WIP Switch back to TCP when RDMA send fails (does not work)

Seems slightly overcomplicated...

.gitea/workflows/buildenv.Dockerfile

@@ -23,7 +23,7 @@ RUN apt-get update
 RUN apt-get -y install etcd qemu-system-x86 qemu-block-extra qemu-utils fio libasan8 \
     libgoogle-perftools-dev devscripts libjerasure-dev cmake libibverbs-dev libisal-dev
 RUN apt-get -y build-dep fio qemu=`dpkg -s qemu-system-x86|grep ^Version:|awk '{print $2}'`
-RUN apt-get update && apt-get -y install jq lp-solve sudo nfs-common fdisk parted libc-ares-dev udev iproute2
+RUN apt-get update && apt-get -y install jq lp-solve sudo nfs-common fdisk parted libc-ares-dev udev
 RUN apt-get --download-only source fio qemu=`dpkg -s qemu-system-x86|grep ^Version:|awk '{print $2}'`
 
 RUN set -ex; \

.gitea/workflows/test.yml

@@ -62,13 +62,8 @@ jobs:
     needs: build
     container: ${{env.TEST_IMAGE}}:${{github.sha}}
     steps:
-    - name: Run tests
-      id: test
-      run: cd /root/vitastor/build && make -j16 build_tests test
-    - name: Print logs
-      if: always() && steps.test.outcome == 'failure'
-      run: |
-        cat /root/vitastor/build/Testing/Temporary/LastTest.log
+    # leak sanitizer sometimes crashes
+    - run: cd /root/vitastor/build && ASAN_OPTIONS=detect_leaks=0 make -j16 build_tests test
 
   npm_lint:
     runs-on: ubuntu-latest
@@ -167,24 +162,6 @@ jobs:
           echo ""
         done
 
-  test_change_pg_minsize:
-    runs-on: ubuntu-latest
-    needs: build
-    container: ${{env.TEST_IMAGE}}:${{github.sha}}
-    steps:
-    - name: Run test
-      id: test
-      timeout-minutes: 3
-      run: /root/vitastor/tests/test_change_pg_minsize.sh
-    - name: Print logs
-      if: always() && steps.test.outcome == 'failure'
-      run: |
-        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
-          echo "-------- $i --------"
-          cat $i
-          echo ""
-        done
-
   test_change_pg_size:
     runs-on: ubuntu-latest
     needs: build
@@ -311,42 +288,6 @@ jobs:
           echo ""
         done
 
-  test_img_online_resize:
-    runs-on: ubuntu-latest
-    needs: build
-    container: ${{env.TEST_IMAGE}}:${{github.sha}}
-    steps:
-    - name: Run test
-      id: test
-      timeout-minutes: 3
-      run: /root/vitastor/tests/test_img_online_resize.sh
-    - name: Print logs
-      if: always() && steps.test.outcome == 'failure'
-      run: |
-        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
-          echo "-------- $i --------"
-          cat $i
-          echo ""
-        done
-
-  test_no_uring:
-    runs-on: ubuntu-latest
-    needs: build
-    container: ${{env.TEST_IMAGE}}:${{github.sha}}
-    steps:
-    - name: Run test
-      id: test
-      timeout-minutes: 3
-      run: /root/vitastor/tests/test_no_uring.sh
-    - name: Print logs
-      if: always() && steps.test.outcome == 'failure'
-      run: |
-        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
-          echo "-------- $i --------"
-          cat $i
-          echo ""
-        done
-
   test_interrupted_rebalance:
     runs-on: ubuntu-latest
     needs: build
@@ -2345,24 +2286,6 @@ jobs:
           echo ""
         done
 
-  test_nfs_long_readdir:
-    runs-on: ubuntu-latest
-    needs: build
-    container: ${{env.TEST_IMAGE}}:${{github.sha}}
-    steps:
-    - name: Run test
-      id: test
-      timeout-minutes: 3
-      run: /root/vitastor/tests/test_nfs_long_readdir.sh
-    - name: Print logs
-      if: always() && steps.test.outcome == 'failure'
-      run: |
-        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
-          echo "-------- $i --------"
-          cat $i
-          echo ""
-        done
-
   test_write_encrypted:
     runs-on: ubuntu-latest
     needs: build

CMakeLists.txt

@@ -2,7 +2,7 @@ cmake_minimum_required(VERSION 2.8...3.30)
 
 project(vitastor)
 
-set(VITASTOR_VERSION "3.2.2")
+set(VITASTOR_VERSION "3.0.15")
 
 include(CTest)
 

csi/Dockerfile

@@ -37,8 +37,8 @@ RUN (echo deb http://vitastor.io/debian trixie main > /etc/apt/sources.list.d/vi
     wget -q -O /etc/apt/trusted.gpg.d/vitastor.gpg https://vitastor.io/debian/pubkey.gpg && \
     apt-get update && \
     apt-get install -y vitastor-client ibverbs-providers && \
-    wget https://vitastor.io/archive/qemu/qemu-trixie-10.0.11%2Bds-0%2Bdeb13u1%2Bvitastor2/qemu-utils_10.0.11%2Bds-0%2Bdeb13u1%2Bvitastor2_amd64.deb && \
-    wget https://vitastor.io/archive/qemu/qemu-trixie-10.0.11%2Bds-0%2Bdeb13u1%2Bvitastor2/qemu-block-extra_10.0.11%2Bds-0%2Bdeb13u1%2Bvitastor2_amd64.deb && \
+    wget https://vitastor.io/archive/qemu/qemu-trixie-10.0.2%2Bds-2%2Bvitastor1/qemu-utils_10.0.2%2Bds-2%2Bvitastor1_amd64.deb && \
+    wget https://vitastor.io/archive/qemu/qemu-trixie-10.0.2%2Bds-2%2Bvitastor1/qemu-block-extra_10.0.2%2Bds-2%2Bvitastor1_amd64.deb && \
     dpkg -x qemu-utils*.deb tmp1 && \
     dpkg -x qemu-block-extra*.deb tmp1 && \
     cp -a tmp1/usr/bin/qemu-storage-daemon /usr/bin/ && \

csi/Dockerfile.local

@@ -25,8 +25,8 @@ RUN apt-get update && \
         # NFS mount dependencies
         nfs-common netbase \
         # dependencies of qemu-storage-daemon
-        libaio1t64 libc6 libfuse3-4 libglib2.0-0t64 libgmp10 libgnutls30t64 \
-        libhogweed6t64 libnettle8t64 libnuma1 libselinux1 liburing2 libzstd1 zlib1g && \
+        libnuma1 liburing2 libglib2.0-0 libfuse3-3 libaio1 libzstd1 libnettle8 \
+        libgmp10 libhogweed6 libp11-kit0 libidn2-0 libunistring2 libtasn1-6 libpcre2-8-0 libffi8 && \
     apt-get clean && \
     (echo options nbd nbds_max=128 > /etc/modprobe.d/nbd.conf)
 
@@ -36,8 +36,8 @@ ADD deb /deb
 
 RUN apt-get update && \
     apt-get -y install /deb/vitastor-client_*.deb && \
-    wget https://vitastor.io/archive/qemu/qemu-trixie-10.0.11%2Bds-0%2Bdeb13u1%2Bvitastor2/qemu-utils_10.0.11%2Bds-0%2Bdeb13u1%2Bvitastor2_amd64.deb && \
-    wget https://vitastor.io/archive/qemu/qemu-trixie-10.0.11%2Bds-0%2Bdeb13u1%2Bvitastor2/qemu-block-extra_10.0.11%2Bds-0%2Bdeb13u1%2Bvitastor2_amd64.deb && \
+    wget https://vitastor.io/archive/qemu/qemu-trixie-9.2.2%2Bds-1%2Bvitastor4/qemu-utils_9.2.2%2Bds-1%2Bvitastor4_amd64.deb && \
+    wget https://vitastor.io/archive/qemu/qemu-trixie-9.2.2%2Bds-1%2Bvitastor4/qemu-block-extra_9.2.2%2Bds-1%2Bvitastor4_amd64.deb && \
     dpkg -x qemu-utils*.deb tmp1 && \
     dpkg -x qemu-block-extra*.deb tmp1 && \
     cp -a tmp1/usr/bin/qemu-storage-daemon /usr/bin/ && \

