Test: replace EPOLLIN with blocking sockets + io_uring reads

2020-03-01 17:23:14 +03:00
169 changed files with 5158 additions and 24348 deletions
--- a/.dockerignore
+++ b/.dockerignore
@ -1,19 +0,0 @@
 .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
--- a/.gitignore
+++ b/.gitignore
@ -1,18 +0,0 @@
 *.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
--- a/.gitmodules
+++ b/.gitmodules
@ -1,6 +0,0 @@
 [submodule "cpp-btree"]
 	path = cpp-btree
 	url = ../cpp-btree.git
 [submodule "json11"]
 	path = json11
 	url = ../json11.git
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@ -1,5 +0,0 @@
 cmake_minimum_required(VERSION 2.8)
 project(vitastor)
 add_subdirectory(src)
--- a/GPL-2.0.txt
+++ b/GPL-2.0.txt
@ -1,339 +0,0 @@
                    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.
--- a/27
+++ b/27
@ -1,27 +0,0 @@
 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).
--- a/64
+++ b/64
@ -0,0 +1,64 @@
 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 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
 	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
 	g++ $(CXXFLAGS) -c -o $@ $<
 osd_receive.o: osd_receive.cpp osd.h osd_ops.h
 	g++ $(CXXFLAGS) -c -o $@ $<
 osd_send.o: osd_send.cpp osd.h osd_ops.h
 	g++ $(CXXFLAGS) -c -o $@ $<
 osd_peering.o: osd_peering.cpp osd.h osd_ops.h osd_peering_pg.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
 	g++ $(CXXFLAGS) -c -o $@ $<
 osd.o: osd.cpp osd.h osd_ops.h osd_peering_pg.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
 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
--- a/README-ru.md
+++ b/README-ru.md
@ -1,491 +0,0 @@
 ## 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).
--- a/README.md
+++ b/README.md
@ -1,449 +0,0 @@
 ## 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).
--- a/VNPL-1.1.txt
+++ b/VNPL-1.1.txt
@ -1,648 +0,0 @@
                     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.
--- a/src/allocator.cpp
+++ b/src/allocator.cpp
@ -1,6 +1,3 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 (see README.md for details)
 #include <stdexcept>
 #include "allocator.h"
@ -13,19 +10,19 @@ allocator::allocator(uint64_t blocks)
    {
        throw std::invalid_argument("blocks");
    }
-    uint64_t p2 = 1;
+    uint64_t p2 = 1, total = 1;
    total = 0;
    while (p2 * 64 < blocks)
    {
        total += p2;
        p2 = p2 * 64;
        total += p2;
    }
    total -= p2;
    total += (blocks+63) / 64;
-    mask = new uint64_t[total];
+    mask = new uint64_t[2 + total];
    size = free = blocks;
    last_one_mask = (blocks % 64) == 0
        ? UINT64_MAX
-        : ((1l << (blocks % 64)) - 1);
+        : ~(UINT64_MAX << (64 - blocks % 64));
    for (uint64_t i = 0; i < total; i++)
    {
        mask[i] = 0;
@ -99,10 +96,6 @@ 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)
        {
@ -117,6 +110,11 @@ 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;
    }
@ -127,35 +125,3 @@ 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++;
        }
    }
 }
--- a/src/allocator.h
+++ b/src/allocator.h
@ -1,6 +1,3 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 (see README.md for details)
 #pragma once
 #include <stdint.h>
@ -8,7 +5,6 @@
 // Hierarchical bitmap allocator
 class allocator
 {
    uint64_t total;
    uint64_t size;
    uint64_t free;
    uint64_t last_one_mask;
@ -20,5 +16,3 @@ 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);
--- a/src/blockstore.cpp
+++ b/src/blockstore.cpp
@ -1,6 +1,3 @@
 // 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)
@ -35,7 +32,12 @@ bool blockstore_t::is_safe_to_stop()
 void blockstore_t::enqueue_op(blockstore_op_t *op)
 {
-    impl->enqueue_op(op);
+    impl->enqueue_op(op, false);
 }
 void blockstore_t::enqueue_op_first(blockstore_op_t *op)
 {
    impl->enqueue_op(op, true);
 }
 std::unordered_map<object_id, uint64_t> & blockstore_t::get_unstable_writes()
@ -43,11 +45,6 @@ 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();
@ -58,12 +55,7 @@ uint64_t blockstore_t::get_block_count()
    return impl->get_block_count();
 }
-uint64_t blockstore_t::get_free_block_count()
+uint32_t blockstore_t::get_disk_alignment()
 {
-    return impl->get_free_block_count();
+    return impl->get_disk_alignment();
 }
 uint32_t blockstore_t::get_bitmap_granularity()
 {
    return impl->get_bitmap_granularity();
 }
--- a/src/blockstore.h
+++ b/src/blockstore.h
@ -1,6 +1,3 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 (see README.md for details)
 #pragma once
 #ifndef _LARGEFILE64_SOURCE
@ -9,7 +6,6 @@
 #include <stdint.h>
 #include <string>
 #include <map>
 #include <unordered_map>
 #include <functional>
@ -19,27 +15,23 @@
 // 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_WRITE_STABLE 3
+#define BS_OP_SYNC 3
-#define BS_OP_SYNC 4
+#define BS_OP_STABLE 4
-#define BS_OP_STABLE 5
+#define BS_OP_DELETE 5
-#define BS_OP_DELETE 6
+#define BS_OP_LIST 6
-#define BS_OP_LIST 7
+#define BS_OP_ROLLBACK 7
-#define BS_OP_ROLLBACK 8
+#define BS_OP_SYNC_STAB_ALL 8
-#define BS_OP_SYNC_STAB_ALL 9
+#define BS_OP_MAX 8
 #define BS_OP_MAX 9
 #define BS_OP_PRIVATE_DATA_SIZE 256
@ -47,9 +39,9 @@
 Blockstore opcode documentation:
-## BS_OP_READ / BS_OP_WRITE / BS_OP_WRITE_STABLE
+## BS_OP_READ / BS_OP_WRITE
-Read or write object data. WRITE_STABLE writes a version that doesn't require marking as stable.
+Read or write object data.
 Input:
 - oid = requested object
@ -58,15 +50,12 @@ 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)
@ -103,7 +92,7 @@ Input:
 - buf = pre-allocated obj_ver_id array <len> units long
 Output:
- retval = 0 or negative error number (-EINVAL, -ENOENT if no such version or -EBUSY if not synced)
+- retval = 0 or negative error number (-EINVAL)
 ## BS_OP_SYNC_STAB_ALL
@ -121,8 +110,6 @@ 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
@ -144,7 +131,6 @@ 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];
@ -179,16 +165,16 @@ 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_bitmap_granularity();
+    uint32_t get_disk_alignment();
 };
--- a/src/blockstore_flush.cpp
+++ b/src/blockstore_flush.cpp
@ -1,24 +1,14 @@
 // 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;
-    syncing_flushers = 0;
+    sync_threshold = flusher_count == 1 ? 1 : flusher_count/2;
-    // FIXME: allow to configure flusher_start_threshold and journal_trim_interval
+    journal_trim_interval = sync_threshold;
    flusher_start_threshold = bs->journal_block_size / sizeof(journal_entry_stable);
    journal_trim_interval = 512;
    journal_trim_counter = 0;
-    trim_wanted = 0;
+    journal_superblock = bs->journal.inmemory ? bs->journal.buffer : memalign(MEM_ALIGNMENT, bs->journal_block_size);
    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++)
    {
@ -65,36 +55,23 @@ journal_flusher_t::~journal_flusher_t()
 bool journal_flusher_t::is_active()
 {
-    return active_flushers > 0 || dequeuing;
+    return active_flushers > 0 || start_forced && flush_queue.size() > 0 || flush_queue.size() >= sync_threshold;
 }
 void journal_flusher_t::loop()
 {
-    target_flusher_count = bs->write_iodepth*2;
+    for (int i = 0; i < flusher_count; i++)
    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)
    {
-        while (target_flusher_count < cur_flusher_count)
+        if (!active_flushers && (start_forced ? !flush_queue.size() : (flush_queue.size() < sync_threshold)))
        {
-            if (co[cur_flusher_count-1].wait_state)
+            return;
                break;
            cur_flusher_count--;
        }
    }
    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())
    {
@ -106,18 +83,10 @@ 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, bool force)
+void journal_flusher_t::unshift_flush(obj_ver_id ov)
 {
 #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())
    {
@ -127,60 +96,16 @@ void journal_flusher_t::unshift_flush(obj_ver_id ov, bool force)
    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::remove_flush(object_id oid)
+void journal_flusher_t::force_start()
 {
-#ifdef BLOCKSTORE_DEBUG
+    start_forced = true;
    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();\
@ -191,7 +116,6 @@ void journal_flusher_t::release_trim()
        }\
        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
@ -230,23 +154,11 @@ 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->dequeuing)
+    if (!flusher->flush_queue.size() ||
        !flusher->start_forced && !flusher->active_flushers && flusher->flush_queue.size() < flusher->sync_threshold)
    {
-stop_flusher:
+        flusher->start_forced = false;
        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;
    }
@ -261,7 +173,7 @@ stop_flusher:
        if (repeat_it != flusher->sync_to_repeat.end())
        {
 #ifdef BLOCKSTORE_DEBUG
-            printf("Postpone %lx:%lx v%lu\n", cur.oid.inode, cur.oid.stripe, cur.version);
+            printf("Postpone %lu:%lu 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
@ -274,110 +186,42 @@ stop_flusher:
        }
        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("Flusher overran writers (dirty_start=%08lx) - searching for older flushes (%d left)\n", bs->journal.dirty_start, search_left);
+        printf("Flushing %lu:%lu v%lu\n", cur.oid.inode, cur.oid.stripe, cur.version);
 #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;
        }
-        // Writes and deletes shouldn't happen at the same time
+        if (copy_count == 0 && clean_loc == UINT64_MAX && !has_delete && !has_empty)
        assert(!has_writes || !has_delete);
        if (!has_writes && !has_delete || has_delete && old_clean_loc == UINT64_MAX)
        {
            // Nothing to flush
-            bs->erase_dirty(dirty_start, std::next(dirty_end), clean_loc);
+            flusher->active_flushers--;
-            goto 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 });
            }
            flusher->sync_to_repeat.erase(repeat_it);
            wait_state = 0;
            goto resume_0;
        }
        // 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 (old_clean_loc == UINT64_MAX)
+            if (copy_count > 0 && has_delete || old_clean_loc == UINT64_MAX)
            {
                // Object not allocated. This is a bug.
