-    }
-    if (st.oid.inode != 0)
-    {
-        // Remember object state
-        st.obj_end = st.ver_end = all.size();
-        remember_object(st, all);
-    }
-    if (pg.pg_cursize < pg.pg_size)
-    {
-        pg.state = pg.state | PG_DEGRADED;
-    }
-    printf(
-        "PG %u is active%s%s%s%s%s (%lu objects)\n", pg.pg_num,
-        (pg.state & PG_DEGRADED) ? " + degraded" : "",
-        (pg.state & PG_HAS_UNFOUND) ? " + has_unfound" : "",
-        (pg.state & PG_HAS_DEGRADED) ? " + has_degraded" : "",
-        (pg.state & PG_HAS_MISPLACED) ? " + has_misplaced" : "",
-        (pg.state & PG_HAS_UNCLEAN) ? " + has_unclean" : "",
-        pg.total_count
-    );
-    pg.state = pg.state | PG_ACTIVE;
-}

osd_primary.cpp

@@ -1,603 +0,0 @@
-#include "osd.h"
-#include "osd_rmw.h"
-
-#define SUBMIT_READ 0
-#define SUBMIT_RMW_READ 1
-#define SUBMIT_WRITE 2
-
-// read: read directly or read paired stripe(s), reconstruct, return
-// write: read paired stripe(s), modify, write
-//
-// nuance: take care to read the same version from paired stripes!
-// to do so, we remember "last readable" version until a write request completes
-// and we postpone other write requests to the same stripe until completion of previous ones
-//
-// sync: sync peers, get unstable versions from somewhere, stabilize them
-
-struct unstable_osd_num_t
-{
-    osd_num_t osd_num;
-    int start, len;
-};
-
-struct osd_primary_op_data_t
-{
-    int st = 0;
-    pg_num_t pg_num;
-    object_id oid;
-    uint64_t target_ver;
-    uint64_t fact_ver = 0;
-    int n_subops = 0, done = 0, errors = 0;
-    int degraded = 0, pg_size, pg_minsize;
-    osd_rmw_stripe_t *stripes;
-    osd_op_t *subops = NULL;
-    // for sync. oops, requires freeing
-    std::vector<unstable_osd_num_t> *unstable_write_osds = NULL;
-    obj_ver_id *unstable_writes = NULL;
-};
-
-void osd_t::finish_primary_op(osd_op_t *cur_op, int retval)
-{
-    // FIXME add separate magic number
-    auto cl_it = clients.find(cur_op->peer_fd);
-    if (cl_it != clients.end())
-    {
-        cur_op->reply.hdr.magic = SECONDARY_OSD_REPLY_MAGIC;
-        cur_op->reply.hdr.id = cur_op->req.hdr.id;
-        cur_op->reply.hdr.opcode = cur_op->req.hdr.opcode;
-        cur_op->reply.hdr.retval = retval;
-        outbox_push(cl_it->second, cur_op);
-    }
-    else
-    {
-        delete cur_op;
-    }
-}
-
-bool osd_t::prepare_primary_rw(osd_op_t *cur_op)
-{
-    // PG number is calculated from the offset
-    // Our EC scheme stores data in fixed chunks equal to (K*block size)
-    // But we must not use K in the process of calculating the PG number
-    // So we calculate the PG number using a separate setting which should be per-inode (FIXME)
-    // FIXME Real pg_num should equal the below expression + 1
-    pg_num_t pg_num = (cur_op->req.rw.inode + cur_op->req.rw.offset / parity_block_size) % pg_count;
-    // FIXME: Postpone operations in inactive PGs
-    if (pg_num > pgs.size() || !(pgs[pg_num].state & PG_ACTIVE))
-    {
-        finish_primary_op(cur_op, -EINVAL);
-        return false;
-    }
-    uint64_t pg_parity_size = bs_block_size * pgs[pg_num].pg_minsize;
-    object_id oid = {
-        .inode = cur_op->req.rw.inode,
-        // oid.stripe = starting offset of the parity stripe, so it can be mapped back to the PG
-        .stripe = (cur_op->req.rw.offset / parity_block_size) * parity_block_size +
-            ((cur_op->req.rw.offset % parity_block_size) / pg_parity_size) * pg_parity_size
-    };
-    if ((cur_op->req.rw.offset + cur_op->req.rw.len) > (oid.stripe + pg_parity_size) ||
-        (cur_op->req.rw.offset % bs_disk_alignment) != 0 ||
-        (cur_op->req.rw.len % bs_disk_alignment) != 0)
-    {
-        finish_primary_op(cur_op, -EINVAL);
-        return false;
-    }
-    osd_primary_op_data_t *op_data = (osd_primary_op_data_t*)calloc(
-        sizeof(osd_primary_op_data_t) + sizeof(osd_rmw_stripe_t) * pgs[pg_num].pg_size, 1
-    );
-    op_data->pg_num = pg_num;
-    op_data->oid = oid;
-    op_data->stripes = ((osd_rmw_stripe_t*)(op_data+1));
-    cur_op->op_data = op_data;
-    split_stripes(pgs[pg_num].pg_minsize, bs_block_size, (uint32_t)(cur_op->req.rw.offset - oid.stripe), cur_op->req.rw.len, op_data->stripes);
-    return true;
-}
-
-void osd_t::continue_primary_read(osd_op_t *cur_op)
-{
-    if (!cur_op->op_data && !prepare_primary_rw(cur_op))
-    {
-        return;
-    }
-    osd_primary_op_data_t *op_data = cur_op->op_data;
-    if (op_data->st == 1)      goto resume_1;
-    else if (op_data->st == 2) goto resume_2;
-    {
-        auto & pg = pgs[op_data->pg_num];
-        for (int role = 0; role < pg.pg_minsize; role++)
-        {
-            op_data->stripes[role].read_start = op_data->stripes[role].req_start;
-            op_data->stripes[role].read_end = op_data->stripes[role].req_end;
-        }
-        // Determine version
-        auto vo_it = pg.ver_override.find(op_data->oid);
-        op_data->target_ver = vo_it != pg.ver_override.end() ? vo_it->second : UINT64_MAX;
-        if (pg.state == PG_ACTIVE)
-        {
-            // Fast happy-path
-            cur_op->buf = alloc_read_buffer(op_data->stripes, pg.pg_minsize, 0);
-            submit_primary_subops(SUBMIT_READ, pg.pg_minsize, pg.cur_set.data(), cur_op);
-            cur_op->send_list.push_back(cur_op->buf, cur_op->req.rw.len);
-            op_data->st = 1;
-        }
-        else
-        {
-            // PG may be degraded or have misplaced objects
-            auto st_it = pg.obj_states.find(op_data->oid);
-            uint64_t* cur_set = (st_it != pg.obj_states.end()
-                ? st_it->second->read_target.data()
-                : pg.cur_set.data());
-            if (extend_missing_stripes(op_data->stripes, cur_set, pg.pg_minsize, pg.pg_size) < 0)
-            {
-                free(op_data);
-                finish_primary_op(cur_op, -EIO);
-                return;
-            }
-            // Submit reads
-            op_data->pg_minsize = pg.pg_minsize;
-            op_data->pg_size = pg.pg_size;
-            op_data->degraded = 1;
-            cur_op->buf = alloc_read_buffer(op_data->stripes, pg.pg_size, 0);
-            submit_primary_subops(SUBMIT_READ, pg.pg_size, cur_set, cur_op);
-            op_data->st = 1;
-        }
-    }
-resume_1:
-    return;
-resume_2:
-    if (op_data->errors > 0)
-    {
-        free(op_data);
-        cur_op->op_data = NULL;
-        finish_primary_op(cur_op, -EIO);
-        return;
-    }
-    if (op_data->degraded)
-    {
-        // Reconstruct missing stripes
-        // FIXME: Always EC(k+1) by now. Add different coding schemes
-        osd_rmw_stripe_t *stripes = op_data->stripes;
-        for (int role = 0; role < op_data->pg_minsize; role++)
-        {
-            if (stripes[role].read_end != 0 && stripes[role].missing)
-            {
-                reconstruct_stripe(stripes, op_data->pg_size, role);
-            }
-            if (stripes[role].req_end != 0)
-            {
-                // Send buffer in parts to avoid copying
-                cur_op->send_list.push_back(
-                    stripes[role].read_buf + (stripes[role].req_start - stripes[role].read_start),
-                    stripes[role].req_end - stripes[role].req_start
-                );
-            }
-        }
-    }
-    free(op_data);
-    cur_op->op_data = NULL;
-    finish_primary_op(cur_op, cur_op->req.rw.len);
-}
-
-void osd_t::submit_primary_subops(int submit_type, int pg_size, const uint64_t* osd_set, osd_op_t *cur_op)
-{
-    bool w = submit_type == SUBMIT_WRITE;
-    osd_primary_op_data_t *op_data = cur_op->op_data;
-    osd_rmw_stripe_t *stripes = op_data->stripes;
-    // Allocate subops
-    int n_subops = 0, zero_read = -1;
-    for (int role = 0; role < pg_size; role++)
-    {
-        if (osd_set[role] == this->osd_num || osd_set[role] != 0 && zero_read == -1)
-        {
-            zero_read = role;
-        }
-        if (osd_set[role] != 0 && (w || stripes[role].read_end != 0))
-        {
-            n_subops++;
-        }
-    }
-    if (!n_subops && submit_type == SUBMIT_RMW_READ)
-    {
-        n_subops = 1;
-    }
-    else
-    {
-        zero_read = -1;
-    }
-    osd_op_t *subops = new osd_op_t[n_subops];
-    op_data->done = op_data->errors = 0;
-    op_data->n_subops = n_subops;
-    op_data->subops = subops;
-    int subop = 0;
-    for (int role = 0; role < pg_size; role++)
-    {
-        // We always submit zero-length writes to all replicas, even if the stripe is not modified
-        if (!(w || stripes[role].read_end != 0 || zero_read == role))
-        {
-            continue;
-        }
-        osd_num_t role_osd_num = osd_set[role];
-        if (role_osd_num != 0)
-        {
-            if (role_osd_num == this->osd_num)
-            {
-                subops[subop].bs_op = new blockstore_op_t({
-                    .opcode = (uint64_t)(w ? BS_OP_WRITE : BS_OP_READ),
-                    .callback = [cur_op, this](blockstore_op_t *subop)
-                    {
-                        handle_primary_subop(cur_op, subop->retval == subop->len, subop->version);
-                    },
-                    .oid = {
-                        .inode = op_data->oid.inode,
-                        .stripe = op_data->oid.stripe | role,
-                    },
-                    .version = w ? 0 : (submit_type == SUBMIT_RMW_READ ? UINT64_MAX : op_data->target_ver),
-                    .offset = w ? stripes[role].write_start : stripes[role].read_start,
-                    .len = w ? stripes[role].write_end - stripes[role].write_start : stripes[role].read_end - stripes[role].read_start,
-                    .buf = w ? stripes[role].write_buf : stripes[role].read_buf,
-                });
-                bs->enqueue_op(subops[subop].bs_op);
-            }
-            else
-            {
-                subops[subop].op_type = OSD_OP_OUT;
-                subops[subop].send_list.push_back(subops[subop].req.buf, OSD_PACKET_SIZE);
-                subops[subop].peer_fd = this->osd_peer_fds.at(role_osd_num);
-                subops[subop].req.sec_rw = {
-                    .header = {
-                        .magic = SECONDARY_OSD_OP_MAGIC,
-                        .id = this->next_subop_id++,
-                        .opcode = (uint64_t)(w ? OSD_OP_SECONDARY_WRITE : OSD_OP_SECONDARY_READ),
-                    },
-                    .oid = {
-                        .inode = op_data->oid.inode,
-                        .stripe = op_data->oid.stripe | role,
-                    },
-                    .version = w ? 0 : (submit_type == SUBMIT_RMW_READ ? UINT64_MAX : op_data->target_ver),
-                    .offset = w ? stripes[role].write_start : stripes[role].read_start,
-                    .len = w ? stripes[role].write_end - stripes[role].write_start : stripes[role].read_end - stripes[role].read_start,
-                };
-                subops[subop].buf = w ? stripes[role].write_buf : stripes[role].read_buf;
-                if (w && stripes[role].write_end > 0)
-                {
-                    subops[subop].send_list.push_back(stripes[role].write_buf, stripes[role].write_end - stripes[role].write_start);
-                }
-                subops[subop].callback = [cur_op, this](osd_op_t *subop)
-                {
-                    // so it doesn't get freed
-                    subop->buf = NULL;
-                    handle_primary_subop(cur_op, subop->reply.hdr.retval == subop->req.sec_rw.len, subop->reply.sec_rw.version);
-                };
-                outbox_push(clients[subops[subop].peer_fd], &subops[subop]);
-            }
-            subop++;
-        }
-    }
-}
-
-void osd_t::handle_primary_subop(osd_op_t *cur_op, int ok, uint64_t version)
-{
-    osd_primary_op_data_t *op_data = cur_op->op_data;
-    op_data->fact_ver = version;
-    if (!ok)
-    {
-        // FIXME: Handle errors
-        op_data->errors++;
-    }
-    else
-    {
-        op_data->done++;
-    }
-    if ((op_data->errors + op_data->done) >= op_data->n_subops)
-    {
-        delete[] op_data->subops;
-        op_data->subops = NULL;
-        op_data->st++;
-        if (cur_op->req.hdr.opcode == OSD_OP_READ)
-        {
-            continue_primary_read(cur_op);
-        }
-        else if (cur_op->req.hdr.opcode == OSD_OP_WRITE)
-        {
-            continue_primary_write(cur_op);
-        }
-        else if (cur_op->req.hdr.opcode == OSD_OP_SYNC)
-        {
-            continue_primary_sync(cur_op);
-        }
-        else
-        {
-            throw std::runtime_error("BUG: unknown opcode");
-        }
-    }
-}
-
-void osd_t::continue_primary_write(osd_op_t *cur_op)
-{
-    if (!cur_op->op_data && !prepare_primary_rw(cur_op))
-    {
-        return;
-    }
-    osd_primary_op_data_t *op_data = cur_op->op_data;
-    // FIXME: Handle operation cancel
-    auto & pg = pgs[op_data->pg_num];
-    if (op_data->st == 1)      goto resume_1;
-    else if (op_data->st == 2) goto resume_2;
-    else if (op_data->st == 3) goto resume_3;
-    else if (op_data->st == 4) goto resume_4;
-    else if (op_data->st == 5) goto resume_5;
-    assert(op_data->st == 0);
-    // Check if actions are pending for this object
-    {
-        auto act_it = pg.obj_stab_actions.lower_bound((obj_piece_id_t){
-            .oid = op_data->oid,
-            .osd_num = 0,
-        });
-        if (act_it != pg.obj_stab_actions.end() &&
-            act_it->first.oid.inode == op_data->oid.inode &&
-            (act_it->first.oid.stripe & ~STRIPE_MASK) == op_data->oid.stripe)
-        {
-            // FIXME postpone the request until actions are done
-            free(op_data);
-            finish_primary_op(cur_op, -EIO);
-            return;
-        }
-    }
-    // Check if there are other write requests to the same object
-    {
-        auto vo_it = pg.write_queue.find(op_data->oid);
-        if (vo_it != pg.write_queue.end())
-        {
-            op_data->st = 1;
-            pg.write_queue.emplace(op_data->oid, cur_op);
-            return;
-        }
-        pg.write_queue.emplace(op_data->oid, cur_op);
-    }
-resume_1:
-    // Determine blocks to read
-    cur_op->rmw_buf = calc_rmw_reads(cur_op->buf, op_data->stripes, pg.cur_set.data(), pg.pg_size, pg.pg_minsize, pg.pg_cursize);
-    // Read required blocks
-    submit_primary_subops(SUBMIT_RMW_READ, pg.pg_size, pg.cur_set.data(), cur_op);
-resume_2:
-    op_data->st = 2;
-    return;
-resume_3:
-    // Save version override for parallel reads
-    pg.ver_override[op_data->oid] = op_data->fact_ver;
-    // Calculate parity
-    calc_rmw_parity(op_data->stripes, pg.pg_size);
-    // Send writes
-    submit_primary_subops(SUBMIT_WRITE, pg.pg_size, pg.cur_set.data(), cur_op);
-resume_4:
-    op_data->st = 4;
-    return;
-resume_5:
-    // Remember version as unstable
-    osd_num_t *osd_set = pg.cur_set.data();
-    for (int role = 0; role < pg.pg_size; role++)
-    {
-        if (osd_set[role] != 0)
-        {
-            this->unstable_writes[(osd_object_id_t){
-                .osd_num = osd_set[role],
-                .oid = {
-                    .inode = op_data->oid.inode,
-                    .stripe = op_data->oid.stripe | role,
-                },
-            }] = op_data->fact_ver;
-        }
-    }
-    // Remember PG as dirty to drop the connection when PG goes offline
-    // (this is required because of the "lazy sync")
-    this->clients[cur_op->peer_fd].dirty_pgs.insert(op_data->pg_num);
-    // Remove version override
-    pg.ver_override.erase(op_data->oid);
-    finish_primary_op(cur_op, cur_op->req.rw.len);
-    // Continue other write operations to the same object
-    {
-        auto next_it = pg.write_queue.find(op_data->oid);
-        auto this_it = next_it;
-        next_it++;
-        pg.write_queue.erase(this_it);
-        if (next_it != pg.write_queue.end() &&
-            next_it->first == op_data->oid)
-        {
-            osd_op_t *next_op = next_it->second;
-            continue_primary_write(next_op);
-        }
-    }
-}
-
-// Save and clear unstable_writes -> SYNC all -> STABLE all
-// FIXME: Run regular automatic syncs based on the number of unstable writes and/or system time
-void osd_t::continue_primary_sync(osd_op_t *cur_op)
-{
-    if (!cur_op->op_data)
-    {
-        cur_op->op_data = (osd_primary_op_data_t*)calloc(sizeof(osd_primary_op_data_t), 1);
-    }
-    if (cur_op->op_data->st == 1)      goto resume_1;
-    else if (cur_op->op_data->st == 2) goto resume_2;
-    else if (cur_op->op_data->st == 3) goto resume_3;
-    else if (cur_op->op_data->st == 4) goto resume_4;
-    else if (cur_op->op_data->st == 5) goto resume_5;
-    else if (cur_op->op_data->st == 6) goto resume_6;
-    assert(cur_op->op_data->st == 0);
-    if (syncs_in_progress.size() > 0)
-    {
-        // Wait for previous syncs, if any
-        // FIXME: We may try to execute the current one in parallel, like in Blockstore, but I'm not sure if it matters at all
-        syncs_in_progress.push_back(cur_op);
-        cur_op->op_data->st = 1;
-resume_1:
-        return;
-    }
-    else
-    {
-        syncs_in_progress.push_back(cur_op);
-    }
-resume_2:
-    // FIXME: Handle operation cancel
-    if (unstable_writes.size() == 0)
-    {
-        // Nothing to sync
-        goto finish;
-    }
-    // Save and clear unstable_writes
-    // FIXME: This is possible to do it on a per-client basis
-    // It would be cool not to copy them here at all, but someone has to deduplicate them by object IDs anyway
-    cur_op->op_data->unstable_write_osds = new std::vector<unstable_osd_num_t>();
-    cur_op->op_data->unstable_writes = new obj_ver_id[unstable_writes.size()];
-    {
-        osd_num_t last_osd = 0;
-        int last_start = 0, last_end = 0;
-        for (auto it = unstable_writes.begin(); it != unstable_writes.end(); it++)
-        {
-            if (last_osd != it->first.osd_num)
-            {
-                if (last_osd != 0)
-                {
-                    cur_op->op_data->unstable_write_osds->push_back((unstable_osd_num_t){
-                        .osd_num = last_osd,
-                        .start = last_start,
-                        .len = last_end - last_start,
-                    });
-                }
-                last_osd = it->first.osd_num;
-                last_start = last_end;
-            }
-            cur_op->op_data->unstable_writes[last_end] = (obj_ver_id){
-                .oid = it->first.oid,
-                .version = it->second,
-            };
-            last_end++;
-        }
-        if (last_osd != 0)
-        {
-            cur_op->op_data->unstable_write_osds->push_back((unstable_osd_num_t){
-                .osd_num = last_osd,
-                .start = last_start,
-                .len = last_end - last_start,
-            });
-        }
-    }
-    unstable_writes.clear();
-    // SYNC
-    submit_primary_sync_subops(cur_op);
-resume_3:
-    cur_op->op_data->st = 3;
-    return;
-resume_4:
-    // Stabilize version sets
-    submit_primary_stab_subops(cur_op);
-resume_5:
-    cur_op->op_data->st = 5;
-    return;
-resume_6:
-    // FIXME: Free them correctly (via a destructor or so)
-    delete cur_op->op_data->unstable_write_osds;
-    delete[] cur_op->op_data->unstable_writes;
-    cur_op->op_data->unstable_writes = NULL;
-    cur_op->op_data->unstable_write_osds = NULL;
-finish:
-    assert(syncs_in_progress.front() == cur_op);
-    syncs_in_progress.pop_front();
-    finish_primary_op(cur_op, 0);
-    if (syncs_in_progress.size() > 0)
-    {
-        cur_op = syncs_in_progress.front();
-        cur_op->op_data->st++;
-        goto resume_2;
-    }
-}
-
-void osd_t::submit_primary_sync_subops(osd_op_t *cur_op)
-{
-    osd_primary_op_data_t *op_data = cur_op->op_data;
-    int n_osds = op_data->unstable_write_osds->size();
-    osd_op_t *subops = new osd_op_t[n_osds];
-    op_data->done = op_data->errors = 0;
-    op_data->n_subops = n_osds;
-    op_data->subops = subops;
-    for (int i = 0; i < n_osds; i++)
-    {
-        osd_num_t sync_osd = (*(op_data->unstable_write_osds))[i].osd_num;
-        if (sync_osd == this->osd_num)
-        {
-            subops[i].bs_op = new blockstore_op_t({
-                .opcode = BS_OP_SYNC,
-                .callback = [cur_op, this](blockstore_op_t *subop)
-                {
-                    handle_primary_subop(cur_op, subop->retval == 0, 0);
-                },
-            });
-            bs->enqueue_op(subops[i].bs_op);
-        }
-        else
-        {
-            subops[i].op_type = OSD_OP_OUT;
-            subops[i].send_list.push_back(subops[i].req.buf, OSD_PACKET_SIZE);
-            subops[i].peer_fd = osd_peer_fds.at(sync_osd);
-            subops[i].req.sec_sync = {
-                .header = {
-                    .magic = SECONDARY_OSD_OP_MAGIC,
-                    .id = this->next_subop_id++,
-                    .opcode = OSD_OP_SECONDARY_SYNC,
-                },
-            };
-            subops[i].callback = [cur_op, this](osd_op_t *subop)
-            {
-                handle_primary_subop(cur_op, subop->reply.hdr.retval == 0, 0);
-            };
-            outbox_push(clients[subops[i].peer_fd], &subops[i]);
-        }
-    }
-}
-
-void osd_t::submit_primary_stab_subops(osd_op_t *cur_op)
-{
-    osd_primary_op_data_t *op_data = cur_op->op_data;
-    int n_osds = op_data->unstable_write_osds->size();
-    osd_op_t *subops = new osd_op_t[n_osds];
-    op_data->done = op_data->errors = 0;
-    op_data->n_subops = n_osds;
-    op_data->subops = subops;
-    for (int i = 0; i < n_osds; i++)
-    {
-        auto & stab_osd = (*(op_data->unstable_write_osds))[i];
-        if (stab_osd.osd_num == this->osd_num)
-        {
-            subops[i].bs_op = new blockstore_op_t({
-                .opcode = BS_OP_STABLE,
-                .callback = [cur_op, this](blockstore_op_t *subop)
-                {
-                    handle_primary_subop(cur_op, subop->retval == 0, 0);
-                },
-                .len = (uint32_t)stab_osd.len,
-                .buf = (void*)(op_data->unstable_writes + stab_osd.start),
-            });
-            bs->enqueue_op(subops[i].bs_op);
-        }
-        else
-        {
-            subops[i].op_type = OSD_OP_OUT;
-            subops[i].send_list.push_back(subops[i].req.buf, OSD_PACKET_SIZE);
-            subops[i].peer_fd = osd_peer_fds.at(stab_osd.osd_num);
-            subops[i].req.sec_stab = {
-                .header = {
-                    .magic = SECONDARY_OSD_OP_MAGIC,
-                    .id = this->next_subop_id++,
-                    .opcode = OSD_OP_SECONDARY_STABILIZE,
-                },
-                .len = (uint64_t)(stab_osd.len * sizeof(obj_ver_id)),
-            };
-            subops[i].send_list.push_back(op_data->unstable_writes + stab_osd.start, stab_osd.len * sizeof(obj_ver_id));
-            subops[i].callback = [cur_op, this](osd_op_t *subop)
-            {
-                handle_primary_subop(cur_op, subop->reply.hdr.retval == 0, 0);
-            };
-            outbox_push(clients[subops[i].peer_fd], &subops[i]);
-        }
-    }
-}

osd_receive.cpp

@@ -1,204 +0,0 @@
-#include "osd.h"
-
-void osd_t::read_requests()
-{
-    for (int i = 0; i < read_ready_clients.size(); i++)
-    {
-        int peer_fd = read_ready_clients[i];
-        auto & cl = clients[peer_fd];
-        io_uring_sqe* sqe = ringloop->get_sqe();
-        if (!sqe)
-        {
-            read_ready_clients.erase(read_ready_clients.begin(), read_ready_clients.begin() + i);
-            return;
-        }
-        ring_data_t* data = ((ring_data_t*)sqe->user_data);
-        if (!cl.read_buf)
-        {
-            // no reads in progress
-            // so this is either a new command or a reply to a previously sent command
-            if (!cl.read_op)
-            {
-                cl.read_op = new osd_op_t;
-                cl.read_op->peer_fd = peer_fd;
-            }
-            cl.read_op->op_type = OSD_OP_IN;
-            cl.read_buf = &cl.read_op->req.buf;
-            cl.read_remaining = OSD_PACKET_SIZE;
-            cl.read_state = CL_READ_OP;
-        }
-        cl.read_iov.iov_base = cl.read_buf;
-        cl.read_iov.iov_len = cl.read_remaining;
-        cl.read_msg.msg_iov = &cl.read_iov;
-        cl.read_msg.msg_iovlen = 1;
-        data->callback = [this, peer_fd](ring_data_t *data) { handle_read(data, peer_fd); };
-        my_uring_prep_recvmsg(sqe, peer_fd, &cl.read_msg, 0);
-    }
-    read_ready_clients.clear();
-}
-
-void osd_t::handle_read(ring_data_t *data, int peer_fd)
-{
-    auto cl_it = clients.find(peer_fd);
-    if (cl_it != clients.end())
-    {
-        auto & cl = cl_it->second;
-        if (data->res == -EAGAIN)
-        {
-            cl.read_ready--;
-            if (cl.read_ready > 0)
-                read_ready_clients.push_back(peer_fd);
-            return;
-        }
-        else if (data->res < 0)
-        {
-            // this is a client socket, so don't panic. just disconnect it
-            printf("Client %d socket read error: %d (%s). Disconnecting client\n", peer_fd, -data->res, strerror(-data->res));
-            stop_client(peer_fd);
-            return;
-        }
-        read_ready_clients.push_back(peer_fd);
-        if (data->res > 0)
-        {
-            cl.read_remaining -= data->res;
-            cl.read_buf += data->res;
-            if (cl.read_remaining <= 0)
-            {
-                cl.read_buf = NULL;
-                if (cl.read_state == CL_READ_OP)
-                {
-                    if (cl.read_op->req.hdr.magic == SECONDARY_OSD_REPLY_MAGIC)
-                    {
-                        handle_reply_hdr(&cl);
-                    }
-                    else
-                    {
-                        handle_op_hdr(&cl);
-                    }
-                }
-                else if (cl.read_state == CL_READ_DATA)
-                {
-                    // Operation is ready
-                    exec_op(cl.read_op);
-                    cl.read_op = NULL;
-                    cl.read_state = 0;
-                }
-                else if (cl.read_state == CL_READ_REPLY_DATA)
-                {
-                    // Reply is ready
-                    auto req_it = cl.sent_ops.find(cl.read_reply_id);
-                    osd_op_t *request = req_it->second;
-                    cl.sent_ops.erase(req_it);
-                    cl.read_reply_id = 0;
-                    cl.read_state = 0;
-                    // Measure subop latency
-                    timespec tv_end;
-                    clock_gettime(CLOCK_REALTIME, &tv_end);
-                    subop_stat_count[request->req.hdr.opcode]++;
-                    subop_stat_sum[request->req.hdr.opcode] += (
-                        (tv_end.tv_sec - request->tv_begin.tv_sec)*1000000 +
-                        (tv_end.tv_nsec - request->tv_begin.tv_nsec)/1000
-                    );
-                    request->callback(request);
-                }
-            }
-        }
-    }
-}
-
-void osd_t::handle_op_hdr(osd_client_t *cl)
-{
-    osd_op_t *cur_op = cl->read_op;
-    if (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_READ)
-    {
-        if (cur_op->req.sec_rw.len > 0)
-            cur_op->buf = memalign(512, cur_op->req.sec_rw.len);
-        cl->read_remaining = 0;
-    }
-    else if (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_WRITE)
-    {
-        if (cur_op->req.sec_rw.len > 0)
-            cur_op->buf = memalign(512, cur_op->req.sec_rw.len);
-        cl->read_remaining = cur_op->req.sec_rw.len;
-    }
-    else if (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_STABILIZE ||
-        cur_op->req.hdr.opcode == OSD_OP_SECONDARY_ROLLBACK)
-    {
-        if (cur_op->req.sec_stab.len > 0)
-            cur_op->buf = memalign(512, cur_op->req.sec_stab.len);
-        cl->read_remaining = cur_op->req.sec_stab.len;
-    }
-    else if (cur_op->req.hdr.opcode == OSD_OP_READ)
-    {
-        if (cur_op->req.rw.len > 0)
-            cur_op->buf = memalign(512, cur_op->req.rw.len);
-        cl->read_remaining = 0;
-    }
-    else if (cur_op->req.hdr.opcode == OSD_OP_WRITE)
-    {
-        if (cur_op->req.rw.len > 0)
-            cur_op->buf = memalign(512, cur_op->req.rw.len);
-        cl->read_remaining = cur_op->req.rw.len;
-    }
-    if (cl->read_remaining > 0)
-    {
-        // Read data
-        cl->read_buf = cur_op->buf;
-        cl->read_state = CL_READ_DATA;
-    }
-    else
-    {
-        // Operation is ready
-        cl->read_op = NULL;
-        cl->read_state = 0;
-        exec_op(cur_op);
-    }
-}
-
-void osd_t::handle_reply_hdr(osd_client_t *cl)
-{
-    osd_op_t *cur_op = cl->read_op;
-    auto req_it = cl->sent_ops.find(cur_op->req.hdr.id);
-    if (req_it == cl->sent_ops.end())
-    {
-        // Command out of sync. Drop connection
-        printf("Client %d command out of sync: id %lu\n", cl->peer_fd, cur_op->req.hdr.id);
-        stop_client(cl->peer_fd);
-        return;
-    }
-    osd_op_t *op = req_it->second;
-    memcpy(op->reply.buf, cur_op->req.buf, OSD_PACKET_SIZE);
-    if (op->reply.hdr.opcode == OSD_OP_SECONDARY_READ &&
-        op->reply.hdr.retval > 0)
-    {
-        // Read data. In this case we assume that the buffer is preallocated by the caller (!)
-        assert(op->buf);
-        cl->read_state = CL_READ_REPLY_DATA;
-        cl->read_reply_id = op->req.hdr.id;
-        cl->read_buf = op->buf;
-        cl->read_remaining = op->reply.hdr.retval;
-    }
-    else if (op->reply.hdr.opcode == OSD_OP_SECONDARY_LIST &&
-        op->reply.hdr.retval > 0)
-    {
-        op->buf = memalign(512, sizeof(obj_ver_id) * op->reply.hdr.retval);
-        cl->read_state = CL_READ_REPLY_DATA;
-        cl->read_reply_id = op->req.hdr.id;
-        cl->read_buf = op->buf;
-        cl->read_remaining = sizeof(obj_ver_id) * op->reply.hdr.retval;
-    }
-    else
-    {
-        cl->read_state = 0;
-        cl->sent_ops.erase(req_it);
-        // Measure subop latency
-        timespec tv_end;
-        clock_gettime(CLOCK_REALTIME, &tv_end);
-        subop_stat_count[op->req.hdr.opcode]++;
-        subop_stat_sum[op->req.hdr.opcode] += (
-            (tv_end.tv_sec - op->tv_begin.tv_sec)*1000000 +
-            (tv_end.tv_nsec - op->tv_begin.tv_nsec)/1000
-        );
-        op->callback(op);
-    }
-}

osd_rmw.cpp

@@ -1,367 +0,0 @@
-#include <malloc.h>
-#include <assert.h>
-#include "xor.h"
-#include "osd_rmw.h"
-
-static inline void extend_read(uint32_t start, uint32_t end, osd_rmw_stripe_t & stripe)
-{
-    if (stripe.read_end == 0)
-    {
-        stripe.read_start = start;
-        stripe.read_end = end;
-    }
-    else
-    {
-        if (stripe.read_end < end)
-            stripe.read_end = end;
-        if (stripe.read_start > start)
-            stripe.read_start = start;
-    }
-}
-
-static inline void cover_read(uint32_t start, uint32_t end, osd_rmw_stripe_t & stripe)
-{
-    // Subtract <to> write request from <from> request
-    if (start >= stripe.req_start &&
-        end <= stripe.req_end)
-    {
-        return;
-    }
-    if (start <= stripe.req_start &&
-        end >= stripe.req_start &&
-        end <= stripe.req_end)
-    {
-        end = stripe.req_start;
-    }
-    else if (start >= stripe.req_start &&
-        start <= stripe.req_end &&
-        end >= stripe.req_end)
-    {
-        start = stripe.req_end;
-    }
-    if (stripe.read_end == 0)
-    {
-        stripe.read_start = start;
-        stripe.read_end = end;
-    }
-    else
-    {
-        if (stripe.read_end < end)
-            stripe.read_end = end;
-        if (stripe.read_start > start)
-            stripe.read_start = start;
-    }
-}
-
-void split_stripes(uint64_t pg_minsize, uint32_t bs_block_size, uint32_t start, uint32_t end, osd_rmw_stripe_t *stripes)
-{
-    end = start+end;
-    for (int role = 0; role < pg_minsize; role++)
-    {
-        if (start < (1+role)*bs_block_size && end > role*bs_block_size)
-        {
-            stripes[role].req_start = start < role*bs_block_size ? 0 : start-role*bs_block_size;
-            stripes[role].req_end = end > (role+1)*bs_block_size ? bs_block_size : end-role*bs_block_size;
-        }
-    }
-}
-
-void reconstruct_stripe(osd_rmw_stripe_t *stripes, int pg_size, int role)
-{
-    int prev = -2;
-    for (int other = 0; other < pg_size; other++)
-    {
-        if (other != role)
-        {
-            if (prev == -2)
-            {
-                prev = other;
-            }
-            else if (prev >= 0)
-            {
-                memxor(
-                    stripes[prev].read_buf + (stripes[prev].read_start - stripes[role].read_start),
-                    stripes[other].read_buf + (stripes[other].read_start - stripes[other].read_start),
-                    stripes[role].read_buf, stripes[role].read_end - stripes[role].read_start
-                );
-                prev = -1;
-            }
-            else
-            {
-                memxor(
-                    stripes[role].read_buf,
-                    stripes[other].read_buf + (stripes[other].read_start - stripes[role].read_start),
-                    stripes[role].read_buf, stripes[role].read_end - stripes[role].read_start
-                );
-            }
-        }
-    }
-}
-
-int extend_missing_stripes(osd_rmw_stripe_t *stripes, osd_num_t *osd_set, int minsize, int size)
-{
-    for (int role = 0; role < minsize; role++)
-    {
-        if (stripes[role].read_end != 0 && osd_set[role] == 0)
-        {
-            stripes[role].missing = true;
-            // Stripe is missing. Extend read to other stripes.
-            // We need at least pg_minsize stripes to recover the lost part.
-            // FIXME: LRC EC and similar don't require to read all other stripes.
-            int exist = 0;
-            for (int j = 0; j < size; j++)
-            {
-                if (osd_set[j] != 0)
-                {
-                    extend_read(stripes[role].read_start, stripes[role].read_end, stripes[j]);
-                    exist++;
-                    if (exist >= minsize)
-                    {
-                        break;
-                    }
-                }
-            }
-            if (exist < minsize)
-            {
-                // Less than minsize stripes are available for this object
-                return -1;
-            }
-        }
-    }
-    return 0;
-}
-
-void* alloc_read_buffer(osd_rmw_stripe_t *stripes, int read_pg_size, uint64_t add_size)
-{
-    // Calculate buffer size
-    uint64_t buf_size = add_size;
-    for (int role = 0; role < read_pg_size; role++)
-    {
-        if (stripes[role].read_end != 0)
-        {
-            buf_size += stripes[role].read_end - stripes[role].read_start;
-        }
-    }
-    // Allocate buffer
-    void *buf = memalign(MEM_ALIGNMENT, buf_size);
-    uint64_t buf_pos = add_size;
-    for (int role = 0; role < read_pg_size; role++)
-    {
-        if (stripes[role].read_end != 0)
-        {
-            stripes[role].read_buf = buf + buf_pos;
-            buf_pos += stripes[role].read_end - stripes[role].read_start;
-        }
-    }
-    return buf;
-}
-
-void* calc_rmw_reads(void *write_buf, osd_rmw_stripe_t *stripes, uint64_t *osd_set, uint64_t pg_size, uint64_t pg_minsize, uint64_t pg_cursize)
-{
-    // Generic parity modification (read-modify-write) algorithm
-    // Reconstruct -> Read -> Calc parity -> Write
-    // Now we always read continuous ranges. This means that an update of the beginning
-    // of one data stripe and the end of another will lead to a read of full paired stripes.
-    // FIXME: (Maybe) read small individual ranges in that case instead.
-    uint32_t start = 0, end = 0;
-    for (int role = 0; role < pg_minsize; role++)
-    {
-        if (stripes[role].req_end != 0)
-        {
-            start = !end || stripes[role].req_start < start ? stripes[role].req_start : start;
-            end = std::max(stripes[role].req_end, end);
-            stripes[role].write_start = stripes[role].req_start;
-            stripes[role].write_end = stripes[role].req_end;
-        }
-    }
-    for (int role = 0; role < pg_minsize; role++)
-    {
-        cover_read(start, end, stripes[role]);
-    }
-    int has_parity = 0;
-    for (int role = pg_minsize; role < pg_size; role++)
-    {
-        if (osd_set[role] != 0)
-        {
-            has_parity++;
-            stripes[role].write_start = start;
-            stripes[role].write_end = end;
-        }
-        else
-            stripes[role].missing = true;
-    }
-    if (pg_cursize < pg_size)
-    {
-        if (has_parity == 0)
-        {
-            // Parity is missing, we don't need to read anything
-            for (int role = 0; role < pg_minsize; role++)
-            {
-                stripes[role].read_end = 0;
-            }
-        }
-        else
-        {
-            // Other stripe(s) are missing
-            for (int role = 0; role < pg_minsize; role++)
-            {
-                if (osd_set[role] == 0 && stripes[role].read_end != 0)
-                {
-                    stripes[role].missing = true;
-                    for (int r2 = 0; r2 < pg_size; r2++)
-                    {
-                        // Read the non-covered range of <role> from all other stripes to reconstruct it
-                        if (r2 != role && osd_set[r2] != 0)
-                        {
-                            extend_read(stripes[role].read_start, stripes[role].read_end, stripes[r2]);
-                        }
-                    }
-                }
-            }
-        }
-    }
-    // Allocate read buffers
-    void *rmw_buf = alloc_read_buffer(stripes, pg_size, has_parity * (end - start));
-    // Position parity & write buffers
-    uint64_t buf_pos = 0, in_pos = 0;
-    for (int role = 0; role < pg_size; role++)
-    {
-        if (stripes[role].req_end != 0)
-        {
-            stripes[role].write_buf = write_buf + in_pos;
-            in_pos += stripes[role].req_end - stripes[role].req_start;
-        }
-        else if (role >= pg_minsize && osd_set[role] != 0)
-        {
-            stripes[role].write_buf = rmw_buf + buf_pos;
-            buf_pos += end - start;
-        }
-    }
-    return rmw_buf;
-}
-
-static void get_old_new_buffers(osd_rmw_stripe_t & stripe, uint32_t wr_start, uint32_t wr_end, buf_len_t *bufs, int & nbufs)
-{
-    uint32_t ns = 0, ne = 0, os = 0, oe = 0;
-    if (stripe.req_end > wr_start &&
-        stripe.req_start < wr_end)
-    {
-        ns = std::max(stripe.req_start, wr_start);
-        ne = std::min(stripe.req_end, wr_end);
-    }
-    if (stripe.read_end > wr_start &&
-        stripe.read_start < wr_end)
-    {
-        os = std::max(stripe.read_start, wr_start);
-        oe = std::min(stripe.read_end, wr_end);
-    }
-    if (ne && (!oe || ns <= os))
-    {
-        // NEW or NEW->OLD
-        bufs[nbufs++] = { .buf = stripe.write_buf + ns - stripe.req_start, .len = ne-ns };
-        if (os < ne)
-            os = ne;
-        if (oe > os)
-        {
-            // NEW->OLD
-            bufs[nbufs++] = { .buf = stripe.read_buf + os - stripe.read_start, .len = oe-os };
-        }
-    }
-    else if (oe)
-    {
-        // OLD or OLD->NEW or OLD->NEW->OLD
-        if (ne)
-        {
-            // OLD->NEW or OLD->NEW->OLD
-            bufs[nbufs++] = { .buf = stripe.read_buf + os - stripe.read_start, .len = ns-os };
-            bufs[nbufs++] = { .buf = stripe.write_buf + ns - stripe.req_start, .len = ne-ns };
-            if (oe > ne)
-            {
-                // OLD->NEW->OLD
-                bufs[nbufs++] = { .buf = stripe.read_buf + ne - stripe.read_start, .len = oe-ne };
-            }
-        }
-        else
-        {
-            // OLD
-            bufs[nbufs++] = { .buf = stripe.read_buf + os - stripe.read_start, .len = oe-os };
-        }
-    }
-}
-
-static void xor_multiple_buffers(buf_len_t *xor1, int n1, buf_len_t *xor2, int n2, void *dest, uint32_t len)
-{
-    assert(n1 > 0 && n2 > 0);
-    int i1 = 0, i2 = 0;
-    uint32_t start1 = 0, start2 = 0, end1 = xor1[0].len, end2 = xor2[0].len;
-    uint32_t pos = 0;
-    while (pos < len)
-    {
-        // We know for sure that ranges overlap
-        uint32_t end = std::min(end1, end2);
-        memxor(xor1[i1].buf + pos-start1, xor2[i2].buf + pos-start2, dest+pos, end-pos);
-        pos = end;
-        if (pos >= end1)
-        {
-            i1++;
-            if (i1 >= n1)
-            {
-                assert(pos >= end2);
-                return;
-            }
-            start1 = end1;
-            end1 += xor1[i1].len;
-        }
-        if (pos >= end2)
-        {
-            i2++;
-            start2 = end2;
-            end2 += xor2[i2].len;
-        }
-    }
-}
-
-void calc_rmw_parity(osd_rmw_stripe_t *stripes, int pg_size)
-{
-    if (stripes[pg_size-1].missing)
-    {
-        // Parity OSD is unavailable
-        return;
-    }
-    for (int role = 0; role < pg_size; role++)
-    {
-        if (stripes[role].read_end != 0 && stripes[role].missing)
-        {
-            // Reconstruct missing stripe (EC k+1)
-            reconstruct_stripe(stripes, pg_size, role);
-            break;
-        }
-    }
-    // Calculate new parity (EC k+1)
-    int parity = pg_size-1, prev = -2;
-    auto wr_end = stripes[parity].write_end;
-    auto wr_start = stripes[parity].write_start;
-    for (int other = 0; other < pg_size-1; other++)
-    {
-        if (prev == -2)
-        {
-            prev = other;
-        }
-        else
-        {
-            int n1 = 0, n2 = 0;
-            buf_len_t xor1[3], xor2[3];
-            if (prev == -1)
-            {
-                xor1[n1++] = { .buf = stripes[parity].write_buf, .len = wr_end-wr_start };
-            }
-            else
-            {
-                get_old_new_buffers(stripes[prev], wr_start, wr_end, xor1, n1);
-                prev = -1;
-            }
-            get_old_new_buffers(stripes[other], wr_start, wr_end, xor2, n2);
-            xor_multiple_buffers(xor1, n1, xor2, n2, stripes[parity].write_buf, wr_end-wr_start);
-        }
-    }
-}