csi/Makefile

@@ -1,4 +1,4 @@
-VITASTOR_VERSION ?= v3.2.2
+VITASTOR_VERSION ?= v3.0.15
 
 all: build push
 

csi/deploy/004-csi-nodeplugin.yaml

@@ -49,7 +49,7 @@ spec:
             capabilities:
               add: ["SYS_ADMIN"]
             allowPrivilegeEscalation: true
-          image: vitalif/vitastor-csi:v3.2.2
+          image: vitalif/vitastor-csi:v3.0.15
           args:
             - "--node=$(NODE_ID)"
             - "--endpoint=$(CSI_ENDPOINT)"

csi/deploy/007-csi-provisioner.yaml

@@ -87,7 +87,7 @@ spec:
             - name: socket-dir
               mountPath: /csi
         - name: csi-attacher
-          image: k8s.gcr.io/sig-storage/csi-attacher:v3.2.2
+          image: k8s.gcr.io/sig-storage/csi-attacher:v3.1.0
           args:
             - "--v=5"
             - "--csi-address=$(ADDRESS)"
@@ -121,7 +121,7 @@ spec:
             privileged: true
             capabilities:
               add: ["SYS_ADMIN"]
-          image: vitalif/vitastor-csi:v3.2.2
+          image: vitalif/vitastor-csi:v3.0.15
           args:
             - "--node=$(NODE_ID)"
             - "--endpoint=$(CSI_ENDPOINT)"

csi/go.mod

@@ -7,7 +7,6 @@ require (
 	github.com/golang/glog v0.0.0-20160126235308-23def4e6c14b
 	github.com/kubernetes-csi/csi-lib-utils v0.9.1
 	golang.org/x/net v0.7.0
-	golang.org/x/sys v0.5.0 // indirect
 	golang.org/x/xerrors v0.0.0-20200804184101-5ec99f83aff1 // indirect
 	google.golang.org/grpc v1.33.1
 	google.golang.org/protobuf v1.24.0

csi/src/config.go

@@ -5,7 +5,7 @@ package vitastor
 
 const (
     vitastorCSIDriverName    = "csi.vitastor.io"
-    vitastorCSIDriverVersion = "3.2.2"
+    vitastorCSIDriverVersion = "3.0.15"
 )
 
 // Config struct fills the parameters of request or user input

csi/src/controllerserver.go

@@ -8,6 +8,7 @@ import (
     "encoding/json"
     "fmt"
     "strings"
+    "time"
     "os"
     "io/ioutil"
 
@@ -26,6 +27,7 @@ const (
     MB int64 = 1024 * KB
     GB int64 = 1024 * MB
     TB int64 = 1024 * GB
+    ETCD_TIMEOUT time.Duration = 15*time.Second
 )
 
 type InodeIndex struct
@@ -685,7 +687,7 @@ func (cs *ControllerServer) ControllerExpandVolume(ctx context.Context, req *csi
 
     return &csi.ControllerExpandVolumeResponse{
         CapacityBytes: int64(inodeCfg[0].Size),
-        NodeExpansionRequired: true,
+        NodeExpansionRequired: false,
     }, nil
 }
 

csi/src/nodeserver.go

@@ -96,16 +96,18 @@ func NewNodeServer(driver *Driver) *NodeServer
     {
         ns.restoreUblkDaemons()
     }
-    ns.restoreNfsDaemons()
-    dur, err := time.ParseDuration(os.Getenv("RESTART_INTERVAL"))
-    if (err != nil)
+    if (ns.method == MOUNT_VDUSE || ns.method == MOUNT_UBLK)
     {
-        dur = 10 * time.Second
-    }
-    ns.restartInterval = dur
-    if (ns.restartInterval != time.Duration(0))
-    {
-        go ns.restarter()
+        dur, err := time.ParseDuration(os.Getenv("RESTART_INTERVAL"))
+        if (err != nil)
+        {
+            dur = 10 * time.Second
+        }
+        ns.restartInterval = dur
+        if (ns.restartInterval != time.Duration(0))
+        {
+            go ns.restarter()
+        }
     }
     return ns
 }
@@ -132,8 +134,10 @@ func (ns *NodeServer) unlockVolume(lockId string)
 
 func (ns *NodeServer) restarter()
 {
-    // Restart dead storage daemons at regular intervals. Otherwise volume I/O
-    // may hang and may even lead to a kernel panic if panic_on_hung_task is set.
+    // Restart dead VDUSE daemons at regular intervals
+    // Otherwise volume I/O may hang in case of a qemu-storage-daemon crash
+    // Moreover, it may lead to a kernel panic of the kernel is configured to
+    // panic on hung tasks
     ticker := time.NewTicker(ns.restartInterval)
     defer ticker.Stop()
     for
@@ -147,7 +151,6 @@ func (ns *NodeServer) restarter()
         {
             ns.restoreUblkDaemons()
         }
-        ns.restoreNfsDaemons()
     }
 }
 
@@ -326,10 +329,15 @@ func (ns *NodeServer) restoreNfsDaemons()
     {
         return
     }
+    activeNFS, err := ns.listActiveNFS()
+    if (err != nil)
+    {
+        return
+    }
     // Check all state files and try to restore active mounts
     for _, stateFile := range stateFiles
     {
-        ns.checkNfsState(stateFile)
+        ns.checkNfsState(stateFile, activeNFS)
     }
 }
 
@@ -355,43 +363,24 @@ func (ns *NodeServer) readNfsState(stateFile string, allowNotExists bool) (*NfsS
     return &state, nil
 }
 
-func (ns *NodeServer) checkNfsState(stateFile string)
+func (ns *NodeServer) checkNfsState(stateFile string, activeNfs map[int][]string)
 {
-    // Read state file to determine the lock ID
+    // Read state file
     state, err := ns.readNfsState(stateFile, false)
     if (err != nil)
     {
         return
     }
     // Lock FS
-    lockId := state.ConfigPath+":fs:"+state.FsName
-    ns.lockVolume(lockId)
-    defer ns.unlockVolume(lockId)
-    // Recheck state and mounts under the lock: publish/unpublish may have changed
-    // both while this recovery attempt was waiting for the lock.
-    state, err = ns.readNfsState(stateFile, true)
-    if (err != nil || state == nil)
-    {
-        return
-    }
-    activeNfs, err := ns.listActiveNFS()
-    if (err != nil)
-    {
-        return
-    }
+    ns.lockVolume(state.ConfigPath+":fs:"+state.FsName)
+    defer ns.unlockVolume(state.ConfigPath+":fs:"+state.FsName)
+    // Check if NFS at this port is still mounted
     pidFile := ns.stateDir + filepath.Base(stateFile)
     pidFile = pidFile[0:len(pidFile)-5] + ".pid"
     if (len(activeNfs[state.Port]) == 0)
     {
-        // This is a stale state file. Remove the root staging mount before
-        // stopping the daemon, otherwise a dead hidden NFS mount may remain.
+        // this is a stale state file, remove it
         klog.Warningf("state file %v contains stale mount at port %d, removing it", stateFile, state.Port)
-        err = mount.CleanupMountPoint(state.Path, ns.mounter, false)
-        if (err != nil)
-        {
-            klog.Errorf("failed to unmount stale NFS mount %v: %v", state.Path, err)
-            return
-        }
         ns.stopNFS(stateFile, pidFile)
         return
     }
@@ -411,7 +400,6 @@ func (ns *NodeServer) checkNfsState(stateFile string)
             "--pidfile", pidFile,
             "--bind", "127.0.0.1",
             "--port", fmt.Sprintf("%d", state.Port),
-            "--config_path", state.ConfigPath,
             "--fs", state.FsName,
             "--pool", state.Pool,
             "--portmap", "0",
@@ -737,7 +725,6 @@ func (ns *NodeServer) mountNFS(ctxVars map[string]string) (string, error)
         "--pidfile", pidFile,
         "--bind", "127.0.0.1",
         "--port", "auto",
-        "--config_path", state.ConfigPath,
         "--fs", state.FsName,
         "--pool", state.Pool,
         "--portmap", "0",
@@ -767,7 +754,7 @@ func (ns *NodeServer) mountNFS(ctxVars map[string]string) (string, error)
     // Mount NFS
     _, _, err = system(
         "mount", "-t", "nfs", "127.0.0.1:/", state.Path,
-        "-o", fmt.Sprintf("port=%d,mountport=%d,nfsvers=3,soft,sync,nolock,tcp", port, port),
+        "-o", fmt.Sprintf("port=%d,mountport=%d,nfsvers=3,soft,nolock,tcp", port, port),
     )
     if (err != nil)
     {
@@ -828,38 +815,26 @@ func (ns *NodeServer) listActiveNFS() (map[int][]string, error)
         klog.Errorf("failed to list mounts: %v", err)
         return nil, err
     }
-    return collectActiveNFS(mounts), nil
-}
-
-func collectActiveNFS(mounts []mount.MountInfo) map[int][]string
-{
     activeNFS := make(map[int][]string)
-    for _, mnt := range mounts
+    for _, mount := range mounts
     {
-        // Only count PVC subdirectory mounts, not the root staging mount. Newer
-        // kernels expose the subdirectory in Source while older ones expose it
-        // in Root. The staging mount may also be visible through several bind
-        // aliases in the CSI container, so its mount point isn't a reliable
-        // discriminator.
-        if (mnt.FsType != "nfs" || !strings.HasPrefix(mnt.Source, "127.0.0.1:/") ||
-            (mnt.Source == "127.0.0.1:/" && mnt.Root == "/"))
+        // Volume mounts always refer to subpaths
+        if (mount.FsType == "nfs" && mount.Root != "/")
         {
-            continue
-        }
-        // NFS-specific options, including port, are per-superblock options.
-        for _, opt := range mnt.SuperOptions
-        {
-            if (strings.HasPrefix(opt, "port="))
+            for _, opt := range mount.MountOptions
             {
-                port64, err := strconv.ParseUint(opt[5:], 10, 16)
-                if (err == nil)
+                if (strings.HasPrefix(opt, "port="))
                 {
-                    activeNFS[int(port64)] = append(activeNFS[int(port64)], mnt.MountPoint)
+                    port64, err := strconv.ParseUint(opt[5:], 10, 16)
+                    if (err == nil)
+                    {
+                        activeNFS[int(port64)] = append(activeNFS[int(port64)], mount.MountPoint)
+                    }
                 }
             }
         }
     }
-    return activeNFS
+    return activeNFS, nil
 }
 
