Release v0.5.4

- Fix a rare hang, more or less reproducible with very slow drives
- Fix a hang with the no_same_sector_overwrites mode

.dockerignore

@@ -0,0 +1,17 @@
+.git
+build
+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

Makefile

@@ -1,10 +1,28 @@
+BINDIR ?= /usr/bin
+LIBDIR ?= /usr/lib/x86_64-linux-gnu
+QEMU_PLUGINDIR ?= /usr/lib/x86_64-linux-gnu/qemu
+
 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
 # -fsanitize=address
-CXXFLAGS := -g -O3 -Wall -Wno-sign-compare -Wno-comment -Wno-parentheses -Wno-pointer-arith -fPIC -fdiagnostics-color=always
+CXXFLAGS := -g -O3 -Wall -Wno-sign-compare -Wno-comment -Wno-parentheses -Wno-pointer-arith -fPIC -fdiagnostics-color=always -I/usr/include/jerasure
 all: libfio_blockstore.so osd libfio_sec_osd.so libfio_cluster.so stub_osd stub_uring_osd stub_bench osd_test dump_journal qemu_driver.so nbd_proxy rm_inode
 clean:
-	rm -f *.o
+	rm -f *.o libblockstore.so libfio_blockstore.so osd libfio_sec_osd.so libfio_cluster.so stub_osd stub_uring_osd stub_bench osd_test dump_journal qemu_driver.so nbd_proxy rm_inode
+
+install: all
+	mkdir -p $(DESTDIR)$(LIBDIR)/vitastor
+	install -m 0755 libfio_sec_osd.so $(DESTDIR)$(LIBDIR)/vitastor/
+	install -m 0755 libfio_cluster.so $(DESTDIR)$(LIBDIR)/vitastor/
+	install -m 0755 libfio_blockstore.so $(DESTDIR)$(LIBDIR)/vitastor/
+	install -m 0755 libblockstore.so $(DESTDIR)$(LIBDIR)/vitastor/
+	mkdir -p $(DESTDIR)$(BINDIR)
+	install -m 0755 osd $(DESTDIR)$(BINDIR)/vitastor-osd
+	install -m 0755 dump_journal $(DESTDIR)$(BINDIR)/vitastor-dump-journal
+	install -m 0755 nbd_proxy $(DESTDIR)$(BINDIR)/vitastor-nbd
+	install -m 0755 rm_inode $(DESTDIR)$(BINDIR)/vitastor-rm
+	mkdir -p $(DESTDIR)$(QEMU_PLUGINDIR)
+	install -m 0755 qemu_driver.so $(DESTDIR)$(QEMU_PLUGINDIR)/block-vitastor.so
 
 dump_journal: dump_journal.cpp crc32c.o blockstore_journal.h
 	g++ $(CXXFLAGS) -o $@ $< crc32c.o
@@ -12,19 +30,19 @@ dump_journal: dump_journal.cpp crc32c.o blockstore_journal.h
 libblockstore.so: $(BLOCKSTORE_OBJS)
 	g++ $(CXXFLAGS) -o $@ -shared $(BLOCKSTORE_OBJS) -ltcmalloc_minimal -luring
 libfio_blockstore.so: ./libblockstore.so fio_engine.o json11.o
-	g++ $(CXXFLAGS) -shared -o $@ fio_engine.o json11.o ./libblockstore.so -ltcmalloc_minimal -luring
+	g++ $(CXXFLAGS) -Wl,-rpath,'$(LIBDIR)/vitastor',-rpath,'$$ORIGIN' -shared -o $@ fio_engine.o json11.o libblockstore.so -ltcmalloc_minimal -luring
 
 OSD_OBJS := osd.o osd_secondary.o msgr_receive.o msgr_send.o osd_peering.o osd_flush.o osd_peering_pg.o \
 	osd_primary.o osd_primary_subops.o etcd_state_client.o messenger.o osd_cluster.o http_client.o osd_ops.o pg_states.o \
 	osd_rmw.o json11.o base64.o timerfd_manager.o epoll_manager.o
 osd: ./libblockstore.so osd_main.cpp osd.h osd_ops.h $(OSD_OBJS)
-	g++ $(CXXFLAGS) -o $@ osd_main.cpp $(OSD_OBJS) ./libblockstore.so -ltcmalloc_minimal -luring
+	g++ $(CXXFLAGS) -Wl,-rpath,'$(LIBDIR)/vitastor',-rpath,'$$ORIGIN' -o $@ osd_main.cpp $(OSD_OBJS) libblockstore.so -ltcmalloc_minimal -luring -lJerasure
 
 stub_osd: stub_osd.o rw_blocking.o
 	g++ $(CXXFLAGS) -o $@ stub_osd.o rw_blocking.o -ltcmalloc_minimal
 
 osd_rmw_test: osd_rmw_test.o
-	g++ $(CXXFLAGS) -o $@ osd_rmw_test.o
+	g++ $(CXXFLAGS) -o $@ osd_rmw_test.o -lJerasure -fsanitize=address
 
 STUB_URING_OSD_OBJS := stub_uring_osd.o epoll_manager.o messenger.o msgr_send.o msgr_receive.o ringloop.o timerfd_manager.o json11.o
 stub_uring_osd: $(STUB_URING_OSD_OBJS)
@@ -55,12 +73,12 @@ qemu_driver.o: qemu_driver.c qemu_proxy.h
 		-I qemu/include $(CXXFLAGS) -c -o $@ $<
 
 qemu_driver.so: qemu_driver.o qemu_proxy.o $(FIO_CLUSTER_OBJS)
-	g++ $(CXXFLAGS) -ltcmalloc_minimal -shared -o $@ $< $(FIO_CLUSTER_OBJS) qemu_driver.o qemu_proxy.o -luring
+	g++ $(CXXFLAGS) -ltcmalloc_minimal -shared -o $@ $(FIO_CLUSTER_OBJS) qemu_driver.o qemu_proxy.o -luring
 
 test_blockstore: ./libblockstore.so test_blockstore.cpp timerfd_interval.o
-	g++ $(CXXFLAGS) -o test_blockstore test_blockstore.cpp timerfd_interval.o ./libblockstore.so -ltcmalloc_minimal -luring
-test: test.cpp osd_peering_pg.o
-	g++ $(CXXFLAGS) -o test test.cpp osd_peering_pg.o -luring -lm
+	g++ $(CXXFLAGS) -Wl,-rpath,'$(LIBDIR)/vitastor',-rpath,'$$ORIGIN' -o test_blockstore test_blockstore.cpp timerfd_interval.o libblockstore.so -ltcmalloc_minimal -luring
+test_shit: test_shit.cpp osd_peering_pg.o
+	g++ $(CXXFLAGS) -o test_shit test_shit.cpp -luring -lm
 test_allocator: test_allocator.cpp allocator.o
 	g++ $(CXXFLAGS) -o test_allocator test_allocator.cpp allocator.o
 
@@ -165,12 +183,12 @@ stub_osd.o: stub_osd.cpp object_id.h osd_id.h osd_ops.h rw_blocking.h
 	g++ $(CXXFLAGS) -c -o $@ $<
 stub_uring_osd.o: stub_uring_osd.cpp epoll_manager.h json11/json11.hpp malloc_or_die.h messenger.h object_id.h osd_id.h osd_ops.h ringloop.h timerfd_manager.h
 	g++ $(CXXFLAGS) -c -o $@ $<
-test.o: test.cpp allocator.h blockstore.h blockstore_flush.h blockstore_impl.h blockstore_init.h blockstore_journal.h cpp-btree/btree_map.h crc32c.h malloc_or_die.h object_id.h osd_id.h osd_ops.h osd_peering_pg.h pg_states.h ringloop.h
-	g++ $(CXXFLAGS) -c -o $@ $<
 test_allocator.o: test_allocator.cpp allocator.h
 	g++ $(CXXFLAGS) -c -o $@ $<
 test_blockstore.o: test_blockstore.cpp blockstore.h object_id.h ringloop.h timerfd_interval.h
 	g++ $(CXXFLAGS) -c -o $@ $<
+test_shit.o: test_shit.cpp allocator.h blockstore.h blockstore_flush.h blockstore_impl.h blockstore_init.h blockstore_journal.h cpp-btree/btree_map.h crc32c.h malloc_or_die.h object_id.h osd_id.h osd_ops.h osd_peering_pg.h pg_states.h ringloop.h
+	g++ $(CXXFLAGS) -c -o $@ $<
 timerfd_interval.o: timerfd_interval.cpp ringloop.h timerfd_interval.h
 	g++ $(CXXFLAGS) -c -o $@ $<
 timerfd_manager.o: timerfd_manager.cpp timerfd_manager.h

README.md

@@ -16,7 +16,8 @@ breaking changes in the future. However, the following is implemented:
 
 - Basic part: highly-available block storage with symmetric clustering and no SPOF
 - Performance ;-D
-- Two redundancy schemes: Replication and XOR n+1 (simplest case of EC)
+- 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
@@ -31,16 +32,14 @@ breaking changes in the future. However, the following is implemented:
 - QEMU driver (built out-of-tree)
 - Loadable fio engine for benchmarks (also built out-of-tree)
 - NBD proxy for kernel mounts
-- Inode removal tool (./rm_inode)
+- Inode removal tool (vitastor-rm)
+- Packaging for Debian and CentOS
 
 ## Roadmap
 
-- Packaging for Debian and, probably, CentOS too
 - OSD creation tool (OSDs currently have to be created by hand)
 - Other administrative tools
 - Per-inode I/O and space usage statistics
-- jerasure EC support with any number of data and parity drives in a group
-- Parallel usage of multiple network interfaces
 - Proxmox and OpenNebula plugins
 - iSCSI proxy
 - Inode metadata storage in etcd
@@ -50,6 +49,7 @@ breaking changes in the future. However, the following is implemented:
 - 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)
 
@@ -280,7 +280,34 @@ Vitastor with single-thread NBD on the same hardware:
 - 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)
 
-## Building
+## 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`
+
+### 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.
@@ -290,10 +317,10 @@ Vitastor with single-thread NBD on the same hardware:
   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 12 or newer.
-- Install gcc and g++ 9.x or later.
+- 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 4.x or 5.x, get its source, begin to build it, stop the build and copy headers:
+- 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
@@ -303,11 +330,15 @@ Vitastor with single-thread NBD on the same hardware:
       * 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`
-      * `<qemu>/qapi` &rarr; `<vitastor>/qemu/b/qemu/qapi`
+      * 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
-- Install fio 3.16 or later, get its source and symlink it into `<vitastor>/fio`.
+- 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 Vitastor with `make -j8`.
-- Copy binaries somewhere.
+- Run `make install` (optionally with `LIBDIR=/usr/lib64 QEMU_PLUGINDIR=/usr/lib64/qemu-kvm`
+  if you're using an RPM-based distro).
 
 ## Running
 
@@ -322,8 +353,9 @@ and calculate disk offsets almost by hand. This will be fixed in near future.
 - 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"}}'`.
-- Calculate offsets for your drives with `node ./mon/simple-offsets.js --device /dev/sdX`.
-- Make systemd units for your OSDs. Look at `./mon/make-units.sh` for example.
+  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"}`.
+- Calculate offsets for your drives with `node /usr/lib/vitastor/mon/simple-offsets.js --device /dev/sdX`.
+- Make systemd units for your OSDs. Look at `/usr/lib/vitastor/mon/make-units.sh` for example.
   Notable configuration variables from the example:
   - `disable_data_fsync 1` - only safe with server-grade drives with capacitors.
   - `immediate_commit all` - use this if all your drives are server-grade.
@@ -342,35 +374,32 @@ and calculate disk offsets almost by hand. This will be fixed in near future.
     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.
-- Start any number of monitors: `cd mon; node mon-main.js --etcd_url 'http://10.115.0.10:2379,http://10.115.0.11:2379,http://10.115.0.12:2379,http://10.115.0.13:2379' --etcd_prefix '/vitastor' --etcd_start_timeout 5`.
+- Start any number of monitors: `node /usr/lib/vitastor/mon/mon-main.js --etcd_url 'http://10.115.0.10:2379,http://10.115.0.11:2379,http://10.115.0.12:2379,http://10.115.0.13:2379' --etcd_prefix '/vitastor' --etcd_start_timeout 5`.
 - 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_cluster.so -name=test -bs=4M -direct=1 -iodepth=16 -rw=write -etcd=10.115.0.10:2379/v3 -pool=1 -inode=1 -size=400G`.
+- Run tests with (for example): `fio -thread -ioengine=/usr/lib/x86_64-linux-gnu/vitastor/libfio_cluster.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):
   ```
-  LD_PRELOAD=./qemu_driver.so qemu-img convert -f qcow2 debian10.qcow2 -p
+  LD_PRELOAD=/usr/lib/x86_64-linux-gnu/qemu/block-vitastor.so 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'
   ```
 - Run QEMU with (for example):
   ```
-  LD_PRELOAD=./qemu_driver.so qemu-system-x86_64 -enable-kvm -m 1024
+  LD_PRELOAD=/usr/lib/x86_64-linux-gnu/qemu/block-vitastor.so 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):
   ```
-  ./rm_inode --etcd_address 10.115.0.10:2379/v3 --pool 1 --inode 1 --parallel_osds 16 --iodepth 32
+  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 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. Inode removal tool currently can't handle unclean
-  objects, so incomplete objects become undeletable. This will be fixed in near future
-  by allowing the inode removal tool to delete unclean objects. With this problem fixed
-  you'll be able just to repeat the removal again.
+  and it's currently not implemented. Just to repeat the removal again in this case.
 
 ## Implementation Principles
 

blockstore_flush.cpp

@@ -76,6 +76,9 @@ void journal_flusher_t::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())
     {
@@ -94,8 +97,11 @@ void journal_flusher_t::enqueue_flush(obj_ver_id ov)
     }
 }
 
-void journal_flusher_t::unshift_flush(obj_ver_id ov)
+void journal_flusher_t::unshift_flush(obj_ver_id ov, bool force)
 {
+#ifdef BLOCKSTORE_DEBUG
+    printf("unshift_flush %lx:%lx v%lu\n", ov.oid.inode, ov.oid.stripe, ov.version);
+#endif
     auto it = flush_versions.find(ov.oid);
     if (it != flush_versions.end())
     {
@@ -105,15 +111,38 @@ void journal_flusher_t::unshift_flush(obj_ver_id ov)
     else
     {
         flush_versions[ov.oid] = ov.version;
-        flush_queue.push_front(ov.oid);
+        if (!force)
+            flush_queue.push_front(ov.oid);
     }
-    if (!dequeuing && (flush_queue.size() >= flusher_start_threshold || trim_wanted > 0))
+    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)
+{
+#ifdef BLOCKSTORE_DEBUG
+    printf("undo_flush %lx:%lx\n", oid.inode, oid.stripe);
+#endif
+    auto v_it = flush_versions.find(oid);
+    if (v_it != flush_versions.end())
+    {
+        flush_versions.erase(v_it);
+        for (auto q_it = flush_queue.begin(); q_it != flush_queue.end(); q_it++)
+        {
+            if (*q_it == oid)
+            {
+                flush_queue.erase(q_it);
+                break;
+            }
+        }
+    }
+}
+
 void journal_flusher_t::request_trim()
 {
     dequeuing = true;
@@ -194,6 +223,7 @@ bool journal_flusher_co::loop()
 resume_0:
     if (!flusher->flush_queue.size() || !flusher->dequeuing)
     {
+stop_flusher:
         if (flusher->trim_wanted > 0 && flusher->journal_trim_counter > 0)
         {
             // Attempt forced trim
@@ -298,9 +328,7 @@ resume_0:
 #ifdef BLOCKSTORE_DEBUG
                     printf("No older flushes, stopping\n");
 #endif
-                    flusher->dequeuing = false;
-                    wait_state = 0;
-                    return true;
+                    goto stop_flusher;
                 }
             }
         }
@@ -319,8 +347,8 @@ resume_1:
             return false;
         }
         // Writes and deletes shouldn't happen at the same time
-        assert(!(copy_count > 0 || has_writes) || !has_delete);
-        if (copy_count == 0 && !has_writes && !has_delete || has_delete && old_clean_loc == UINT64_MAX)
+        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);
@@ -445,8 +473,8 @@ resume_1:
             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 %lx (%lx:%lx) with %lx:%lx\n",
-                    clean_loc, new_entry->oid.inode, new_entry->oid.stripe, cur.oid.inode, cur.oid.stripe);
+                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;
@@ -513,7 +541,7 @@ resume_1:
         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->unshift_flush({ .oid = cur.oid, .version = repeat_it->second }, false);
         }
         flusher->sync_to_repeat.erase(repeat_it);
     trim_journal:
@@ -602,7 +630,7 @@ bool journal_flusher_co::scan_dirty(int wait_base)
         {
             char err[1024];
             snprintf(
-                err, 1024, "BUG: Unexpected dirty_entry %lx:%lx v%lu state during flush: %d",
+                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);

blockstore_flush.h

@@ -107,5 +107,6 @@ public:
     void request_trim();
     void release_trim();
     void enqueue_flush(obj_ver_id oid);
-    void unshift_flush(obj_ver_id oid);
+    void unshift_flush(obj_ver_id oid, bool force);
+    void remove_flush(object_id oid);
 };

blockstore_impl.cpp

@@ -287,7 +287,7 @@ void blockstore_impl_t::check_wait(blockstore_op_t *op)
     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 ||
+        if (journal.sector_info[next].flush_count > 0 ||
             journal.sector_info[next].dirty)
         {
             // do not submit

blockstore_impl.h

@@ -30,12 +30,13 @@
 #define BS_ST_BIG_WRITE 0x02
 #define BS_ST_DELETE 0x03
 
-#define BS_ST_WAIT_BIG 0x10
-#define BS_ST_IN_FLIGHT 0x20
-#define BS_ST_SUBMITTED 0x30
-#define BS_ST_WRITTEN 0x40
-#define BS_ST_SYNCED 0x50
-#define BS_ST_STABLE 0x60
+#define BS_ST_WAIT_DEL 0x10
+#define BS_ST_WAIT_BIG 0x20
+#define BS_ST_IN_FLIGHT 0x30
+#define BS_ST_SUBMITTED 0x40
+#define BS_ST_WRITTEN 0x50
+#define BS_ST_SYNCED 0x60
+#define BS_ST_STABLE 0x70
 
 #define BS_ST_INSTANT 0x100
 
@@ -153,6 +154,8 @@ struct blockstore_op_private_t
 
     // 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;

blockstore_init.cpp

@@ -111,7 +111,7 @@ 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 >> block_order);
+                    printf("Free block %lu (new location is %lu)\n", clean_it->second.location >> block_order, done_cnt+i);
 #endif
                     bs->data_alloc->set(clean_it->second.location >> block_order, false);
                 }
@@ -557,11 +557,56 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
             {
 #ifdef BLOCKSTORE_DEBUG
                 printf(
-                    "je_big_write%s oid=%lx:%lx ver=%lu loc=%08lx\n",
+                    "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
+                    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)

blockstore_journal.cpp

@@ -6,7 +6,7 @@
 blockstore_journal_check_t::blockstore_journal_check_t(blockstore_impl_t *bs)
 {
     this->bs = bs;
-    sectors_required = 0;
+    sectors_to_write = 0;
     next_pos = bs->journal.next_free;
     next_sector = bs->journal.cur_sector;
     first_sector = -1;
@@ -20,23 +20,26 @@ int blockstore_journal_check_t::check_available(blockstore_op_t *op, int entries
     int required = entries_required;
     while (1)
     {
-        int fits = bs->journal.no_same_sector_overwrites && bs->journal.sector_info[next_sector].written
+        int fits = bs->journal.no_same_sector_overwrites && next_pos == bs->journal.next_free && bs->journal.sector_info[next_sector].written
             ? 0
             : (bs->journal.block_size - next_in_pos) / size;
         if (fits > 0)
         {
+            if (fits > required)
+            {
+                fits = required;
+            }
             if (first_sector == -1)
             {
                 first_sector = next_sector;
             }
             required -= fits;
             next_in_pos += fits * size;
-            sectors_required++;
+            sectors_to_write++;
         }
         else if (bs->journal.sector_info[next_sector].dirty)
         {
-            // sectors_required is more like "sectors to write"
-            sectors_required++;
+            sectors_to_write++;
         }
         if (required <= 0)
         {
@@ -59,7 +62,7 @@ int blockstore_journal_check_t::check_available(blockstore_op_t *op, int entries
                 " is too small for a batch of "+std::to_string(entries_required)+" entries of "+std::to_string(size)+" bytes"
             );
         }
-        if (bs->journal.sector_info[next_sector].usage_count > 0 ||
+        if (bs->journal.sector_info[next_sector].flush_count > 0 ||
             bs->journal.sector_info[next_sector].dirty)
         {
             // No memory buffer available. Wait for it.
@@ -71,17 +74,18 @@ int blockstore_journal_check_t::check_available(blockstore_op_t *op, int entries
                     dirty++;
                     used++;
                 }
-                if (bs->journal.sector_info[i].usage_count > 0)
+                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\n",
+                "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].usage_count
+                bs->journal.sector_info[next_sector].flush_count
             );
             PRIV(op)->wait_for = WAIT_JOURNAL_BUFFER;
             return 0;
@@ -100,11 +104,8 @@ 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 (free space: %lu bytes, sectors to write: %d)\n",
-            (bs->journal.next_free >= bs->journal.used_start
-                ? bs->journal.len-bs->journal.block_size - (bs->journal.next_free-bs->journal.used_start)
-                : bs->journal.used_start - bs->journal.next_free),
-            sectors_required
+            "Ran out of journal space (used_start=%08lx, next_free=%08lx, dirty_start=%08lx)\n",
+            bs->journal.used_start, bs->journal.next_free, bs->journal.dirty_start
         );
         PRIV(op)->wait_for = WAIT_JOURNAL;
         bs->flusher->request_trim();
@@ -116,22 +117,21 @@ 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.block_size - journal.in_sector_pos < size ||
-        journal.no_same_sector_overwrites && journal.sector_info[journal.cur_sector].written)
+    if (!journal.entry_fits(size))
     {
         assert(!journal.sector_info[journal.cur_sector].dirty);
         // Move to the next journal sector
-        journal.sector_info[journal.cur_sector].written = false;
-        if (journal.sector_info[journal.cur_sector].usage_count > 0)
+        if (journal.sector_info[journal.cur_sector].flush_count > 0)
         {
             // Also select next sector buffer in memory
             journal.cur_sector = ((journal.cur_sector + 1) % journal.sector_count);
-            assert(!journal.sector_info[journal.cur_sector].usage_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;
@@ -157,7 +157,7 @@ void prepare_journal_sector_write(journal_t & journal, int cur_sector, io_uring_
 {
     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

blockstore_journal.h

@@ -133,7 +133,7 @@ inline uint32_t je_crc32(journal_entry *je)
 struct journal_sector_info_t
 {
     uint64_t offset;
-    uint64_t usage_count;
+    uint64_t flush_count;
     bool written;
     bool dirty;
 };
@@ -170,13 +170,18 @@ struct journal_t
     ~journal_t();
     bool trim();
     uint64_t get_trim_pos();
+    inline bool entry_fits(int size)
+    {
+        return !(block_size - in_sector_pos < size ||
+            no_same_sector_overwrites && sector_info[cur_sector].written);
+    }
 };
 
 struct blockstore_journal_check_t
 {
     blockstore_impl_t *bs;
     uint64_t next_pos, next_sector, next_in_pos;
-    int sectors_required, first_sector;
+    int sectors_to_write, first_sector;
     bool right_dir; // writing to the end or the beginning of the ring buffer
 
     blockstore_journal_check_t(blockstore_impl_t *bs);

blockstore_rollback.cpp

@@ -75,44 +75,35 @@ skip_ov:
         return 0;
     }
     // There is sufficient space. Get SQEs
-    struct io_uring_sqe *sqe[space_check.sectors_required];
-    for (i = 0; i < space_check.sectors_required; i++)
+    struct io_uring_sqe *sqe[space_check.sectors_to_write];
+    for (i = 0; i < space_check.sectors_to_write; i++)
     {
         BS_SUBMIT_GET_SQE_DECL(sqe[i]);
     }
     // Prepare and submit journal entries
     auto cb = [this, op](ring_data_t *data) { handle_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_flushed_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++)
     {
+        if (!journal.entry_fits(sizeof(journal_entry_rollback)) &&
+            journal.sector_info[journal.cur_sector].dirty)
+        {
+            if (cur_sector == -1)
+                PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
+            prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], cb);
+            cur_sector = journal.cur_sector;
+        }
         journal_entry_rollback *je = (journal_entry_rollback*)
             prefill_single_journal_entry(journal, JE_ROLLBACK, sizeof(journal_entry_rollback));
-        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)
-        {
-            // Write previous sector. We should write the sector only after filling it,
-            // because otherwise we'll write a lot more sectors in the "no_same_sector_overwrite" mode
-            if (cur_sector != -1)
-                prepare_journal_sector_write(journal, cur_sector, sqe[s++], cb);
-            else
-                PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
-            cur_sector = journal.cur_sector;
-        }
     }
-    if (cur_sector != -1)
-        prepare_journal_sector_write(journal, cur_sector, sqe[s++], cb);
+    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;
@@ -234,10 +225,38 @@ void blockstore_impl_t::handle_rollback_event(ring_data_t *data, blockstore_op_t
 