                char err[1024];
                snprintf(
-                    err, 1024, "BUG: Object %lx:%lx v%lu that we are trying to flush is not allocated on the data device",
+                    err, 1024, "BUG: Object %lu:%lu 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);
@ -426,18 +270,18 @@ 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)*(2*bs->clean_entry_bitmap_size));
+                : bs->clean_bitmap + (clean_loc >> bs->block_order)*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, bs->bitmap_granularity);
+                bitmap_set(new_clean_bitmap, clean_bitmap_offset, clean_bitmap_len);
            }
        }
        for (it = v.begin(); it != v.end(); it++)
        {
            if (new_clean_bitmap)
            {
-                bitmap_set(new_clean_bitmap, it->offset, it->len, bs->bitmap_granularity);
+                bitmap_set(new_clean_bitmap, it->offset, it->len);
            }
            await_sqe(4);
            data->iov = (struct iovec){ it->buf, (size_t)it->len };
@ -471,7 +315,6 @@ 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 };
@ -483,30 +326,17 @@ 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 };
@ -556,35 +386,13 @@ 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) || flusher->trim_wanted > 0)
+        if (!((++flusher->journal_trim_counter) % flusher->journal_trim_interval))
        {
            flusher->journal_trim_counter = 0;
-            new_trim_pos = bs->journal.get_trim_pos();
+            if (bs->journal.trim())
            if (new_trim_pos != bs->journal.used_start)
            {
-            resume_19:
+                // Update journal "superblock"
                // 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,
@ -592,7 +400,7 @@ resume_1:
                    .type = JE_START,
                    .size = sizeof(journal_entry_start),
                    .reserved = 0,
-                    .journal_start = new_trim_pos,
+                    .journal_start = bs->journal.used_start,
                };
                ((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 };
@ -605,28 +413,20 @@ 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;
    }
@ -645,25 +445,19 @@ bool journal_flusher_co::scan_dirty(int wait_base)
    copy_count = 0;
    clean_loc = UINT64_MAX;
    has_delete = false;
-    has_writes = false;
+    has_empty = false;
    skip_copy = false;
    clean_init_bitmap = false;
    while (1)
    {
-        if (!IS_STABLE(dirty_it->second.state))
+        if (dirty_it->second.state == ST_J_STABLE && !skip_copy)
        {
            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
-            has_writes = true;
+            if (dirty_it->second.len == 0)
-            if (dirty_it->second.len != 0)
+            {
                has_empty = true;
            }
            else
            {
                offset = dirty_it->second.offset;
                end_offset = dirty_it->second.offset + dirty_it->second.len;
@ -677,18 +471,18 @@ 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_or_die(MEM_ALIGNMENT, submit_len) });
+                        it = v.insert(it, (copy_buffer_t){ .offset = offset, .len = submit_len, .buf = memalign(MEM_ALIGNMENT, submit_len) });
                        copy_count++;
                        if (bs->journal.inmemory)
                        {
                            // Take it from memory
-                            memcpy(it->buf, bs->journal.buffer + submit_offset, submit_len);
+                            memcpy(v.back().buf, bs->journal.buffer + submit_offset, submit_len);
                        }
                        else
                        {
                            // Read it from disk
                            await_sqe(0);
-                            data->iov = (struct iovec){ it->buf, (size_t)submit_len };
+                            data->iov = (struct iovec){ v.back().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
@ -702,22 +496,30 @@ bool journal_flusher_co::scan_dirty(int wait_base)
                }
            }
        }
-        else if (IS_BIG_WRITE(dirty_it->second.state) && !skip_copy)
+        else if (dirty_it->second.state == ST_D_STABLE && !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 (IS_DELETE(dirty_it->second.state) && !skip_copy)
+        else if (dirty_it->second.state == ST_DEL_STABLE && !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())
        {
@ -753,7 +555,7 @@ 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_or_die(MEM_ALIGNMENT, bs->meta_block_size);
+        wr.buf = memalign(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,
@ -783,7 +585,7 @@ 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 (new location is %lu)\n", old_clean_loc >> bs->block_order, clean_loc >> bs->block_order);
+        printf("Free block %lu\n", old_clean_loc >> bs->block_order);
 #endif
        bs->data_alloc->set(old_clean_loc >> bs->block_order, false);
    }
@ -830,37 +632,32 @@ bool journal_flusher_co::fsync_batch(bool fsync_meta, int wait_base)
        });
    sync_found:
        cur_sync->ready_count++;
-        flusher->syncing_flushers++;
+        if (cur_sync->ready_count >= flusher->sync_threshold || !flusher->flush_queue.size())
    resume_1:
        if (!cur_sync->state)
        {
-            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_1:
            if (wait_count > 0)
            {
                // 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)
        {
@ -869,3 +666,35 @@ 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++;
        }
    }
 }
--- a/src/blockstore_flush.h
+++ b/src/blockstore_flush.h
@ -1,6 +1,3 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 (see README.md for details)
 struct copy_buffer_t
 {
    uint64_t offset, len;
@ -48,8 +45,8 @@ 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_writes;
+    bool skip_copy, has_delete, has_empty;
-    blockstore_clean_db_t::iterator clean_it;
+    spp::sparse_hash_map<object_id, clean_entry>::iterator clean_it;
    std::vector<copy_buffer_t> v;
    std::vector<copy_buffer_t>::iterator it;
    int copy_count;
@ -59,8 +56,6 @@ 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;
@ -69,6 +64,7 @@ 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();
@ -77,20 +73,17 @@ public:
 // Journal flusher itself
 class journal_flusher_t
 {
-    int trim_wanted = 0;
+    bool start_forced = false;
-    bool dequeuing;
+    int flusher_count;
-    int flusher_count, cur_flusher_count, target_flusher_count;
+    int sync_threshold;
    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;
@ -102,10 +95,7 @@ public:
    ~journal_flusher_t();
    void loop();
    bool is_active();
-    void mark_trim_possible();
+    void force_start();
    void request_trim();
    void release_trim();
    void enqueue_flush(obj_ver_id oid);
-    void unshift_flush(obj_ver_id oid, bool force);
+    void unshift_flush(obj_ver_id oid);
    void remove_flush(object_id oid);
 };
--- a/blockstore_impl.cpp
+++ b/blockstore_impl.cpp
@ -0,0 +1,458 @@
 #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);
 }
--- a/src/blockstore_impl.h
+++ b/src/blockstore_impl.h
@ -1,15 +1,14 @@
 // 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>
@ -17,41 +16,43 @@
 #include <deque>
 #include <new>
-#include "cpp-btree/btree_map.h"
+#include "sparsepp/sparsepp/spp.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 BS_ST_SMALL_WRITE 0x01
+#define ST_J_IN_FLIGHT 1
-#define BS_ST_BIG_WRITE 0x02
+#define ST_J_SUBMITTED 2
-#define BS_ST_DELETE 0x03
+#define ST_J_WRITTEN 3
 #define ST_J_SYNCED 4
 #define ST_J_STABLE 5
-#define BS_ST_WAIT_DEL 0x10
+#define ST_D_IN_FLIGHT 15
-#define BS_ST_WAIT_BIG 0x20
+#define ST_D_SUBMITTED 16
-#define BS_ST_IN_FLIGHT 0x30
+#define ST_D_WRITTEN 17
-#define BS_ST_SUBMITTED 0x40
+#define ST_D_META_WRITTEN 19
-#define BS_ST_WRITTEN 0x50
+#define ST_D_META_SYNCED 20
-#define BS_ST_SYNCED 0x60
+#define ST_D_STABLE 21
 #define BS_ST_STABLE 0x70
-#define BS_ST_INSTANT 0x100
+#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 IMMEDIATE_NONE 0
+#define ST_CURRENT 48
 #define IMMEDIATE_SMALL 1
 #define IMMEDIATE_ALL 2
-#define BS_ST_TYPE_MASK 0x0F
+#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 BS_ST_WORKFLOW_MASK 0xF0
+#define IS_STABLE(st) (st == ST_J_STABLE || st == ST_D_STABLE || st == ST_DEL_STABLE || st == ST_CURRENT)
-#define IS_IN_FLIGHT(st) (((st) & 0xF0) <= BS_ST_SUBMITTED)
+#define IS_SYNCED(st) (IS_STABLE(st) || st == ST_J_SYNCED || st == ST_D_META_SYNCED || st == ST_DEL_SYNCED)
-#define IS_STABLE(st) (((st) & 0xF0) == BS_ST_STABLE)
+#define IS_JOURNAL(st) (st >= ST_J_SUBMITTED && st <= ST_J_STABLE)
-#define IS_SYNCED(st) (((st) & 0xF0) >= BS_ST_SYNCED)
+#define IS_BIG_WRITE(st) (st >= ST_D_SUBMITTED && st <= ST_D_STABLE)
-#define IS_JOURNAL(st) (((st) & 0x0F) == BS_ST_SMALL_WRITE)
+#define IS_DELETE(st) (st >= ST_DEL_SUBMITTED && st <= ST_DEL_STABLE)
-#define IS_BIG_WRITE(st) (((st) & 0x0F) == BS_ST_BIG_WRITE)
+#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 IS_DELETE(st) (((st) & 0x0F) == BS_ST_DELETE)
 #define BS_SUBMIT_GET_SQE(sqe, data) \
    BS_SUBMIT_GET_ONLY_SQE(sqe); \
@ -77,8 +78,7 @@
 #include "blockstore_journal.h"
-// 32 bytes = 24 bytes + block bitmap (4 bytes by default) + external attributes (also bitmap, 4 bytes by default)
+// 24 bytes + block bitmap per "clean" entry on disk with fixed metadata tables
 // 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;
 };
-// 64 = 24 + 40 bytes per dirty entry in memory (obj_ver_id => dirty_entry)
+// 56 = 24 + 32 bytes per dirty entry in memory (obj_ver_id => dirty_entry)
 struct __attribute__((__packed__)) dirty_entry
 {
    uint32_t state;
@ -103,7 +103,6 @@ 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!)
@ -125,6 +124,8 @@ 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
@ -138,7 +139,7 @@ struct fulfill_read_t
 };
 #define PRIV(op) ((blockstore_op_private_t*)(op)->private_data)
-#define FINISH_OP(op) PRIV(op)->~blockstore_op_private_t(); std::function<void (blockstore_op_t*)>(op->callback)(op)
+#define FINISH_OP(op) PRIV(op)->~blockstore_op_private_t(); op->callback(op)
 struct blockstore_op_private_t
 {
@ -146,28 +147,23 @@ 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_flushed_journal_sector, max_flushed_journal_sector;
+    uint64_t min_used_journal_sector, max_used_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"
@ -181,37 +177,34 @@ class blockstore_impl_t
    uint32_t block_size;
    uint64_t meta_offset;
    uint64_t data_offset;
-    uint64_t cfg_journal_size, cfg_data_size;
+    uint64_t cfg_journal_size;
    // Required write alignment and journal/metadata/data areas' location alignment
-    uint32_t disk_alignment = 4096;
+    uint32_t disk_alignment = 512;
    // Journal block size - minimum_io_size of the journal device is the best choice
-    uint64_t journal_block_size = 4096;
+    uint64_t journal_block_size = 512;
    // Metadata block size - minimum_io_size of the metadata device is the best choice
-    uint64_t meta_block_size = 4096;
+    uint64_t meta_block_size = 512;
    // 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"
-    // Suitable only for server SSDs with capacitors, requires disabled data and journal fsyncs
+    // FIXME Not implemented yet
-    int immediate_commit = IMMEDIATE_NONE;
+    bool immediate_commit = false;
    bool inmemory_meta = false;
-    // Maximum flusher count
+    int flusher_count;
    unsigned flusher_count;
    // Maximum queue depth
    unsigned max_write_iodepth = 128;
    /******* END OF OPTIONS *******/
    struct ring_consumer_t ring_consumer;
-    blockstore_clean_db_t clean_db;
+    // Another option is https://github.com/algorithm-ninja/cpp-btree
    spp::sparse_hash_map<object_id, clean_entry> clean_db;
    uint8_t *clean_bitmap = NULL;
    blockstore_dirty_db_t dirty_db;
-    std::vector<blockstore_op_t*> submit_queue;
+    std::list<blockstore_op_t*> submit_queue; // FIXME: funny thing is that vector is better here