 // NodePublishVolume mounts the volume mounted to the staging path to the target path
@@ -1055,131 +1030,10 @@ func (ns *NodeServer) NodeGetVolumeStats(ctx context.Context, req *csi.NodeGetVo
     return nil, status.Error(codes.Unimplemented, "")
 }
 
-// NodeExpandVolume expands the block device and, for filesystem volumes, the filesystem on the node
+// NodeExpandVolume expanding the file system on the node
 func (ns *NodeServer) NodeExpandVolume(ctx context.Context, req *csi.NodeExpandVolumeRequest) (*csi.NodeExpandVolumeResponse, error)
 {
-    klog.Infof("received node expand volume request %+v", protosanitizer.StripSecrets(req))
-    if (req == nil)
-    {
-        return nil, status.Error(codes.InvalidArgument, "request cannot be empty")
-    }
-    if (req.VolumeId == "")
-    {
-        return nil, status.Error(codes.InvalidArgument, "VolumeId is required")
-    }
-
-    ctxVars := make(map[string]string)
-    err := json.Unmarshal([]byte(req.VolumeId), &ctxVars)
-    if (err != nil)
-    {
-        return nil, status.Error(codes.Internal, "volume ID not in JSON format")
-    }
-    volName := ctxVars["name"]
-
-    if (ctxVars["vitastorfs"] != "")
-    {
-        // VitastorFS CSI volumes are NFS-backed and currently don't have per-volume quotas to resize here.
-        return &csi.NodeExpandVolumeResponse{CapacityBytes: req.GetCapacityRange().GetRequiredBytes()}, nil
-    }
-
-    ns.lockVolume(ctxVars["configPath"]+":block:"+volName)
-    defer ns.unlockVolume(ctxVars["configPath"]+":block:"+volName)
-
-    targetPath := req.GetStagingTargetPath()
-    if (targetPath == "")
-    {
-        targetPath = req.GetVolumePath()
-    }
-    if (targetPath == "")
-    {
-        return nil, status.Error(codes.InvalidArgument, "VolumePath or StagingTargetPath is required")
-    }
-
-    devicePath, err := GetDeviceNameFromMount(targetPath)
-    if (err != nil)
-    {
-        if (os.IsNotExist(err))
-        {
-            return nil, status.Error(codes.NotFound, "Target path not found")
-        }
-        return nil, err
-    }
-    if (devicePath == "")
-    {
-        return nil, status.Errorf(codes.NotFound, "%s is not a mountpoint", targetPath)
-    }
-    if (devicePath[0] != '/')
-    {
-        return nil, status.Errorf(codes.Internal, "failed to resolve device for %s: %s", targetPath, devicePath)
-    }
-    if (len(devicePath) < 5 || devicePath[0:5] != "/dev/")
-    {
-        // NodeExpandVolume may be called with the published target path. For block volumes it is
-        // usually a bind mount of the staging target, so resolve one more mount layer if needed.
-        var realDevicePath string
-        realDevicePath, err = GetDeviceNameFromMount(devicePath)
-        if (err == nil && realDevicePath != "")
-        {
-            devicePath = realDevicePath
-        }
-    }
-
-    // All newest Vitastor driver versions (nbd, ublk and vduse) support online resize.
-    // So all we need to do is just wait a bit until the device is resized by the driver.
-    requiredBytes := req.GetCapacityRange().GetRequiredBytes()
-    if (requiredBytes > 0)
-    {
-        timeout := 30
-        var actualSize uint64
-        var err error
-        for i := 0; i < timeout && actualSize < uint64(requiredBytes); i++
-        {
-            actualSize, err = GetBlockDeviceSize(devicePath)
-            if (err != nil)
-            {
-                return nil, err
-            }
-            time.Sleep(1 * time.Second)
-        }
-        if (actualSize < uint64(requiredBytes))
-        {
-            return nil, status.Errorf(codes.NotFound, "Failed to wait for online resize of %s to %v bytes: actual size is still %v after %v seconds",
-                devicePath, requiredBytes, actualSize, timeout)
-        }
-    }
-
-    isBlock := req.GetVolumeCapability().GetBlock() != nil
-    if (isBlock)
-    {
-        return &csi.NodeExpandVolumeResponse{CapacityBytes: requiredBytes}, nil
-    }
-
-    diskMounter := &mount.SafeFormatAndMount{Interface: ns.mounter, Exec: utilexec.New()}
-    fsType := req.GetVolumeCapability().GetMount().GetFsType()
-    if (fsType == "")
-    {
-        fsType, err = diskMounter.GetDiskFormat(devicePath)
-        if (err != nil)
-        {
-            return nil, err
-        }
-    }
-
-    switch (fsType)
-    {
-        case "ext4":
-            _, err = systemCombined("resize2fs", devicePath)
-        case "xfs":
-            _, err = systemCombined("xfs_growfs", targetPath)
-        default:
-            return nil, status.Errorf(codes.InvalidArgument, "unsupported filesystem for resize: %s", fsType)
-    }
-    if (err != nil)
-    {
-        return nil, err
-    }
-
-    return &csi.NodeExpandVolumeResponse{CapacityBytes: requiredBytes}, nil
+    return nil, status.Error(codes.Unimplemented, "")
 }
 
 // NodeGetCapabilities returns the supported capabilities of the node server
@@ -1194,13 +1048,6 @@ func (ns *NodeServer) NodeGetCapabilities(ctx context.Context, req *csi.NodeGetC
                     },
                 },
             },
-            &csi.NodeServiceCapability{
-                Type: &csi.NodeServiceCapability_Rpc{
-                    Rpc: &csi.NodeServiceCapability_RPC{
-                        Type: csi.NodeServiceCapability_RPC_EXPAND_VOLUME,
-                    },
-                },
-            },
         },
     }, nil
 }