 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)
+    if (dirty_end == dirty_start)
     {
+        return;
+    }
+    auto dirty_it = dirty_end;
+    dirty_it--;
+    if (IS_DELETE(dirty_it->second.state))
+    {
+        object_id oid = dirty_it->first.oid;
+#ifdef BLOCKSTORE_DEBUG
+        printf("Unblock writes-after-delete %lx:%lx v%lx\n", oid.inode, oid.stripe, dirty_it->first.version);
+#endif
+        dirty_it = dirty_end;
+        // Unblock operations blocked by delete flushing
+        uint32_t next_state = BS_ST_IN_FLIGHT;
+        while (dirty_it != dirty_db.end() && dirty_it->first.oid == oid)
+        {
+            if ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_WAIT_DEL)
+            {
+                dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | next_state;
+                if (IS_BIG_WRITE(dirty_it->second.state))
+                {
+                    next_state = BS_ST_WAIT_BIG;
+                }
+            }
+            dirty_it++;
+        }
+        dirty_it = dirty_end;
         dirty_it--;
+    }
+    while (1)
+    {
         if (IS_BIG_WRITE(dirty_it->second.state) && dirty_it->second.location != clean_loc)
         {
 #ifdef BLOCKSTORE_DEBUG
@@ -256,6 +275,11 @@ void blockstore_impl_t::erase_dirty(blockstore_dirty_db_t::iterator dirty_start,
         {
             journal.used_sectors.erase(dirty_it->second.journal_sector);
         }
+        if (dirty_it == dirty_start)
+        {
+            break;
+        }
+        dirty_it--;
     }
     dirty_db.erase(dirty_start, dirty_end);
 }

blockstore_stable.cpp

@@ -98,45 +98,36 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
         return 0;
     }
     // There is sufficient space. Get SQEs
-    struct io_uring_sqe *sqe[space_check.sectors_required];
-    for (i = 0; i < space_check.sectors_required; i++)
+    struct io_uring_sqe *sqe[space_check.sectors_to_write];
+    for (i = 0; i < space_check.sectors_to_write; i++)
     {
         BS_SUBMIT_GET_SQE_DECL(sqe[i]);
     }
     // Prepare and submit journal entries
     auto cb = [this, op](ring_data_t *data) { handle_stable_event(data, op); };
     int s = 0, cur_sector = -1;
-    if ((journal_block_size - journal.in_sector_pos) < sizeof(journal_entry_stable) &&
-        journal.sector_info[journal.cur_sector].dirty)
-    {
-        if (cur_sector == -1)
-            PRIV(op)->min_flushed_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
+        if (!journal.entry_fits(sizeof(journal_entry_stable)) &&
+            journal.sector_info[journal.cur_sector].dirty)
+        {
+            if (cur_sector == -1)
+                PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
+            prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], cb);
+            cur_sector = journal.cur_sector;
+        }
         journal_entry_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)
-        {
-            // Write previous sector. We should write the sector only after filling it,
-            // because otherwise we'll write a lot more sectors in the "no_same_sector_overwrite" mode
-            if (cur_sector != -1)
-                prepare_journal_sector_write(journal, cur_sector, sqe[s++], cb);
-            else
-                PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
-            cur_sector = journal.cur_sector;
-        }
     }
-    if (cur_sector != -1)
-        prepare_journal_sector_write(journal, cur_sector, sqe[s++], cb);
+    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;
@@ -213,9 +204,6 @@ void blockstore_impl_t::mark_stable(const obj_ver_id & v)
                 break;
             }
         }
-#ifdef BLOCKSTORE_DEBUG
-        printf("enqueue_flush %lx:%lx v%lu\n", v.oid.inode, v.oid.stripe, v.version);
-#endif
         flusher->enqueue_flush(v);
     }
     auto unstab_it = unstable_writes.find(v.oid);

blockstore_sync.cpp

@@ -112,30 +112,29 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op)
             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++)
+        struct io_uring_sqe *sqe[space_check.sectors_to_write];
+        for (int i = 0; i < space_check.sectors_to_write; i++)
         {
             BS_SUBMIT_GET_SQE_DECL(sqe[i]);
         }
         // Prepare and submit journal entries
         auto it = PRIV(op)->sync_big_writes.begin();
         int s = 0, cur_sector = -1;
-        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_flushed_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())
         {
+            if (!journal.entry_fits(sizeof(journal_entry_big_write)) &&
+                journal.sector_info[journal.cur_sector].dirty)
+            {
+                if (cur_sector == -1)
+                    PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
+                prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], cb);
+                cur_sector = journal.cur_sector;
+            }
             journal_entry_big_write *je = (journal_entry_big_write*)prefill_single_journal_entry(
                 journal, (dirty_db[*it].state & BS_ST_INSTANT) ? JE_BIG_WRITE_INSTANT : JE_BIG_WRITE,
                 sizeof(journal_entry_big_write)
             );
             dirty_db[*it].journal_sector = journal.sector_info[journal.cur_sector].offset;
-            journal.sector_info[journal.cur_sector].dirty = false;
             journal.used_sectors[journal.sector_info[journal.cur_sector].offset]++;
 #ifdef BLOCKSTORE_DEBUG
             printf(
@@ -152,19 +151,11 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op)
             je->crc32 = je_crc32((journal_entry*)je);
             journal.crc32_last = je->crc32;
             it++;
-            if (cur_sector != journal.cur_sector)
-            {
-                // Write previous sector. We should write the sector only after filling it,
-                // because otherwise we'll write a lot more sectors in the "no_same_sector_overwrite" mode
-                if (cur_sector != -1)
-                    prepare_journal_sector_write(journal, cur_sector, sqe[s++], cb);
-                else
-                    PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
-                cur_sector = journal.cur_sector;
-            }
         }
-        if (cur_sector != -1)
-            prepare_journal_sector_write(journal, cur_sector, sqe[s++], cb);
+        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 = SYNC_JOURNAL_WRITE_SENT;

blockstore_write.cpp

@@ -7,7 +7,7 @@ bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
 {
     // Check or assign version number
     bool found = false, deleted = false, is_del = (op->opcode == BS_OP_DELETE);
-    bool is_inflight_big = false;
+    bool wait_big = false, wait_del = false;
     uint64_t version = 1;
     if (dirty_db.size() > 0)
     {
@@ -21,7 +21,8 @@ bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
             found = true;
             version = dirty_it->first.version + 1;
             deleted = IS_DELETE(dirty_it->second.state);
-            is_inflight_big = (dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_BIG_WRITE
+            wait_del = ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_WAIT_DEL);
+            wait_big = (dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_BIG_WRITE
                 ? !IS_SYNCED(dirty_it->second.state)
                 : ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_WAIT_BIG);
         }
@@ -38,23 +39,43 @@ bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
             deleted = true;
         }
     }
-    if (op->version == 0)
-    {
-        op->version = version;
-    }
-    else if (op->version < version)
-    {
-        // Invalid version requested
-        op->retval = -EEXIST;
-        return false;
-    }
     if (deleted && is_del)
     {
         // Already deleted
         op->retval = 0;
         return false;
     }
-    if (is_inflight_big && !is_del && !deleted && op->len < block_size &&
+    PRIV(op)->real_version = 0;
+    if (op->version == 0)
+    {
+        op->version = version;
+    }
+    else if (op->version < version)
+    {
+        // Implicit operations must be added like that: DEL [FLUSH] BIG [SYNC] SMALL SMALL
+        if (deleted || wait_del)
+        {
+            // It's allowed to write versions with low numbers over deletes
+            // However, we have to flush those deletes first as we use version number for ordering
+#ifdef BLOCKSTORE_DEBUG
+            printf("Write %lx:%lx v%lu over delete (real v%lu) offset=%u len=%u\n", op->oid.inode, op->oid.stripe, version, op->version, op->offset, op->len);
+#endif
+            wait_del = true;
+            PRIV(op)->real_version = op->version;
+            op->version = version;
+            flusher->unshift_flush((obj_ver_id){
+                .oid = op->oid,
+                .version = version-1,
+            }, true);
+        }
+        else
+        {
+            // Invalid version requested
+            op->retval = -EEXIST;
+            return false;
+        }
+    }
+    if (wait_big && !is_del && !deleted && op->len < block_size &&
         immediate_commit != IMMEDIATE_ALL)
     {
         // Issue an additional sync so that the previous big write can reach the journal
@@ -69,22 +90,31 @@ bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
 #ifdef BLOCKSTORE_DEBUG
     if (is_del)
         printf("Delete %lx:%lx v%lu\n", op->oid.inode, op->oid.stripe, op->version);
-    else
+    else if (!wait_del)
         printf("Write %lx:%lx v%lu offset=%u len=%u\n", op->oid.inode, op->oid.stripe, op->version, op->offset, op->len);
 #endif
-    // No strict need to add it into dirty_db here, it's just left
+    // FIXME No strict need to add it into dirty_db here, it's just left
     // from the previous implementation where reads waited for writes
+    uint32_t state;
+    if (is_del)
+        state = BS_ST_DELETE | BS_ST_IN_FLIGHT;
+    else
+    {
+        state = (op->len == block_size || deleted ? BS_ST_BIG_WRITE : BS_ST_SMALL_WRITE);
+        if (wait_del)
+            state |= BS_ST_WAIT_DEL;
+        else if (state == BS_ST_SMALL_WRITE && wait_big)
+            state |= BS_ST_WAIT_BIG;
+        else
+            state |= BS_ST_IN_FLIGHT;
+        if (op->opcode == BS_OP_WRITE_STABLE)
+            state |= BS_ST_INSTANT;
+    }
     dirty_db.emplace((obj_ver_id){
         .oid = op->oid,
         .version = op->version,
     }, (dirty_entry){
-        .state = (uint32_t)(
-            is_del
-                ? (BS_ST_DELETE | BS_ST_IN_FLIGHT)
-                : (op->opcode == BS_OP_WRITE_STABLE ? BS_ST_INSTANT : 0) | (op->len == block_size || deleted
-                    ? (BS_ST_BIG_WRITE | BS_ST_IN_FLIGHT)
-                    : (is_inflight_big ? (BS_ST_SMALL_WRITE | BS_ST_WAIT_BIG) : (BS_ST_SMALL_WRITE | BS_ST_IN_FLIGHT)))
-        ),
+        .state = state,
         .flags = 0,
         .location = 0,
         .offset = is_del ? 0 : op->offset,
@@ -106,12 +136,38 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
         .version = op->version,
     });
     assert(dirty_it != dirty_db.end());
-    if ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_WAIT_BIG)
+    if ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) < BS_ST_IN_FLIGHT)
     {
         // Don't dequeue
         return 0;
     }
-    else if ((dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_BIG_WRITE)
+    if (PRIV(op)->real_version != 0)
+    {
+        // Restore original low version number for unblocked operations
+#ifdef BLOCKSTORE_DEBUG
+        printf("Restoring %lx:%lx version: v%lu -> v%lu\n", op->oid.inode, op->oid.stripe, op->version, PRIV(op)->real_version);
+#endif
+        auto prev_it = dirty_it;
+        prev_it--;
+        if (prev_it->first.oid == op->oid && prev_it->first.version >= PRIV(op)->real_version)
+        {
+            // Original version is still invalid
+            // FIXME Oops. Successive small writes will currently break in an unexpected way. Fix it
+            dirty_db.erase(dirty_it);
+            op->retval = -EEXIST;
+            FINISH_OP(op);
+            return 1;
+        }
+        op->version = PRIV(op)->real_version;
+        PRIV(op)->real_version = 0;
+        dirty_entry e = dirty_it->second;
+        dirty_db.erase(dirty_it);
+        dirty_it = dirty_db.emplace((obj_ver_id){
+            .oid = op->oid,
+            .version = op->version,
+        }, e).first;
+    }
+    if ((dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_BIG_WRITE)
     {
         blockstore_journal_check_t space_check(this);
         if (!space_check.check_available(op, unsynced_big_writes.size() + 1, sizeof(journal_entry_big_write), JOURNAL_STABILIZE_RESERVATION))
@@ -129,6 +185,8 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
                 PRIV(op)->wait_for = WAIT_FREE;
                 return 0;
             }
+            // FIXME Oops. Successive small writes will currently break in an unexpected way. Fix it
+            dirty_db.erase(dirty_it);
             op->retval = -ENOSPC;
             FINISH_OP(op);
             return 1;
@@ -319,7 +377,6 @@ resume_2:
         sizeof(journal_entry_big_write)
     );
     dirty_it->second.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(
@@ -344,7 +401,7 @@ resume_2:
 resume_4:
     // Switch object state
 #ifdef BLOCKSTORE_DEBUG
-    printf("Ack write %lx:%lx v%lu = %d\n", op->oid.inode, op->oid.stripe, op->version, dirty_it->second.state);
+    printf("Ack write %lx:%lx v%lu = state %x\n", op->oid.inode, op->oid.stripe, op->version, dirty_it->second.state);
 #endif
     bool imm = (dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_BIG_WRITE
         ? (immediate_commit == IMMEDIATE_ALL)
@@ -411,8 +468,8 @@ void blockstore_impl_t::release_journal_sectors(blockstore_op_t *op)
         uint64_t s = PRIV(op)->min_flushed_journal_sector;
         while (1)
         {
-            journal.sector_info[s-1].usage_count--;
-            if (s != (1+journal.cur_sector) && journal.sector_info[s-1].usage_count == 0)
+            journal.sector_info[s-1].flush_count--;
+            if (s != (1+journal.cur_sector) && journal.sector_info[s-1].flush_count == 0)
             {
                 // We know for sure that we won't write into this sector anymore
                 uint64_t new_ds = journal.sector_info[s-1].offset + journal.block_size;
@@ -492,7 +549,10 @@ int blockstore_impl_t::dequeue_del(blockstore_op_t *op)
         prepare_journal_sector_write(journal, journal.cur_sector, sqe, cb);
         PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
         PRIV(op)->pending_ops++;
-        // Remember small write as unsynced
+    }
+    else
+    {
+        // Remember delete as unsynced
         unsynced_small_writes.push_back((obj_ver_id){
             .oid = op->oid,
             .version = op->version,

cluster_client.cpp

@@ -473,7 +473,7 @@ void cluster_client_t::slice_rw(cluster_op_t *op)
     // Primary OSDs still operate individual stripes, but their size is multiplied by PG minsize in case of EC
     auto & pool_cfg = st_cli.pool_config[INODE_POOL(op->inode)];
     uint64_t pg_block_size = bs_block_size * (
-        pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_minsize
+        pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks
     );
     uint64_t first_stripe = (op->offset / pg_block_size) * pg_block_size;
     uint64_t last_stripe = ((op->offset + op->len + pg_block_size - 1) / pg_block_size - 1) * pg_block_size;
@@ -488,7 +488,7 @@ void cluster_client_t::slice_rw(cluster_op_t *op)
         uint64_t begin = (op->offset < stripe ? stripe : op->offset);
         uint64_t end = (op->offset + op->len) > (stripe + pg_block_size)
             ? (stripe + pg_block_size) : (op->offset + op->len);
-        op->parts[i] = {
+        op->parts[i] = (cluster_op_part_t){
             .parent = op,
             .offset = begin,
             .len = (uint32_t)(end - begin),
@@ -533,7 +533,7 @@ bool cluster_client_t::try_send(cluster_op_t *op, cluster_op_part_t *part)
             part->osd_num = primary_osd;
             part->sent = true;
             op->sent_count++;
-            part->op = {
+            part->op = (osd_op_t){
                 .op_type = OSD_OP_OUT,
                 .peer_fd = peer_fd,
                 .req = { .rw = {
@@ -694,7 +694,7 @@ void cluster_client_t::send_sync(cluster_op_t *op, cluster_op_part_t *part)
     assert(peer_it != msgr.osd_peer_fds.end());
     part->sent = true;
     op->sent_count++;
-    part->op = {
+    part->op = (osd_op_t){
         .op_type = OSD_OP_OUT,
         .peer_fd = peer_it->second,
         .req = {

copy-fio-includes.sh

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

copy-qemu-includes.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+
+#cd qemu
+#debian/rules b/configure-stamp
+#cd b/qemu; make qapi
+
+gcc -I qemu/b/qemu `pkg-config glib-2.0 --cflags` \
+    -I qemu/include -E -o qemu_driver.i 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

debian/changelog

@@ -0,0 +1,17 @@
+vitastor (0.5.4-1) unstable; urgency=medium
+
+  * Bugfixes
+
+ -- Vitaliy Filippov <vitalif@yourcmc.ru>  Tue, 02 Feb 2021 23:01:24 +0300
+
+vitastor (0.5.1-1) unstable; urgency=medium
+
+  * Add jerasure support
+
+ -- Vitaliy Filippov <vitalif@yourcmc.ru>  Sat, 05 Dec 2020 17:02:26 +0300
+
+vitastor (0.5-1) unstable; urgency=medium
+
+  * First packaging for Debian
+
+ -- Vitaliy Filippov <vitalif@yourcmc.ru>  Thu, 05 Nov 2020 02:20:59 +0300

debian/compat

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

debian/control

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

debian/copyright

@@ -0,0 +1,20 @@
+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-1.0 and/or GPL-2.0+
+ All server-side code (OSD, Monitor and so on) is licensed under the terms of
+ Vitastor Network Public License 1.0 (VNPL 1.0), 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 ("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 1.0 and also GNU GPL 2.0 or later to allow for compatibility with GPLed
+ software like QEMU and fio.

debian/install

@@ -0,0 +1,3 @@
+VNPL-1.0.txt usr/share/doc/vitastor
+GPL-2.0.txt usr/share/doc/vitastor
+mon usr/lib/vitastor

debian/patched-qemu.Dockerfile

@@ -0,0 +1,49 @@
+# Build patched QEMU for Debian Buster or Bullseye/Sid inside a container
+# cd ..; podman build --build-arg REL=bullseye -v `pwd`/build:/root/build -f debian/patched-qemu.Dockerfile .
+
+ARG REL=bullseye
+
+FROM debian:$REL
+
+# again, it doesn't work otherwise
+ARG REL=bullseye
+
+WORKDIR /root
+
+RUN if [ "$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/build/qemu-$REL; \
+    rm -rf /root/build/qemu-$REL/*; \
+    cd /root/build/qemu-$REL; \
+    dpkg-source -x /root/qemu*.dsc; \
+    if [ -d /root/build/qemu-$REL/qemu-5.0 ]; then \
+        cp /root/vitastor/qemu-5.0-vitastor.patch /root/build/qemu-$REL/qemu-5.0/debian/patches; \
+        echo qemu-5.0-vitastor.patch >> /root/build/qemu-$REL/qemu-5.0/debian/patches/series; \
+    else \
+        cp /root/vitastor/qemu-5.1-vitastor.patch /root/build/qemu-$REL/qemu-*/debian/patches; \
+        P=`ls -d /root/build/qemu-$REL/qemu-*/debian/patches`; \
+        echo qemu-5.1-vitastor.patch >> $P/series; \
+    fi; \
+    cd /root/build/qemu-$REL/qemu-*/; \
+    V=$(head -n1 debian/changelog | perl -pe 's/^.*\((.*?)(~bpo[\d\+]*)?\).*$/$1/')+vitastor1; \
+    DEBFULLNAME="Vitaliy Filippov <vitalif@yourcmc.ru>" dch -D $REL -v $V 'Plug Vitastor block driver'; \
+    DEB_BUILD_OPTIONS=nocheck dpkg-buildpackage --jobs=auto -sa; \
+    rm -rf /root/build/qemu-$REL/qemu-*/

debian/rules

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

debian/source/format

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

debian/substvars

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

debian/vitastor.Dockerfile

@@ -0,0 +1,72 @@
+# Build Vitastor packages for Debian Buster or Bullseye/Sid inside a container
+# cd ..; podman build --build-arg REL=bullseye -v `pwd`/build:/root/build -f debian/vitastor.Dockerfile .
+
+ARG REL=bullseye
+
+FROM debian:$REL
+
+# again, it doesn't work otherwise
+ARG REL=bullseye
+
+WORKDIR /root
+
+RUN if [ "$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
+
+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/build/qemu-$REL/; \
+    rm -rf qemu*/; \
+    dpkg-source -x qemu*.dsc; \
+    cd /root/build/qemu-$REL/qemu*/; \
+    debian/rules b/configure-stamp; \
+    cd b/qemu; \
+    make -j8 qapi; \
+    mkdir -p /root/build/vitastor-$REL; \
+    rm -rf /root/build/vitastor-$REL/*; \
+    cd /root/build/vitastor-$REL; \
+    cp -r /root/vitastor vitastor-0.5.4; \
+    ln -s /root/build/qemu-$REL/qemu-*/ vitastor-0.5.4/qemu; \
+    ln -s /root/fio-build/fio-*/ vitastor-0.5.4/fio; \
+    cd vitastor-0.5.4; \
+    FIO=$(head -n1 fio/debian/changelog | perl -pe 's/^.*\((.*?)\).*$/$1/'); \
+    QEMU=$(head -n1 qemu/debian/changelog | perl -pe 's/^.*\((.*?)\).*$/$1/'); \
+    sh copy-qemu-includes.sh; \
+    sh copy-fio-includes.sh; \
+    rm qemu fio; \
+    mkdir -p a b debian/patches; \
+    mv qemu-copy b/qemu; \
+    mv fio-copy b/fio; \
+    diff -NaurpbB a b > debian/patches/qemu-fio-headers.patch || true; \
+    echo qemu-fio-headers.patch >> debian/patches/series; \
+    rm -rf a b; \
+    rm -rf /root/build/qemu-$REL/qemu*/; \
+    echo "dep:fio=$FIO" > debian/substvars; \
+    echo "dep:qemu=$QEMU" >> debian/substvars; \
+    cd /root/build/vitastor-$REL; \
+    tar --sort=name --mtime='2020-01-01' --owner=0 --group=0 --exclude=debian -cJf vitastor_0.5.4.orig.tar.xz vitastor-0.5.4; \
+    cd vitastor-0.5.4; \
+    V=$(head -n1 debian/changelog | perl -pe 's/^.*\((.*?)\).*$/$1/'); \
+    DEBFULLNAME="Vitaliy Filippov <vitalif@yourcmc.ru>" dch -D $REL -v "$V""$REL" "Rebuild for $REL"; \
+    DEB_BUILD_OPTIONS=nocheck dpkg-buildpackage --jobs=auto -sa; \
+    rm -rf /root/build/vitastor-$REL/vitastor-*/

dump_journal.cpp

@@ -26,23 +26,32 @@ struct journal_dump_t
     uint64_t journal_offset;
     uint64_t journal_len;
     uint64_t journal_pos;
+    bool all;
+    bool started;
     int fd;
+    uint32_t crc32_last;
 