    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;
@ -228,7 +221,6 @@ 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;
@ -251,7 +243,6 @@ 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;
@ -271,29 +262,26 @@ 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);
-    int continue_write(blockstore_op_t *op);
+    void ack_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 continue_sync(blockstore_op_t *op, bool queue_has_in_progress_sync);
+    int dequeue_sync(blockstore_op_t *op);
    void handle_sync_event(ring_data_t *data, blockstore_op_t *op);
-    void ack_sync(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);
    // 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);
@ -321,16 +309,12 @@ public:
    bool is_stalled();
    // Submission
-    void enqueue_op(blockstore_op_t *op);
+    void enqueue_op(blockstore_op_t *op, bool first = false);
    // 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 uint64_t get_free_block_count() { return data_alloc->get_free_count(); }
+    inline uint32_t get_disk_alignment() { return disk_alignment; }
    inline uint32_t get_bitmap_granularity() { return disk_alignment; }
 };
--- a/src/blockstore_init.cpp
+++ b/src/blockstore_init.cpp
@ -1,6 +1,3 @@
 // 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)
@ -100,7 +97,7 @@ 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)*2*bs->clean_entry_bitmap_size, &entry->bitmap, 2*bs->clean_entry_bitmap_size);
+            memcpy(bs->clean_bitmap + (done_cnt+i)*bs->clean_entry_bitmap_size, &entry->bitmap, bs->clean_entry_bitmap_size);
        }
        if (entry->oid.inode > 0)
        {
@ -111,17 +108,13 @@ void blockstore_init_meta::handle_entries(void* entries, unsigned count, int blo
                {
                    // free the previous block
 #ifdef BLOCKSTORE_DEBUG
-                    printf("Free block %lu (new location is %lu)\n", clean_it->second.location >> block_order, done_cnt+i);
+                    printf("Free block %lu\n", clean_it->second.location >> bs->block_order);
 #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: %lx:%lx v%lu\n", done_cnt+i, entry->oid.inode, entry->oid.stripe, entry->version);
+                printf("Allocate block (clean entry) %lu: %lu:%lu 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){
@ -132,7 +125,7 @@ void blockstore_init_meta::handle_entries(void* entries, unsigned count, int blo
            else
            {
 #ifdef BLOCKSTORE_DEBUG
-                printf("Old clean entry %lu: %lx:%lx v%lu\n", done_cnt+i, entry->oid.inode, entry->oid.stripe, entry->version);
+                printf("Old clean entry %lu: %lu:%lu v%lu\n", done_cnt+i, entry->oid.inode, entry->oid.stripe, entry->version);
 #endif
            }
        }
@ -209,7 +202,11 @@ int blockstore_init_journal::loop()
        goto resume_7;
    printf("Reading blockstore journal\n");
    if (!bs->journal.inmemory)
-        submitted_buf = memalign_or_die(MEM_ALIGNMENT, 2*bs->journal.block_size);
+    {
        submitted_buf = memalign(MEM_ALIGNMENT, 2*bs->journal.block_size);
        if (!submitted_buf)
            throw std::bad_alloc();
    }
    else
        submitted_buf = bs->journal.buffer;
    // Read first block of the journal
@ -320,7 +317,7 @@ resume_1:
                if (journal_pos < bs->journal.used_start)
                    end = bs->journal.used_start;
                if (!bs->journal.inmemory)
-                    submitted_buf = memalign_or_die(MEM_ALIGNMENT, JOURNAL_BUFFER_SIZE);
+                    submitted_buf = memalign(MEM_ALIGNMENT, JOURNAL_BUFFER_SIZE);
                else
                    submitted_buf = bs->journal.buffer + journal_pos;
                data->iov = {
@ -403,10 +400,10 @@ resume_1:
            }
        }
    }
-    bs->flusher->mark_trim_possible();
+    // Trim journal on start so we don't stall when all entries are older
-    bs->journal.dirty_start = bs->journal.next_free;
+    bs->journal.trim();
    printf(
-        "Journal entries loaded: %lu, free journal space: %lu bytes (%08lx..%08lx is used), free blocks: %lu / %lu\n",
+        "Journal entries loaded: %lu, free journal space: %lu bytes (%lu..%lu 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)
@ -442,7 +439,7 @@ 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_MIN || je->type > JE_MAX || started && je->crc32_prev != crc32_last)
+                je->type < JE_SMALL_WRITE || je->type > JE_DELETE || started && je->crc32_prev != crc32_last)
            {
                if (pos == 0)
                {
@ -456,15 +453,10 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
                    break;
                }
            }
-            if (je->type == JE_SMALL_WRITE || je->type == JE_SMALL_WRITE_INSTANT)
+            if (je->type == JE_SMALL_WRITE)
            {
 #ifdef BLOCKSTORE_DEBUG
-                printf(
+                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);
                    "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;
@ -482,7 +474,7 @@ 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 (%08lx) != stored journal data offset (%08lx)", location, je->small_write.data_offset);
+                    snprintf(err, 1024, "BUG: calculated journal data offset (%lu) != stored journal data offset (%lu)", location, je->small_write.data_offset);
                    throw std::runtime_error(err);
                }
                uint32_t data_crc32 = 0;
@ -517,9 +509,7 @@ 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;
@ -528,105 +518,33 @@ 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->small_write.version)
+                    clean_it->second.version < je->big_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 = (BS_ST_SMALL_WRITE | BS_ST_SYNCED),
+                        .state = ST_J_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(
+                    printf("journal offset %lu is used by %lu:%lu v%lu\n", proc_pos, ov.oid.inode, ov.oid.stripe, ov.version);
                        "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 || je->type == JE_BIG_WRITE_INSTANT)
+            else if (je->type == JE_BIG_WRITE)
            {
 #ifdef BLOCKSTORE_DEBUG
-                printf(
+                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);
                    "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)
@ -636,100 +554,131 @@ 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 = (BS_ST_BIG_WRITE | BS_ST_SYNCED),
+                        .state = ST_D_META_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=%lx:%lx ver=%lu\n", je->stable.oid.inode, je->stable.oid.stripe, je->stable.version);
+                printf("je_stable oid=%lu:%lu 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,
                };
-                bs->mark_stable(ov);
+                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);
                }
            }
            else if (je->type == JE_ROLLBACK)
            {
 #ifdef BLOCKSTORE_DEBUG
-                printf("je_rollback oid=%lx:%lx ver=%lu\n", je->rollback.oid.inode, je->rollback.oid.stripe, je->rollback.version);
+                printf("je_rollback oid=%lu:%lu ver=%lu\n", je->rollback.oid.inode, je->rollback.oid.stripe, je->rollback.version);
 #endif
                // rollback dirty writes of <oid> up to <version>
-                obj_ver_id ov = {
+                auto it = bs->dirty_db.lower_bound((obj_ver_id){
                    .oid = je->rollback.oid,
-                    .version = je->rollback.version,
+                    .version = UINT64_MAX,
-                };
+                });
-                bs->mark_rolled_back(ov);
+                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;
                    }
                }
            }
            else if (je->type == JE_DELETE)
            {
 #ifdef BLOCKSTORE_DEBUG
-                printf("je_delete oid=%lx:%lx ver=%lu\n", je->del.oid.inode, je->del.oid.stripe, je->del.version);
+                printf("je_delete oid=%lu:%lu ver=%lu\n", je->del.oid.inode, je->del.oid.stripe, je->del.version);
 #endif
-                auto clean_it = bs->clean_db.find(je->del.oid);
+                // oid, version
-                if (clean_it != bs->clean_db.end() &&
+                obj_ver_id ov = {
-                    clean_it->second.version < je->del.version)
+                    .oid = je->del.oid,
-                {
+                    .version = je->del.version,
-                    // oid, version
+                };
-                    obj_ver_id ov = {
+                bs->dirty_db.emplace(ov, (dirty_entry){
-                        .oid = je->del.oid,
+                    .state = ST_DEL_SYNCED,
-                        .version = je->del.version,
+                    .flags = 0,
-                    };
+                    .location = 0,
-                    bs->dirty_db.emplace(ov, (dirty_entry){
+                    .offset = 0,
-                        .state = (BS_ST_DELETE | BS_ST_SYNCED),
+                    .len = 0,
-                        .flags = 0,
+                    .journal_sector = proc_pos,
-                        .location = 0,
+                });
-                        .offset = 0,
+                bs->journal.used_sectors[proc_pos]++;
                        .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;
--- a/src/blockstore_init.h
+++ b/src/blockstore_init.h
@ -1,6 +1,3 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 (see README.md for details)
 #pragma once
 class blockstore_init_meta
--- a/src/blockstore_journal.cpp
+++ b/src/blockstore_journal.cpp
@ -1,45 +1,31 @@
 // 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_to_write = 0;
+    sectors_required = 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 entries_required, int size, int data_after)
+int blockstore_journal_check_t::check_available(blockstore_op_t *op, int required, int size, int data_after)
 {
    int required = entries_required;
    while (1)
    {
-        int fits = bs->journal.no_same_sector_overwrites && next_pos == bs->journal.next_free && bs->journal.sector_info[next_sector].written
+        int fits = (bs->journal.block_size - next_in_pos) / size;
            ? 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_to_write++;
+            sectors_required++;
        }
        else if (bs->journal.sector_info[next_sector].dirty)
        {
-            sectors_to_write++;
+            // sectors_required is more like "sectors to write"
            sectors_required++;
        }
        if (required <= 0)
        {
@ -52,41 +38,19 @@ int blockstore_journal_check_t::check_available(blockstore_op_t *op, int entries
            right_dir = false;
        }
        next_in_pos = 0;
-        next_sector = ((next_sector + 1) % bs->journal.sector_count);
+        if (bs->journal.sector_info[next_sector].usage_count > 0 ||
-        if (next_sector == first_sector)
+            bs->journal.sector_info[next_sector].dirty)
        {
-            // next_sector may wrap when all sectors are flushed and the incoming batch is too big
+            next_sector = ((next_sector + 1) % bs->journal.sector_count);
            // 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].flush_count > 0 ||
+        if (bs->journal.sector_info[next_sector].usage_count > 0 ||
            bs->journal.sector_info[next_sector].dirty)
        {
            // No memory buffer available. Wait for it.
-            int used = 0, dirty = 0;
+#ifdef BLOCKSTORE_DEBUG
-            for (int i = 0; i < bs->journal.sector_count; i++)
+            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);
-                if (bs->journal.sector_info[i].dirty)
+#endif
                {
                    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;
        }
@ -104,11 +68,13 @@ int blockstore_journal_check_t::check_available(blockstore_op_t *op, int entries
    {
        // No space in the journal. Wait until used_start changes.