csi/src/nodeserver_test.go

@@ -1,73 +0,0 @@
-// Copyright (c) Vitaliy Filippov, 2019+
-// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
-
-package vitastor
-
-import (
-    "reflect"
-    "testing"
-
-    "k8s.io/utils/mount"
-)
-
-func TestCollectActiveNFS(t *testing.T)
-{
-    mounts := []mount.MountInfo{
-        {
-            Root:         "/",
-            Source:       "127.0.0.1:/",
-            MountPoint:   "/var/lib/kubelet/plugins/csi.vitastor.io/nfs/hash",
-            FsType:       "nfs",
-            MountOptions: []string{"rw", "relatime"},
-            SuperOptions: []string{"rw", "vers=3", "port=45059"},
-        },
-        {
-            // This is how NFS subdirectory bind mounts appear on Linux 6.19:
-            // Root is still / and the subdirectory is only visible in Source.
-            Root:         "/",
-            Source:       "127.0.0.1:/pvc-one",
-            MountPoint:   "/var/lib/kubelet/pods/pod-one/volumes/kubernetes.io~csi/pvc-one/mount",
-            FsType:       "nfs",
-            MountOptions: []string{"rw", "relatime"},
-            SuperOptions: []string{"rw", "vers=3", "port=45059"},
-        },
-        {
-            Root:         "/",
-            Source:       "127.0.0.1:/pvc-two",
-            MountPoint:   "/var/lib/kubelet/pods/pod-two/volumes/kubernetes.io~csi/pvc-two/mount",
-            FsType:       "nfs",
-            MountOptions: []string{"rw", "relatime"},
-            SuperOptions: []string{"rw", "vers=3", "port=45059"},
-        },
-        {
-            // Older kernels expose the bind-mounted subdirectory in Root.
-            Root:         "/pvc-three",
-            Source:       "127.0.0.1:/",
-            MountPoint:   "/var/lib/kubelet/pods/pod-three/volumes/kubernetes.io~csi/pvc-three/mount",
-            FsType:       "nfs",
-            MountOptions: []string{"rw", "relatime"},
-            SuperOptions: []string{"rw", "vers=3", "port=45059"},
-        },
-        {
-            Root:         "/",
-            Source:       "server:/other",
-            MountPoint:   "/mnt/other",
-            FsType:       "nfs",
-            MountOptions: []string{"rw"},
-            SuperOptions: []string{"rw", "port=2049"},
-        },
-    }
-
-    got := collectActiveNFS(mounts)
-    want := map[int][]string{
-        45059: {
-            "/var/lib/kubelet/pods/pod-one/volumes/kubernetes.io~csi/pvc-one/mount",
-            "/var/lib/kubelet/pods/pod-two/volumes/kubernetes.io~csi/pvc-two/mount",
-            "/var/lib/kubelet/pods/pod-three/volumes/kubernetes.io~csi/pvc-three/mount",
-        },
-    }
-    if (!reflect.DeepEqual(got, want))
-    {
-        t.Fatalf("unexpected active NFS mounts: got %#v, want %#v", got, want)
-    }
-}

csi/src/utils.go

@@ -20,7 +20,6 @@ import (
 
     "google.golang.org/grpc/codes"
     "google.golang.org/grpc/status"
-    "golang.org/x/sys/unix"
 )
 
 type MountMethod int
@@ -520,22 +519,3 @@ func GetDeviceNameFromMount(mountPath string) (string, error)
 
     return device, nil
 }
-
-func GetBlockDeviceSize(devicePath string) (uint64, error)
-{
-    file, err := os.Open(devicePath)
-    if (err != nil)
-    {
-        return 0, err
-    }
-    defer file.Close()
-
-    // BLKGETSIZE64 returns the size in bytes as a uint64/int64
-    size, err := unix.IoctlGetInt(int(file.Fd()), unix.BLKGETSIZE64)
-    if (err != nil)
-    {
-        return 0, fmt.Errorf("ioctl BLKGETSIZE64 failed: %w", err)
-    }
-
-    return uint64(size), nil
-}

debian/changelog

@@ -1,4 +1,4 @@
-vitastor (3.2.2-1) unstable; urgency=medium
+vitastor (3.0.15-1) unstable; urgency=medium
 
   * Bugfixes
 

debian/vitastor-buildenv.Dockerfile

@@ -12,9 +12,7 @@ ARG REL=
 WORKDIR /root
 
 RUN set -e -x; \
-    if [ "$REL" = "buster" ]; then \
-        perl -i -pe 's/deb.debian.org/archive.debian.org/' /etc/apt/sources.list; \
-    fi; \
+    perl -i -pe 's/deb.debian.org/archive.debian.org/' /etc/apt/sources.list; \
     apt-get update; \
     apt-get -y install wget; \
     wget https://vitastor.io/debian/pubkey.gpg -O /etc/apt/trusted.gpg.d/vitastor.gpg; \
@@ -26,6 +24,6 @@ RUN set -e -x; \
 
 RUN apt-get update && \
     apt-get -y install fio libgoogle-perftools-dev devscripts libjerasure-dev cmake libc-ares-dev libisal-crypto-dev \
-        libibverbs-dev librdmacm-dev libisal-dev libnl-3-dev libnl-genl-3-dev libssl-dev curl nodejs npm node-nan node-bindings && \
+        libibverbs-dev librdmacm-dev libisal-dev libnl-3-dev libnl-genl-3-dev curl nodejs npm node-nan node-bindings && \
     apt-get -y build-dep fio && \
     apt-get --download-only source fio

debian/vitastor-client.install

@@ -4,8 +4,6 @@ usr/bin/vitastor-rm
 usr/bin/vitastor-nbd
 usr/bin/vitastor-ublk
 usr/bin/vitastor-nfs
-usr/sbin/mount.vitastorfs
-usr/sbin/mount.vitastorblk
 usr/bin/vitastor-kv
 usr/bin/vitastor-kv-stress
 usr/lib/*/libvitastor*.so*

docker/Makefile

@@ -1,4 +1,4 @@
-VITASTOR_VERSION ?= v3.2.2
+VITASTOR_VERSION ?= v3.0.15
 
 all: build push
 

docker/etc/vitastor/docker.conf

@@ -4,7 +4,7 @@
 #
 
 # Desired Vitastor version
-VITASTOR_VERSION=v3.2.2
+VITASTOR_VERSION=v3.0.15
 
 # Additional arguments for all containers
 # For example, you may want to specify a custom logging driver here

docs/config/network.en.md

@@ -17,9 +17,6 @@ between clients, OSDs and etcd.
 - [rdma_device](#rdma_device)
 - [rdma_port_num](#rdma_port_num)
 - [rdma_gid_index](#rdma_gid_index)
-- [rdma_traffic_class](#rdma_traffic_class)
-- [rdma_sl](#rdma_sl)
-- [rdma_set_flow_label](#rdma_set_flow_label)
 - [rdma_mtu](#rdma_mtu)
 - [rdma_max_sge](#rdma_max_sge)
 - [rdma_max_msg](#rdma_max_msg)
@@ -36,7 +33,6 @@ between clients, OSDs and etcd.
 - [etcd_ws_keepalive_interval](#etcd_ws_keepalive_interval)
 - [etcd_min_reload_interval](#etcd_min_reload_interval)
 - [tcp_header_buffer_size](#tcp_header_buffer_size)
-- [rxbounce](#rxbounce)
 - [min_zerocopy_send_size](#min_zerocopy_send_size)
 - [use_sync_send_recv](#use_sync_send_recv)
 
@@ -158,40 +154,6 @@ A correct rdma_gid_index for RoCEv2 is usually 1 (IPv6) or 3 (IPv4).
 
 Not relevant for RDMA-CM (use_rdmacm).
 
-## rdma_traffic_class
-
-- Type: integer
-- Default: 0
-
-Full TOS byte for RDMA traffic in the IP header (RoCEv2): DSCP in the
-higher 6 bits and ECN in the lower 2 bits. For example, 106 (0x6A) =
-DSCP 26 + ECT(0). Works with both libibverbs and RDMA-CM modes and should
-be set on both OSDs and clients to mark traffic in both directions.
-Default 0 = don't mark.
-
-## rdma_sl
-
-- Type: integer
-- Default: 0
-
-Service Level (802.1p priority analogue) for RDMA traffic, put into the
-address vector of every connection. Only applies to the libibverbs mode:
-in RDMA-CM mode the service level is derived from rdma_traffic_class.
-Default 0 = as is without PFC-configured networks.
-
-## rdma_set_flow_label
-
-- Type: boolean
-- Default: false
-
-Automatically derive an IPv6 Flow Label for RDMA connections from the
-queue pair number pair, so that every connection gets its own label and
-connections spread over ECMP links while each connection keeps a single
-stable path.  Only applied to IPv6 GIDs: an IPv4-mapped GID
-(::ffff:a.b.c.d) produces IPv4 packets which have no flow label.  Only
-applies to the libibverbs mode; RDMA-CM generates flow labels on its own.
-Default false = don't set.
-
 ## rdma_mtu
 
 - Type: integer
@@ -341,21 +303,6 @@ is received without an additional copy. You can try to play with this
 parameter and see how it affects random iops and linear bandwidth if you
 want.
 