-    void dump_block(void *buf);
+    int dump_block(void *buf);
 };
 
 int main(int argc, char *argv[])
 {
-    if (argc < 5)
+    journal_dump_t self = { 0 };
+    int b = 1;
+    if (argc >= 2 && !strcmp(argv[1], "--all"))
     {
-        printf("USAGE: %s <journal_file> <journal_block_size> <offset> <size>\n", argv[0]);
+        self.all = true;
+        b = 2;
+    }
+    if (argc < b+4)
+    {
+        printf("USAGE: %s [--all] <journal_file> <journal_block_size> <offset> <size>\n", argv[0]);
         return 1;
     }
-    journal_dump_t self;
-    self.journal_device = argv[1];
-    self.journal_block = strtoul(argv[2], NULL, 10);
-    self.journal_offset = strtoull(argv[3], NULL, 10);
-    self.journal_len = strtoull(argv[4], NULL, 10);
+    self.journal_device = argv[b];
+    self.journal_block = strtoul(argv[b+1], NULL, 10);
+    self.journal_offset = strtoull(argv[b+2], NULL, 10);
+    self.journal_len = strtoull(argv[b+3], NULL, 10);
     if (self.journal_block < MEM_ALIGNMENT || (self.journal_block % MEM_ALIGNMENT) ||
         self.journal_block > 128*1024)
     {
@@ -57,30 +66,64 @@ int main(int argc, char *argv[])
     }
     void *data = memalign(MEM_ALIGNMENT, self.journal_block);
     self.journal_pos = 0;
-    while (self.journal_pos < self.journal_len)
+    if (self.all)
+    {
+        while (self.journal_pos < self.journal_len)
+        {
+            int r = pread(self.fd, data, self.journal_block, self.journal_offset+self.journal_pos);
+            assert(r == self.journal_block);
+            uint64_t s;
+            for (s = 0; s < self.journal_block; s += 8)
+            {
+                if (*((uint64_t*)(data+s)) != 0)
+                    break;
+            }
+            if (s == self.journal_block)
+            {
+                printf("offset %08lx: zeroes\n", self.journal_pos);
+                self.journal_pos += self.journal_block;
+            }
+            else if (((journal_entry*)data)->magic == JOURNAL_MAGIC)
+            {
+                printf("offset %08lx:\n", self.journal_pos);
+                self.dump_block(data);
+            }
+            else
+            {
+                printf("offset %08lx: no magic in the beginning, looks like random data (pattern=%lx)\n", self.journal_pos, *((uint64_t*)data));
+                self.journal_pos += self.journal_block;
+            }
+        }
+    }
+    else
     {
         int r = pread(self.fd, data, self.journal_block, self.journal_offset+self.journal_pos);
         assert(r == self.journal_block);
-        uint64_t s;
-        for (s = 0; s < self.journal_block; s += 8)
+        journal_entry *je = (journal_entry*)(data);
+        if (je->magic != JOURNAL_MAGIC || je->type != JE_START || je_crc32(je) != je->crc32)
         {
-            if (*((uint64_t*)(data+s)) != 0)
-                break;
-        }
-        if (s == self.journal_block)
-        {
-            printf("offset %08lx: zeroes\n", self.journal_pos);
-            self.journal_pos += self.journal_block;
-        }
-        else if (((journal_entry*)data)->magic == JOURNAL_MAGIC)
-        {
-            printf("offset %08lx:\n", self.journal_pos);
-            self.dump_block(data);
+            printf("offset %08lx: journal superblock is invalid\n", self.journal_pos);
         }
         else
         {
-            printf("offset %08lx: no magic in the beginning, looks like random data (pattern=%lx)\n", self.journal_pos, *((uint64_t*)data));
-            self.journal_pos += self.journal_block;
+            printf("offset %08lx:\n", self.journal_pos);
+            self.dump_block(data);
+            self.started = false;
+            self.journal_pos = je->start.journal_start;
+            while (1)
+            {
+                if (self.journal_pos >= self.journal_len)
+                    self.journal_pos = self.journal_block;
+                r = pread(self.fd, data, self.journal_block, self.journal_offset+self.journal_pos);
+                assert(r == self.journal_block);
+                printf("offset %08lx:\n", self.journal_pos);
+                r = self.dump_block(data);
+                if (r <= 0)
+                {
+                    printf("end of the journal\n");
+                    break;
+                }
+            }
         }
     }
     free(data);
@@ -88,7 +131,7 @@ int main(int argc, char *argv[])
     return 0;
 }
 
-void journal_dump_t::dump_block(void *buf)
+int journal_dump_t::dump_block(void *buf)
 {
     uint32_t pos = 0;
     journal_pos += journal_block;
@@ -97,12 +140,19 @@ void journal_dump_t::dump_block(void *buf)
     while (pos < journal_block)
     {
         journal_entry *je = (journal_entry*)(buf + pos);
-        if (je->magic != JOURNAL_MAGIC || je->type < JE_MIN || je->type > JE_MAX)
+        if (je->magic != JOURNAL_MAGIC || je->type < JE_MIN || je->type > JE_MAX ||
+            !all && started && je->crc32_prev != crc32_last)
         {
             break;
         }
-        const char *crc32_valid = je_crc32(je) == je->crc32 ? "(valid)" : "(invalid)";
-        printf("entry % 3d: crc32=%08x %s prev=%08x ", entry, je->crc32, crc32_valid, je->crc32_prev);
+        bool crc32_valid = je_crc32(je) == je->crc32;
+        if (!all && !crc32_valid)
+        {
+            break;
+        }
+        started = true;
+        crc32_last = je->crc32;
+        printf("entry % 3d: crc32=%08x %s prev=%08x ", entry, je->crc32, (crc32_valid ? "(valid)" : "(invalid)"), je->crc32_prev);
         if (je->type == JE_START)
         {
             printf("je_start start=%08lx\n", je->start.journal_start);
@@ -170,4 +220,5 @@ void journal_dump_t::dump_block(void *buf)
     {
         journal_pos = journal_len;
     }
+    return entry;
 }

epoll_manager.cpp

@@ -84,8 +84,12 @@ void epoll_manager_t::handle_epoll_events()
         nfds = epoll_wait(epoll_fd, events, MAX_EPOLL_EVENTS, 0);
         for (int i = 0; i < nfds; i++)
         {
-            auto & cb = epoll_handlers[events[i].data.fd];
-            cb(events[i].data.fd, events[i].events);
+            auto cb_it = epoll_handlers.find(events[i].data.fd);
+            if (cb_it != epoll_handlers.end())
+            {
+                auto & cb = cb_it->second;
+                cb(events[i].data.fd, events[i].events);
+            }
         }
     } while (nfds == MAX_EPOLL_EVENTS);
 }

etcd_state_client.cpp

@@ -173,6 +173,7 @@ void etcd_state_client_t::start_etcd_watcher()
             { "range_end", base64_encode(etcd_prefix+"/config0") },
             { "start_revision", etcd_watch_revision+1 },
             { "watch_id", ETCD_CONFIG_WATCH_ID },
+            { "progress_notify", true },
         } }
     }).dump());
     etcd_watch_ws->post_message(WS_TEXT, json11::Json(json11::Json::object {
@@ -181,6 +182,7 @@ void etcd_state_client_t::start_etcd_watcher()
             { "range_end", base64_encode(etcd_prefix+"/osd/state0") },
             { "start_revision", etcd_watch_revision+1 },
             { "watch_id", ETCD_OSD_STATE_WATCH_ID },
+            { "progress_notify", true },
         } }
     }).dump());
     etcd_watch_ws->post_message(WS_TEXT, json11::Json(json11::Json::object {
@@ -189,6 +191,7 @@ void etcd_state_client_t::start_etcd_watcher()
             { "range_end", base64_encode(etcd_prefix+"/pg/state0") },
             { "start_revision", etcd_watch_revision+1 },
             { "watch_id", ETCD_PG_STATE_WATCH_ID },
+            { "progress_notify", true },
         } }
     }).dump());
     etcd_watch_ws->post_message(WS_TEXT, json11::Json(json11::Json::object {
@@ -197,6 +200,7 @@ void etcd_state_client_t::start_etcd_watcher()
             { "range_end", base64_encode(etcd_prefix+"/pg/history0") },
             { "start_revision", etcd_watch_revision+1 },
             { "watch_id", ETCD_PG_HISTORY_WATCH_ID },
+            { "progress_notify", true },
         } }
     }).dump());
 }
@@ -315,67 +319,99 @@ void etcd_state_client_t::parse_state(const std::string & key, const json11::Jso
         }
         for (auto & pool_item: value.object_items())
         {
+            pool_config_t pc;
+            // ID
             pool_id_t pool_id = stoull_full(pool_item.first);
             if (!pool_id || pool_id >= POOL_ID_MAX)
             {
                 printf("Pool ID %s is invalid (must be a number less than 0x%x), skipping pool\n", pool_item.first.c_str(), POOL_ID_MAX);
                 continue;
             }
-            if (pool_item.second["pg_size"].uint64_value() < 1 ||
-                pool_item.second["scheme"] == "xor" && pool_item.second["pg_size"].uint64_value() < 3)
-            {
-                printf("Pool %u has invalid pg_size, skipping pool\n", pool_id);
-                continue;
-            }
-            if (pool_item.second["pg_minsize"].uint64_value() < 1 ||
-                pool_item.second["pg_minsize"].uint64_value() > pool_item.second["pg_size"].uint64_value() ||
-                pool_item.second["pg_minsize"].uint64_value() < (pool_item.second["pg_size"].uint64_value() - 1))
-            {
-                printf("Pool %u has invalid pg_minsize, skipping pool\n", pool_id);
-                continue;
-            }
-            if (pool_item.second["pg_count"].uint64_value() < 1)
-            {
-                printf("Pool %u has invalid pg_count, skipping pool\n", pool_id);
-                continue;
-            }
-            if (pool_item.second["name"].string_value() == "")
+            pc.id = pool_id;
+            // Pool Name
+            pc.name = pool_item.second["name"].string_value();
+            if (pc.name == "")
             {
                 printf("Pool %u has empty name, skipping pool\n", pool_id);
                 continue;
             }
-            if (pool_item.second["scheme"] != "replicated" && pool_item.second["scheme"] != "xor")
+            // Failure Domain
+            pc.failure_domain = pool_item.second["failure_domain"].string_value();
+            // Coding Scheme
+            if (pool_item.second["scheme"] == "replicated")
+                pc.scheme = POOL_SCHEME_REPLICATED;
+            else if (pool_item.second["scheme"] == "xor")
+                pc.scheme = POOL_SCHEME_XOR;
+            else if (pool_item.second["scheme"] == "jerasure")
+                pc.scheme = POOL_SCHEME_JERASURE;
+            else
             {
-                printf("Pool %u has invalid coding scheme (only \"xor\" and \"replicated\" are allowed), skipping pool\n", pool_id);
+                printf("Pool %u has invalid coding scheme (one of \"xor\", \"replicated\" or \"jerasure\" required), skipping pool\n", pool_id);
                 continue;
             }
-            if (pool_item.second["max_osd_combinations"].uint64_value() > 0 &&
-                pool_item.second["max_osd_combinations"].uint64_value() < 100)
+            // PG Size
+            pc.pg_size = pool_item.second["pg_size"].uint64_value();
+            if (pc.pg_size < 1 ||
+                pool_item.second["pg_size"].uint64_value() < 3 &&
+                (pc.scheme == POOL_SCHEME_XOR || pc.scheme == POOL_SCHEME_JERASURE) ||
+                pool_item.second["pg_size"].uint64_value() > 256)
+            {
+                printf("Pool %u has invalid pg_size, skipping pool\n", pool_id);
+                continue;
+            }
+            // Parity Chunks
+            pc.parity_chunks = pool_item.second["parity_chunks"].uint64_value();
+            if (pc.scheme == POOL_SCHEME_XOR)
+            {
+                if (pc.parity_chunks > 1)
+                {
+                    printf("Pool %u has invalid parity_chunks (must be 1), skipping pool\n", pool_id);
+                    continue;
+                }
+                pc.parity_chunks = 1;
+            }
+            if (pc.scheme == POOL_SCHEME_JERASURE &&
+                (pc.parity_chunks < 1 || pc.parity_chunks > pc.pg_size-2))
+            {
+                printf("Pool %u has invalid parity_chunks (must be between 1 and pg_size-2), skipping pool\n", pool_id);
+                continue;
+            }
+            // PG MinSize
+            pc.pg_minsize = pool_item.second["pg_minsize"].uint64_value();
+            if (pc.pg_minsize < 1 || pc.pg_minsize > pc.pg_size ||
+                (pc.scheme == POOL_SCHEME_XOR || pc.scheme == POOL_SCHEME_JERASURE) &&
+                pc.pg_minsize < (pc.pg_size-pc.parity_chunks))
+            {
+                printf("Pool %u has invalid pg_minsize, skipping pool\n", pool_id);
+                continue;
+            }
+            // PG Count
+            pc.pg_count = pool_item.second["pg_count"].uint64_value();
+            if (pc.pg_count < 1)
+            {
+                printf("Pool %u has invalid pg_count, skipping pool\n", pool_id);
+                continue;
+            }
+            // Max OSD Combinations
+            pc.max_osd_combinations = pool_item.second["max_osd_combinations"].uint64_value();
+            if (!pc.max_osd_combinations)
+                pc.max_osd_combinations = 10000;
+            if (pc.max_osd_combinations > 0 && pc.max_osd_combinations < 100)
             {
                 printf("Pool %u has invalid max_osd_combinations (must be at least 100), skipping pool\n", pool_id);
                 continue;
             }
+            // PG Stripe Size
+            pc.pg_stripe_size = pool_item.second["pg_stripe_size"].uint64_value();
+            uint64_t min_stripe_size = bs_block_size * (pc.scheme == POOL_SCHEME_REPLICATED ? 1 : (pc.pg_size-pc.parity_chunks));
+            if (pc.pg_stripe_size < min_stripe_size)
+                pc.pg_stripe_size = min_stripe_size;
+            // Save
+            pc.real_pg_count = this->pool_config[pool_id].real_pg_count;
+            std::swap(pc.pg_config, this->pool_config[pool_id].pg_config);
+            std::swap(this->pool_config[pool_id], pc);
             auto & parsed_cfg = this->pool_config[pool_id];
             parsed_cfg.exists = true;
-            parsed_cfg.id = pool_id;
-            parsed_cfg.name = pool_item.second["name"].string_value();
-            parsed_cfg.scheme = pool_item.second["scheme"] == "replicated" ? POOL_SCHEME_REPLICATED : POOL_SCHEME_XOR;
-            parsed_cfg.pg_size = pool_item.second["pg_size"].uint64_value();
-            parsed_cfg.pg_minsize = pool_item.second["pg_minsize"].uint64_value();
-            parsed_cfg.pg_count = pool_item.second["pg_count"].uint64_value();
-            parsed_cfg.failure_domain = pool_item.second["failure_domain"].string_value();
-            parsed_cfg.pg_stripe_size = pool_item.second["pg_stripe_size"].uint64_value();
-            uint64_t min_stripe_size = bs_block_size *
-                (parsed_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : parsed_cfg.pg_minsize);
-            if (parsed_cfg.pg_stripe_size < min_stripe_size)
-            {
-                parsed_cfg.pg_stripe_size = min_stripe_size;
-            }
-            parsed_cfg.max_osd_combinations = pool_item.second["max_osd_combinations"].uint64_value();
-            if (!parsed_cfg.max_osd_combinations)
-            {
-                parsed_cfg.max_osd_combinations = 10000;
-            }
             for (auto & pg_item: parsed_cfg.pg_config)
             {
                 if (pg_item.second.target_set.size() != parsed_cfg.pg_size)

etcd_state_client.h

@@ -43,7 +43,7 @@ struct pool_config_t
     pool_id_t id;
     std::string name;
     uint64_t scheme;
-    uint64_t pg_size, pg_minsize;
+    uint64_t pg_size, pg_minsize, parity_chunks;
     uint64_t pg_count;
     uint64_t real_pg_count;
     std::string failure_domain;

fio_cluster.cpp

@@ -93,7 +93,7 @@ static struct fio_option options[] = {
     {
         .name   = "cluster_log_level",
         .lname  = "cluster log level",
-        .type   = FIO_OPT_BOOL,
+        .type   = FIO_OPT_INT,
         .off1   = offsetof(struct sec_options, cluster_log),
         .help   = "Set log level for the Vitastor client",
         .def    = "0",
@@ -145,9 +145,7 @@ static void sec_cleanup(struct thread_data *td)
         delete bsd->cli;
         delete bsd->epmgr;
         delete bsd->ringloop;
-        bsd->cli = NULL;
-        bsd->epmgr = NULL;
-        bsd->ringloop = NULL;
+        delete bsd;
     }
 }
 

fio_sec_osd.cpp

@@ -140,6 +140,7 @@ static void sec_cleanup(struct thread_data *td)
     if (bsd)
     {
         close(bsd->connect_fd);
+        delete bsd;
     }
 }
 
@@ -312,6 +313,7 @@ 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)

messenger.cpp

@@ -30,7 +30,7 @@ osd_messenger_t::~osd_messenger_t()
 {
     while (clients.size() > 0)
     {
-        stop_client(clients.begin()->first);
+        stop_client(clients.begin()->first, true);
     }
 }
 
@@ -111,12 +111,12 @@ void osd_messenger_t::try_connect_peer_addr(osd_num_t peer_osd, const char *peer
         timeout_id = tfd->set_timer(1000*peer_connect_timeout, false, [this, peer_fd](int timer_id)
         {
             osd_num_t peer_osd = clients.at(peer_fd)->osd_num;
-            stop_client(peer_fd);
+            stop_client(peer_fd, true);
             on_connect_peer(peer_osd, -EIO);
             return;
         });
     }
-    clients[peer_fd] = new osd_client_t({
+    clients[peer_fd] = new osd_client_t((osd_client_t){
         .peer_addr = addr,
         .peer_port = peer_port,
         .peer_fd = peer_fd,
@@ -149,7 +149,7 @@ void osd_messenger_t::handle_connect_epoll(int peer_fd)
     }
     if (result != 0)
     {
-        stop_client(peer_fd);
+        stop_client(peer_fd, true);
         on_connect_peer(peer_osd, -result);
         return;
     }
@@ -171,7 +171,7 @@ void osd_messenger_t::handle_peer_epoll(int peer_fd, int epoll_events)
     {
         // Stop client
         printf("[OSD %lu] client %d disconnected\n", this->osd_num, peer_fd);
-        stop_client(peer_fd);
+        stop_client(peer_fd, true);
     }
     else if (epoll_events & EPOLLIN)
     {
@@ -233,7 +233,7 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
     osd_op_t *op = new osd_op_t();
     op->op_type = OSD_OP_OUT;
     op->peer_fd = cl->peer_fd;
-    op->req = {
+    op->req = (osd_any_op_t){
         .show_conf = {
             .header = {
                 .magic = SECONDARY_OSD_OP_MAGIC,
@@ -309,7 +309,7 @@ void osd_messenger_t::cancel_op(osd_op_t *op)
     }
 }
 
-void osd_messenger_t::stop_client(int peer_fd)
+void osd_messenger_t::stop_client(int peer_fd, bool force)
 {
     assert(peer_fd != 0);
     auto it = clients.find(peer_fd);
@@ -334,6 +334,10 @@ void osd_messenger_t::stop_client(int peer_fd)
                 printf("[OSD %lu] Stopping client %d (regular client)\n", osd_num, peer_fd);
         }
     }
+    else if (!force)
+    {
+        return;
+    }
     cl->peer_state = PEER_STOPPED;
     clients.erase(it);
     tfd->set_fd_handler(peer_fd, false, NULL);
@@ -348,7 +352,14 @@ void osd_messenger_t::stop_client(int peer_fd)
     }
     if (cl->read_op)
     {
-        delete cl->read_op;
+        if (cl->read_op->callback)
+        {
+            cancel_op(cl->read_op);
+        }
+        else
+        {
+            delete cl->read_op;
+        }
         cl->read_op = NULL;
     }
     for (auto rit = read_ready_clients.begin(); rit != read_ready_clients.end(); rit++)
@@ -402,7 +413,7 @@ void osd_messenger_t::accept_connections(int listen_fd)
         fcntl(peer_fd, F_SETFL, fcntl(peer_fd, F_GETFL, 0) | O_NONBLOCK);
         int one = 1;
         setsockopt(peer_fd, SOL_TCP, TCP_NODELAY, &one, sizeof(one));
-        clients[peer_fd] = new osd_client_t({
+        clients[peer_fd] = new osd_client_t((osd_client_t){
             .peer_addr = addr,
             .peer_port = ntohs(addr.sin_port),
             .peer_fd = peer_fd,

messenger.h

@@ -275,7 +275,7 @@ struct osd_messenger_t
 
 public:
     void connect_peer(uint64_t osd_num, json11::Json peer_state);
-    void stop_client(int peer_fd);
+    void stop_client(int peer_fd, bool force = false);
     void outbox_push(osd_op_t *cur_op);
     std::function<void(osd_op_t*)> exec_op;
     std::function<void(osd_num_t)> repeer_pgs;

mon/mon.js

@@ -9,212 +9,215 @@ const LPOptimizer = require('./lp-optimizer.js');
 const stableStringify = require('./stable-stringify.js');
 const PGUtil = require('./PGUtil.js');
 