osd_rmw.h

@@ -1,36 +0,0 @@
-#pragma once
-
-#include <stdint.h>
-#include "object_id.h"
-#include "osd_id.h"
-
-#ifndef MEM_ALIGNMENT
-#define MEM_ALIGNMENT 512
-#endif
-
-struct buf_len_t
-{
-    void *buf;
-    uint64_t len;
-};
-
-struct osd_rmw_stripe_t
-{
-    void *read_buf, *write_buf;
-    uint32_t req_start, req_end;
-    uint32_t read_start, read_end;
-    uint32_t write_start, write_end;
-    bool missing;
-};
-
-void split_stripes(uint64_t pg_minsize, uint32_t bs_block_size, uint32_t start, uint32_t len, osd_rmw_stripe_t *stripes);
-
-void reconstruct_stripe(osd_rmw_stripe_t *stripes, int pg_size, int role);
-
-int extend_missing_stripes(osd_rmw_stripe_t *stripes, osd_num_t *osd_set, int minsize, int size);
-
-void* alloc_read_buffer(osd_rmw_stripe_t *stripes, int read_pg_size, uint64_t add_size);
-
-void* calc_rmw_reads(void *write_buf, osd_rmw_stripe_t *stripes, uint64_t *osd_set, uint64_t pg_size, uint64_t pg_minsize, uint64_t pg_cursize);
-
-void calc_rmw_parity(osd_rmw_stripe_t *stripes, int pg_size);

osd_rmw_test.cpp

@@ -1,93 +0,0 @@
-#include <string.h>
-#include "osd_rmw.cpp"
-#include "test_pattern.h"
-
-int main(int narg, char *args[])
-{
-    osd_num_t osd_set[3] = { 1, 0, 3 };
-    osd_rmw_stripe_t stripes[3] = { 0 };
-    // Test 1
-    split_stripes(2, 128*1024, 128*1024-4096, 8192, stripes);
-    assert(stripes[0].req_start == 128*1024-4096 && stripes[0].req_end == 128*1024);
-    assert(stripes[1].req_start == 0 && stripes[1].req_end == 4096);
-    assert(stripes[2].req_end == 0);
-    // Test 2
-    for (int i = 0; i < 3; i++)
-    {
-        stripes[i].read_start = stripes[i].req_start;
-        stripes[i].read_end = stripes[i].req_end;
-    }
-    assert(extend_missing_stripes(stripes, osd_set, 2, 3) == 0);
-    assert(stripes[0].read_start == 0 && stripes[0].read_end == 128*1024);
-    assert(stripes[2].read_start == 0 && stripes[2].read_end == 4096);
-    // Test 3
-    stripes[0] = { .req_start = 128*1024-4096, .req_end = 128*1024 };
-    cover_read(0, 128*1024, stripes[0]);
-    assert(stripes[0].read_start == 0 && stripes[0].read_end == 128*1024-4096);
-    // Test 4.1
-    memset(stripes, 0, sizeof(stripes));
-    split_stripes(2, 128*1024, 128*1024-4096, 8192, stripes);
-    void* write_buf = malloc(8192);
-    void* rmw_buf = calc_rmw_reads(write_buf, stripes, osd_set, 3, 2, 2);
-    assert(stripes[0].read_start == 0 && stripes[0].read_end == 128*1024);
-    assert(stripes[1].read_start == 4096 && stripes[1].read_end == 128*1024);
-    assert(stripes[2].read_start == 4096 && stripes[2].read_end == 128*1024);
-    assert(stripes[0].read_buf == rmw_buf+128*1024);
-    assert(stripes[1].read_buf == rmw_buf+128*1024*2);
-    assert(stripes[2].read_buf == rmw_buf+128*1024*3-4096);
-    assert(stripes[0].write_buf == write_buf);
-    assert(stripes[1].write_buf == write_buf+4096);
-    assert(stripes[2].write_buf == rmw_buf);
-    // Test 4.2
-    set_pattern(write_buf, 8192, PATTERN0);
-    set_pattern(stripes[0].read_buf, 128*1024, PATTERN1); // old data
-    set_pattern(stripes[1].read_buf, 128*1024-4096, UINT64_MAX); // didn't read it, it's missing
-    set_pattern(stripes[2].read_buf, 128*1024-4096, 0); // old parity = 0
-    calc_rmw_parity(stripes, 3);
-    check_pattern(stripes[2].write_buf, 4096, PATTERN0^PATTERN1); // new parity
-    check_pattern(stripes[2].write_buf+4096, 128*1024-4096*2, 0); // new parity
-    check_pattern(stripes[2].write_buf+128*1024-4096, 4096, PATTERN0^PATTERN1); // new parity
-    free(rmw_buf);
-    free(write_buf);
-    // Test 5.1
-    memset(stripes, 0, sizeof(stripes));
-    split_stripes(2, 128*1024, 0, 64*1024*3, stripes);
-    assert(stripes[0].req_start == 0 && stripes[0].req_end == 128*1024);
-    assert(stripes[1].req_start == 0 && stripes[1].req_end == 64*1024);
-    assert(stripes[2].req_end == 0);
-    // Test 5.2
-    write_buf = malloc(64*1024*3);
-    rmw_buf = calc_rmw_reads(write_buf, stripes, osd_set, 3, 2, 2);
-    assert(stripes[0].read_start == 64*1024 && stripes[0].read_end == 128*1024);
-    assert(stripes[1].read_start == 64*1024 && stripes[1].read_end == 128*1024);
-    assert(stripes[2].read_start == 64*1024 && stripes[2].read_end == 128*1024);
-    assert(stripes[0].read_buf == rmw_buf+128*1024);
-    assert(stripes[1].read_buf == rmw_buf+64*3*1024);
-    assert(stripes[2].read_buf == rmw_buf+64*4*1024);
-    assert(stripes[0].write_buf == write_buf);
-    assert(stripes[1].write_buf == write_buf+128*1024);
-    assert(stripes[2].write_buf == rmw_buf);
-    free(rmw_buf);
-    free(write_buf);
-    // Test 6.1
-    memset(stripes, 0, sizeof(stripes));
-    split_stripes(2, 128*1024, 0, 64*1024*3, stripes);
-    osd_set[1] = 2;
-    write_buf = malloc(64*1024*3);
-    rmw_buf = calc_rmw_reads(write_buf, stripes, osd_set, 3, 2, 3);
-    assert(stripes[0].read_end == 0);
-    assert(stripes[1].read_start == 64*1024 && stripes[1].read_end == 128*1024);
-    assert(stripes[2].read_end == 0);
-    assert(stripes[0].read_buf == 0);
-    assert(stripes[1].read_buf == rmw_buf+128*1024);
-    assert(stripes[2].read_buf == 0);
-    assert(stripes[0].write_buf == write_buf);
-    assert(stripes[1].write_buf == write_buf+128*1024);
-    assert(stripes[2].write_buf == rmw_buf);
-    free(rmw_buf);
-    free(write_buf);
-    osd_set[1] = 0;
-    // End
-    printf("all ok\n");
-    return 0;
-}

osd_secondary.cpp

@@ -1,150 +0,0 @@
-#include "osd.h"
-
-#include "json11/json11.hpp"
-
-void osd_t::secondary_op_callback(osd_op_t *op)
-{
-    inflight_ops--;
-    auto cl_it = clients.find(op->peer_fd);
-    if (cl_it != clients.end())
-    {
-        op->reply.hdr.magic = SECONDARY_OSD_REPLY_MAGIC;
-        op->reply.hdr.id = op->req.hdr.id;
-        op->reply.hdr.opcode = op->req.hdr.opcode;
-        op->reply.hdr.retval = op->bs_op->retval;
-        if (op->req.hdr.opcode == OSD_OP_SECONDARY_READ ||
-            op->req.hdr.opcode == OSD_OP_SECONDARY_WRITE)
-        {
-            op->reply.sec_rw.version = op->bs_op->version;
-        }
-        else if (op->req.hdr.opcode == OSD_OP_SECONDARY_DELETE)
-        {
-            op->reply.sec_del.version = op->bs_op->version;
-        }
-        if (op->req.hdr.opcode == OSD_OP_SECONDARY_READ &&
-            op->reply.hdr.retval > 0)
-        {
-            op->send_list.push_back(op->buf, op->reply.hdr.retval);
-        }
-        else if (op->req.hdr.opcode == OSD_OP_SECONDARY_LIST)
-        {
-            // allocated by blockstore
-            op->buf = op->bs_op->buf;
-            if (op->reply.hdr.retval > 0)
-            {
-                op->send_list.push_back(op->buf, op->reply.hdr.retval * sizeof(obj_ver_id));
-            }
-            op->reply.sec_list.stable_count = op->bs_op->version;
-        }
-        auto & cl = cl_it->second;
-        outbox_push(cl, op);
-    }
-    else
-    {
-        delete op;
-    }
-}
-
-void osd_t::exec_secondary(osd_op_t *cur_op)
-{
-    cur_op->bs_op = new blockstore_op_t();
-    cur_op->bs_op->callback = [this, cur_op](blockstore_op_t* bs_op) { secondary_op_callback(cur_op); };
-    cur_op->bs_op->opcode = (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_READ ? BS_OP_READ
-        : (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_WRITE ? BS_OP_WRITE
-        : (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_SYNC ? BS_OP_SYNC
-        : (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_STABILIZE ? BS_OP_STABLE
-        : (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_ROLLBACK ? BS_OP_ROLLBACK
-        : (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_DELETE ? BS_OP_DELETE
-        : (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_LIST ? BS_OP_LIST
-        : -1)))))));
-    if (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_READ ||
-        cur_op->req.hdr.opcode == OSD_OP_SECONDARY_WRITE)
-    {
-        cur_op->bs_op->oid = cur_op->req.sec_rw.oid;
-        cur_op->bs_op->version = cur_op->req.sec_rw.version;
-        cur_op->bs_op->offset = cur_op->req.sec_rw.offset;
-        cur_op->bs_op->len = cur_op->req.sec_rw.len;
-        cur_op->bs_op->buf = cur_op->buf;
-#ifdef OSD_STUB
-        cur_op->bs_op->retval = cur_op->bs_op->len;
-#endif
-    }
-    else if (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_DELETE)
-    {
-        cur_op->bs_op->oid = cur_op->req.sec_del.oid;
-        cur_op->bs_op->version = cur_op->req.sec_del.version;
-#ifdef OSD_STUB
-        cur_op->bs_op->retval = 0;
-#endif
-    }
-    else if (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_STABILIZE ||
-        cur_op->req.hdr.opcode == OSD_OP_SECONDARY_ROLLBACK)
-    {
-        cur_op->bs_op->len = cur_op->req.sec_stab.len/sizeof(obj_ver_id);
-        cur_op->bs_op->buf = cur_op->buf;
-#ifdef OSD_STUB
-        cur_op->bs_op->retval = 0;
-#endif
-    }
-    else if (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_LIST)
-    {
-        if (cur_op->req.sec_list.pg_count < cur_op->req.sec_list.list_pg)
-        {
-            // requested pg number is greater than total pg count
-            cur_op->bs_op->retval = -EINVAL;
-            secondary_op_callback(cur_op);
-            return;
-        }
-        cur_op->bs_op->oid.stripe = cur_op->req.sec_list.parity_block_size;
-        cur_op->bs_op->len = cur_op->req.sec_list.pg_count;
-        cur_op->bs_op->offset = cur_op->req.sec_list.list_pg - 1;
-#ifdef OSD_STUB
-        cur_op->bs_op->retval = 0;
-        cur_op->bs_op->buf = NULL;
-#endif
-    }
-#ifdef OSD_STUB
-    secondary_op_callback(cur_op);
-#else
-    bs->enqueue_op(cur_op->bs_op);
-#endif
-}
-
-void osd_t::exec_show_config(osd_op_t *cur_op)
-{
-    // FIXME: Send the real config, not its source
-    std::string cfg_str = json11::Json(config).dump();
-    cur_op->reply.hdr.magic = SECONDARY_OSD_REPLY_MAGIC;
-    cur_op->reply.hdr.id = cur_op->req.hdr.id;
-    cur_op->reply.hdr.opcode = cur_op->req.hdr.opcode;
-    cur_op->reply.hdr.retval = cfg_str.size()+1;
-    cur_op->buf = malloc(cfg_str.size()+1);
-    memcpy(cur_op->buf, cfg_str.c_str(), cfg_str.size()+1);
-    auto & cl = clients[cur_op->peer_fd];
-    cur_op->send_list.push_back(cur_op->buf, cur_op->reply.hdr.retval);
-    outbox_push(cl, cur_op);
-}
-
-void osd_t::exec_sync_stab_all(osd_op_t *cur_op)
-{
-    // Sync and stabilize all objects
-    // This command is only valid for tests
-    cur_op->bs_op = new blockstore_op_t();
-    if (!allow_test_ops)
-    {
-        cur_op->bs_op->retval = -EINVAL;
-        secondary_op_callback(cur_op);
-        return;
-    }
-    cur_op->bs_op->opcode = BS_OP_SYNC_STAB_ALL;
-    cur_op->bs_op->callback = [this, cur_op](blockstore_op_t *bs_op)
-    {
-        secondary_op_callback(cur_op);
-    };
-#ifdef OSD_STUB
-    cur_op->bs_op->retval = 0;
-    secondary_op_callback(cur_op);
-#else
-    bs->enqueue_op(cur_op->bs_op);
-#endif
-}

osd_send.cpp

@@ -1,131 +0,0 @@
-#include "osd.h"
-
-void osd_t::outbox_push(osd_client_t & cl, osd_op_t *cur_op)
-{
-    assert(cur_op->peer_fd);
-    if (cur_op->op_type == OSD_OP_OUT)
-    {
-        clock_gettime(CLOCK_REALTIME, &cur_op->tv_begin);
-    }
-    cl.outbox.push_back(cur_op);
-    if (cl.write_op || cl.outbox.size() > 1 || !try_send(cl))
-    {
-        if (cl.write_state == 0)
-        {
-            cl.write_state = CL_WRITE_READY;
-            write_ready_clients.push_back(cur_op->peer_fd);
-        }
-        ringloop->wakeup();
-    }
-}
-
-bool osd_t::try_send(osd_client_t & cl)
-{
-    int peer_fd = cl.peer_fd;
-    io_uring_sqe* sqe = ringloop->get_sqe();
-    if (!sqe)
-    {
-        return false;
-    }
-    ring_data_t* data = ((ring_data_t*)sqe->user_data);
-    if (!cl.write_op)
-    {
-        // pick next command
-        cl.write_op = cl.outbox.front();
-        cl.outbox.pop_front();
-        cl.write_state = CL_WRITE_REPLY;
-        clock_gettime(CLOCK_REALTIME, &cl.write_op->tv_send);
-        if (cl.write_op->op_type == OSD_OP_IN)
-        {
-            // Measure execution latency
-            timespec tv_end = cl.write_op->tv_send;
-            op_stat_count[cl.write_op->req.hdr.opcode]++;
-            op_stat_sum[cl.write_op->req.hdr.opcode] += (
-                (tv_end.tv_sec - cl.write_op->tv_begin.tv_sec)*1000000 +
-                (tv_end.tv_nsec - cl.write_op->tv_begin.tv_nsec)/1000
-            );
-        }
-    }
-    cl.write_msg.msg_iov = cl.write_op->send_list.get_iovec();
-    cl.write_msg.msg_iovlen = cl.write_op->send_list.get_size();
-    data->callback = [this, peer_fd](ring_data_t *data) { handle_send(data, peer_fd); };
-    my_uring_prep_sendmsg(sqe, peer_fd, &cl.write_msg, 0);
-    return true;
-}
-
-void osd_t::send_replies()
-{
-    for (int i = 0; i < write_ready_clients.size(); i++)
-    {
-        int peer_fd = write_ready_clients[i];
-        if (!try_send(clients[peer_fd]))
-        {
-            write_ready_clients.erase(write_ready_clients.begin(), write_ready_clients.begin() + i);
-            return;
-        }
-    }
-    write_ready_clients.clear();
-}
-
-void osd_t::handle_send(ring_data_t *data, int peer_fd)
-{
-    auto cl_it = clients.find(peer_fd);
-    if (cl_it != clients.end())
-    {
-        auto & cl = cl_it->second;
-        if (data->res < 0 && data->res != -EAGAIN)
-        {
-            // this is a client socket, so don't panic. just disconnect it
-            printf("Client %d socket write error: %d (%s). Disconnecting client\n", peer_fd, -data->res, strerror(-data->res));
-            stop_client(peer_fd);
-            return;
-        }
-        if (data->res >= 0)
-        {
-            osd_op_t *cur_op = cl.write_op;
-            while (data->res > 0 && cur_op->send_list.sent < cur_op->send_list.count)
-            {
-                iovec & iov = cur_op->send_list.buf[cur_op->send_list.sent];
-                if (iov.iov_len <= data->res)
-                {
-                    data->res -= iov.iov_len;
-                    cur_op->send_list.sent++;
-                }
-                else
-                {
-                    iov.iov_len -= data->res;
-                    iov.iov_base += data->res;
-                    break;
-                }
-            }
-            if (cur_op->send_list.sent >= cur_op->send_list.count)
-            {
-                // Done
-                if (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_STABILIZE)
-                {
-                    timespec tv_end;
-                    clock_gettime(CLOCK_REALTIME, &tv_end);
-                    send_stat_count++;
-                    send_stat_sum += (
-                        (tv_end.tv_sec - cl.write_op->tv_send.tv_sec)*1000000 +
-                        (tv_end.tv_nsec - cl.write_op->tv_send.tv_nsec)/1000
-                    );
-                }
-                if (cur_op->op_type == OSD_OP_IN)
-                {
-                    delete cur_op;
-                }
-                else
-                {
-                    cl.sent_ops[cl.write_op->req.hdr.id] = cl.write_op;
-                }
-                cl.write_op = NULL;
-                cl.write_state = cl.outbox.size() > 0 ? CL_WRITE_READY : 0;
-            }
-        }
-        if (cl.write_state != 0)
-        {
-            write_ready_clients.push_back(peer_fd);
-        }
-    }
-}

qemu-3.1-vitastor.patch

@@ -0,0 +1,84 @@
+Index: qemu-3.1+dfsg/qapi/block-core.json
+===================================================================
+--- qemu-3.1+dfsg.orig/qapi/block-core.json
++++ qemu-3.1+dfsg/qapi/block-core.json
+@@ -2617,7 +2617,7 @@
+ ##
+ { 'enum': 'BlockdevDriver',
+   'data': [ 'blkdebug', 'blklogwrites', 'blkverify', 'bochs', 'cloop',
+-            'copy-on-read', 'dmg', 'file', 'ftp', 'ftps', 'gluster',
++            'copy-on-read', 'dmg', 'file', 'ftp', 'ftps', 'gluster', 'vitastor',
+             'host_cdrom', 'host_device', 'http', 'https', 'iscsi', 'luks',
+             'nbd', 'nfs', 'null-aio', 'null-co', 'nvme', 'parallels', 'qcow',
+             'qcow2', 'qed', 'quorum', 'raw', 'rbd', 'replication', 'sheepdog',
+@@ -3367,6 +3367,24 @@
+             '*tag': 'str' } }
+ 
+ ##
++# @BlockdevOptionsVitastor:
++#
++# Driver specific block device options for vitastor
++#
++# @inode:       Inode number
++# @pool:        Pool ID
++# @size:        Desired image size in bytes
++# @etcd_host:   etcd connection address
++# @etcd_prefix: etcd key/value prefix
++##
++{ 'struct': 'BlockdevOptionsVitastor',
++  'data': { 'inode': 'uint64',
++            'pool': 'uint64',
++            'size': 'uint64',
++            'etcd_host': 'str',
++            '*etcd_prefix': 'str' } }
++
++##
+ # @ReplicationMode:
+ #
+ # An enumeration of replication modes.
+@@ -3713,6 +3731,7 @@
+       'rbd':        'BlockdevOptionsRbd',
+       'replication':'BlockdevOptionsReplication',
+       'sheepdog':   'BlockdevOptionsSheepdog',
++      'vitastor':   'BlockdevOptionsVitastor',
+       'ssh':        'BlockdevOptionsSsh',
+       'throttle':   'BlockdevOptionsThrottle',
+       'vdi':        'BlockdevOptionsGenericFormat',
+@@ -4158,6 +4177,17 @@
+             '*block-state-zero':    'bool' } }
+ 
+ ##
++# @BlockdevCreateOptionsVitastor:
++#
++# Driver specific image creation options for Vitastor.
++#
++# @size: Size of the virtual disk in bytes
++##
++{ 'struct': 'BlockdevCreateOptionsVitastor',
++  'data': { 'location':         'BlockdevOptionsVitastor',
++            'size':             'size' } }
++
++##
+ # @BlockdevVpcSubformat:
+ #
+ # @dynamic: Growing image file
+@@ -4212,6 +4242,7 @@
+       'qed':            'BlockdevCreateOptionsQed',
+       'rbd':            'BlockdevCreateOptionsRbd',
+       'sheepdog':       'BlockdevCreateOptionsSheepdog',
++      'vitastor':       'BlockdevCreateOptionsVitastor',
+       'ssh':            'BlockdevCreateOptionsSsh',
+       'vdi':            'BlockdevCreateOptionsVdi',
+       'vhdx':           'BlockdevCreateOptionsVhdx',
+Index: qemu-3.1+dfsg/scripts/modules/module_block.py
+===================================================================
+--- qemu-3.1+dfsg.orig/scripts/modules/module_block.py
++++ qemu-3.1+dfsg/scripts/modules/module_block.py
+@@ -88,6 +88,7 @@ def print_bottom(fheader):
+ output_file = sys.argv[1]
+ with open(output_file, 'w') as fheader:
+     print_top(fheader)
++    add_module(fheader, "vitastor", "vitastor", "vitastor")
+ 
+     for filename in sys.argv[2:]:
+         if os.path.isfile(filename):

qemu-4.2-vitastor.patch

@@ -0,0 +1,84 @@
+Index: qemu/qapi/block-core.json
+===================================================================
+--- qemu.orig/qapi/block-core.json	2020-11-07 22:57:38.932613674 +0000
++++ qemu.orig/qapi/block-core.json	2020-11-07 22:59:49.890722862 +0000
+@@ -2907,7 +2907,7 @@
+             'nbd', 'nfs', 'null-aio', 'null-co', 'nvme', 'parallels', 'qcow',
+             'qcow2', 'qed', 'quorum', 'raw', 'rbd',
+             { 'name': 'replication', 'if': 'defined(CONFIG_REPLICATION)' },
+-            'sheepdog',
++            'sheepdog', 'vitastor',
+             'ssh', 'throttle', 'vdi', 'vhdx', 'vmdk', 'vpc', 'vvfat', 'vxhs' ] }
+ 
+ ##
+@@ -3725,6 +3725,24 @@
+             '*tag': 'str' } }
+ 
+ ##
++# @BlockdevOptionsVitastor:
++#
++# Driver specific block device options for vitastor
++#
++# @inode:       Inode number
++# @pool:        Pool ID
++# @size:        Desired image size in bytes
++# @etcd_host:   etcd connection address
++# @etcd_prefix: etcd key/value prefix
++##
++{ 'struct': 'BlockdevOptionsVitastor',
++  'data': { 'inode': 'uint64',
++            'pool': 'uint64',
++            'size': 'uint64',
++            'etcd_host': 'str',
++            '*etcd_prefix': 'str' } }
++
++##
+ # @ReplicationMode:
+ #
+ # An enumeration of replication modes.
+@@ -4084,6 +4102,7 @@
+       'replication': { 'type': 'BlockdevOptionsReplication',
+                        'if': 'defined(CONFIG_REPLICATION)' },
+       'sheepdog':   'BlockdevOptionsSheepdog',
++      'vitastor':   'BlockdevOptionsVitastor',
+       'ssh':        'BlockdevOptionsSsh',
+       'throttle':   'BlockdevOptionsThrottle',
+       'vdi':        'BlockdevOptionsGenericFormat',
+@@ -4461,6 +4480,17 @@
+             '*cluster-size' :   'size' } }
+ 
+ ##
++# @BlockdevCreateOptionsVitastor:
++#
++# Driver specific image creation options for Vitastor.
++#
++# @size: Size of the virtual disk in bytes
++##
++{ 'struct': 'BlockdevCreateOptionsVitastor',
++  'data': { 'location':         'BlockdevOptionsVitastor',
++            'size':             'size' } }
++
++##
+ # @BlockdevVmdkSubformat:
+ #
+ # Subformat options for VMDK images
+@@ -4722,6 +4752,7 @@
+       'qed':            'BlockdevCreateOptionsQed',
+       'rbd':            'BlockdevCreateOptionsRbd',
+       'sheepdog':       'BlockdevCreateOptionsSheepdog',
++      'vitastor':       'BlockdevCreateOptionsVitastor',
+       'ssh':            'BlockdevCreateOptionsSsh',
+       'vdi':            'BlockdevCreateOptionsVdi',
+       'vhdx':           'BlockdevCreateOptionsVhdx',
+Index: qemu/scripts/modules/module_block.py
+===================================================================
+--- qemu.orig/scripts/modules/module_block.py	2020-11-07 22:57:38.936613739 +0000
++++ qemu/scripts/modules/module_block.py	2020-11-07 22:59:49.890722862 +0000
+@@ -86,6 +86,7 @@ def print_bottom(fheader):
+ output_file = sys.argv[1]
+ with open(output_file, 'w') as fheader:
+     print_top(fheader)
++    add_module(fheader, "vitastor", "vitastor", "vitastor")
+ 
+     for filename in sys.argv[2:]:
+         if os.path.isfile(filename):

qemu-5.0-vitastor.patch

@@ -0,0 +1,84 @@
+Index: qemu/qapi/block-core.json
+===================================================================
+--- qemu.orig/qapi/block-core.json
++++ qemu/qapi/block-core.json
+@@ -2798,7 +2798,7 @@
+             'luks', 'nbd', 'nfs', 'null-aio', 'null-co', 'nvme', 'parallels',
+             'qcow', 'qcow2', 'qed', 'quorum', 'raw', 'rbd',
+             { 'name': 'replication', 'if': 'defined(CONFIG_REPLICATION)' },
+-            'sheepdog',
++            'sheepdog', 'vitastor',
+             'ssh', 'throttle', 'vdi', 'vhdx', 'vmdk', 'vpc', 'vvfat', 'vxhs' ] }
+ 
+ ##
+@@ -3635,6 +3635,24 @@
+             '*tag': 'str' } }
+ 
+ ##
++# @BlockdevOptionsVitastor:
++#
++# Driver specific block device options for vitastor
++#
++# @inode:       Inode number
++# @pool:        Pool ID
++# @size:        Desired image size in bytes
++# @etcd_host:   etcd connection address
++# @etcd_prefix: etcd key/value prefix
++##
++{ 'struct': 'BlockdevOptionsVitastor',
++  'data': { 'inode': 'uint64',
++            'pool': 'uint64',
++            'size': 'uint64',
++            'etcd_host': 'str',
++            '*etcd_prefix': 'str' } }
++
++##
+ # @ReplicationMode:
+ #
+ # An enumeration of replication modes.
+@@ -3995,6 +4013,7 @@
+       'replication': { 'type': 'BlockdevOptionsReplication',
+                        'if': 'defined(CONFIG_REPLICATION)' },
+       'sheepdog':   'BlockdevOptionsSheepdog',
++      'vitastor':   'BlockdevOptionsVitastor',
+       'ssh':        'BlockdevOptionsSsh',
+       'throttle':   'BlockdevOptionsThrottle',
+       'vdi':        'BlockdevOptionsGenericFormat',
+@@ -4365,6 +4384,17 @@
+             '*cluster-size' :   'size' } }
+ 
+ ##
++# @BlockdevCreateOptionsVitastor:
++#
++# Driver specific image creation options for Vitastor.
++#
++# @size: Size of the virtual disk in bytes
++##
++{ 'struct': 'BlockdevCreateOptionsVitastor',
++  'data': { 'location':         'BlockdevOptionsVitastor',
++            'size':             'size' } }
++
++##
+ # @BlockdevVmdkSubformat:
+ #
+ # Subformat options for VMDK images
+@@ -4626,6 +4656,7 @@
+       'qed':            'BlockdevCreateOptionsQed',
+       'rbd':            'BlockdevCreateOptionsRbd',
+       'sheepdog':       'BlockdevCreateOptionsSheepdog',
++      'vitastor':       'BlockdevCreateOptionsVitastor',
+       'ssh':            'BlockdevCreateOptionsSsh',
+       'vdi':            'BlockdevCreateOptionsVdi',
+       'vhdx':           'BlockdevCreateOptionsVhdx',
+Index: qemu/scripts/modules/module_block.py
+===================================================================
+--- qemu.orig/scripts/modules/module_block.py
++++ qemu/scripts/modules/module_block.py
+@@ -85,6 +85,7 @@ def print_bottom(fheader):
+ output_file = sys.argv[1]
+ with open(output_file, 'w') as fheader:
+     print_top(fheader)
++    add_module(fheader, "vitastor", "vitastor", "vitastor")
+ 
+     for filename in sys.argv[2:]:
+         if os.path.isfile(filename):

qemu-5.1-vitastor.patch

@@ -0,0 +1,84 @@
+Index: qemu-5.1+dfsg/qapi/block-core.json
+===================================================================
+--- qemu-5.1+dfsg.orig/qapi/block-core.json
++++ qemu-5.1+dfsg/qapi/block-core.json
+@@ -2807,7 +2807,7 @@
+             'luks', 'nbd', 'nfs', 'null-aio', 'null-co', 'nvme', 'parallels',
+             'qcow', 'qcow2', 'qed', 'quorum', 'raw', 'rbd',
+             { 'name': 'replication', 'if': 'defined(CONFIG_REPLICATION)' },
+-            'sheepdog',
++            'sheepdog', 'vitastor',
+             'ssh', 'throttle', 'vdi', 'vhdx', 'vmdk', 'vpc', 'vvfat' ] }
+ 
+ ##
+@@ -3644,6 +3644,24 @@
+             '*tag': 'str' } }
+ 
+ ##
++# @BlockdevOptionsVitastor:
++#
++# Driver specific block device options for vitastor
++#
++# @inode:       Inode number
++# @pool:        Pool ID
++# @size:        Desired image size in bytes
++# @etcd_host:   etcd connection address
++# @etcd_prefix: etcd key/value prefix
++##
++{ 'struct': 'BlockdevOptionsVitastor',
++  'data': { 'inode': 'uint64',
++            'pool': 'uint64',
++            'size': 'uint64',
++            'etcd_host': 'str',
++            '*etcd_prefix': 'str' } }
++
++##
+ # @ReplicationMode:
+ #
+ # An enumeration of replication modes.
+@@ -3988,6 +4006,7 @@
+       'replication': { 'type': 'BlockdevOptionsReplication',
+                        'if': 'defined(CONFIG_REPLICATION)' },
+       'sheepdog':   'BlockdevOptionsSheepdog',
++      'vitastor':   'BlockdevOptionsVitastor',
+       'ssh':        'BlockdevOptionsSsh',
+       'throttle':   'BlockdevOptionsThrottle',
+       'vdi':        'BlockdevOptionsGenericFormat',
+@@ -4376,6 +4395,17 @@
+             '*cluster-size' :   'size' } }
+ 
+ ##
++# @BlockdevCreateOptionsVitastor:
++#
++# Driver specific image creation options for Vitastor.
++#
++# @size: Size of the virtual disk in bytes
++##
++{ 'struct': 'BlockdevCreateOptionsVitastor',
++  'data': { 'location':         'BlockdevOptionsVitastor',
++            'size':             'size' } }
++
++##
+ # @BlockdevVmdkSubformat:
+ #
+ # Subformat options for VMDK images
+@@ -4637,6 +4667,7 @@
+       'qed':            'BlockdevCreateOptionsQed',
+       'rbd':            'BlockdevCreateOptionsRbd',
+       'sheepdog':       'BlockdevCreateOptionsSheepdog',
++      'vitastor':       'BlockdevCreateOptionsVitastor',
+       'ssh':            'BlockdevCreateOptionsSsh',
+       'vdi':            'BlockdevCreateOptionsVdi',
+       'vhdx':           'BlockdevCreateOptionsVhdx',
+Index: qemu-5.1+dfsg/scripts/modules/module_block.py
+===================================================================
+--- qemu-5.1+dfsg.orig/scripts/modules/module_block.py
++++ qemu-5.1+dfsg/scripts/modules/module_block.py
+@@ -86,6 +86,7 @@ if __name__ == '__main__':
+     output_file = sys.argv[1]
+     with open(output_file, 'w') as fheader:
+         print_top(fheader)
++        add_module(fheader, "vitastor", "vitastor", "vitastor")
+ 
+         for filename in sys.argv[2:]:
+             if os.path.isfile(filename):

rpm/build-tarball.sh

@@ -0,0 +1,51 @@
+#!/bin/bash
+# Vitastor depends on QEMU and FIO headers, but QEMU and FIO don't have -devel packages
+# So we have to copy their headers into the source tarball
+
+set -e
+
+VITASTOR=$(dirname $0)
+VITASTOR=$(realpath "$VITASTOR/..")
+
+if [ -d /opt/rh/gcc-toolset-9 ]; then
+    # CentOS 8
+    EL=8
+    . /opt/rh/gcc-toolset-9/enable
+else
+    # CentOS 7
+    EL=7
+    . /opt/rh/devtoolset-9/enable
+fi
+cd ~/rpmbuild/SPECS
+rpmbuild -bp fio.spec
+perl -i -pe 's/^make V=1/exit 0; make V=1/' qemu*.spec
+rpmbuild -bc qemu*.spec
+perl -i -pe 's/^exit 0; make V=1/make V=1/' qemu*.spec
+cd ~/rpmbuild/BUILD/qemu*/
+rm -rf $VITASTOR/qemu $VITASTOR/fio
+mkdir -p $VITASTOR/qemu/b/qemu
+make -j8 config-host.h
+cp config-host.h $VITASTOR/qemu/b/qemu
+cp -r include $VITASTOR/qemu
+if [ -f qapi-schema.json ]; then
+    # QEMU 2.0
+    make qapi-types.h
+    cp qapi-types.h $VITASTOR/qemu/b/qemu
+else
+    # QEMU 3.0+
+    make qapi
+    cp -r qapi $VITASTOR/qemu/b/qemu
+fi
+cd $VITASTOR
+sh copy-qemu-includes.sh
+rm -rf qemu
+mv qemu-copy qemu
+ln -s ~/rpmbuild/BUILD/fio*/ fio
+sh copy-fio-includes.sh
+rm fio
+mv fio-copy fio
+FIO=`rpm -qi fio | perl -e 'while(<>) { /^Epoch[\s:]+(\S+)/ && print "$1:"; /^Version[\s:]+(\S+)/ && print $1; /^Release[\s:]+(\S+)/ && print "-$1"; }'`
+QEMU=`rpm -qi qemu qemu-kvm | perl -e 'while(<>) { /^Epoch[\s:]+(\S+)/ && print "$1:"; /^Version[\s:]+(\S+)/ && print $1; /^Release[\s:]+(\S+)/ && print "-$1"; }'`
+perl -i -pe 's/(Requires:\s*fio)([^\n]+)?/$1 = '$FIO'/' $VITASTOR/rpm/vitastor-el$EL.spec
+perl -i -pe 's/(Requires:\s*qemu(?:-kvm)?)([^\n]+)?/$1 = '$QEMU'/' $VITASTOR/rpm/vitastor-el$EL.spec
+tar --transform 's#^#vitastor-0.5.10/#' --exclude 'rpm/*.rpm' -czf $VITASTOR/../vitastor-0.5.10$(rpm --eval '%dist').tar.gz *

rpm/qemu-el8.Dockerfile

@@ -0,0 +1,31 @@
+# Build packages for CentOS 8 inside a container
+# cd ..; podman build -t qemu-el8 -v `pwd`/packages:/root/packages -f rpm/qemu-el8.Dockerfile .
+
+FROM centos:8
+
+WORKDIR /root
+
+RUN rm -f /etc/yum.repos.d/CentOS-Media.repo
+RUN dnf -y install centos-release-advanced-virtualization epel-release dnf-plugins-core rpm-build
+RUN rm -rf /var/lib/dnf/*; dnf download --disablerepo='*' --enablerepo='centos-advanced-virtualization-source' --source qemu-kvm
+RUN rpm --nomd5 -i qemu*.src.rpm
+RUN cd ~/rpmbuild/SPECS && dnf builddep -y --enablerepo=PowerTools --spec qemu-kvm.spec
+
+ADD qemu-*-vitastor.patch /root/vitastor/
+
+RUN set -e; \
+    mkdir -p /root/packages/qemu-el8; \
+    rm -rf /root/packages/qemu-el8/*; \
+    rpm --nomd5 -i /root/qemu*.src.rpm; \
+    cd ~/rpmbuild/SPECS; \
+    PN=$(grep ^Patch qemu-kvm.spec | tail -n1 | perl -pe 's/Patch(\d+).*/$1/'); \
+    csplit qemu-kvm.spec "/^Patch$PN/"; \
+    cat xx00 > qemu-kvm.spec; \
+    head -n 1 xx01 >> qemu-kvm.spec; \
+    echo "Patch$((PN+1)): qemu-4.2-vitastor.patch" >> qemu-kvm.spec; \
+    tail -n +2 xx01 >> qemu-kvm.spec; \
+    perl -i -pe 's/(^Release:\s*\d+)/$1.vitastor/' qemu-kvm.spec; \
+    cp /root/vitastor/qemu-4.2-vitastor.patch ~/rpmbuild/SOURCES; \
+    rpmbuild --nocheck -ba qemu-kvm.spec; \
+    cp ~/rpmbuild/RPMS/*/*qemu* /root/packages/qemu-el8/; \
+    cp ~/rpmbuild/SRPMS/*qemu* /root/packages/qemu-el8/