        printf(
-            "Ran out of journal space (used_start=%08lx, next_free=%08lx, dirty_start=%08lx)\n",
+            "Ran out of journal space (free space: %lu bytes)\n",
-            bs->journal.used_start, bs->journal.next_free, bs->journal.dirty_start
+            (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)
        );
        PRIV(op)->wait_for = WAIT_JOURNAL;
-        bs->flusher->request_trim();
+        bs->flusher->force_start();
        PRIV(op)->wait_detail = bs->journal.used_start;
        return 0;
    }
@ -117,21 +83,15 @@ int blockstore_journal_check_t::check_available(blockstore_op_t *op, int entries
 journal_entry* prefill_single_journal_entry(journal_t & journal, uint16_t type, uint32_t size)
 {
-    if (!journal.entry_fits(size))
+    if (journal.block_size - journal.in_sector_pos < size)
    {
        assert(!journal.sector_info[journal.cur_sector].dirty);
        // Move to the next journal sector
-        if (journal.sector_info[journal.cur_sector].flush_count > 0)
+        if (journal.sector_info[journal.cur_sector].usage_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;
@ -156,8 +116,7 @@ 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].written = true;
+    journal.sector_info[cur_sector].usage_count++;
    journal.sector_info[cur_sector].flush_count++;
    ring_data_t *data = ((ring_data_t*)sqe->user_data);
    data->iov = (struct iovec){
        (journal.inmemory
@ -184,13 +143,13 @@ journal_t::~journal_t()
    buffer = NULL;
 }
-uint64_t journal_t::get_trim_pos()
+bool journal_t::trim()
 {
    auto journal_used_it = used_sectors.lower_bound(used_start);
 #ifdef BLOCKSTORE_DEBUG
    printf(
-        "Trimming journal (used_start=%08lx, next_free=%08lx, dirty_start=%08lx, new_start=%08lx, new_refcount=%ld)\n",
+        "Trimming journal (used_start=%lu, next_free=%lu, first_used=%lu, usage_count=%lu)\n",
-        used_start, next_free, dirty_start,
+        used_start, next_free,
        journal_used_it == used_sectors.end() ? 0 : journal_used_it->first,
        journal_used_it == used_sectors.end() ? 0 : journal_used_it->second
    );
@ -202,19 +161,26 @@ uint64_t journal_t::get_trim_pos()
        if (journal_used_it == used_sectors.end())
        {
            // Journal is empty
-            return next_free;
+            used_start = next_free;
        }
        else
        {
-            // next_free does not need updating during trim
+            used_start = journal_used_it->first;
-            return journal_used_it->first;
+            // next_free does not need updating here
        }
    }
    else if (journal_used_it->first > used_start)
    {
        // Journal is cleared up to <journal_used_it>
-        return journal_used_it->first;
+        used_start = journal_used_it->first;
    }
-    // Can't trim journal
+    else
-    return used_start;
+    {
        // 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;
 }
--- a/src/blockstore_journal.h
+++ b/src/blockstore_journal.h
@ -1,6 +1,3 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 (see README.md for details)
 #pragma once
 #include "crc32c.h"
@ -10,23 +7,17 @@
 #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
@ -54,9 +45,6 @@ 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
@ -71,9 +59,6 @@ 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
@ -139,8 +124,7 @@ inline uint32_t je_crc32(journal_entry *je)
 struct journal_sector_info_t
 {
    uint64_t offset;
-    uint64_t flush_count;
+    uint64_t usage_count;
    bool written;
    bool dirty;
 };
@ -151,21 +135,16 @@ struct journal_t
    bool inmemory = false;
    void *buffer = NULL;
-    uint64_t block_size;
+    uint64_t block_size = 512;
    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;
@ -175,19 +154,13 @@ 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_to_write, first_sector;
+    int sectors_required;
    bool right_dir; // writing to the end or the beginning of the ring buffer
    blockstore_journal_check_t(blockstore_impl_t *bs);
--- a/src/blockstore_open.cpp
+++ b/src/blockstore_open.cpp
@ -1,7 +1,3 @@
 // 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)
@ -38,23 +34,10 @@ 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);
@ -62,15 +45,12 @@ 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)
    {
@ -84,21 +64,17 @@ 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 = 4096;
+        disk_alignment = 512;
    }
    else if (disk_alignment % MEM_ALIGNMENT)
    {
-        throw std::runtime_error("disk_alignment must be a multiple of "+std::to_string(MEM_ALIGNMENT));
+        throw std::runtime_error("disk_alingment must be a multiple of "+std::to_string(MEM_ALIGNMENT));
    }
    if (!journal_block_size)
    {
-        journal_block_size = 4096;
+        journal_block_size = 512;
    }
    else if (journal_block_size % MEM_ALIGNMENT)
    {
@ -106,7 +82,7 @@ void blockstore_impl_t::parse_config(blockstore_config_t & config)
    }
    if (!meta_block_size)
    {
-        meta_block_size = 4096;
+        meta_block_size = 512;
    }
    else if (meta_block_size % MEM_ALIGNMENT)
    {
@ -118,7 +94,7 @@ void blockstore_impl_t::parse_config(blockstore_config_t & config)
    }
    if (!bitmap_granularity)
    {
-        bitmap_granularity = DEFAULT_BITMAP_GRANULARITY;
+        bitmap_granularity = 4096;
    }
    else if (bitmap_granularity % disk_alignment)
    {
@ -152,25 +128,9 @@ 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) + 2*clean_entry_bitmap_size;
+    clean_entry_size = sizeof(clean_disk_entry) + clean_entry_bitmap_size;
    journal.block_size = journal_block_size;
    journal.next_free = journal_block_size;
    journal.used_start = journal_block_size;
@ -191,15 +151,6 @@ 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)
@ -237,7 +188,7 @@ void blockstore_impl_t::calc_lengths()
    }
    else if (clean_entry_bitmap_size)
    {
-        clean_bitmap = (uint8_t*)malloc(block_count * 2*clean_entry_bitmap_size);
+        clean_bitmap = (uint8_t*)malloc(block_count * clean_entry_bitmap_size);
        if (!clean_bitmap)
            throw std::runtime_error("Failed to allocate memory for the metadata sparse write bitmap");
    }
@ -301,10 +252,6 @@ 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()
@ -322,14 +269,11 @@ 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)
        {
@ -347,15 +291,12 @@ void blockstore_impl_t::open_journal()
        {
            throw std::runtime_error("Failed to open journal device");
        }
-        check_size(journal.fd, &journal.device_size, "journal device");
+        check_size(journal.fd, &journal.device_size, "metadata 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)
        {
--- a/src/blockstore_read.cpp
+++ b/src/blockstore_read.cpp
@ -1,6 +1,3 @@
 // 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,
@ -11,10 +8,12 @@ int blockstore_impl_t::fulfill_read_push(blockstore_op_t *op, void *buf, uint64_
        // Zero-length version - skip
        return 1;
    }
-    else if (IS_IN_FLIGHT(item_state))
+    if (IS_IN_FLIGHT(item_state))
    {
-        // Write not finished yet - skip
+        // Pause until it's written somewhere
-        return 1;
+        PRIV(op)->wait_for = WAIT_IN_FLIGHT;
        PRIV(op)->wait_detail = item_version;
        return 0;
    }
    else if (IS_DELETE(item_state))
    {
@ -40,7 +39,6 @@ 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)
 {
@ -53,20 +51,8 @@ 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 = {
@ -85,30 +71,12 @@ 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);
@ -127,7 +95,7 @@ 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 2;
+        return 1;
    }
    uint64_t fulfilled = 0;
    PRIV(read_op)->pending_ops = 0;
@ -149,11 +117,6 @@ 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)))
@ -170,61 +133,63 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
            dirty_it--;
        }
    }
-    if (clean_it != clean_db.end())
+    if (clean_it != clean_db.end() && fulfilled < read_op->len)
    {
        if (!result_version)
        {
            result_version = clean_it->second.version;
-            if (read_op->bitmap)
+        }
        if (!clean_entry_bitmap_size)
        {
            if (!fulfill_read(read_op, fulfilled, 0, block_size, ST_CURRENT, 0, clean_it->second.location))
            {
-                void *bmp_ptr = get_clean_entry_bitmap(clean_it->second.location, clean_entry_bitmap_size);
+                // need to wait. undo added requests, don't dequeue op
-                memcpy(read_op->bitmap, bmp_ptr, clean_entry_bitmap_size);
+                PRIV(read_op)->read_vec.clear();
                return 0;
            }
        }
-        if (fulfilled < read_op->len)
+        else
        {
-            if (!clean_entry_bitmap_size)
+            uint64_t meta_loc = clean_it->second.location >> block_order;
            uint8_t *clean_entry_bitmap;
            if (inmemory_meta)
            {
-                if (!fulfill_read(read_op, fulfilled, 0, block_size, (BS_ST_BIG_WRITE | BS_ST_STABLE), 0, clean_it->second.location))
+                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));
-                    // need to wait. undo added requests, don't dequeue op
+                clean_entry_bitmap = (uint8_t*)(metadata_buffer + sector + pos*clean_entry_size + sizeof(clean_disk_entry));
                    PRIV(read_op)->read_vec.clear();
                    return 0;
                }
            }
            else
            {
-                uint8_t *clean_entry_bitmap = get_clean_entry_bitmap(clean_it->second.location, 0);
+                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)
+            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)
                {
-                    while (!(clean_entry_bitmap[bmp_end >> 3] & (1 << (bmp_end & 0x7))) && bmp_end < 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))
                    {
-                        bmp_end++;
+                        // need to wait. undo added requests, don't dequeue op
-                    }
+                        PRIV(read_op)->read_vec.clear();
-                    if (bmp_end > bmp_start)
+                        return 0;
                    {
                        // 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;
                    }
                }
            }
        }
@ -232,7 +197,7 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
    else if (fulfilled < read_op->len)
    {
        // fill remaining parts with zeroes
-        assert(fulfill_read(read_op, fulfilled, 0, block_size, (BS_ST_DELETE | BS_ST_STABLE), 0, 0));
+        fulfill_read(read_op, fulfilled, 0, block_size, ST_DEL_STABLE, 0, 0);
    }
    assert(fulfilled == read_op->len);
    read_op->version = result_version;
@ -246,10 +211,10 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
        }
        read_op->retval = read_op->len;
        FINISH_OP(read_op);
-        return 2;
+        return 1;
    }
    read_op->retval = 0;
-    return 2;
+    return 1;
 }
 void blockstore_impl_t::handle_read_event(ring_data_t *data, blockstore_op_t *op)
--- a/blockstore_rollback.cpp
+++ b/blockstore_rollback.cpp
@ -0,0 +1,187 @@
 #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);
 }
--- a/src/blockstore_stable.cpp
+++ b/src/blockstore_stable.cpp
@ -1,6 +1,3 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 (see README.md for details)
 #include "blockstore_impl.h"
 // Stabilize small write:
@ -43,10 +40,6 @@
 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++)
@ -58,26 +51,21 @@ 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 = -ENOENT;
+                op->retval = -EINVAL;
                FINISH_OP(op);
-                return 2;
+                return 1;
            }
            else
            {
                // Already stable
            }
        }
-        else if (IS_IN_FLIGHT(dirty_it->second.state))
+        else if (IS_UNSYNCED(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 = -EBUSY;
+            op->retval = EAGAIN;
            FINISH_OP(op);
-            return 2;
+            return 1;
        }
        else if (!IS_STABLE(dirty_it->second.state))
        {
@ -89,7 +77,7 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
        // Already stable
        op->retval = 0;
        FINISH_OP(op);
-        return 2;
+        return 1;
    }
    // Check journal space
    blockstore_journal_check_t space_check(this);
@ -98,126 +86,48 @@ 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_to_write];
+    struct io_uring_sqe *sqe[space_check.sectors_required];
-    for (i = 0; i < space_check.sectors_to_write; i++)
+    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_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++)
    {
-        // FIXME: Only stabilize versions that aren't stable yet
+        auto unstab_it = unstable_writes.find(v->oid);
-        if (!journal.entry_fits(sizeof(journal_entry_stable)) &&
+        if (unstab_it != unstable_writes.end() &&
-            journal.sector_info[journal.cur_sector].dirty)
+            unstab_it->second <= v->version)
        {
-            if (cur_sector == -1)
+            unstable_writes.erase(unstab_it);
                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)
    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)
+            if (cur_sector == -1)
-            {
+                PRIV(op)->min_used_journal_sector = 1 + journal.cur_sector;
-                dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | BS_ST_STABLE;
+            cur_sector = journal.cur_sector;
-                // Allocations and deletions are counted when they're stabilized
+            prepare_journal_sector_write(journal, cur_sector, sqe[s++], cb);
                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);
    }
    PRIV(op)->max_used_journal_sector = 1 + journal.cur_sector;
    PRIV(op)->pending_ops = s;
    return 1;
 }
 void blockstore_impl_t::handle_stable_event(ring_data_t *data, blockstore_op_t *op)
@ -233,7 +143,53 @@ void blockstore_impl_t::handle_stable_event(ring_data_t *data, blockstore_op_t *
    PRIV(op)->pending_ops--;
    if (PRIV(op)->pending_ops == 0)
    {
-        PRIV(op)->op_state++;
+        // Release used journal sectors
-        ringloop->wakeup();
+        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);
    }
 }
--- a/blockstore_sync.cpp
+++ b/blockstore_sync.cpp
@ -0,0 +1,269 @@
 #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);
 }
--- a/blockstore_write.cpp
+++ b/blockstore_write.cpp
@ -0,0 +1,328 @@
 #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;
 }
--- a/copy-fio-includes.sh
+++ b/copy-fio-includes.sh
@ -1,13 +0,0 @@
 #!/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
--- a/copy-qemu-includes.sh
+++ b/copy-qemu-includes.sh
@ -1,18 +0,0 @@
 #!/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
--- a/1
+++ b/1
@ -1 +0,0 @@
 Subproject commit 5dc108754ad40d3b1d024f9bd7cca0595ef1a1db
--- a/src/crc32c.c
+++ b/src/crc32c.c
--- a/src/crc32c.h
+++ b/src/crc32c.h
@ -8,10 +8,4 @@
 // 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
--- a/debian/build-vitastor-bullseye.sh
+++ b/debian/build-vitastor-bullseye.sh
@ -1,7 +0,0 @@
 #!/bin/bash
 sed 's/$REL/bullseye/g' < vitastor.Dockerfile > ../Dockerfile
 cd ..
 mkdir -p packages
 sudo podman build -v `pwd`/packages:/root/packages -f Dockerfile .
 rm Dockerfile
--- a/debian/build-vitastor-buster.sh
+++ b/debian/build-vitastor-buster.sh
@ -1,7 +0,0 @@
 #!/bin/bash
 sed 's/$REL/buster/g' < vitastor.Dockerfile > ../Dockerfile
 cd ..
 mkdir -p packages
 sudo podman build -v `pwd`/packages:/root/packages -f Dockerfile .
 rm Dockerfile
--- a/debian/changelog
+++ b/debian/changelog
@ -1,17 +0,0 @@
 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
--- a/debian/compat
+++ b/debian/compat
@ -1 +0,0 @@
 13
--- a/debian/control
+++ b/debian/control
@ -1,17 +0,0 @@
 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).
--- a/debian/copyright
+++ b/debian/copyright
@ -1,21 +0,0 @@
 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.