+// FIXME document all etcd keys and config variables in the form of JSON schema or similar
+const etcd_allow = new RegExp('^'+[
+    'config/global',
+    'config/node_placement',
+    'config/pools',
+    'config/osd/[1-9]\\d*',
+    'config/pgs',
+    'osd/state/[1-9]\\d*',
+    'osd/stats/[1-9]\\d*',
+    'mon/master',
+    'pg/state/[1-9]\\d*/[1-9]\\d*',
+    'pg/stats/[1-9]\\d*/[1-9]\\d*',
+    'pg/history/[1-9]\\d*/[1-9]\\d*',
+    'stats',
+].join('$|^')+'$');
+
+const etcd_tree = {
+    config: {
+        /* global: {
+            // mon
+            etcd_mon_ttl: 30, // min: 10
+            etcd_mon_timeout: 1000, // ms. min: 0
+            etcd_mon_retries: 5, // min: 0
+            mon_change_timeout: 1000, // ms. min: 100
+            mon_stats_timeout: 1000, // ms. min: 100
+            osd_out_time: 1800, // seconds. min: 0
+            placement_levels: { datacenter: 1, rack: 2, host: 3, osd: 4, ... },
+            // client and osd
+            use_sync_send_recv: false,
+            log_level: 0,
+            block_size: 131072,
+            disk_alignment: 4096,
+            bitmap_granularity: 4096,
+            immediate_commit: false, // 'all' or 'small'
+            client_dirty_limit: 33554432,
+            peer_connect_interval: 5, // seconds. min: 1
+            peer_connect_timeout: 5, // seconds. min: 1
+            up_wait_retry_interval: 500, // ms. min: 50
+            // osd
+            etcd_report_interval: 30, // min: 10
+            run_primary: true,
+            bind_address: "0.0.0.0",
+            bind_port: 0,
+            autosync_interval: 5,
+            client_queue_depth: 128, // unused
+            recovery_queue_depth: 4,
+            readonly: false,
+            print_stats_interval: 3,
+            // blockstore - fixed in superblock
+            block_size,
+            disk_alignment,
+            journal_block_size,
+            meta_block_size,
+            bitmap_granularity,
+            journal_device,
+            journal_offset,
+            journal_size,
+            disable_journal_fsync,
+            data_device,
+            data_offset,
+            data_size,
+            disable_data_fsync,
+            meta_device,
+            meta_offset,
+            disable_meta_fsync,
+            disable_device_lock,
+            // blockstore - configurable
+            flusher_count,
+            inmemory_metadata,
+            inmemory_journal,
+            journal_sector_buffer_count,
+            journal_no_same_sector_overwrites,
+        }, */
+        global: {},
+        /* node_placement: {
+            host1: { level: 'host', parent: 'rack1' },
+            ...
+        }, */
+        node_placement: {},
+        /* pools: {
+            <id>: {
+                name: 'testpool',
+                // jerasure uses Reed-Solomon-Vandermonde codes
+                scheme: 'replicated' | 'xor' | 'jerasure',
+                pg_size: 3,
+                pg_minsize: 2,
+                // number of parity chunks, required for jerasure
+                parity_chunks?: 1,
+                pg_count: 100,
+                failure_domain: 'host',
+                max_osd_combinations: 10000,
+                pg_stripe_size: 4194304,
+                root_node?: 'rack1',
+                // restrict pool to OSDs having all of these tags
+                osd_tags?: 'nvme' | [ 'nvme', ... ],
+            },
+            ...
+        }, */
+        pools: {},
+        osd: {
+            /* <id>: { reweight?: 1, tags?: [ 'nvme', ... ] }, ... */
+        },
+        /* pgs: {
+            hash: string,
+            items: {
+                <pool_id>: {
+                    <pg_id>: {
+                        osd_set: [ 1, 2, 3 ],
+                        primary: 1,
+                        pause: false,
+                    }
+                }
+            }
+        }, */
+        pgs: {},
+    },
+    osd: {
+        state: {
+            /* <osd_num_t>: {
+                state: "up",
+                addresses: string[],
+                host: string,
+                port: uint16_t,
+                primary_enabled: boolean,
+                blockstore_enabled: boolean,
+            }, */
+        },
+        stats: {
+            /* <osd_num_t>: {
+                time: number, // unix time
+                blockstore_ready: boolean,
+                size: uint64_t, // bytes
+                free: uint64_t, // bytes
+                host: string,
+                op_stats: {
+                    <string>: { count: uint64_t, usec: uint64_t, bytes: uint64_t },
+                },
+                subop_stats: {
+                    <string>: { count: uint64_t, usec: uint64_t },
+                },
+                recovery_stats: {
+                    degraded: { count: uint64_t, bytes: uint64_t },
+                    misplaced: { count: uint64_t, bytes: uint64_t },
+                },
+            }, */
+        },
+    },
+    mon: {
+        master: {
+            /* ip: [ string ], */
+        },
+    },
+    pg: {
+        state: {
+            /* <pool_id>: {
+                <pg_id>: {
+                    primary: osd_num_t,
+                    state: ("starting"|"peering"|"incomplete"|"active"|"stopping"|"offline"|
+                        "degraded"|"has_incomplete"|"has_degraded"|"has_misplaced"|"has_unclean"|
+                        "has_invalid"|"left_on_dead")[],
+                }
+            }, */
+        },
+        stats: {
+            /* <pool_id>: {
+                <pg_id>: {
+                    object_count: uint64_t,
+                    clean_count: uint64_t,
+                    misplaced_count: uint64_t,
+                    degraded_count: uint64_t,
+                    incomplete_count: uint64_t,
+                    write_osd_set: osd_num_t[],
+                },
+            }, */
+        },
+        history: {
+            /* <pool_id>: {
+                <pg_id>: {
+                    osd_sets: osd_num_t[][],
+                    all_peers: osd_num_t[],
+                    epoch: uint32_t,
+                },
+            }, */
+        },
+    },
+    stats: {
+        /* op_stats: {
+            <string>: { count: uint64_t, usec: uint64_t, bytes: uint64_t },
+        },
+        subop_stats: {
+            <string>: { count: uint64_t, usec: uint64_t },
+        },
+        recovery_stats: {
+            degraded: { count: uint64_t, bytes: uint64_t },
+            misplaced: { count: uint64_t, bytes: uint64_t },
+        },
+        object_counts: {
+            object: uint64_t,
+            clean: uint64_t,
+            misplaced: uint64_t,
+            degraded: uint64_t,
+            incomplete: uint64_t,
+        }, */
+    },
+};
+
 // FIXME Split into several files
 class Mon
 {
-    // FIXME document all etcd keys and config variables in the form of JSON schema or similar
-    static etcd_allow = new RegExp('^'+[
-        'config/global',
-        'config/node_placement',
-        'config/pools',
-        'config/osd/[1-9]\\d*',
-        'config/pgs',
-        'osd/state/[1-9]\\d*',
-        'osd/stats/[1-9]\\d*',
-        'mon/master',
-        'pg/state/[1-9]\\d*/[1-9]\\d*',
-        'pg/stats/[1-9]\\d*/[1-9]\\d*',
-        'pg/history/[1-9]\\d*/[1-9]\\d*',
-        'stats',
-    ].join('$|^')+'$')
-
-    static etcd_tree = {
-        config: {
-            /* global: {
-                // mon
-                etcd_mon_ttl: 30, // min: 10
-                etcd_mon_timeout: 1000, // ms. min: 0
-                etcd_mon_retries: 5, // min: 0
-                mon_change_timeout: 1000, // ms. min: 100
-                mon_stats_timeout: 1000, // ms. min: 100
-                osd_out_time: 1800, // seconds. min: 0
-                placement_levels: { datacenter: 1, rack: 2, host: 3, osd: 4, ... },
-                // client and osd
-                use_sync_send_recv: false,
-                log_level: 0,
-                block_size: 131072,
-                disk_alignment: 4096,
-                bitmap_granularity: 4096,
-                immediate_commit: false, // 'all' or 'small'
-                client_dirty_limit: 33554432,
-                peer_connect_interval: 5, // seconds. min: 1
-                peer_connect_timeout: 5, // seconds. min: 1
-                up_wait_retry_interval: 500, // ms. min: 50
-                // osd
-                etcd_report_interval: 30, // min: 10
-                run_primary: true,
-                bind_address: "0.0.0.0",
-                bind_port: 0,
-                autosync_interval: 5,
-                client_queue_depth: 128, // unused
-                recovery_queue_depth: 4,
-                readonly: false,
-                print_stats_interval: 3,
-                // blockstore - fixed in superblock
-                block_size,
-                disk_alignment,
-                journal_block_size,
-                meta_block_size,
-                bitmap_granularity,
-                journal_device,
-                journal_offset,
-                journal_size,
-                disable_journal_fsync,
-                data_device,
-                data_offset,
-                data_size,
-                disable_data_fsync,
-                meta_device,
-                meta_offset,
-                disable_meta_fsync,
-                disable_device_lock,
-                // blockstore - configurable
-                flusher_count,
-                inmemory_metadata,
-                inmemory_journal,
-                journal_sector_buffer_count,
-                journal_no_same_sector_overwrites,
-            }, */
-            global: {},
-            /* node_placement: {
-                host1: { level: 'host', parent: 'rack1' },
-                ...
-            }, */
-            node_placement: {},
-            /* pools: {
-                <id>: {
-                    name: 'testpool',
-                    scheme: 'xor',
-                    pg_size: 3,
-                    pg_minsize: 2,
-                    pg_count: 100,
-                    failure_domain: 'host',
-                    max_osd_combinations: 10000,
-                    pg_stripe_size: 4194304,
-                    root_node?: 'rack1',
-                    // restrict pool to OSDs having all of these tags
-                    osd_tags?: 'nvme' | [ 'nvme', ... ],
-                },
-                ...
-            }, */
-            pools: {},
-            osd: {
-                /* <id>: { reweight?: 1, tags?: [ 'nvme', ... ] }, ... */
-            },
-            /* pgs: {
-                hash: string,
-                items: {
-                    <pool_id>: {
-                        <pg_id>: {
-                            osd_set: [ 1, 2, 3 ],
-                            primary: 1,
-                            pause: false,
-                        }
-                    }
-                }
-            }, */
-            pgs: {},
-        },
-        osd: {
-            state: {
-                /* <osd_num_t>: {
-                    state: "up",
-                    addresses: string[],
-                    host: string,
-                    port: uint16_t,
-                    primary_enabled: boolean,
-                    blockstore_enabled: boolean,
-                }, */
-            },
-            stats: {
-                /* <osd_num_t>: {
-                    time: number, // unix time
-                    blockstore_ready: boolean,
-                    size: uint64_t, // bytes
-                    free: uint64_t, // bytes
-                    host: string,
-                    op_stats: {
-                        <string>: { count: uint64_t, usec: uint64_t, bytes: uint64_t },
-                    },
-                    subop_stats: {
-                        <string>: { count: uint64_t, usec: uint64_t },
-                    },
-                    recovery_stats: {
-                        degraded: { count: uint64_t, bytes: uint64_t },
-                        misplaced: { count: uint64_t, bytes: uint64_t },
-                    },
-                }, */
-            },
-        },
-        mon: {
-            master: {
-                /* ip: [ string ], */
-            },
-        },
-        pg: {
-            state: {
-                /* <pool_id>: {
-                    <pg_id>: {
-                        primary: osd_num_t,
-                        state: ("starting"|"peering"|"incomplete"|"active"|"stopping"|"offline"|
-                            "degraded"|"has_incomplete"|"has_degraded"|"has_misplaced"|"has_unclean"|
-                            "has_invalid"|"left_on_dead")[],
-                    }
-                }, */
-            },
-            stats: {
-                /* <pool_id>: {
-                    <pg_id>: {
-                        object_count: uint64_t,
-                        clean_count: uint64_t,
-                        misplaced_count: uint64_t,
-                        degraded_count: uint64_t,
-                        incomplete_count: uint64_t,
-                        write_osd_set: osd_num_t[],
-                    },
-                }, */
-            },
-            history: {
-                /* <pool_id>: {
-                    <pg_id>: {
-                        osd_sets: osd_num_t[][],
-                        all_peers: osd_num_t[],
-                        epoch: uint32_t,
-                    },
-                }, */
-            },
-        },
-        stats: {
-            /* op_stats: {
-                <string>: { count: uint64_t, usec: uint64_t, bytes: uint64_t },
-            },
-            subop_stats: {
-                <string>: { count: uint64_t, usec: uint64_t },
-            },
-            recovery_stats: {
-                degraded: { count: uint64_t, bytes: uint64_t },
-                misplaced: { count: uint64_t, bytes: uint64_t },
-            },
-            object_counts: {
-                object: uint64_t,
-                clean: uint64_t,
-                misplaced: uint64_t,
-                degraded: uint64_t,
-                incomplete: uint64_t,
-            }, */
-        },
-    }
-
     constructor(config)
     {
         // FIXME: Maybe prefer local etcd
@@ -250,7 +253,10 @@ class Mon
         const res = await this.etcd_call('/kv/txn', { success: [
             { requestRange: { key: b64(this.etcd_prefix+'/config/global') } }
         ] }, this.etcd_start_timeout, -1);
-        this.parse_kv(res.responses[0].response_range.kvs[0]);
+        if (res.responses[0].response_range.kvs)
+        {
+            this.parse_kv(res.responses[0].response_range.kvs[0]);
+        }
         this.check_config();
     }
 
@@ -331,6 +337,7 @@ class Mon
                 range_end: b64(this.etcd_prefix+'0'),
                 start_revision: ''+this.etcd_watch_revision,
                 watch_id: 1,
+                progress_notify: true,
             },
         }));
         this.ws.on('message', (msg) =>
@@ -561,19 +568,15 @@ class Mon
                     { requestPut: { key: b64(this.etcd_prefix+'/config/pgs'), value: b64(JSON.stringify(new_cfg)) } },
                 ],
             }, this.config.etcd_mon_timeout, 0);
-            if (!res.succeeded)
-            {
-                return false;
-            }
-            this.state.config.pgs = new_cfg;
+            return false;
         }
         return !has_online;
     }
 
     save_new_pgs_txn(request, pool_id, up_osds, prev_pgs, new_pgs, pg_history)
     {
-        const replicated = this.state.config.pools[pool_id].scheme === 'replicated';
-        const pg_minsize = this.state.config.pools[pool_id].pg_minsize;
+        const replicated = new_pgs.length && this.state.config.pools[pool_id].scheme === 'replicated';
+        const pg_minsize = new_pgs.length && this.state.config.pools[pool_id].pg_minsize;
         const pg_items = {};
         new_pgs.map((osd_set, i) =>
         {
@@ -628,13 +631,21 @@ class Mon
             }
         }
         this.state.config.pgs.items = this.state.config.pgs.items || {};
-        this.state.config.pgs.items[pool_id] = pg_items;
+        if (!new_pgs.length)
+        {
+            delete this.state.config.pgs.items[pool_id];
+        }
+        else
+        {
+            this.state.config.pgs.items[pool_id] = pg_items;
+        }
     }
 
     validate_pool_cfg(pool_id, pool_cfg, warn)
     {
         pool_cfg.pg_size = Math.floor(pool_cfg.pg_size);
         pool_cfg.pg_minsize = Math.floor(pool_cfg.pg_minsize);
+        pool_cfg.parity_chunks = Math.floor(pool_cfg.parity_chunks) || undefined;
         pool_cfg.pg_count = Math.floor(pool_cfg.pg_count);
         pool_cfg.failure_domain = pool_cfg.failure_domain || 'host';
         pool_cfg.max_osd_combinations = Math.floor(pool_cfg.max_osd_combinations) || 10000;
@@ -644,8 +655,14 @@ class Mon
                 console.log('Pool ID '+pool_id+' is invalid');
             return false;
         }
-        if (!pool_cfg.pg_size || pool_cfg.pg_size < 1 ||
-            pool_cfg.scheme === 'xor' && pool_cfg.pg_size < 3)
+        if (pool_cfg.scheme !== 'xor' && pool_cfg.scheme !== 'replicated' && pool_cfg.scheme !== 'jerasure')
+        {
+            if (warn)
+                console.log('Pool '+pool_id+' has invalid coding scheme (one of "xor", "replicated" and "jerasure" required)');
+            return false;
+        }
+        if (!pool_cfg.pg_size || pool_cfg.pg_size < 1 || pool_cfg.pg_size > 256 ||
+            (pool_cfg.scheme === 'xor' || pool_cfg.scheme == 'jerasure') && pool_cfg.pg_size < 3)
         {
             if (warn)
                 console.log('Pool '+pool_id+' has invalid pg_size');
@@ -658,6 +675,18 @@ class Mon
                 console.log('Pool '+pool_id+' has invalid pg_minsize');
             return false;
         }
+        if (pool_cfg.scheme === 'xor' && pool_cfg.parity_chunks != 0 && pool_cfg.parity_chunks != 1)
+        {
+            if (warn)
+                console.log('Pool '+pool_id+' has invalid parity_chunks (must be 1)');
+            return false;
+        }
+        if (pool_cfg.scheme === 'jerasure' && (pool_cfg.parity_chunks < 1 || pool_cfg.parity_chunks > pool_cfg.pg_size-2))
+        {
+            if (warn)
+                console.log('Pool '+pool_id+' has invalid parity_chunks (must be between 1 and pg_size-2)');
+            return false;
+        }
         if (!pool_cfg.pg_count || pool_cfg.pg_count < 1)
         {
             if (warn)
@@ -670,12 +699,6 @@ class Mon
                 console.log('Pool '+pool_id+' has empty name');
             return false;
         }
-        if (pool_cfg.scheme !== 'xor' && pool_cfg.scheme !== 'replicated')
-        {
-            if (warn)
-                console.log('Pool '+pool_id+' has invalid coding scheme (only "xor" and "replicated" are allowed)');
-            return false;
-        }
         if (pool_cfg.max_osd_combinations < 100)
         {
             if (warn)
@@ -739,6 +762,24 @@ class Mon
         {
             // Something has changed
             const etcd_request = { compare: [], success: [] };
+            for (const pool_id in (this.state.config.pgs||{}).items||{})
+            {
+                if (!this.state.config.pools[pool_id])
+                {
+                    // Pool deleted. Delete all PGs, but first stop them.
+                    if (!await this.stop_all_pgs(pool_id))
+                    {
+                        this.schedule_recheck();
+                        return;
+                    }
+                    const prev_pgs = [];
+                    for (const pg in this.state.config.pgs.items[pool_id]||{})
+                    {
+                        prev_pgs[pg-1] = this.state.config.pgs.items[pool_id][pg].osd_set;
+                    }
+                    this.save_new_pgs_txn(etcd_request, pool_id, up_osds, prev_pgs, [], []);
+                }
+            }
             for (const pool_id in this.state.config.pools)
             {
                 const pool_cfg = this.state.config.pools[pool_id];
@@ -1228,4 +1269,7 @@ function sha1hex(str)
     return hash.digest('hex');
 }
 
+Mon.etcd_allow = etcd_allow;
+Mon.etcd_tree = etcd_tree;
+
 module.exports = Mon;

mon/simple-offsets.js

@@ -4,6 +4,7 @@
 // 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()
@@ -15,6 +16,7 @@ async function run()
         device_block_size: 4096,
         journal_offset: 0,
         device_size: 0,
+        format: 'text',
     };
     for (let i = 2; i < process.argv.length; i++)
     {
@@ -24,7 +26,22 @@ async function run()
             i++;
         }
     }
-    const device_size = Number(options.device_size || await system("blockdev --getsize64 "+options.device));
+    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');
@@ -32,25 +49,45 @@ async function run()
     }
     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 + options.object_size/options.bitmap_granularity/8));
+    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");
-    process.stdout.write(
-        `Metadata size: ${meta_size_fmt}\n`+
-        `Options for the OSD:\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` : '')
-    );
+    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) : ok(stdout))));
+    return new Promise((ok, no) => child_process.exec(cmd, { maxBuffer: 64*1024*1024 }, (err, stdout, stderr) => (err ? no(err.message) : ok(stdout))));
 }
 
-run().catch(console.error);
+run().catch(err => { console.error(err); process.exit(1); });

msgr_receive.cpp

@@ -9,6 +9,10 @@ void osd_messenger_t::read_requests()
     {
         int peer_fd = read_ready_clients[i];
         osd_client_t *cl = clients[peer_fd];
+        if (cl->read_msg.msg_iovlen)
+        {
+            continue;
+        }
         if (cl->read_remaining < receive_buffer_size)
         {
             cl->read_iov.iov_base = cl->in_buf;
@@ -29,6 +33,7 @@ void osd_messenger_t::read_requests()
             io_uring_sqe* sqe = ringloop->get_sqe();
             if (!sqe)
             {
+                cl->read_msg.msg_iovlen = 0;
                 read_ready_clients.erase(read_ready_clients.begin(), read_ready_clients.begin() + i);
                 return;
             }
@@ -52,6 +57,7 @@ void osd_messenger_t::read_requests()
 bool osd_messenger_t::handle_read(int result, osd_client_t *cl)
 {
     bool ret = false;
+    cl->read_msg.msg_iovlen = 0;
     cl->refs--;
     if (cl->peer_state == PEER_STOPPED)
     {
@@ -160,8 +166,14 @@ bool osd_messenger_t::handle_finished_read(osd_client_t *cl)
     {
         if (cl->read_op->req.hdr.magic == SECONDARY_OSD_REPLY_MAGIC)
             return handle_reply_hdr(cl);
-        else
+        else if (cl->read_op->req.hdr.magic == SECONDARY_OSD_OP_MAGIC)
             handle_op_hdr(cl);
+        else
+        {
+            printf("Received garbage: magic=%lx id=%lu opcode=%lx from %d\n", cl->read_op->req.hdr.magic, cl->read_op->req.hdr.id, cl->read_op->req.hdr.opcode, cl->peer_fd);
+            stop_client(cl->peer_fd);
+            return false;
+        }
     }
     else if (cl->read_state == CL_READ_DATA)
     {

msgr_send.cpp

@@ -46,7 +46,8 @@ void osd_messenger_t::outbox_push(osd_op_t *cur_op)
         to_send_list.push_back((iovec){ .iov_base = cur_op->req.buf, .iov_len = OSD_PACKET_SIZE });
         cl->sent_ops[cur_op->req.hdr.id] = cur_op;
     }
-    // Pre-defined send_lists
+    to_outbox.push_back(NULL);
+    // Operation data
     if ((cur_op->op_type == OSD_OP_IN
         ? (cur_op->req.hdr.opcode == OSD_OP_READ ||
         cur_op->req.hdr.opcode == OSD_OP_SEC_READ ||
@@ -58,17 +59,17 @@ void osd_messenger_t::outbox_push(osd_op_t *cur_op)
         cur_op->req.hdr.opcode == OSD_OP_SEC_STABILIZE ||
         cur_op->req.hdr.opcode == OSD_OP_SEC_ROLLBACK)) && cur_op->iov.count > 0)
     {
-        to_outbox.push_back(NULL);
         for (int i = 0; i < cur_op->iov.count; i++)
         {
             assert(cur_op->iov.buf[i].iov_base);
             to_send_list.push_back(cur_op->iov.buf[i]);
-            to_outbox.push_back(i == cur_op->iov.count-1 ? cur_op : NULL);
+            to_outbox.push_back(NULL);
         }
     }
-    else
+    if (cur_op->op_type == OSD_OP_IN)
     {
-        to_outbox.push_back(cur_op);
+        // To free it later
+        to_outbox[to_outbox.size()-1] = cur_op;
     }
     if (!ringloop)
     {
@@ -92,6 +93,10 @@ void osd_messenger_t::outbox_push(osd_op_t *cur_op)
 void osd_messenger_t::measure_exec(osd_op_t *cur_op)
 {
     // Measure execution latency
+    if (cur_op->req.hdr.opcode > OSD_OP_MAX)
+    {
+        return;
+    }
     timespec tv_end;
     clock_gettime(CLOCK_REALTIME, &tv_end);
     stats.op_stat_count[cur_op->req.hdr.opcode]++;
@@ -198,11 +203,8 @@ void osd_messenger_t::handle_send(int result, osd_client_t *cl)
             {
                 if (cl->outbox[done])
                 {
-                    // Operation fully sent
-                    if (cl->outbox[done]->op_type == OSD_OP_IN)
-                    {
-                        delete cl->outbox[done];
-                    }
+                    // Reply fully sent
+                    delete cl->outbox[done];
                 }
                 result -= iov.iov_len;
                 done++;

nbd_proxy.cpp

@@ -17,6 +17,10 @@
 #include "epoll_manager.h"
 #include "cluster_client.h"
 
+#ifndef MSG_ZEROCOPY
+#define MSG_ZEROCOPY 0
+#endif
+
 const char *exe_name = NULL;
 
 class nbd_proxy

osd_cluster.cpp

@@ -4,6 +4,7 @@
 #include "osd.h"
 #include "base64.h"
 #include "etcd_state_client.h"
+#include "osd_rmw.h"
 
 // Startup sequence:
 //   Start etcd watcher -> Load global OSD configuration -> Bind socket -> Acquire lease -> Report&lock OSD state
@@ -32,12 +33,26 @@ void osd_t::init_cluster()
             }
             pgs[{ 1, 1 }] = (pg_t){
                 .state = PG_PEERING,
+                .scheme = POOL_SCHEME_XOR,
                 .pg_cursize = 0,
+                .pg_size = 3,
+                .pg_minsize = 2,
+                .parity_chunks = 1,
                 .pool_id = 1,
                 .pg_num = 1,
                 .target_set = { 1, 2, 3 },
                 .cur_set = { 0, 0, 0 },
             };
+            st_cli.pool_config[1] = (pool_config_t){
+                .exists = true,
+                .id = 1,
+                .name = "testpool",
+                .scheme = POOL_SCHEME_XOR,
+                .pg_size = 3,
+                .pg_minsize = 2,
+                .pg_count = 1,
+                .real_pg_count = 1,
+            };
             report_pg_state(pgs[{ 1, 1 }]);
             pg_counts[1] = 1;
         }
@@ -369,6 +384,7 @@ void osd_t::create_osd_state()
         {
             st_cli.load_pgs();
         }
+        report_statistics();
     });
 }
 