rpm/qemu-kvm-el7.spec.patch

@@ -0,0 +1,257 @@
+--- qemu-kvm.spec.orig	2020-11-09 23:41:03.000000000 +0000
++++ qemu-kvm.spec	2020-12-06 10:44:24.207640963 +0000
+@@ -2,7 +2,7 @@
+ %global SLOF_gittagcommit 899d9883
+ 
+ %global have_usbredir 1
+-%global have_spice    1
++%global have_spice    0
+ %global have_opengl   1
+ %global have_fdt      0
+ %global have_gluster  1
+@@ -56,7 +56,7 @@ Requires: %{name}-block-curl = %{epoch}:
+ Requires: %{name}-block-gluster = %{epoch}:%{version}-%{release} \
+ %endif                                                           \
+ Requires: %{name}-block-iscsi = %{epoch}:%{version}-%{release}   \
+-Requires: %{name}-block-rbd = %{epoch}:%{version}-%{release}     \
++#Requires: %{name}-block-rbd = %{epoch}:%{version}-%{release}     \
+ Requires: %{name}-block-ssh = %{epoch}:%{version}-%{release}
+ 
+ # Macro to properly setup RHEL/RHEV conflict handling
+@@ -67,7 +67,7 @@ Obsoletes: %1-rhev
+ Summary: QEMU is a machine emulator and virtualizer
+ Name: qemu-kvm
+ Version: 4.2.0
+-Release: 29.vitastor%{?dist}.6
++Release: 30.vitastor%{?dist}.6
+ # Epoch because we pushed a qemu-1.0 package. AIUI this can't ever be dropped
+ Epoch: 15
+ License: GPLv2 and GPLv2+ and CC-BY
+@@ -99,8 +99,8 @@ Source30: kvm-s390x.conf
+ Source31: kvm-x86.conf
+ Source32: qemu-pr-helper.service
+ Source33: qemu-pr-helper.socket
+-Source34: 81-kvm-rhel.rules
+-Source35: udev-kvm-check.c
++#Source34: 81-kvm-rhel.rules
++#Source35: udev-kvm-check.c
+ Source36: README.tests
+ 
+ 
+@@ -825,7 +825,9 @@ Patch331: kvm-Drop-bogus-IPv6-messages.p
+ Patch333: kvm-virtiofsd-Whitelist-fchmod.patch
+ # For bz#1883869 - virtiofsd core dump in KATA Container [rhel-8.2.1.z]
+ Patch334: kvm-virtiofsd-avoid-proc-self-fd-tempdir.patch
+-Patch335: qemu-4.2-vitastor.patch
++Patch335: qemu-use-sphinx-1.2.patch
++Patch336: qemu-config-tcmalloc-warning.patch
++Patch337: qemu-4.2-vitastor.patch
+ 
+ BuildRequires: wget
+ BuildRequires: rpm-build
+@@ -842,7 +844,8 @@ BuildRequires: pciutils-devel
+ BuildRequires: libiscsi-devel
+ BuildRequires: ncurses-devel
+ BuildRequires: libattr-devel
+-BuildRequires: libusbx-devel >= 1.0.22
++BuildRequires: gperftools-devel
++BuildRequires: libusbx-devel >= 1.0.21
+ %if %{have_usbredir}
+ BuildRequires: usbredir-devel >= 0.7.1
+ %endif
+@@ -856,12 +859,12 @@ BuildRequires: virglrenderer-devel
+ # For smartcard NSS support
+ BuildRequires: nss-devel
+ %endif
+-BuildRequires: libseccomp-devel >= 2.4.0
++#Requires: libseccomp >= 2.4.0
+ # For network block driver
+ BuildRequires: libcurl-devel
+ BuildRequires: libssh-devel
+-BuildRequires: librados-devel
+-BuildRequires: librbd-devel
++#BuildRequires: librados-devel
++#BuildRequires: librbd-devel
+ %if %{have_gluster}
+ # For gluster block driver
+ BuildRequires: glusterfs-api-devel
+@@ -955,25 +958,25 @@ hardware for a full system such as a PC
+ 
+ %package -n qemu-kvm-core
+ Summary: qemu-kvm core components
++Requires: gperftools-libs
+ Requires: qemu-img = %{epoch}:%{version}-%{release}
+ %ifarch %{ix86} x86_64
+ Requires: seabios-bin >= 1.10.2-1
+ Requires: sgabios-bin
+-Requires: edk2-ovmf
+ %endif
+ %ifarch aarch64
+ Requires: edk2-aarch64
+ %endif
+ 
+ %ifnarch aarch64 s390x
+-Requires: seavgabios-bin >= 1.12.0-3
+-Requires: ipxe-roms-qemu >= 20170123-1
++Requires: seavgabios-bin >= 1.11.0-1
++Requires: ipxe-roms-qemu >= 20181214-1
++Requires: /usr/share/ipxe.efi
+ %endif
+ %ifarch %{power64}
+ Requires: SLOF >= %{SLOF_gittagdate}-1.git%{SLOF_gittagcommit}
+ %endif
+ Requires: %{name}-common = %{epoch}:%{version}-%{release}
+-Requires: libseccomp >= 2.4.0
+ # For compressed guest memory dumps
+ Requires: lzo snappy
+ %if %{have_kvm_setup}
+@@ -1085,15 +1088,15 @@ This package provides the additional iSC
+ Install this package if you want to access iSCSI volumes.
+ 
+ 
+-%package  block-rbd
+-Summary: QEMU Ceph/RBD block driver
+-Requires: %{name}-common%{?_isa} = %{epoch}:%{version}-%{release}
+-
+-%description block-rbd
+-This package provides the additional Ceph/RBD block driver for QEMU.
+-
+-Install this package if you want to access remote Ceph volumes
+-using the rbd protocol.
++#%package  block-rbd
++#Summary: QEMU Ceph/RBD block driver
++#Requires: %{name}-common%{?_isa} = %{epoch}:%{version}-%{release}
++#
++#%description block-rbd
++#This package provides the additional Ceph/RBD block driver for QEMU.
++#
++#Install this package if you want to access remote Ceph volumes
++#using the rbd protocol.
+ 
+ 
+ %package  block-ssh
+@@ -1117,12 +1120,14 @@ the Secure Shell (SSH) protocol.
+ # --build-id option is used for giving info to the debug packages.
+ buildldflags="VL_LDFLAGS=-Wl,--build-id"
+ 
+-%global block_drivers_list qcow2,raw,file,host_device,nbd,iscsi,rbd,blkdebug,luks,null-co,nvme,copy-on-read,throttle
++#%global block_drivers_list qcow2,raw,file,host_device,nbd,iscsi,rbd,blkdebug,luks,null-co,nvme,copy-on-read,throttle
++%global block_drivers_list qcow2,raw,file,host_device,nbd,iscsi,blkdebug,luks,null-co,nvme,copy-on-read,throttle
+ 
+ %if 0%{have_gluster}
+     %global block_drivers_list %{block_drivers_list},gluster
+ %endif
+ 
++[ -e /usr/bin/sphinx-build ] || ln -s sphinx-build-3 /usr/bin/sphinx-build
+ ./configure  \
+  --prefix="%{_prefix}" \
+  --libdir="%{_libdir}" \
+@@ -1152,15 +1157,15 @@ buildldflags="VL_LDFLAGS=-Wl,--build-id"
+ %else
+   --disable-numa \
+ %endif
+-  --enable-rbd \
++  --disable-rbd \
+ %if 0%{have_librdma}
+   --enable-rdma \
+ %else
+   --disable-rdma \
+ %endif
+   --disable-pvrdma \
+-  --enable-seccomp \
+-%if 0%{have_spice}
++  --disable-seccomp \
++%if %{have_spice}
+   --enable-spice \
+   --enable-smartcard \
+   --enable-virglrenderer \
+@@ -1179,7 +1184,7 @@ buildldflags="VL_LDFLAGS=-Wl,--build-id"
+ %else
+   --disable-usb-redir \
+ %endif
+-  --disable-tcmalloc \
++  --enable-tcmalloc \
+ %ifarch x86_64
+   --enable-libpmem \
+ %else
+@@ -1193,9 +1198,7 @@ buildldflags="VL_LDFLAGS=-Wl,--build-id"
+ %endif
+   --python=%{__python3} \
+   --target-list="%{buildarch}" \
+-  --block-drv-rw-whitelist=%{block_drivers_list} \
+   --audio-drv-list= \
+-  --block-drv-ro-whitelist=vmdk,vhdx,vpc,https,ssh \
+   --with-coroutine=ucontext \
+   --tls-priority=NORMAL \
+   --disable-bluez \
+@@ -1262,7 +1265,7 @@ buildldflags="VL_LDFLAGS=-Wl,--build-id"
+   --disable-sanitizers \
+   --disable-hvf \
+   --disable-whpx \
+-  --enable-malloc-trim \
++  --disable-malloc-trim \
+   --disable-membarrier \
+   --disable-vhost-crypto \
+   --disable-libxml2 \
+@@ -1308,7 +1311,7 @@ make V=1 %{?_smp_mflags} $buildldflags
+ cp -a %{kvm_target}-softmmu/qemu-system-%{kvm_target} qemu-kvm
+ 
+ gcc %{SOURCE6} $RPM_OPT_FLAGS $RPM_LD_FLAGS -o ksmctl
+-gcc %{SOURCE35} $RPM_OPT_FLAGS $RPM_LD_FLAGS -o udev-kvm-check
++#gcc %{SOURCE35} $RPM_OPT_FLAGS $RPM_LD_FLAGS -o udev-kvm-check
+ 
+ %install
+ %define _udevdir %(pkg-config --variable=udevdir udev)
+@@ -1343,8 +1346,8 @@ mkdir -p $RPM_BUILD_ROOT%{testsdir}/test
+ mkdir -p $RPM_BUILD_ROOT%{testsdir}/tests/qemu-iotests
+ mkdir -p $RPM_BUILD_ROOT%{testsdir}/scripts/qmp
+ 
+-install -p -m 0755 udev-kvm-check $RPM_BUILD_ROOT%{_udevdir}
+-install -p -m 0644 %{SOURCE34} $RPM_BUILD_ROOT%{_udevrulesdir}
++#install -p -m 0755 udev-kvm-check $RPM_BUILD_ROOT%{_udevdir}
++#install -p -m 0644 %{SOURCE34} $RPM_BUILD_ROOT%{_udevrulesdir}
+ 
+ install -m 0644 scripts/dump-guest-memory.py \
+                 $RPM_BUILD_ROOT%{_datadir}/%{name}
+@@ -1562,6 +1565,8 @@ rm -rf $RPM_BUILD_ROOT%{qemudocdir}/inte
+ # Remove spec
+ rm -rf $RPM_BUILD_ROOT%{qemudocdir}/specs
+ 
++%global __os_install_post %(echo '%{__os_install_post}' | sed -e 's!/usr/lib[^[:space:]]*/brp-python-bytecompile[[:space:]].*$!!g')
++
+ %check
+ export DIFF=diff; make check V=1
+ 
+@@ -1645,8 +1650,8 @@ useradd -r -u 107 -g qemu -G kvm -d / -s
+ %config(noreplace) %{_sysconfdir}/sysconfig/ksm
+ %{_unitdir}/ksmtuned.service
+ %{_sbindir}/ksmtuned
+-%{_udevdir}/udev-kvm-check
+-%{_udevrulesdir}/81-kvm-rhel.rules
++#%{_udevdir}/udev-kvm-check
++#%{_udevrulesdir}/81-kvm-rhel.rules
+ %ghost %{_sysconfdir}/kvm
+ %config(noreplace) %{_sysconfdir}/ksmtuned.conf
+ %dir %{_sysconfdir}/%{name}
+@@ -1711,8 +1716,8 @@ useradd -r -u 107 -g qemu -G kvm -d / -s
+ %{_libexecdir}/vhost-user-gpu
+ %{_datadir}/%{name}/vhost-user/50-qemu-gpu.json
+ %endif
+-%{_libexecdir}/virtiofsd
+-%{_datadir}/%{name}/vhost-user/50-qemu-virtiofsd.json
++#%{_libexecdir}/virtiofsd
++#%{_datadir}/%{name}/vhost-user/50-qemu-virtiofsd.json
+ 
+ %files -n qemu-img
+ %defattr(-,root,root)
+@@ -1748,8 +1753,8 @@ useradd -r -u 107 -g qemu -G kvm -d / -s
+ %files block-iscsi
+ %{_libdir}/qemu-kvm/block-iscsi.so
+ 
+-%files block-rbd
+-%{_libdir}/qemu-kvm/block-rbd.so
++#%files block-rbd
++#%{_libdir}/qemu-kvm/block-rbd.so
+ 
+ %files block-ssh
+ %{_libdir}/qemu-kvm/block-ssh.so

rpm/qemu-kvm.spec.patch

@@ -0,0 +1,29 @@
+--- qemu-kvm.spec	2020-12-05 13:13:54.388623517 +0000
++++ qemu-kvm.spec	2020-12-05 13:13:58.728696598 +0000
+@@ -67,7 +67,7 @@ Obsoletes: %1-rhev
+ Summary: QEMU is a machine emulator and virtualizer
+ Name: qemu-kvm
+ Version: 4.2.0
+-Release: 29%{?dist}.6
++Release: 29.vitastor%{?dist}.6
+ # Epoch because we pushed a qemu-1.0 package. AIUI this can't ever be dropped
+ Epoch: 15
+ License: GPLv2 and GPLv2+ and CC-BY
+@@ -825,6 +825,7 @@ Patch331: kvm-Drop-bogus-IPv6-messages.p
+ Patch333: kvm-virtiofsd-Whitelist-fchmod.patch
+ # For bz#1883869 - virtiofsd core dump in KATA Container [rhel-8.2.1.z]
+ Patch334: kvm-virtiofsd-avoid-proc-self-fd-tempdir.patch
++Patch335: qemu-4.2-vitastor.patch
+ 
+ BuildRequires: wget
+ BuildRequires: rpm-build
+@@ -1192,9 +1193,7 @@ buildldflags="VL_LDFLAGS=-Wl,--build-id"
+ %endif
+   --python=%{__python3} \
+   --target-list="%{buildarch}" \
+-  --block-drv-rw-whitelist=%{block_drivers_list} \
+   --audio-drv-list= \
+-  --block-drv-ro-whitelist=vmdk,vhdx,vpc,https,ssh \
+   --with-coroutine=ucontext \
+   --tls-priority=NORMAL \
+   --disable-bluez \

rpm/vitastor-el7.Dockerfile

@@ -0,0 +1,47 @@
+# Build packages for CentOS 7 inside a container
+# cd ..; podman build -t vitastor-el7 -v `pwd`/packages:/root/packages -f rpm/vitastor-el7.Dockerfile .
+# localedef -i ru_RU -f UTF-8 ru_RU.UTF-8
+
+FROM centos:7
+
+WORKDIR /root
+
+RUN rm -f /etc/yum.repos.d/CentOS-Media.repo
+RUN yum -y --enablerepo=extras install centos-release-scl epel-release yum-utils rpm-build
+RUN yum -y install https://vitastor.io/rpms/centos/7/vitastor-release-1.0-1.el7.noarch.rpm
+RUN yum -y install devtoolset-9-gcc-c++ devtoolset-9-libatomic-devel gperftools-devel qemu-kvm fio rh-nodejs12 jerasure-devel gf-complete-devel
+RUN yumdownloader --disablerepo=centos-sclo-rh --source qemu-kvm
+RUN yumdownloader --disablerepo=centos-sclo-rh --source fio
+RUN rpm --nomd5 -i qemu*.src.rpm
+RUN rpm --nomd5 -i fio*.src.rpm
+RUN rm -f /etc/yum.repos.d/CentOS-Media.repo
+RUN cd ~/rpmbuild/SPECS && yum-builddep -y --enablerepo='*' --disablerepo=centos-sclo-rh --disablerepo=centos-sclo-rh-source --disablerepo=centos-sclo-sclo-testing qemu-kvm.spec
+RUN cd ~/rpmbuild/SPECS && yum-builddep -y --enablerepo='*' --disablerepo=centos-sclo-rh --disablerepo=centos-sclo-rh-source --disablerepo=centos-sclo-sclo-testing fio.spec
+
+ADD https://vitastor.io/rpms/liburing-el7/liburing-0.7-2.el7.src.rpm /root
+
+RUN set -e; \
+    rpm -i liburing*.src.rpm; \
+    cd ~/rpmbuild/SPECS/; \
+    . /opt/rh/devtoolset-9/enable; \
+    rpmbuild -ba liburing.spec; \
+    mkdir -p /root/packages/liburing-el7; \
+    rm -rf /root/packages/liburing-el7/*; \
+    cp ~/rpmbuild/RPMS/*/liburing* /root/packages/liburing-el7/; \
+    cp ~/rpmbuild/SRPMS/liburing* /root/packages/liburing-el7/
+
+RUN rpm -i `ls /root/packages/liburing-el7/liburing-*.x86_64.rpm | grep -v debug`
+
+ADD . /root/vitastor
+
+RUN set -e; \
+    cd /root/vitastor/rpm; \
+    sh build-tarball.sh; \
+    cp /root/vitastor-0.5.10.el7.tar.gz ~/rpmbuild/SOURCES; \
+    cp vitastor-el7.spec ~/rpmbuild/SPECS/vitastor.spec; \
+    cd ~/rpmbuild/SPECS/; \
+    rpmbuild -ba vitastor.spec; \
+    mkdir -p /root/packages/vitastor-el7; \
+    rm -rf /root/packages/vitastor-el7/*; \
+    cp ~/rpmbuild/RPMS/*/vitastor* /root/packages/vitastor-el7/; \
+    cp ~/rpmbuild/SRPMS/vitastor* /root/packages/vitastor-el7/

rpm/vitastor-el7.spec

@@ -0,0 +1,69 @@
+Name:           vitastor
+Version:        0.5.10
+Release:        1%{?dist}
+Summary:        Vitastor, a fast software-defined clustered block storage
+
+License:        Vitastor Network Public License 1.1
+URL:            https://vitastor.io/
+Source0:        vitastor-0.5.10.el7.tar.gz
+
+BuildRequires:  liburing-devel >= 0.6
+BuildRequires:  gperftools-devel
+BuildRequires:  devtoolset-9-gcc-c++
+BuildRequires:  rh-nodejs12
+BuildRequires:  rh-nodejs12-npm
+BuildRequires:  jerasure-devel
+BuildRequires:  gf-complete-devel
+BuildRequires:  cmake
+Requires:       fio = 3.7-1.el7
+Requires:       qemu-kvm = 2.0.0-1.el7.6
+Requires:       rh-nodejs12
+Requires:       rh-nodejs12-npm
+Requires:       liburing >= 0.6
+Requires:       libJerasure2
+Requires:       lpsolve
+
+%description
+Vitastor is a small, simple and fast clustered block storage (storage for VM drives),
+architecturally similar to Ceph which means strong consistency, primary-replication,
+symmetric clustering and automatic data distribution over any number of drives of any
+size with configurable redundancy (replication or erasure codes/XOR).
+
+
+%prep
+%setup -q
+
+
+%build
+. /opt/rh/devtoolset-9/enable
+%cmake . -DQEMU_PLUGINDIR=qemu-kvm
+%make_build
+
+
+%install
+rm -rf $RPM_BUILD_ROOT
+%make_install
+. /opt/rh/rh-nodejs12/enable
+cd mon
+npm install
+cd ..
+mkdir -p %buildroot/usr/lib/vitastor
+cp -r mon %buildroot/usr/lib/vitastor/mon
+
+
+%files
+%doc
+%_bindir/vitastor-dump-journal
+%_bindir/vitastor-nbd
+%_bindir/vitastor-osd
+%_bindir/vitastor-rm
+%_libdir/qemu-kvm/block-vitastor.so
+%_libdir/libfio_vitastor.so
+%_libdir/libfio_vitastor_blk.so
+%_libdir/libfio_vitastor_sec.so
+%_libdir/libvitastor_blk.so
+%_libdir/libvitastor_client.so
+/usr/lib/vitastor
+
+
+%changelog

rpm/vitastor-el8.Dockerfile

@@ -0,0 +1,45 @@
+# Build packages for CentOS 8 inside a container
+# cd ..; podman build -t vitastor-el8 -v `pwd`/packages:/root/packages -f rpm/vitastor-el8.Dockerfile .
+
+FROM centos:8
+
+WORKDIR /root
+
+RUN rm -f /etc/yum.repos.d/CentOS-Media.repo
+RUN dnf -y install centos-release-advanced-virtualization epel-release dnf-plugins-core
+RUN yum -y install https://vitastor.io/rpms/centos/8/vitastor-release-1.0-1.el8.noarch.rpm
+RUN dnf --enablerepo='centos-advanced-virtualization' -y install gcc-toolset-9 gcc-toolset-9-gcc-c++ gperftools-devel qemu-kvm fio nodejs rpm-build jerasure-devel gf-complete-devel
+RUN rm -rf /var/lib/dnf/*; dnf download --disablerepo='*' --enablerepo='vitastor' --source qemu-kvm
+RUN dnf download --source fio
+RUN rpm --nomd5 -i qemu*.src.rpm
+RUN rpm --nomd5 -i fio*.src.rpm
+RUN cd ~/rpmbuild/SPECS && dnf builddep -y --enablerepo=powertools --spec qemu-kvm.spec
+RUN cd ~/rpmbuild/SPECS && dnf builddep -y --enablerepo=powertools --spec fio.spec && dnf install -y cmake
+
+ADD https://vitastor.io/rpms/liburing-el7/liburing-0.7-2.el7.src.rpm /root
+
+RUN set -e; \
+    rpm -i liburing*.src.rpm; \
+    cd ~/rpmbuild/SPECS/; \
+    . /opt/rh/gcc-toolset-9/enable; \
+    rpmbuild -ba liburing.spec; \
+    mkdir -p /root/packages/liburing-el8; \
+    rm -rf /root/packages/liburing-el8/*; \
+    cp ~/rpmbuild/RPMS/*/liburing* /root/packages/liburing-el8/; \
+    cp ~/rpmbuild/SRPMS/liburing* /root/packages/liburing-el8/
+
+RUN rpm -i `ls /root/packages/liburing-el7/liburing-*.x86_64.rpm | grep -v debug`
+
+ADD . /root/vitastor
+
+RUN set -e; \
+    cd /root/vitastor/rpm; \
+    sh build-tarball.sh; \
+    cp /root/vitastor-0.5.10.el8.tar.gz ~/rpmbuild/SOURCES; \
+    cp vitastor-el8.spec ~/rpmbuild/SPECS/vitastor.spec; \
+    cd ~/rpmbuild/SPECS/; \
+    rpmbuild -ba vitastor.spec; \
+    mkdir -p /root/packages/vitastor-el8; \
+    rm -rf /root/packages/vitastor-el8/*; \
+    cp ~/rpmbuild/RPMS/*/vitastor* /root/packages/vitastor-el8/; \
+    cp ~/rpmbuild/SRPMS/vitastor* /root/packages/vitastor-el8/

rpm/vitastor-el8.spec

@@ -0,0 +1,66 @@
+Name:           vitastor
+Version:        0.5.10
+Release:        1%{?dist}
+Summary:        Vitastor, a fast software-defined clustered block storage
+
+License:        Vitastor Network Public License 1.1
+URL:            https://vitastor.io/
+Source0:        vitastor-0.5.10.el8.tar.gz
+
+BuildRequires:  liburing-devel >= 0.6
+BuildRequires:  gperftools-devel
+BuildRequires:  gcc-toolset-9-gcc-c++
+BuildRequires:  nodejs >= 10
+BuildRequires:  jerasure-devel
+BuildRequires:  gf-complete-devel
+BuildRequires:  cmake
+Requires:       fio = 3.7-3.el8
+Requires:       qemu-kvm = 4.2.0-29.el8.6
+Requires:       nodejs >= 10
+Requires:       liburing >= 0.6
+Requires:       libJerasure2
+Requires:       lpsolve
+
+%description
+Vitastor is a small, simple and fast clustered block storage (storage for VM drives),
+architecturally similar to Ceph which means strong consistency, primary-replication,
+symmetric clustering and automatic data distribution over any number of drives of any
+size with configurable redundancy (replication or erasure codes/XOR).
+
+
+%prep
+%setup -q
+
+
+%build
+. /opt/rh/gcc-toolset-9/enable
+%cmake . -DQEMU_PLUGINDIR=qemu-kvm
+%make_build
+
+
+%install
+rm -rf $RPM_BUILD_ROOT
+%make_install
+cd mon
+npm install
+cd ..
+mkdir -p %buildroot/usr/lib/vitastor
+cp -r mon %buildroot/usr/lib/vitastor
+
+
+%files
+%doc
+%_bindir/vitastor-dump-journal
+%_bindir/vitastor-nbd
+%_bindir/vitastor-osd
+%_bindir/vitastor-rm
+%_libdir/qemu-kvm/block-vitastor.so
+%_libdir/libfio_vitastor.so
+%_libdir/libfio_vitastor_blk.so
+%_libdir/libfio_vitastor_sec.so
+%_libdir/libvitastor_blk.so
+%_libdir/libvitastor_client.so
+/usr/lib/vitastor
+
+
+%changelog

src/CMakeLists.txt

@@ -0,0 +1,188 @@
+cmake_minimum_required(VERSION 2.8)
+
+project(vitastor)
+
+include(GNUInstallDirs)
+
+set(QEMU_PLUGINDIR qemu CACHE STRING "QEMU plugin directory suffix (qemu-kvm on RHEL)")
+set(WITH_ASAN false CACHE BOOL "Build with AddressSanitizer")
+if("${CMAKE_INSTALL_PREFIX}" MATCHES "^/usr/local/?$")
+	if(EXISTS "/etc/debian_version")
+		set(CMAKE_INSTALL_LIBDIR "lib/${CMAKE_LIBRARY_ARCHITECTURE}")
+	endif()
+	set(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_LIBDIR}")
+endif()
+
+add_definitions(-DVERSION="0.6-dev")
+add_definitions(-Wall -Wno-sign-compare -Wno-comment -Wno-parentheses -Wno-pointer-arith)
+if (${WITH_ASAN})
+	add_definitions(-fsanitize=address -fno-omit-frame-pointer)
+	add_link_options(-fsanitize=address -fno-omit-frame-pointer)
+endif (${WITH_ASAN})
+
+set(CMAKE_BUILD_TYPE RelWithDebInfo)
+string(REGEX REPLACE "([\\/\\-]O)[12]?" "\\13" CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE}")
+string(REGEX REPLACE "([\\/\\-]O)[12]?" "\\13" CMAKE_CXX_FLAGS_MINSIZEREL "${CMAKE_CXX_FLAGS_MINSIZEREL}")
+string(REGEX REPLACE "([\\/\\-]O)[12]?" "\\13" CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO}")
+string(REGEX REPLACE "([\\/\\-]D) *NDEBUG" "" CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE}")
+string(REGEX REPLACE "([\\/\\-]D) *NDEBUG" "" CMAKE_CXX_FLAGS_MINSIZEREL "${CMAKE_CXX_FLAGS_MINSIZEREL}")
+string(REGEX REPLACE "([\\/\\-]D) *NDEBUG" "" CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO}")
+string(REGEX REPLACE "([\\/\\-]O)[12]?" "\\13" CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_RELEASE}")
+string(REGEX REPLACE "([\\/\\-]O)[12]?" "\\13" CMAKE_C_FLAGS_MINSIZEREL "${CMAKE_C_FLAGS_MINSIZEREL}")
+string(REGEX REPLACE "([\\/\\-]O)[12]?" "\\13" CMAKE_C_FLAGS_RELWITHDEBINFO "${CMAKE_C_FLAGS_RELWITHDEBINFO}")
+string(REGEX REPLACE "([\\/\\-]D) *NDEBUG" "" CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_RELEASE}")
+string(REGEX REPLACE "([\\/\\-]D) *NDEBUG" "" CMAKE_C_FLAGS_MINSIZEREL "${CMAKE_C_FLAGS_MINSIZEREL}")
+string(REGEX REPLACE "([\\/\\-]D) *NDEBUG" "" CMAKE_C_FLAGS_RELWITHDEBINFO "${CMAKE_C_FLAGS_RELWITHDEBINFO}")
+
+find_package(PkgConfig)
+pkg_check_modules(LIBURING REQUIRED liburing)
+pkg_check_modules(GLIB REQUIRED glib-2.0)
+
+include_directories(
+	../
+	/usr/include/jerasure
+	${LIBURING_INCLUDE_DIRS}
+)
+
+# libvitastor_blk.so
+add_library(vitastor_blk SHARED
+	allocator.cpp blockstore.cpp blockstore_impl.cpp blockstore_init.cpp blockstore_open.cpp blockstore_journal.cpp blockstore_read.cpp
+	blockstore_write.cpp blockstore_sync.cpp blockstore_stable.cpp blockstore_rollback.cpp blockstore_flush.cpp crc32c.c ringloop.cpp
+)
+target_link_libraries(vitastor_blk
+	${LIBURING_LIBRARIES}
+	tcmalloc_minimal
+)
+
+# libfio_vitastor_blk.so
+add_library(fio_vitastor_blk SHARED
+	fio_engine.cpp
+	../json11/json11.cpp
+)
+target_link_libraries(fio_vitastor_blk
+	vitastor_blk
+)
+
+# vitastor-osd
+add_executable(vitastor-osd
+	osd_main.cpp osd.cpp osd_secondary.cpp msgr_receive.cpp msgr_send.cpp osd_peering.cpp osd_flush.cpp osd_peering_pg.cpp
+	osd_primary.cpp osd_primary_subops.cpp etcd_state_client.cpp messenger.cpp osd_cluster.cpp http_client.cpp osd_ops.cpp pg_states.cpp
+	osd_rmw.cpp base64.cpp timerfd_manager.cpp epoll_manager.cpp ../json11/json11.cpp
+)
+target_link_libraries(vitastor-osd
+	vitastor_blk
+	Jerasure
+)
+
+# libfio_vitastor_sec.so
+add_library(fio_vitastor_sec SHARED
+	fio_sec_osd.cpp
+	rw_blocking.cpp
+)
+target_link_libraries(fio_vitastor_sec
+	tcmalloc_minimal
+)
+
+# libvitastor_client.so
+add_library(vitastor_client SHARED
+	cluster_client.cpp epoll_manager.cpp etcd_state_client.cpp
+	messenger.cpp msgr_send.cpp msgr_receive.cpp ringloop.cpp ../json11/json11.cpp
+	http_client.cpp osd_ops.cpp pg_states.cpp timerfd_manager.cpp base64.cpp
+)
+target_link_libraries(vitastor_client
+	tcmalloc_minimal
+	${LIBURING_LIBRARIES}
+)
+
+# libfio_vitastor.so
+add_library(fio_vitastor SHARED
+	fio_cluster.cpp
+)
+target_link_libraries(fio_vitastor
+	vitastor_client
+)
+
+# vitastor-nbd
+add_executable(vitastor-nbd
+	nbd_proxy.cpp
+)
+target_link_libraries(vitastor-nbd
+	vitastor_client
+)
+
+# vitastor-rm
+add_executable(vitastor-rm
+	rm_inode.cpp
+)
+target_link_libraries(vitastor-rm
+	vitastor_client
+)
+
+# vitastor-dump-journal
+add_executable(vitastor-dump-journal
+	dump_journal.cpp crc32c.c
+)
+
+# qemu_driver.so
+add_library(qemu_proxy STATIC qemu_proxy.cpp)
+target_compile_options(qemu_proxy PUBLIC -fPIC)
+target_include_directories(qemu_proxy PUBLIC
+	../qemu/b/qemu
+	../qemu/include
+	${GLIB_INCLUDE_DIRS}
+)
+target_link_libraries(qemu_proxy
+	vitastor_client
+)
+add_library(qemu_vitastor SHARED
+	qemu_driver.c
+)
+target_link_libraries(qemu_vitastor
+	qemu_proxy
+)
+set_target_properties(qemu_vitastor PROPERTIES
+	PREFIX ""
+	OUTPUT_NAME "block-vitastor"
+)
+
+### Test stubs
+
+# stub_osd, stub_bench, osd_test
+add_executable(stub_osd stub_osd.cpp rw_blocking.cpp)
+target_link_libraries(stub_osd tcmalloc_minimal)
+add_executable(stub_bench stub_bench.cpp rw_blocking.cpp)
+target_link_libraries(stub_bench tcmalloc_minimal)
+add_executable(osd_test osd_test.cpp rw_blocking.cpp)
+target_link_libraries(osd_test tcmalloc_minimal)
+
+# osd_rmw_test
+add_executable(osd_rmw_test osd_rmw_test.cpp allocator.cpp)
+target_link_libraries(osd_rmw_test Jerasure tcmalloc_minimal)
+
+# stub_uring_osd
+add_executable(stub_uring_osd
+	stub_uring_osd.cpp epoll_manager.cpp messenger.cpp msgr_send.cpp msgr_receive.cpp ringloop.cpp timerfd_manager.cpp ../json11/json11.cpp
+)
+target_link_libraries(stub_uring_osd
+	${LIBURING_LIBRARIES}
+	tcmalloc_minimal
+)
+
+# osd_peering_pg_test
+add_executable(osd_peering_pg_test osd_peering_pg_test.cpp osd_peering_pg.cpp)
+target_link_libraries(osd_peering_pg_test tcmalloc_minimal)
+
+# test_allocator
+add_executable(test_allocator test_allocator.cpp allocator.cpp)
+
+## test_blockstore, test_shit
+#add_executable(test_blockstore test_blockstore.cpp timerfd_interval.cpp)
+#target_link_libraries(test_blockstore blockstore)
+#add_executable(test_shit test_shit.cpp osd_peering_pg.cpp)
+#target_link_libraries(test_shit ${LIBURING_LIBRARIES} m)
+
+### Install
+
+install(TARGETS vitastor-osd vitastor-dump-journal vitastor-nbd vitastor-rm RUNTIME DESTINATION ${CMAKE_INSTALL_BINDIR})
+install(TARGETS fio_vitastor fio_vitastor_blk fio_vitastor_sec vitastor_blk vitastor_client LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR})
+install(TARGETS qemu_vitastor LIBRARY DESTINATION /usr/${CMAKE_INSTALL_LIBDIR}/${QEMU_PLUGINDIR})

src/allocator.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
 #include <stdexcept>
 #include "allocator.h"
 
@@ -10,19 +13,19 @@ allocator::allocator(uint64_t blocks)
     {
         throw std::invalid_argument("blocks");
     }
-    uint64_t p2 = 1, total = 1;
+    uint64_t p2 = 1;
+    total = 0;
     while (p2 * 64 < blocks)
     {
-        p2 = p2 * 64;
         total += p2;
+        p2 = p2 * 64;
     }
-    total -= p2;
     total += (blocks+63) / 64;
-    mask = new uint64_t[2 + total];
+    mask = new uint64_t[total];
     size = free = blocks;
     last_one_mask = (blocks % 64) == 0
         ? UINT64_MAX
-        : ~(UINT64_MAX << (64 - blocks % 64));
+        : ((1l << (blocks % 64)) - 1);
     for (uint64_t i = 0; i < total; i++)
     {
         mask[i] = 0;
@@ -96,6 +99,10 @@ uint64_t allocator::find_free()
     uint64_t p2 = 1, offset = 0, addr = 0, f, i;
     while (p2 < size)
     {
+        if (offset+addr >= total)
+        {
+            return UINT64_MAX;
+        }
         uint64_t m = mask[offset + addr];
         for (i = 0, f = 1; i < 64; i++, f <<= 1)
         {
@@ -110,11 +117,6 @@ uint64_t allocator::find_free()
             return UINT64_MAX;
         }
         addr = (addr * 64) | i;
-        if (addr >= size)
-        {
-            // No space
-            return UINT64_MAX;
-        }
         offset += p2;
         p2 = p2 * 64;
     }
@@ -125,3 +127,35 @@ uint64_t allocator::get_free_count()
 {
     return free;
 }
+
+void bitmap_set(void *bitmap, uint64_t start, uint64_t len, uint64_t bitmap_granularity)
+{
+    if (start == 0)
+    {
+        if (len == 32*bitmap_granularity)
+        {
+            *((uint32_t*)bitmap) = UINT32_MAX;
+            return;
+        }
+        else if (len == 64*bitmap_granularity)
+        {
+            *((uint64_t*)bitmap) = UINT64_MAX;
+            return;
+        }
+    }
+    unsigned bit_start = start / bitmap_granularity;
+    unsigned bit_end = ((start + len) + bitmap_granularity - 1) / bitmap_granularity;
+    while (bit_start < bit_end)
+    {
+        if (!(bit_start & 7) && bit_end >= bit_start+8)
+        {
+            ((uint8_t*)bitmap)[bit_start / 8] = UINT8_MAX;
+            bit_start += 8;
+        }
+        else
+        {
+            ((uint8_t*)bitmap)[bit_start / 8] |= 1 << (bit_start % 8);
+            bit_start++;
+        }
+    }
+}

src/allocator.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
 #pragma once
 
 #include <stdint.h>
@@ -5,6 +8,7 @@
 // Hierarchical bitmap allocator
 class allocator
 {
+    uint64_t total;
     uint64_t size;
     uint64_t free;
     uint64_t last_one_mask;
@@ -16,3 +20,5 @@ public:
     uint64_t find_free();
     uint64_t get_free_count();
 };
+
+void bitmap_set(void *bitmap, uint64_t start, uint64_t len, uint64_t bitmap_granularity);

src/base64.cpp

@@ -0,0 +1,55 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+#include "base64.h"
+
+std::string base64_encode(const std::string &in)
+{
+    std::string out;
+    unsigned val = 0;
+    int valb = -6;
+    for (unsigned char c: in)
+    {
+        val = (val << 8) + c;
+        valb += 8;
+        while (valb >= 0)
+        {
+            out.push_back("ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/"[(val>>valb) & 0x3F]);
+            valb -= 6;
+        }
+    }
+    if (valb > -6)
+        out.push_back("ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/"[((val<<8)>>(valb+8)) & 0x3F]);
+    while (out.size() % 4)
+        out.push_back('=');
+    return out;
+}
+
+static char T[256] = { 0 };
+
+std::string base64_decode(const std::string &in)
+{
+    std::string out;
+    if (T[0] == 0)
+    {
+        for (int i = 0; i < 256; i++)
+            T[i] = -1;
+        for (int i = 0; i < 64; i++)
+            T[(unsigned char)("ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/"[i])] = i;
+    }
+    unsigned val = 0;
+    int valb = -8;
+    for (unsigned char c: in)
+    {
+        if (T[c] == -1)
+            break;
+        val = (val<<6) + T[c];
+        valb += 6;
+        if (valb >= 0)
+        {
+            out.push_back(char((val >> valb) & 0xFF));
+            valb -= 8;
+        }
+    }
+    return out;
+}

src/base64.h

@@ -0,0 +1,8 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+#pragma once
+#include <string>
+
+std::string base64_encode(const std::string &in);
+std::string base64_decode(const std::string &in);

src/blockstore.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
 #include "blockstore_impl.h"
 
 blockstore_t::blockstore_t(blockstore_config_t & config, ring_loop_t *ringloop)
@@ -32,12 +35,7 @@ bool blockstore_t::is_safe_to_stop()
 
 void blockstore_t::enqueue_op(blockstore_op_t *op)
 {
-    impl->enqueue_op(op, false);
-}
-
-void blockstore_t::enqueue_op_first(blockstore_op_t *op)
-{
-    impl->enqueue_op(op, true);
+    impl->enqueue_op(op);
 }
 
 std::unordered_map<object_id, uint64_t> & blockstore_t::get_unstable_writes()
@@ -45,6 +43,11 @@ std::unordered_map<object_id, uint64_t> & blockstore_t::get_unstable_writes()
     return impl->unstable_writes;
 }
 
+std::map<uint64_t, uint64_t> & blockstore_t::get_inode_space_stats()
+{
+    return impl->inode_space_stats;
+}
+
 uint32_t blockstore_t::get_block_size()
 {
     return impl->get_block_size();
@@ -55,7 +58,12 @@ uint64_t blockstore_t::get_block_count()
     return impl->get_block_count();
 }
 
-uint32_t blockstore_t::get_disk_alignment()
+uint64_t blockstore_t::get_free_block_count()
 {
-    return impl->get_disk_alignment();
+    return impl->get_free_block_count();
+}
+
+uint32_t blockstore_t::get_bitmap_granularity()
+{
+    return impl->get_bitmap_granularity();
 }

src/blockstore.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
 #pragma once
 
 #ifndef _LARGEFILE64_SOURCE
@@ -6,6 +9,7 @@
 
 #include <stdint.h>
 
+#include <string>
 #include <map>
 #include <unordered_map>
 #include <functional>
@@ -15,23 +19,27 @@
 
 // Memory alignment for direct I/O (usually 512 bytes)
 // All other alignments must be a multiple of this one
+#ifndef MEM_ALIGNMENT
 #define MEM_ALIGNMENT 512
+#endif
 
 // Default block size is 128 KB, current allowed range is 4K - 128M
 #define DEFAULT_ORDER 17
 #define MIN_BLOCK_SIZE 4*1024
 #define MAX_BLOCK_SIZE 128*1024*1024
+#define DEFAULT_BITMAP_GRANULARITY 4096
 
 #define BS_OP_MIN 1
 #define BS_OP_READ 1
 #define BS_OP_WRITE 2
-#define BS_OP_SYNC 3
-#define BS_OP_STABLE 4
-#define BS_OP_DELETE 5
-#define BS_OP_LIST 6
-#define BS_OP_ROLLBACK 7
-#define BS_OP_SYNC_STAB_ALL 8
-#define BS_OP_MAX 8
+#define BS_OP_WRITE_STABLE 3
+#define BS_OP_SYNC 4
+#define BS_OP_STABLE 5
+#define BS_OP_DELETE 6
+#define BS_OP_LIST 7
+#define BS_OP_ROLLBACK 8
+#define BS_OP_SYNC_STAB_ALL 9
+#define BS_OP_MAX 9
 
 #define BS_OP_PRIVATE_DATA_SIZE 256
 
@@ -39,9 +47,9 @@
 
 Blockstore opcode documentation:
 
-## BS_OP_READ / BS_OP_WRITE
+## BS_OP_READ / BS_OP_WRITE / BS_OP_WRITE_STABLE
 
-Read or write object data.
+Read or write object data. WRITE_STABLE writes a version that doesn't require marking as stable.
 
 Input:
 - oid = requested object
@@ -50,12 +58,15 @@ Input:
   - version == 0: read the last stable version,
   - version == UINT64_MAX: read the last version,
   - otherwise: read the newest version that is <= the specified version
+  - reads aren't guaranteed to return data from previous unfinished writes
   For writes:
   - if version == 0, a new version is assigned automatically
   - if version != 0, it is assigned for the new write if possible, otherwise -EINVAL is returned
 - offset, len = offset and length within object. length may be zero, in that case
   read operation only returns the version / write operation only bumps the version
 - buf = pre-allocated buffer for data (read) / with data (write). may be NULL if len == 0.
+- bitmap = pointer to the new 'external' object bitmap data. Its part which is respective to the
+  write request is copied into the metadata area bitwise and stored there.
 