--- a/debian/install
+++ b/debian/install
@ -1,3 +0,0 @@
 VNPL-1.1.txt usr/share/doc/vitastor
 GPL-2.0.txt usr/share/doc/vitastor
 mon usr/lib/vitastor
--- a/debian/patched-qemu.Dockerfile
+++ b/debian/patched-qemu.Dockerfile
@ -1,44 +0,0 @@
 # 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-*/
--- a/debian/rules
+++ b/debian/rules
@ -1,9 +0,0 @@
 #!/usr/bin/make -f
 export DH_VERBOSE = 1
 %:
 	dh $@
 override_dh_installdeb:
 	cat debian/substvars >> debian/vitastor.substvars
 	dh_installdeb
--- a/debian/source/format
+++ b/debian/source/format
@ -1 +0,0 @@
 3.0 (quilt)
--- a/debian/substvars
+++ b/debian/substvars
@ -1,2 +0,0 @@
 dep:fio=3.16-1
 dep:qemu=1:5.1+dfsg-4+vitastor1
--- a/debian/vitastor.Dockerfile
+++ b/debian/vitastor.Dockerfile
@ -1,67 +0,0 @@
 # Build Vitastor packages for Debian Buster or Bullseye/Sid inside a container
 # cd ..; podman build --build-arg REL=bullseye -v `pwd`/packages:/root/packages -f debian/vitastor.Dockerfile .
 FROM debian:$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-*/
--- a/src/fio_engine.cpp
+++ b/src/fio_engine.cpp
@ -1,6 +1,3 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 (see README.md for details)
 // FIO engine to test Blockstore
 //
 // Initialize storage for tests:
@ -25,7 +22,11 @@
 //     -bs_config='{"data_device":"./test_data.bin"}' -size=1000M
 #include "blockstore.h"
-#include "fio_headers.h"
+extern "C" {
 #define CONFIG_PWRITEV2
 #include "fio/fio.h"
 #include "fio/optgroup.h"
 }
 #include "json11/json11.hpp"
@ -99,7 +100,7 @@ static void bs_cleanup(struct thread_data *td)
                bsd->ringloop->loop();
                if (bsd->bs->is_safe_to_stop())
                    goto safe;
-            } while (bsd->ringloop->has_work());
+            } while (bsd->ringloop->get_loop_again());
            bsd->ringloop->wait();
        }
    safe:
@ -288,7 +289,7 @@ static int bs_invalidate(struct thread_data *td, struct fio_file *f)
 }
 struct ioengine_ops ioengine = {
-    .name               = "vitastor_blockstore",
+    .name               = "microceph_blockstore",
    .version            = FIO_IOOPS_VERSION,
    .flags              = FIO_MEMALIGN | FIO_DISKLESSIO | FIO_NOEXTEND,
    .setup              = bs_setup,
--- a/src/fio_sec_osd.cpp
+++ b/src/fio_sec_osd.cpp
@ -1,6 +1,3 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 // FIO engine to test Blockstore through Secondary OSD interface
 //
 // Prepare storage like in fio_engine.cpp, then start OSD with ./osd, then test it
@ -8,7 +5,7 @@
 // Random write:
 //
 // fio -thread -ioengine=./libfio_sec_osd.so -name=test -bs=4k -direct=1 -fsync=16 -iodepth=16 -rw=randwrite \
-//     -host=127.0.0.1 -port=11203 [-block_size_order=17] [-single_primary=1] -size=1000M
+//     -host=127.0.0.1 -port=11203 [-single_primary=1] -size=1000M
 //
 // Linear write:
 //
@ -30,7 +27,11 @@
 #include "rw_blocking.h"
 #include "osd_ops.h"
-#include "fio_headers.h"
+extern "C" {
 #define CONFIG_PWRITEV2
 #include "fio/fio.h"
 #include "fio/optgroup.h"
 }
 struct sec_data
 {
@ -51,7 +52,6 @@ struct sec_options
    int port = 0;
    int single_primary = 0;
    int trace = 0;
    int block_order = 17;
 };
 static struct fio_option options[] = {
@ -73,15 +73,6 @@ static struct fio_option options[] = {
        .category = FIO_OPT_C_ENGINE,
        .group  = FIO_OPT_G_FILENAME,
    },
    {
        .name   = "block_size_order",
        .lname  = "Blockstore block size order",
        .type   = FIO_OPT_INT,
        .off1   = offsetof(struct sec_options, block_order),
        .help   = "Blockstore block size order (size = 2^order)",
        .category = FIO_OPT_C_ENGINE,
        .group  = FIO_OPT_G_FILENAME,
    },
    {
        .name   = "single_primary",
        .lname  = "Single Primary",
@ -140,7 +131,6 @@ static void sec_cleanup(struct thread_data *td)
    if (bsd)
    {
        close(bsd->connect_fd);
        delete bsd;
    }
 }
@ -149,8 +139,6 @@ static int sec_init(struct thread_data *td)
 {
    sec_options *o = (sec_options*)td->eo;
    sec_data *bsd = (sec_data*)td->io_ops_data;
    bsd->block_order = o->block_order == 0 ? 17 : o->block_order;
    bsd->block_size = 1 << o->block_order;
    struct sockaddr_in addr;
    int r;
@ -204,7 +192,7 @@ static enum fio_q_status sec_queue(struct thread_data *td, struct io_u *io)
    case DDIR_READ:
        if (!opt->single_primary)
        {
-            op.hdr.opcode = OSD_OP_SEC_READ;
+            op.hdr.opcode = OSD_OP_SECONDARY_READ;
            op.sec_rw.oid = {
                .inode = 1,
                .stripe = io->offset >> bsd->block_order,
@ -225,7 +213,7 @@ static enum fio_q_status sec_queue(struct thread_data *td, struct io_u *io)
    case DDIR_WRITE:
        if (!opt->single_primary)
        {
-            op.hdr.opcode = OSD_OP_SEC_WRITE;
+            op.hdr.opcode = OSD_OP_SECONDARY_WRITE;
            op.sec_rw.oid = {
                .inode = 1,
                .stripe = io->offset >> bsd->block_order,
@ -273,18 +261,15 @@ static enum fio_q_status sec_queue(struct thread_data *td, struct io_u *io)
    bsd->op_n++;
    bsd->queue[n] = io;
-    iovec iov[2] = { { .iov_base = op.buf, .iov_len = OSD_PACKET_SIZE } };
+    if (write(bsd->connect_fd, op.buf, OSD_PACKET_SIZE) != OSD_PACKET_SIZE)
-    int iovcnt = 1, wtotal = OSD_PACKET_SIZE;
+    {
        perror("write");
        exit(1);
    }
    if (io->ddir == DDIR_WRITE)
    {
-        iov[1] = { .iov_base = io->xfer_buf, .iov_len = io->xfer_buflen };
+        // Send data
-        wtotal += io->xfer_buflen;
+        write_blocking(bsd->connect_fd, io->xfer_buf, io->xfer_buflen);
        iovcnt++;
    }
    if (writev_blocking(bsd->connect_fd, iov, iovcnt) != wtotal)
    {
        perror("writev");
        exit(1);
    }
    if (io->error != 0)
@ -313,7 +298,6 @@ static int sec_getevents(struct thread_data *td, unsigned int min, unsigned int
            exit(1);
        }
        io_u* io = it->second;
        bsd->queue.erase(it);
        if (io->ddir == DDIR_READ)
        {
            if (reply.hdr.retval != io->xfer_buflen)
@ -381,7 +365,7 @@ static int sec_invalidate(struct thread_data *td, struct fio_file *f)
 }
 struct ioengine_ops ioengine = {
-    .name               = "vitastor_secondary_osd",
+    .name               = "microceph_secondary_osd",
    .version            = FIO_IOOPS_VERSION,
    .flags              = FIO_MEMALIGN | FIO_DISKLESSIO | FIO_NOEXTEND,
    .setup              = sec_setup,
--- a/1
+++ b/1
@ -1 +0,0 @@
 Subproject commit 97f06cb20c1e136fd37d58fb40f57dd8f8a3a4a7
--- a/lambda_size.cpp
+++ b/lambda_size.cpp
@ -0,0 +1,48 @@
 #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();
 }
--- a/mon/PGUtil.js
+++ b/mon/PGUtil.js
@ -1,104 +0,0 @@
 // 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);
    }
 }
--- a/mon/afr.js
+++ b/mon/afr.js
@ -1,89 +0,0 @@
 // 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;
 }
--- a/mon/afr_test.js
+++ b/mon/afr_test.js
@ -1,28 +0,0 @@
 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`)+
        `, ${config.pgs||1} PG per OSD`+
        (config.degraded_replacement ? `\n...and you don't let the rebalance finish before replacing drives` : '')
    );
    console.log('-> '+sprintf("%.7f%%", 100*cluster_afr(config))+'\n');
 }
--- a/mon/lp-optimizer.js
+++ b/mon/lp-optimizer.js
@ -1,716 +0,0 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 (see README.md for details)
 // Data distribution optimizer using linear programming (lp_solve)
 const child_process = require('child_process');
 const NO_OSD = 'Z';
 async function lp_solve(text)
 {
    const cp = child_process.spawn('lp_solve');
    let stdout = '', stderr = '', finish_cb;
    cp.stdout.on('data', buf => stdout += buf.toString());
    cp.stderr.on('data', buf => stderr += buf.toString());
    cp.on('exit', () => finish_cb && finish_cb());
    cp.stdin.write(text);
    cp.stdin.end();
    if (cp.exitCode == null)
    {
        await new Promise(ok => finish_cb = ok);
    }
    if (!stdout.trim())
    {
        return null;
    }
    let score = 0;
    let vars = {};
    for (const line of stdout.split(/\n/))
    {
        let m = /^(^Value of objective function: (-?[\d\.]+)|Actual values of the variables:)\s*$/.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,
 };
--- a/mon/make-osd.sh
+++ b/mon/make-osd.sh
@ -1,76 +0,0 @@
 #!/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
--- a/mon/make-units.sh
+++ b/mon/make-units.sh
@ -1,85 +0,0 @@
 #!/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
--- a/mon/merge.js
+++ b/mon/merge.js
@ -1,23 +0,0 @@
 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]);
--- a/mon/mon-main.js
+++ b/mon/mon-main.js
@ -1,25 +0,0 @@
 #!/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(); });
--- a/mon/mon.js
+++ b/mon/mon.js
--- a/mon/package.json
+++ b/mon/package.json
@ -1,15 +0,0 @@
 {
  "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"
  }
 }
--- a/mon/simple-offsets.js
+++ b/mon/simple-offsets.js
@ -1,93 +0,0 @@
 // 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); });
--- a/mon/stable-stringify.js
+++ b/mon/stable-stringify.js
@ -1,78 +0,0 @@
 // 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;
--- a/mon/test-nonuniform.js
+++ b/mon/test-nonuniform.js
@ -1,130 +0,0 @@
 // 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);
--- a/mon/test-optimize-simple.js
+++ b/mon/test-optimize-simple.js
@ -1,25 +0,0 @@
 // 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);
--- a/mon/test-optimize-undersized.js
+++ b/mon/test-optimize-undersized.js
@ -1,74 +0,0 @@
 // 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);
--- a/mon/test-optimize.js
+++ b/mon/test-optimize.js
@ -1,115 +0,0 @@
 // 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);
--- a/src/object_id.h
+++ b/src/object_id.h
@ -1,19 +1,14 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 #pragma once
 #include <stdint.h>
 #include <functional>
 typedef uint64_t inode_t;
 // 16 bytes per object/stripe id
 // stripe = (start of the parity stripe + peer role)
 // i.e. for example (256KB + one of 0,1,2)
 struct __attribute__((__packed__)) object_id
 {
-    inode_t inode;
+    uint64_t inode;
    uint64_t stripe;
 };
--- a/osd.cpp
+++ b/osd.cpp
@ -0,0 +1,399 @@
 #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 subops with data: %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()
 {
    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));