@@ -479,7 +495,11 @@ void osd_t::apply_pg_count()
             }
             if (still_active > 0)
             {
-                printf("[OSD %lu] PG count change detected, but %d PG(s) are still active. This is not allowed. Exiting\n", this->osd_num, still_active);
+                printf(
+                    "[OSD %lu] PG count change detected for pool %u (new is %lu, old is %u),"
+                    " but %u PG(s) are still active. This is not allowed. Exiting\n",
+                    this->osd_num, pool_item.first, pool_item.second.real_pg_count, pg_counts[pool_item.first], still_active
+                );
                 force_stop(1);
                 return;
             }
@@ -583,6 +603,7 @@ void osd_t::apply_pg_config()
                     .pg_cursize = 0,
                     .pg_size = pool_item.second.pg_size,
                     .pg_minsize = pool_item.second.pg_minsize,
+                    .parity_chunks = pool_item.second.parity_chunks,
                     .pool_id = pool_id,
                     .pg_num = pg_num,
                     .reported_epoch = pg_cfg.epoch,
@@ -590,6 +611,10 @@ void osd_t::apply_pg_config()
                     .all_peers = std::vector<osd_num_t>(all_peers.begin(), all_peers.end()),
                     .target_set = pg_cfg.target_set,
                 };
+                if (pg.scheme == POOL_SCHEME_JERASURE)
+                {
+                    use_jerasure(pg.pg_size, pg.pg_size-pg.parity_chunks, true);
+                }
                 this->pg_state_dirty.insert({ .pool_id = pool_id, .pg_num = pg_num });
                 pg.print_state();
                 if (pg_cfg.cur_primary == this->osd_num)
@@ -778,6 +803,10 @@ void osd_t::report_pg_states()
                     {
                         // Remove offline PGs after reporting their state
                         this->pgs.erase(pg_it);
+                        if (pg_it->second.scheme == POOL_SCHEME_JERASURE)
+                        {
+                            use_jerasure(pg_it->second.pg_size, pg_it->second.pg_size-pg_it->second.parity_chunks, false);
+                        }
                     }
                 }
             }

osd_flush.cpp

@@ -166,7 +166,7 @@ void osd_t::submit_flush_op(pool_id_t pool_id, pg_num_t pg_num, pg_flush_batch_t
     {
         // local
         clock_gettime(CLOCK_REALTIME, &op->tv_begin);
-        op->bs_op = new blockstore_op_t({
+        op->bs_op = new blockstore_op_t((blockstore_op_t){
             .opcode = (uint64_t)(rollback ? BS_OP_ROLLBACK : BS_OP_STABLE),
             .callback = [this, op, pool_id, pg_num, fb](blockstore_op_t *bs_op)
             {
@@ -188,7 +188,7 @@ void osd_t::submit_flush_op(pool_id_t pool_id, pg_num_t pg_num, pg_flush_batch_t
         op->op_type = OSD_OP_OUT;
         op->iov.push_back(op->buf, count * sizeof(obj_ver_id));
         op->peer_fd = peer_fd;
-        op->req = {
+        op->req = (osd_any_op_t){
             .sec_stab = {
                 .header = {
                     .magic = SECONDARY_OSD_OP_MAGIC,
@@ -246,7 +246,7 @@ void osd_t::submit_recovery_op(osd_recovery_op_t *op)
 {
     op->osd_op = new osd_op_t();
     op->osd_op->op_type = OSD_OP_OUT;
-    op->osd_op->req = {
+    op->osd_op->req = (osd_any_op_t){
         .rw = {
             .header = {
                 .magic = SECONDARY_OSD_OP_MAGIC,

osd_id.h

@@ -5,6 +5,7 @@
 
 #define POOL_SCHEME_REPLICATED 1
 #define POOL_SCHEME_XOR 2
+#define POOL_SCHEME_JERASURE 3
 #define POOL_ID_MAX 0x10000
 #define POOL_ID_BITS 16
 #define INODE_POOL(inode) (pool_id_t)((inode) >> (64 - POOL_ID_BITS))

osd_peering.cpp

@@ -307,7 +307,7 @@ void osd_t::submit_sync_and_list_subop(osd_num_t role_osd, pg_peering_state_t *p
         osd_op_t *op = new osd_op_t();
         op->op_type = OSD_OP_OUT;
         op->peer_fd = cl->peer_fd;
-        op->req = {
+        op->req = (osd_any_op_t){
             .sec_sync = {
                 .header = {
                     .magic = SECONDARY_OSD_OP_MAGIC,
@@ -382,7 +382,7 @@ void osd_t::submit_list_subop(osd_num_t role_osd, pg_peering_state_t *ps)
         osd_op_t *op = new osd_op_t();
         op->op_type = OSD_OP_OUT;
         op->peer_fd = c_cli.osd_peer_fds[role_osd];
-        op->req = {
+        op->req = (osd_any_op_t){
             .sec_list = {
                 .header = {
                     .magic = SECONDARY_OSD_OP_MAGIC,

osd_peering_pg.h

@@ -75,7 +75,7 @@ struct pg_t
 {
     int state = 0;
     uint64_t scheme = 0;
-    uint64_t pg_cursize = 0, pg_size = 0, pg_minsize = 0;
+    uint64_t pg_cursize = 0, pg_size = 0, pg_minsize = 0, parity_chunks = 0;
     pool_id_t pool_id = 0;
     pg_num_t pg_num = 0;
     uint64_t clean_count = 0, total_count = 0;
@@ -94,7 +94,7 @@ struct pg_t
     std::vector<osd_num_t> cur_set;
     // same thing in state_dict-like format
     pg_osd_set_t cur_loc_set;
-    // moved object map. by default, each object is considered to reside on the cur_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

osd_primary.cpp

@@ -16,8 +16,9 @@ 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)
-    // K = pg_minsize in case of EC/XOR, or 1 for replicated pools
+    // K = (pg_size-parity_chunks) in case of EC/XOR, or 1 for replicated pools
     pool_id_t pool_id = INODE_POOL(cur_op->req.rw.inode);
+    // FIXME: We have to access pool config here, so make sure that it doesn't change while its PGs are active...
     auto pool_cfg_it = st_cli.pool_config.find(pool_id);
     if (pool_cfg_it == st_cli.pool_config.end())
     {
@@ -26,7 +27,8 @@ bool osd_t::prepare_primary_rw(osd_op_t *cur_op)
         return false;
     }
     auto & pool_cfg = pool_cfg_it->second;
-    uint64_t pg_block_size = bs_block_size * (pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_minsize);
+    uint64_t pg_data_size = (pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks);
+    uint64_t pg_block_size = bs_block_size * pg_data_size;
     object_id oid = {
         .inode = cur_op->req.rw.inode,
         // oid.stripe = starting offset of the parity stripe
@@ -37,6 +39,7 @@ bool osd_t::prepare_primary_rw(osd_op_t *cur_op)
     if (pg_it == pgs.end() || !(pg_it->second.state & PG_ACTIVE))
     {
         // This OSD is not primary for this PG or the PG is inactive
+        // FIXME: Allow reads from PGs degraded under pg_minsize, but don't allow writes
         finish_op(cur_op, -EPIPE);
         return false;
     }
@@ -54,9 +57,9 @@ bool osd_t::prepare_primary_rw(osd_op_t *cur_op)
     op_data->oid = oid;
     op_data->stripes = ((osd_rmw_stripe_t*)(op_data+1));
     op_data->scheme = pool_cfg.scheme;
+    op_data->pg_data_size = pg_data_size;
     cur_op->op_data = op_data;
-    split_stripes((pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pg_it->second.pg_minsize),
-        bs_block_size, (uint32_t)(cur_op->req.rw.offset - oid.stripe), cur_op->req.rw.len, op_data->stripes);
+    split_stripes(pg_data_size, bs_block_size, (uint32_t)(cur_op->req.rw.offset - oid.stripe), cur_op->req.rw.len, op_data->stripes);
     pg_it->second.inflight++;
     return true;
 }
@@ -101,7 +104,7 @@ void osd_t::continue_primary_read(osd_op_t *cur_op)
     else if (op_data->st == 2) goto resume_2;
     {
         auto & pg = pgs[{ .pool_id = INODE_POOL(op_data->oid.inode), .pg_num = op_data->pg_num }];
-        for (int role = 0; role < (op_data->scheme == POOL_SCHEME_REPLICATED ? 1 : pg.pg_minsize); role++)
+        for (int role = 0; role < op_data->pg_data_size; 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;
@@ -112,24 +115,23 @@ void osd_t::continue_primary_read(osd_op_t *cur_op)
         if (pg.state == PG_ACTIVE || op_data->scheme == POOL_SCHEME_REPLICATED)
         {
             // Fast happy-path
-            cur_op->buf = alloc_read_buffer(op_data->stripes,
-                (op_data->scheme == POOL_SCHEME_REPLICATED ? 1 : pg.pg_minsize), 0);
+            cur_op->buf = alloc_read_buffer(op_data->stripes, op_data->pg_data_size, 0);
             submit_primary_subops(SUBMIT_READ, op_data->target_ver,
-                (op_data->scheme == POOL_SCHEME_REPLICATED ? pg.pg_size : pg.pg_minsize), pg.cur_set.data(), cur_op);
+                (op_data->scheme == POOL_SCHEME_REPLICATED ? pg.pg_size : op_data->pg_data_size), pg.cur_set.data(), cur_op);
             op_data->st = 1;
         }
         else
         {
             // PG may be degraded or have misplaced objects
             uint64_t* cur_set = get_object_osd_set(pg, op_data->oid, pg.cur_set.data(), &op_data->object_state);
-            if (extend_missing_stripes(op_data->stripes, cur_set, pg.pg_minsize, pg.pg_size) < 0)
+            if (extend_missing_stripes(op_data->stripes, cur_set, op_data->pg_data_size, pg.pg_size) < 0)
             {
                 finish_op(cur_op, -EIO);
                 return;
             }
             // Submit reads
-            op_data->pg_minsize = pg.pg_minsize;
             op_data->pg_size = pg.pg_size;
+            op_data->scheme = pg.scheme;
             op_data->degraded = 1;
             cur_op->buf = alloc_read_buffer(op_data->stripes, pg.pg_size, 0);
             submit_primary_subops(SUBMIT_READ, op_data->target_ver, pg.pg_size, cur_set, cur_op);
@@ -147,14 +149,17 @@ resume_2:
     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 (op_data->scheme == POOL_SCHEME_XOR)
+        {
+            reconstruct_stripes_xor(stripes, op_data->pg_size);
+        }
+        else if (op_data->scheme == POOL_SCHEME_JERASURE)
+        {
+            reconstruct_stripes_jerasure(stripes, op_data->pg_size, op_data->pg_data_size);
+        }
+        for (int role = 0; role < op_data->pg_size; role++)
         {
-            if (stripes[role].read_end != 0 && stripes[role].missing)
-            {
-                reconstruct_stripe_xor(stripes, op_data->pg_size, role);
-            }
             if (stripes[role].req_end != 0)
             {
                 // Send buffer in parts to avoid copying
@@ -245,7 +250,7 @@ resume_1:
     else
     {
         cur_op->rmw_buf = calc_rmw(cur_op->buf, op_data->stripes, op_data->prev_set,
-            pg.pg_size, pg.pg_minsize, pg.pg_cursize, pg.cur_set.data(), bs_block_size);
+            pg.pg_size, op_data->pg_data_size, pg.pg_cursize, pg.cur_set.data(), bs_block_size);
         if (!cur_op->rmw_buf)
         {
             // Refuse partial overwrite of an incomplete object
@@ -285,7 +290,14 @@ resume_3:
     else
     {
         // Recover missing stripes, calculate parity
-        calc_rmw_parity_xor(op_data->stripes, pg.pg_size, op_data->prev_set, pg.cur_set.data(), bs_block_size);
+        if (pg.scheme == POOL_SCHEME_XOR)
+        {
+            calc_rmw_parity_xor(op_data->stripes, pg.pg_size, op_data->prev_set, pg.cur_set.data(), bs_block_size);
+        }
+        else if (pg.scheme == POOL_SCHEME_JERASURE)
+        {
+            calc_rmw_parity_jerasure(op_data->stripes, pg.pg_size, op_data->pg_data_size, op_data->prev_set, pg.cur_set.data(), bs_block_size);
+        }
     }
     // Send writes
     if ((op_data->fact_ver >> (64-PG_EPOCH_BITS)) < pg.epoch)
@@ -477,7 +489,11 @@ resume_7:
         }
         // Remember PG as dirty to drop the connection when PG goes offline
         // (this is required because of the "lazy sync")
-        c_cli.clients[cur_op->peer_fd]->dirty_pgs.insert({ .pool_id = pg.pool_id, .pg_num = pg.pg_num });
+        auto cl_it = c_cli.clients.find(cur_op->peer_fd);
+        if (cl_it != c_cli.clients.end())
+        {
+            cl_it->second->dirty_pgs.insert({ .pool_id = pg.pool_id, .pg_num = pg.pg_num });
+        }
         dirty_pgs.insert({ .pool_id = pg.pool_id, .pg_num = pg.pg_num });
     }
     return true;

osd_primary.h

@@ -25,7 +25,7 @@ struct osd_primary_op_data_t
     uint64_t fact_ver = 0;
     uint64_t scheme = 0;
     int n_subops = 0, done = 0, errors = 0, epipe = 0;
-    int degraded = 0, pg_size, pg_minsize;
+    int degraded = 0, pg_size, pg_data_size;
     osd_rmw_stripe_t *stripes;
     osd_op_t *subops = NULL;
     uint64_t *prev_set = NULL;

osd_primary_subops.cpp

@@ -11,7 +11,7 @@ void osd_t::autosync()
     {
         autosync_op = new osd_op_t();
         autosync_op->op_type = OSD_OP_IN;
-        autosync_op->req = {
+        autosync_op->req = (osd_any_op_t){
             .sync = {
                 .header = {
                     .magic = SECONDARY_OSD_OP_MAGIC,
@@ -295,7 +295,7 @@ void osd_t::handle_primary_subop(osd_op_t *subop, osd_op_t *cur_op)
             uint64_t version = subop->reply.sec_rw.version;
 #ifdef OSD_DEBUG
             uint64_t peer_osd = c_cli.clients.find(subop->peer_fd) != c_cli.clients.end()
-                ? c_cli.clients[subop->peer_fd].osd_num : osd_num;
+                ? c_cli.clients[subop->peer_fd]->osd_num : osd_num;
             printf("subop %lu from osd %lu: version = %lu\n", opcode, peer_osd, version);
 #endif
             if (op_data->fact_ver != 0 && op_data->fact_ver != version)
@@ -510,7 +510,7 @@ void osd_t::submit_primary_stab_subops(osd_op_t *cur_op)
         {
             clock_gettime(CLOCK_REALTIME, &subops[i].tv_begin);
             subops[i].op_type = (uint64_t)cur_op;
-            subops[i].bs_op = new blockstore_op_t({
+            subops[i].bs_op = new blockstore_op_t((blockstore_op_t){
                 .opcode = BS_OP_STABLE,
                 .callback = [subop = &subops[i], this](blockstore_op_t *bs_subop)
                 {

osd_rmw.cpp

@@ -1,12 +1,18 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.0 (see README.md for details)
 
+#include <stdexcept>
 #include <string.h>
 #include <assert.h>
+#include <jerasure/reed_sol.h>
+#include <jerasure.h>
+#include <map>
 #include "xor.h"
 #include "osd_rmw.h"
 #include "malloc_or_die.h"
 
+#define OSD_JERASURE_W 32
+
 static inline void extend_read(uint32_t start, uint32_t end, osd_rmw_stripe_t & stripe)
 {
     if (stripe.read_end == 0)
@@ -75,44 +81,189 @@ void split_stripes(uint64_t pg_minsize, uint32_t bs_block_size, uint32_t start,
     }
 }
 
-void reconstruct_stripe_xor(osd_rmw_stripe_t *stripes, int pg_size, int role)
+void reconstruct_stripes_xor(osd_rmw_stripe_t *stripes, int pg_size)
 {
-    int prev = -2;
-    for (int other = 0; other < pg_size; other++)
+    for (int role = 0; role < pg_size; role++)
     {
-        if (other != role)
+        if (stripes[role].read_end != 0 && stripes[role].missing)
         {
-            if (prev == -2)
+            // Reconstruct missing stripe (XOR k+1)
+            int prev = -2;
+            for (int other = 0; other < pg_size; other++)
             {
-                prev = other;
-            }
-            else if (prev >= 0)
-            {
-                assert(stripes[role].read_start >= stripes[prev].read_start &&
-                    stripes[role].read_start >= stripes[other].read_start);
-                memxor(
-                    stripes[prev].read_buf + (stripes[role].read_start - stripes[prev].read_start),
-                    stripes[other].read_buf + (stripes[role].read_start - stripes[other].read_start),
-                    stripes[role].read_buf, stripes[role].read_end - stripes[role].read_start
-                );
-                prev = -1;
-            }
-            else
-            {
-                assert(stripes[role].read_start >= stripes[other].read_start);
-                memxor(
-                    stripes[role].read_buf,
-                    stripes[other].read_buf + (stripes[role].read_start - stripes[other].read_start),
-                    stripes[role].read_buf, stripes[role].read_end - stripes[role].read_start
-                );
+                if (other != role)
+                {
+                    if (prev == -2)
+                    {
+                        prev = other;
+                    }
+                    else if (prev >= 0)
+                    {
+                        assert(stripes[role].read_start >= stripes[prev].read_start &&
+                            stripes[role].read_start >= stripes[other].read_start);
+                        memxor(
+                            stripes[prev].read_buf + (stripes[role].read_start - stripes[prev].read_start),
+                            stripes[other].read_buf + (stripes[role].read_start - stripes[other].read_start),
+                            stripes[role].read_buf, stripes[role].read_end - stripes[role].read_start
+                        );
+                        prev = -1;
+                    }
+                    else
+                    {
+                        assert(stripes[role].read_start >= stripes[other].read_start);
+                        memxor(
+                            stripes[role].read_buf,
+                            stripes[other].read_buf + (stripes[role].read_start - stripes[other].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)
+struct reed_sol_erased_t
 {
-    for (int role = 0; role < minsize; role++)
+    int *data;
+    int size;
+};
+
+inline bool operator < (const reed_sol_erased_t &a, const reed_sol_erased_t &b)
+{
+    for (int i = 0; i < a.size && i < b.size; i++)
+    {
+        if (a.data[i] < b.data[i])
+            return -1;
+        else if (a.data[i] > b.data[i])
+            return 1;
+    }
+    return 0;
+}
+
+struct reed_sol_matrix_t
+{
+    int refs = 0;
+    int *data;
+    std::map<reed_sol_erased_t, int*> decodings;
+};
+
+std::map<uint64_t, reed_sol_matrix_t> matrices;
+
+void use_jerasure(int pg_size, int pg_minsize, bool use)
+{
+    uint64_t key = (uint64_t)pg_size | ((uint64_t)pg_minsize) << 32;
+    auto rs_it = matrices.find(key);
+    if (rs_it == matrices.end())
+    {
+        if (!use)
+        {
+            return;
+        }
+        int *matrix = reed_sol_vandermonde_coding_matrix(pg_minsize, pg_size-pg_minsize, OSD_JERASURE_W);
+        matrices[key] = (reed_sol_matrix_t){
+            .refs = 0,
+            .data = matrix,
+        };
+        rs_it = matrices.find(key);
+    }
+    rs_it->second.refs += (!use ? -1 : 1);
+    if (rs_it->second.refs <= 0)
+    {
+        free(rs_it->second.data);
+        for (auto dec_it = rs_it->second.decodings.begin(); dec_it != rs_it->second.decodings.end();)
+        {
+            int *data = dec_it->second;
+            rs_it->second.decodings.erase(dec_it++);
+            free(data);
+        }
+        matrices.erase(rs_it);
+    }
+}
+
+reed_sol_matrix_t* get_jerasure_matrix(int pg_size, int pg_minsize)
+{
+    uint64_t key = (uint64_t)pg_size | ((uint64_t)pg_minsize) << 32;
+    auto rs_it = matrices.find(key);
+    if (rs_it == matrices.end())
+    {
+        throw std::runtime_error("jerasure matrix not initialized");
+    }
+    return &rs_it->second;
+}
+
+// jerasure_matrix_decode() decodes all chunks at once and tries to reencode all missing coding chunks.
+// we don't need it. also it makes an extra allocation of int *erased on every call and doesn't cache
+// the decoding matrix.
+// all these flaws are fixed in this function:
+int* get_jerasure_decoding_matrix(osd_rmw_stripe_t *stripes, int pg_size, int pg_minsize)
+{
+    int edd = 0;
+    int erased[pg_size] = { 0 };
+    for (int i = 0; i < pg_size; i++)
+        if (stripes[i].read_end == 0 || stripes[i].missing)
+            erased[i] = 1;
+    for (int i = 0; i < pg_minsize; i++)
+        if (stripes[i].read_end != 0 && stripes[i].missing)
+            edd++;
+    if (edd == 0)
+        return NULL;
+    reed_sol_matrix_t *matrix = get_jerasure_matrix(pg_size, pg_minsize);
+    auto dec_it = matrix->decodings.find((reed_sol_erased_t){ .data = erased, .size = pg_size });
+    if (dec_it == matrix->decodings.end())
+    {
+        int *dm_ids = (int*)malloc(sizeof(int)*(pg_minsize + pg_minsize*pg_minsize + pg_size));
+        int *decoding_matrix = dm_ids + pg_minsize;
+        if (!dm_ids)
+            throw std::bad_alloc();
+        // we always use row_k_ones=1 and w=8 (OSD_JERASURE_W)
+        if (jerasure_make_decoding_matrix(pg_minsize, pg_size-pg_minsize, OSD_JERASURE_W, matrix->data, erased, decoding_matrix, dm_ids) < 0)
+        {
+            free(dm_ids);
+            throw std::runtime_error("jerasure_make_decoding_matrix() failed");
+        }
+        int *erased_copy = dm_ids + pg_minsize + pg_minsize*pg_minsize;
+        memcpy(erased_copy, erased, pg_size*sizeof(int));
+        matrix->decodings.emplace((reed_sol_erased_t){ .data = erased_copy, .size = pg_size }, dm_ids);
+        return dm_ids;
+    }
+    return dec_it->second;
+}
+
+void reconstruct_stripes_jerasure(osd_rmw_stripe_t *stripes, int pg_size, int pg_minsize)
+{
+    int *dm_ids = get_jerasure_decoding_matrix(stripes, pg_size, pg_minsize);
+    if (!dm_ids)
+    {
+        return;
+    }
+    int *decoding_matrix = dm_ids + pg_minsize;
+    char *data_ptrs[pg_size] = { 0 };
+    for (int role = 0; role < pg_minsize; role++)
+    {
+        if (stripes[role].read_end != 0 && stripes[role].missing)
+        {
+            for (int other = 0; other < pg_size; other++)
+            {
+                if (stripes[other].read_end != 0 && !stripes[other].missing)
+                {
+                    assert(stripes[other].read_start <= stripes[role].read_start);
+                    assert(stripes[other].read_end >= stripes[role].read_end);
+                    data_ptrs[other] = (char*)(stripes[other].read_buf + (stripes[role].read_start - stripes[other].read_start));
+                }
+            }
+            data_ptrs[role] = (char*)stripes[role].read_buf;
+            jerasure_matrix_dotprod(
+                pg_minsize, OSD_JERASURE_W, decoding_matrix+(role*pg_minsize), dm_ids, role,
+                data_ptrs, data_ptrs+pg_minsize, stripes[role].read_end - stripes[role].read_start
+            );
+        }
+    }
+}
+
+int extend_missing_stripes(osd_rmw_stripe_t *stripes, osd_num_t *osd_set, int pg_minsize, int pg_size)
+{
+    for (int role = 0; role < pg_minsize; role++)
     {
         if (stripes[role].read_end != 0 && osd_set[role] == 0)
         {
@@ -121,21 +272,21 @@ int extend_missing_stripes(osd_rmw_stripe_t *stripes, osd_num_t *osd_set, int mi
             // 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++)
+            for (int j = 0; j < pg_size; j++)
             {
                 if (osd_set[j] != 0)
                 {
                     extend_read(stripes[role].read_start, stripes[role].read_end, stripes[j]);
                     exist++;
-                    if (exist >= minsize)
+                    if (exist >= pg_minsize)
                     {
                         break;
                     }
                 }
             }
-            if (exist < minsize)
+            if (exist < pg_minsize)
             {
-                // Less than minsize stripes are available for this object
+                // Less than pg_minsize stripes are available for this object
                 return -1;
             }
         }
@@ -369,19 +520,9 @@ static void xor_multiple_buffers(buf_len_t *xor1, int n1, buf_len_t *xor2, int n
     }
 }
 