-## rxbounce
-
-- Type: boolean
-- Default: false
-- Can be changed online: yes
-
-Receive all TCP data into an intermediate buffer before copying to the actual
-destination. Enabling this option is required for Windows guests if you use TCP
-and [security.en.md#proto_checksums](proto_checksums) because Windows uses a
-shared system-wide 'dummy page' to skip unneeded blocks during reads and breaks
-checksum verification with it when data copy is avoided. rxbounce is only needed
-when you use TCP and don't use transport encryption (AES-GCM); it's not required
-with RDMA or with GCM encryption because both always copy data anyway (yes, RDMA
-isn't actually zerocopy in Vitastor).
-
 ## min_zerocopy_send_size
 
 - Type: integer

docs/config/network.ru.md

@@ -17,9 +17,6 @@
 - [rdma_device](#rdma_device)
 - [rdma_port_num](#rdma_port_num)
 - [rdma_gid_index](#rdma_gid_index)
-- [rdma_traffic_class](#rdma_traffic_class)
-- [rdma_sl](#rdma_sl)
-- [rdma_set_flow_label](#rdma_set_flow_label)
 - [rdma_mtu](#rdma_mtu)
 - [rdma_max_sge](#rdma_max_sge)
 - [rdma_max_msg](#rdma_max_msg)
@@ -36,7 +33,6 @@
 - [etcd_ws_keepalive_interval](#etcd_ws_keepalive_interval)
 - [etcd_min_reload_interval](#etcd_min_reload_interval)
 - [tcp_header_buffer_size](#tcp_header_buffer_size)
-- [rxbounce](#rxbounce)
 - [min_zerocopy_send_size](#min_zerocopy_send_size)
 - [use_sync_send_recv](#use_sync_send_recv)
 
@@ -161,40 +157,6 @@ libibverbs < v32.
 
 Опция неприменима к RDMA-CM (use_rdmacm).
 
-## rdma_traffic_class
-
-- Тип: целое число
-- Значение по умолчанию: 0
-
-Полный байт TOS для RDMA-трафика в заголовке IP (RoCEv2): DSCP в старших
-6 битах и ECN в младших 2. Например, 106 (0x6A) = DSCP 26 + ECT(0).
-Работает в обоих режимах RDMA - и libibverbs, и RDMA-CM; должен быть задан
-и на OSD, и на клиентах, чтобы трафик помечался в обе стороны. По умолчанию
-0 - не помечать.
-
-## rdma_sl
-
-- Тип: целое число
-- Значение по умолчанию: 0
-
-Service Level (аналог приоритета 802.1p) для RDMA-трафика; попадает в
-адресный вектор каждого соединения. Применим только к libibverbs-режиму:
-в RDMA-CM уровень обслуживания выводится из rdma_traffic_class.
-По умолчанию 0 = как в сетях без PFC.
-
-## rdma_set_flow_label
-
-- Тип: булево (да/нет)
-- Значение по умолчанию: false
-
-Автоматически выводить IPv6 Flow Label для RDMA-соединений из пары
-номеров очередей (QPN), чтобы каждое соединение получало свою метку,
-соединения распределялись по ECMP-каналам, но каждое соединение шло по
-одному стабильному пути. Применяется только к IPv6-GID: IPv4-mapped GID
-(::ffff:a.b.c.d) даёт IPv4-пакеты, где поля flow label нет. Применим
-только к libibverbs-режиму - RDMA-CM всегда генерирует flow label сам.
-По умолчанию false = не задавать.
-
 ## rdma_mtu
 
 - Тип: целое число
@@ -348,22 +310,6 @@ Vitastor содержат 128-байтные заголовки, за котор
 поменять этот параметр и посмотреть, как он влияет на производительность
 случайного и линейного доступа.
 
-## rxbounce
-
-- Тип: булево (да/нет)
-- Значение по умолчанию: false
-- Можно менять на лету: да
-
-Принимать все данные по TCP в промежуточный буфер и потом копировать их в
-буфер клиента (ВМ). Это нужно для гостевых ОС Windows с TCP и включёнными
-[security.ru.md#proto_checksums](proto_checksums), потому что Windows использует
-одну разделяемую общесистемную страницу-заглушку для пропуска ненужных блоков
-при чтении и ломает проверку контрольных сумм, если данные не копируются лишний
-раз. Опция rxbounce нужна только если вы используете TCP и не используете
-транспортное шифрование (AES-GCM); с RDMA или шифрованием опция не нужна, так
-как в этом случае данные копируются всегда (да, RDMA в Vitastor на самом деле
-не zero-copy).
-
 ## min_zerocopy_send_size
 
 - Тип: целое число

docs/config/security.en.md

@@ -19,7 +19,6 @@ a PEM string with certificate or a private key. In the latter case, the string m
 - [pkey](#pkey)
 - [etcd_ca](#etcd_ca)
 - [client_ca](#client_ca)
-- [admin_ca](#admin_ca)
 - [osd_ca](#osd_ca)
 - [mon_ca](#mon_ca)
 - [antietcd_cert](#antietcd_cert)
@@ -87,12 +86,6 @@ certificate itself - it's fine to use it for etcd_ca.
 Trusted TLS CA to verify Vitastor client certificates.
 Mandatory for Vitastor protocol encryption.
 
-## admin_ca
-
-- Type: string
-
-Trusted TLS CA to verify Vitastor administrator certificates.
-
 ## osd_ca
 
 - Type: string

docs/config/security.ru.md

@@ -21,7 +21,6 @@ OSD, мониторами и клиентами.
 - [pkey](#pkey)
 - [etcd_ca](#etcd_ca)
 - [client_ca](#client_ca)
-- [admin_ca](#admin_ca)
 - [osd_ca](#osd_ca)
 - [mon_ca](#mon_ca)
 - [antietcd_cert](#antietcd_cert)
@@ -89,12 +88,6 @@ OSD, мониторами и клиентами.
 Доверенный TLS-сертификат для проверки сертификатов клиентов Vitastor.
 Требуется для шифрования протокола Vitastor.
 
-## admin_ca
-
-- Тип: строка
-
-Доверенный TLS-сертификат для проверки сертификатов администраторов Vitastor.
-
 ## osd_ca
 
 - Тип: строка

docs/config/src/make.js

@@ -13,7 +13,6 @@ const L = {
         toc_installation: 'Installation',
         toc_config: '[Configuration](../config.en.md)',
         toc_usage: 'Usage',
-        toc_tests: 'Tests',
         toc_performance: 'Performance',
         online: 'Can be changed online: yes',
     },
@@ -29,7 +28,6 @@ const L = {
         toc_installation: 'Установка',
         toc_config: '[Конфигурация](../config.ru.md)',
         toc_usage: 'Использование',
-        toc_tests: 'Тесты',
         toc_performance: 'Производительность',
         online: 'Можно менять на лету: да',
     },

docs/config/src/network.yml

@@ -188,53 +188,6 @@
     Правильный rdma_gid_index для RoCEv2, как правило, 1 (IPv6) или 3 (IPv4).
 