                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 = 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;
    }
    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();
    };
 }
 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)
 {
    gettimeofday(&cur_op->tv_begin, NULL);
    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);
    }
 }
--- a/osd.h
+++ b/osd.h
@ -0,0 +1,274 @@
 #pragma once
 #include <sys/types.h>
 #include <sys/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
 {
    timeval tv_begin;
    timeval 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 = { 0 };
    msghdr read_msg = { 0 };
    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();
    bool try_receive(osd_client_t & cl);
    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);
    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
    );
 }
--- a/osd_client.cpp
+++ b/osd_client.cpp
@ -0,0 +1,40 @@
 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),
            });
        }
    }
 }
--- a/osd_id.h
+++ b/osd_id.h
@ -0,0 +1,4 @@
 #pragma once
 typedef uint64_t osd_num_t;
 typedef uint32_t pg_num_t;
--- a/src/osd_main.cpp
+++ b/src/osd_main.cpp
@ -1,28 +1,14 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 (see README.md for details)
 #include "osd.h"
 #include <signal.h>
-static osd_t *osd = NULL;
+void handle_sigint(int sig)
 static bool force_stopping = false;
 static void handle_sigint(int sig)
 {
    if (osd && !force_stopping)
    {
        force_stopping = true;
        osd->force_stop(0);
        return;
    }
    exit(0);
 }
 int main(int narg, char *args[])
 {
    setvbuf(stdout, NULL, _IONBF, 0);
    setvbuf(stderr, NULL, _IONBF, 0);
    if (sizeof(osd_any_op_t) > OSD_PACKET_SIZE ||
        sizeof(osd_any_reply_t) > OSD_PACKET_SIZE)
    {
@ -39,15 +25,16 @@ int main(int narg, char *args[])
        }
    }
    signal(SIGINT, handle_sigint);
    signal(SIGTERM, handle_sigint);
    ring_loop_t *ringloop = new ring_loop_t(512);
-    osd = new osd_t(config, ringloop);
+    blockstore_t *bs = new blockstore_t(config, ringloop);
    osd_t *osd = new osd_t(config, bs, ringloop);
    while (1)
    {
        ringloop->loop();
        ringloop->wait();
    }
    delete osd;
    delete bs;
    delete ringloop;
    return 0;
 }
--- a/src/osd_ops.h
+++ b/src/osd_ops.h
@ -1,6 +1,3 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 #pragma once
 #include "object_id.h"
@ -13,26 +10,21 @@
 #define OSD_PACKET_SIZE             0x80
 // Opcodes
 #define OSD_OP_MIN                  1
-#define OSD_OP_SEC_READ             1
+#define OSD_OP_SECONDARY_READ       1
-#define OSD_OP_SEC_WRITE            2
+#define OSD_OP_SECONDARY_WRITE      2
-#define OSD_OP_SEC_WRITE_STABLE     3
+#define OSD_OP_SECONDARY_SYNC       3
-#define OSD_OP_SEC_SYNC             4
+#define OSD_OP_SECONDARY_STABILIZE  4
-#define OSD_OP_SEC_STABILIZE        5
+#define OSD_OP_SECONDARY_ROLLBACK   5
-#define OSD_OP_SEC_ROLLBACK         6
+#define OSD_OP_SECONDARY_DELETE     6
-#define OSD_OP_SEC_DELETE           7
+#define OSD_OP_TEST_SYNC_STAB_ALL   7
-#define OSD_OP_TEST_SYNC_STAB_ALL   8
+#define OSD_OP_SECONDARY_LIST       8
-#define OSD_OP_SEC_LIST             9
+#define OSD_OP_SHOW_CONFIG          9
-#define OSD_OP_SHOW_CONFIG          10
+#define OSD_OP_READ                 10
-#define OSD_OP_READ                 11
+#define OSD_OP_WRITE                11
-#define OSD_OP_WRITE                12
+#define OSD_OP_SYNC                 12
-#define OSD_OP_SYNC                 13
+#define OSD_OP_MAX                  12
 #define OSD_OP_DELETE               14
 #define OSD_OP_PING                 15
 #define OSD_OP_MAX                  15
 // Alignment & limit for read/write operations
-#ifndef MEM_ALIGNMENT
+#define OSD_RW_ALIGN                512
 #define MEM_ALIGNMENT               512
 #endif
 #define OSD_RW_MAX                  64*1024*1024
 // common request and reply headers
@ -65,15 +57,11 @@ struct __attribute__((__packed__)) osd_op_secondary_rw_t
    // object
    object_id oid;
    // read/write version (automatic or specific)
    // FIXME deny values close to UINT64_MAX
    uint64_t version;
    // offset
    uint32_t offset;
    // length
    uint32_t len;
    // bitmap/attribute length - bitmap comes after header, but before data
    uint32_t attr_len;
    uint32_t pad0;
 };
 struct __attribute__((__packed__)) osd_reply_secondary_rw_t
@ -81,9 +69,6 @@ struct __attribute__((__packed__)) osd_reply_secondary_rw_t
    osd_reply_header_t header;
    // for reads and writes: assigned or read version number
    uint64_t version;
    // for reads: bitmap/attribute length (just to double-check)
    uint32_t attr_len;
    uint32_t pad0;
 };
 // delete object on the secondary OSD
@ -145,10 +130,7 @@ struct __attribute__((__packed__)) osd_op_secondary_list_t
    osd_op_header_t header;
    // placement group total number and total count
    pg_num_t list_pg, pg_count;
-    // size of an area that maps to one PG continuously
+    uint64_t parity_block_size;
    uint64_t pg_stripe_size;
    // inode range (used to select pools)
    uint64_t min_inode, max_inode;
 };
 struct __attribute__((__packed__)) osd_reply_secondary_list_t
@ -174,9 +156,6 @@ struct __attribute__((__packed__)) osd_op_rw_t
 struct __attribute__((__packed__)) osd_reply_rw_t
 {
    osd_reply_header_t header;
    // for reads: bitmap length
    uint32_t bitmap_len;
    uint32_t pad0;
 };
 // sync to the primary OSD
@ -190,7 +169,6 @@ struct __attribute__((__packed__)) osd_reply_sync_t
    osd_reply_header_t header;
 };
 // FIXME it would be interesting to try to unify blockstore_op and osd_op formats
 union osd_any_op_t
 {
    osd_op_header_t hdr;
@ -218,5 +196,3 @@ union osd_any_reply_t
    osd_reply_sync_t sync;
    uint8_t buf[OSD_PACKET_SIZE];
 };
 extern const char* osd_op_names[];
--- a/osd_peering.cpp
+++ b/osd_peering.cpp
@ -0,0 +1,407 @@
 #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 | 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));
    fcntl(peer_fd, F_SETFL, fcntl(peer_fd, F_GETFL, 0) & ~O_NONBLOCK);
    // Disable EPOLLOUT on this fd
    cl.connect_callback = NULL;
    cl.peer_state = PEER_CONNECTED;
    epoll_event ev;
    ev.data.fd = peer_fd;
    ev.events = 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();
 }
--- a/osd_peering_pg.cpp
+++ b/osd_peering_pg.cpp
@ -0,0 +1,265 @@
 #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;
 }
--- a/src/osd_peering_pg.h
+++ b/src/osd_peering_pg.h
@ -1,23 +1,43 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 (see README.md for details)
 #include <map>
 #include <vector>
 #include <algorithm>
 #include "cpp-btree/btree_map.h"
 #include "object_id.h"
 #include "osd_ops.h"
 #include "pg_states.h"
-#define PG_EPOCH_BITS 48
+#include "sparsepp/sparsepp/spp.h"
 // Placement group states
 // Exactly one of these:
 #define PG_OFFLINE (1<<0)
 #define PG_PEERING (1<<1)
 #define PG_INCOMPLETE (1<<2)
 #define PG_ACTIVE (1<<3)
 // Plus any of these:
 #define PG_DEGRADED (1<<4)
 #define PG_HAS_UNFOUND (1<<5)
 #define PG_HAS_DEGRADED (1<<6)
 #define PG_HAS_MISPLACED (1<<7)
 #define PG_HAS_UNCLEAN (1<<8)
 // FIXME: Safe default that doesn't depend on parity_block_size of pg_parity_size
 #define STRIPE_MASK ((uint64_t)4096 - 1)
 // OSD object states
 #define OBJ_CLEAN 0x01
 #define OBJ_MISPLACED 0x02
 #define OBJ_DEGRADED 0x03
 #define OBJ_INCOMPLETE 0x04
 #define OBJ_NEEDS_STABLE 0x10000
 #define OBJ_NEEDS_ROLLBACK 0x20000
 #define OBJ_OVERCOPIED 0x40000
 #define OBJ_BUGGY 0x80000
 struct pg_obj_loc_t
 {
    uint64_t role;
    osd_num_t osd_num;
-    bool outdated;
+    bool stable;
 };
 typedef std::vector<pg_obj_loc_t> pg_osd_set_t;
@ -44,10 +64,28 @@ struct osd_op_t;
 struct pg_peering_state_t
 {
    // osd_num -> list result
-    std::map<osd_num_t, osd_op_t*> list_ops;
+    spp::sparse_hash_map<osd_num_t, osd_op_t*> list_ops;
-    std::map<osd_num_t, pg_list_result_t> list_results;
+    spp::sparse_hash_map<osd_num_t, pg_list_result_t> list_results;
-    pool_id_t pool_id = 0;
+    int list_done = 0;
-    pg_num_t pg_num = 0;
+};
 struct pg_obj_state_check_t
 {
    int obj_start = 0, obj_end = 0, ver_start = 0, ver_end = 0;
    object_id oid = { 0 };
    uint64_t max_ver = 0;
    uint64_t target_ver = 0;
    uint64_t n_copies = 0, has_roles = 0, n_roles = 0, n_stable = 0, n_matched = 0;
    bool is_buggy = false, has_old_unstable = false;
    pg_osd_set_t osd_set;
 };
 struct obj_ver_role
 {
    object_id oid;
    uint64_t version;
    uint64_t osd_num;
    bool is_stable;
 };
 struct obj_piece_id_t
@ -56,75 +94,60 @@ struct obj_piece_id_t
    uint64_t osd_num;
 };
-struct obj_ver_osd_t
+struct obj_piece_ver_t
 {
-    uint64_t osd_num;
+    uint64_t max_ver = 0;
-    object_id oid;
+    uint64_t stable_ver = 0;
    uint64_t version;
 };
-struct flush_action_t
+struct obj_stab_action_t
 {
    bool rollback = false, make_stable = false;
    uint64_t stable_to = 0, rollback_to = 0;
    bool submitted = false;
 };
 struct pg_flush_batch_t
 {
    std::map<osd_num_t, std::vector<obj_ver_id>> rollback_lists;
    std::map<osd_num_t, std::vector<obj_ver_id>> stable_lists;
    int flush_ops = 0, flush_done = 0;
    int flush_objects = 0;
 };
 struct pg_t
 {
-    int state = 0;
+    int state;
-    uint64_t scheme = 0;
+    uint64_t pg_cursize = 3, pg_size = 3, pg_minsize = 2;
-    uint64_t pg_cursize = 0, pg_size = 0, pg_minsize = 0, pg_data_size = 0;
+    pg_num_t pg_num;
    pool_id_t pool_id = 0;
    pg_num_t pg_num = 0;
    uint64_t clean_count = 0, total_count = 0;
    // epoch number - should increase with each non-clean activation of the PG
    uint64_t epoch = 0, reported_epoch = 0;
    // target history and all potential peers
    std::vector<std::vector<osd_num_t>> target_history;
    std::vector<osd_num_t> all_peers;
    bool history_changed = false;
    // peer list from the last peering event
    std::vector<osd_num_t> cur_peers;
    // target_set is the "correct" peer OSD set for this PG
    std::vector<osd_num_t> target_set;
    // cur_set is the current set of connected peer OSDs for this PG
    // cur_set = (role => osd_num or UINT64_MAX if missing). role numbers begin with zero
    std::vector<osd_num_t> cur_set;
-    // same thing in state_dict-like format
+    // moved object map. by default, each object is considered to reside on the cur_set.
    pg_osd_set_t cur_loc_set;
    // moved object map. by default, each object is considered to reside on cur_set.
    // this map stores all objects that differ.