-void calc_rmw_parity_xor(osd_rmw_stripe_t *stripes, int pg_size, uint64_t *read_osd_set, uint64_t *write_osd_set, uint32_t chunk_size)
+static void calc_rmw_parity_copy_mod(osd_rmw_stripe_t *stripes, int pg_size, int pg_minsize,
+    uint64_t *read_osd_set, uint64_t *write_osd_set, uint32_t chunk_size, uint32_t &start, uint32_t &end)
 {
-    int pg_minsize = pg_size-1;
-    for (int role = 0; role < pg_size; role++)
-    {
-        if (stripes[role].read_end != 0 && stripes[role].missing)
-        {
-            // Reconstruct missing stripe (XOR k+1)
-            reconstruct_stripe_xor(stripes, pg_size, role);
-            break;
-        }
-    }
-    uint32_t start = 0, end = 0;
     if (write_osd_set[pg_minsize] != 0 || write_osd_set != read_osd_set)
     {
         // Required for the next two if()s
@@ -421,6 +562,53 @@ void calc_rmw_parity_xor(osd_rmw_stripe_t *stripes, int pg_size, uint64_t *read_
             }
         }
     }
+}
+
+static void calc_rmw_parity_copy_parity(osd_rmw_stripe_t *stripes, int pg_size, int pg_minsize,
+    uint64_t *read_osd_set, uint64_t *write_osd_set, uint32_t chunk_size, uint32_t start, uint32_t end)
+{
+    if (write_osd_set != read_osd_set)
+    {
+        for (int role = pg_minsize; role < pg_size; role++)
+        {
+            if (write_osd_set[role] != read_osd_set[role] && (start != 0 || end != chunk_size))
+            {
+                // Copy new parity into the read buffer to write it back
+                memcpy(
+                    stripes[role].read_buf + start,
+                    stripes[role].write_buf,
+                    end - start
+                );
+                stripes[role].write_buf = stripes[role].read_buf;
+                stripes[role].write_start = 0;
+                stripes[role].write_end = chunk_size;
+            }
+        }
+    }
+#ifdef RMW_DEBUG
+    printf("calc_rmw_parity:\n");
+    for (int role = 0; role < pg_size; role++)
+    {
+        auto & s = stripes[role];
+        printf(
+            "Tr=%lu Tw=%lu Q=%x-%x R=%x-%x W=%x-%x Rb=%lx Wb=%lx\n",
+            read_osd_set[role], write_osd_set[role],
+            s.req_start, s.req_end,
+            s.read_start, s.read_end,
+            s.write_start, s.write_end,
+            (uint64_t)s.read_buf,
+            (uint64_t)s.write_buf
+        );
+    }
+#endif
+}
+
+void calc_rmw_parity_xor(osd_rmw_stripe_t *stripes, int pg_size, uint64_t *read_osd_set, uint64_t *write_osd_set, uint32_t chunk_size)
+{
+    int pg_minsize = pg_size-1;
+    reconstruct_stripes_xor(stripes, pg_size);
+    uint32_t start = 0, end = 0;
+    calc_rmw_parity_copy_mod(stripes, pg_size, pg_minsize, read_osd_set, write_osd_set, chunk_size, start, end);
     if (write_osd_set[pg_minsize] != 0 && end != 0)
     {
         // Calculate new parity (XOR k+1)
@@ -449,38 +637,71 @@ void calc_rmw_parity_xor(osd_rmw_stripe_t *stripes, int pg_size, uint64_t *read_
             }
         }
     }
-    if (write_osd_set != read_osd_set)
+    calc_rmw_parity_copy_parity(stripes, pg_size, pg_minsize, read_osd_set, write_osd_set, chunk_size, start, end);
+}
+
+void calc_rmw_parity_jerasure(osd_rmw_stripe_t *stripes, int pg_size, int pg_minsize,
+    uint64_t *read_osd_set, uint64_t *write_osd_set, uint32_t chunk_size)
+{
+    reed_sol_matrix_t *matrix = get_jerasure_matrix(pg_size, pg_minsize);
+    reconstruct_stripes_jerasure(stripes, pg_size, pg_minsize);
+    uint32_t start = 0, end = 0;
+    calc_rmw_parity_copy_mod(stripes, pg_size, pg_minsize, read_osd_set, write_osd_set, chunk_size, start, end);
+    if (end != 0)
     {
-        for (int role = pg_minsize; role < pg_size; role++)
+        int i;
+        for (i = pg_minsize; i < pg_size; i++)
         {
-            if (write_osd_set[role] != read_osd_set[role] && (start != 0 || end != chunk_size))
+            if (write_osd_set[i] != 0)
+                break;
+        }
+        if (i < pg_size)
+        {
+            // Calculate new coding chunks
+            buf_len_t bufs[pg_size][3];
+            int nbuf[pg_size] = { 0 }, curbuf[pg_size] = { 0 };
+            uint32_t positions[pg_size];
+            void *data_ptrs[pg_size] = { 0 };
+            for (int i = 0; i < pg_minsize; i++)
             {
-                // Copy new parity into the read buffer to write it back
-                memcpy(
-                    stripes[role].read_buf + start,
-                    stripes[role].write_buf,
-                    end - start
+                get_old_new_buffers(stripes[i], start, end, bufs[i], nbuf[i]);
+                positions[i] = start;
+            }
+            for (int i = pg_minsize; i < pg_size; i++)
+            {
+                bufs[i][nbuf[i]++] = { .buf = stripes[i].write_buf, .len = end-start };
+                positions[i] = start;
+            }
+            uint32_t pos = start;
+            while (pos < end)
+            {
+                uint32_t next_end = end;
+                for (int i = 0; i < pg_size; i++)
+                {
+                    assert(curbuf[i] < nbuf[i]);
+                    assert(bufs[i][curbuf[i]].buf);
+                    data_ptrs[i] = bufs[i][curbuf[i]].buf + pos-positions[i];
+                    uint32_t this_end = bufs[i][curbuf[i]].len + positions[i];
+                    if (next_end > this_end)
+                        next_end = this_end;
+                }
+                assert(next_end > pos);
+                for (int i = 0; i < pg_size; i++)
+                {
+                    uint32_t this_end = bufs[i][curbuf[i]].len + positions[i];
+                    if (next_end >= this_end)
+                    {
+                        positions[i] += bufs[i][curbuf[i]].len;
+                        curbuf[i]++;
+                    }
+                }
+                jerasure_matrix_encode(
+                    pg_minsize, pg_size-pg_minsize, OSD_JERASURE_W, matrix->data,
+                    (char**)data_ptrs, (char**)data_ptrs+pg_minsize, next_end-pos
                 );
-                stripes[role].write_buf = stripes[role].read_buf;
-                stripes[role].write_start = 0;
-                stripes[role].write_end = chunk_size;
+                pos = next_end;
             }
         }
     }
-#ifdef RMW_DEBUG
-    printf("calc_rmw_xor:\n");
-    for (int role = 0; role < pg_size; role++)
-    {
-        auto & s = stripes[role];
-        printf(
-            "Tr=%lu Tw=%lu Q=%x-%x R=%x-%x W=%x-%x Rb=%lx Wb=%lx\n",
-            read_osd_set[role], write_osd_set[role],
-            s.req_start, s.req_end,
-            s.read_start, s.read_end,
-            s.write_start, s.write_end,
-            (uint64_t)s.read_buf,
-            (uint64_t)s.write_buf
-        );
-    }
-#endif
+    calc_rmw_parity_copy_parity(stripes, pg_size, pg_minsize, read_osd_set, write_osd_set, chunk_size, start, end);
 }

osd_rmw.h

@@ -26,11 +26,13 @@ struct osd_rmw_stripe_t
     bool missing;
 };
 
+// Here pg_minsize is the number of data chunks, not the minimum number of alive OSDs for the PG to operate
+
 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_xor(osd_rmw_stripe_t *stripes, int pg_size, int role);
+void reconstruct_stripes_xor(osd_rmw_stripe_t *stripes, int pg_size);
 
-int extend_missing_stripes(osd_rmw_stripe_t *stripes, osd_num_t *osd_set, int minsize, int size);
+int extend_missing_stripes(osd_rmw_stripe_t *stripes, osd_num_t *osd_set, int pg_minsize, int pg_size);
 
 void* alloc_read_buffer(osd_rmw_stripe_t *stripes, int read_pg_size, uint64_t add_size);
 
@@ -38,3 +40,10 @@ void* calc_rmw(void *request_buf, osd_rmw_stripe_t *stripes, uint64_t *read_osd_
     uint64_t pg_size, uint64_t pg_minsize, uint64_t pg_cursize, uint64_t *write_osd_set, uint64_t chunk_size);
 
 void calc_rmw_parity_xor(osd_rmw_stripe_t *stripes, int pg_size, uint64_t *read_osd_set, uint64_t *write_osd_set, uint32_t chunk_size);
+
+void use_jerasure(int pg_size, int pg_minsize, bool use);
+
+void reconstruct_stripes_jerasure(osd_rmw_stripe_t *stripes, int pg_size, int pg_minsize);
+
+void calc_rmw_parity_jerasure(osd_rmw_stripe_t *stripes, int pg_size, int pg_minsize,
+    uint64_t *read_osd_set, uint64_t *write_osd_set, uint32_t chunk_size);

osd_rmw_test.cpp

@@ -18,107 +18,8 @@ void test9();
 void test10();
 void test11();
 void test12();
-
-/***
-
-Cases:
-
-1. split(offset=128K-4K, len=8K)
-   = [ [ 128K-4K, 128K ], [ 0, 4K ], [ 0, 0 ] ]
-
-2. read(offset=128K-4K, len=8K, osd_set=[1,0,3])
-   = { read: [ [ 0, 128K ], [ 0, 4K ], [ 0, 4K ] ] }
-
-3. cover_read(0, 128K, { req: [ 128K-4K, 4K ] })
-   = { read: [ 0, 128K-4K ] }
-
-4. write(offset=128K-4K, len=8K, osd_set=[1,0,3])
-   = {
-     read: [ [ 0, 128K ], [ 4K, 128K ], [ 4K, 128K ] ],
-     write: [ [ 128K-4K, 128K ], [ 0, 4K ], [ 0, 128K ] ],
-     input buffer: [ write0, write1 ],
-     rmw buffer: [ write2, read0, read1, read2 ],
-   }
-   + check write2 buffer
-
-5. write(offset=0, len=128K+64K, osd_set=[1,0,3])
-   = {
-     req: [ [ 0, 128K ], [ 0, 64K ], [ 0, 0 ] ],
-     read: [ [ 64K, 128K ], [ 64K, 128K ], [ 64K, 128K ] ],
-     write: [ [ 0, 128K ], [ 0, 64K ], [ 0, 128K ] ],
-     input buffer: [ write0, write1 ],
-     rmw buffer: [ write2, read0, read1, read2 ],
-   }
-
-6. write(offset=0, len=128K+64K, osd_set=[1,2,3])
-   = {
-     req: [ [ 0, 128K ], [ 0, 64K ], [ 0, 0 ] ],
-     read: [ [ 0, 0 ], [ 64K, 128K ], [ 0, 0 ] ],
-     write: [ [ 0, 128K ], [ 0, 64K ], [ 0, 128K ] ],
-     input buffer: [ write0, write1 ],
-     rmw buffer: [ write2, read1 ],
-   }
-
-7. calc_rmw(offset=128K-4K, len=8K, osd_set=[1,0,3], write_set=[1,2,3])
-   = {
-     read: [ [ 0, 128K ], [ 0, 128K ], [ 0, 128K ] ],
-     write: [ [ 128K-4K, 128K ], [ 0, 4K ], [ 0, 128K ] ],
-     input buffer: [ write0, write1 ],
-     rmw buffer: [ write2, read0, read1, read2 ],
-   }
-   then, after calc_rmw_parity_xor(): {
-     write: [ [ 128K-4K, 128K ], [ 0, 128K ], [ 0, 128K ] ],
-     write1==read1,
-   }
-   + check write1 buffer
-   + check write2 buffer
-
-8. calc_rmw(offset=0, len=128K+4K, osd_set=[0,2,3], write_set=[1,2,3])
-   = {
-     read: [ [ 0, 0 ], [ 4K, 128K ], [ 0, 0 ] ],
-     write: [ [ 0, 128K ], [ 0, 4K ], [ 0, 128K ] ],
-     input buffer: [ write0, write1 ],
-     rmw buffer: [ write2, read1 ],
-   }
-   + check write2 buffer
-
-9. object recovery case:
-   calc_rmw(offset=0, len=0, read_osd_set=[0,2,3], write_osd_set=[1,2,3])
-   = {
-     read: [ [ 0, 128K ], [ 0, 128K ], [ 0, 128K ] ],
-     write: [ [ 0, 0 ], [ 0, 0 ], [ 0, 0 ] ],
-     input buffer: NULL,
-     rmw buffer: [ read0, read1, read2 ],
-   }
-   then, after calc_rmw_parity_xor(): {
-     write: [ [ 0, 128K ], [ 0, 0 ], [ 0, 0 ] ],
-     write0==read0,
-   }
-   + check write0 buffer
-
-10. full overwrite/recovery case:
-   calc_rmw(offset=0, len=256K, read_osd_set=[1,0,0], write_osd_set=[1,2,3])
-   = {
-     read: [ [ 0, 0 ], [ 0, 0 ], [ 0, 0 ] ],
-     write: [ [ 0, 128K ], [ 0, 128K ], [ 0, 128K ] ],
-     input buffer: [ write0, write1 ],
-     rmw buffer: [ write2 ],
-   }
-   then, after calc_rmw_parity_xor(): all the same
-   + check write2 buffer
-
-10. partial recovery case:
-   calc_rmw(offset=128K, len=128K, read_osd_set=[1,0,0], write_osd_set=[1,2,3])
-   = {
-     read: [ [ 0, 128K ], [ 0, 0 ], [ 0, 0 ] ],
-     write: [ [ 0, 0 ], [ 0, 128K ], [ 0, 128K ] ],
-     input buffer: [ write1 ],
-     rmw buffer: [ write2, read0 ],
-   }
-   then, after calc_rmw_parity_xor(): all the same
-   + check write2 buffer
-
-***/
+void test13();
+void test14();
 
 int main(int narg, char *args[])
 {
@@ -142,6 +43,10 @@ int main(int narg, char *args[])
     test11();
     // Test 12
     test12();
+    // Test 13
+    test13();
+    // Test 14
+    test14();
     // End
     printf("all ok\n");
     return 0;
@@ -169,6 +74,19 @@ void dump_stripes(osd_rmw_stripe_t *stripes, int pg_size)
     printf("\n");
 }
 
+/***
+
+1. split(offset=128K-4K, len=8K)
+   = [ [ 128K-4K, 128K ], [ 0, 4K ], [ 0, 0 ] ]
+
+   read(offset=128K-4K, len=8K, osd_set=[1,0,3])
+   = { read: [ [ 0, 128K ], [ 0, 4K ], [ 0, 4K ] ] }
+
+   cover_read(0, 128K, { req: [ 128K-4K, 4K ] })
+   = { read: [ 0, 128K-4K ] }
+
+***/
+
 void test1()
 {
     osd_num_t osd_set[3] = { 1, 0, 3 };
@@ -188,11 +106,24 @@ void test1()
     assert(stripes[0].read_start == 0 && stripes[0].read_end == 128*1024);
     assert(stripes[2].read_start == 0 && stripes[2].read_end == 4096);
     // Test 1.3
-    stripes[0] = { .req_start = 128*1024-4096, .req_end = 128*1024 };
+    stripes[0] = (osd_rmw_stripe_t){ .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);
 }
 
+/***
+
+4. write(offset=128K-4K, len=8K, osd_set=[1,0,3])
+   = {
+     read: [ [ 0, 128K ], [ 4K, 128K ], [ 4K, 128K ] ],
+     write: [ [ 128K-4K, 128K ], [ 0, 4K ], [ 0, 128K ] ],
+     input buffer: [ write0, write1 ],
+     rmw buffer: [ write2, read0, read1, read2 ],
+   }
+   + check write2 buffer
+
+***/
+
 void test4()
 {
     osd_num_t osd_set[3] = { 1, 0, 3 };
@@ -226,6 +157,19 @@ void test4()
     free(write_buf);
 }
 
+/***
+
+5. write(offset=0, len=128K+64K, osd_set=[1,0,3])
+   = {
+     req: [ [ 0, 128K ], [ 0, 64K ], [ 0, 0 ] ],
+     read: [ [ 64K, 128K ], [ 64K, 128K ], [ 64K, 128K ] ],
+     write: [ [ 0, 128K ], [ 0, 64K ], [ 0, 128K ] ],
+     input buffer: [ write0, write1 ],
+     rmw buffer: [ write2, read0, read1, read2 ],
+   }
+
+***/
+
 void test5()
 {
     osd_num_t osd_set[3] = { 1, 0, 3 };
@@ -254,6 +198,19 @@ void test5()
     free(write_buf);
 }
 
+/***
+
+6. write(offset=0, len=128K+64K, osd_set=[1,2,3])
+   = {
+     req: [ [ 0, 128K ], [ 0, 64K ], [ 0, 0 ] ],
+     read: [ [ 0, 0 ], [ 64K, 128K ], [ 0, 0 ] ],
+     write: [ [ 0, 128K ], [ 0, 64K ], [ 0, 128K ] ],
+     input buffer: [ write0, write1 ],
+     rmw buffer: [ write2, read1 ],
+   }
+
+***/
+
 void test6()
 {
     osd_num_t osd_set[3] = { 1, 2, 3 };
@@ -278,6 +235,24 @@ void test6()
     free(write_buf);
 }
 
+/***
+
+7. calc_rmw(offset=128K-4K, len=8K, osd_set=[1,0,3], write_set=[1,2,3])
+   = {
+     read: [ [ 0, 128K ], [ 0, 128K ], [ 0, 128K ] ],
+     write: [ [ 128K-4K, 128K ], [ 0, 4K ], [ 0, 128K ] ],
+     input buffer: [ write0, write1 ],
+     rmw buffer: [ write2, read0, read1, read2 ],
+   }
+   then, after calc_rmw_parity_xor(): {
+     write: [ [ 128K-4K, 128K ], [ 0, 128K ], [ 0, 128K ] ],
+     write1==read1,
+   }
+   + check write1 buffer
+   + check write2 buffer
+
+***/
+
 void test7()
 {
     osd_num_t osd_set[3] = { 1, 0, 3 };
@@ -318,6 +293,19 @@ void test7()
     free(write_buf);
 }
 
+/***
+
+8. calc_rmw(offset=0, len=128K+4K, osd_set=[0,2,3], write_set=[1,2,3])
+   = {
+     read: [ [ 0, 0 ], [ 4K, 128K ], [ 0, 0 ] ],
+     write: [ [ 0, 128K ], [ 0, 4K ], [ 0, 128K ] ],
+     input buffer: [ write0, write1 ],
+     rmw buffer: [ write2, read1 ],
+   }
+   + check write2 buffer
+
+***/
+
 void test8()
 {
     osd_num_t osd_set[3] = { 0, 2, 3 };
@@ -355,6 +343,24 @@ void test8()
     free(write_buf);
 }
 
+/***
+
+9. object recovery case:
+   calc_rmw(offset=0, len=0, read_osd_set=[0,2,3], write_osd_set=[1,2,3])
+   = {
+     read: [ [ 0, 128K ], [ 0, 128K ], [ 0, 128K ] ],
+     write: [ [ 0, 0 ], [ 0, 0 ], [ 0, 0 ] ],
+     input buffer: NULL,
+     rmw buffer: [ read0, read1, read2 ],
+   }
+   then, after calc_rmw_parity_xor(): {
+     write: [ [ 0, 128K ], [ 0, 0 ], [ 0, 0 ] ],
+     write0==read0,
+   }
+   + check write0 buffer
+
+***/
+
 void test9()
 {
     osd_num_t osd_set[3] = { 0, 2, 3 };
@@ -395,6 +401,21 @@ void test9()
     free(rmw_buf);
 }
 
+/***
+
+10. full overwrite/recovery case:
+   calc_rmw(offset=0, len=256K, read_osd_set=[1,0,0], write_osd_set=[1,2,3])
+   = {
+     read: [ [ 0, 0 ], [ 0, 0 ], [ 0, 0 ] ],
+     write: [ [ 0, 128K ], [ 0, 128K ], [ 0, 128K ] ],
+     input buffer: [ write0, write1 ],
+     rmw buffer: [ write2 ],
+   }
+   then, after calc_rmw_parity_xor(): all the same
+   + check write2 buffer
+
+***/
+
 void test10()
 {
     osd_num_t osd_set[3] = { 1, 0, 0 };
@@ -436,6 +457,21 @@ void test10()
     free(write_buf);
 }
 
+/***
+
+11. partial recovery case:
+   calc_rmw(offset=128K, len=128K, read_osd_set=[1,0,0], write_osd_set=[1,2,3])
+   = {
+     read: [ [ 0, 128K ], [ 0, 0 ], [ 0, 0 ] ],
+     write: [ [ 0, 0 ], [ 0, 128K ], [ 0, 128K ] ],
+     input buffer: [ write1 ],
+     rmw buffer: [ write2, read0 ],
+   }
+   then, after calc_rmw_parity_xor(): all the same
+   + check write2 buffer
+
+***/
+
 void test11()
 {
     osd_num_t osd_set[3] = { 1, 0, 0 };
@@ -477,17 +513,32 @@ void test11()
     free(write_buf);
 }
 