 Output:
 - retval = number of bytes actually read/written or negative error number (-EINVAL or -ENOSPC)
@@ -92,7 +103,7 @@ Input:
 - buf = pre-allocated obj_ver_id array <len> units long
 
 Output:
-- retval = 0 or negative error number (-EINVAL)
+- retval = 0 or negative error number (-EINVAL, -ENOENT if no such version or -EBUSY if not synced)
 
 ## BS_OP_SYNC_STAB_ALL
 
@@ -110,6 +121,8 @@ Input:
 - oid.stripe = PG alignment
 - len = PG count or 0 to list all objects
 - offset = PG number
+- oid.inode = min inode number or 0 to list all inodes
+- version = max inode number or 0 to list all inodes
 
 Output:
 - retval = total obj_ver_id count
@@ -131,6 +144,7 @@ struct blockstore_op_t
     uint32_t offset;
     uint32_t len;
     void *buf;
+    void *bitmap;
     int retval;
 
     uint8_t private_data[BS_OP_PRIVATE_DATA_SIZE];
@@ -165,16 +179,16 @@ public:
     // Submission
     void enqueue_op(blockstore_op_t *op);
 
-    // Insert operation into the beginning of the queue
-    // Intended for the OSD syncer "thread" to be able to stabilize something when the journal is full
-    void enqueue_op_first(blockstore_op_t *op);
-
     // Unstable writes are added here (map of object_id -> version)
     std::unordered_map<object_id, uint64_t> & get_unstable_writes();
 
+    // Get per-inode space usage statistics
+    std::map<uint64_t, uint64_t> & get_inode_space_stats();
+
     // FIXME rename to object_size
     uint32_t get_block_size();
     uint64_t get_block_count();
+    uint64_t get_free_block_count();
 
-    uint32_t get_disk_alignment();
+    uint32_t get_bitmap_granularity();
 };

src/blockstore_flush.cpp

@@ -1,14 +1,24 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
 #include "blockstore_impl.h"
 
 journal_flusher_t::journal_flusher_t(int flusher_count, blockstore_impl_t *bs)
 {
     this->bs = bs;
     this->flusher_count = flusher_count;
+    this->cur_flusher_count = 1;
+    this->target_flusher_count = 1;
+    dequeuing = false;
+    trimming = false;
     active_flushers = 0;
-    sync_threshold = flusher_count == 1 ? 1 : flusher_count/2;
-    journal_trim_interval = sync_threshold;
+    syncing_flushers = 0;
+    // FIXME: allow to configure flusher_start_threshold and journal_trim_interval
+    flusher_start_threshold = bs->journal_block_size / sizeof(journal_entry_stable);
+    journal_trim_interval = 512;
     journal_trim_counter = 0;
-    journal_superblock = bs->journal.inmemory ? bs->journal.buffer : memalign(MEM_ALIGNMENT, bs->journal_block_size);
+    trim_wanted = 0;
+    journal_superblock = bs->journal.inmemory ? bs->journal.buffer : memalign_or_die(MEM_ALIGNMENT, bs->journal_block_size);
     co = new journal_flusher_co[flusher_count];
     for (int i = 0; i < flusher_count; i++)
     {
@@ -55,23 +65,36 @@ journal_flusher_t::~journal_flusher_t()
 
 bool journal_flusher_t::is_active()
 {
-    return active_flushers > 0 || start_forced && flush_queue.size() > 0 || flush_queue.size() >= sync_threshold;
+    return active_flushers > 0 || dequeuing;
 }
 
 void journal_flusher_t::loop()
 {
-    for (int i = 0; i < flusher_count; i++)
+    target_flusher_count = bs->write_iodepth*2;
+    if (target_flusher_count <= 0)
+        target_flusher_count = 1;
+    else if (target_flusher_count > flusher_count)
+        target_flusher_count = flusher_count;
+    if (target_flusher_count > cur_flusher_count)
+        cur_flusher_count = target_flusher_count;
+    else if (target_flusher_count < cur_flusher_count)
     {
-        if (!active_flushers && (start_forced ? !flush_queue.size() : (flush_queue.size() < sync_threshold)))
+        while (target_flusher_count < cur_flusher_count)
         {
-            return;
+            if (co[cur_flusher_count-1].wait_state)
+                break;
+            cur_flusher_count--;
         }
-        co[i].loop();
     }
+    for (int i = 0; (active_flushers > 0 || dequeuing) && i < cur_flusher_count; i++)
+        co[i].loop();
 }
 
 void journal_flusher_t::enqueue_flush(obj_ver_id ov)
 {
+#ifdef BLOCKSTORE_DEBUG
+    printf("enqueue_flush %lx:%lx v%lu\n", ov.oid.inode, ov.oid.stripe, ov.version);
+#endif
     auto it = flush_versions.find(ov.oid);
     if (it != flush_versions.end())
     {
@@ -83,10 +106,18 @@ void journal_flusher_t::enqueue_flush(obj_ver_id ov)
         flush_versions[ov.oid] = ov.version;
         flush_queue.push_back(ov.oid);
     }
+    if (!dequeuing && (flush_queue.size() >= flusher_start_threshold || trim_wanted > 0))
+    {
+        dequeuing = true;
+        bs->ringloop->wakeup();
+    }
 }
 
-void journal_flusher_t::unshift_flush(obj_ver_id ov)
+void journal_flusher_t::unshift_flush(obj_ver_id ov, bool force)
 {
+#ifdef BLOCKSTORE_DEBUG
+    printf("unshift_flush %lx:%lx v%lu\n", ov.oid.inode, ov.oid.stripe, ov.version);
+#endif
     auto it = flush_versions.find(ov.oid);
     if (it != flush_versions.end())
     {
@@ -96,16 +127,60 @@ void journal_flusher_t::unshift_flush(obj_ver_id ov)
     else
     {
         flush_versions[ov.oid] = ov.version;
+        if (!force)
+            flush_queue.push_front(ov.oid);
+    }
+    if (force)
         flush_queue.push_front(ov.oid);
+    if (force || !dequeuing && (flush_queue.size() >= flusher_start_threshold || trim_wanted > 0))
+    {
+        dequeuing = true;
+        bs->ringloop->wakeup();
     }
 }
 
-void journal_flusher_t::force_start()
+void journal_flusher_t::remove_flush(object_id oid)
 {
-    start_forced = true;
+#ifdef BLOCKSTORE_DEBUG
+    printf("undo_flush %lx:%lx\n", oid.inode, oid.stripe);
+#endif
+    auto v_it = flush_versions.find(oid);
+    if (v_it != flush_versions.end())
+    {
+        flush_versions.erase(v_it);
+        for (auto q_it = flush_queue.begin(); q_it != flush_queue.end(); q_it++)
+        {
+            if (*q_it == oid)
+            {
+                flush_queue.erase(q_it);
+                break;
+            }
+        }
+    }
+}
+
+void journal_flusher_t::request_trim()
+{
+    dequeuing = true;
+    trim_wanted++;
     bs->ringloop->wakeup();
 }
 
+void journal_flusher_t::mark_trim_possible()
+{
+    if (trim_wanted > 0)
+    {
+        dequeuing = true;
+        journal_trim_counter++;
+        bs->ringloop->wakeup();
+    }
+}
+
+void journal_flusher_t::release_trim()
+{
+    trim_wanted--;
+}
+
 #define await_sqe(label) \
     resume_##label:\
         sqe = bs->get_sqe();\
@@ -116,6 +191,7 @@ void journal_flusher_t::force_start()
         }\
         data = ((ring_data_t*)sqe->user_data);
 
+// FIXME: Implement batch flushing
 bool journal_flusher_co::loop()
 {
     // This is much better than implementing the whole function as an FSM
@@ -154,11 +230,23 @@ bool journal_flusher_co::loop()
         goto resume_17;
     else if (wait_state == 18)
         goto resume_18;
+    else if (wait_state == 19)
+        goto resume_19;
+    else if (wait_state == 20)
+        goto resume_20;
+    else if (wait_state == 21)
+        goto resume_21;
 resume_0:
-    if (!flusher->flush_queue.size() ||
-        !flusher->start_forced && !flusher->active_flushers && flusher->flush_queue.size() < flusher->sync_threshold)
+    if (!flusher->flush_queue.size() || !flusher->dequeuing)
     {
-        flusher->start_forced = false;
+stop_flusher:
+        if (flusher->trim_wanted > 0 && flusher->journal_trim_counter > 0)
+        {
+            // Attempt forced trim
+            flusher->active_flushers++;
+            goto trim_journal;
+        }
+        flusher->dequeuing = false;
         wait_state = 0;
         return true;
     }
@@ -173,7 +261,7 @@ resume_0:
         if (repeat_it != flusher->sync_to_repeat.end())
         {
 #ifdef BLOCKSTORE_DEBUG
-            printf("Postpone %lu:%lu v%lu\n", cur.oid.inode, cur.oid.stripe, cur.version);
+            printf("Postpone %lx:%lx v%lu\n", cur.oid.inode, cur.oid.stripe, cur.version);
 #endif
             // We don't flush different parts of history of the same object in parallel
             // So we check if someone is already flushing this object
@@ -186,42 +274,110 @@ resume_0:
         }
         else
             flusher->sync_to_repeat[cur.oid] = 0;
+        if (dirty_end->second.journal_sector >= bs->journal.dirty_start &&
+            (bs->journal.dirty_start >= bs->journal.used_start ||
+            dirty_end->second.journal_sector < bs->journal.used_start))
+        {
+            flusher->enqueue_flush(cur);
+            // We can't flush journal sectors that are still written to
+            // However, as we group flushes by oid, current oid may have older writes to flush!
+            // And it may even block writes if we don't flush the older version
+            // (if it's in the beginning of the journal)...
+            // So first try to find an older version of the same object to flush.
+            bool found = false;
+            while (dirty_end != bs->dirty_db.begin())
+            {
+                dirty_end--;
+                if (dirty_end->first.oid != cur.oid)
+                {
+                    break;
+                }
+                if (!(dirty_end->second.journal_sector >= bs->journal.dirty_start &&
+                    (bs->journal.dirty_start >= bs->journal.used_start ||
+                    dirty_end->second.journal_sector < bs->journal.used_start)))
+                {
+                    found = true;
+                    cur.version = dirty_end->first.version;
+                    break;
+                }
+            }
+            if (!found)
+            {
+                // Try other objects
+                flusher->sync_to_repeat.erase(cur.oid);
+                int search_left = flusher->flush_queue.size() - 1;
 #ifdef BLOCKSTORE_DEBUG
-        printf("Flushing %lu:%lu v%lu\n", cur.oid.inode, cur.oid.stripe, cur.version);
+                printf("Flusher overran writers (dirty_start=%08lx) - searching for older flushes (%d left)\n", bs->journal.dirty_start, search_left);
+#endif
+                while (search_left > 0)
+                {
+                    cur.oid = flusher->flush_queue.front();
+                    cur.version = flusher->flush_versions[cur.oid];
+                    flusher->flush_queue.pop_front();
+                    flusher->flush_versions.erase(cur.oid);
+                    dirty_end = bs->dirty_db.find(cur);
+                    if (dirty_end != bs->dirty_db.end())
+                    {
+                        if (dirty_end->second.journal_sector >= bs->journal.dirty_start &&
+                            (bs->journal.dirty_start >= bs->journal.used_start ||
+                            dirty_end->second.journal_sector < bs->journal.used_start))
+                        {
+#ifdef BLOCKSTORE_DEBUG
+                            printf("Write %lx:%lx v%lu is too new: offset=%08lx\n", cur.oid.inode, cur.oid.stripe, cur.version, dirty_end->second.journal_sector);
+#endif
+                            flusher->enqueue_flush(cur);
+                        }
+                        else
+                        {
+                            repeat_it = flusher->sync_to_repeat.find(cur.oid);
+                            if (repeat_it == flusher->sync_to_repeat.end())
+                            {
+                                flusher->sync_to_repeat[cur.oid] = 0;
+                                break;
+                            }
+                        }
+                    }
+                    search_left--;
+                }
+                if (search_left <= 0)
+                {
+#ifdef BLOCKSTORE_DEBUG
+                    printf("No older flushes, stopping\n");
+#endif
+                    goto stop_flusher;
+                }
+            }
+        }
+#ifdef BLOCKSTORE_DEBUG
+        printf("Flushing %lx:%lx v%lu\n", cur.oid.inode, cur.oid.stripe, cur.version);
 #endif
         flusher->active_flushers++;
 resume_1:
+        // Find it in clean_db
+        clean_it = bs->clean_db.find(cur.oid);
+        old_clean_loc = (clean_it != bs->clean_db.end() ? clean_it->second.location : UINT64_MAX);
         // Scan dirty versions of the object
         if (!scan_dirty(1))
         {
             wait_state += 1;
             return false;
         }
-        if (copy_count == 0 && clean_loc == UINT64_MAX && !has_delete && !has_empty)
+        // Writes and deletes shouldn't happen at the same time
+        assert(!has_writes || !has_delete);
+        if (!has_writes && !has_delete || has_delete && old_clean_loc == UINT64_MAX)
         {
             // Nothing to flush
-            flusher->active_flushers--;
-            repeat_it = flusher->sync_to_repeat.find(cur.oid);
-            if (repeat_it != flusher->sync_to_repeat.end() && repeat_it->second > cur.version)
-            {
-                // Requeue version
-                flusher->unshift_flush({ .oid = cur.oid, .version = repeat_it->second });
-            }
-            flusher->sync_to_repeat.erase(repeat_it);
-            wait_state = 0;
-            goto resume_0;
+            bs->erase_dirty(dirty_start, std::next(dirty_end), clean_loc);
+            goto release_oid;
         }
-        // Find it in clean_db
-        clean_it = bs->clean_db.find(cur.oid);
-        old_clean_loc = (clean_it != bs->clean_db.end() ? clean_it->second.location : UINT64_MAX);
         if (clean_loc == UINT64_MAX)
         {
-            if (copy_count > 0 && has_delete || old_clean_loc == UINT64_MAX)
+            if (old_clean_loc == UINT64_MAX)
             {
                 // Object not allocated. This is a bug.
                 char err[1024];
                 snprintf(
-                    err, 1024, "BUG: Object %lu:%lu v%lu that we are trying to flush is not allocated on the data device",
+                    err, 1024, "BUG: Object %lx:%lx v%lu that we are trying to flush is not allocated on the data device",
                     cur.oid.inode, cur.oid.stripe, cur.version
                 );
                 throw std::runtime_error(err);
@@ -270,18 +426,18 @@ resume_1:
         {
             new_clean_bitmap = (bs->inmemory_meta
                 ? meta_new.buf + meta_new.pos*bs->clean_entry_size + sizeof(clean_disk_entry)
-                : bs->clean_bitmap + (clean_loc >> bs->block_order)*bs->clean_entry_bitmap_size);
+                : bs->clean_bitmap + (clean_loc >> bs->block_order)*(2*bs->clean_entry_bitmap_size));
             if (clean_init_bitmap)
             {
                 memset(new_clean_bitmap, 0, bs->clean_entry_bitmap_size);
-                bitmap_set(new_clean_bitmap, clean_bitmap_offset, clean_bitmap_len);
+                bitmap_set(new_clean_bitmap, clean_bitmap_offset, clean_bitmap_len, bs->bitmap_granularity);
             }
         }
         for (it = v.begin(); it != v.end(); it++)
         {
             if (new_clean_bitmap)
             {
-                bitmap_set(new_clean_bitmap, it->offset, it->len);
+                bitmap_set(new_clean_bitmap, it->offset, it->len, bs->bitmap_granularity);
             }
             await_sqe(4);
             data->iov = (struct iovec){ it->buf, (size_t)it->len };
@@ -315,6 +471,7 @@ resume_1:
                 wait_state = 5;
                 return false;
             }
+            // zero out old metadata entry
             memset(meta_old.buf + meta_old.pos*bs->clean_entry_size, 0, bs->clean_entry_size);
             await_sqe(15);
             data->iov = (struct iovec){ meta_old.buf, bs->meta_block_size };
@@ -326,17 +483,30 @@ resume_1:
         }
         if (has_delete)
         {
+            // zero out new metadata entry
             memset(meta_new.buf + meta_new.pos*bs->clean_entry_size, 0, bs->clean_entry_size);
         }
         else
         {
             clean_disk_entry *new_entry = (clean_disk_entry*)(meta_new.buf + meta_new.pos*bs->clean_entry_size);
+            if (new_entry->oid.inode != 0 && new_entry->oid != cur.oid)
+            {
+                printf("Fatal error (metadata corruption or bug): tried to overwrite non-zero metadata entry %lu (%lx:%lx) with %lx:%lx\n",
+                    clean_loc >> bs->block_order, new_entry->oid.inode, new_entry->oid.stripe, cur.oid.inode, cur.oid.stripe);
+                exit(1);
+            }
             new_entry->oid = cur.oid;
             new_entry->version = cur.version;
             if (!bs->inmemory_meta)
             {
                 memcpy(&new_entry->bitmap, new_clean_bitmap, bs->clean_entry_bitmap_size);
             }
+            // copy latest external bitmap/attributes
+            if (bs->clean_entry_bitmap_size)
+            {
+                void *bmp_ptr = bs->clean_entry_bitmap_size > sizeof(void*) ? dirty_end->second.bitmap : &dirty_end->second.bitmap;
+                memcpy((void*)(new_entry+1) + bs->clean_entry_bitmap_size, bmp_ptr, bs->clean_entry_bitmap_size);
+            }
         }
         await_sqe(6);
         data->iov = (struct iovec){ meta_new.buf, bs->meta_block_size };
@@ -386,13 +556,35 @@ resume_1:
         }
         // Update clean_db and dirty_db, free old data locations
         update_clean_db();
+#ifdef BLOCKSTORE_DEBUG
+        printf("Flushed %lx:%lx v%lu (%d copies, wr:%d, del:%d), %ld left\n", cur.oid.inode, cur.oid.stripe, cur.version,
+            copy_count, has_writes, has_delete, flusher->flush_queue.size());
+#endif
+    release_oid:
+        repeat_it = flusher->sync_to_repeat.find(cur.oid);
+        if (repeat_it != flusher->sync_to_repeat.end() && repeat_it->second > cur.version)
+        {
+            // Requeue version
+            flusher->unshift_flush({ .oid = cur.oid, .version = repeat_it->second }, false);
+        }
+        flusher->sync_to_repeat.erase(repeat_it);
+    trim_journal:
         // Clear unused part of the journal every <journal_trim_interval> flushes
-        if (!((++flusher->journal_trim_counter) % flusher->journal_trim_interval))
+        if (!((++flusher->journal_trim_counter) % flusher->journal_trim_interval) || flusher->trim_wanted > 0)
         {
             flusher->journal_trim_counter = 0;
-            if (bs->journal.trim())
+            new_trim_pos = bs->journal.get_trim_pos();
+            if (new_trim_pos != bs->journal.used_start)
             {
-                // Update journal "superblock"
+            resume_19:
+                // Wait for other coroutines trimming the journal, if any
+                if (flusher->trimming)
+                {
+                    wait_state = 19;
+                    return false;
+                }
+                flusher->trimming = true;
+                // First update journal "superblock" and only then update <used_start> in memory
                 await_sqe(12);
                 *((journal_entry_start*)flusher->journal_superblock) = {
                     .crc32 = 0,
@@ -400,7 +592,7 @@ resume_1:
                     .type = JE_START,
                     .size = sizeof(journal_entry_start),
                     .reserved = 0,
-                    .journal_start = bs->journal.used_start,
+                    .journal_start = new_trim_pos,
                 };
                 ((journal_entry_start*)flusher->journal_superblock)->crc32 = je_crc32((journal_entry*)flusher->journal_superblock);
                 data->iov = (struct iovec){ flusher->journal_superblock, bs->journal_block_size };
@@ -413,20 +605,28 @@ resume_1:
                     wait_state = 13;
                     return false;
                 }
+                if (!bs->disable_journal_fsync)
+                {
+                    await_sqe(20);
+                    my_uring_prep_fsync(sqe, bs->journal.fd, IORING_FSYNC_DATASYNC);
+                    data->iov = { 0 };
+                    data->callback = simple_callback_w;
+                resume_21:
+                    if (wait_count > 0)
+                    {
+                        wait_state = 21;
+                        return false;
+                    }
+                }
+                bs->journal.used_start = new_trim_pos;
+#ifdef BLOCKSTORE_DEBUG
+                printf("Journal trimmed to %08lx (next_free=%08lx)\n", bs->journal.used_start, bs->journal.next_free);
+#endif
+                flusher->trimming = false;
             }
         }
         // All done
-#ifdef BLOCKSTORE_DEBUG
-        printf("Flushed %lu:%lu v%lu\n", cur.oid.inode, cur.oid.stripe, cur.version);
-#endif
         flusher->active_flushers--;
-        repeat_it = flusher->sync_to_repeat.find(cur.oid);
-        if (repeat_it != flusher->sync_to_repeat.end() && repeat_it->second > cur.version)
-        {
-            // Requeue version
-            flusher->unshift_flush({ .oid = cur.oid, .version = repeat_it->second });
-        }
-        flusher->sync_to_repeat.erase(repeat_it);
         wait_state = 0;
         goto resume_0;
     }
@@ -445,19 +645,25 @@ bool journal_flusher_co::scan_dirty(int wait_base)
     copy_count = 0;
     clean_loc = UINT64_MAX;
     has_delete = false;
-    has_empty = false;
+    has_writes = false;
     skip_copy = false;
     clean_init_bitmap = false;
     while (1)
     {
-        if (dirty_it->second.state == ST_J_STABLE && !skip_copy)
+        if (!IS_STABLE(dirty_it->second.state))
+        {
+            char err[1024];
+            snprintf(
+                err, 1024, "BUG: Unexpected dirty_entry %lx:%lx v%lu unstable state during flush: %d",
+                dirty_it->first.oid.inode, dirty_it->first.oid.stripe, dirty_it->first.version, dirty_it->second.state
+            );
+            throw std::runtime_error(err);
+        }
+        else if (IS_JOURNAL(dirty_it->second.state) && !skip_copy)
         {
             // First we submit all reads
-            if (dirty_it->second.len == 0)
-            {
-                has_empty = true;
-            }
-            else
+            has_writes = true;
+            if (dirty_it->second.len != 0)
             {
                 offset = dirty_it->second.offset;
                 end_offset = dirty_it->second.offset + dirty_it->second.len;
@@ -471,18 +677,18 @@ bool journal_flusher_co::scan_dirty(int wait_base)
                     {
                         submit_offset = dirty_it->second.location + offset - dirty_it->second.offset;
                         submit_len = it == v.end() || it->offset >= end_offset ? end_offset-offset : it->offset-offset;
-                        it = v.insert(it, (copy_buffer_t){ .offset = offset, .len = submit_len, .buf = memalign(MEM_ALIGNMENT, submit_len) });
+                        it = v.insert(it, (copy_buffer_t){ .offset = offset, .len = submit_len, .buf = memalign_or_die(MEM_ALIGNMENT, submit_len) });
                         copy_count++;
                         if (bs->journal.inmemory)
                         {
                             // Take it from memory
-                            memcpy(v.back().buf, bs->journal.buffer + submit_offset, submit_len);
+                            memcpy(it->buf, bs->journal.buffer + submit_offset, submit_len);
                         }
                         else
                         {
                             // Read it from disk
                             await_sqe(0);
-                            data->iov = (struct iovec){ v.back().buf, (size_t)submit_len };
+                            data->iov = (struct iovec){ it->buf, (size_t)submit_len };
                             data->callback = simple_callback_r;
                             my_uring_prep_readv(
                                 sqe, bs->journal.fd, &data->iov, 1, bs->journal.offset + submit_offset
@@ -496,30 +702,22 @@ bool journal_flusher_co::scan_dirty(int wait_base)
                 }
             }
         }
-        else if (dirty_it->second.state == ST_D_STABLE && !skip_copy)
+        else if (IS_BIG_WRITE(dirty_it->second.state) && !skip_copy)
         {
             // There is an unflushed big write. Copy small writes in its position
+            has_writes = true;
             clean_loc = dirty_it->second.location;
             clean_init_bitmap = true;
             clean_bitmap_offset = dirty_it->second.offset;
             clean_bitmap_len = dirty_it->second.len;
             skip_copy = true;
         }
-        else if (dirty_it->second.state == ST_DEL_STABLE && !skip_copy)
+        else if (IS_DELETE(dirty_it->second.state) && !skip_copy)
         {
             // There is an unflushed delete
             has_delete = true;
             skip_copy = true;
         }
-        else if (!IS_STABLE(dirty_it->second.state))
-        {
-            char err[1024];
-            snprintf(
-                err, 1024, "BUG: Unexpected dirty_entry %lu:%lu v%lu state during flush: %d",
-                dirty_it->first.oid.inode, dirty_it->first.oid.stripe, dirty_it->first.version, dirty_it->second.state
-            );
-            throw std::runtime_error(err);
-        }
         dirty_start = dirty_it;
         if (dirty_it == bs->dirty_db.begin())
         {
@@ -555,7 +753,7 @@ bool journal_flusher_co::modify_meta_read(uint64_t meta_loc, flusher_meta_write_
     if (wr.it == flusher->meta_sectors.end())
     {
         // Not in memory yet, read it
-        wr.buf = memalign(MEM_ALIGNMENT, bs->meta_block_size);
+        wr.buf = memalign_or_die(MEM_ALIGNMENT, bs->meta_block_size);
         wr.it = flusher->meta_sectors.emplace(wr.sector, (meta_sector_t){
             .offset = wr.sector,
             .len = bs->meta_block_size,
@@ -585,7 +783,7 @@ void journal_flusher_co::update_clean_db()
     if (old_clean_loc != UINT64_MAX && old_clean_loc != clean_loc)
     {
 #ifdef BLOCKSTORE_DEBUG
-        printf("Free block %lu\n", old_clean_loc >> bs->block_order);
+        printf("Free block %lu (new location is %lu)\n", old_clean_loc >> bs->block_order, clean_loc >> bs->block_order);
 #endif
         bs->data_alloc->set(old_clean_loc >> bs->block_order, false);
     }
@@ -632,32 +830,37 @@ bool journal_flusher_co::fsync_batch(bool fsync_meta, int wait_base)
         });
     sync_found:
         cur_sync->ready_count++;
-        if (cur_sync->ready_count >= flusher->sync_threshold || !flusher->flush_queue.size())
+        flusher->syncing_flushers++;
+    resume_1:
+        if (!cur_sync->state)
         {
-            // Sync batch is ready. Do it.
-            await_sqe(0);
-            data->iov = { 0 };
-            data->callback = simple_callback_w;
-            my_uring_prep_fsync(sqe, fsync_meta ? bs->meta_fd : bs->data_fd, IORING_FSYNC_DATASYNC);
-            cur_sync->state = 1;
-            wait_count++;
-        resume_1:
-            if (wait_count > 0)
+            if (flusher->syncing_flushers >= flusher->cur_flusher_count || !flusher->flush_queue.size())
             {
+                // Sync batch is ready. Do it.
+                await_sqe(0);
+                data->iov = { 0 };
+                data->callback = simple_callback_w;
+                my_uring_prep_fsync(sqe, fsync_meta ? bs->meta_fd : bs->data_fd, IORING_FSYNC_DATASYNC);
+                cur_sync->state = 1;
+                wait_count++;
+            resume_2:
+                if (wait_count > 0)
+                {
+                    wait_state = 2;
+                    return false;
+                }
+                // Sync completed. All previous coroutines waiting for it must be resumed
+                cur_sync->state = 2;
+                bs->ringloop->wakeup();
+            }
+            else
+            {
+                // Wait until someone else sends and completes a sync.
                 wait_state = 1;
                 return false;
             }
-            // Sync completed. All previous coroutines waiting for it must be resumed
-            cur_sync->state = 2;
-            bs->ringloop->wakeup();
-        }
-        // Wait until someone else sends and completes a sync.
-    resume_2:
-        if (!cur_sync->state)
-        {
-            wait_state = 2;
-            return false;
         }
+        flusher->syncing_flushers--;
         cur_sync->ready_count--;
         if (cur_sync->ready_count == 0)
         {
@@ -666,35 +869,3 @@ bool journal_flusher_co::fsync_batch(bool fsync_meta, int wait_base)
     }
     return true;
 }
-
-void journal_flusher_co::bitmap_set(void *bitmap, uint64_t start, uint64_t len)
-{
-    if (start == 0)
-    {
-        if (len == 32*bs->bitmap_granularity)
-        {
-            *((uint32_t*)bitmap) = UINT32_MAX;
-            return;
-        }
-        else if (len == 64*bs->bitmap_granularity)
-        {
-            *((uint64_t*)bitmap) = UINT64_MAX;
-            return;
-        }
-    }
-    unsigned bit_start = start / bs->bitmap_granularity;
-    unsigned bit_end = ((start + len) + bs->bitmap_granularity - 1) / bs->bitmap_granularity;
-    while (bit_start < bit_end)
-    {
-        if (!(bit_start & 7) && bit_end >= bit_start+8)
-        {
-            ((uint8_t*)bitmap)[bit_start / 8] = UINT8_MAX;
-            bit_start += 8;
-        }
-        else
-        {
-            ((uint8_t*)bitmap)[bit_start / 8] |= 1 << (bit_start % 8);
-            bit_start++;
-        }
-    }
-}

src/blockstore_flush.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
 struct copy_buffer_t
 {
     uint64_t offset, len;
@@ -45,8 +48,8 @@ class journal_flusher_co
     std::map<object_id, uint64_t>::iterator repeat_it;
     std::function<void(ring_data_t*)> simple_callback_r, simple_callback_w;
 
-    bool skip_copy, has_delete, has_empty;
-    spp::sparse_hash_map<object_id, clean_entry>::iterator clean_it;
+    bool skip_copy, has_delete, has_writes;
+    blockstore_clean_db_t::iterator clean_it;
     std::vector<copy_buffer_t> v;
     std::vector<copy_buffer_t>::iterator it;
     int copy_count;
@@ -56,6 +59,8 @@ class journal_flusher_co
     uint64_t clean_bitmap_offset, clean_bitmap_len;
     void *new_clean_bitmap;
 
+    uint64_t new_trim_pos;
+
     // local: scan_dirty()
     uint64_t offset, end_offset, submit_offset, submit_len;
 
@@ -64,7 +69,6 @@ class journal_flusher_co
     bool modify_meta_read(uint64_t meta_loc, flusher_meta_write_t &wr, int wait_base);
     void update_clean_db();
     bool fsync_batch(bool fsync_meta, int wait_base);
-    void bitmap_set(void *bitmap, uint64_t start, uint64_t len);
 public:
     journal_flusher_co();
     bool loop();
@@ -73,17 +77,20 @@ public:
 // Journal flusher itself
 class journal_flusher_t
 {
-    bool start_forced = false;
-    int flusher_count;
-    int sync_threshold;
+    int trim_wanted = 0;
+    bool dequeuing;
+    int flusher_count, cur_flusher_count, target_flusher_count;
+    int flusher_start_threshold;
     journal_flusher_co *co;
     blockstore_impl_t *bs;
     friend class journal_flusher_co;
 
     int journal_trim_counter, journal_trim_interval;
+    bool trimming;
     void* journal_superblock;
 
     int active_flushers;
+    int syncing_flushers;
     std::list<flusher_sync_t> syncs;
     std::map<object_id, uint64_t> sync_to_repeat;
 
@@ -95,7 +102,10 @@ public:
     ~journal_flusher_t();
     void loop();
     bool is_active();
-    void force_start();
+    void mark_trim_possible();
+    void request_trim();
+    void release_trim();
     void enqueue_flush(obj_ver_id oid);
-    void unshift_flush(obj_ver_id oid);
+    void unshift_flush(obj_ver_id oid, bool force);
+    void remove_flush(object_id oid);
 };

src/blockstore_impl.cpp

@@ -0,0 +1,583 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+#include "blockstore_impl.h"
+
+blockstore_impl_t::blockstore_impl_t(blockstore_config_t & config, ring_loop_t *ringloop)
+{
+    assert(sizeof(blockstore_op_private_t) <= BS_OP_PRIVATE_DATA_SIZE);
+    this->ringloop = ringloop;
+    ring_consumer.loop = [this]() { loop(); };
+    ringloop->register_consumer(&ring_consumer);
+    initialized = 0;
+    data_fd = meta_fd = journal.fd = -1;
+    parse_config(config);
+    zero_object = (uint8_t*)memalign_or_die(MEM_ALIGNMENT, block_size);
+    try
+    {
+        open_data();
+        open_meta();
+        open_journal();
+        calc_lengths();
+        data_alloc = new allocator(block_count);
+    }
+    catch (std::exception & e)
+    {
+        if (data_fd >= 0)
+            close(data_fd);
+        if (meta_fd >= 0 && meta_fd != data_fd)
+            close(meta_fd);
+        if (journal.fd >= 0 && journal.fd != meta_fd)
+            close(journal.fd);
+        throw;
+    }
+    flusher = new journal_flusher_t(flusher_count, this);
+}
+
+blockstore_impl_t::~blockstore_impl_t()
+{
+    delete data_alloc;
+    delete flusher;
+    free(zero_object);
+    ringloop->unregister_consumer(&ring_consumer);
+    if (data_fd >= 0)
+        close(data_fd);
+    if (meta_fd >= 0 && meta_fd != data_fd)
+        close(meta_fd);
+    if (journal.fd >= 0 && journal.fd != meta_fd)
+        close(journal.fd);
+    if (metadata_buffer)
+        free(metadata_buffer);
+    if (clean_bitmap)
+        free(clean_bitmap);
+}
+
+bool blockstore_impl_t::is_started()
+{
+    return initialized == 10;
+}
+
+bool blockstore_impl_t::is_stalled()
+{
+    return queue_stall;
+}
+
+// main event loop - produce requests
+void blockstore_impl_t::loop()
+{
+    // FIXME: initialized == 10 is ugly
+    if (initialized != 10)
+    {
+        // read metadata, then journal
+        if (initialized == 0)
+        {
+            metadata_init_reader = new blockstore_init_meta(this);
+            initialized = 1;
+        }
+        if (initialized == 1)
+        {
+            int res = metadata_init_reader->loop();
+            if (!res)
+            {
+                delete metadata_init_reader;
+                metadata_init_reader = NULL;
+                journal_init_reader = new blockstore_init_journal(this);
+                initialized = 2;
+            }
+        }
+        if (initialized == 2)
+        {
+            int res = journal_init_reader->loop();
+            if (!res)
+            {
+                delete journal_init_reader;
+                journal_init_reader = NULL;
+                initialized = 10;
+                ringloop->wakeup();
+            }
+        }
+    }
+    else
+    {
+        // try to submit ops
+        unsigned initial_ring_space = ringloop->space_left();
+        // has_writes == 0 - no writes before the current queue item
+        // has_writes == 1 - some writes in progress
+        // has_writes == 2 - tried to submit some writes, but failed
+        int has_writes = 0, op_idx = 0, new_idx = 0;
+        for (; op_idx < submit_queue.size(); op_idx++, new_idx++)
+        {
+            auto op = submit_queue[op_idx];
+            submit_queue[new_idx] = op;
+            // FIXME: This needs some simplification
+            // Writes should not block reads if the ring is not full and reads don't depend on them
+            // In all other cases we should stop submission
+            if (PRIV(op)->wait_for)
+            {
+                check_wait(op);
+                if (PRIV(op)->wait_for == WAIT_SQE)
+                {
+                    break;
+                }
+                else if (PRIV(op)->wait_for)
+                {
+                    if (op->opcode == BS_OP_WRITE || op->opcode == BS_OP_WRITE_STABLE || op->opcode == BS_OP_DELETE)
+                    {
+                        has_writes = 2;
+                    }
+                    continue;
+                }
+            }
+            unsigned ring_space = ringloop->space_left();
+            unsigned prev_sqe_pos = ringloop->save();
+            // 0 = can't submit
+            // 1 = in progress
+            // 2 = can be removed from queue
+            int wr_st = 0;
+            if (op->opcode == BS_OP_READ)
+            {
+                wr_st = dequeue_read(op);
+            }
+            else if (op->opcode == BS_OP_WRITE || op->opcode == BS_OP_WRITE_STABLE)
+            {
+                if (has_writes == 2)
+                {
+                    // Some writes already could not be submitted
+                    continue;
+                }
+                wr_st = dequeue_write(op);
+                has_writes = wr_st > 0 ? 1 : 2;
+            }
+            else if (op->opcode == BS_OP_DELETE)
+            {
+                if (has_writes == 2)
+                {
+                    // Some writes already could not be submitted
+                    continue;
+                }
+                wr_st = dequeue_del(op);
+                has_writes = wr_st > 0 ? 1 : 2;
+            }
+            else if (op->opcode == BS_OP_SYNC)
+            {
+                // wait for all small writes to be submitted
+                // wait for all big writes to complete, submit data device fsync
+                // wait for the data device fsync to complete, then submit journal writes for big writes
+                // then submit an fsync operation
+                if (has_writes)
+                {
+                    // Can't submit SYNC before previous writes
+                    continue;
+                }
+                wr_st = continue_sync(op, false);
+                if (wr_st != 2)
+                {
+                    has_writes = wr_st > 0 ? 1 : 2;
+                }
+            }
+            else if (op->opcode == BS_OP_STABLE)
+            {
+                wr_st = dequeue_stable(op);
+            }
+            else if (op->opcode == BS_OP_ROLLBACK)
+            {
+                wr_st = dequeue_rollback(op);
+            }
+            else if (op->opcode == BS_OP_LIST)
+            {
+                // LIST doesn't need to be blocked by previous modifications
+                process_list(op);
+                wr_st = 2;
+            }
+            if (wr_st == 2)
+            {
+                new_idx--;
+            }
+            if (wr_st == 0)
+            {
+                ringloop->restore(prev_sqe_pos);
+                if (PRIV(op)->wait_for == WAIT_SQE)
+                {
+                    PRIV(op)->wait_detail = 1 + ring_space;
+                    // ring is full, stop submission
+                    break;
+                }
+            }
+        }
+        if (op_idx != new_idx)
+        {
+            while (op_idx < submit_queue.size())
+            {
+                submit_queue[new_idx++] = submit_queue[op_idx++];
+            }
+            submit_queue.resize(new_idx);
+        }
+        if (!readonly)
+        {
+            flusher->loop();
+        }
+        int ret = ringloop->submit();
+        if (ret < 0)
+        {
+            throw std::runtime_error(std::string("io_uring_submit: ") + strerror(-ret));
+        }
+        if ((initial_ring_space - ringloop->space_left()) > 0)
+        {
+            live = true;
+        }
+        queue_stall = !live && !ringloop->has_work();
+        live = false;
+    }
+}
+
+bool blockstore_impl_t::is_safe_to_stop()
+{
+    // It's safe to stop blockstore when there are no in-flight operations,
+    // no in-progress syncs and flusher isn't doing anything
+    if (submit_queue.size() > 0 || !readonly && flusher->is_active())
+    {
+        return false;
+    }
+    if (unsynced_big_writes.size() > 0 || unsynced_small_writes.size() > 0)
+    {
+        if (!readonly && !stop_sync_submitted)
+        {
+            // We should sync the blockstore before unmounting
+            blockstore_op_t *op = new blockstore_op_t;
+            op->opcode = BS_OP_SYNC;
+            op->buf = NULL;
+            op->callback = [](blockstore_op_t *op)
+            {
+                delete op;
+            };
+            enqueue_op(op);
+            stop_sync_submitted = true;
+        }
+        return false;
+    }
+    return true;
+}
+
+void blockstore_impl_t::check_wait(blockstore_op_t *op)
+{
+    if (PRIV(op)->wait_for == WAIT_SQE)
+    {
+        if (ringloop->space_left() < PRIV(op)->wait_detail)
+        {
+            // stop submission if there's still no free space
+#ifdef BLOCKSTORE_DEBUG
+            printf("Still waiting for %lu SQE(s)\n", PRIV(op)->wait_detail);
+#endif
+            return;
+        }
+        PRIV(op)->wait_for = 0;
+    }
+    else if (PRIV(op)->wait_for == WAIT_JOURNAL)
+    {
+        if (journal.used_start == PRIV(op)->wait_detail)
+        {
+            // do not submit
+#ifdef BLOCKSTORE_DEBUG
+            printf("Still waiting to flush journal offset %08lx\n", PRIV(op)->wait_detail);
+#endif
+            return;
+        }
+        flusher->release_trim();
+        PRIV(op)->wait_for = 0;
+    }
+    else if (PRIV(op)->wait_for == WAIT_JOURNAL_BUFFER)
+    {
+        int next = ((journal.cur_sector + 1) % journal.sector_count);
+        if (journal.sector_info[next].flush_count > 0 ||
+            journal.sector_info[next].dirty)
+        {
+            // do not submit
+#ifdef BLOCKSTORE_DEBUG
+            printf("Still waiting for a journal buffer\n");
+#endif
+            return;
+        }
+        PRIV(op)->wait_for = 0;
+    }
+    else if (PRIV(op)->wait_for == WAIT_FREE)
+    {
+        if (!data_alloc->get_free_count() && flusher->is_active())
+        {
+#ifdef BLOCKSTORE_DEBUG
+            printf("Still waiting for free space on the data device\n");
+#endif
+            return;
+        }
+        PRIV(op)->wait_for = 0;
+    }
+    else
+    {
+        throw std::runtime_error("BUG: op->wait_for value is unexpected");
+    }
+}
+
+void blockstore_impl_t::enqueue_op(blockstore_op_t *op)
+{
+    if (op->opcode < BS_OP_MIN || op->opcode > BS_OP_MAX ||
+        ((op->opcode == BS_OP_READ || op->opcode == BS_OP_WRITE || op->opcode == BS_OP_WRITE_STABLE) && (
+            op->offset >= block_size ||
+            op->len > block_size-op->offset ||
+            (op->len % disk_alignment)
+        )) ||
+        readonly && op->opcode != BS_OP_READ && op->opcode != BS_OP_LIST)
+    {
+        // Basic verification not passed
+        op->retval = -EINVAL;
+        std::function<void (blockstore_op_t*)>(op->callback)(op);
+        return;
+    }
+    if (op->opcode == BS_OP_SYNC_STAB_ALL)
+    {
+        std::function<void(blockstore_op_t*)> *old_callback = new std::function<void(blockstore_op_t*)>(op->callback);
+        op->opcode = BS_OP_SYNC;
+        op->callback = [this, old_callback](blockstore_op_t *op)
+        {
+            if (op->retval >= 0 && unstable_writes.size() > 0)
+            {
+                op->opcode = BS_OP_STABLE;
+                op->len = unstable_writes.size();
+                obj_ver_id *vers = new obj_ver_id[op->len];
+                op->buf = vers;
+                int i = 0;
+                for (auto it = unstable_writes.begin(); it != unstable_writes.end(); it++, i++)
+                {
+                    vers[i] = {
+                        .oid = it->first,
+                        .version = it->second,
+                    };
+                }
+                unstable_writes.clear();
+                op->callback = [this, old_callback](blockstore_op_t *op)
+                {
+                    obj_ver_id *vers = (obj_ver_id*)op->buf;
+                    delete[] vers;
+                    op->buf = NULL;
+                    (*old_callback)(op);
+                    delete old_callback;
+                };
+                this->enqueue_op(op);
+            }
+            else
+            {
+                (*old_callback)(op);
+                delete old_callback;
+            }
+        };
+    }
+    if ((op->opcode == BS_OP_WRITE || op->opcode == BS_OP_WRITE_STABLE || op->opcode == BS_OP_DELETE) && !enqueue_write(op))
+    {
+        std::function<void (blockstore_op_t*)>(op->callback)(op);
+        return;
+    }
+    // Call constructor without allocating memory. We'll call destructor before returning op back
+    new ((void*)op->private_data) blockstore_op_private_t;
+    PRIV(op)->wait_for = 0;
+    PRIV(op)->op_state = 0;
+    PRIV(op)->pending_ops = 0;
+    submit_queue.push_back(op);
+    ringloop->wakeup();
+}
+
+static bool replace_stable(object_id oid, uint64_t version, int search_start, int search_end, obj_ver_id* list)
+{
+    while (search_start < search_end)
+    {
+        int pos = search_start+(search_end-search_start)/2;
+        if (oid < list[pos].oid)
+        {
+            search_end = pos;
+        }
+        else if (list[pos].oid < oid)
+        {
+            search_start = pos+1;
+        }
+        else
+        {
+            list[pos].version = version;
+            return true;
+        }
+    }
+    return false;
+}
+
+void blockstore_impl_t::process_list(blockstore_op_t *op)
+{
+    uint32_t list_pg = op->offset;
+    uint32_t pg_count = op->len;
+    uint64_t pg_stripe_size = op->oid.stripe;
+    uint64_t min_inode = op->oid.inode;
+    uint64_t max_inode = op->version;
+    // Check PG
+    if (pg_count != 0 && (pg_stripe_size < MIN_BLOCK_SIZE || list_pg >= pg_count))
+    {
+        op->retval = -EINVAL;
+        FINISH_OP(op);
+        return;
+    }
+    // Copy clean_db entries (sorted)
+    int stable_count = 0, stable_alloc = clean_db.size() / (pg_count ? pg_count : 1);
+    obj_ver_id *stable = (obj_ver_id*)malloc(sizeof(obj_ver_id) * stable_alloc);
+    if (!stable)
+    {
+        op->retval = -ENOMEM;
+        FINISH_OP(op);
+        return;
+    }
+    {
+        auto clean_it = clean_db.begin(), clean_end = clean_db.end();
+        if ((min_inode != 0 || max_inode != 0) && min_inode <= max_inode)
+        {
+            clean_it = clean_db.lower_bound({
+                .inode = min_inode,
+                .stripe = 0,
+            });
+            clean_end = clean_db.upper_bound({
+                .inode = max_inode,
+                .stripe = UINT64_MAX,
+            });
+        }
+        for (; clean_it != clean_end; clean_it++)
+        {
+            if (!pg_count || ((clean_it->first.inode + clean_it->first.stripe / pg_stripe_size) % pg_count) == list_pg)
+            {
+                if (stable_count >= stable_alloc)
+                {
+                    stable_alloc += 32768;
+                    stable = (obj_ver_id*)realloc(stable, sizeof(obj_ver_id) * stable_alloc);
+                    if (!stable)
+                    {
+                        op->retval = -ENOMEM;
+                        FINISH_OP(op);
+                        return;
+                    }
+                }
+                stable[stable_count++] = {
+                    .oid = clean_it->first,
+                    .version = clean_it->second.version,
+                };
+            }
+        }
+    }
+    int clean_stable_count = stable_count;
+    // Copy dirty_db entries (sorted, too)
+    int unstable_count = 0, unstable_alloc = 0;
+    obj_ver_id *unstable = NULL;
+    {
+        auto dirty_it = dirty_db.begin(), dirty_end = dirty_db.end();
+        if ((min_inode != 0 || max_inode != 0) && min_inode <= max_inode)
+        {
+            dirty_it = dirty_db.lower_bound({
+                .oid = {
+                    .inode = min_inode,
+                    .stripe = 0,
+                },
+                .version = 0,
+            });
+            dirty_end = dirty_db.upper_bound({
+                .oid = {
+                    .inode = max_inode,
+                    .stripe = UINT64_MAX,
+                },
+                .version = UINT64_MAX,
+            });
+        }
+        for (; dirty_it != dirty_end; dirty_it++)
+        {
+            if (!pg_count || ((dirty_it->first.oid.inode + dirty_it->first.oid.stripe / pg_stripe_size) % pg_count) == list_pg)
+            {
+                if (IS_DELETE(dirty_it->second.state))
+                {
+                    // Deletions are always stable, so try to zero out two possible entries
+                    if (!replace_stable(dirty_it->first.oid, 0, 0, clean_stable_count, stable))
+                    {
+                        replace_stable(dirty_it->first.oid, 0, clean_stable_count, stable_count, stable);
+                    }
+                }
+                else if (IS_STABLE(dirty_it->second.state))
+                {
+                    // First try to replace a clean stable version in the first part of the list
+                    if (!replace_stable(dirty_it->first.oid, dirty_it->first.version, 0, clean_stable_count, stable))
+                    {
+                        // Then try to replace the last dirty stable version in the second part of the list
+                        if (stable_count > 0 && stable[stable_count-1].oid == dirty_it->first.oid)
+                        {
+                            stable[stable_count-1].version = dirty_it->first.version;
+                        }
+                        else
+                        {
+                            if (stable_count >= stable_alloc)
+                            {
+                                stable_alloc += 32768;
+                                stable = (obj_ver_id*)realloc(stable, sizeof(obj_ver_id) * stable_alloc);
+                                if (!stable)
+                                {
+                                    if (unstable)
+                                        free(unstable);
+                                    op->retval = -ENOMEM;
+                                    FINISH_OP(op);
+                                    return;
+                                }
+                            }
+                            stable[stable_count++] = dirty_it->first;
+                        }
+                    }
+                }
+                else
+                {
+                    if (unstable_count >= unstable_alloc)
+                    {
+                        unstable_alloc += 32768;
+                        unstable = (obj_ver_id*)realloc(unstable, sizeof(obj_ver_id) * unstable_alloc);
+                        if (!unstable)
+                        {
+                            if (stable)
+                                free(stable);
+                            op->retval = -ENOMEM;
+                            FINISH_OP(op);
+                            return;
+                        }
+                    }
+                    unstable[unstable_count++] = dirty_it->first;
+                }
+            }
+        }
+    }
+    // Remove zeroed out stable entries
+    int j = 0;
+    for (int i = 0; i < stable_count; i++)
+    {
+        if (stable[i].version != 0)
+        {
+            stable[j++] = stable[i];
+        }
+    }
+    stable_count = j;
+    if (stable_count+unstable_count > stable_alloc)
+    {
+        stable_alloc = stable_count+unstable_count;
+        stable = (obj_ver_id*)realloc(stable, sizeof(obj_ver_id) * stable_alloc);
+        if (!stable)
+        {
+            if (unstable)
+                free(unstable);
+            op->retval = -ENOMEM;
+            FINISH_OP(op);
+            return;
+        }
+    }
+    // Copy unstable entries
+    for (int i = 0; i < unstable_count; i++)
+    {
+        stable[j++] = unstable[i];
+    }
+    free(unstable);
+    op->version = stable_count;
+    op->retval = stable_count+unstable_count;
+    op->buf = stable;
+    FINISH_OP(op);
+}