    // it may consume up to ~ (raw storage / object size) * 24 bytes in the worst case scenario
    // which is up to ~192 MB per 1 TB in the worst case scenario
    std::map<pg_osd_set_t, pg_osd_set_state_t> state_dict;
-    btree::btree_map<object_id, pg_osd_set_state_t*> incomplete_objects, misplaced_objects, degraded_objects;
+    spp::sparse_hash_map<object_id, pg_osd_set_state_t*> obj_states;
-    std::map<obj_piece_id_t, flush_action_t> flush_actions;
+    std::map<obj_piece_id_t, obj_stab_action_t> obj_stab_actions;
-    std::vector<obj_ver_osd_t> copies_to_delete_after_sync;
+    spp::sparse_hash_map<object_id, uint64_t> ver_override;
    btree::btree_map<object_id, uint64_t> ver_override;
    pg_peering_state_t *peering_state = NULL;
    pg_flush_batch_t *flush_batch = NULL;
    int inflight = 0; // including write_queue
    std::multimap<object_id, osd_op_t*> write_queue;
-    void calc_object_states(int log_level);
+    void calc_object_states();
-    void print_state();
+    void remember_object(pg_obj_state_check_t &st, std::vector<obj_ver_role> &all);
 };
 inline bool operator < (const pg_obj_loc_t &a, const pg_obj_loc_t &b)
 {
-    return a.outdated < b.outdated ||
+    return a.role < b.role || a.role == b.role && a.osd_num < b.osd_num ||
-        a.outdated == b.outdated && a.role < b.role ||
+        a.role == b.role && a.osd_num == b.osd_num && a.stable < b.stable;
-        a.outdated == b.outdated && a.role == b.role && a.osd_num < b.osd_num;
+}
 inline bool operator < (const obj_ver_role & a, const obj_ver_role & b)
 {
    // ORDER BY inode ASC, stripe & ~STRIPE_MASK ASC, version DESC, osd_num ASC
    return a.oid.inode < b.oid.inode || a.oid.inode == b.oid.inode && (
        (a.oid.stripe & ~STRIPE_MASK) < (b.oid.stripe & ~STRIPE_MASK) ||
        (a.oid.stripe & ~STRIPE_MASK) == (b.oid.stripe & ~STRIPE_MASK) && (
            a.version > b.version || a.version == b.version && a.osd_num < b.osd_num
        )
    );
 }
 inline bool operator == (const obj_piece_id_t & a, const obj_piece_id_t & b)
@ -149,6 +172,7 @@ namespace std
                // Copy-pasted from spp::hash_combine()
                seed ^= (e.role + 0xc6a4a7935bd1e995 + (seed << 6) + (seed >> 2));
                seed ^= (e.osd_num + 0xc6a4a7935bd1e995 + (seed << 6) + (seed >> 2));
                seed ^= ((e.stable ? 1 : 0) + 0xc6a4a7935bd1e995 + (seed << 6) + (seed >> 2));
            }
            return seed;
        }
--- a/osd_primary.cpp
+++ b/osd_primary.cpp
@ -0,0 +1,603 @@
 #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]);
        }
    }
 }
--- a/osd_receive.cpp
+++ b/osd_receive.cpp
@ -0,0 +1,207 @@
 #include "osd.h"
 bool osd_t::try_receive(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.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);
    return true;
 }
 void osd_t::read_requests()
 {
    for (auto & p: clients)
    {
        if (p.second.peer_state == PEER_CONNECTED && p.second.read_iov.iov_len == 0)
        {
            try_receive(p.second);
        }
    }
 }
 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;
        cl.read_iov.iov_len = 0;
        if (data->res == -EAGAIN)
        {
            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;
        }
        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
                    timeval tv_end;
                    gettimeofday(&tv_end, NULL);
                    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_usec - request->tv_begin.tv_usec
                    );
                    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
        timeval tv_end;
        gettimeofday(&tv_end, NULL);
        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_usec - op->tv_begin.tv_usec
        );
        op->callback(op);
    }
 }
--- a/osd_rmw.cpp
+++ b/osd_rmw.cpp
@ -0,0 +1,367 @@
 #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);
        }
    }
 }
--- a/osd_rmw.h
+++ b/osd_rmw.h
@ -0,0 +1,36 @@
 #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);
--- a/osd_rmw_test.cpp
+++ b/osd_rmw_test.cpp
@ -0,0 +1,93 @@
 #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;
 }
--- a/osd_secondary.cpp
+++ b/osd_secondary.cpp
@ -0,0 +1,150 @@
 #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
 }
--- a/osd_send.cpp
+++ b/osd_send.cpp
@ -0,0 +1,125 @@
 #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)
    {
        gettimeofday(&cur_op->tv_begin, NULL);
    }
    if (cl.write_state == 0)
    {
        cl.write_state = CL_WRITE_READY;
        write_ready_clients.push_back(cur_op->peer_fd);
    }
    cl.outbox.push_back(cur_op);
    ringloop->wakeup();
 }
 void osd_t::send_replies()
 {
    for (int i = 0; i < write_ready_clients.size(); i++)
    {
        int peer_fd = write_ready_clients[i];
        auto & cl = clients[peer_fd];
        io_uring_sqe* sqe = ringloop->get_sqe();
        if (!sqe)
        {
            write_ready_clients.erase(write_ready_clients.begin(), write_ready_clients.begin() + i);
            return;
        }
        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;
            if (cl.write_op->op_type == OSD_OP_OUT)
            {
                gettimeofday(&cl.write_op->tv_send, NULL);
            }
            else
            {
                // Measure execution latency
                timeval tv_end;
                gettimeofday(&tv_end, NULL);
                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_usec - cl.write_op->tv_begin.tv_usec
                );
            }
        }
        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);
    }
    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->op_type == OSD_OP_IN)
                {
                    delete cur_op;
                }
                else
                {
                    // Measure subops with data
                    if (cur_op->req.hdr.opcode == OSD_OP_SECONDARY_STABILIZE ||
                        cur_op->req.hdr.opcode == OSD_OP_SECONDARY_WRITE)
                    {
                        timeval tv_end;
                        gettimeofday(&tv_end, NULL);
                        send_stat_count++;
                        send_stat_sum += (
                            (tv_end.tv_sec - cl.write_op->tv_send.tv_sec)*1000000 +
                            tv_end.tv_usec - cl.write_op->tv_send.tv_usec
                        );
                    }
                    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);
        }
    }
 }
--- a/src/osd_test.cpp
+++ b/src/osd_test.cpp
@ -1,6 +1,3 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 (see README.md for details)
 #include <sys/types.h>
 #include <sys/socket.h>
 #include <netinet/in.h>
@ -22,8 +19,6 @@
 int connect_osd(const char *osd_address, int osd_port);
 uint64_t test_read(int connect_fd, uint64_t inode, uint64_t stripe, uint64_t version, uint64_t offset, uint64_t len);
 uint64_t test_write(int connect_fd, uint64_t inode, uint64_t stripe, uint64_t version, uint64_t pattern);
 void* test_primary_read(int connect_fd, uint64_t inode, uint64_t offset, uint64_t len);
@ -34,8 +29,6 @@ void test_primary_sync(int connect_fd);
 void test_sync_stab_all(int connect_fd);
 void test_list_stab(int connect_fd);
 int main0(int narg, char *args[])
 {
    int connect_fd;
@ -101,16 +94,7 @@ int main2(int narg, char *args[])
    return 0;
 }
-int main3(int narg, char *args[])
+int main(int narg, char *args[])
 {
    int connect_fd;
    connect_fd = connect_osd("127.0.0.1", 11203);
    test_list_stab(connect_fd);
    close(connect_fd);
    return 0;
 }
 int main4(int narg, char *args[])
 {
    int connect_fd;
    // Cluster write (sync not implemented yet)
@ -122,15 +106,6 @@ int main4(int narg, char *args[])
    return 0;
 }
 int main(int narg, char *args[])
 {
    int connect_fd;
    connect_fd = connect_osd("192.168.7.2", 43051);
    test_read(connect_fd, 1, 1039663104, UINT64_MAX, 0, 128*1024);
    close(connect_fd);
    return 0;
 }
 int connect_osd(const char *osd_address, int osd_port)
 {
    struct sockaddr_in addr;
@ -173,7 +148,7 @@ bool check_reply(int r, osd_any_op_t & op, osd_any_reply_t & reply, int expected
        printf("bad reply: magic, id or opcode does not match request\n");
        return false;
    }
-    if (expected >= 0 && reply.hdr.retval != expected)
+    if (reply.hdr.retval != expected)
    {
        printf("operation failed, retval=%ld\n", reply.hdr.retval);
        return false;
@ -181,73 +156,13 @@ bool check_reply(int r, osd_any_op_t & op, osd_any_reply_t & reply, int expected
    return true;
 }
 uint64_t test_read(int connect_fd, uint64_t inode, uint64_t stripe, uint64_t version, uint64_t offset, uint64_t len)
 {
    osd_any_op_t op;
    osd_any_reply_t reply;
    op.hdr.magic = SECONDARY_OSD_OP_MAGIC;
    op.hdr.id = 1;
    op.hdr.opcode = OSD_OP_SEC_READ;
    op.sec_rw.oid = {
        .inode = inode,
        .stripe = stripe,
    };
    op.sec_rw.version = version;
    op.sec_rw.offset = offset;
    op.sec_rw.len = len;
    void *data = memalign(MEM_ALIGNMENT, op.sec_rw.len);
    write_blocking(connect_fd, op.buf, OSD_PACKET_SIZE);
    int r = read_blocking(connect_fd, reply.buf, OSD_PACKET_SIZE);
    if (!check_reply(r, op, reply, op.sec_rw.len))
    {
        free(data);
        return 0;
    }
    r = read_blocking(connect_fd, data, len);
    if (r != len)
    {
        free(data);
        perror("read data");
        return 0;
    }
    free(data);
    printf("Read %lx:%lx v%lu = v%lu\n", inode, stripe, version, reply.sec_rw.version);
    op.hdr.opcode = OSD_OP_SEC_LIST;
    op.sec_list.list_pg = 1;
    op.sec_list.pg_count = 1;
    op.sec_list.pg_stripe_size = 4*1024*1024;
    write_blocking(connect_fd, op.buf, OSD_PACKET_SIZE);
    r = read_blocking(connect_fd, reply.buf, OSD_PACKET_SIZE);
    if (reply.hdr.retval < 0 || !check_reply(r, op, reply, reply.hdr.retval))
    {
        return 0;
    }
    data = memalign(MEM_ALIGNMENT, sizeof(obj_ver_id)*reply.hdr.retval);
    r = read_blocking(connect_fd, data, sizeof(obj_ver_id)*reply.hdr.retval);
    if (r != sizeof(obj_ver_id)*reply.hdr.retval)
    {
        free(data);
        perror("read data");
        return 0;
    }
    obj_ver_id *ov = (obj_ver_id*)data;
    for (int i = 0; i < reply.hdr.retval; i++)
    {
        if (ov[i].oid.inode == inode && (ov[i].oid.stripe & ~(4096-1)) == (stripe & ~(4096-1)))
        {
            printf("list: %lx:%lx v%lu stable=%d\n", ov[i].oid.inode, ov[i].oid.stripe, ov[i].version, i < reply.sec_list.stable_count ? 1 : 0);
        }
    }
    return 0;
 }
 uint64_t test_write(int connect_fd, uint64_t inode, uint64_t stripe, uint64_t version, uint64_t pattern)
 {
    osd_any_op_t op;
    osd_any_reply_t reply;
    op.hdr.magic = SECONDARY_OSD_OP_MAGIC;
    op.hdr.id = 1;
-    op.hdr.opcode = OSD_OP_SEC_WRITE;
+    op.hdr.opcode = OSD_OP_SECONDARY_WRITE;
    op.sec_rw.oid = {
        .inode = inode,
        .stripe = stripe,
@ -255,7 +170,7 @@ uint64_t test_write(int connect_fd, uint64_t inode, uint64_t stripe, uint64_t ve
    op.sec_rw.version = version;
    op.sec_rw.offset = 0;
    op.sec_rw.len = 128*1024;
-    void *data = memalign(MEM_ALIGNMENT, op.sec_rw.len);