+/***
+
+12. parity recovery case:
+   calc_rmw(offset=0, len=0, read_osd_set=[1,2,0], write_osd_set=[1,2,3])
+   = {
+     read: [ [ 0, 128K ], [ 0, 128K ], [ 0, 0 ] ],
+     write: [ [ 0, 0 ], [ 0, 0 ], [ 0, 128K ] ],
+     input buffer: [],
+     rmw buffer: [ write2, read0, read1 ],
+   }
+   then, after calc_rmw_parity_xor(): all the same
+   + check write2 buffer
+
+***/
+
 void test12()
 {
     osd_num_t osd_set[3] = { 1, 2, 0 };
     osd_num_t write_osd_set[3] = { 1, 2, 3 };
     osd_rmw_stripe_t stripes[3] = { 0 };
-    // Test 11.0
+    // Test 12.0
     split_stripes(2, 128*1024, 0, 0, stripes);
     assert(stripes[0].req_start == 0 && stripes[0].req_end == 0);
     assert(stripes[1].req_start == 0 && stripes[1].req_end == 0);
     assert(stripes[2].req_start == 0 && stripes[2].req_end == 0);
-    // Test 11.1
+    // Test 12.1
     void *rmw_buf = calc_rmw(NULL, stripes, osd_set, 3, 2, 3, write_osd_set, 128*1024);
     assert(rmw_buf);
     assert(stripes[0].read_start == 0 && stripes[0].read_end == 128*1024);
@@ -502,7 +553,7 @@ void test12()
     assert(stripes[0].write_buf == NULL);
     assert(stripes[1].write_buf == NULL);
     assert(stripes[2].write_buf == rmw_buf);
-    // Test 11.2
+    // Test 12.2
     set_pattern(stripes[0].read_buf, 128*1024, PATTERN1);
     set_pattern(stripes[1].read_buf, 128*1024, PATTERN2);
     calc_rmw_parity_xor(stripes, 3, osd_set, write_osd_set, 128*1024);
@@ -515,3 +566,217 @@ void test12()
     check_pattern(stripes[2].write_buf, 128*1024, PATTERN1^PATTERN2);
     free(rmw_buf);
 }
+
+/***
+
+13. basic jerasure 2+2 test
+   calc_rmw(offset=128K-4K, len=8K, osd_set=[1,2,0,0], write_set=[1,2,3,4])
+   = {
+     read: [ [ 0, 128K ], [ 0, 128K ], [ 0, 0 ], [ 0, 0 ] ],
+     write: [ [ 128K-4K, 128K ], [ 0, 4K ], [ 0, 128K ], [ 0, 128K ] ],
+     input buffer: [ write0, write1 ],
+     rmw buffer: [ write2, write3, read0, read1 ],
+   }
+   then, after calc_rmw_parity_jerasure(): all the same
+   then simulate read with read_osd_set=[0,0,3,4] and check read0,read1 buffers
+
+***/
+
+void test13()
+{
+    use_jerasure(4, 2, true);
+    osd_num_t osd_set[4] = { 1, 2, 0, 0 };
+    osd_num_t write_osd_set[4] = { 1, 2, 3, 4 };
+    osd_rmw_stripe_t stripes[4] = { 0 };
+    // Test 13.0
+    void *write_buf = malloc_or_die(8192);
+    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_start == 0 && stripes[2].req_end == 0);
+    assert(stripes[3].req_start == 0 && stripes[3].req_end == 0);
+    // Test 13.1
+    void *rmw_buf = calc_rmw(write_buf, stripes, osd_set, 4, 2, 4, write_osd_set, 128*1024);
+    assert(rmw_buf);
+    assert(stripes[0].read_start == 0 && stripes[0].read_end == 128*1024-4096);
+    assert(stripes[1].read_start == 4096 && stripes[1].read_end == 128*1024);
+    assert(stripes[2].read_start == 0 && stripes[2].read_end == 0);
+    assert(stripes[3].read_start == 0 && stripes[3].read_end == 0);
+    assert(stripes[0].write_start == 128*1024-4096 && stripes[0].write_end == 128*1024);
+    assert(stripes[1].write_start == 0 && stripes[1].write_end == 4096);
+    assert(stripes[2].write_start == 0 && stripes[2].write_end == 128*1024);
+    assert(stripes[3].write_start == 0 && stripes[3].write_end == 128*1024);
+    assert(stripes[0].read_buf == rmw_buf+2*128*1024);
+    assert(stripes[1].read_buf == rmw_buf+3*128*1024-4096);
+    assert(stripes[2].read_buf == NULL);
+    assert(stripes[3].read_buf == NULL);
+    assert(stripes[0].write_buf == write_buf);
+    assert(stripes[1].write_buf == write_buf+4096);
+    assert(stripes[2].write_buf == rmw_buf);
+    assert(stripes[3].write_buf == rmw_buf+128*1024);
+    // Test 13.2 - encode
+    set_pattern(write_buf, 8192, PATTERN3);
+    set_pattern(stripes[0].read_buf, 128*1024-4096, PATTERN1);
+    set_pattern(stripes[1].read_buf, 128*1024-4096, PATTERN2);
+    calc_rmw_parity_jerasure(stripes, 4, 2, osd_set, write_osd_set, 128*1024);
+    assert(stripes[0].write_start == 128*1024-4096 && stripes[0].write_end == 128*1024);
+    assert(stripes[1].write_start == 0 && stripes[1].write_end == 4096);
+    assert(stripes[2].write_start == 0 && stripes[2].write_end == 128*1024);
+    assert(stripes[3].write_start == 0 && stripes[3].write_end == 128*1024);
+    assert(stripes[0].write_buf == write_buf);
+    assert(stripes[1].write_buf == write_buf+4096);
+    assert(stripes[2].write_buf == rmw_buf);
+    assert(stripes[3].write_buf == rmw_buf+128*1024);
+    // Test 13.3 - full decode and verify
+    osd_num_t read_osd_set[4] = { 0, 0, 3, 4 };
+    memset(stripes, 0, sizeof(stripes));
+    split_stripes(2, 128*1024, 0, 256*1024, stripes);
+    assert(stripes[0].req_start == 0 && stripes[0].req_end == 128*1024);
+    assert(stripes[1].req_start == 0 && stripes[1].req_end == 128*1024);
+    assert(stripes[2].req_start == 0 && stripes[2].req_end == 0);
+    assert(stripes[3].req_start == 0 && stripes[3].req_end == 0);
+    for (int role = 0; role < 4; role++)
+    {
+        stripes[role].read_start = stripes[role].req_start;
+        stripes[role].read_end = stripes[role].req_end;
+    }
+    assert(extend_missing_stripes(stripes, read_osd_set, 2, 4) == 0);
+    assert(stripes[0].read_start == 0 && stripes[0].read_end == 128*1024);
+    assert(stripes[1].read_start == 0 && stripes[1].read_end == 128*1024);
+    assert(stripes[2].read_start == 0 && stripes[2].read_end == 128*1024);
+    assert(stripes[3].read_start == 0 && stripes[3].read_end == 128*1024);
+    void *read_buf = alloc_read_buffer(stripes, 4, 0);
+    assert(read_buf);
+    assert(stripes[0].read_buf == read_buf);
+    assert(stripes[1].read_buf == read_buf+128*1024);
+    assert(stripes[2].read_buf == read_buf+2*128*1024);
+    assert(stripes[3].read_buf == read_buf+3*128*1024);
+    memcpy(read_buf+2*128*1024, rmw_buf, 128*1024);
+    memcpy(read_buf+3*128*1024, rmw_buf+128*1024, 128*1024);
+    reconstruct_stripes_jerasure(stripes, 4, 2);
+    check_pattern(stripes[0].read_buf, 128*1024-4096, PATTERN1);
+    check_pattern(stripes[0].read_buf+128*1024-4096, 4096, PATTERN3);
+    check_pattern(stripes[1].read_buf, 4096, PATTERN3);
+    check_pattern(stripes[1].read_buf+4096, 128*1024-4096, PATTERN2);
+    free(read_buf);
+    // Test 13.4 - partial decode (only 1st chunk) and verify
+    memset(stripes, 0, sizeof(stripes));
+    split_stripes(2, 128*1024, 0, 128*1024, stripes);
+    assert(stripes[0].req_start == 0 && stripes[0].req_end == 128*1024);
+    assert(stripes[1].req_start == 0 && stripes[1].req_end == 0);
+    assert(stripes[2].req_start == 0 && stripes[2].req_end == 0);
+    assert(stripes[3].req_start == 0 && stripes[3].req_end == 0);
+    for (int role = 0; role < 4; role++)
+    {
+        stripes[role].read_start = stripes[role].req_start;
+        stripes[role].read_end = stripes[role].req_end;
+    }
+    assert(extend_missing_stripes(stripes, read_osd_set, 2, 4) == 0);
+    assert(stripes[0].read_start == 0 && stripes[0].read_end == 128*1024);
+    assert(stripes[1].read_start == 0 && stripes[1].read_end == 0);
+    assert(stripes[2].read_start == 0 && stripes[2].read_end == 128*1024);
+    assert(stripes[3].read_start == 0 && stripes[3].read_end == 128*1024);
+    read_buf = alloc_read_buffer(stripes, 4, 0);
+    assert(read_buf);
+    assert(stripes[0].read_buf == read_buf);
+    assert(stripes[1].read_buf == NULL);
+    assert(stripes[2].read_buf == read_buf+128*1024);
+    assert(stripes[3].read_buf == read_buf+2*128*1024);
+    memcpy(read_buf+128*1024, rmw_buf, 128*1024);
+    memcpy(read_buf+2*128*1024, rmw_buf+128*1024, 128*1024);
+    reconstruct_stripes_jerasure(stripes, 4, 2);
+    check_pattern(stripes[0].read_buf, 128*1024-4096, PATTERN1);
+    check_pattern(stripes[0].read_buf+128*1024-4096, 4096, PATTERN3);
+    free(read_buf);
+    // Huh done
+    free(rmw_buf);
+    free(write_buf);
+    use_jerasure(4, 2, false);
+}
+
+/***
+
+13. basic jerasure 2+1 test
+   calc_rmw(offset=128K-4K, len=8K, osd_set=[1,2,0], write_set=[1,2,3])
+   = {
+     read: [ [ 0, 128K ], [ 0, 128K ], [ 0, 0 ] ],
+     write: [ [ 128K-4K, 128K ], [ 0, 4K ], [ 0, 128K ] ],
+     input buffer: [ write0, write1 ],
+     rmw buffer: [ write2, read0, read1 ],
+   }
+   then, after calc_rmw_parity_jerasure(): all the same
+   then simulate read with read_osd_set=[0,2,3] and check read0 buffer
+
+***/
+
+void test14()
+{
+    use_jerasure(3, 2, true);
+    osd_num_t osd_set[3] = { 1, 2, 0 };
+    osd_num_t write_osd_set[3] = { 1, 2, 3 };
+    osd_rmw_stripe_t stripes[3] = { 0 };
+    // Test 13.0
+    void *write_buf = malloc_or_die(8192);
+    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_start == 0 && stripes[2].req_end == 0);
+    // Test 13.1
+    void *rmw_buf = calc_rmw(write_buf, stripes, osd_set, 3, 2, 3, write_osd_set, 128*1024);
+    assert(rmw_buf);
+    assert(stripes[0].read_start == 0 && stripes[0].read_end == 128*1024-4096);
+    assert(stripes[1].read_start == 4096 && stripes[1].read_end == 128*1024);
+    assert(stripes[2].read_start == 0 && stripes[2].read_end == 0);
+    assert(stripes[0].write_start == 128*1024-4096 && stripes[0].write_end == 128*1024);
+    assert(stripes[1].write_start == 0 && stripes[1].write_end == 4096);
+    assert(stripes[2].write_start == 0 && stripes[2].write_end == 128*1024);
+    assert(stripes[0].read_buf == rmw_buf+128*1024);
+    assert(stripes[1].read_buf == rmw_buf+2*128*1024-4096);
+    assert(stripes[2].read_buf == NULL);
+    assert(stripes[0].write_buf == write_buf);
+    assert(stripes[1].write_buf == write_buf+4096);
+    assert(stripes[2].write_buf == rmw_buf);
+    // Test 13.2 - encode
+    set_pattern(write_buf, 8192, PATTERN3);
+    set_pattern(stripes[0].read_buf, 128*1024-4096, PATTERN1);
+    set_pattern(stripes[1].read_buf, 128*1024-4096, PATTERN2);
+    calc_rmw_parity_jerasure(stripes, 3, 2, osd_set, write_osd_set, 128*1024);
+    assert(stripes[0].write_start == 128*1024-4096 && stripes[0].write_end == 128*1024);
+    assert(stripes[1].write_start == 0 && stripes[1].write_end == 4096);
+    assert(stripes[2].write_start == 0 && stripes[2].write_end == 128*1024);
+    assert(stripes[0].write_buf == write_buf);
+    assert(stripes[1].write_buf == write_buf+4096);
+    assert(stripes[2].write_buf == rmw_buf);
+    // Test 13.3 - decode and verify
+    osd_num_t read_osd_set[4] = { 0, 2, 3 };
+    memset(stripes, 0, sizeof(stripes));
+    split_stripes(2, 128*1024, 0, 128*1024, stripes);
+    assert(stripes[0].req_start == 0 && stripes[0].req_end == 128*1024);
+    assert(stripes[1].req_start == 0 && stripes[1].req_end == 0);
+    assert(stripes[2].req_start == 0 && stripes[2].req_end == 0);
+    for (int role = 0; role < 3; role++)
+    {
+        stripes[role].read_start = stripes[role].req_start;
+        stripes[role].read_end = stripes[role].req_end;
+    }
+    assert(extend_missing_stripes(stripes, read_osd_set, 2, 3) == 0);
+    assert(stripes[0].read_start == 0 && stripes[0].read_end == 128*1024);
+    assert(stripes[1].read_start == 0 && stripes[1].read_end == 128*1024);
+    assert(stripes[2].read_start == 0 && stripes[2].read_end == 128*1024);
+    void *read_buf = alloc_read_buffer(stripes, 3, 0);
+    assert(read_buf);
+    assert(stripes[0].read_buf == read_buf);
+    assert(stripes[1].read_buf == read_buf+128*1024);
+    assert(stripes[2].read_buf == read_buf+2*128*1024);
+    set_pattern(stripes[1].read_buf, 4096, PATTERN3);
+    set_pattern(stripes[1].read_buf+4096, 128*1024-4096, PATTERN2);
+    memcpy(stripes[2].read_buf, rmw_buf, 128*1024);
+    reconstruct_stripes_jerasure(stripes, 3, 2);
+    check_pattern(stripes[0].read_buf, 128*1024-4096, PATTERN1);
+    check_pattern(stripes[0].read_buf+128*1024-4096, 4096, PATTERN3);
+    free(read_buf);
+    // Huh done
+    free(rmw_buf);
+    free(write_buf);
+    use_jerasure(3, 2, false);
+}

qemu-3.1-vitastor.patch

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

qemu-4.2-vitastor.patch

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

qemu-5.0-vitastor.patch

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

qemu-5.1-vitastor.patch

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

qemu_driver.c

@@ -3,11 +3,10 @@
 
 // QEMU block driver
 
+#define BUILD_DSO
 #define _GNU_SOURCE
 #include "qemu/osdep.h"
-#include "qemu/units.h"
 #include "block/block_int.h"
-#include "block/qdict.h"
 #include "qapi/error.h"
 #include "qapi/qmp/qdict.h"
 #include "qapi/qmp/qerror.h"
@@ -15,10 +14,28 @@
 #include "qemu/error-report.h"
 #include "qemu/module.h"
 #include "qemu/option.h"
+
+#if QEMU_VERSION_MAJOR >= 3
+#include "qemu/units.h"
+#include "block/qdict.h"
 #include "qemu/cutils.h"
+#else
+#include "qapi/qmp/qint.h"
+#define qdict_put_int(options, name, num_val) qdict_put_obj(options, name, QOBJECT(qint_from_int(num_val)))
+#define qdict_put_str(options, name, value) qdict_put_obj(options, name, QOBJECT(qstring_from_str(value)))
+#define qobject_unref QDECREF
+#endif
 
 #include "qemu_proxy.h"
 
+void qemu_module_dummy(void)
+{
+}
+
+void DSO_STAMP_FUN(void)
+{
+}
+
 typedef struct VitastorClient
 {
     void *proxy;
@@ -176,12 +193,14 @@ static void vitastor_close(BlockDriverState *bs)
         g_free(client->etcd_prefix);
 }
 
+#if QEMU_VERSION_MAJOR >= 3
 static int vitastor_probe_blocksizes(BlockDriverState *bs, BlockSizes *bsz)
 {
     bsz->phys = 4096;
     bsz->log = 4096;
     return 0;
 }
+#endif
 
 static int coroutine_fn vitastor_co_create_opts(
 #if QEMU_VERSION_MAJOR >= 4
@@ -208,6 +227,7 @@ out:
     return ret;
 }
 
+#if QEMU_VERSION_MAJOR >= 3
 static int coroutine_fn vitastor_co_truncate(BlockDriverState *bs, int64_t offset,
 #if QEMU_VERSION_MAJOR >= 4
     bool exact,
@@ -231,6 +251,7 @@ static int coroutine_fn vitastor_co_truncate(BlockDriverState *bs, int64_t offse
 
     return 0;
 }
+#endif
 
 static int vitastor_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
 {
@@ -244,11 +265,22 @@ static int64_t vitastor_getlength(BlockDriverState *bs)
     return client->size;
 }
 
+#if QEMU_VERSION_MAJOR >= 3
 static void vitastor_refresh_limits(BlockDriverState *bs, Error **errp)
+#else
+static int vitastor_refresh_limits(BlockDriverState *bs)
+#endif
 {
+#if QEMU_VERSION_MAJOR >= 4
     bs->bl.request_alignment = 4096;
     bs->bl.min_mem_alignment = 4096;
+#else
+    bs->request_alignment = 4096;
+#endif
     bs->bl.opt_mem_alignment = 4096;
+#if QEMU_VERSION_MAJOR < 3
+    return 0;
+#endif
 }
 
 static int64_t vitastor_get_allocated_file_size(BlockDriverState *bs)
@@ -271,7 +303,12 @@ static void vitastor_co_generic_bh_cb(int retval, void *opaque)
     task->complete = 1;
     if (qemu_coroutine_self() != task->co)
     {
+#if QEMU_VERSION_MAJOR >= 3
         aio_co_wake(task->co);
+#else
+        qemu_coroutine_enter(task->co, NULL);
+        qemu_aio_release(task);
+#endif
     }
 }
 
@@ -313,6 +350,18 @@ static int coroutine_fn vitastor_co_pwritev(BlockDriverState *bs, uint64_t offse
     return task.ret;
 }
 
+#if QEMU_VERSION_MAJOR < 3
+static int coroutine_fn vitastor_co_readv(BlockDriverState *bs, int64_t sector_num, int nb_sectors, QEMUIOVector *iov)
+{
+    return vitastor_co_preadv(bs, sector_num*BDRV_SECTOR_SIZE, nb_sectors*BDRV_SECTOR_SIZE, iov, 0);
+}
+
+static int coroutine_fn vitastor_co_writev(BlockDriverState *bs, int64_t sector_num, int nb_sectors, QEMUIOVector *iov)
+{
+    return vitastor_co_pwritev(bs, sector_num*BDRV_SECTOR_SIZE, nb_sectors*BDRV_SECTOR_SIZE, iov, 0);
+}
+#endif
+
 static int coroutine_fn vitastor_co_flush(BlockDriverState *bs)
 {
     VitastorClient *client = bs->opaque;
@@ -331,6 +380,7 @@ static int coroutine_fn vitastor_co_flush(BlockDriverState *bs)
     return task.ret;
 }
 
+#if QEMU_VERSION_MAJOR >= 3
 static QemuOptsList vitastor_create_opts = {
     .name = "vitastor-create-opts",
     .head = QTAILQ_HEAD_INITIALIZER(vitastor_create_opts.head),
@@ -343,6 +393,16 @@ static QemuOptsList vitastor_create_opts = {
         { /* end of list */ }
     }
 };
+#else
+static QEMUOptionParameter vitastor_create_opts[] = {
+    {
+        .name = BLOCK_OPT_SIZE,
+        .type = OPT_SIZE,
+        .help = "Virtual disk size"
+    },
+    { NULL }
+};
+#endif
 
 static const char *vitastor_strong_runtime_opts[] = {
     "inode",
@@ -363,7 +423,9 @@ static BlockDriver bdrv_vitastor = {
     .bdrv_has_zero_init             = bdrv_has_zero_init_1,
     .bdrv_get_info                  = vitastor_get_info,
     .bdrv_getlength                 = vitastor_getlength,
+#if QEMU_VERSION_MAJOR >= 3
     .bdrv_probe_blocksizes          = vitastor_probe_blocksizes,
+#endif
     .bdrv_refresh_limits            = vitastor_refresh_limits,
 
     // FIXME: Implement it along with per-inode statistics
@@ -373,12 +435,17 @@ static BlockDriver bdrv_vitastor = {
     .bdrv_close                     = vitastor_close,
 
     // Option list for the create operation
+#if QEMU_VERSION_MAJOR >= 3
     .create_opts                    = &vitastor_create_opts,
+#else
+    .create_options                 = vitastor_create_opts,
+#endif
 
     // For qmp_blockdev_create(), used by the qemu monitor / QAPI
     // Requires patching QAPI IDL, thus unimplemented
     //.bdrv_co_create                 = vitastor_co_create,
 
+#if QEMU_VERSION_MAJOR >= 3
     // For bdrv_create(), used by qemu-img
     .bdrv_co_create_opts            = vitastor_co_create_opts,
 
@@ -386,6 +453,11 @@ static BlockDriver bdrv_vitastor = {
 
     .bdrv_co_preadv                 = vitastor_co_preadv,
     .bdrv_co_pwritev                = vitastor_co_pwritev,
+#else
+    .bdrv_co_readv                  = vitastor_co_readv,
+    .bdrv_co_writev                 = vitastor_co_writev,
+#endif
+
     .bdrv_co_flush_to_disk          = vitastor_co_flush,
 
 #if QEMU_VERSION_MAJOR >= 4

ringloop.cpp

@@ -77,7 +77,10 @@ void ring_loop_t::loop()
             dl.callback(&dl);
         }
         else
+        {
+            printf("Warning: empty callback in SQE\n");
             free_ring_data[free_ring_data_ptr++] = d - ring_datas;
+        }
         io_uring_cqe_seen(&ring, cqe);
     }
     while (get_sqe_queue.size() > 0)

ringloop.h

@@ -142,7 +142,10 @@ 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)

rm_inode.cpp

@@ -6,26 +6,38 @@
  * May be included into a bigger "command-line management interface" in the future
  */
 
+#include <vector>
 #include <algorithm>
 
 #include "epoll_manager.h"
 #include "cluster_client.h"
 #include "pg_states.h"
 
-#define RM_NO_LIST 1
-#define RM_LIST_SENT 2
-#define RM_REMOVING 3
-#define RM_END 4
+#define RM_LISTING 1
+#define RM_REMOVING 2
+#define RM_END 3
 
 const char *exe_name = NULL;
 
+struct rm_pg_t;
+
 struct rm_pg_osd_t
 {
-    pg_num_t pg_num;
+    rm_pg_t *pg = NULL;
     osd_num_t osd_num;
+    bool sent = false;
+};
+
+struct rm_pg_t
+{
+    pg_num_t pg_num;
+    osd_num_t rm_osd_num;
+    std::vector<rm_pg_osd_t> list_osds;
     int state = 0;
-    obj_ver_id *obj_list = NULL;
-    uint64_t obj_count = 0, obj_pos = 0, obj_done = 0, obj_prev_done = 0;
+    int to_list;
+    std::set<object_id> objects;
+    std::set<object_id>::iterator obj_pos;
+    uint64_t obj_count = 0, obj_done = 0, obj_prev_done = 0;
     int in_flight = 0;
 };
 
@@ -41,11 +53,12 @@ protected:
     cluster_client_t *cli = NULL;
     ring_consumer_t consumer;
 
-    std::vector<rm_pg_osd_t*> lists;
+    std::vector<rm_pg_t*> lists;
     uint64_t total_count = 0, total_done = 0, total_prev_pct = 0;
     uint64_t pgs_to_list = 0;
     bool started = false;
     bool progress = true;
+    bool list_first = false;
     int log_level = 0;
 
 public:
@@ -62,7 +75,7 @@ public:
             else if (args[i][0] == '-' && args[i][1] == '-')
             {
                 const char *opt = args[i]+2;
-                cfg[opt] = !strcmp(opt, "json") || i == narg-1 ? "1" : args[++i];
+                cfg[opt] = !strcmp(opt, "json") || !strcmp(opt, "wait-list") || i == narg-1 ? "1" : args[++i];
             }
         }
         return cfg;
@@ -74,7 +87,7 @@ public:
             "Vitastor inode removal tool\n"
             "(c) Vitaliy Filippov, 2020 (VNPL-1.0)\n\n"
             "USAGE:\n"
-            "  %s --etcd_address <etcd_address> --pool <pool> --inode <inode>\n",
+            "  %s --etcd_address <etcd_address> --pool <pool> --inode <inode> [--wait-list]\n",
             exe_name
         );
         exit(0);
@@ -105,6 +118,7 @@ public:
             parallel_osds = 4;
         log_level = cfg["log_level"].int64_value();
         progress = cfg["progress"].uint64_value() ? true : false;
+        list_first = cfg["wait-list"].uint64_value() ? true : false;
         // Create client
         ringloop = new ring_loop_t(512);
         epmgr = new epoll_manager_t(ringloop);
@@ -137,21 +151,57 @@ public:
         for (auto & pg_item: pool_cfg.pg_config)
         {
             auto & pg = pg_item.second;
-            if (pg.pause || !pg.cur_primary || pg.cur_state != PG_ACTIVE)
+            if (pg.pause || !pg.cur_primary || !(pg.cur_state & PG_ACTIVE))
             {
-                // FIXME Support deletion in non-clean active PGs by introducing a "primary-list" command
-                fprintf(stderr, "PG %u is not active+clean, skipping\n", pg_item.first);
+                fprintf(stderr, "PG %u is inactive, skipping\n", pg_item.first);
                 continue;
             }
-            rm_pg_osd_t *r = new rm_pg_osd_t();
+            rm_pg_t *r = new rm_pg_t();
             r->pg_num = pg_item.first;
-            r->osd_num = pg.cur_primary;
-            r->state = RM_NO_LIST;
+            r->rm_osd_num = pg.cur_primary;
+            r->state = RM_LISTING;
+            if (pg.cur_state != PG_ACTIVE)
+            {
+                std::set<osd_num_t> all_peers;
+                for (osd_num_t pg_osd: pg.target_set)
+                {
+                    if (pg_osd != 0)
+                    {
+                        all_peers.insert(pg_osd);
+                    }
+                }
+                for (osd_num_t pg_osd: pg.all_peers)
+                {
+                    if (pg_osd != 0)
+                    {
+                        all_peers.insert(pg_osd);
+                    }
+                }
+                for (auto & hist_item: pg.target_history)
+                {
+                    for (auto pg_osd: hist_item)
+                    {
+                        if (pg_osd != 0)
+                        {
+                            all_peers.insert(pg_osd);
+                        }
+                    }
+                }
+                for (osd_num_t peer_osd: all_peers)
+                {
+                    r->list_osds.push_back((rm_pg_osd_t){ .pg = r, .osd_num = peer_osd, .sent = false });
+                }
+            }
+            else
+            {
+                r->list_osds.push_back((rm_pg_osd_t){ .pg = r, .osd_num = pg.cur_primary, .sent = false });
+            }
+            r->to_list = r->list_osds.size();
             lists.push_back(r);
         }
-        std::sort(lists.begin(), lists.end(), [](rm_pg_osd_t *a, rm_pg_osd_t *b)
+        std::sort(lists.begin(), lists.end(), [](rm_pg_t *a, rm_pg_t *b)
         {
-            return a->osd_num < b->osd_num ? true : false;
+            return a->rm_osd_num < b->rm_osd_num ? true : false;
         });
         pgs_to_list = lists.size();
         started = true;
@@ -160,6 +210,10 @@ public:
 