src/blockstore_impl.h

@@ -1,14 +1,15 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
 #pragma once
 
 #include "blockstore.h"
-#include "timerfd_interval.h"
 
 #include <sys/types.h>
 #include <sys/ioctl.h>
 #include <sys/stat.h>
 #include <fcntl.h>
 #include <unistd.h>
-#include <malloc.h>
 #include <linux/fs.h>
 
 #include <vector>
@@ -16,43 +17,41 @@
 #include <deque>
 #include <new>
 
-#include "sparsepp/sparsepp/spp.h"
+#include "cpp-btree/btree_map.h"
 
+#include "malloc_or_die.h"
 #include "allocator.h"
 
 //#define BLOCKSTORE_DEBUG
 
 // States are not stored on disk. Instead, they're deduced from the journal
-// FIXME: Rename to BS_ST_*
 
-#define ST_J_IN_FLIGHT 1
-#define ST_J_SUBMITTED 2
-#define ST_J_WRITTEN 3
-#define ST_J_SYNCED 4
-#define ST_J_STABLE 5
+#define BS_ST_SMALL_WRITE 0x01
+#define BS_ST_BIG_WRITE 0x02
+#define BS_ST_DELETE 0x03
 
-#define ST_D_IN_FLIGHT 15
-#define ST_D_SUBMITTED 16
-#define ST_D_WRITTEN 17
-#define ST_D_META_WRITTEN 19
-#define ST_D_META_SYNCED 20
-#define ST_D_STABLE 21
+#define BS_ST_WAIT_DEL 0x10
+#define BS_ST_WAIT_BIG 0x20
+#define BS_ST_IN_FLIGHT 0x30
+#define BS_ST_SUBMITTED 0x40
+#define BS_ST_WRITTEN 0x50
+#define BS_ST_SYNCED 0x60
+#define BS_ST_STABLE 0x70
 
-#define ST_DEL_IN_FLIGHT 31
-#define ST_DEL_SUBMITTED 32
-#define ST_DEL_WRITTEN 33
-#define ST_DEL_SYNCED 34
-#define ST_DEL_STABLE 35
+#define BS_ST_INSTANT 0x100
 
-#define ST_CURRENT 48
+#define IMMEDIATE_NONE 0
+#define IMMEDIATE_SMALL 1
+#define IMMEDIATE_ALL 2
 
-#define IS_IN_FLIGHT(st) (st == ST_J_IN_FLIGHT || st == ST_D_IN_FLIGHT || st == ST_DEL_IN_FLIGHT || st == ST_J_SUBMITTED || st == ST_D_SUBMITTED || st == ST_DEL_SUBMITTED)
-#define IS_STABLE(st) (st == ST_J_STABLE || st == ST_D_STABLE || st == ST_DEL_STABLE || st == ST_CURRENT)
-#define IS_SYNCED(st) (IS_STABLE(st) || st == ST_J_SYNCED || st == ST_D_META_SYNCED || st == ST_DEL_SYNCED)
-#define IS_JOURNAL(st) (st >= ST_J_SUBMITTED && st <= ST_J_STABLE)
-#define IS_BIG_WRITE(st) (st >= ST_D_SUBMITTED && st <= ST_D_STABLE)
-#define IS_DELETE(st) (st >= ST_DEL_SUBMITTED && st <= ST_DEL_STABLE)
-#define IS_UNSYNCED(st) (st >= ST_J_SUBMITTED && st <= ST_J_WRITTEN || st >= ST_D_SUBMITTED && st <= ST_D_META_WRITTEN || st >= ST_DEL_SUBMITTED && st <= ST_DEL_WRITTEN)
+#define BS_ST_TYPE_MASK 0x0F
+#define BS_ST_WORKFLOW_MASK 0xF0
+#define IS_IN_FLIGHT(st) (((st) & 0xF0) <= BS_ST_SUBMITTED)
+#define IS_STABLE(st) (((st) & 0xF0) == BS_ST_STABLE)
+#define IS_SYNCED(st) (((st) & 0xF0) >= BS_ST_SYNCED)
+#define IS_JOURNAL(st) (((st) & 0x0F) == BS_ST_SMALL_WRITE)
+#define IS_BIG_WRITE(st) (((st) & 0x0F) == BS_ST_BIG_WRITE)
+#define IS_DELETE(st) (((st) & 0x0F) == BS_ST_DELETE)
 
 #define BS_SUBMIT_GET_SQE(sqe, data) \
     BS_SUBMIT_GET_ONLY_SQE(sqe); \
@@ -78,7 +77,8 @@
 
 #include "blockstore_journal.h"
 
-// 24 bytes + block bitmap per "clean" entry on disk with fixed metadata tables
+// 32 bytes = 24 bytes + block bitmap (4 bytes by default) + external attributes (also bitmap, 4 bytes by default)
+// per "clean" entry on disk with fixed metadata tables
 // FIXME: maybe add crc32's to metadata
 struct __attribute__((__packed__)) clean_disk_entry
 {
@@ -94,7 +94,7 @@ struct __attribute__((__packed__)) clean_entry
     uint64_t location;
 };
 
-// 56 = 24 + 32 bytes per dirty entry in memory (obj_ver_id => dirty_entry)
+// 64 = 24 + 40 bytes per dirty entry in memory (obj_ver_id => dirty_entry)
 struct __attribute__((__packed__)) dirty_entry
 {
     uint32_t state;
@@ -103,6 +103,7 @@ struct __attribute__((__packed__)) dirty_entry
     uint32_t offset;   // data offset within object (stripe)
     uint32_t len;      // data length
     uint64_t journal_sector; // journal sector used for this entry
+    void* bitmap;   // either external bitmap itself when it fits, or a pointer to it when it doesn't
 };
 
 // - Sync must be submitted after previous writes/deletes (not before!)
@@ -124,8 +125,6 @@ struct __attribute__((__packed__)) dirty_entry
 
 // Suspend operation until there are more free SQEs
 #define WAIT_SQE 1
-// Suspend operation until version <wait_detail> of object <oid> is written
-#define WAIT_IN_FLIGHT 2
 // Suspend operation until there are <wait_detail> bytes of free space in the journal on disk
 #define WAIT_JOURNAL 3
 // Suspend operation until the next journal sector buffer is free
@@ -139,7 +138,7 @@ struct fulfill_read_t
 };
 
 #define PRIV(op) ((blockstore_op_private_t*)(op)->private_data)
-#define FINISH_OP(op) PRIV(op)->~blockstore_op_private_t(); op->callback(op)
+#define FINISH_OP(op) PRIV(op)->~blockstore_op_private_t(); std::function<void (blockstore_op_t*)>(op->callback)(op)
 
 struct blockstore_op_private_t
 {
@@ -147,23 +146,28 @@ struct blockstore_op_private_t
     int wait_for;
     uint64_t wait_detail;
     int pending_ops;
+    int op_state;
 
     // Read
     std::vector<fulfill_read_t> read_vec;
 
     // Sync, write
-    uint64_t min_used_journal_sector, max_used_journal_sector;
+    uint64_t min_flushed_journal_sector, max_flushed_journal_sector;
 
     // Write
     struct iovec iov_zerofill[3];
+    // Warning: must not have a default value here because it's written to before calling constructor in blockstore_write.cpp O_o
+    uint64_t real_version;
 
     // Sync
     std::vector<obj_ver_id> sync_big_writes, sync_small_writes;
     int sync_small_checked, sync_big_checked;
-    std::list<blockstore_op_t*>::iterator in_progress_ptr;
-    int sync_state, prev_sync_count;
 };
 
+// https://github.com/algorithm-ninja/cpp-btree
+// https://github.com/greg7mdp/sparsepp/ was used previously, but it was TERRIBLY slow after resizing
+// with sparsepp, random reads dropped to ~700 iops very fast with just as much as ~32k objects in the DB
+typedef btree::btree_map<object_id, clean_entry> blockstore_clean_db_t;
 typedef std::map<obj_ver_id, dirty_entry> blockstore_dirty_db_t;
 
 #include "blockstore_init.h"
@@ -177,34 +181,37 @@ class blockstore_impl_t
     uint32_t block_size;
     uint64_t meta_offset;
     uint64_t data_offset;
-    uint64_t cfg_journal_size;
+    uint64_t cfg_journal_size, cfg_data_size;
     // Required write alignment and journal/metadata/data areas' location alignment
-    uint32_t disk_alignment = 512;
+    uint32_t disk_alignment = 4096;
     // Journal block size - minimum_io_size of the journal device is the best choice
-    uint64_t journal_block_size = 512;
+    uint64_t journal_block_size = 4096;
     // Metadata block size - minimum_io_size of the metadata device is the best choice
-    uint64_t meta_block_size = 512;
+    uint64_t meta_block_size = 4096;
     // Sparse write tracking granularity. 4 KB is a good choice. Must be a multiple of disk_alignment
     uint64_t bitmap_granularity = 4096;
     bool readonly = false;
+    // By default, Blockstore locks all opened devices exclusively. This option can be used to disable locking
+    bool disable_flock = false;
     // It is safe to disable fsync() if drive write cache is writethrough
     bool disable_data_fsync = false, disable_meta_fsync = false, disable_journal_fsync = false;
     // Enable if you want every operation to be executed with an "implicit fsync"
-    // FIXME Not implemented yet
-    bool immediate_commit = false;
+    // Suitable only for server SSDs with capacitors, requires disabled data and journal fsyncs
+    int immediate_commit = IMMEDIATE_NONE;
     bool inmemory_meta = false;
-    int flusher_count;
+    // Maximum flusher count
+    unsigned flusher_count;
+    // Maximum queue depth
+    unsigned max_write_iodepth = 128;
     /******* END OF OPTIONS *******/
 
     struct ring_consumer_t ring_consumer;
 
-    // Another option is https://github.com/algorithm-ninja/cpp-btree
-    spp::sparse_hash_map<object_id, clean_entry> clean_db;
+    blockstore_clean_db_t clean_db;
     uint8_t *clean_bitmap = NULL;
     blockstore_dirty_db_t dirty_db;
-    std::list<blockstore_op_t*> submit_queue; // FIXME: funny thing is that vector is better here
+    std::vector<blockstore_op_t*> submit_queue;
     std::vector<obj_ver_id> unsynced_big_writes, unsynced_small_writes;
-    std::list<blockstore_op_t*> in_progress_syncs; // ...and probably here, too
     allocator *data_alloc = NULL;
     uint8_t *zero_object;
 
@@ -221,6 +228,7 @@ class blockstore_impl_t
 
     struct journal_t journal;
     journal_flusher_t *flusher;
+    int write_iodepth = 0;
 
     bool live = false, queue_stall = false;
     ring_loop_t *ringloop;
@@ -243,6 +251,7 @@ class blockstore_impl_t
     void open_data();
     void open_meta();
     void open_journal();
+    uint8_t* get_clean_entry_bitmap(uint64_t block_loc, int offset);
 
     // Asynchronous init
     int initialized;
@@ -262,26 +271,29 @@ class blockstore_impl_t
 
     // Write
     bool enqueue_write(blockstore_op_t *op);
+    void cancel_all_writes(blockstore_op_t *op, blockstore_dirty_db_t::iterator dirty_it, int retval);
     int dequeue_write(blockstore_op_t *op);
     int dequeue_del(blockstore_op_t *op);
-    void ack_write(blockstore_op_t *op);
+    int continue_write(blockstore_op_t *op);
     void release_journal_sectors(blockstore_op_t *op);
     void handle_write_event(ring_data_t *data, blockstore_op_t *op);
 
     // Sync
-    int dequeue_sync(blockstore_op_t *op);
+    int continue_sync(blockstore_op_t *op, bool queue_has_in_progress_sync);
     void handle_sync_event(ring_data_t *data, blockstore_op_t *op);
-    int continue_sync(blockstore_op_t *op);
-    void ack_one_sync(blockstore_op_t *op);
-    int ack_sync(blockstore_op_t *op);
+    void ack_sync(blockstore_op_t *op);
 
     // Stabilize
     int dequeue_stable(blockstore_op_t *op);
+    int continue_stable(blockstore_op_t *op);
+    void mark_stable(const obj_ver_id & ov);
     void handle_stable_event(ring_data_t *data, blockstore_op_t *op);
     void stabilize_object(object_id oid, uint64_t max_ver);
 
     // Rollback
     int dequeue_rollback(blockstore_op_t *op);
+    int continue_rollback(blockstore_op_t *op);
+    void mark_rolled_back(const obj_ver_id & ov);
     void handle_rollback_event(ring_data_t *data, blockstore_op_t *op);
     void erase_dirty(blockstore_dirty_db_t::iterator dirty_start, blockstore_dirty_db_t::iterator dirty_end, uint64_t clean_loc);
 
@@ -309,12 +321,16 @@ public:
     bool is_stalled();
 
     // Submission
-    void enqueue_op(blockstore_op_t *op, bool first = false);
+    void enqueue_op(blockstore_op_t *op);
 
     // Unstable writes are added here (map of object_id -> version)
     std::unordered_map<object_id, uint64_t> unstable_writes;
 
+    // Space usage statistics
+    std::map<uint64_t, uint64_t> inode_space_stats;
+
     inline uint32_t get_block_size() { return block_size; }
     inline uint64_t get_block_count() { return block_count; }
-    inline uint32_t get_disk_alignment() { return disk_alignment; }
+    inline uint64_t get_free_block_count() { return data_alloc->get_free_count(); }
+    inline uint32_t get_bitmap_granularity() { return disk_alignment; }
 };

src/blockstore_init.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
 #include "blockstore_impl.h"
 
 blockstore_init_meta::blockstore_init_meta(blockstore_impl_t *bs)
@@ -97,7 +100,7 @@ void blockstore_init_meta::handle_entries(void* entries, unsigned count, int blo
         clean_disk_entry *entry = (clean_disk_entry*)(entries + i*bs->clean_entry_size);
         if (!bs->inmemory_meta && bs->clean_entry_bitmap_size)
         {
-            memcpy(bs->clean_bitmap + (done_cnt+i)*bs->clean_entry_bitmap_size, &entry->bitmap, bs->clean_entry_bitmap_size);
+            memcpy(bs->clean_bitmap + (done_cnt+i)*2*bs->clean_entry_bitmap_size, &entry->bitmap, 2*bs->clean_entry_bitmap_size);
         }
         if (entry->oid.inode > 0)
         {
@@ -108,13 +111,17 @@ void blockstore_init_meta::handle_entries(void* entries, unsigned count, int blo
                 {
                     // free the previous block
 #ifdef BLOCKSTORE_DEBUG
-                    printf("Free block %lu\n", clean_it->second.location >> bs->block_order);
+                    printf("Free block %lu (new location is %lu)\n", clean_it->second.location >> block_order, done_cnt+i);
 #endif
                     bs->data_alloc->set(clean_it->second.location >> block_order, false);
                 }
+                else
+                {
+                    bs->inode_space_stats[entry->oid.inode] += bs->block_size;
+                }
                 entries_loaded++;
 #ifdef BLOCKSTORE_DEBUG
-                printf("Allocate block (clean entry) %lu: %lu:%lu v%lu\n", done_cnt+i, entry->oid.inode, entry->oid.stripe, entry->version);
+                printf("Allocate block (clean entry) %lu: %lx:%lx v%lu\n", done_cnt+i, entry->oid.inode, entry->oid.stripe, entry->version);
 #endif
                 bs->data_alloc->set(done_cnt+i, true);
                 bs->clean_db[entry->oid] = (struct clean_entry){
@@ -125,7 +132,7 @@ void blockstore_init_meta::handle_entries(void* entries, unsigned count, int blo
             else
             {
 #ifdef BLOCKSTORE_DEBUG
-                printf("Old clean entry %lu: %lu:%lu v%lu\n", done_cnt+i, entry->oid.inode, entry->oid.stripe, entry->version);
+                printf("Old clean entry %lu: %lx:%lx v%lu\n", done_cnt+i, entry->oid.inode, entry->oid.stripe, entry->version);
 #endif
             }
         }
@@ -202,11 +209,7 @@ int blockstore_init_journal::loop()
         goto resume_7;
     printf("Reading blockstore journal\n");
     if (!bs->journal.inmemory)
-    {
-        submitted_buf = memalign(MEM_ALIGNMENT, 2*bs->journal.block_size);
-        if (!submitted_buf)
-            throw std::bad_alloc();
-    }
+        submitted_buf = memalign_or_die(MEM_ALIGNMENT, 2*bs->journal.block_size);
     else
         submitted_buf = bs->journal.buffer;
     // Read first block of the journal
@@ -317,7 +320,7 @@ resume_1:
                 if (journal_pos < bs->journal.used_start)
                     end = bs->journal.used_start;
                 if (!bs->journal.inmemory)
-                    submitted_buf = memalign(MEM_ALIGNMENT, JOURNAL_BUFFER_SIZE);
+                    submitted_buf = memalign_or_die(MEM_ALIGNMENT, JOURNAL_BUFFER_SIZE);
                 else
                     submitted_buf = bs->journal.buffer + journal_pos;
                 data->iov = {
@@ -400,10 +403,10 @@ resume_1:
             }
         }
     }
-    // Trim journal on start so we don't stall when all entries are older
-    bs->journal.trim();
+    bs->flusher->mark_trim_possible();
+    bs->journal.dirty_start = bs->journal.next_free;
     printf(
-        "Journal entries loaded: %lu, free journal space: %lu bytes (%lu..%lu is used), free blocks: %lu / %lu\n",
+        "Journal entries loaded: %lu, free journal space: %lu bytes (%08lx..%08lx is used), free blocks: %lu / %lu\n",
         entries_loaded,
         (bs->journal.next_free >= bs->journal.used_start
             ? bs->journal.len-bs->journal.block_size - (bs->journal.next_free-bs->journal.used_start)
@@ -439,7 +442,7 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
         {
             journal_entry *je = (journal_entry*)(buf + proc_pos - done_pos + pos);
             if (je->magic != JOURNAL_MAGIC || je_crc32(je) != je->crc32 ||
-                je->type < JE_SMALL_WRITE || je->type > JE_DELETE || started && je->crc32_prev != crc32_last)
+                je->type < JE_MIN || je->type > JE_MAX || started && je->crc32_prev != crc32_last)
             {
                 if (pos == 0)
                 {
@@ -453,10 +456,15 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
                     break;
                 }
             }
-            if (je->type == JE_SMALL_WRITE)
+            if (je->type == JE_SMALL_WRITE || je->type == JE_SMALL_WRITE_INSTANT)
             {
 #ifdef BLOCKSTORE_DEBUG
-                printf("je_small_write oid=%lu:%lu ver=%lu offset=%u len=%u\n", je->small_write.oid.inode, je->small_write.oid.stripe, je->small_write.version, je->small_write.offset, je->small_write.len);
+                printf(
+                    "je_small_write%s oid=%lx:%lx ver=%lu offset=%u len=%u\n",
+                    je->type == JE_SMALL_WRITE_INSTANT ? "_instant" : "",
+                    je->small_write.oid.inode, je->small_write.oid.stripe, je->small_write.version,
+                    je->small_write.offset, je->small_write.len
+                );
 #endif
                 // oid, version, offset, len
                 uint64_t prev_free = next_free;
@@ -474,7 +482,7 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
                 if (location != je->small_write.data_offset)
                 {
                     char err[1024];
-                    snprintf(err, 1024, "BUG: calculated journal data offset (%lu) != stored journal data offset (%lu)", location, je->small_write.data_offset);
+                    snprintf(err, 1024, "BUG: calculated journal data offset (%08lx) != stored journal data offset (%08lx)", location, je->small_write.data_offset);
                     throw std::runtime_error(err);
                 }
                 uint32_t data_crc32 = 0;
@@ -509,7 +517,9 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
                 if (data_crc32 != je->small_write.crc32_data)
                 {
                     // journal entry is corrupt, stop here
-                    // interesting thing is that we must clear the corrupt entry if we're not readonly
+                    // interesting thing is that we must clear the corrupt entry if we're not readonly,
+                    // because we don't write next entries in the same journal block
+                    printf("Journal entry data is corrupt (data crc32 %x != %x)\n", data_crc32, je->small_write.crc32_data);
                     memset(buf + proc_pos - done_pos + pos, 0, bs->journal.block_size - pos);
                     bs->journal.next_free = prev_free;
                     init_write_buf = buf + proc_pos - done_pos;
@@ -518,33 +528,105 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
                 }
                 auto clean_it = bs->clean_db.find(je->small_write.oid);
                 if (clean_it == bs->clean_db.end() ||
-                    clean_it->second.version < je->big_write.version)
+                    clean_it->second.version < je->small_write.version)
                 {
                     obj_ver_id ov = {
                         .oid = je->small_write.oid,
                         .version = je->small_write.version,
                     };
+                    void *bmp = (void*)je + sizeof(journal_entry_small_write);
+                    if (bs->clean_entry_bitmap_size <= sizeof(void*))
+                    {
+                        memcpy(&bmp, bmp, bs->clean_entry_bitmap_size);
+                    }
+                    else if (!bs->journal.inmemory)
+                    {
+                        // FIXME Using large blockstore objects and not keeping journal in memory
+                        // will result in a lot of small allocations for entry bitmaps. This can
+                        // only be fixed by using a patched map with dynamic entry size, but not
+                        // the btree_map, because it doesn't keep iterators valid all the time.
+                        void *bmp_cp = malloc_or_die(bs->clean_entry_bitmap_size);
+                        memcpy(bmp_cp, bmp, bs->clean_entry_bitmap_size);
+                        bmp = bmp_cp;
+                    }
                     bs->dirty_db.emplace(ov, (dirty_entry){
-                        .state = ST_J_SYNCED,
+                        .state = (BS_ST_SMALL_WRITE | BS_ST_SYNCED),
                         .flags = 0,
                         .location = location,
                         .offset = je->small_write.offset,
                         .len = je->small_write.len,
                         .journal_sector = proc_pos,
+                        .bitmap = bmp,
                     });
                     bs->journal.used_sectors[proc_pos]++;
 #ifdef BLOCKSTORE_DEBUG
-                    printf("journal offset %lu is used by %lu:%lu v%lu\n", proc_pos, ov.oid.inode, ov.oid.stripe, ov.version);
+                    printf(
+                        "journal offset %08lx is used by %lx:%lx v%lu (%lu refs)\n",
+                        proc_pos, ov.oid.inode, ov.oid.stripe, ov.version, bs->journal.used_sectors[proc_pos]
+                    );
 #endif
                     auto & unstab = bs->unstable_writes[ov.oid];
                     unstab = unstab < ov.version ? ov.version : unstab;
+                    if (je->type == JE_SMALL_WRITE_INSTANT)
+                    {
+                        bs->mark_stable(ov);
+                    }
                 }
             }
-            else if (je->type == JE_BIG_WRITE)
+            else if (je->type == JE_BIG_WRITE || je->type == JE_BIG_WRITE_INSTANT)
             {
 #ifdef BLOCKSTORE_DEBUG
-                printf("je_big_write oid=%lu:%lu ver=%lu loc=%lu\n", je->big_write.oid.inode, je->big_write.oid.stripe, je->big_write.version, je->big_write.location);
+                printf(
+                    "je_big_write%s oid=%lx:%lx ver=%lu loc=%lu\n",
+                    je->type == JE_BIG_WRITE_INSTANT ? "_instant" : "",
+                    je->big_write.oid.inode, je->big_write.oid.stripe, je->big_write.version, je->big_write.location >> bs->block_order
+                );
 #endif
+                auto dirty_it = bs->dirty_db.upper_bound((obj_ver_id){
+                    .oid = je->big_write.oid,
+                    .version = UINT64_MAX,
+                });
+                if (dirty_it != bs->dirty_db.begin() && bs->dirty_db.size() > 0)
+                {
+                    dirty_it--;
+                    if (dirty_it->first.oid == je->big_write.oid &&
+                        dirty_it->first.version >= je->big_write.version &&
+                        (dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_DELETE)
+                    {
+                        // It is allowed to overwrite a deleted object with a
+                        // version number smaller than deletion version number,
+                        // because the presence of a BIG_WRITE entry means that
+                        // its data and metadata are already flushed.
+                        // We don't know if newer versions are flushed, but
+                        // the previous delete definitely is.
+                        // So we flush previous dirty entries, but retain the clean one.
+                        // This feature is required for writes happening shortly
+                        // after deletes.
+                        auto dirty_end = dirty_it;
+                        dirty_end++;
+                        while (1)
+                        {
+                            if (dirty_it == bs->dirty_db.begin())
+                            {
+                                break;
+                            }
+                            dirty_it--;
+                            if (dirty_it->first.oid != je->big_write.oid)
+                            {
+                                dirty_it++;
+                                break;
+                            }
+                        }
+                        auto clean_it = bs->clean_db.find(je->big_write.oid);
+                        bs->erase_dirty(
+                            dirty_it, dirty_end,
+                            clean_it != bs->clean_db.end() ? clean_it->second.location : UINT64_MAX
+                        );
+                        // Remove it from the flusher's queue, too
+                        // Otherwise it may end up referring to a small unstable write after reading the rest of the journal
+                        bs->flusher->remove_flush(je->big_write.oid);
+                    }
+                }
                 auto clean_it = bs->clean_db.find(je->big_write.oid);
                 if (clean_it == bs->clean_db.end() ||
                     clean_it->second.version < je->big_write.version)
@@ -554,131 +636,100 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
                         .oid = je->big_write.oid,
                         .version = je->big_write.version,
                     };
+                    void *bmp = (void*)je + sizeof(journal_entry_big_write);
+                    if (bs->clean_entry_bitmap_size <= sizeof(void*))
+                    {
+                        memcpy(&bmp, bmp, bs->clean_entry_bitmap_size);
+                    }
+                    else if (!bs->journal.inmemory)
+                    {
+                        // FIXME Using large blockstore objects and not keeping journal in memory
+                        // will result in a lot of small allocations for entry bitmaps. This can
+                        // only be fixed by using a patched map with dynamic entry size, but not
+                        // the btree_map, because it doesn't keep iterators valid all the time.
+                        void *bmp_cp = malloc_or_die(bs->clean_entry_bitmap_size);
+                        memcpy(bmp_cp, bmp, bs->clean_entry_bitmap_size);
+                        bmp = bmp_cp;
+                    }
                     bs->dirty_db.emplace(ov, (dirty_entry){
-                        .state = ST_D_META_SYNCED,
+                        .state = (BS_ST_BIG_WRITE | BS_ST_SYNCED),
                         .flags = 0,
                         .location = je->big_write.location,
                         .offset = je->big_write.offset,
                         .len = je->big_write.len,
                         .journal_sector = proc_pos,
+                        .bitmap = bmp,
                     });
 #ifdef BLOCKSTORE_DEBUG
                     printf("Allocate block %lu\n", je->big_write.location >> bs->block_order);
 #endif
                     bs->data_alloc->set(je->big_write.location >> bs->block_order, true);
                     bs->journal.used_sectors[proc_pos]++;
+#ifdef BLOCKSTORE_DEBUG
+                    printf(
+                        "journal offset %08lx is used by %lx:%lx v%lu (%lu refs)\n",
+                        proc_pos, ov.oid.inode, ov.oid.stripe, ov.version, bs->journal.used_sectors[proc_pos]
+                    );
+#endif
                     auto & unstab = bs->unstable_writes[ov.oid];
                     unstab = unstab < ov.version ? ov.version : unstab;
+                    if (je->type == JE_BIG_WRITE_INSTANT)
+                    {
+                        bs->mark_stable(ov);
+                    }
                 }
             }
             else if (je->type == JE_STABLE)
             {
 #ifdef BLOCKSTORE_DEBUG
-                printf("je_stable oid=%lu:%lu ver=%lu\n", je->stable.oid.inode, je->stable.oid.stripe, je->stable.version);
+                printf("je_stable oid=%lx:%lx ver=%lu\n", je->stable.oid.inode, je->stable.oid.stripe, je->stable.version);
 #endif
                 // oid, version
                 obj_ver_id ov = {
                     .oid = je->stable.oid,
                     .version = je->stable.version,
                 };
-                auto it = bs->dirty_db.find(ov);
-                if (it == bs->dirty_db.end())
-                {
-                    // journal contains a legitimate STABLE entry for a non-existing dirty write
-                    // this probably means that journal was trimmed between WRITE and STABLE entries
-                    // skip it
-                }
-                else
-                {
-                    while (1)
-                    {
-                        it->second.state = (it->second.state == ST_D_META_SYNCED
-                            ? ST_D_STABLE
-                            : (it->second.state == ST_DEL_SYNCED ? ST_DEL_STABLE : ST_J_STABLE));
-                        if (it == bs->dirty_db.begin())
-                            break;
-                        it--;
-                        if (it->first.oid != ov.oid || IS_STABLE(it->second.state))
-                            break;
-                    }
-                    bs->flusher->enqueue_flush(ov);
-                }
-                auto unstab_it = bs->unstable_writes.find(ov.oid);
-                if (unstab_it != bs->unstable_writes.end() && unstab_it->second <= ov.version)
-                {
-                    bs->unstable_writes.erase(unstab_it);
-                }
+                bs->mark_stable(ov);
             }
             else if (je->type == JE_ROLLBACK)
             {
 #ifdef BLOCKSTORE_DEBUG
-                printf("je_rollback oid=%lu:%lu ver=%lu\n", je->rollback.oid.inode, je->rollback.oid.stripe, je->rollback.version);
+                printf("je_rollback oid=%lx:%lx ver=%lu\n", je->rollback.oid.inode, je->rollback.oid.stripe, je->rollback.version);
 #endif
                 // rollback dirty writes of <oid> up to <version>
-                auto it = bs->dirty_db.lower_bound((obj_ver_id){
+                obj_ver_id ov = {
                     .oid = je->rollback.oid,
-                    .version = UINT64_MAX,
-                });
-                if (it != bs->dirty_db.begin())
-                {
-                    uint64_t max_unstable = 0;
-                    auto rm_start = it;
-                    auto rm_end = it;
-                    it--;
-                    while (it->first.oid == je->rollback.oid &&
-                        it->first.version > je->rollback.version &&
-                        !IS_IN_FLIGHT(it->second.state) &&
-                        !IS_STABLE(it->second.state))
-                    {
-                        if (it->first.oid != je->rollback.oid)
-                            break;
-                        else if (it->first.version <= je->rollback.version)
-                        {
-                            if (!IS_STABLE(it->second.state))
-                                max_unstable = it->first.version;
-                            break;
-                        }
-                        else if (IS_STABLE(it->second.state))
-                            break;
-                        // Remove entry
-                        rm_start = it;
-                        if (it == bs->dirty_db.begin())
-                            break;
-                        it--;
-                    }
-                    if (rm_start != rm_end)
-                    {
-                        bs->erase_dirty(rm_start, rm_end, UINT64_MAX);
-                    }
-                    auto unstab_it = bs->unstable_writes.find(je->rollback.oid);
-                    if (unstab_it != bs->unstable_writes.end())
-                    {
-                        if (max_unstable == 0)
-                            bs->unstable_writes.erase(unstab_it);
-                        else
-                            unstab_it->second = max_unstable;
-                    }
-                }
+                    .version = je->rollback.version,
+                };
+                bs->mark_rolled_back(ov);
             }
             else if (je->type == JE_DELETE)
             {
 #ifdef BLOCKSTORE_DEBUG
-                printf("je_delete oid=%lu:%lu ver=%lu\n", je->del.oid.inode, je->del.oid.stripe, je->del.version);
+                printf("je_delete oid=%lx:%lx ver=%lu\n", je->del.oid.inode, je->del.oid.stripe, je->del.version);
 #endif
-                // oid, version
-                obj_ver_id ov = {
-                    .oid = je->del.oid,
-                    .version = je->del.version,
-                };
-                bs->dirty_db.emplace(ov, (dirty_entry){
-                    .state = ST_DEL_SYNCED,
-                    .flags = 0,
-                    .location = 0,
-                    .offset = 0,
-                    .len = 0,
-                    .journal_sector = proc_pos,
-                });
-                bs->journal.used_sectors[proc_pos]++;
+                auto clean_it = bs->clean_db.find(je->del.oid);
+                if (clean_it != bs->clean_db.end() &&
+                    clean_it->second.version < je->del.version)
+                {
+                    // oid, version
+                    obj_ver_id ov = {
+                        .oid = je->del.oid,
+                        .version = je->del.version,
+                    };
+                    bs->dirty_db.emplace(ov, (dirty_entry){
+                        .state = (BS_ST_DELETE | BS_ST_SYNCED),
+                        .flags = 0,
+                        .location = 0,
+                        .offset = 0,
+                        .len = 0,
+                        .journal_sector = proc_pos,
+                    });
+                    bs->journal.used_sectors[proc_pos]++;
+                    // Deletions are treated as immediately stable, because
+                    // "2-phase commit" (write->stabilize) isn't sufficient for them anyway
+                    bs->mark_stable(ov);
+                }
             }
             started = true;
             pos += je->size;

src/blockstore_init.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
 #pragma once
 
 class blockstore_init_meta

src/blockstore_journal.cpp

@@ -1,31 +1,45 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
 #include "blockstore_impl.h"
 
 blockstore_journal_check_t::blockstore_journal_check_t(blockstore_impl_t *bs)
 {
     this->bs = bs;
-    sectors_required = 0;
+    sectors_to_write = 0;
     next_pos = bs->journal.next_free;
     next_sector = bs->journal.cur_sector;
+    first_sector = -1;
     next_in_pos = bs->journal.in_sector_pos;
     right_dir = next_pos >= bs->journal.used_start;
 }
 