     Опция неприменима к RDMA-CM (use_rdmacm).
-- name: rdma_traffic_class
-  type: int
-  default: 0
-  info: |
-    Full TOS byte for RDMA traffic in the IP header (RoCEv2): DSCP in the
-    higher 6 bits and ECN in the lower 2 bits. For example, 106 (0x6A) =
-    DSCP 26 + ECT(0). Works with both libibverbs and RDMA-CM modes and should
-    be set on both OSDs and clients to mark traffic in both directions.
-    Default 0 = don't mark.
-  info_ru: |
-    Полный байт TOS для RDMA-трафика в заголовке IP (RoCEv2): DSCP в старших
-    6 битах и ECN в младших 2. Например, 106 (0x6A) = DSCP 26 + ECT(0).
-    Работает в обоих режимах RDMA - и libibverbs, и RDMA-CM; должен быть задан
-    и на OSD, и на клиентах, чтобы трафик помечался в обе стороны. По умолчанию
-    0 - не помечать.
-- name: rdma_sl
-  type: int
-  default: 0
-  info: |
-    Service Level (802.1p priority analogue) for RDMA traffic, put into the
-    address vector of every connection. Only applies to the libibverbs mode:
-    in RDMA-CM mode the service level is derived from rdma_traffic_class.
-    Default 0 = as is without PFC-configured networks.
-  info_ru: |
-    Service Level (аналог приоритета 802.1p) для RDMA-трафика; попадает в
-    адресный вектор каждого соединения. Применим только к libibverbs-режиму:
-    в RDMA-CM уровень обслуживания выводится из rdma_traffic_class.
-    По умолчанию 0 = как в сетях без PFC.
-- name: rdma_set_flow_label
-  type: bool
-  default: false
-  info: |
-    Automatically derive an IPv6 Flow Label for RDMA connections from the
-    queue pair number pair, so that every connection gets its own label and
-    connections spread over ECMP links while each connection keeps a single
-    stable path.  Only applied to IPv6 GIDs: an IPv4-mapped GID
-    (::ffff:a.b.c.d) produces IPv4 packets which have no flow label.  Only
-    applies to the libibverbs mode; RDMA-CM generates flow labels on its own.
-    Default false = don't set.
-  info_ru: |
-    Автоматически выводить IPv6 Flow Label для RDMA-соединений из пары
-    номеров очередей (QPN), чтобы каждое соединение получало свою метку,
-    соединения распределялись по ECMP-каналам, но каждое соединение шло по
-    одному стабильному пути. Применяется только к IPv6-GID: IPv4-mapped GID
-    (::ffff:a.b.c.d) даёт IPv4-пакеты, где поля flow label нет. Применим
-    только к libibverbs-режиму - RDMA-CM всегда генерирует flow label сам.
-    По умолчанию false = не задавать.
 - name: rdma_mtu
   type: int
   info: |
@@ -405,29 +358,6 @@
     параметра читается без дополнительного копирования. Вы можете попробовать
     поменять этот параметр и посмотреть, как он влияет на производительность
     случайного и линейного доступа.
-- name: rxbounce
-  type: bool
-  default: false
-  online: true
-  info: |
-    Receive all TCP data into an intermediate buffer before copying to the actual
-    destination. Enabling this option is required for Windows guests if you use TCP
-    and [security.en.md#proto_checksums](proto_checksums) because Windows uses a
-    shared system-wide 'dummy page' to skip unneeded blocks during reads and breaks
-    checksum verification with it when data copy is avoided. rxbounce is only needed
-    when you use TCP and don't use transport encryption (AES-GCM); it's not required
-    with RDMA or with GCM encryption because both always copy data anyway (yes, RDMA
-    isn't actually zerocopy in Vitastor).
-  info_ru: |
-    Принимать все данные по TCP в промежуточный буфер и потом копировать их в
-    буфер клиента (ВМ). Это нужно для гостевых ОС Windows с TCP и включёнными
-    [security.ru.md#proto_checksums](proto_checksums), потому что Windows использует
-    одну разделяемую общесистемную страницу-заглушку для пропуска ненужных блоков
-    при чтении и ломает проверку контрольных сумм, если данные не копируются лишний
-    раз. Опция rxbounce нужна только если вы используете TCP и не используете
-    транспортное шифрование (AES-GCM); с RDMA или шифрованием опция не нужна, так
-    как в этом случае данные копируются всегда (да, RDMA в Vitastor на самом деле
-    не zero-copy).
 - name: min_zerocopy_send_size
   type: int
   default: 32768

docs/config/src/security.yml

@@ -39,12 +39,6 @@
   info_ru: |
     Доверенный TLS-сертификат для проверки сертификатов клиентов Vitastor.
     Требуется для шифрования протокола Vitastor.
-- name: admin_ca
-  type: string
-  info: |
-    Trusted TLS CA to verify Vitastor administrator certificates.
-  info_ru: |
-    Доверенный TLS-сертификат для проверки сертификатов администраторов Vitastor.
 - name: osd_ca
   type: string
   info: |

docs/installation/docker.en.md

@@ -26,9 +26,9 @@ at Vitastor Kubernetes operator: https://github.com/Antilles7227/vitastor-operat
 The instruction is very simple.
 
 1. Download a Docker image of the desired version: \
-   `docker pull vitalif/vitastor:v3.2.2`
+   `docker pull vitalif/vitastor:v3.0.15`
 2. Install scripts to the host system: \
-   `docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.2.2 install.sh`
+   `docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.0.15 install.sh`
 3. Reload udev rules: \
    `udevadm control --reload-rules`
 4. Enable the vitastor-host service: \

docs/installation/docker.ru.md

@@ -25,9 +25,9 @@ Vitastor можно установить в Docker/Podman. При этом etcd,
 Инструкция по установке максимально простая.
 
 1. Скачайте Docker-образ желаемой версии: \
-   `docker pull vitalif/vitastor:v3.2.2`
+   `docker pull vitalif/vitastor:v3.0.15`
 2. Установите скрипты в хост-систему командой: \
-   `docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.2.2 install.sh`
+   `docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.0.15 install.sh`
 3. Перезагрузите правила udev: \
    `udevadm control --reload-rules`
 4. Включите сервис vitastor-host: \

docs/installation/packages.en.md

@@ -14,9 +14,12 @@
   - Debian 13 (Trixie/Sid): `deb https://vitastor.io/debian trixie main`
   - Debian 12 (Bookworm): `deb https://vitastor.io/debian bookworm main`
   - Debian 11 (Bullseye): `deb https://vitastor.io/debian bullseye main`
-  - Ubuntu 26.04 (Resolute): `deb https://vitastor.io/debian resolute main`
-  - Ubuntu 24.04 (Noble): `deb https://vitastor.io/debian noble main`
+  - Debian 10 (Buster): `deb https://vitastor.io/debian buster main`
   - Ubuntu 22.04 (Jammy): `deb https://vitastor.io/debian jammy main`
+  - Ubuntu 24.04 (Noble): `deb https://vitastor.io/debian noble main`
+  - Ubuntu 26.04 (Resolute): `deb https://vitastor.io/debian resolute main`
+  - Add `-oldstable` to bookworm/bullseye/buster in this line to install the last
+    stable version from 0.9.x branch instead of 1.x
 - To always prefer vitastor-patched QEMU and Libvirt versions, add the following to `/etc/apt/preferences`:
   ```
   Package: *
@@ -28,14 +31,17 @@
 ## CentOS
 
 - Add Vitastor package repository:
+  - CentOS 7: `yum install https://vitastor.io/rpms/centos/7/vitastor-release.rpm`
   - CentOS 8: `dnf install https://vitastor.io/rpms/centos/8/vitastor-release.rpm`
   - AlmaLinux 9 and other RHEL 9 clones (Rocky, Oracle...): `dnf install https://vitastor.io/rpms/centos/9/vitastor-release.rpm`
   - AlmaLinux 10 and other RHEL 10 clones: `dnf install https://vitastor.io/rpms/centos/10/vitastor-release.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`
   - RHEL 9/10 clones: not required
 - 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`
   - RHEL 9 clones: `dnf install https://www.elrepo.org/elrepo-release-9.el9.elrepo.noarch.rpm`
   - RHEL 10 clones: not required

docs/installation/packages.ru.md

@@ -14,9 +14,12 @@
   - Debian 13 (Trixie/Sid): `deb https://vitastor.io/debian trixie main`
   - Debian 12 (Bookworm): `deb https://vitastor.io/debian bookworm main`
   - Debian 11 (Bullseye): `deb https://vitastor.io/debian bullseye main`
-  - Ubuntu 26.04 (Resolute): `deb https://vitastor.io/debian resolute main`
-  - Ubuntu 24.04 (Noble): `deb https://vitastor.io/debian noble main`
+  - Debian 10 (Buster): `deb https://vitastor.io/debian buster main`
   - Ubuntu 22.04 (Jammy): `deb https://vitastor.io/debian jammy main`
+  - Ubuntu 24.04 (Noble): `deb https://vitastor.io/debian noble main`
+  - Ubuntu 26.04 (Resolute): `deb https://vitastor.io/debian resolute main`
+  - Добавьте `-oldstable` к слову bookworm/bullseye/buster в этой строке, чтобы
+    установить последнюю стабильную версию из ветки 0.9.x вместо 1.x
 - Чтобы всегда предпочитались версии пакетов QEMU и Libvirt с патчами Vitastor, добавьте в `/etc/apt/preferences`:
   ```
   Package: *
@@ -28,14 +31,17 @@
 ## CentOS
 