     void send_list(rm_pg_osd_t *cur_list)
     {
+        if (cur_list->sent)
+        {
+            return;
+        }
         if (cli->msgr.osd_peer_fds.find(cur_list->osd_num) ==
             cli->msgr.osd_peer_fds.end())
         {
@@ -170,14 +224,14 @@ public:
         osd_op_t *op = new osd_op_t();
         op->op_type = OSD_OP_OUT;
         op->peer_fd = cli->msgr.osd_peer_fds[cur_list->osd_num];
-        op->req = {
+        op->req = (osd_any_op_t){
             .sec_list = {
                 .header = {
                     .magic = SECONDARY_OSD_OP_MAGIC,
                     .id = cli->msgr.next_subop_id++,
                     .opcode = OSD_OP_SEC_LIST,
                 },
-                .list_pg = cur_list->pg_num,
+                .list_pg = cur_list->pg->pg_num,
                 .pg_count = (pg_num_t)cli->st_cli.pool_config[pool_id].real_pg_count,
                 .pg_stripe_size = cli->st_cli.pool_config[pool_id].pg_stripe_size,
                 .min_inode = inode,
@@ -186,62 +240,76 @@ public:
         };
         op->callback = [this, cur_list](osd_op_t *op)
         {
-            pgs_to_list--;
+            cur_list->pg->to_list--;
             if (op->reply.hdr.retval < 0)
             {
-                fprintf(stderr, "Failed to get object list from OSD %lu (retval=%ld), skipping the PG\n",
-                    cur_list->osd_num, op->reply.hdr.retval);
-                cli->msgr.stop_client(cur_list->osd_num);
-                delete op;
-                cur_list->state = RM_END;
-                continue_delete();
-                return;
+                fprintf(stderr, "Failed to get PG %u/%u object list from OSD %lu (retval=%ld), skipping\n",
+                    pool_id, cur_list->pg->pg_num, cur_list->osd_num, op->reply.hdr.retval);
             }
-            if (log_level > 0)
+            else
             {
-                printf(
-                    "[PG %u/%u] Got inode object list from OSD %lu: %ld object versions\n",
-                    pool_id, cur_list->pg_num, cur_list->osd_num, op->reply.hdr.retval
-                );
+                if (op->reply.sec_list.stable_count < op->reply.hdr.retval)
+                {
+                    // Unstable objects, if present, mean that someone still writes into the inode. Warn the user about it.
+                    printf(
+                        "[PG %u/%u] Inode still has %lu unstable object versions - is it still open? Not a good idea to delete it.\n",
+                        pool_id, cur_list->pg->pg_num, op->reply.hdr.retval - op->reply.sec_list.stable_count
+                    );
+                }
+                if (log_level > 0)
+                {
+                    printf(
+                        "[PG %u/%u] Got inode object list from OSD %lu: %ld object versions\n",
+                        pool_id, cur_list->pg->pg_num, cur_list->osd_num, op->reply.hdr.retval
+                    );
+                }
+                for (uint64_t i = 0; i < op->reply.hdr.retval; i++)
+                {
+                    object_id oid = ((obj_ver_id*)op->buf)[i].oid;
+                    oid.stripe = oid.stripe & ~STRIPE_MASK;
+                    cur_list->pg->objects.insert(oid);
+                }
             }
-            cur_list->obj_list = (obj_ver_id*)op->buf;
-            cur_list->obj_count = (uint64_t)op->reply.hdr.retval;
-            cur_list->obj_done = cur_list->obj_prev_done = cur_list->obj_pos = 0;
-            total_count += cur_list->obj_count;
-            total_prev_pct = 0;
-            // set op->buf to NULL so it doesn't get freed
-            op->buf = NULL;
             delete op;
-            cur_list->state = RM_REMOVING;
+            if (cur_list->pg->to_list <= 0)
+            {
+                cur_list->pg->obj_done = cur_list->pg->obj_prev_done = 0;
+                cur_list->pg->obj_pos = cur_list->pg->objects.begin();
+                cur_list->pg->obj_count = cur_list->pg->objects.size();
+                total_count += cur_list->pg->obj_count;
+                total_prev_pct = 0;
+                cur_list->pg->state = RM_REMOVING;
+                pgs_to_list--;
+            }
             continue_delete();
         };
-        cur_list->state = RM_LIST_SENT;
         cli->msgr.outbox_push(op);
+        cur_list->sent = true;
     }
 
-    void send_ops(rm_pg_osd_t *cur_list)
+    void send_ops(rm_pg_t *cur_list)
     {
-        if (cli->msgr.osd_peer_fds.find(cur_list->osd_num) ==
+        if (cli->msgr.osd_peer_fds.find(cur_list->rm_osd_num) ==
             cli->msgr.osd_peer_fds.end())
         {
             // Initiate connection
-            cli->msgr.connect_peer(cur_list->osd_num, cli->st_cli.peer_states[cur_list->osd_num]);
+            cli->msgr.connect_peer(cur_list->rm_osd_num, cli->st_cli.peer_states[cur_list->rm_osd_num]);
             return;
         }
-        while (cur_list->in_flight < iodepth && cur_list->obj_pos < cur_list->obj_count)
+        while (cur_list->in_flight < iodepth && cur_list->obj_pos != cur_list->objects.end())
         {
             osd_op_t *op = new osd_op_t();
             op->op_type = OSD_OP_OUT;
-            op->peer_fd = cli->msgr.osd_peer_fds[cur_list->osd_num];
-            op->req = {
+            op->peer_fd = cli->msgr.osd_peer_fds[cur_list->rm_osd_num];
+            op->req = (osd_any_op_t){
                 .rw = {
                     .header = {
                         .magic = SECONDARY_OSD_OP_MAGIC,
                         .id = cli->msgr.next_subop_id++,
                         .opcode = OSD_OP_DELETE,
                     },
-                    .inode = cur_list->obj_list[cur_list->obj_pos].oid.inode,
-                    .offset = (cur_list->obj_list[cur_list->obj_pos].oid.stripe & ~STRIPE_MASK),
+                    .inode = cur_list->obj_pos->inode,
+                    .offset = (cur_list->obj_pos->stripe & ~STRIPE_MASK),
                     .len = 0,
                 },
             };
@@ -251,7 +319,7 @@ public:
                 if (op->reply.hdr.retval < 0)
                 {
                     fprintf(stderr, "Failed to remove object from PG %u (OSD %lu) (retval=%ld)\n",
-                        cur_list->pg_num, cur_list->osd_num, op->reply.hdr.retval);
+                        cur_list->pg_num, cur_list->rm_osd_num, op->reply.hdr.retval);
                 }
                 delete op;
                 cur_list->obj_done++;
@@ -262,12 +330,10 @@ public:
             cur_list->obj_pos++;
             cur_list->in_flight++;
         }
-        if (!cur_list->in_flight && cur_list->obj_pos >= cur_list->obj_count)
+        if (!cur_list->in_flight && cur_list->obj_pos == cur_list->objects.end())
         {
-            free(cur_list->obj_list);
-            cur_list->obj_list = NULL;
             cur_list->obj_count = 0;
-            cur_list->obj_done = cur_list->obj_prev_done = cur_list->obj_pos = 0;
+            cur_list->obj_done = cur_list->obj_prev_done = 0;
             cur_list->state = RM_END;
         }
     }
@@ -276,6 +342,22 @@ public:
     {
         int par_osd = 0;
         osd_num_t max_seen_osd = 0;
+        bool no_del = false;
+        if (list_first)
+        {
+            int i, n = 0;
+            for (i = 0; i < lists.size(); i++)
+            {
+                if (lists[i]->state == RM_LISTING)
+                {
+                    n++;
+                }
+            }
+            if (n > 0)
+            {
+                no_del = true;
+            }
+        }
         for (int i = 0; i < lists.size(); i++)
         {
             if (lists[i]->state == RM_END)
@@ -284,18 +366,25 @@ public:
                 lists.erase(lists.begin()+i, lists.begin()+i+1);
                 i--;
             }
-            else if (lists[i]->osd_num > max_seen_osd)
+            else if (lists[i]->rm_osd_num > max_seen_osd)
             {
-                if (lists[i]->state == RM_NO_LIST)
+                if (lists[i]->state == RM_LISTING)
                 {
-                    send_list(lists[i]);
+                    for (int j = 0; j < lists[i]->list_osds.size(); j++)
+                    {
+                        send_list(&lists[i]->list_osds[j]);
+                    }
                 }
                 else if (lists[i]->state == RM_REMOVING)
                 {
+                    if (no_del)
+                    {
+                        continue;
+                    }
                     send_ops(lists[i]);
                 }
                 par_osd++;
-                max_seen_osd = lists[i]->osd_num;
+                max_seen_osd = lists[i]->rm_osd_num;
                 if (par_osd >= parallel_osds)
                 {
                     break;

rpm/build-tarball.sh

@@ -0,0 +1,51 @@
+#!/bin/bash
+# Vitastor depends on QEMU and FIO headers, but QEMU and FIO don't have -devel packages
+# So we have to copy their headers into the source tarball
+
+set -e
+
+VITASTOR=$(dirname $0)
+VITASTOR=$(realpath "$VITASTOR/..")
+
+if [ -d /opt/rh/gcc-toolset-9 ]; then
+    # CentOS 8
+    EL=8
+    . /opt/rh/gcc-toolset-9/enable
+else
+    # CentOS 7
+    EL=7
+    . /opt/rh/devtoolset-9/enable
+fi
+cd ~/rpmbuild/SPECS
+rpmbuild -bp fio.spec
+perl -i -pe 's/^make V=1/exit 0; make V=1/' qemu*.spec
+rpmbuild -bc qemu*.spec
+perl -i -pe 's/^exit 0; make V=1/make V=1/' qemu*.spec
+cd ~/rpmbuild/BUILD/qemu*/
+rm -rf $VITASTOR/qemu $VITASTOR/fio
+mkdir -p $VITASTOR/qemu/b/qemu
+make -j8 config-host.h
+cp config-host.h $VITASTOR/qemu/b/qemu
+cp -r include $VITASTOR/qemu
+if [ -f qapi-schema.json ]; then
+    # QEMU 2.0
+    make qapi-types.h
+    cp qapi-types.h $VITASTOR/qemu/b/qemu
+else
+    # QEMU 3.0+
+    make qapi
+    cp -r qapi $VITASTOR/qemu/b/qemu
+fi
+cd $VITASTOR
+sh copy-qemu-includes.sh
+rm -rf qemu
+mv qemu-copy qemu
+ln -s ~/rpmbuild/BUILD/fio*/ fio
+sh copy-fio-includes.sh
+rm fio
+mv fio-copy fio
+FIO=`rpm -qi fio | perl -e 'while(<>) { /^Epoch[\s:]+(\S+)/ && print "$1:"; /^Version[\s:]+(\S+)/ && print $1; /^Release[\s:]+(\S+)/ && print "-$1"; }'`
+QEMU=`rpm -qi qemu qemu-kvm | perl -e 'while(<>) { /^Epoch[\s:]+(\S+)/ && print "$1:"; /^Version[\s:]+(\S+)/ && print $1; /^Release[\s:]+(\S+)/ && print "-$1"; }'`
+perl -i -pe 's/(Requires:\s*fio)([^\n]+)?/$1 = '$FIO'/' $VITASTOR/rpm/vitastor-el$EL.spec
+perl -i -pe 's/(Requires:\s*qemu(?:-kvm)?)([^\n]+)?/$1 = '$QEMU'/' $VITASTOR/rpm/vitastor-el$EL.spec
+tar --transform 's#^#vitastor-0.5.4/#' --exclude 'rpm/*.rpm' -czf $VITASTOR/../vitastor-0.5.4$(rpm --eval '%dist').tar.gz *

rpm/qemu-el8.Dockerfile

@@ -0,0 +1,31 @@
+# Build packages for CentOS 8 inside a container
+# cd ..; podman build -t qemu-el8 -v `pwd`/build:/root/build -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/build/qemu-el8; \
+    rm -rf /root/build/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/build/qemu-el8/; \
+    cp ~/rpmbuild/SRPMS/*qemu* /root/build/qemu-el8/

rpm/qemu-kvm-el7.spec.patch

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

rpm/qemu-kvm.spec.patch

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

rpm/vitastor-el7.Dockerfile

@@ -0,0 +1,47 @@
+# Build packages for CentOS 7 inside a container
+# cd ..; podman build -t vitastor-el7 -v `pwd`/build:/root/build -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/build/liburing-el7; \
+    rm -rf /root/build/liburing-el7/*; \
+    cp ~/rpmbuild/RPMS/*/liburing* /root/build/liburing-el7/; \
+    cp ~/rpmbuild/SRPMS/liburing* /root/build/liburing-el7/
+
+RUN rpm -i `ls /root/build/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-0.5.4.el7.tar.gz ~/rpmbuild/SOURCES; \
+    cp vitastor-el7.spec ~/rpmbuild/SPECS/vitastor.spec; \
+    cd ~/rpmbuild/SPECS/; \
+    rpmbuild -ba vitastor.spec; \
+    mkdir -p /root/build/vitastor-el7; \
+    rm -rf /root/build/vitastor-el7/*; \
+    cp ~/rpmbuild/RPMS/*/vitastor* /root/build/vitastor-el7/; \
+    cp ~/rpmbuild/SRPMS/vitastor* /root/build/vitastor-el7/

rpm/vitastor-el7.spec

@@ -0,0 +1,63 @@
+Name:           vitastor
+Version:        0.5.4
+Release:        2%{?dist}
+Summary:        Vitastor, a fast software-defined clustered block storage
+
+License:        Vitastor Network Public License 1.0
+URL:            https://vitastor.io/
+Source0:        vitastor-0.5.4.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
+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
+make %{?_smp_mflags} BINDIR=%_bindir LIBDIR=%_libdir QEMU_PLUGINDIR=%_libdir/qemu-kvm
+
+
+%install
+rm -rf $RPM_BUILD_ROOT
+%make_install BINDIR=%_bindir LIBDIR=%_libdir QEMU_PLUGINDIR=%_libdir/qemu-kvm
+. /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/vitastor
+/usr/lib/vitastor
+
+
+%changelog

rpm/vitastor-el8.Dockerfile

@@ -0,0 +1,45 @@
+# Build packages for CentOS 8 inside a container
+# cd ..; podman build -t vitastor-el8 -v `pwd`/build:/root/build -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
+
+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/build/liburing-el8; \
+    rm -rf /root/build/liburing-el8/*; \
+    cp ~/rpmbuild/RPMS/*/liburing* /root/build/liburing-el8/; \
+    cp ~/rpmbuild/SRPMS/liburing* /root/build/liburing-el8/
+
+RUN rpm -i `ls /root/build/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-0.5.4.el8.tar.gz ~/rpmbuild/SOURCES; \
+    cp vitastor-el8.spec ~/rpmbuild/SPECS/vitastor.spec; \
+    cd ~/rpmbuild/SPECS/; \
+    rpmbuild -ba vitastor.spec; \
+    mkdir -p /root/build/vitastor-el8; \
+    rm -rf /root/build/vitastor-el8/*; \
+    cp ~/rpmbuild/RPMS/*/vitastor* /root/build/vitastor-el8/; \
+    cp ~/rpmbuild/SRPMS/vitastor* /root/build/vitastor-el8/

rpm/vitastor-el8.spec

@@ -0,0 +1,60 @@
+Name:           vitastor
+Version:        0.5.4
+Release:        2%{?dist}
+Summary:        Vitastor, a fast software-defined clustered block storage
+
+License:        Vitastor Network Public License 1.0
+URL:            https://vitastor.io/
+Source0:        vitastor-0.5.4.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
+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
+make %{?_smp_mflags} BINDIR=%_bindir LIBDIR=%_libdir QEMU_PLUGINDIR=%_libdir/qemu-kvm
+
+
+%install
+rm -rf $RPM_BUILD_ROOT
+%make_install BINDIR=%_bindir LIBDIR=%_libdir QEMU_PLUGINDIR=%_libdir/qemu-kvm
+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/vitastor
+/usr/lib/vitastor
+
+
+%changelog

run_tests.sh

@@ -0,0 +1,77 @@
+#!/bin/bash
+
+if [ ! "$BASH_VERSION" ] ; then
+    echo "Use bash to run this script ($0)" 1>&2
+    exit 1
+fi
+
+format_error()
+{
+    echo $(echo -n -e "\033[1;31m")$1$(echo -n -e "\033[m")
+    $ETCDCTL get --prefix /vitastor > ./testdata/etcd-dump.txt
+    exit 1
+}
+format_green()
+{
+    echo $(echo -n -e "\033[1;32m")$1$(echo -n -e "\033[m")
+}
+
+set -e -x
+
+trap 'kill -9 $(jobs -p)' EXIT
+
+ETCD=${ETCD:-etcd}
+ETCD_PORT=${ETCD_PORT:-12379}
+
+rm -rf ./testdata
+mkdir -p ./testdata
+dd if=/dev/zero of=./testdata/test_osd1.bin bs=1024 count=1 seek=$((1024*1024-1))
+dd if=/dev/zero of=./testdata/test_osd2.bin bs=1024 count=1 seek=$((1024*1024-1))
+dd if=/dev/zero of=./testdata/test_osd3.bin bs=1024 count=1 seek=$((1024*1024-1))
+
+$ETCD -name etcd_test --data-dir ./testdata/etcd \
+    --advertise-client-urls http://127.0.0.1:$ETCD_PORT --listen-client-urls http://127.0.0.1:$ETCD_PORT \
+    --initial-advertise-peer-urls http://127.0.0.1:$((ETCD_PORT+1)) --listen-peer-urls http://127.0.0.1:$((ETCD_PORT+1)) \
+    --max-txn-ops=100000 --auto-compaction-retention=10 --auto-compaction-mode=revision &>./testdata/etcd.log &
+ETCD_PID=$!
+ETCD_URL=127.0.0.1:$ETCD_PORT/v3
+ETCDCTL="${ETCD}ctl --endpoints=http://$ETCD_URL"
+
+./osd --osd_num 1 --bind_address 127.0.0.1 --etcd_address $ETCD_URL $(node mon/simple-offsets.js --format options --device ./testdata/test_osd1.bin 2>/dev/null) &>./testdata/osd1.log &
+OSD1_PID=$!
+./osd --osd_num 2 --bind_address 127.0.0.1 --etcd_address $ETCD_URL $(node mon/simple-offsets.js --format options --device ./testdata/test_osd2.bin 2>/dev/null) &>./testdata/osd2.log &
+OSD2_PID=$!
+./osd --osd_num 3 --bind_address 127.0.0.1 --etcd_address $ETCD_URL $(node mon/simple-offsets.js --format options --device ./testdata/test_osd3.bin 2>/dev/null) &>./testdata/osd3.log &
+OSD3_PID=$!
+
+cd mon
+npm install
+cd ..
+node mon/mon-main.js --etcd_url http://$ETCD_URL --etcd_prefix "/vitastor" &>./testdata/mon.log &
+MON_PID=$!
+
+$ETCDCTL put /vitastor/config/pools '{"1":{"name":"testpool","scheme":"xor","pg_size":3,"pg_minsize":2,"parity_chunks":1,"pg_count":1,"failure_domain":"osd"}}'
+
+sleep 2
+
+if ! ($ETCDCTL get /vitastor/config/pgs --print-value-only | jq -s -e '(. | length) != 0 and (.[0].items["1"]["1"].osd_set | sort) == ["1","2","3"]'); then
+    format_error "FAILED: 1 PG NOT CONFIGURED"
+fi
+
+if ! ($ETCDCTL get /vitastor/pg/state/1/1 --print-value-only | jq -s -e '(. | length) != 0 and .[0].state == ["active"]'); then
+    format_error "FAILED: 1 PG NOT UP"
+fi
+
+echo leak:fio >> testdata/lsan-suppress.txt
+echo leak:tcmalloc >> testdata/lsan-suppress.txt
+echo leak:ceph >> testdata/lsan-suppress.txt
+echo leak:librbd >> testdata/lsan-suppress.txt
+echo leak:_M_mutate >> testdata/lsan-suppress.txt
+echo leak:_M_assign >> testdata/lsan-suppress.txt
+#LSAN_OPTIONS=suppressions=`pwd`/testdata/lsan-suppress.txt LD_PRELOAD=libasan.so.5 \
+#    fio -thread -name=test -ioengine=./libfio_sec_osd.so -bs=4k -fsync=128 `$ETCDCTL get /vitastor/osd/state/1 --print-value-only | jq -r '"-host="+.addresses[0]+" -port="+(.port|tostring)'` -rw=write -size=32M
+
+LSAN_OPTIONS=suppressions=`pwd`/testdata/lsan-suppress.txt LD_PRELOAD=libasan.so.5 \
+    fio -thread -name=test -ioengine=./libfio_cluster.so -bs=4M -direct=1 -iodepth=1 -fsync=1 -rw=write -etcd=$ETCD_URL -pool=1 -inode=1 -size=1G -cluster_log_level=10
+
+format_green OK

test-build-el7.sh

@@ -0,0 +1,34 @@
+#!/bin/bash
+
+# Cheatsheet for CentOS 7 packaging (not a build script)
+
+set -e
+rm -f /etc/yum.repos.d/CentOS-Media.repo
+yum -y --enablerepo=extras install centos-release-scl epel-release
+yum -y --enablerepo='*' install devtoolset-9-gcc-c++ devtoolset-9-libatomic-devel gperftools-devel
+yumdownloader --source qemu
+yumdownloader --source fio
+yum-builddep -y --enablerepo='*' qemu
+yum -y install rpm-build
+. /opt/rh/devtoolset-9/enable
+rpm --nomd5 -i qemu*.src.rpm
+rpm --nomd5 -i fio*.src.rpm
+cd ~/rpmbuild/SPECS
+rpmbuild -bp fio.spec
+perl -i -pe 's/^make V=1/exit 1; make V=1/' qemu.spec
+rpmbuild -bc qemu.spec
+perl -i -pe 's/^exit 1; make V=1/make V=1/' qemu.spec
+cd ~/rpmbuild/BUILD/qemu*/
+make qapi-types.h
+mkdir -p ~/vitastor/qemu/b/qemu
+cp config-host.h ~/vitastor/qemu/b/qemu 
+cp qapi-types.h ~/vitastor/qemu/b/qemu
+cp -r include ~/vitastor/qemu
+cd ~/vitastor
+sh copy-qemu-includes.sh
+mv qemu qemu-old
+mv qemu-copy qemu
+ln -s ~/rpmbuild/BUILD/fio*/ fio
+sh copy-fio-includes.sh
+rm fio
+mv fio-copy fio

test_pattern.h

@@ -12,4 +12,4 @@
 #define PATTERN3 0x426bd7854eb08509
 
 #define set_pattern(buf, len, pattern) for (uint64_t i = 0; i < len; i += 8) { *(uint64_t*)((void*)buf + i) = pattern; }
-#define check_pattern(buf, len, pattern) for (uint64_t i = 0; i < len; i += 8) { assert(*(uint64_t*)(buf + i) == pattern); }
+#define check_pattern(buf, len, pattern) { uint64_t bad = UINT64_MAX; for (uint64_t i = 0; i < len; i += 8) { if ((*(uint64_t*)(buf + i)) != (pattern)) { bad = i; break; } } if (bad != UINT64_MAX) { printf("mismatch at %lx\n", bad); } assert(bad == UINT64_MAX); }

test_shit.cpp

@@ -30,7 +30,7 @@
 
 #include "blockstore.h"
 #include "blockstore_impl.h"
-#include "osd_peering_pg.h"
+#include "osd_peering_pg.cpp"
 //#include "cpp-btree/btree_map.h"
 
 static int setup_context(unsigned entries, struct io_uring *ring)
@@ -168,7 +168,7 @@ int main0(int argc, char *argv[])
             },
             .version = 1,
         }] = (dirty_entry){
-            .state = ST_D_SYNCED,
+            .state = BS_ST_SYNCED | BS_ST_BIG_WRITE,
             .flags = 0,
             .location = (uint64_t)i << 17,
             .offset = 0,