 // Check if we can write <required> entries of <size> bytes and <data_after> data bytes after them to the journal
-int blockstore_journal_check_t::check_available(blockstore_op_t *op, int required, int size, int data_after)
+int blockstore_journal_check_t::check_available(blockstore_op_t *op, int entries_required, int size, int data_after)
 {
+    int required = entries_required;
     while (1)
     {
-        int fits = (bs->journal.block_size - next_in_pos) / size;
+        int fits = bs->journal.no_same_sector_overwrites && next_pos == bs->journal.next_free && bs->journal.sector_info[next_sector].written
+            ? 0
+            : (bs->journal.block_size - next_in_pos) / size;
         if (fits > 0)
         {
+            if (fits > required)
+            {
+                fits = required;
+            }
+            if (first_sector == -1)
+            {
+                first_sector = next_sector;
+            }
             required -= fits;
             next_in_pos += fits * size;
-            sectors_required++;
+            sectors_to_write++;
         }
         else if (bs->journal.sector_info[next_sector].dirty)
         {
-            // sectors_required is more like "sectors to write"
-            sectors_required++;
+            sectors_to_write++;
         }
         if (required <= 0)
         {
@@ -38,19 +52,41 @@ int blockstore_journal_check_t::check_available(blockstore_op_t *op, int require
             right_dir = false;
         }
         next_in_pos = 0;
-        if (bs->journal.sector_info[next_sector].usage_count > 0 ||
-            bs->journal.sector_info[next_sector].dirty)
+        next_sector = ((next_sector + 1) % bs->journal.sector_count);
+        if (next_sector == first_sector)
         {
-            next_sector = ((next_sector + 1) % bs->journal.sector_count);
+            // next_sector may wrap when all sectors are flushed and the incoming batch is too big
+            // This is an error condition, we can't wait for anything in this case
+            throw std::runtime_error(
+                "Blockstore journal_sector_buffer_count="+std::to_string(bs->journal.sector_count)+
+                " is too small for a batch of "+std::to_string(entries_required)+" entries of "+std::to_string(size)+" bytes"
+            );
         }
-        if (bs->journal.sector_info[next_sector].usage_count > 0 ||
+        if (bs->journal.sector_info[next_sector].flush_count > 0 ||
             bs->journal.sector_info[next_sector].dirty)
         {
             // No memory buffer available. Wait for it.
-#ifdef BLOCKSTORE_DEBUG
-            printf("next journal buffer %d is still dirty=%d used=%d\n", next_sector,
-                bs->journal.sector_info[next_sector].dirty, bs->journal.sector_info[next_sector].usage_count);
-#endif
+            int used = 0, dirty = 0;
+            for (int i = 0; i < bs->journal.sector_count; i++)
+            {
+                if (bs->journal.sector_info[i].dirty)
+                {
+                    dirty++;
+                    used++;
+                }
+                if (bs->journal.sector_info[i].flush_count > 0)
+                {
+                    used++;
+                }
+            }
+            // In fact, it's even more rare than "ran out of journal space", so print a warning
+            printf(
+                "Ran out of journal sector buffers: %d/%lu buffers used (%d dirty), next buffer (%ld)"
+                " is %s and flushed %lu times. Consider increasing \'journal_sector_buffer_count\'\n",
+                used, bs->journal.sector_count, dirty, next_sector,
+                bs->journal.sector_info[next_sector].dirty ? "dirty" : "not dirty",
+                bs->journal.sector_info[next_sector].flush_count
+            );
             PRIV(op)->wait_for = WAIT_JOURNAL_BUFFER;
             return 0;
         }
@@ -68,13 +104,11 @@ int blockstore_journal_check_t::check_available(blockstore_op_t *op, int require
     {
         // No space in the journal. Wait until used_start changes.
         printf(
-            "Ran out of journal space (free space: %lu bytes)\n",
-            (bs->journal.next_free >= bs->journal.used_start
-                ? bs->journal.len-bs->journal.block_size - (bs->journal.next_free-bs->journal.used_start)
-                : bs->journal.used_start - bs->journal.next_free)
+            "Ran out of journal space (used_start=%08lx, next_free=%08lx, dirty_start=%08lx)\n",
+            bs->journal.used_start, bs->journal.next_free, bs->journal.dirty_start
         );
         PRIV(op)->wait_for = WAIT_JOURNAL;
-        bs->flusher->force_start();
+        bs->flusher->request_trim();
         PRIV(op)->wait_detail = bs->journal.used_start;
         return 0;
     }
@@ -83,15 +117,21 @@ int blockstore_journal_check_t::check_available(blockstore_op_t *op, int require
 
 journal_entry* prefill_single_journal_entry(journal_t & journal, uint16_t type, uint32_t size)
 {
-    if (journal.block_size - journal.in_sector_pos < size)
+    if (!journal.entry_fits(size))
     {
         assert(!journal.sector_info[journal.cur_sector].dirty);
         // Move to the next journal sector
-        if (journal.sector_info[journal.cur_sector].usage_count > 0)
+        if (journal.sector_info[journal.cur_sector].flush_count > 0)
         {
             // Also select next sector buffer in memory
             journal.cur_sector = ((journal.cur_sector + 1) % journal.sector_count);
+            assert(!journal.sector_info[journal.cur_sector].flush_count);
         }
+        else
+        {
+            journal.dirty_start = journal.next_free;
+        }
+        journal.sector_info[journal.cur_sector].written = false;
         journal.sector_info[journal.cur_sector].offset = journal.next_free;
         journal.in_sector_pos = 0;
         journal.next_free = (journal.next_free+journal.block_size) < journal.len ? journal.next_free + journal.block_size : journal.block_size;
@@ -116,7 +156,8 @@ journal_entry* prefill_single_journal_entry(journal_t & journal, uint16_t type,
 void prepare_journal_sector_write(journal_t & journal, int cur_sector, io_uring_sqe *sqe, std::function<void(ring_data_t*)> cb)
 {
     journal.sector_info[cur_sector].dirty = false;
-    journal.sector_info[cur_sector].usage_count++;
+    journal.sector_info[cur_sector].written = true;
+    journal.sector_info[cur_sector].flush_count++;
     ring_data_t *data = ((ring_data_t*)sqe->user_data);
     data->iov = (struct iovec){
         (journal.inmemory
@@ -143,13 +184,13 @@ journal_t::~journal_t()
     buffer = NULL;
 }
 
-bool journal_t::trim()
+uint64_t journal_t::get_trim_pos()
 {
     auto journal_used_it = used_sectors.lower_bound(used_start);
 #ifdef BLOCKSTORE_DEBUG
     printf(
-        "Trimming journal (used_start=%lu, next_free=%lu, first_used=%lu, usage_count=%lu)\n",
-        used_start, next_free,
+        "Trimming journal (used_start=%08lx, next_free=%08lx, dirty_start=%08lx, new_start=%08lx, new_refcount=%ld)\n",
+        used_start, next_free, dirty_start,
         journal_used_it == used_sectors.end() ? 0 : journal_used_it->first,
         journal_used_it == used_sectors.end() ? 0 : journal_used_it->second
     );
@@ -161,26 +202,19 @@ bool journal_t::trim()
         if (journal_used_it == used_sectors.end())
         {
             // Journal is empty
-            used_start = next_free;
+            return next_free;
         }
         else
         {
-            used_start = journal_used_it->first;
-            // next_free does not need updating here
+            // next_free does not need updating during trim
+            return journal_used_it->first;
         }
     }
     else if (journal_used_it->first > used_start)
     {
         // Journal is cleared up to <journal_used_it>
-        used_start = journal_used_it->first;
+        return journal_used_it->first;
     }
-    else
-    {
-        // Can't trim journal
-        return false;
-    }
-#ifdef BLOCKSTORE_DEBUG
-    printf("Journal trimmed to %lu (next_free=%lu)\n", used_start, next_free);
-#endif
-    return true;
+    // Can't trim journal
+    return used_start;
 }

src/blockstore_journal.h

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
 #pragma once
 
 #include "crc32c.h"
@@ -7,17 +10,23 @@
 #define JOURNAL_BUFFER_SIZE 4*1024*1024
 
 // We reserve some extra space for future stabilize requests during writes
+// FIXME: This value should be dynamic i.e. Blockstore ideally shouldn't allow
+// writing more than can be stabilized afterwards
 #define JOURNAL_STABILIZE_RESERVATION 65536
 
 // Journal entries
 // Journal entries are linked to each other by their crc32 value
 // The journal is almost a blockchain, because object versions constantly increase
+#define JE_MIN         0x01
 #define JE_START       0x01
 #define JE_SMALL_WRITE 0x02
 #define JE_BIG_WRITE   0x03
 #define JE_STABLE      0x04
 #define JE_DELETE      0x05
 #define JE_ROLLBACK    0x06
+#define JE_SMALL_WRITE_INSTANT 0x07
+#define JE_BIG_WRITE_INSTANT   0x08
+#define JE_MAX         0x08
 
 // crc32c comes first to ease calculation and is equal to crc32()
 struct __attribute__((__packed__)) journal_entry_start
@@ -45,6 +54,9 @@ struct __attribute__((__packed__)) journal_entry_small_write
     // data_offset is its offset within journal
     uint64_t data_offset;
     uint32_t crc32_data;
+    // small_write and big_write entries are followed by the "external" bitmap
+    // its size is dynamic and included in journal entry's <size> field
+    uint8_t bitmap[];
 };
 
 struct __attribute__((__packed__)) journal_entry_big_write
@@ -59,6 +71,9 @@ struct __attribute__((__packed__)) journal_entry_big_write
     uint32_t offset;
     uint32_t len;
     uint64_t location;
+    // small_write and big_write entries are followed by the "external" bitmap
+    // its size is dynamic and included in journal entry's <size> field
+    uint8_t bitmap[];
 };
 
 struct __attribute__((__packed__)) journal_entry_stable
@@ -124,7 +139,8 @@ inline uint32_t je_crc32(journal_entry *je)
 struct journal_sector_info_t
 {
     uint64_t offset;
-    uint64_t usage_count;
+    uint64_t flush_count;
+    bool written;
     bool dirty;
 };
 
@@ -135,16 +151,21 @@ struct journal_t
     bool inmemory = false;
     void *buffer = NULL;
 
-    uint64_t block_size = 512;
+    uint64_t block_size;
     uint64_t offset, len;
+    // Next free block offset
     uint64_t next_free = 0;
+    // First occupied block offset
     uint64_t used_start = 0;
+    // End of the last block not used for writing anymore
+    uint64_t dirty_start = 0;
     uint32_t crc32_last = 0;
 
     // Current sector(s) used for writing
     void *sector_buf = NULL;
     journal_sector_info_t *sector_info = NULL;
     uint64_t sector_count;
+    bool no_same_sector_overwrites = false;
     int cur_sector = 0;
     int in_sector_pos = 0;
 
@@ -154,13 +175,19 @@ struct journal_t
 
     ~journal_t();
     bool trim();
+    uint64_t get_trim_pos();
+    inline bool entry_fits(int size)
+    {
+        return !(block_size - in_sector_pos < size ||
+            no_same_sector_overwrites && sector_info[cur_sector].written);
+    }
 };
 
 struct blockstore_journal_check_t
 {
     blockstore_impl_t *bs;
     uint64_t next_pos, next_sector, next_in_pos;
-    int sectors_required;
+    int sectors_to_write, first_sector;
     bool right_dir; // writing to the end or the beginning of the ring buffer
 
     blockstore_journal_check_t(blockstore_impl_t *bs);

src/blockstore_open.cpp

@@ -1,3 +1,7 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+#include <sys/file.h>
 #include "blockstore_impl.h"
 
 static uint32_t is_power_of_two(uint64_t value)
@@ -34,10 +38,23 @@ void blockstore_impl_t::parse_config(blockstore_config_t & config)
     {
         disable_journal_fsync = true;
     }
+    if (config["disable_device_lock"] == "true" || config["disable_device_lock"] == "1" || config["disable_device_lock"] == "yes")
+    {
+        disable_flock = true;
+    }
+    if (config["immediate_commit"] == "all")
+    {
+        immediate_commit = IMMEDIATE_ALL;
+    }
+    else if (config["immediate_commit"] == "small")
+    {
+        immediate_commit = IMMEDIATE_SMALL;
+    }
     metadata_buf_size = strtoull(config["meta_buf_size"].c_str(), NULL, 10);
     cfg_journal_size = strtoull(config["journal_size"].c_str(), NULL, 10);
     data_device = config["data_device"];
     data_offset = strtoull(config["data_offset"].c_str(), NULL, 10);
+    cfg_data_size = strtoull(config["data_size"].c_str(), NULL, 10);
     meta_device = config["meta_device"];
     meta_offset = strtoull(config["meta_offset"].c_str(), NULL, 10);
     block_size = strtoull(config["block_size"].c_str(), NULL, 10);
@@ -45,12 +62,15 @@ void blockstore_impl_t::parse_config(blockstore_config_t & config)
     journal_device = config["journal_device"];
     journal.offset = strtoull(config["journal_offset"].c_str(), NULL, 10);
     journal.sector_count = strtoull(config["journal_sector_buffer_count"].c_str(), NULL, 10);
+    journal.no_same_sector_overwrites = config["journal_no_same_sector_overwrites"] == "true" ||
+        config["journal_no_same_sector_overwrites"] == "1" || config["journal_no_same_sector_overwrites"] == "yes";
     journal.inmemory = config["inmemory_journal"] != "false";
     disk_alignment = strtoull(config["disk_alignment"].c_str(), NULL, 10);
     journal_block_size = strtoull(config["journal_block_size"].c_str(), NULL, 10);
     meta_block_size = strtoull(config["meta_block_size"].c_str(), NULL, 10);
     bitmap_granularity = strtoull(config["bitmap_granularity"].c_str(), NULL, 10);
     flusher_count = strtoull(config["flusher_count"].c_str(), NULL, 10);
+    max_write_iodepth = strtoull(config["max_write_iodepth"].c_str(), NULL, 10);
     // Validate
     if (!block_size)
     {
@@ -64,17 +84,21 @@ void blockstore_impl_t::parse_config(blockstore_config_t & config)
     {
         flusher_count = 32;
     }
+    if (!max_write_iodepth)
+    {
+        max_write_iodepth = 128;
+    }
     if (!disk_alignment)
     {
-        disk_alignment = 512;
+        disk_alignment = 4096;
     }
     else if (disk_alignment % MEM_ALIGNMENT)
     {
-        throw std::runtime_error("disk_alingment must be a multiple of "+std::to_string(MEM_ALIGNMENT));
+        throw std::runtime_error("disk_alignment must be a multiple of "+std::to_string(MEM_ALIGNMENT));
     }
     if (!journal_block_size)
     {
-        journal_block_size = 512;
+        journal_block_size = 4096;
     }
     else if (journal_block_size % MEM_ALIGNMENT)
     {
@@ -82,7 +106,7 @@ void blockstore_impl_t::parse_config(blockstore_config_t & config)
     }
     if (!meta_block_size)
     {
-        meta_block_size = 512;
+        meta_block_size = 4096;
     }
     else if (meta_block_size % MEM_ALIGNMENT)
     {
@@ -94,7 +118,7 @@ void blockstore_impl_t::parse_config(blockstore_config_t & config)
     }
     if (!bitmap_granularity)
     {
-        bitmap_granularity = 4096;
+        bitmap_granularity = DEFAULT_BITMAP_GRANULARITY;
     }
     else if (bitmap_granularity % disk_alignment)
     {
@@ -128,9 +152,25 @@ void blockstore_impl_t::parse_config(blockstore_config_t & config)
     {
         metadata_buf_size = 4*1024*1024;
     }
+    if (meta_device == "")
+    {
+        disable_meta_fsync = disable_data_fsync;
+    }
+    if (journal_device == "")
+    {
+        disable_journal_fsync = disable_meta_fsync;
+    }
+    if (immediate_commit != IMMEDIATE_NONE && !disable_journal_fsync)
+    {
+        throw std::runtime_error("immediate_commit requires disable_journal_fsync");
+    }
+    if (immediate_commit == IMMEDIATE_ALL && !disable_data_fsync)
+    {
+        throw std::runtime_error("immediate_commit=all requires disable_journal_fsync and disable_data_fsync");
+    }
     // init some fields
     clean_entry_bitmap_size = block_size / bitmap_granularity / 8;
-    clean_entry_size = sizeof(clean_disk_entry) + clean_entry_bitmap_size;
+    clean_entry_size = sizeof(clean_disk_entry) + 2*clean_entry_bitmap_size;
     journal.block_size = journal_block_size;
     journal.next_free = journal_block_size;
     journal.used_start = journal_block_size;
@@ -151,6 +191,15 @@ void blockstore_impl_t::calc_lengths()
         data_len = data_len < journal.offset-data_offset
             ? data_len : journal.offset-data_offset;
     }
+    if (cfg_data_size != 0)
+    {
+        if (data_len < cfg_data_size)
+        {
+            throw std::runtime_error("Data area ("+std::to_string(data_len)+
+                " bytes) is less than configured size ("+std::to_string(cfg_data_size)+" bytes)");
+        }
+        data_len = cfg_data_size;
+    }
     // meta
     meta_area = (meta_fd == data_fd ? data_size : meta_size) - meta_offset;
     if (meta_fd == data_fd && meta_offset <= data_offset)
@@ -188,7 +237,7 @@ void blockstore_impl_t::calc_lengths()
     }
     else if (clean_entry_bitmap_size)
     {
-        clean_bitmap = (uint8_t*)malloc(block_count * clean_entry_bitmap_size);
+        clean_bitmap = (uint8_t*)malloc(block_count * 2*clean_entry_bitmap_size);
         if (!clean_bitmap)
             throw std::runtime_error("Failed to allocate memory for the metadata sparse write bitmap");
     }
@@ -252,6 +301,10 @@ void blockstore_impl_t::open_data()
     {
         throw std::runtime_error("data_offset exceeds device size = "+std::to_string(data_size));
     }
+    if (!disable_flock && flock(data_fd, LOCK_EX|LOCK_NB) != 0)
+    {
+        throw std::runtime_error(std::string("Failed to lock data device: ") + strerror(errno));
+    }
 }
 
 void blockstore_impl_t::open_meta()
@@ -269,11 +322,14 @@ void blockstore_impl_t::open_meta()
         {
             throw std::runtime_error("meta_offset exceeds device size = "+std::to_string(meta_size));
         }
+        if (!disable_flock && flock(meta_fd, LOCK_EX|LOCK_NB) != 0)
+        {
+            throw std::runtime_error(std::string("Failed to lock metadata device: ") + strerror(errno));
+        }
     }
     else
     {
         meta_fd = data_fd;
-        disable_meta_fsync = disable_data_fsync;
         meta_size = 0;
         if (meta_offset >= data_size)
         {
@@ -291,12 +347,15 @@ void blockstore_impl_t::open_journal()
         {
             throw std::runtime_error("Failed to open journal device");
         }
-        check_size(journal.fd, &journal.device_size, "metadata device");
+        check_size(journal.fd, &journal.device_size, "journal device");
+        if (!disable_flock && flock(journal.fd, LOCK_EX|LOCK_NB) != 0)
+        {
+            throw std::runtime_error(std::string("Failed to lock journal device: ") + strerror(errno));
+        }
     }
     else
     {
         journal.fd = meta_fd;
-        disable_journal_fsync = disable_meta_fsync;
         journal.device_size = 0;
         if (journal.offset >= data_size)
         {

src/blockstore_read.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
 #include "blockstore_impl.h"
 
 int blockstore_impl_t::fulfill_read_push(blockstore_op_t *op, void *buf, uint64_t offset, uint64_t len,
@@ -8,12 +11,10 @@ int blockstore_impl_t::fulfill_read_push(blockstore_op_t *op, void *buf, uint64_
         // Zero-length version - skip
         return 1;
     }
-    if (IS_IN_FLIGHT(item_state))
+    else if (IS_IN_FLIGHT(item_state))
     {
-        // Pause until it's written somewhere
-        PRIV(op)->wait_for = WAIT_IN_FLIGHT;
-        PRIV(op)->wait_detail = item_version;
-        return 0;
+        // Write not finished yet - skip
+        return 1;
     }
     else if (IS_DELETE(item_state))
     {
@@ -39,6 +40,7 @@ int blockstore_impl_t::fulfill_read_push(blockstore_op_t *op, void *buf, uint64_
     return 1;
 }
 
+// FIXME I've seen a bug here so I want some tests
 int blockstore_impl_t::fulfill_read(blockstore_op_t *read_op, uint64_t &fulfilled, uint32_t item_start, uint32_t item_end,
     uint32_t item_state, uint64_t item_version, uint64_t item_location)
 {
@@ -51,8 +53,20 @@ int blockstore_impl_t::fulfill_read(blockstore_op_t *read_op, uint64_t &fulfille
         while (1)
         {
             for (; it != PRIV(read_op)->read_vec.end(); it++)
+            {
                 if (it->offset >= cur_start)
+                {
                     break;
+                }
+                else if (it->offset + it->len > cur_start)
+                {
+                    cur_start = it->offset + it->len;
+                    if (cur_start >= item_end)
+                    {
+                        goto endwhile;
+                    }
+                }
+            }
             if (it == PRIV(read_op)->read_vec.end() || it->offset > cur_start)
             {
                 fulfill_read_t el = {
@@ -71,12 +85,30 @@ int blockstore_impl_t::fulfill_read(blockstore_op_t *read_op, uint64_t &fulfille
             }
             cur_start = it->offset + it->len;
             if (it == PRIV(read_op)->read_vec.end() || cur_start >= item_end)
+            {
                 break;
+            }
         }
     }
+endwhile:
     return 1;
 }
 
+uint8_t* blockstore_impl_t::get_clean_entry_bitmap(uint64_t block_loc, int offset)
+{
+    uint8_t *clean_entry_bitmap;
+    uint64_t meta_loc = block_loc >> block_order;
+    if (inmemory_meta)
+    {
+        uint64_t sector = (meta_loc / (meta_block_size / clean_entry_size)) * meta_block_size;
+        uint64_t pos = (meta_loc % (meta_block_size / clean_entry_size));
+        clean_entry_bitmap = (uint8_t*)(metadata_buffer + sector + pos*clean_entry_size + sizeof(clean_disk_entry) + offset);
+    }
+    else
+        clean_entry_bitmap = (uint8_t*)(clean_bitmap + meta_loc*2*clean_entry_bitmap_size + offset);
+    return clean_entry_bitmap;
+}
+
 int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
 {
     auto clean_it = clean_db.find(read_op->oid);
@@ -95,7 +127,7 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
         read_op->version = 0;
         read_op->retval = read_op->len;
         FINISH_OP(read_op);
-        return 1;
+        return 2;
     }
     uint64_t fulfilled = 0;
     PRIV(read_op)->pending_ops = 0;
@@ -117,6 +149,11 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
                 if (!result_version)
                 {
                     result_version = dirty_it->first.version;
+                    if (read_op->bitmap)
+                    {
+                        void *bmp_ptr = (clean_entry_bitmap_size > sizeof(void*) ? dirty_it->second.bitmap : &dirty_it->second.bitmap);
+                        memcpy(read_op->bitmap, bmp_ptr, clean_entry_bitmap_size);
+                    }
                 }
                 if (!fulfill_read(read_op, fulfilled, dirty.offset, dirty.offset + dirty.len,
                     dirty.state, dirty_it->first.version, dirty.location + (IS_JOURNAL(dirty.state) ? 0 : dirty.offset)))
@@ -133,63 +170,61 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
             dirty_it--;
         }
     }
-    if (clean_it != clean_db.end() && fulfilled < read_op->len)
+    if (clean_it != clean_db.end())
     {
         if (!result_version)
         {
             result_version = clean_it->second.version;
-        }
-        if (!clean_entry_bitmap_size)
-        {
-            if (!fulfill_read(read_op, fulfilled, 0, block_size, ST_CURRENT, 0, clean_it->second.location))
+            if (read_op->bitmap)
             {
-                // need to wait. undo added requests, don't dequeue op
-                PRIV(read_op)->read_vec.clear();
-                return 0;
+                void *bmp_ptr = get_clean_entry_bitmap(clean_it->second.location, clean_entry_bitmap_size);
+                memcpy(read_op->bitmap, bmp_ptr, clean_entry_bitmap_size);
             }
         }
-        else
+        if (fulfilled < read_op->len)
         {
-            uint64_t meta_loc = clean_it->second.location >> block_order;
-            uint8_t *clean_entry_bitmap;
-            if (inmemory_meta)
+            if (!clean_entry_bitmap_size)
             {
-                uint64_t sector = (meta_loc / (meta_block_size / clean_entry_size)) * meta_block_size;
-                uint64_t pos = (meta_loc % (meta_block_size / clean_entry_size));
-                clean_entry_bitmap = (uint8_t*)(metadata_buffer + sector + pos*clean_entry_size + sizeof(clean_disk_entry));
+                if (!fulfill_read(read_op, fulfilled, 0, block_size, (BS_ST_BIG_WRITE | BS_ST_STABLE), 0, clean_it->second.location))
+                {
+                    // need to wait. undo added requests, don't dequeue op
+                    PRIV(read_op)->read_vec.clear();
+                    return 0;
+                }
             }
             else
             {
-                clean_entry_bitmap = (uint8_t*)(clean_bitmap + meta_loc*clean_entry_bitmap_size);
-            }
-            uint64_t bmp_start = 0, bmp_end = 0, bmp_size = block_size/bitmap_granularity;
-            while (bmp_start < bmp_size)
-            {
-                while (!(clean_entry_bitmap[bmp_end >> 3] & (1 << (bmp_end & 0x7))) && bmp_end < bmp_size)
+                uint8_t *clean_entry_bitmap = get_clean_entry_bitmap(clean_it->second.location, 0);
+                uint64_t bmp_start = 0, bmp_end = 0, bmp_size = block_size/bitmap_granularity;
+                while (bmp_start < bmp_size)
                 {
-                    bmp_end++;
-                }
-                if (bmp_end > bmp_start)
-                {
-                    // fill with zeroes
-                    fulfill_read(read_op, fulfilled, bmp_start * bitmap_granularity,
-                        bmp_end * bitmap_granularity, ST_DEL_STABLE, 0, 0);
-                }
-                bmp_start = bmp_end;
-                while (clean_entry_bitmap[bmp_end >> 3] & (1 << (bmp_end & 0x7)) && bmp_end < bmp_size)
-                {
-                    bmp_end++;
-                }
-                if (bmp_end > bmp_start)
-                {
-                    if (!fulfill_read(read_op, fulfilled, bmp_start * bitmap_granularity,
-                        bmp_end * bitmap_granularity, ST_CURRENT, 0, clean_it->second.location + bmp_start * bitmap_granularity))
+                    while (!(clean_entry_bitmap[bmp_end >> 3] & (1 << (bmp_end & 0x7))) && bmp_end < bmp_size)
                     {
-                        // need to wait. undo added requests, don't dequeue op
-                        PRIV(read_op)->read_vec.clear();
-                        return 0;
+                        bmp_end++;
+                    }
+                    if (bmp_end > bmp_start)
+                    {
+                        // fill with zeroes
+                        assert(fulfill_read(read_op, fulfilled, bmp_start * bitmap_granularity,
+                            bmp_end * bitmap_granularity, (BS_ST_DELETE | BS_ST_STABLE), 0, 0));
                     }
                     bmp_start = bmp_end;
+                    while (clean_entry_bitmap[bmp_end >> 3] & (1 << (bmp_end & 0x7)) && bmp_end < bmp_size)
+                    {
+                        bmp_end++;
+                    }
+                    if (bmp_end > bmp_start)
+                    {
+                        if (!fulfill_read(read_op, fulfilled, bmp_start * bitmap_granularity,
+                            bmp_end * bitmap_granularity, (BS_ST_BIG_WRITE | BS_ST_STABLE), 0,
+                            clean_it->second.location + bmp_start * bitmap_granularity))
+                        {
+                            // need to wait. undo added requests, don't dequeue op
+                            PRIV(read_op)->read_vec.clear();
+                            return 0;
+                        }
+                        bmp_start = bmp_end;
+                    }
                 }
             }
         }
@@ -197,7 +232,7 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
     else if (fulfilled < read_op->len)
     {
         // fill remaining parts with zeroes
-        fulfill_read(read_op, fulfilled, 0, block_size, ST_DEL_STABLE, 0, 0);
+        assert(fulfill_read(read_op, fulfilled, 0, block_size, (BS_ST_DELETE | BS_ST_STABLE), 0, 0));
     }
     assert(fulfilled == read_op->len);
     read_op->version = result_version;
@@ -211,10 +246,10 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
         }
         read_op->retval = read_op->len;
         FINISH_OP(read_op);
-        return 1;
+        return 2;
     }
     read_op->retval = 0;
-    return 1;
+    return 2;
 }
 
 void blockstore_impl_t::handle_read_event(ring_data_t *data, blockstore_op_t *op)

src/blockstore_rollback.cpp

@@ -0,0 +1,281 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+#include "blockstore_impl.h"
+
+int blockstore_impl_t::dequeue_rollback(blockstore_op_t *op)
+{
+    if (PRIV(op)->op_state)
+    {
+        return continue_rollback(op);
+    }
+    obj_ver_id *v, *nv;
+    int i, todo = op->len;
+    for (i = 0, v = (obj_ver_id*)op->buf, nv = (obj_ver_id*)op->buf; i < op->len; i++, v++, nv++)
+    {
+        if (nv != v)
+        {
+            *nv = *v;
+        }
+        // Check that there are some versions greater than v->version (which may be zero),
+        // check that they're unstable, synced, and not currently written to
+        auto dirty_it = dirty_db.lower_bound((obj_ver_id){
+            .oid = v->oid,
+            .version = UINT64_MAX,
+        });
+        if (dirty_it == dirty_db.begin())
+        {
+skip_ov:
+            // Already rolled back, skip this object version
+            todo--;
+            nv--;
+            continue;
+        }
+        else
+        {
+            dirty_it--;
+            if (dirty_it->first.oid != v->oid || dirty_it->first.version < v->version)
+            {
+                goto skip_ov;
+            }
+            while (dirty_it->first.oid == v->oid && dirty_it->first.version > v->version)
+            {
+                if (IS_IN_FLIGHT(dirty_it->second.state))
+                {
+                    // Object write is still in progress. Wait until the write request completes
+                    return 0;
+                }
+                else if (!IS_SYNCED(dirty_it->second.state) ||
+                    IS_STABLE(dirty_it->second.state))
+                {
+                    op->retval = -EBUSY;
+                    FINISH_OP(op);
+                    return 2;
+                }
+                if (dirty_it == dirty_db.begin())
+                {
+                    break;
+                }
+                dirty_it--;
+            }
+        }
+    }
+    op->len = todo;
+    if (!todo)
+    {
+        // Already rolled back
+        op->retval = 0;
+        FINISH_OP(op);
+        return 2;
+    }
+    // Check journal space
+    blockstore_journal_check_t space_check(this);
+    if (!space_check.check_available(op, todo, sizeof(journal_entry_rollback), 0))
+    {
+        return 0;
+    }
+    // There is sufficient space. Get SQEs
+    struct io_uring_sqe *sqe[space_check.sectors_to_write];
+    for (i = 0; i < space_check.sectors_to_write; i++)
+    {
+        BS_SUBMIT_GET_SQE_DECL(sqe[i]);
+    }
+    // Prepare and submit journal entries
+    auto cb = [this, op](ring_data_t *data) { handle_rollback_event(data, op); };
+    int s = 0, cur_sector = -1;
+    for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
+    {
+        if (!journal.entry_fits(sizeof(journal_entry_rollback)) &&
+            journal.sector_info[journal.cur_sector].dirty)
+        {
+            if (cur_sector == -1)
+                PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
+            prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], cb);
+            cur_sector = journal.cur_sector;
+        }
+        journal_entry_rollback *je = (journal_entry_rollback*)
+            prefill_single_journal_entry(journal, JE_ROLLBACK, sizeof(journal_entry_rollback));
+        je->oid = v->oid;
+        je->version = v->version;
+        je->crc32 = je_crc32((journal_entry*)je);
+        journal.crc32_last = je->crc32;
+    }
+    prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], cb);
+    assert(s == space_check.sectors_to_write);
+    if (cur_sector == -1)
+        PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
+    PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
+    PRIV(op)->pending_ops = s;
+    PRIV(op)->op_state = 1;
+    return 1;
+}
+
+int blockstore_impl_t::continue_rollback(blockstore_op_t *op)
+{
+    if (PRIV(op)->op_state == 2)
+        goto resume_2;
+    else if (PRIV(op)->op_state == 3)
+        goto resume_3;
+    else if (PRIV(op)->op_state == 5)
+        goto resume_5;
+    else
+        return 1;
+resume_2:
+    // Release used journal sectors
+    release_journal_sectors(op);
+resume_3:
+    if (!disable_journal_fsync)
+    {
+        io_uring_sqe *sqe;
+        BS_SUBMIT_GET_SQE_DECL(sqe);
+        ring_data_t *data = ((ring_data_t*)sqe->user_data);
+        my_uring_prep_fsync(sqe, journal.fd, IORING_FSYNC_DATASYNC);
+        data->iov = { 0 };
+        data->callback = [this, op](ring_data_t *data) { handle_rollback_event(data, op); };
+        PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 0;
+        PRIV(op)->pending_ops = 1;
+        PRIV(op)->op_state = 4;
+        return 1;
+    }
+resume_5:
+    obj_ver_id* v;
+    int i;
+    for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
+    {
+        mark_rolled_back(*v);
+    }
+    flusher->mark_trim_possible();
+    // Acknowledge op
+    op->retval = 0;
+    FINISH_OP(op);
+    return 2;
+}
+
+void blockstore_impl_t::mark_rolled_back(const obj_ver_id & ov)
+{
+    auto it = dirty_db.lower_bound((obj_ver_id){
+        .oid = ov.oid,
+        .version = UINT64_MAX,
+    });
+    if (it != dirty_db.begin())
+    {
+        uint64_t max_unstable = 0;
+        auto rm_start = it;
+        auto rm_end = it;
+        it--;
+        while (1)
+        {
+            if (it->first.oid != ov.oid)
+                break;
+            else if (it->first.version <= ov.version)
+            {
+                if (!IS_STABLE(it->second.state))
+                    max_unstable = it->first.version;
+                break;
+            }
+            else if (IS_IN_FLIGHT(it->second.state) || IS_STABLE(it->second.state))
+                break;
+            // Remove entry
+            rm_start = it;
+            if (it == dirty_db.begin())
+                break;
+            it--;
+        }
+        if (rm_start != rm_end)
+        {
+            erase_dirty(rm_start, rm_end, UINT64_MAX);
+            auto unstab_it = unstable_writes.find(ov.oid);
+            if (unstab_it != unstable_writes.end())
+            {
+                if (max_unstable == 0)
+                    unstable_writes.erase(unstab_it);
+                else
+                    unstab_it->second = max_unstable;
+            }
+        }
+    }
+}
+
+void blockstore_impl_t::handle_rollback_event(ring_data_t *data, blockstore_op_t *op)
+{
+    live = true;
+    if (data->res != data->iov.iov_len)
+    {
+        throw std::runtime_error(
+            "write operation failed ("+std::to_string(data->res)+" != "+std::to_string(data->iov.iov_len)+
+            "). in-memory state is corrupted. AAAAAAAaaaaaaaaa!!!111"
+        );
+    }
+    PRIV(op)->pending_ops--;
+    if (PRIV(op)->pending_ops == 0)
+    {
+        PRIV(op)->op_state++;
+        ringloop->wakeup();
+    }
+}
+
+void blockstore_impl_t::erase_dirty(blockstore_dirty_db_t::iterator dirty_start, blockstore_dirty_db_t::iterator dirty_end, uint64_t clean_loc)
+{
+    if (dirty_end == dirty_start)
+    {
+        return;
+    }
+    auto dirty_it = dirty_end;
+    dirty_it--;
+    if (IS_DELETE(dirty_it->second.state))
+    {
+        object_id oid = dirty_it->first.oid;
+#ifdef BLOCKSTORE_DEBUG
+        printf("Unblock writes-after-delete %lx:%lx v%lx\n", oid.inode, oid.stripe, dirty_it->first.version);
+#endif
+        dirty_it = dirty_end;
+        // Unblock operations blocked by delete flushing
+        uint32_t next_state = BS_ST_IN_FLIGHT;
+        while (dirty_it != dirty_db.end() && dirty_it->first.oid == oid)
+        {
+            if ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_WAIT_DEL)
+            {
+                dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | next_state;
+                if (IS_BIG_WRITE(dirty_it->second.state))
+                {
+                    next_state = BS_ST_WAIT_BIG;
+                }
+            }
+            dirty_it++;
+        }
+        dirty_it = dirty_end;
+        dirty_it--;
+    }
+    while (1)
+    {
+        if (IS_BIG_WRITE(dirty_it->second.state) && dirty_it->second.location != clean_loc)
+        {
+#ifdef BLOCKSTORE_DEBUG
+            printf("Free block %lu\n", dirty_it->second.location >> block_order);
+#endif
+            data_alloc->set(dirty_it->second.location >> block_order, false);
+        }
+        int used = --journal.used_sectors[dirty_it->second.journal_sector];
+#ifdef BLOCKSTORE_DEBUG
+        printf(
+            "remove usage of journal offset %08lx by %lx:%lx v%lu (%d refs)\n", dirty_it->second.journal_sector,
+            dirty_it->first.oid.inode, dirty_it->first.oid.stripe, dirty_it->first.version, used
+        );
+#endif
+        if (used == 0)
+        {
+            journal.used_sectors.erase(dirty_it->second.journal_sector);
+        }
+        if (clean_entry_bitmap_size > sizeof(void*))
+        {
+            free(dirty_it->second.bitmap);
+            dirty_it->second.bitmap = NULL;
+        }
+        if (dirty_it == dirty_start)
+        {
+            break;
+        }
+        dirty_it--;
+    }
+    dirty_db.erase(dirty_start, dirty_end);
+}

src/blockstore_stable.cpp

@@ -1,3 +1,6 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
 #include "blockstore_impl.h"
 
 // Stabilize small write:
@@ -40,6 +43,10 @@
 
 int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
 {
+    if (PRIV(op)->op_state)
+    {
+        return continue_stable(op);
+    }
     obj_ver_id* v;
     int i, todo = 0;
     for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
@@ -51,21 +58,26 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
             if (clean_it == clean_db.end() || clean_it->second.version < v->version)
             {
                 // No such object version
-                op->retval = -EINVAL;
+                op->retval = -ENOENT;
                 FINISH_OP(op);
-                return 1;
+                return 2;
             }
             else
             {
                 // Already stable
             }
         }
-        else if (IS_UNSYNCED(dirty_it->second.state))
+        else if (IS_IN_FLIGHT(dirty_it->second.state))
+        {
+            // Object write is still in progress. Wait until the write request completes
+            return 0;
+        }
+        else if (!IS_SYNCED(dirty_it->second.state))
         {
             // Object not synced yet. Caller must sync it first
-            op->retval = EAGAIN;
+            op->retval = -EBUSY;
             FINISH_OP(op);
-            return 1;
+            return 2;
         }
         else if (!IS_STABLE(dirty_it->second.state))
         {
@@ -77,7 +89,7 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
         // Already stable
         op->retval = 0;
         FINISH_OP(op);
-        return 1;
+        return 2;
     }
     // Check journal space
     blockstore_journal_check_t space_check(this);
@@ -86,50 +98,128 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
         return 0;
     }
     // There is sufficient space. Get SQEs
-    struct io_uring_sqe *sqe[space_check.sectors_required];
-    for (i = 0; i < space_check.sectors_required; i++)
+    struct io_uring_sqe *sqe[space_check.sectors_to_write];
+    for (i = 0; i < space_check.sectors_to_write; i++)
     {
         BS_SUBMIT_GET_SQE_DECL(sqe[i]);
     }
     // Prepare and submit journal entries
     auto cb = [this, op](ring_data_t *data) { handle_stable_event(data, op); };
     int s = 0, cur_sector = -1;
-    if ((journal_block_size - journal.in_sector_pos) < sizeof(journal_entry_stable) &&
-        journal.sector_info[journal.cur_sector].dirty)
-    {
-        if (cur_sector == -1)
-            PRIV(op)->min_used_journal_sector = 1 + journal.cur_sector;
-        cur_sector = journal.cur_sector;
-        prepare_journal_sector_write(journal, cur_sector, sqe[s++], cb);
-    }
     for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
     {
-        auto unstab_it = unstable_writes.find(v->oid);
-        if (unstab_it != unstable_writes.end() &&
-            unstab_it->second <= v->version)
+        // FIXME: Only stabilize versions that aren't stable yet
+        if (!journal.entry_fits(sizeof(journal_entry_stable)) &&
+            journal.sector_info[journal.cur_sector].dirty)
         {
-            unstable_writes.erase(unstab_it);
+            if (cur_sector == -1)
+                PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
+            prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], cb);
+            cur_sector = journal.cur_sector;
         }
         journal_entry_stable *je = (journal_entry_stable*)
             prefill_single_journal_entry(journal, JE_STABLE, sizeof(journal_entry_stable));
-        journal.sector_info[journal.cur_sector].dirty = false;
         je->oid = v->oid;
         je->version = v->version;
         je->crc32 = je_crc32((journal_entry*)je);
         journal.crc32_last = je->crc32;
-        if (cur_sector != journal.cur_sector)
-        {
-            if (cur_sector == -1)
-                PRIV(op)->min_used_journal_sector = 1 + journal.cur_sector;
-            cur_sector = journal.cur_sector;
-            prepare_journal_sector_write(journal, cur_sector, sqe[s++], cb);
-        }
     }
-    PRIV(op)->max_used_journal_sector = 1 + journal.cur_sector;
+    prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], cb);
+    assert(s == space_check.sectors_to_write);
+    if (cur_sector == -1)
+        PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
+    PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
     PRIV(op)->pending_ops = s;
+    PRIV(op)->op_state = 1;
     return 1;
 }
 