 - Добавьте в систему репозиторий Vitastor:
+  - CentOS 7: `yum install https://vitastor.io/rpms/centos/7/vitastor-release.rpm`
   - CentOS 8: `dnf install https://vitastor.io/rpms/centos/8/vitastor-release.rpm`
   - AlmaLinux 9 и другие клоны RHEL 9 (Rocky, Oracle...): `dnf install https://vitastor.io/rpms/centos/9/vitastor-release.rpm`
   - AlmaLinux 10 и другие клоны RHEL 10: `dnf install https://vitastor.io/rpms/centos/10/vitastor-release.rpm`
 - Включите EPEL: `yum/dnf install epel-release`
 - Включите дополнительные репозитории CentOS:
+  - CentOS 7: `yum install centos-release-scl`
   - CentOS 8: `dnf install centos-release-advanced-virtualization`
   - Клоны RHEL 9/10: не нужно
 - Включите elrepo-kernel:
+  - CentOS 7: `yum install https://www.elrepo.org/elrepo-release-7.el7.elrepo.noarch.rpm`
   - CentOS 8: `dnf install https://www.elrepo.org/elrepo-release-8.el8.elrepo.noarch.rpm`
   - Клоны RHEL 9: `dnf install https://www.elrepo.org/elrepo-release-9.el9.elrepo.noarch.rpm`
   - Клоны RHEL 10: не нужно

docs/installation/proxmox.en.md

@@ -6,10 +6,10 @@
 
 # Proxmox VE
 
-To enable Vitastor support in Proxmox Virtual Environment (7.x-9.x are supported):
+To enable Vitastor support in Proxmox Virtual Environment (6.4-9.x are supported):
 
 - Add the corresponding Vitastor Debian repository into sources.list on Proxmox hosts:
-  trixie for 9.0+, bookworm for 8.1+, pve8.0 for 8.0, bullseye for 7.4, pve7.3 for 7.3, pve7.2 for 7.2, pve7.1 for 7.1
+  trixie for 9.0+, bookworm for 8.1+, pve8.0 for 8.0, bullseye for 7.4, pve7.3 for 7.3, pve7.2 for 7.2, pve7.1 for 7.1, buster for 6.4
 - Install vitastor-client, pve-qemu-kvm, pve-storage-vitastor (* or see note) packages from Vitastor repository
 - Define storage in `/etc/pve/storage.cfg` (see below)
 - Block network access from VMs to Vitastor network (to OSDs and etcd),

docs/installation/proxmox.ru.md

@@ -6,10 +6,10 @@
 
 # Proxmox VE
 
-Чтобы подключить Vitastor к Proxmox Virtual Environment (поддерживаются версии 7.x-9.x):
+Чтобы подключить Vitastor к Proxmox Virtual Environment (поддерживаются версии 6.4-9.x):
 
 - Добавьте соответствующий Debian-репозиторий Vitastor в sources.list на хостах Proxmox:
-  trixie для 9.0+, bookworm для 8.1+, pve8.0 для 8.0, bullseye для 7.4, pve7.3 для 7.3, pve7.2 для 7.2, pve7.1 для 7.1
+  trixie для 9.0+, bookworm для 8.1+, pve8.0 для 8.0, bullseye для 7.4, pve7.3 для 7.3, pve7.2 для 7.2, pve7.1 для 7.1, buster для 6.4
 - Установите пакеты vitastor-client, pve-qemu-kvm, pve-storage-vitastor (* или см. сноску) из репозитория Vitastor
 - Определите тип хранилища в `/etc/pve/storage.cfg` (см. ниже)
 - Обязательно заблокируйте доступ от виртуальных машин к сети Vitastor (OSD и etcd), т.к. Vitastor (пока) не поддерживает аутентификацию