+    void *data = memalign(512, op.sec_rw.len);
    for (int i = 0; i < (op.sec_rw.len)/sizeof(uint64_t); i++)
        ((uint64_t*)data)[i] = pattern;
    write_blocking(connect_fd, op.buf, OSD_PACKET_SIZE);
@ -290,7 +205,7 @@ void* test_primary_read(int connect_fd, uint64_t inode, uint64_t offset, uint64_
    op.rw.inode = inode;
    op.rw.offset = offset;
    op.rw.len = len;
-    void *data = memalign(MEM_ALIGNMENT, len);
+    void *data = memalign(512, len);
    write_blocking(connect_fd, op.buf, OSD_PACKET_SIZE);
    int r = read_blocking(connect_fd, reply.buf, OSD_PACKET_SIZE);
    if (!check_reply(r, op, reply, len))
@ -318,7 +233,7 @@ void test_primary_write(int connect_fd, uint64_t inode, uint64_t offset, uint64_
    op.rw.inode = inode;
    op.rw.offset = offset;
    op.rw.len = len;
-    void *data = memalign(MEM_ALIGNMENT, len);
+    void *data = memalign(512, len);
    set_pattern(data, len, pattern);
    write_blocking(connect_fd, op.buf, OSD_PACKET_SIZE);
    write_blocking(connect_fd, data, len);
@ -350,40 +265,3 @@ void test_sync_stab_all(int connect_fd)
    int r = read_blocking(connect_fd, reply.buf, OSD_PACKET_SIZE);
    assert(check_reply(r, op, reply, 0));
 }
 void test_list_stab(int connect_fd)
 {
    osd_any_op_t op;
    osd_any_reply_t reply;
    op.hdr.magic = SECONDARY_OSD_OP_MAGIC;
    op.hdr.id = 1;
    op.hdr.opcode = OSD_OP_SEC_LIST;
    op.sec_list.pg_count = 0;
    assert(write_blocking(connect_fd, op.buf, OSD_PACKET_SIZE) == OSD_PACKET_SIZE);
    int r = read_blocking(connect_fd, reply.buf, OSD_PACKET_SIZE);
    assert(check_reply(r, op, reply, -1));
    int total_count = reply.hdr.retval;
    int stable_count = reply.sec_list.stable_count;
    obj_ver_id *data = (obj_ver_id*)malloc(total_count * sizeof(obj_ver_id));
    assert(data);
    assert(read_blocking(connect_fd, data, total_count * sizeof(obj_ver_id)) == (total_count * sizeof(obj_ver_id)));
    int last_start = stable_count;
    for (int i = stable_count; i <= total_count; i++)
    {
        // Stabilize in portions of 32 entries
        if (i - last_start >= 32 || i == total_count)
        {
            op.hdr.opcode = OSD_OP_SEC_STABILIZE;
            op.sec_stab.len = sizeof(obj_ver_id) * (i - last_start);
            assert(write_blocking(connect_fd, op.buf, OSD_PACKET_SIZE) == OSD_PACKET_SIZE);
            assert(write_blocking(connect_fd, data + last_start, op.sec_stab.len) == op.sec_stab.len);
            r = read_blocking(connect_fd, reply.buf, OSD_PACKET_SIZE);
            assert(check_reply(r, op, reply, 0));
            last_start = i;
        }
    }
    obj_ver_id *data2 = (obj_ver_id*)malloc(sizeof(obj_ver_id) * 32);
    assert(data2);
    free(data2);
    free(data);
 }
--- a/qemu-3.1-vitastor.patch
+++ b/qemu-3.1-vitastor.patch
@ -1,84 +0,0 @@
 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):
--- a/qemu-4.2-vitastor.patch
+++ b/qemu-4.2-vitastor.patch
@ -1,84 +0,0 @@
 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):
--- a/qemu-5.0-vitastor.patch
+++ b/qemu-5.0-vitastor.patch
@ -1,84 +0,0 @@
 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):
--- a/qemu-5.1-vitastor.patch
+++ b/qemu-5.1-vitastor.patch
@ -1,84 +0,0 @@
 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):
--- a/src/ringloop.cpp
+++ b/src/ringloop.cpp
@ -1,10 +1,3 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 #include <stdlib.h>
 #include <stdexcept>
 #include "ringloop.h"
 ring_loop_t::ring_loop_t(int qd)
@ -15,7 +8,7 @@ ring_loop_t::ring_loop_t(int qd)
        throw std::runtime_error(std::string("io_uring_queue_init: ") + strerror(-ret));
    }
    free_ring_data_ptr = *ring.cq.kring_entries;
-    ring_datas = (struct ring_data_t*)calloc(free_ring_data_ptr, sizeof(ring_data_t));
+    ring_datas = (struct ring_data_t*)malloc(sizeof(ring_data_t) * free_ring_data_ptr);
    free_ring_data = (int*)malloc(sizeof(int) * free_ring_data_ptr);
    if (!ring_datas || !free_ring_data)
    {
@ -25,7 +18,6 @@ ring_loop_t::ring_loop_t(int qd)
    {
        free_ring_data[i] = i;
    }
    wait_sqe_id = 1;
 }
 ring_loop_t::~ring_loop_t()
@ -35,10 +27,11 @@ ring_loop_t::~ring_loop_t()
    io_uring_queue_exit(&ring);
 }
-void ring_loop_t::register_consumer(ring_consumer_t *consumer)
+int ring_loop_t::register_consumer(ring_consumer_t & consumer)
 {
-    unregister_consumer(consumer);
+    consumer.number = consumers.size();
    consumers.push_back(consumer);
    return consumer.number;
 }
 void ring_loop_t::wakeup()
@ -46,15 +39,12 @@ void ring_loop_t::wakeup()
    loop_again = true;
 }
-void ring_loop_t::unregister_consumer(ring_consumer_t *consumer)
+void ring_loop_t::unregister_consumer(ring_consumer_t & consumer)
 {
-    for (int i = 0; i < consumers.size(); i++)
+    if (consumer.number >= 0 && consumer.number < consumers.size())
    {
-        if (consumers[i] == consumer)
+        consumers[consumer.number].loop = NULL;
-        {
+        consumer.number = -1;
            consumers.erase(consumers.begin()+i, consumers.begin()+i+1);
            break;
        }
    }
 }
@ -66,34 +56,18 @@ void ring_loop_t::loop()
        struct ring_data_t *d = (struct ring_data_t*)cqe->user_data;
        if (d->callback)
        {
-            // First free ring_data item, then call the callback
+            d->res = cqe->res;
-            // so it has at least 1 free slot for the next event
+            d->callback(d);
            // which is required for EPOLLET to function properly
            struct ring_data_t dl;
            dl.iov = d->iov;
            dl.res = cqe->res;
            dl.callback.swap(d->callback);
            free_ring_data[free_ring_data_ptr++] = d - ring_datas;
            dl.callback(&dl);
        }
        else
        {
            printf("Warning: empty callback in SQE\n");
            free_ring_data[free_ring_data_ptr++] = d - ring_datas;
        }
        free_ring_data[free_ring_data_ptr++] = d - ring_datas;
        io_uring_cqe_seen(&ring, cqe);
    }
    while (get_sqe_queue.size() > 0)
    {
        (get_sqe_queue[0].second)();
        get_sqe_queue.erase(get_sqe_queue.begin());
    }
    do
    {
        loop_again = false;
        for (int i = 0; i < consumers.size(); i++)
        {
-            consumers[i]->loop();
+            consumers[i].loop();
        }
    } while (loop_again);
 }
--- a/src/ringloop.h
+++ b/src/ringloop.h
@ -1,6 +1,3 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 #pragma once
 #ifndef _LARGEFILE64_SOURCE
@ -11,7 +8,6 @@
 #include <assert.h>
 #include <liburing.h>
 #include <string>
 #include <functional>
 #include <vector>
@ -117,24 +113,23 @@ struct ring_data_t
 struct ring_consumer_t
 {
    int number;
    std::function<void(void)> loop;
 };
 class ring_loop_t
 {
-    std::vector<std::pair<int,std::function<void()>>> get_sqe_queue;
+    std::vector<ring_consumer_t> consumers;
    std::vector<ring_consumer_t*> consumers;
    struct ring_data_t *ring_datas;
    int *free_ring_data;
    int wait_sqe_id;
    unsigned free_ring_data_ptr;
    bool loop_again;
    struct io_uring ring;
 public:
    ring_loop_t(int qd);
    ~ring_loop_t();
-    void register_consumer(ring_consumer_t *consumer);
+    int register_consumer(ring_consumer_t & consumer);
-    void unregister_consumer(ring_consumer_t *consumer);
+    void unregister_consumer(ring_consumer_t & consumer);
    inline struct io_uring_sqe* get_sqe()
    {
@ -142,27 +137,9 @@ public:
            return NULL;
        struct io_uring_sqe* sqe = io_uring_get_sqe(&ring);
        if (sqe)
        {
            *sqe = { 0 };
            io_uring_sqe_set_data(sqe, ring_datas + free_ring_data[--free_ring_data_ptr]);
        }
        return sqe;
    }
    inline int wait_sqe(std::function<void()> cb)
    {
        get_sqe_queue.push_back({ wait_sqe_id, cb });
        return wait_sqe_id++;
    }
    inline void cancel_wait_sqe(int wait_id)
    {
        for (int i = 0; i < get_sqe_queue.size(); i++)
        {
            if (get_sqe_queue[i].first == wait_id)
            {
                get_sqe_queue.erase(get_sqe_queue.begin()+i, get_sqe_queue.begin()+i+1);
            }
        }
    }
    inline int submit()
    {
        return io_uring_submit(&ring);
@ -176,7 +153,7 @@ public:
    {
        return free_ring_data_ptr;
    }
-    inline bool has_work()
+    inline bool get_loop_again()
    {
        return loop_again;
    }
--- a/rpm/build-tarball.sh
+++ b/rpm/build-tarball.sh
@ -1,51 +0,0 @@
 #!/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 *
--- a/rpm/qemu-el8.Dockerfile
+++ b/rpm/qemu-el8.Dockerfile
@ -1,31 +0,0 @@
 # 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/
--- a/rpm/qemu-kvm-el7.spec.patch
+++ b/rpm/qemu-kvm-el7.spec.patch
@ -1,257 +0,0 @@
 --- 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
--- a/rpm/qemu-kvm.spec.patch
+++ b/rpm/qemu-kvm.spec.patch
@ -1,29 +0,0 @@
 --- 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 \
--- a/rpm/vitastor-el7.Dockerfile
+++ b/rpm/vitastor-el7.Dockerfile
@ -1,47 +0,0 @@
 # 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/
--- a/rpm/vitastor-el7.spec
+++ b/rpm/vitastor-el7.spec
@ -1,69 +0,0 @@
 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
--- a/rpm/vitastor-el8.Dockerfile
+++ b/rpm/vitastor-el8.Dockerfile
@ -1,45 +0,0 @@
 # 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/
--- a/rpm/vitastor-el8.spec
+++ b/rpm/vitastor-el8.spec
@ -1,66 +0,0 @@
 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
--- a/rw_blocking.cpp
+++ b/rw_blocking.cpp
@ -0,0 +1,52 @@
 #include <errno.h>
 #include <stdlib.h>
 #include <stdio.h>
 #include "rw_blocking.h"
 int read_blocking(int fd, void *read_buf, size_t remaining)
 {
    size_t done = 0;
    while (done < remaining)
    {
        size_t r = read(fd, read_buf, remaining-done);
        if (r <= 0)
        {
            if (!errno)
            {
                // EOF
                return done;
            }
            else if (errno != EAGAIN && errno != EPIPE)
            {
                perror("read");
                exit(1);
            }
            continue;
        }
        done += r;
        read_buf += r;
    }
    return done;
 }
 int write_blocking(int fd, void *write_buf, size_t remaining)
 {
    size_t done = 0;
    while (done < remaining)
    {
        size_t r = write(fd, write_buf, remaining-done);
        if (r < 0)
        {
            if (errno != EAGAIN && errno != EPIPE)
            {
                perror("write");
                exit(1);
            }
            continue;
        }
        done += r;
        write_buf += r;
    }
    return done;
 }
--- a/rw_blocking.h
+++ b/rw_blocking.h
@ -0,0 +1,6 @@
 #pragma once
 #include <unistd.h>
 int read_blocking(int fd, void *read_buf, size_t remaining);
 int write_blocking(int fd, void *write_buf, size_t remaining);
--- a/src/CMakeLists.txt
+++ b/src/CMakeLists.txt
@ -1,188 +0,0 @@
 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})
--- a/Show More
+++ b/Show More
		`@ -1 +0,0 @@`
			`Subproject commit 5dc108754ad40d3b1d024f9bd7cca0595ef1a1db`
		`@ -1,2 +0,0 @@`
			`dep:fio=3.16-1`
			`dep:qemu=1:5.1+dfsg-4+vitastor1`
		`@ -1 +0,0 @@`
			`Subproject commit 97f06cb20c1e136fd37d58fb40f57dd8f8a3a4a7`