+int blockstore_impl_t::continue_stable(blockstore_op_t *op)
+{
+    if (PRIV(op)->op_state == 2)
+        goto resume_2;
+    else if (PRIV(op)->op_state == 3)
+        goto resume_3;
+    else if (PRIV(op)->op_state == 5)
+        goto resume_5;
+    else
+        return 1;
+resume_2:
+    // Release used journal sectors
+    release_journal_sectors(op);
+resume_3:
+    if (!disable_journal_fsync)
+    {
+        io_uring_sqe *sqe;
+        BS_SUBMIT_GET_SQE_DECL(sqe);
+        ring_data_t *data = ((ring_data_t*)sqe->user_data);
+        my_uring_prep_fsync(sqe, journal.fd, IORING_FSYNC_DATASYNC);
+        data->iov = { 0 };
+        data->callback = [this, op](ring_data_t *data) { handle_stable_event(data, op); };
+        PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 0;
+        PRIV(op)->pending_ops = 1;
+        PRIV(op)->op_state = 4;
+        return 1;
+    }
+resume_5:
+    // Mark dirty_db entries as stable, acknowledge op completion
+    obj_ver_id* v;
+    int i;
+    for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
+    {
+        // Mark all dirty_db entries up to op->version as stable
+        mark_stable(*v);
+    }
+    // Acknowledge op
+    op->retval = 0;
+    FINISH_OP(op);
+    return 2;
+}
+
+void blockstore_impl_t::mark_stable(const obj_ver_id & v)
+{
+    auto dirty_it = dirty_db.find(v);
+    if (dirty_it != dirty_db.end())
+    {
+        while (1)
+        {
+            if ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_SYNCED)
+            {
+                dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | BS_ST_STABLE;
+                // Allocations and deletions are counted when they're stabilized
+                if (IS_BIG_WRITE(dirty_it->second.state))
+                {
+                    inode_space_stats[dirty_it->first.oid.inode] += block_size;
+                }
+                else if (IS_DELETE(dirty_it->second.state))
+                {
+                    inode_space_stats[dirty_it->first.oid.inode] -= block_size;
+                }
+            }
+            else if (IS_STABLE(dirty_it->second.state))
+            {
+                break;
+            }
+            if (dirty_it == dirty_db.begin())
+            {
+                break;
+            }
+            dirty_it--;
+            if (dirty_it->first.oid != v.oid)
+            {
+                break;
+            }
+        }
+        flusher->enqueue_flush(v);
+    }
+    auto unstab_it = unstable_writes.find(v.oid);
+    if (unstab_it != unstable_writes.end() &&
+        unstab_it->second <= v.version)
+    {
+        unstable_writes.erase(unstab_it);
+    }
+}
+
 void blockstore_impl_t::handle_stable_event(ring_data_t *data, blockstore_op_t *op)
 {
     live = true;
@@ -143,53 +233,7 @@ void blockstore_impl_t::handle_stable_event(ring_data_t *data, blockstore_op_t *
     PRIV(op)->pending_ops--;
     if (PRIV(op)->pending_ops == 0)
     {
-        // Release used journal sectors
-        release_journal_sectors(op);
-        // Mark dirty_db entries as stable, acknowledge op completion
-        obj_ver_id* v;
-        int i;
-        for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
-        {
-            // Mark all dirty_db entries up to op->version as stable
-            auto dirty_it = dirty_db.find(*v);
-            if (dirty_it != dirty_db.end())
-            {
-                while (1)
-                {
-                    if (dirty_it->second.state == ST_J_SYNCED)
-                    {
-                        dirty_it->second.state = ST_J_STABLE;
-                    }
-                    else if (dirty_it->second.state == ST_D_META_SYNCED)
-                    {
-                        dirty_it->second.state = ST_D_STABLE;
-                    }
-                    else if (dirty_it->second.state == ST_DEL_SYNCED)
-                    {
-                        dirty_it->second.state = ST_DEL_STABLE;
-                    }
-                    else if (IS_STABLE(dirty_it->second.state))
-                    {
-                        break;
-                    }
-                    if (dirty_it == dirty_db.begin())
-                    {
-                        break;
-                    }
-                    dirty_it--;
-                    if (dirty_it->first.oid != v->oid)
-                    {
-                        break;
-                    }
-                }
-#ifdef BLOCKSTORE_DEBUG
-                printf("enqueue_flush %lu:%lu v%lu\n", v->oid.inode, v->oid.stripe, v->version);
-#endif
-                flusher->enqueue_flush(*v);
-            }
-        }
-        // Acknowledge op
-        op->retval = 0;
-        FINISH_OP(op);
+        PRIV(op)->op_state++;
+        ringloop->wakeup();
     }
 }

src/blockstore_sync.cpp

@@ -0,0 +1,247 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+#include "blockstore_impl.h"
+
+#define SYNC_HAS_SMALL 1
+#define SYNC_HAS_BIG 2
+#define SYNC_DATA_SYNC_SENT 3
+#define SYNC_DATA_SYNC_DONE 4
+#define SYNC_JOURNAL_WRITE_SENT 5
+#define SYNC_JOURNAL_WRITE_DONE 6
+#define SYNC_JOURNAL_SYNC_SENT 7
+#define SYNC_DONE 8
+
+int blockstore_impl_t::continue_sync(blockstore_op_t *op, bool queue_has_in_progress_sync)
+{
+    if (immediate_commit == IMMEDIATE_ALL)
+    {
+        // We can return immediately because sync is only dequeued after all previous writes
+        op->retval = 0;
+        FINISH_OP(op);
+        return 2;
+    }
+    if (PRIV(op)->op_state == 0)
+    {
+        stop_sync_submitted = false;
+        PRIV(op)->sync_big_writes.swap(unsynced_big_writes);
+        PRIV(op)->sync_small_writes.swap(unsynced_small_writes);
+        PRIV(op)->sync_small_checked = 0;
+        PRIV(op)->sync_big_checked = 0;
+        unsynced_big_writes.clear();
+        unsynced_small_writes.clear();
+        if (PRIV(op)->sync_big_writes.size() > 0)
+            PRIV(op)->op_state = SYNC_HAS_BIG;
+        else if (PRIV(op)->sync_small_writes.size() > 0)
+            PRIV(op)->op_state = SYNC_HAS_SMALL;
+        else
+            PRIV(op)->op_state = SYNC_DONE;
+    }
+    if (PRIV(op)->op_state == SYNC_HAS_SMALL)
+    {
+        // No big writes, just fsync the journal
+        if (journal.sector_info[journal.cur_sector].dirty)
+        {
+            // Write out the last journal sector if it happens to be dirty
+            BS_SUBMIT_GET_ONLY_SQE(sqe);
+            prepare_journal_sector_write(journal, journal.cur_sector, sqe, [this, op](ring_data_t *data) { handle_sync_event(data, op); });
+            PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
+            PRIV(op)->pending_ops = 1;
+            PRIV(op)->op_state = SYNC_JOURNAL_WRITE_SENT;
+            return 1;
+        }
+        else
+        {
+            PRIV(op)->op_state = SYNC_JOURNAL_WRITE_DONE;
+        }
+    }
+    if (PRIV(op)->op_state == SYNC_HAS_BIG)
+    {
+        // 1st step: fsync data
+        if (!disable_data_fsync)
+        {
+            BS_SUBMIT_GET_SQE(sqe, data);
+            my_uring_prep_fsync(sqe, data_fd, IORING_FSYNC_DATASYNC);
+            data->iov = { 0 };
+            data->callback = [this, op](ring_data_t *data) { handle_sync_event(data, op); };
+            PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 0;
+            PRIV(op)->pending_ops = 1;
+            PRIV(op)->op_state = SYNC_DATA_SYNC_SENT;
+            return 1;
+        }
+        else
+        {
+            PRIV(op)->op_state = SYNC_DATA_SYNC_DONE;
+        }
+    }
+    if (PRIV(op)->op_state == SYNC_DATA_SYNC_DONE)
+    {
+        // 2nd step: Data device is synced, prepare & write journal entries
+        // Check space in the journal and journal memory buffers
+        blockstore_journal_check_t space_check(this);
+        if (!space_check.check_available(op, PRIV(op)->sync_big_writes.size(), sizeof(journal_entry_big_write), JOURNAL_STABILIZE_RESERVATION))
+        {
+            return 0;
+        }
+        // Get SQEs. Don't bother about merging, submit each journal sector as a separate request
+        struct io_uring_sqe *sqe[space_check.sectors_to_write];
+        for (int i = 0; i < space_check.sectors_to_write; i++)
+        {
+            BS_SUBMIT_GET_SQE_DECL(sqe[i]);
+        }
+        // Prepare and submit journal entries
+        auto it = PRIV(op)->sync_big_writes.begin();
+        int s = 0, cur_sector = -1;
+        while (it != PRIV(op)->sync_big_writes.end())
+        {
+            if (!journal.entry_fits(sizeof(journal_entry_big_write)) &&
+                journal.sector_info[journal.cur_sector].dirty)
+            {
+                if (cur_sector == -1)
+                    PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
+                prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], [this, op](ring_data_t *data) { handle_sync_event(data, op); });
+                cur_sector = journal.cur_sector;
+            }
+            journal_entry_big_write *je = (journal_entry_big_write*)prefill_single_journal_entry(
+                journal, (dirty_db[*it].state & BS_ST_INSTANT) ? JE_BIG_WRITE_INSTANT : JE_BIG_WRITE,
+                sizeof(journal_entry_big_write)
+            );
+            dirty_db[*it].journal_sector = journal.sector_info[journal.cur_sector].offset;
+            journal.used_sectors[journal.sector_info[journal.cur_sector].offset]++;
+#ifdef BLOCKSTORE_DEBUG
+            printf(
+                "journal offset %08lx is used by %lx:%lx v%lu (%lu refs)\n",
+                dirty_db[*it].journal_sector, it->oid.inode, it->oid.stripe, it->version,
+                journal.used_sectors[journal.sector_info[journal.cur_sector].offset]
+            );
+#endif
+            je->oid = it->oid;
+            je->version = it->version;
+            je->offset = dirty_db[*it].offset;
+            je->len = dirty_db[*it].len;
+            je->location = dirty_db[*it].location;
+            je->crc32 = je_crc32((journal_entry*)je);
+            journal.crc32_last = je->crc32;
+            it++;
+        }
+        prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], [this, op](ring_data_t *data) { handle_sync_event(data, op); });
+        assert(s == space_check.sectors_to_write);
+        if (cur_sector == -1)
+            PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
+        PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
+        PRIV(op)->pending_ops = s;
+        PRIV(op)->op_state = SYNC_JOURNAL_WRITE_SENT;
+        return 1;
+    }
+    if (PRIV(op)->op_state == SYNC_JOURNAL_WRITE_DONE)
+    {
+        if (!disable_journal_fsync)
+        {
+            BS_SUBMIT_GET_SQE(sqe, data);
+            my_uring_prep_fsync(sqe, journal.fd, IORING_FSYNC_DATASYNC);
+            data->iov = { 0 };
+            data->callback = [this, op](ring_data_t *data) { handle_sync_event(data, op); };
+            PRIV(op)->pending_ops = 1;
+            PRIV(op)->op_state = SYNC_JOURNAL_SYNC_SENT;
+            return 1;
+        }
+        else
+        {
+            PRIV(op)->op_state = SYNC_DONE;
+        }
+    }
+    if (PRIV(op)->op_state == SYNC_DONE && !queue_has_in_progress_sync)
+    {
+        ack_sync(op);
+        return 2;
+    }
+    return 1;
+}
+
+void blockstore_impl_t::handle_sync_event(ring_data_t *data, blockstore_op_t *op)
+{
+    live = true;
+    if (data->res != data->iov.iov_len)
+    {
+        throw std::runtime_error(
+            "write operation failed ("+std::to_string(data->res)+" != "+std::to_string(data->iov.iov_len)+
+            "). in-memory state is corrupted. AAAAAAAaaaaaaaaa!!!111"
+        );
+    }
+    PRIV(op)->pending_ops--;
+    if (PRIV(op)->pending_ops == 0)
+    {
+        // Release used journal sectors
+        release_journal_sectors(op);
+        // Handle states
+        if (PRIV(op)->op_state == SYNC_DATA_SYNC_SENT)
+        {
+            PRIV(op)->op_state = SYNC_DATA_SYNC_DONE;
+        }
+        else if (PRIV(op)->op_state == SYNC_JOURNAL_WRITE_SENT)
+        {
+            PRIV(op)->op_state = SYNC_JOURNAL_WRITE_DONE;
+        }
+        else if (PRIV(op)->op_state == SYNC_JOURNAL_SYNC_SENT)
+        {
+            PRIV(op)->op_state = SYNC_DONE;
+        }
+        else
+        {
+            throw std::runtime_error("BUG: unexpected sync op state");
+        }
+        ringloop->wakeup();
+    }
+}
+
+void blockstore_impl_t::ack_sync(blockstore_op_t *op)
+{
+    // Handle states
+    for (auto it = PRIV(op)->sync_big_writes.begin(); it != PRIV(op)->sync_big_writes.end(); it++)
+    {
+#ifdef BLOCKSTORE_DEBUG
+        printf("Ack sync big %lx:%lx v%lu\n", it->oid.inode, it->oid.stripe, it->version);
+#endif
+        auto & unstab = unstable_writes[it->oid];
+        unstab = unstab < it->version ? it->version : unstab;
+        auto dirty_it = dirty_db.find(*it);
+        dirty_it->second.state = ((dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | BS_ST_SYNCED);
+        if (dirty_it->second.state & BS_ST_INSTANT)
+        {
+            mark_stable(dirty_it->first);
+        }
+        dirty_it++;
+        while (dirty_it != dirty_db.end() && dirty_it->first.oid == it->oid)
+        {
+            if ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_WAIT_BIG)
+            {
+                dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | BS_ST_IN_FLIGHT;
+            }
+            dirty_it++;
+        }
+    }
+    for (auto it = PRIV(op)->sync_small_writes.begin(); it != PRIV(op)->sync_small_writes.end(); it++)
+    {
+#ifdef BLOCKSTORE_DEBUG
+        printf("Ack sync small %lx:%lx v%lu\n", it->oid.inode, it->oid.stripe, it->version);
+#endif
+        auto & unstab = unstable_writes[it->oid];
+        unstab = unstab < it->version ? it->version : unstab;
+        if (dirty_db[*it].state == (BS_ST_DELETE | BS_ST_WRITTEN))
+        {
+            dirty_db[*it].state = (BS_ST_DELETE | BS_ST_SYNCED);
+            // Deletions are treated as immediately stable
+            mark_stable(*it);
+        }
+        else /* (BS_ST_INSTANT?) | BS_ST_SMALL_WRITE | BS_ST_WRITTEN */
+        {
+            dirty_db[*it].state = (dirty_db[*it].state & ~BS_ST_WORKFLOW_MASK) | BS_ST_SYNCED;
+            if (dirty_db[*it].state & BS_ST_INSTANT)
+            {
+                mark_stable(*it);
+            }
+        }
+    }
+    op->retval = 0;
+    FINISH_OP(op);
+}

src/blockstore_write.cpp

@@ -0,0 +1,646 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+#include "blockstore_impl.h"
+
+bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
+{
+    // Check or assign version number
+    bool found = false, deleted = false, is_del = (op->opcode == BS_OP_DELETE);
+    bool wait_big = false, wait_del = false;
+    void *bmp = NULL;
+    uint64_t version = 1;
+    if (!is_del && clean_entry_bitmap_size > sizeof(void*))
+    {
+        bmp = calloc_or_die(1, clean_entry_bitmap_size);
+    }
+    if (dirty_db.size() > 0)
+    {
+        auto dirty_it = dirty_db.upper_bound((obj_ver_id){
+            .oid = op->oid,
+            .version = UINT64_MAX,
+        });
+        dirty_it--; // segfaults when dirty_db is empty
+        if (dirty_it != dirty_db.end() && dirty_it->first.oid == op->oid)
+        {
+            found = true;
+            version = dirty_it->first.version + 1;
+            deleted = IS_DELETE(dirty_it->second.state);
+            wait_del = ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_WAIT_DEL);
+            wait_big = (dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_BIG_WRITE
+                ? !IS_SYNCED(dirty_it->second.state)
+                : ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_WAIT_BIG);
+            if (clean_entry_bitmap_size > sizeof(void*))
+                memcpy(bmp, dirty_it->second.bitmap, clean_entry_bitmap_size);
+            else
+                bmp = dirty_it->second.bitmap;
+        }
+    }
+    if (!found)
+    {
+        auto clean_it = clean_db.find(op->oid);
+        if (clean_it != clean_db.end())
+        {
+            version = clean_it->second.version + 1;
+            void *bmp_ptr = get_clean_entry_bitmap(clean_it->second.location, clean_entry_bitmap_size);
+            memcpy((clean_entry_bitmap_size > sizeof(void*) ? bmp : &bmp), bmp_ptr, clean_entry_bitmap_size);
+        }
+        else
+        {
+            deleted = true;
+        }
+    }
+    if (deleted && is_del)
+    {
+        // Already deleted
+        op->retval = 0;
+        return false;
+    }
+    PRIV(op)->real_version = 0;
+    if (op->version == 0)
+    {
+        op->version = version;
+    }
+    else if (op->version < version)
+    {
+        // Implicit operations must be added like that: DEL [FLUSH] BIG [SYNC] SMALL SMALL
+        if (deleted || wait_del)
+        {
+            // It's allowed to write versions with low numbers over deletes
+            // However, we have to flush those deletes first as we use version number for ordering
+#ifdef BLOCKSTORE_DEBUG
+            printf("Write %lx:%lx v%lu over delete (real v%lu) offset=%u len=%u\n", op->oid.inode, op->oid.stripe, version, op->version, op->offset, op->len);
+#endif
+            wait_del = true;
+            PRIV(op)->real_version = op->version;
+            op->version = version;
+            flusher->unshift_flush((obj_ver_id){
+                .oid = op->oid,
+                .version = version-1,
+            }, true);
+        }
+        else
+        {
+            // Invalid version requested
+            op->retval = -EEXIST;
+            if (!is_del && clean_entry_bitmap_size > sizeof(void*))
+            {
+                free(bmp);
+            }
+            return false;
+        }
+    }
+    if (wait_big && !is_del && !deleted && op->len < block_size &&
+        immediate_commit != IMMEDIATE_ALL)
+    {
+        // Issue an additional sync so that the previous big write can reach the journal
+        blockstore_op_t *sync_op = new blockstore_op_t;
+        sync_op->opcode = BS_OP_SYNC;
+        sync_op->callback = [this, op](blockstore_op_t *sync_op)
+        {
+            delete sync_op;
+        };
+        enqueue_op(sync_op);
+    }
+#ifdef BLOCKSTORE_DEBUG
+    if (is_del)
+        printf("Delete %lx:%lx v%lu\n", op->oid.inode, op->oid.stripe, op->version);
+    else if (!wait_del)
+        printf("Write %lx:%lx v%lu offset=%u len=%u\n", op->oid.inode, op->oid.stripe, op->version, op->offset, op->len);
+#endif
+    // FIXME No strict need to add it into dirty_db here, it's just left
+    // from the previous implementation where reads waited for writes
+    uint32_t state;
+    if (is_del)
+        state = BS_ST_DELETE | BS_ST_IN_FLIGHT;
+    else
+    {
+        state = (op->len == block_size || deleted ? BS_ST_BIG_WRITE : BS_ST_SMALL_WRITE);
+        if (wait_del)
+            state |= BS_ST_WAIT_DEL;
+        else if (state == BS_ST_SMALL_WRITE && wait_big)
+            state |= BS_ST_WAIT_BIG;
+        else
+            state |= BS_ST_IN_FLIGHT;
+        if (op->opcode == BS_OP_WRITE_STABLE)
+            state |= BS_ST_INSTANT;
+        if (op->bitmap)
+        {
+            // Only allow to overwrite part of the object bitmap respective to the write's offset/len
+            uint8_t *bmp_ptr = (uint8_t*)(clean_entry_bitmap_size > sizeof(void*) ? bmp : &bmp);
+            uint32_t bit = op->offset/bitmap_granularity;
+            uint32_t bits_left = op->len/bitmap_granularity;
+            while (!(bit % 8) && bits_left > 8)
+            {
+                // Copy bytes
+                bmp_ptr[bit/8] = ((uint8_t*)op->bitmap)[bit/8];
+                bit += 8;
+                bits_left -= 8;
+            }
+            while (bits_left > 0)
+            {
+                // Copy bits
+                bmp_ptr[bit/8] = (bmp_ptr[bit/8] & ~(1 << (bit%8)))
+                    | (((uint8_t*)op->bitmap)[bit/8] & (1 << bit%8));
+                bit++;
+                bits_left--;
+            }
+        }
+    }
+    dirty_db.emplace((obj_ver_id){
+        .oid = op->oid,
+        .version = op->version,
+    }, (dirty_entry){
+        .state = state,
+        .flags = 0,
+        .location = 0,
+        .offset = is_del ? 0 : op->offset,
+        .len = is_del ? 0 : op->len,
+        .journal_sector = 0,
+        .bitmap = bmp,
+    });
+    return true;
+}
+
+void blockstore_impl_t::cancel_all_writes(blockstore_op_t *op, blockstore_dirty_db_t::iterator dirty_it, int retval)
+{
+    while (dirty_it != dirty_db.end() && dirty_it->first.oid == op->oid)
+    {
+        if (clean_entry_bitmap_size > sizeof(void*))
+            free(dirty_it->second.bitmap);
+        dirty_db.erase(dirty_it++);
+    }
+    bool found = false;
+    for (auto other_op: submit_queue)
+    {
+        if (!found && other_op == op)
+            found = true;
+        else if (found && other_op->oid == op->oid &&
+            (other_op->opcode == BS_OP_WRITE || other_op->opcode == BS_OP_WRITE_STABLE))
+        {
+            // Mark operations to cancel them
+            PRIV(other_op)->real_version = UINT64_MAX;
+            other_op->retval = retval;
+        }
+    }
+    op->retval = retval;
+    FINISH_OP(op);
+}
+
+// First step of the write algorithm: dequeue operation and submit initial write(s)
+int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
+{
+    if (PRIV(op)->op_state)
+    {
+        return continue_write(op);
+    }
+    auto dirty_it = dirty_db.find((obj_ver_id){
+        .oid = op->oid,
+        .version = op->version,
+    });
+    assert(dirty_it != dirty_db.end());
+    if ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) < BS_ST_IN_FLIGHT)
+    {
+        // Don't dequeue
+        return 0;
+    }
+    if (PRIV(op)->real_version != 0)
+    {
+        if (PRIV(op)->real_version == UINT64_MAX)
+        {
+            // This is the flag value used to cancel operations
+            FINISH_OP(op);
+            return 2;
+        }
+        // Restore original low version number for unblocked operations
+#ifdef BLOCKSTORE_DEBUG
+        printf("Restoring %lx:%lx version: v%lu -> v%lu\n", op->oid.inode, op->oid.stripe, op->version, PRIV(op)->real_version);
+#endif
+        auto prev_it = dirty_it;
+        prev_it--;
+        if (prev_it->first.oid == op->oid && prev_it->first.version >= PRIV(op)->real_version)
+        {
+            // Original version is still invalid
+            // All subsequent writes to the same object must be canceled too
+            cancel_all_writes(op, dirty_it, -EEXIST);
+            return 2;
+        }
+        op->version = PRIV(op)->real_version;
+        PRIV(op)->real_version = 0;
+        dirty_entry e = dirty_it->second;
+        dirty_db.erase(dirty_it);
+        dirty_it = dirty_db.emplace((obj_ver_id){
+            .oid = op->oid,
+            .version = op->version,
+        }, e).first;
+    }
+    if (write_iodepth >= max_write_iodepth)
+    {
+        return 0;
+    }
+    if ((dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_BIG_WRITE)
+    {
+        blockstore_journal_check_t space_check(this);
+        if (!space_check.check_available(op, unsynced_big_writes.size() + 1, sizeof(journal_entry_big_write), JOURNAL_STABILIZE_RESERVATION))
+        {
+            return 0;
+        }
+        // Big (redirect) write
+        uint64_t loc = data_alloc->find_free();
+        if (loc == UINT64_MAX)
+        {
+            // no space
+            if (flusher->is_active())
+            {
+                // hope that some space will be available after flush
+                PRIV(op)->wait_for = WAIT_FREE;
+                return 0;
+            }
+            cancel_all_writes(op, dirty_it, -ENOSPC);
+            return 2;
+        }
+        write_iodepth++;
+        BS_SUBMIT_GET_SQE(sqe, data);
+        dirty_it->second.location = loc << block_order;
+        dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | BS_ST_SUBMITTED;
+#ifdef BLOCKSTORE_DEBUG
+        printf("Allocate block %lu\n", loc);
+#endif
+        data_alloc->set(loc, true);
+        uint64_t stripe_offset = (op->offset % bitmap_granularity);
+        uint64_t stripe_end = (op->offset + op->len) % bitmap_granularity;
+        // Zero fill up to bitmap_granularity
+        int vcnt = 0;
+        if (stripe_offset)
+        {
+            PRIV(op)->iov_zerofill[vcnt++] = (struct iovec){ zero_object, stripe_offset };
+        }
+        PRIV(op)->iov_zerofill[vcnt++] = (struct iovec){ op->buf, op->len };
+        if (stripe_end)
+        {
+            stripe_end = bitmap_granularity - stripe_end;
+            PRIV(op)->iov_zerofill[vcnt++] = (struct iovec){ zero_object, stripe_end };
+        }
+        data->iov.iov_len = op->len + stripe_offset + stripe_end; // to check it in the callback
+        data->callback = [this, op](ring_data_t *data) { handle_write_event(data, op); };
+        my_uring_prep_writev(
+            sqe, data_fd, PRIV(op)->iov_zerofill, vcnt, data_offset + (loc << block_order) + op->offset - stripe_offset
+        );
+        PRIV(op)->pending_ops = 1;
+        PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 0;
+        if (immediate_commit != IMMEDIATE_ALL)
+        {
+            // Remember big write as unsynced
+            unsynced_big_writes.push_back((obj_ver_id){
+                .oid = op->oid,
+                .version = op->version,
+            });
+            PRIV(op)->op_state = 3;
+        }
+        else
+        {
+            PRIV(op)->op_state = 1;
+        }
+    }
+    else /* if ((dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_SMALL_WRITE) */
+    {
+        // Small (journaled) write
+        // First check if the journal has sufficient space
+        blockstore_journal_check_t space_check(this);
+        if (unsynced_big_writes.size() && !space_check.check_available(op, unsynced_big_writes.size(), sizeof(journal_entry_big_write), 0)
+            || !space_check.check_available(op, 1, sizeof(journal_entry_small_write), op->len + JOURNAL_STABILIZE_RESERVATION))
+        {
+            return 0;
+        }
+        write_iodepth++;
+        // There is sufficient space. Get SQE(s)
+        struct io_uring_sqe *sqe1 = NULL;
+        if (immediate_commit != IMMEDIATE_NONE ||
+            (journal_block_size - journal.in_sector_pos) < sizeof(journal_entry_small_write) &&
+            journal.sector_info[journal.cur_sector].dirty)
+        {
+            // Write current journal sector only if it's dirty and full, or in the immediate_commit mode
+            BS_SUBMIT_GET_SQE_DECL(sqe1);
+        }
+        struct io_uring_sqe *sqe2 = NULL;
+        if (op->len > 0)
+        {
+            BS_SUBMIT_GET_SQE_DECL(sqe2);
+        }
+        // Got SQEs. Prepare previous journal sector write if required
+        auto cb = [this, op](ring_data_t *data) { handle_write_event(data, op); };
+        if (immediate_commit == IMMEDIATE_NONE)
+        {
+            if (sqe1)
+            {
+                prepare_journal_sector_write(journal, journal.cur_sector, sqe1, cb);
+                PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
+                PRIV(op)->pending_ops++;
+            }
+            else
+            {
+                PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 0;
+            }
+        }
+        // Then pre-fill journal entry
+        journal_entry_small_write *je = (journal_entry_small_write*)prefill_single_journal_entry(
+            journal, op->opcode == BS_OP_WRITE_STABLE ? JE_SMALL_WRITE_INSTANT : JE_SMALL_WRITE,
+            sizeof(journal_entry_small_write) + clean_entry_bitmap_size
+        );
+        dirty_it->second.journal_sector = journal.sector_info[journal.cur_sector].offset;
+        journal.used_sectors[journal.sector_info[journal.cur_sector].offset]++;
+#ifdef BLOCKSTORE_DEBUG
+        printf(
+            "journal offset %08lx is used by %lx:%lx v%lu (%lu refs)\n",
+            dirty_it->second.journal_sector, dirty_it->first.oid.inode, dirty_it->first.oid.stripe, dirty_it->first.version,
+            journal.used_sectors[journal.sector_info[journal.cur_sector].offset]
+        );
+#endif
+        // Figure out where data will be
+        journal.next_free = (journal.next_free + op->len) <= journal.len ? journal.next_free : journal_block_size;
+        je->oid = op->oid;
+        je->version = op->version;
+        je->offset = op->offset;
+        je->len = op->len;
+        je->data_offset = journal.next_free;
+        je->crc32_data = crc32c(0, op->buf, op->len);
+        memcpy((void*)(je+1), (clean_entry_bitmap_size > sizeof(void*) ? dirty_it->second.bitmap : &dirty_it->second.bitmap), clean_entry_bitmap_size);
+        je->crc32 = je_crc32((journal_entry*)je);
+        journal.crc32_last = je->crc32;
+        if (immediate_commit != IMMEDIATE_NONE)
+        {
+            prepare_journal_sector_write(journal, journal.cur_sector, sqe1, cb);
+            PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
+            PRIV(op)->pending_ops++;
+        }
+        if (op->len > 0)
+        {
+            // Prepare journal data write
+            if (journal.inmemory)
+            {
+                // Copy data
+                memcpy(journal.buffer + journal.next_free, op->buf, op->len);
+            }
+            ring_data_t *data2 = ((ring_data_t*)sqe2->user_data);
+            data2->iov = (struct iovec){ op->buf, op->len };
+            data2->callback = cb;
+            my_uring_prep_writev(
+                sqe2, journal.fd, &data2->iov, 1, journal.offset + journal.next_free
+            );
+            PRIV(op)->pending_ops++;
+        }
+        else
+        {
+            // Zero-length overwrite. Allowed to bump object version in EC placement groups without actually writing data
+        }
+        dirty_it->second.location = journal.next_free;
+        dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | BS_ST_SUBMITTED;
+        journal.next_free += op->len;
+        if (journal.next_free >= journal.len)
+        {
+            journal.next_free = journal_block_size;
+        }
+        if (immediate_commit == IMMEDIATE_NONE)
+        {
+            // Remember small write as unsynced
+            unsynced_small_writes.push_back((obj_ver_id){
+                .oid = op->oid,
+                .version = op->version,
+            });
+        }
+        if (!PRIV(op)->pending_ops)
+        {
+            PRIV(op)->op_state = 4;
+            return continue_write(op);
+        }
+        else
+        {
+            PRIV(op)->op_state = 3;
+        }
+    }
+    return 1;
+}
+
+int blockstore_impl_t::continue_write(blockstore_op_t *op)
+{
+    io_uring_sqe *sqe = NULL;
+    journal_entry_big_write *je;
+    int op_state = PRIV(op)->op_state;
+    if (op_state != 2 && op_state != 4)
+    {
+        // In progress
+        return 1;
+    }
+    auto dirty_it = dirty_db.find((obj_ver_id){
+        .oid = op->oid,
+        .version = op->version,
+    });
+    assert(dirty_it != dirty_db.end());
+    if (op_state == 2)
+        goto resume_2;
+    else if (op_state == 4)
+        goto resume_4;
+resume_2:
+    // Only for the immediate_commit mode: prepare and submit big_write journal entry
+    BS_SUBMIT_GET_SQE_DECL(sqe);
+    je = (journal_entry_big_write*)prefill_single_journal_entry(
+        journal, op->opcode == BS_OP_WRITE_STABLE ? JE_BIG_WRITE_INSTANT : JE_BIG_WRITE,
+        sizeof(journal_entry_big_write) + clean_entry_bitmap_size
+    );
+    dirty_it->second.journal_sector = journal.sector_info[journal.cur_sector].offset;
+    journal.used_sectors[journal.sector_info[journal.cur_sector].offset]++;
+#ifdef BLOCKSTORE_DEBUG
+    printf(
+        "journal offset %08lx is used by %lx:%lx v%lu (%lu refs)\n",
+        journal.sector_info[journal.cur_sector].offset, op->oid.inode, op->oid.stripe, op->version,
+        journal.used_sectors[journal.sector_info[journal.cur_sector].offset]
+    );
+#endif
+    je->oid = op->oid;
+    je->version = op->version;
+    je->offset = op->offset;
+    je->len = op->len;
+    je->location = dirty_it->second.location;
+    memcpy((void*)(je+1), (clean_entry_bitmap_size > sizeof(void*) ? dirty_it->second.bitmap : &dirty_it->second.bitmap), clean_entry_bitmap_size);
+    je->crc32 = je_crc32((journal_entry*)je);
+    journal.crc32_last = je->crc32;
+    prepare_journal_sector_write(journal, journal.cur_sector, sqe,
+        [this, op](ring_data_t *data) { handle_write_event(data, op); });
+    PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
+    PRIV(op)->pending_ops = 1;
+    PRIV(op)->op_state = 3;
+    return 1;
+resume_4:
+    // Switch object state
+#ifdef BLOCKSTORE_DEBUG
+    printf("Ack write %lx:%lx v%lu = state %x\n", op->oid.inode, op->oid.stripe, op->version, dirty_it->second.state);
+#endif
+    bool imm = (dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_BIG_WRITE
+        ? (immediate_commit == IMMEDIATE_ALL)
+        : (immediate_commit != IMMEDIATE_NONE);
+    if (imm)
+    {
+        auto & unstab = unstable_writes[op->oid];
+        unstab = unstab < op->version ? op->version : unstab;
+    }
+    dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK)
+        | (imm ? BS_ST_SYNCED : BS_ST_WRITTEN);
+    if (imm && ((dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_DELETE || (dirty_it->second.state & BS_ST_INSTANT)))
+    {
+        // Deletions are treated as immediately stable
+        mark_stable(dirty_it->first);
+    }
+    if (immediate_commit == IMMEDIATE_ALL)
+    {
+        dirty_it++;
+        while (dirty_it != dirty_db.end() && dirty_it->first.oid == op->oid)
+        {
+            if ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_WAIT_BIG)
+            {
+                dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | BS_ST_IN_FLIGHT;
+            }
+            dirty_it++;
+        }
+    }
+    // Acknowledge write
+    op->retval = op->len;
+    write_iodepth--;
+    FINISH_OP(op);
+    return 2;
+}
+
+void blockstore_impl_t::handle_write_event(ring_data_t *data, blockstore_op_t *op)
+{
+    live = true;
+    if (data->res != data->iov.iov_len)
+    {
+        // FIXME: our state becomes corrupted after a write error. maybe do something better than just die
+        throw std::runtime_error(
+            "write operation failed ("+std::to_string(data->res)+" != "+std::to_string(data->iov.iov_len)+
+            "). in-memory state is corrupted. AAAAAAAaaaaaaaaa!!!111"
+        );
+    }
+    PRIV(op)->pending_ops--;
+    if (PRIV(op)->pending_ops == 0)
+    {
+        release_journal_sectors(op);
+        PRIV(op)->op_state++;
+        ringloop->wakeup();
+    }
+}
+
+void blockstore_impl_t::release_journal_sectors(blockstore_op_t *op)
+{
+    // Release flushed journal sectors
+    if (PRIV(op)->min_flushed_journal_sector > 0 &&
+        PRIV(op)->max_flushed_journal_sector > 0)
+    {
+        uint64_t s = PRIV(op)->min_flushed_journal_sector;
+        while (1)
+        {
+            journal.sector_info[s-1].flush_count--;
+            if (s != (1+journal.cur_sector) && journal.sector_info[s-1].flush_count == 0)
+            {
+                // We know for sure that we won't write into this sector anymore
+                uint64_t new_ds = journal.sector_info[s-1].offset + journal.block_size;
+                if (new_ds >= journal.len)
+                {
+                    new_ds = journal.block_size;
+                }
+                if ((journal.dirty_start + (journal.dirty_start >= journal.used_start ? 0 : journal.len)) <
+                    (new_ds + (new_ds >= journal.used_start ? 0 : journal.len)))
+                {
+                    journal.dirty_start = new_ds;
+                }
+            }
+            if (s == PRIV(op)->max_flushed_journal_sector)
+                break;
+            s = 1 + s % journal.sector_count;
+        }
+        PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 0;
+    }
+}
+
+int blockstore_impl_t::dequeue_del(blockstore_op_t *op)
+{
+    if (PRIV(op)->op_state)
+    {
+        return continue_write(op);
+    }
+    auto dirty_it = dirty_db.find((obj_ver_id){
+        .oid = op->oid,
+        .version = op->version,
+    });
+    assert(dirty_it != dirty_db.end());
+    blockstore_journal_check_t space_check(this);
+    if (!space_check.check_available(op, 1, sizeof(journal_entry_del), JOURNAL_STABILIZE_RESERVATION))
+    {
+        return 0;
+    }
+    write_iodepth++;
+    io_uring_sqe *sqe = NULL;
+    if (immediate_commit != IMMEDIATE_NONE ||
+        (journal_block_size - journal.in_sector_pos) < sizeof(journal_entry_del) &&
+        journal.sector_info[journal.cur_sector].dirty)
+    {
+        // Write current journal sector only if it's dirty and full, or in the immediate_commit mode
+        BS_SUBMIT_GET_SQE_DECL(sqe);
+    }
+    auto cb = [this, op](ring_data_t *data) { handle_write_event(data, op); };
+    // Prepare journal sector write
+    if (immediate_commit == IMMEDIATE_NONE)
+    {
+        if (sqe)
+        {
+            prepare_journal_sector_write(journal, journal.cur_sector, sqe, cb);
+            PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
+            PRIV(op)->pending_ops++;
+        }
+        else
+        {
+            PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 0;
+        }
+    }
+    // Pre-fill journal entry
+    journal_entry_del *je = (journal_entry_del*)prefill_single_journal_entry(
+        journal, JE_DELETE, sizeof(struct journal_entry_del)
+    );
+    dirty_it->second.journal_sector = journal.sector_info[journal.cur_sector].offset;
+    journal.used_sectors[journal.sector_info[journal.cur_sector].offset]++;
+#ifdef BLOCKSTORE_DEBUG
+    printf(
+        "journal offset %08lx is used by %lx:%lx v%lu (%lu refs)\n",
+        dirty_it->second.journal_sector, dirty_it->first.oid.inode, dirty_it->first.oid.stripe, dirty_it->first.version,
+        journal.used_sectors[journal.sector_info[journal.cur_sector].offset]
+    );
+#endif
+    je->oid = op->oid;
+    je->version = op->version;
+    je->crc32 = je_crc32((journal_entry*)je);
+    journal.crc32_last = je->crc32;
+    dirty_it->second.state = BS_ST_DELETE | BS_ST_SUBMITTED;
+    if (immediate_commit != IMMEDIATE_NONE)
+    {
+        prepare_journal_sector_write(journal, journal.cur_sector, sqe, cb);
+        PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
+        PRIV(op)->pending_ops++;
+    }
+    else
+    {
+        // Remember delete as unsynced
+        unsynced_small_writes.push_back((obj_ver_id){
+            .oid = op->oid,
+            .version = op->version,
+        });
+    }
+    if (!PRIV(op)->pending_ops)
+    {
+        PRIV(op)->op_state = 4;
+        return continue_write(op);
+    }
+    else
+    {
+        PRIV(op)->op_state = 3;
+    }
+    return 1;
+}

src/cluster_client.cpp

@@ -0,0 +1,931 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#include <stdexcept>
+#include "cluster_client.h"
+
+#define SCRAP_BUFFER_SIZE 4*1024*1024
+
+cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json & config)
+{
+    this->ringloop = ringloop;
+    this->tfd = tfd;
+    this->config = config;
+
+    msgr.osd_num = 0;
+    msgr.tfd = tfd;
+    msgr.ringloop = ringloop;
+    msgr.repeer_pgs = [this](osd_num_t peer_osd)
+    {
+        if (msgr.osd_peer_fds.find(peer_osd) != msgr.osd_peer_fds.end())
+        {
+            // peer_osd just connected
+            continue_ops();
+        }
+        else if (unsynced_writes.size())
+        {
+            // peer_osd just dropped connection
+            for (auto op: syncing_writes)
+            {
+                for (auto & part: op->parts)
+                {
+                    if (part.osd_num == peer_osd && part.done)
+                    {
+                        // repeat this operation
+                        part.osd_num = 0;
+                        part.done = false;
+                        assert(!part.sent);
+                        op->done_count--;
+                    }
+                }
+            }
+            for (auto op: unsynced_writes)
+            {
+                for (auto & part: op->parts)
+                {
+                    if (part.osd_num == peer_osd && part.done)
+                    {
+                        // repeat this operation
+                        part.osd_num = 0;
+                        part.done = false;
+                        assert(!part.sent);
+                        op->done_count--;
+                    }
+                }
+                if (op->done_count < op->parts.size())
+                {
+                    cur_ops.insert(op);
+                }
+            }
+            continue_ops();
+        }
+    };
+    msgr.exec_op = [this](osd_op_t *op)
+    {
+        // Garbage in
+        printf("Incoming garbage from peer %d\n", op->peer_fd);
+        msgr.stop_client(op->peer_fd);
+        delete op;
+    };
+    msgr.init();
+
+    st_cli.tfd = tfd;
+    st_cli.on_load_config_hook = [this](json11::Json::object & cfg) { on_load_config_hook(cfg); };
+    st_cli.on_change_osd_state_hook = [this](uint64_t peer_osd) { on_change_osd_state_hook(peer_osd); };
+    st_cli.on_change_hook = [this](json11::Json::object & changes) { on_change_hook(changes); };
+    st_cli.on_load_pgs_hook = [this](bool success) { on_load_pgs_hook(success); };
+
+    st_cli.parse_config(config);
+    st_cli.load_global_config();
+
+    scrap_buffer_size = SCRAP_BUFFER_SIZE;
+    scrap_buffer = malloc_or_die(scrap_buffer_size);
+
+    if (ringloop)
+    {
+        consumer.loop = [this]()
+        {
+            msgr.read_requests();
+            msgr.send_replies();
+            this->ringloop->submit();
+        };
+        ringloop->register_consumer(&consumer);
+    }
+}
+
+cluster_client_t::~cluster_client_t()
+{
+    if (ringloop)
+    {
+        ringloop->unregister_consumer(&consumer);
+    }
+    free(scrap_buffer);
+}
+
+cluster_op_t::~cluster_op_t()
+{
+    if (buf)
+    {
+        free(buf);
+        buf = NULL;
+    }
+    if (bitmap_buf)
+    {
+        free(bitmap_buf);
+        part_bitmaps = NULL;
+        bitmap_buf = NULL;
+    }
+}
+
+void cluster_client_t::continue_ops(bool up_retry)
+{
+    for (auto op_it = cur_ops.begin(); op_it != cur_ops.end(); )
+    {
+        if ((*op_it)->up_wait)
+        {
+            if (up_retry)
+            {
+                (*op_it)->up_wait = false;
+                continue_rw(*op_it++);
+            }
+            else
+                op_it++;
+        }
+        else
+            continue_rw(*op_it++);
+    }
+}
+
+static uint32_t is_power_of_two(uint64_t value)
+{
+    uint32_t l = 0;
+    while (value > 1)
+    {
+        if (value & 1)
+        {
+            return 64;
+        }
+        value = value >> 1;
+        l++;
+    }
+    return l;
+}
+
+void cluster_client_t::on_load_config_hook(json11::Json::object & config)
+{
+    bs_block_size = config["block_size"].uint64_value();
+    bs_bitmap_granularity = config["bitmap_granularity"].uint64_value();
+    if (!bs_block_size)
+    {
+        bs_block_size = DEFAULT_BLOCK_SIZE;
+    }
+    if (!bs_bitmap_granularity)
+    {
+        bs_bitmap_granularity = DEFAULT_BITMAP_GRANULARITY;
+    }
+    bs_bitmap_size = bs_block_size / bs_bitmap_granularity / 8;
+    uint32_t block_order;
+    if ((block_order = is_power_of_two(bs_block_size)) >= 64 || bs_block_size < MIN_BLOCK_SIZE || bs_block_size >= MAX_BLOCK_SIZE)
+    {
+        throw std::runtime_error("Bad block size");
+    }
+    if (config["immediate_commit"] == "all")
+    {
+        // Cluster-wide immediate_commit mode
+        immediate_commit = true;
+    }
+    else if (config.find("client_dirty_limit") != config.end())
+    {
+        client_dirty_limit = config["client_dirty_limit"].uint64_value();
+    }
+    if (!client_dirty_limit)
+    {
+        client_dirty_limit = DEFAULT_CLIENT_DIRTY_LIMIT;
+    }
+    up_wait_retry_interval = config["up_wait_retry_interval"].uint64_value();
+    if (!up_wait_retry_interval)
+    {
+        up_wait_retry_interval = 500;
+    }
+    else if (up_wait_retry_interval < 50)
+    {
+        up_wait_retry_interval = 50;
+    }
+    msgr.parse_config(config);
+    msgr.parse_config(this->config);
+    st_cli.load_pgs();
+}
+
+void cluster_client_t::on_load_pgs_hook(bool success)
+{
+    for (auto pool_item: st_cli.pool_config)
+    {
+        pg_counts[pool_item.first] = pool_item.second.real_pg_count;
+    }
+    pgs_loaded = true;
+    for (auto fn: on_ready_hooks)
+    {
+        fn();
+    }
+    on_ready_hooks.clear();
+    for (auto op: offline_ops)
+    {
+        execute(op);
+    }
+    offline_ops.clear();
+    continue_ops();
+}
+
+void cluster_client_t::on_change_hook(json11::Json::object & changes)
+{
+    for (auto pool_item: st_cli.pool_config)
+    {
+        if (pg_counts[pool_item.first] != pool_item.second.real_pg_count)
+        {
+            // At this point, all pool operations should have been suspended
+            // And now they have to be resliced!
+            for (auto op: cur_ops)
+            {
+                if (INODE_POOL(op->cur_inode) == pool_item.first)
+                {
+                    op->needs_reslice = true;
+                }
+            }
+            for (auto op: unsynced_writes)
+            {
+                if (INODE_POOL(op->cur_inode) == pool_item.first)
+                {
+                    op->needs_reslice = true;
+                }
+            }
+            for (auto op: syncing_writes)
+            {
+                if (INODE_POOL(op->cur_inode) == pool_item.first)
+                {
+                    op->needs_reslice = true;
+                }
+            }
+            pg_counts[pool_item.first] = pool_item.second.real_pg_count;
+        }
+    }
+    continue_ops();
+}
+
+void cluster_client_t::on_change_osd_state_hook(uint64_t peer_osd)
+{
+    if (msgr.wanted_peers.find(peer_osd) != msgr.wanted_peers.end())
+    {
+        msgr.connect_peer(peer_osd, st_cli.peer_states[peer_osd]);
+    }
+}
+
+bool cluster_client_t::is_ready()
+{
+    return pgs_loaded;
+}
+
+void cluster_client_t::on_ready(std::function<void(void)> fn)
+{
+    if (pgs_loaded)
+    {
+        fn();
+    }
+    else
+    {
+        on_ready_hooks.push_back(fn);
+    }
+}
+
+/**
+ * How writes are synced when immediate_commit is false
+ *
+ * 1) accept up to <client_dirty_limit> write operations for execution,
+ *    queue all subsequent writes into <next_writes>
+ * 2) accept exactly one SYNC, queue all subsequent SYNCs into <next_writes>, too
+ * 3) "continue" all accepted writes
+ *
+ * "Continue" WRITE:
+ * 1) if the operation is not a copy yet - copy it (required for replay)
+ * 2) if the operation is not sliced yet - slice it
+ * 3) if the operation doesn't require reslice - try to connect & send all remaining parts
+ * 4) if any of them fail due to disconnected peers or PGs not up, repeat after reconnecting or small timeout
+ * 5) if any of them fail due to other errors, fail the operation and forget it from the current "unsynced batch"
+ * 6) if PG count changes before all parts are done, wait for all in-progress parts to finish,
+ *    throw all results away, reslice and resubmit op
+ * 7) when all parts are done, try to "continue" the current SYNC
+ * 8) if the operation succeeds, but then some OSDs drop their connections, repeat
+ *    parts from the current "unsynced batch" previously sent to those OSDs in any order
+ *
+ * "Continue" current SYNC:
+ * 1) take all unsynced operations from the current batch
+ * 2) check if all affected OSDs are still alive
+ * 3) if yes, send all SYNCs. otherwise, leave current SYNC as is.
+ * 4) if any of them fail due to disconnected peers, repeat SYNC after repeating all writes
+ * 5) if any of them fail due to other errors, fail the SYNC operation
+ */
+
+void cluster_client_t::execute(cluster_op_t *op)
+{
+    if (!pgs_loaded)
+    {
+        // We're offline
+        offline_ops.push_back(op);
+        return;
+    }
+    op->retval = 0;
+    if (op->opcode != OSD_OP_SYNC && op->opcode != OSD_OP_READ && op->opcode != OSD_OP_WRITE ||
+        (op->opcode == OSD_OP_READ || op->opcode == OSD_OP_WRITE) && (!op->inode || !op->len ||
+        op->offset % bs_bitmap_granularity || op->len % bs_bitmap_granularity))
+    {
+        op->retval = -EINVAL;
+        std::function<void(cluster_op_t*)>(op->callback)(op);
+        return;
+    }
+    if (op->opcode == OSD_OP_SYNC)
+    {
+        execute_sync(op);
+        return;
+    }
+    op->cur_inode = op->inode;
+    if (op->opcode == OSD_OP_WRITE)
+    {
+        auto ino_it = st_cli.inode_config.find(op->inode);
+        if (ino_it != st_cli.inode_config.end() && ino_it->second.readonly)
+        {
+            op->retval = -EINVAL;
+            std::function<void(cluster_op_t*)>(op->callback)(op);
+            return;
+        }
+        if (!immediate_commit)
+        {
+            if (next_writes.size() > 0)
+            {
+                assert(cur_sync);
+                next_writes.push_back(op);
+                return;
+            }
+            if (queued_bytes >= client_dirty_limit)
+            {
+                // Push an extra SYNC operation to flush previous writes
+                next_writes.push_back(op);
+                cluster_op_t *sync_op = new cluster_op_t;
+                sync_op->is_internal = true;
+                sync_op->opcode = OSD_OP_SYNC;
+                sync_op->callback = [](cluster_op_t* sync_op) {};
+                execute_sync(sync_op);
+                return;
+            }
+            queued_bytes += op->len;
+            op = copy_write(op);
+            unsynced_writes.push_back(op);
+        }
+    }
+    cur_ops.insert(op);
+    continue_rw(op);
+}
+
+cluster_op_t *cluster_client_t::copy_write(cluster_op_t *op)
+{
+    // Save operation for replay when one of PGs goes out of sync
+    // (primary OSD drops our connection in this case)
+    cluster_op_t *op_copy = new cluster_op_t();
+    op_copy->is_internal = true;
+    op_copy->orig_op = op;
+    op_copy->opcode = op->opcode;
+    op_copy->inode = op->inode;
+    op_copy->cur_inode = op->inode;
+    op_copy->offset = op->offset;
+    op_copy->len = op->len;
+    op_copy->buf = malloc_or_die(op->len);
+    op_copy->iov.push_back(op_copy->buf, op->len);
+    op_copy->callback = [](cluster_op_t* op_copy)
+    {
+        if (op_copy->orig_op)
+        {
+            // Acknowledge write and forget the original pointer
+            op_copy->orig_op->retval = op_copy->retval;
+            std::function<void(cluster_op_t*)>(op_copy->orig_op->callback)(op_copy->orig_op);
+            op_copy->orig_op = NULL;
+        }
+    };
+    void *cur_buf = op_copy->buf;
+    for (int i = 0; i < op->iov.count; i++)
+    {
+        memcpy(cur_buf, op->iov.buf[i].iov_base, op->iov.buf[i].iov_len);
+        cur_buf += op->iov.buf[i].iov_len;
+    }
+    return op_copy;
+}
+
+// FIXME Reimplement it using "coroutine emulation"
+void cluster_client_t::continue_rw(cluster_op_t *op)
+{
+    pool_id_t pool_id = INODE_POOL(op->cur_inode);
+    if (!pool_id)
+    {
+        op->retval = -EINVAL;
+        std::function<void(cluster_op_t*)>(op->callback)(op);
+        return;
+    }
+    if (st_cli.pool_config.find(pool_id) == st_cli.pool_config.end() ||
+        st_cli.pool_config[pool_id].real_pg_count == 0)
+    {
+        // Postpone operations to unknown pools
+        return;
+    }
+    if (!op->parts.size())
+    {
+        // Slice the operation into parts
+        slice_rw(op);
+    }
+    if (!op->needs_reslice)
+    {
+        // Send unsent parts, if they're not subject to change
+        for (int i = 0; i < op->parts.size(); i++)
+        {
+            if (!op->parts[i].sent && !op->parts[i].done)
+            {
+                try_send(op, i);
+            }
+        }
+    }
+    if (!op->sent_count)
+    {
+        if (op->done_count >= op->parts.size())
+        {
+            // Finished successfully
+            // Even if the PG count has changed in meanwhile we treat it as success
+            // because if some operations were invalid for the new PG count we'd get errors
+            bool is_read = op->opcode == OSD_OP_READ;
+            if (is_read)
+            {
+                // Check parent inode
+                auto ino_it = st_cli.inode_config.find(op->cur_inode);
+                if (ino_it != st_cli.inode_config.end() &&
+                    ino_it->second.parent_id)
+                {
+                    // Continue reading from the parent inode
+                    // FIXME: This obviously requires optimizations for long snapshot chains
+                    op->cur_inode = ino_it->second.parent_id;
+                    op->parts.clear();
+                    op->done_count = 0;
+                    op->needs_reslice = true;
+                    continue_rw(op);
+                    return;
+                }
+            }
+            cur_ops.erase(op);
+            op->retval = op->len;
+            std::function<void(cluster_op_t*)>(op->callback)(op);
+            if (!is_read)
+            {
+                continue_sync();
+            }
+            return;
+        }
+        else if (op->retval != 0 && op->retval != -EPIPE)
+        {
+            // Fatal error (not -EPIPE)
+            bool is_read = op->opcode == OSD_OP_READ;
+            cur_ops.erase(op);
+            if (!immediate_commit && op->opcode == OSD_OP_WRITE)
+            {
+                for (int i = 0; i < unsynced_writes.size(); i++)
+                {
+                    if (unsynced_writes[i] == op)
+                    {
+                        unsynced_writes.erase(unsynced_writes.begin()+i, unsynced_writes.begin()+i+1);
+                        break;
+                    }
+                }
+            }
+            bool del = op->is_internal;
+            std::function<void(cluster_op_t*)>(op->callback)(op);
+            if (del)
+            {
+                delete op;
+            }
+            if (!is_read)
+            {
+                continue_sync();
+            }
+            return;
+        }
+        else
+        {
+            // -EPIPE or no error - clear the error
+            op->retval = 0;
+            if (op->needs_reslice)
+            {
+                op->parts.clear();
+                op->done_count = 0;
+                op->needs_reslice = false;
+                continue_rw(op);
+            }
+        }
+    }
+}
+
+static void add_iov(int size, bool skip, cluster_op_t *op, int &iov_idx, size_t &iov_pos, osd_op_buf_list_t &iov, void *scrap, int scrap_len)
+{
+    int left = size;
+    while (left > 0 && iov_idx < op->iov.count)
+    {
+        int cur_left = op->iov.buf[iov_idx].iov_len - iov_pos;
+        if (cur_left < left)
+        {
+            if (!skip)
+            {
+                iov.push_back(op->iov.buf[iov_idx].iov_base + iov_pos, cur_left);
+            }
+            left -= cur_left;
+            iov_pos = 0;
+            iov_idx++;
+        }
+        else
+        {
+            if (!skip)
+            {
+                iov.push_back(op->iov.buf[iov_idx].iov_base + iov_pos, left);
+            }
+            iov_pos += left;
+            left = 0;
+        }
+    }
+    assert(left == 0);
+    if (skip && scrap_len > 0)
+    {
+        // All skipped ranges are read into the same useless buffer
+        left = size;
+        while (left > 0)
+        {
+            int cur_left = scrap_len < left ? scrap_len : left;
+            iov.push_back(scrap, cur_left);
+            left -= cur_left;
+        }
+    }
+}
+
+void cluster_client_t::slice_rw(cluster_op_t *op)
+{
+    // Slice the request into individual object stripe requests
+    // Primary OSDs still operate individual stripes, but their size is multiplied by PG minsize in case of EC
+    auto & pool_cfg = st_cli.pool_config[INODE_POOL(op->cur_inode)];
+    uint32_t pg_data_size = (
+        pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks
+    );
+    uint64_t pg_block_size = bs_block_size * pg_data_size;
+    uint64_t first_stripe = (op->offset / pg_block_size) * pg_block_size;
+    uint64_t last_stripe = ((op->offset + op->len + pg_block_size - 1) / pg_block_size - 1) * pg_block_size;
+    op->retval = 0;
+    op->parts.resize((last_stripe - first_stripe) / pg_block_size + 1);
+    if (op->opcode == OSD_OP_READ)
+    {
+        // Allocate memory for the bitmap
+        unsigned object_bitmap_size = ((op->len / bs_bitmap_granularity + 7) / 8);
+        object_bitmap_size = (object_bitmap_size < 8 ? 8 : object_bitmap_size);
+        unsigned bitmap_mem = object_bitmap_size + (bs_bitmap_size * pg_data_size) * op->parts.size();
+        if (op->bitmap_buf_size < bitmap_mem)
+        {
+            op->bitmap_buf = realloc_or_die(op->bitmap_buf, bitmap_mem);
+            if (!op->bitmap_buf_size)
+            {
+                // First allocation
+                memset(op->bitmap_buf, 0, object_bitmap_size);
+            }
+            op->part_bitmaps = op->bitmap_buf + object_bitmap_size;
+            op->bitmap_buf_size = bitmap_mem;
+        }
+    }
+    int iov_idx = 0;
+    size_t iov_pos = 0;
+    int i = 0;
+    for (uint64_t stripe = first_stripe; stripe <= last_stripe; stripe += pg_block_size)
+    {
+        pg_num_t pg_num = (op->cur_inode + stripe/pool_cfg.pg_stripe_size) % pool_cfg.real_pg_count + 1; // like map_to_pg()
+        uint64_t begin = (op->offset < stripe ? stripe : op->offset);
+        uint64_t end = (op->offset + op->len) > (stripe + pg_block_size)
+            ? (stripe + pg_block_size) : (op->offset + op->len);
+        op->parts[i].iov.reset();
+        if (op->cur_inode != op->inode)
+        {
+            // Read remaining parts from upper layers
+            uint64_t prev = begin, cur = begin;
+            bool skip_prev = true;
+            while (cur < end)
+            {
+                unsigned bmp_loc = (cur - op->offset)/bs_bitmap_granularity;
+                bool skip = (((*(uint8_t*)(op->bitmap_buf + bmp_loc/8)) >> (bmp_loc%8)) & 0x1);
+                if (skip_prev != skip)
+                {
+                    if (cur > prev)
+                    {
+                        if (prev == begin && skip_prev)
+                        {
+                            begin = cur;
+                            // Just advance iov_idx & iov_pos
+                            add_iov(cur-prev, true, op, iov_idx, iov_pos, op->parts[i].iov, NULL, 0);
+                        }
+                        else
+                            add_iov(cur-prev, skip_prev, op, iov_idx, iov_pos, op->parts[i].iov, scrap_buffer, scrap_buffer_size);
+                    }
+                    skip_prev = skip;
+                    prev = cur;
+                }
+                cur += bs_bitmap_granularity;
+            }
+            assert(cur > prev);
+            if (skip_prev)
+            {
+                // Just advance iov_idx & iov_pos
+                add_iov(end-prev, true, op, iov_idx, iov_pos, op->parts[i].iov, NULL, 0);
+                end = prev;
+            }
+            else
+                add_iov(cur-prev, skip_prev, op, iov_idx, iov_pos, op->parts[i].iov, scrap_buffer, scrap_buffer_size);
+            if (end == begin)
+                op->done_count++;
+        }
+        else
+        {
+            add_iov(end-begin, false, op, iov_idx, iov_pos, op->parts[i].iov, NULL, 0);
+        }
+        op->parts[i].parent = op;
+        op->parts[i].offset = begin;
+        op->parts[i].len = (uint32_t)(end - begin);
+        op->parts[i].pg_num = pg_num;
+        op->parts[i].osd_num = 0;
+        op->parts[i].sent = end <= begin;
+        op->parts[i].done = end <= begin;
+        i++;
+    }
+}
+
+bool cluster_client_t::try_send(cluster_op_t *op, int i)
+{
+    cluster_op_part_t *part = &op->parts[i];
+    auto & pool_cfg = st_cli.pool_config[INODE_POOL(op->cur_inode)];
+    auto pg_it = pool_cfg.pg_config.find(part->pg_num);
+    if (pg_it != pool_cfg.pg_config.end() &&
+        !pg_it->second.pause && pg_it->second.cur_primary)
+    {
+        osd_num_t primary_osd = pg_it->second.cur_primary;
+        auto peer_it = msgr.osd_peer_fds.find(primary_osd);
+        if (peer_it != msgr.osd_peer_fds.end())
+        {
+            int peer_fd = peer_it->second;
+            part->osd_num = primary_osd;
+            part->sent = true;
+            op->sent_count++;
+            uint64_t pg_bitmap_size = bs_bitmap_size * (
+                pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks
+            );
+            part->op = (osd_op_t){
+                .op_type = OSD_OP_OUT,
+                .peer_fd = peer_fd,
+                .req = { .rw = {
+                    .header = {
+                        .magic = SECONDARY_OSD_OP_MAGIC,
+                        .id = op_id++,
+                        .opcode = op->opcode,
+                    },
+                    .inode = op->cur_inode,
+                    .offset = part->offset,
+                    .len = part->len,
+                } },
+                .bitmap = op->opcode == OSD_OP_WRITE ? NULL : op->part_bitmaps + pg_bitmap_size*i,
+                .bitmap_len = (unsigned)(op->opcode == OSD_OP_WRITE ? 0 : pg_bitmap_size),
+                .callback = [this, part](osd_op_t *op_part)
+                {
+                    handle_op_part(part);
+                },
+            };
+            part->op.iov = part->iov;
+            msgr.outbox_push(&part->op);
+            return true;
+        }
+        else if (msgr.wanted_peers.find(primary_osd) == msgr.wanted_peers.end())
+        {
+            msgr.connect_peer(primary_osd, st_cli.peer_states[primary_osd]);
+        }
+    }
+    return false;
+}
+
+void cluster_client_t::execute_sync(cluster_op_t *op)
+{
+    if (immediate_commit)
+    {
+        // Syncs are not required in the immediate_commit mode
+        op->retval = 0;
+        std::function<void(cluster_op_t*)>(op->callback)(op);
+    }
+    else if (cur_sync != NULL)
+    {
+        next_writes.push_back(op);
+    }
+    else
+    {
+        cur_sync = op;
+        continue_sync();
+    }
+}
+
+void cluster_client_t::continue_sync()
+{
+    if (!cur_sync || cur_sync->parts.size() > 0)
+    {
+        // Already submitted
+        return;
+    }
+    cur_sync->retval = 0;
+    std::set<osd_num_t> sync_osds;
+    for (auto prev_op: unsynced_writes)
+    {
+        if (prev_op->done_count < prev_op->parts.size())
+        {
+            // Writes not finished yet
+            return;
+        }
+        for (auto & part: prev_op->parts)
+        {
+            if (part.osd_num)
+            {
+                sync_osds.insert(part.osd_num);
+            }
+        }
+    }
+    if (!sync_osds.size())
+    {
+        // No dirty writes
+        finish_sync();
+        return;
+    }
+    // Check that all OSD connections are still alive
+    for (auto sync_osd: sync_osds)
+    {
+        auto peer_it = msgr.osd_peer_fds.find(sync_osd);
+        if (peer_it == msgr.osd_peer_fds.end())
+        {
+            // SYNC is pointless to send to a non connected OSD
+            return;
+        }
+    }
+    syncing_writes.swap(unsynced_writes);
+    // Post sync to affected OSDs
+    cur_sync->parts.resize(sync_osds.size());
+    int i = 0;
+    for (auto sync_osd: sync_osds)
+    {
+        cur_sync->parts[i] = {
+            .parent = cur_sync,
+            .osd_num = sync_osd,
+            .sent = false,
+            .done = false,
+        };
+        send_sync(cur_sync, &cur_sync->parts[i]);
+        i++;
+    }
+}
+
+void cluster_client_t::finish_sync()
+{
+    int retval = cur_sync->retval;
+    if (retval != 0)
+    {
+        for (auto op: syncing_writes)
+        {
+            if (op->done_count < op->parts.size())
+            {
+                cur_ops.insert(op);
+            }
+        }
+        unsynced_writes.insert(unsynced_writes.begin(), syncing_writes.begin(), syncing_writes.end());
+        syncing_writes.clear();
+    }
+    if (retval == -EPIPE)
+    {
+        // Retry later
+        cur_sync->parts.clear();
+        cur_sync->retval = 0;
+        cur_sync->sent_count = 0;
+        cur_sync->done_count = 0;
+        return;
+    }
+    std::function<void(cluster_op_t*)>(cur_sync->callback)(cur_sync);
+    if (!retval)
+    {
+        for (auto op: syncing_writes)
+        {
+            assert(op->sent_count == 0);
+            if (op->is_internal)
+            {
+                delete op;
+            }
+        }
+        syncing_writes.clear();
+    }
+    cur_sync = NULL;
+    queued_bytes = 0;
+    std::vector<cluster_op_t*> next_wr_copy;
+    next_wr_copy.swap(next_writes);
+    for (auto next_op: next_wr_copy)
+    {
+        execute(next_op);
+    }
+}
+
+void cluster_client_t::send_sync(cluster_op_t *op, cluster_op_part_t *part)
+{
+    auto peer_it = msgr.osd_peer_fds.find(part->osd_num);
+    assert(peer_it != msgr.osd_peer_fds.end());
+    part->sent = true;
+    op->sent_count++;
+    part->op = (osd_op_t){
+        .op_type = OSD_OP_OUT,
+        .peer_fd = peer_it->second,
+        .req = {
+            .hdr = {
+                .magic = SECONDARY_OSD_OP_MAGIC,
+                .id = op_id++,
+                .opcode = OSD_OP_SYNC,
+            },
+        },
+        .callback = [this, part](osd_op_t *op_part)
+        {
+            handle_op_part(part);
+        },
+    };
+    msgr.outbox_push(&part->op);
+}
+
+static inline void mem_or(void *res, const void *r2, unsigned int len)
+{
+    unsigned int i;
+    for (i = 0; i < len; ++i)
+    {
+        // Hope the compiler vectorizes this
+        ((uint8_t*)res)[i] = ((uint8_t*)res)[i] | ((uint8_t*)r2)[i];
+    }
+}
+
+void cluster_client_t::handle_op_part(cluster_op_part_t *part)
+{
+    cluster_op_t *op = part->parent;
+    part->sent = false;
+    op->sent_count--;
+    int expected = part->op.req.hdr.opcode == OSD_OP_SYNC ? 0 : part->op.req.rw.len;
+    if (part->op.reply.hdr.retval != expected)
+    {
+        // Operation failed, retry
+        printf(
+            "Operation failed on OSD %lu: retval=%ld (expected %d), dropping connection\n",
+            part->osd_num, part->op.reply.hdr.retval, expected
+        );
+        msgr.stop_client(part->op.peer_fd);
+        if (part->op.reply.hdr.retval == -EPIPE)
+        {
+            op->up_wait = true;
+            if (!retry_timeout_id)
+            {
+                retry_timeout_id = tfd->set_timer(up_wait_retry_interval, false, [this](int)
+                {
+                    retry_timeout_id = 0;
+                    continue_ops(true);
+                });
+            }
+        }
+        if (!op->retval || op->retval == -EPIPE)
+        {
+            // Don't overwrite other errors with -EPIPE
+            op->retval = part->op.reply.hdr.retval;
+        }
+    }
+    else
+    {
+        // OK
+        part->done = true;
+        op->done_count++;
+        if (op->opcode == OSD_OP_READ)
+        {
+            // Copy (OR) bitmap
+            auto & pool_cfg = st_cli.pool_config[INODE_POOL(op->cur_inode)];
+            uint32_t pg_block_size = bs_block_size * (
+                pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks
+            );
+            uint32_t object_offset = (part->op.req.rw.offset - op->offset) / bs_bitmap_granularity;
+            uint32_t part_offset = (part->op.req.rw.offset % pg_block_size) / bs_bitmap_granularity;
+            uint32_t part_len = part->op.req.rw.len / bs_bitmap_granularity;
+            if (!(object_offset & 0x7) && !(part_offset & 0x7) && (part_len >= 8))
+            {
+                // Copy bytes
+                mem_or(op->bitmap_buf + object_offset/8, part->op.bitmap + part_offset/8, part_len/8);
+                object_offset += (part_len & ~0x7);
+                part_offset += (part_len & ~0x7);
+                part_len = (part_len & 0x7);
+            }
+            while (part_len > 0)
+            {
+                // Copy bits
+                (*(uint8_t*)(op->bitmap_buf + (object_offset >> 3))) |= (
+                    (((*(uint8_t*)(part->op.bitmap + (part_offset >> 3))) >> (part_offset & 0x7)) & 0x1) << (object_offset & 0x7)
+                );
+                part_offset++;
+                object_offset++;
+                part_len--;
+            }
+        }
+    }
+    if (op->sent_count == 0)
+    {
+        if (op->opcode == OSD_OP_SYNC)
+        {
+            assert(op == cur_sync);
+            finish_sync();
+        }
+        else if (!op->up_wait)
+        {
+            continue_rw(op);
+        }
+    }
+}

src/cluster_client.h

@@ -0,0 +1,111 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#pragma once
+
+#include "messenger.h"
+#include "etcd_state_client.h"
+
+#define MIN_BLOCK_SIZE 4*1024
+#define MAX_BLOCK_SIZE 128*1024*1024
+#define DEFAULT_CLIENT_DIRTY_LIMIT 32*1024*1024
+
+struct cluster_op_t;
+
+struct cluster_op_part_t
+{
+    cluster_op_t *parent;
+    uint64_t offset;
+    uint32_t len;
+    pg_num_t pg_num;
+    osd_num_t osd_num;
+    osd_op_buf_list_t iov;
+    bool sent;
+    bool done;
+    osd_op_t op;
+};
+
+struct cluster_op_t
+{
+    uint64_t opcode; // OSD_OP_READ, OSD_OP_WRITE, OSD_OP_SYNC
+    uint64_t inode;
+    uint64_t offset;
+    uint64_t len;
+    int retval;
+    osd_op_buf_list_t iov;
+    std::function<void(cluster_op_t*)> callback;
+    ~cluster_op_t();
+protected:
+    uint64_t cur_inode; // for snapshot reads
+    void *buf = NULL;
+    cluster_op_t *orig_op = NULL;
+    bool is_internal = false;
+    bool needs_reslice = false;
+    bool up_wait = false;
+    int sent_count = 0, done_count = 0;
+    std::vector<cluster_op_part_t> parts;
+    void *bitmap_buf = NULL, *part_bitmaps = NULL;
+    unsigned bitmap_buf_size = 0;
+    friend class cluster_client_t;
+};
+
+class cluster_client_t
+{
+    timerfd_manager_t *tfd;
+    ring_loop_t *ringloop;
+
+    uint64_t bs_block_size = 0;
+    uint32_t bs_bitmap_granularity = 0, bs_bitmap_size = 0;
+    std::map<pool_id_t, uint64_t> pg_counts;
+    bool immediate_commit = false;
+    // FIXME: Implement inmemory_commit mode. Note that it requires to return overlapping reads from memory.
+    uint64_t client_dirty_limit = 0;
+    int log_level;
+    int up_wait_retry_interval = 500; // ms
+
+    uint64_t op_id = 1;
+    ring_consumer_t consumer;
+    // operations currently in progress
+    std::set<cluster_op_t*> cur_ops;
+    int retry_timeout_id = 0;
+    // unsynced operations are copied in memory to allow replay when cluster isn't in the immediate_commit mode
+    // unsynced_writes are replayed in any order (because only the SYNC operation guarantees ordering)
+    std::vector<cluster_op_t*> unsynced_writes;
+    std::vector<cluster_op_t*> syncing_writes;
+    cluster_op_t* cur_sync = NULL;
+    std::vector<cluster_op_t*> next_writes;
+    std::vector<cluster_op_t*> offline_ops;
+    uint64_t queued_bytes = 0;
+    void *scrap_buffer = NULL;
+    unsigned scrap_buffer_size = 0;
+
+    bool pgs_loaded = false;
+    std::vector<std::function<void(void)>> on_ready_hooks;
+
+public:
+    etcd_state_client_t st_cli;
+    osd_messenger_t msgr;
+    json11::Json config;
+
+    cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json & config);
+    ~cluster_client_t();
+    void execute(cluster_op_t *op);
+    bool is_ready();
+    void on_ready(std::function<void(void)> fn);
+
+protected:
+    void continue_ops(bool up_retry = false);
+    void on_load_config_hook(json11::Json::object & config);
+    void on_load_pgs_hook(bool success);
+    void on_change_hook(json11::Json::object & changes);
+    void on_change_osd_state_hook(uint64_t peer_osd);
+    cluster_op_t *copy_write(cluster_op_t *op);
+    void continue_rw(cluster_op_t *op);
+    void slice_rw(cluster_op_t *op);
+    bool try_send(cluster_op_t *op, int i);
+    void execute_sync(cluster_op_t *op);
+    void continue_sync();
+    void finish_sync();
+    void send_sync(cluster_op_t *op, cluster_op_part_t *part);
+    void handle_op_part(cluster_op_part_t *part);
+};

src/crc32c.h

@@ -8,4 +8,10 @@
 // unsigned __int64 _mm_crc32_u64 (unsigned __int64 crc, unsigned __int64 v)
 // unsigned int _mm_crc32_u8 (unsigned int crc, unsigned char v)
 
+#ifdef __cplusplus
+extern "C" {
+#endif
 uint32_t crc32c(uint32_t crc, const void *buf, size_t len);
+#ifdef __cplusplus
+};
+#endif

src/dump_journal.cpp

@@ -0,0 +1,224 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+#define _LARGEFILE64_SOURCE
+#include <sys/types.h>
+#include <sys/ioctl.h>
+#include <sys/stat.h>
+#include <sys/time.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <stdint.h>
+#include <malloc.h>
+#include <linux/fs.h>
+#include <string.h>
+#include <errno.h>
+#include <assert.h>
+#include <stdio.h>
+
+#include "blockstore_impl.h"
+#include "crc32c.h"
+
+struct journal_dump_t
+{
+    char *journal_device;
+    uint32_t journal_block;
+    uint64_t journal_offset;
+    uint64_t journal_len;
+    uint64_t journal_pos;
+    bool all;
+    bool started;
+    int fd;
+    uint32_t crc32_last;
+
+    int dump_block(void *buf);
+};
+
+int main(int argc, char *argv[])
+{
+    journal_dump_t self = { 0 };
+    int b = 1;
+    if (argc >= 2 && !strcmp(argv[1], "--all"))
+    {
+        self.all = true;
+        b = 2;
+    }
+    if (argc < b+4)
+    {
+        printf("USAGE: %s [--all] <journal_file> <journal_block_size> <offset> <size>\n", argv[0]);
+        return 1;
+    }
+    self.journal_device = argv[b];
+    self.journal_block = strtoul(argv[b+1], NULL, 10);
+    self.journal_offset = strtoull(argv[b+2], NULL, 10);
+    self.journal_len = strtoull(argv[b+3], NULL, 10);
+    if (self.journal_block < MEM_ALIGNMENT || (self.journal_block % MEM_ALIGNMENT) ||
+        self.journal_block > 128*1024)
+    {
+        printf("Invalid journal block size\n");
+        return 1;
+    }
+    self.fd = open(self.journal_device, O_DIRECT|O_RDONLY);
+    if (self.fd == -1)
+    {
+        printf("Failed to open journal\n");
+        return 1;
+    }
+    void *data = memalign(MEM_ALIGNMENT, self.journal_block);
+    self.journal_pos = 0;
+    if (self.all)
+    {
+        while (self.journal_pos < self.journal_len)
+        {
+            int r = pread(self.fd, data, self.journal_block, self.journal_offset+self.journal_pos);
+            assert(r == self.journal_block);
+            uint64_t s;
+            for (s = 0; s < self.journal_block; s += 8)
+            {
+                if (*((uint64_t*)(data+s)) != 0)
+                    break;
+            }
+            if (s == self.journal_block)
+            {
+                printf("offset %08lx: zeroes\n", self.journal_pos);
+                self.journal_pos += self.journal_block;
+            }
+            else if (((journal_entry*)data)->magic == JOURNAL_MAGIC)
+            {
+                printf("offset %08lx:\n", self.journal_pos);
+                self.dump_block(data);
+            }
+            else
+            {
+                printf("offset %08lx: no magic in the beginning, looks like random data (pattern=%lx)\n", self.journal_pos, *((uint64_t*)data));
+                self.journal_pos += self.journal_block;
+            }
+        }
+    }
+    else
+    {
+        int r = pread(self.fd, data, self.journal_block, self.journal_offset+self.journal_pos);
+        assert(r == self.journal_block);
+        journal_entry *je = (journal_entry*)(data);
+        if (je->magic != JOURNAL_MAGIC || je->type != JE_START || je_crc32(je) != je->crc32)
+        {
+            printf("offset %08lx: journal superblock is invalid\n", self.journal_pos);
+        }
+        else
+        {
+            printf("offset %08lx:\n", self.journal_pos);
+            self.dump_block(data);
+            self.started = false;
+            self.journal_pos = je->start.journal_start;
+            while (1)
+            {
+                if (self.journal_pos >= self.journal_len)
+                    self.journal_pos = self.journal_block;
+                r = pread(self.fd, data, self.journal_block, self.journal_offset+self.journal_pos);
+                assert(r == self.journal_block);
+                printf("offset %08lx:\n", self.journal_pos);
+                r = self.dump_block(data);
+                if (r <= 0)
+                {
+                    printf("end of the journal\n");
+                    break;
+                }
+            }
+        }
+    }
+    free(data);
+    close(self.fd);
+    return 0;
+}
+
+int journal_dump_t::dump_block(void *buf)
+{
+    uint32_t pos = 0;
+    journal_pos += journal_block;
+    int entry = 0;
+    bool wrapped = false;
+    while (pos < journal_block)
+    {
+        journal_entry *je = (journal_entry*)(buf + pos);
+        if (je->magic != JOURNAL_MAGIC || je->type < JE_MIN || je->type > JE_MAX ||
+            !all && started && je->crc32_prev != crc32_last)
+        {
+            break;
+        }
+        bool crc32_valid = je_crc32(je) == je->crc32;
+        if (!all && !crc32_valid)
+        {
+            break;
+        }
+        started = true;
+        crc32_last = je->crc32;
+        printf("entry % 3d: crc32=%08x %s prev=%08x ", entry, je->crc32, (crc32_valid ? "(valid)" : "(invalid)"), je->crc32_prev);
+        if (je->type == JE_START)
+        {
+            printf("je_start start=%08lx\n", je->start.journal_start);
+        }
+        else if (je->type == JE_SMALL_WRITE || je->type == JE_SMALL_WRITE_INSTANT)
+        {
+            printf(
+                "je_small_write%s oid=%lx:%lx ver=%lu offset=%u len=%u loc=%08lx",
+                je->type == JE_SMALL_WRITE_INSTANT ? "_instant" : "",
+                je->small_write.oid.inode, je->small_write.oid.stripe,
+                je->small_write.version, je->small_write.offset, je->small_write.len,
+                je->small_write.data_offset
+            );
+            if (journal_pos + je->small_write.len > journal_len)
+            {
+                // data continues from the beginning of the journal
+                journal_pos = journal_block;
+                wrapped = true;
+            }
+            if (journal_pos != je->small_write.data_offset)
+            {
+                printf(" (mismatched, calculated = %lu)", journal_pos);
+            }
+            journal_pos += je->small_write.len;
+            if (journal_pos >= journal_len)
+            {
+                journal_pos = journal_block;
+                wrapped = true;
+            }
+            uint32_t data_crc32 = 0;
+            void *data = memalign(MEM_ALIGNMENT, je->small_write.len);
+            assert(pread(fd, data, je->small_write.len, journal_offset+je->small_write.data_offset) == je->small_write.len);
+            data_crc32 = crc32c(0, data, je->small_write.len);
+            free(data);
+            printf(
+                " data_crc32=%08x%s", je->small_write.crc32_data,
+                (data_crc32 != je->small_write.crc32_data) ? " (invalid)" : " (valid)"
+            );
+            printf("\n");
+        }
+        else if (je->type == JE_BIG_WRITE || je->type == JE_BIG_WRITE_INSTANT)
+        {
+            printf(
+                "je_big_write%s oid=%lx:%lx ver=%lu loc=%08lx\n",
+                je->type == JE_BIG_WRITE_INSTANT ? "_instant" : "",
+                je->big_write.oid.inode, je->big_write.oid.stripe, je->big_write.version, je->big_write.location
+            );
+        }
+        else if (je->type == JE_STABLE)
+        {
+            printf("je_stable oid=%lx:%lx ver=%lu\n", je->stable.oid.inode, je->stable.oid.stripe, je->stable.version);
+        }
+        else if (je->type == JE_ROLLBACK)
+        {
+            printf("je_rollback oid=%lx:%lx ver=%lu\n", je->rollback.oid.inode, je->rollback.oid.stripe, je->rollback.version);
+        }
+        else if (je->type == JE_DELETE)
+        {
+            printf("je_delete oid=%lx:%lx ver=%lu\n", je->del.oid.inode, je->del.oid.stripe, je->del.version);
+        }
+        pos += je->size;
+        entry++;
+    }
+    if (wrapped)
+    {
+        journal_pos = journal_len;
+    }
+    return entry;
+}

src/epoll_manager.cpp

@@ -0,0 +1,95 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#include <sys/epoll.h>
+#include <sys/poll.h>
+#include <unistd.h>
+#include <stdexcept>
+
+#include "epoll_manager.h"
+
+#define MAX_EPOLL_EVENTS 64
+
+epoll_manager_t::epoll_manager_t(ring_loop_t *ringloop)
+{
+    this->ringloop = ringloop;
+
+    epoll_fd = epoll_create(1);
+    if (epoll_fd < 0)
+    {
+        throw std::runtime_error(std::string("epoll_create: ") + strerror(errno));
+    }
+
+    tfd = new timerfd_manager_t([this](int fd, bool wr, std::function<void(int, int)> handler) { set_fd_handler(fd, wr, handler); });
+
+    handle_epoll_events();
+}
+
+epoll_manager_t::~epoll_manager_t()
+{
+    if (tfd)
+    {
+        delete tfd;
+        tfd = NULL;
+    }
+    close(epoll_fd);
+}
+
+void epoll_manager_t::set_fd_handler(int fd, bool wr, std::function<void(int, int)> handler)
+{
+    if (handler != NULL)
+    {
+        bool exists = epoll_handlers.find(fd) != epoll_handlers.end();
+        epoll_event ev;
+        ev.data.fd = fd;
+        ev.events = (wr ? EPOLLOUT : 0) | EPOLLIN | EPOLLRDHUP | EPOLLET;
+        if (epoll_ctl(epoll_fd, exists ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, fd, &ev) < 0)
+        {
+            throw std::runtime_error(std::string("epoll_ctl: ") + strerror(errno));
+        }
+        epoll_handlers[fd] = handler;
+    }
+    else
+    {
+        if (epoll_ctl(epoll_fd, EPOLL_CTL_DEL, fd, NULL) < 0 && errno != ENOENT)
+        {
+            throw std::runtime_error(std::string("epoll_ctl: ") + strerror(errno));
+        }
+        epoll_handlers.erase(fd);
+    }
+}
+
+void epoll_manager_t::handle_epoll_events()
+{
+    io_uring_sqe *sqe = ringloop->get_sqe();
+    if (!sqe)
+    {
+        throw std::runtime_error("can't get SQE, will fall out of sync with EPOLLET");
+    }
+    ring_data_t *data = ((ring_data_t*)sqe->user_data);
+    my_uring_prep_poll_add(sqe, epoll_fd, POLLIN);
+    data->callback = [this](ring_data_t *data)
+    {
+        if (data->res < 0)
+        {
+            throw std::runtime_error(std::string("epoll failed: ") + strerror(-data->res));
+        }
+        handle_epoll_events();
+    };
+    ringloop->submit();
+    int nfds;
+    epoll_event events[MAX_EPOLL_EVENTS];
+    do
+    {
+        nfds = epoll_wait(epoll_fd, events, MAX_EPOLL_EVENTS, 0);
+        for (int i = 0; i < nfds; i++)
+        {
+            auto cb_it = epoll_handlers.find(events[i].data.fd);
+            if (cb_it != epoll_handlers.end())
+            {
+                auto & cb = cb_it->second;
+                cb(events[i].data.fd, events[i].events);
+            }
+        }
+    } while (nfds == MAX_EPOLL_EVENTS);
+}

src/epoll_manager.h

@@ -0,0 +1,23 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#pragma once
+
+#include <map>
+
+#include "ringloop.h"
+#include "timerfd_manager.h"
+
+class epoll_manager_t
+{
+    int epoll_fd;
+    ring_loop_t *ringloop;
+    std::map<int, std::function<void(int, int)>> epoll_handlers;
+public:
+    epoll_manager_t(ring_loop_t *ringloop);
+    ~epoll_manager_t();
+    void set_fd_handler(int fd, bool wr, std::function<void(int, int)> handler);
+    void handle_epoll_events();
+
+    timerfd_manager_t *tfd;
+};