docs/tests/pg-chaos.en.md

@@ -1,220 +0,0 @@
-[Documentation](../../README.md#documentation) → Tests → The PG chaos-test model
-
------
-
-[Читать на русском](pg-chaos.ru.md)
-
-# The PG chaos-test model
-
-A seed-driven stress test of the PG machinery (`src/test/test_pg_chaos.cpp`): ONE
-real `osd_t` (the primary) with a real blockstore over a mock disk (entirely in
-RAM, `ringloop_mock`) owns a single PG (number 1, pool 1, four 128 K objects, 4 K
-granularity). The other PG members are fake secondary OSDs inside the test. The
-pool is either replicated x3 (members 2,3, spare 4) or EC 2+2 (members 2..4,
-spare 5, one object = one parity stripe). The seed uniquely determines the
-configuration and the whole event sequence; a failing run is reproduced with
-the same `--seed`.
-
-What it mainly verifies is the PG machinery (the subop fan-out,
-peering/LIST/rollback/stabilize, repeer daemons, stop_pg, INCOMPLETE, recovery) and
-partly the primary OSD's storage; checking the storage is not the test's main goal -
-test_blockstore_chaos covers it.
-
-## Configurations (10)
-
-- replicated/EC ×
-- immediate_commit none/all ×
-- csum_block none/4096/16384
-
-`--recovery` enables recovery/rebalance after a set change (off by default so that
-degraded states get exercised).
-
-## The world model
-
-Each object has an ordered version history; a version enters the model at SUBMIT
-time (the delta = the request's exact bytes/length/granule bitmap), with a state:
-- `V_SUBMITTED` - the op was sent and has no answer yet (in flight). It behaves
-  like V_OPTIONAL in every respect but one: concurrent reads are entitled not to
-  show this write's result, and that must not mark the version DEAD.
-- `V_OPTIONAL` - the write was sent but not acknowledged: it either applied (then
-  as a whole) or did not apply at all. Versions reach this state from V_SUBMITTED
-  and, in the explicit-fsync modes, from V_ACKED when the client session dies
-  before a SYNC request (that counts as a marker that the write is unacknowledged).
-  Observing an OPTIONAL write's effect in a later read restores it to ACKED;
-  observing its absence (not visible over its own range) makes it DEAD.
-- `V_ACKED` - the write was acknowledged; the model records `ack_live` (the live set
-  at ack time - used to check that the failure model wasn't exceeded). With
-  immediate_commit=all it is immediately considered on-disk durable. An ACKED write
-  must not roll back while the client's session is alive.
-- `V_SYNCED` - acknowledged and SYNCed: durable on disk, and it may never roll back
-  again at all.
-- `V_PARTIAL` - deleted, but possibly not from every OSD (the PG stayed LEFT_ON_DEAD).
-- `V_DEAD` - the version was either definitely refused (-EAGAIN, possibly -ENOSPC in
-  the future), or it was OPTIONAL and it later became certain that it was refused
-  because a subsequent read did NOT observe it. A DEAD version showing again fails
-  the test.
-
-Object contents are self-identifying: the pattern encodes the object, the version and
-the offset, so a read result decodes back into "which version wrote these bytes".
-
-Verification reads ONLY through the primary OSD and checks that the answer is
-explained by a merge of ONE consistent subset of the written versions, where every
-durable version is definitely present, every refused one is definitely absent, and
-the contents never lie about their authorship. With an incomplete deletion, EC may
-legitimately answer INCOMPLETE, and replicated pools may resurrect the object.
-
-Everything derives from version states and events, with no extra flags: an ACKED
-version cannot vanish except through an event that first breaks the session's
-guarantee - a crash (taking the session down with the process) or a
-repeer/stop_pg (which stop dirty sessions BEFORE peering starts).
-
-The test's checks look ONLY at what the write path exposes to the client; the model
-deliberately ignores things the client cannot know about, such as paired OSD
-deaths per se.
-
-## The simulated reality
-
-- **Disk simulator**: requests carry delays; `power_loss()` = power outage -
-  uncommitted writes are lost, committed ones survive. An unknown FD/opcode aborts
-  the test (a stub guard).
-- **etcd simulator**: keys/values + leases; revoke/TTL are emulated on a crash (the
-  OSD's keys are removed, its lease revoked).
-- **Peer OSD simulator**: answers per the secondary-OSD/blockstore contract -
-  read/write/delete, SYNC, LIST. When a peer OSD dies, non-SYNCed writes die with it
-  with probability `peer_cache_loss`% (default 50 - modelling a volatile disk cache);
-  SYNCed ones always survive. A peer crash by itself does NOT change the world model:
-  the model only knows what the client sees. The crash repeers the PG, the repeer
-  kills dirty client sessions (volatile ACKED writes are exactly what makes a client
-  dirty), the session dies - and THAT is what the client observes
-  (PEER_STOPPED/EPIPE), and only session_died() demotes. If a crash rolled something
-  back under a live session, that would be a rollback invisible to the client -
-  check 5 catches it.
-- **Network simulator**: service_peers runs every round. The request-phase delay:
-  a random `--p-wire`% (default 70) of each peer's queued subops gets applied; the
-  response-phase delay: a random `p_reply`% of the already-applied replies gets
-  delivered. A reply belongs to its connection: the connection died - undelivered
-  replies die with `-EPIPE` (the apply stays). When an OSD instance dies, however,
-  mutating ops still in `sent_ops` are copied into a separate "wire" queue; they are
-  applied later at the `--p-wire` rate WITHOUT an answer (the connection fell).
-  Non-mutating ops (READ/LIST) are dropped (nobody left to answer), and the harvested
-  queue is capped at 64 entries.
-
-## Operations (one per step)
-
-Each step picks an action by weight (`--p-*`; 0 disables it):
-
-- **wait (500)** - send nothing, just run the simulation (event windows).
-- **write (60)** - write a random range of a random object; bump the version; the
-  delta is remembered as OPTIONAL; at most max_inflight (8) in flight.
-- **delete (6)** - delete a random object; the version is remembered as OPTIONAL with
-  zero length (a full delete); LEFT_ON_DEAD/ack are tracked in account().
-- **read (40)** - read an object: whole with probability `full_read`%, otherwise an
-  arbitrary range; a whole read verifies the object.
-- **sync (30)** - flush the cache (fsync); on confirmation the versions are marked
-  SYNCED (durable).
-- **restart (1)** - a CLEAN OSD restart: drain the queue (all inflight finish), wait
-  for the disk to go idle, destroy, recreate, wait for the OSD to come up, then,
-  with `--p-post` (50%) chance, read and verify all object contents ("a clean restart
-  that lost nothing").
-- **takeover (1)** - a PG handover: first stop the PG by pointing the primary at a
-  fake member in the etcd stub -> stop_pg stops the PG; right after the
-  `/pg/state/1/1` key disappears (meaning the OSD released the PG) -> recreate the OSD
-  and point the primary back at it -> re-verify the objects as in restart.
-- **crash (2)** - the primary OSD crashes: a disk power loss (un-synced writes lost),
-  lease cancel + removal of the OSD's keys without cancelling the ops in flight to
-  peer OSDs, a new instance starts, objects re-verified.
-- **peer_down (3)** - a peer OSD crashes: a victim is chosen that breaks no object's
-  quorum (all latest OPTIONAL versions down to the first SYNCED are checked), the peer
-  OSD stops and its connection drops.
-- **peer_up (3)** - a peer OSD comes back (connects - if the PG or its history still
-  involves it); then settle(): wait for the PG to come up (ACTIVE|INCOMPLETE +
-  quiescence), then run recover_pending - deferred checks of objects whose chains may
-  have re-settled.
-- **set_change (1)** - the PG moves to another OSD set through etcd; with
-  `--recovery` - recovery reads; without it, degradation is legal and the
-  "surviving copies" world is protected by the checks.
-
-At the end - wait for completion: clean up until "the PG is healthy and every op has
-finished", at most MAX_STEPS=1e6 steps -> exceeding that aborts with a "did not
-settle" error.
-
-## Op outcome accounting (account)
-
-- Completed with the expected retval -> V_ACKED. A delete without LEFT_ON_DEAD closes
-  the partial-deletion window. A delete with LEFT_ON_DEAD must be repeated after the
-  peers come back, and in EC it allows a future INCOMPLETE unless the object is
-  rewritten in the meantime.
-- Completed with EAGAIN -> V_DEAD: the version was definitely not applied; currently
-  unreachable in the test, kept as a placeholder for the future - it would also apply
-  to e.g. ENOSPC.
-- Completed with EPIPE -> the version stays OPTIONAL: a refusal is not an ack, but the
-  op may have applied (or not). An optional delete must also be repeated after the
-  peers come back and likewise allows EC INCOMPLETE until the next rewrite.
-- Completed with any other error -> the test aborts ("failed with retval").
-
-## Checks (every abort point)
-
-**Data consistency (world_check):**
-1. world_check - the answer must be explained by a subset of the versions: every
-   required (durable) one is present; bytes are answered by exactly their own version
-   ("record obj/seq/off mismatch", a bitmap granule "carries data from wrong
-   version").
-2. "shows %s but the highest..." - an old version of the object was read while later
-   writes are lost (monotonicity broken).
-3. "read shows seq over acked v's granule" - a granule's contents (obj/seq/off in the
-   record) contradict the version that owns the granule in the model: stitched or
-   torn contents instead of the acknowledged bytes.
-4. "DEAD version showed again" - refused versions must not come back.
-5. "committed version read fully absent over floor" / "established version read
-   absent over its granule" - acknowledged ops must not disappear.
-6. A read -EIO "OSD deems unrecoverable" is legal in exactly one window: an
-   unfinished EC deletion. Outside it, -EIO aborts the test - including a "clean" ack
-   delete + INCOMPLETE after a revival (the OSD lied about completeness) and a chain
-   rollback under a live session with no deletion window; any other retval != len
-   aborts the test.
-7. bitmap_len != the expected read answer length -> abort the test.
-8. "came back INCOMPLETE" after set_change/crash - same criterion: an unfinished
-   deletion window makes INCOMPLETE legal and the deletion is repeated; anything
-   else aborts the test.
-9. "came back ... which no world of the recorded versions explains" - the final
-   check of object contents after the test settles.
-10. "changed across a clean restart that lost nothing" - object immutability across
-    a clean restart.
-
-**Liveness/timing:**
-
-11. "did not settle in MAX_STEPS" - the ops never completed within a huge step count
-    (a possible self-deadlock);
-12. "operations did not complete" - stuck ops, printed in the dump; "a read of
-    object N did not complete in MAX_STEPS ... deadlock?"; "OSD did not load PGs
-    from etcd in MAX_STEPS" / "OSD did not finish peering in MAX_STEPS" - start-up
-    checks.
-13. "did not release the PG" - the OSD didn't give the PG up when commanded;
-14. "disk did not go idle" - a use-after-free guard: every disk op must finish
-    before the OSD is destroyed;
-15. "PG did not become healthy after reviving every peer" - the PG stayed down after
-    all its OSDs came back;
-16. "pending recovery/repeated deletion could not complete on a healthy PG" - a
-    deletion/recovery that never finished on a live PG.
-
-**Test-stub guards:**
-
-17. An unexpected SQE (fd/opcode) on the disk stub; "unexpected secondary opcode" on
-    a peer; arguments like `--p-reply >= 1`, `--max-inflight >= 1`, etc.
-
-**Coverage checks:**
-
-18. `peer_cache_loss<100` + a volatile config + >=8 peer deaths -> verify that the
-    variant where a peer OSD kept its journal ran at least once.
-19. >=10 dirty OSD deaths/PG handovers -> wire_harvested > 0: at least one op of the
-    primary OSD that was in flight must have been harvested into the "wire" on the
-    way down - the "delivery from a dead OSD" path was exercised.
-
-## Reporting and reproducibility
-
-The OK line: writes(refused)/deletes/syncs/crashes/restarts/takeovers/peer
-deaths(kept journal)/recoveries/set changes/verified reads/stragglers
-(applied/captured/lost)/disk ops.
-
-Determinism: one seed -> the same configuration and the same schedule; `-v` traces
-disk and peer-OSD operations, op harvesting on peer deaths, and etcd changes.

docs/tests/pg-chaos.ru.md

@@ -1,197 +0,0 @@
-[Документация](../../README-ru.md#документация) → Тесты → Модель хаос-теста PG (`src/test/test_pg_chaos.cpp`)
-
------
-
-[Read in English](pg-chaos.en.md)
-
-# Модель хаос-теста PG
-
-Seed-driven стресс-тест PG-машинерии (`src/test/test_pg_chaos.cpp`): ОДИН настоящий
-`osd_t` (primary) с настоящим хранилищем на заглушке диска (полностью в RAM, `ringloop_mock`)
-владеет одной PG (номер 1, пул 1, объекты — 4 шт по 128 К, гранула 4 К). Остальные члены PG —
-фейковые вторичные OSD внутри теста. PG: реплицированная x3 (members 2,3, spare 4) или
-EC 2+2 (members 2..4, spare 5, объект = один parity-stripe). Сид однозначно задаёт
-конфигурацию и всю последовательность событий; упавший прогон воспроизводится тем же `--seed`.
-
-Проверяется в первую очередь PG-машинерия (subop-веер, peering/LIST/rollback/stabilize, repeer-демоны,
-stop_pg, INCOMPLETE, recovery) и частично хранилище первичного OSD; однако, проверка хранилища —
-не главная цель теста, его покрывает test_blockstore_chaos.
-
-## Конфигурации (10)
-
-- replicated/EC ×
-- immediate_commit none/all ×
-- csum_block нет/4096/16384
-
-`--recovery` включает recovery/rebalance после set-change (по умолчанию выкл, чтобы
-гонять degraded-состояния).
-
-## Модель мира
-
-Для каждого объекта — упорядоченная история версий; версия пишется в модель в момент
-SUBMIT (дельта = точные байты/длина/битмап гранул запроса), состояние:
-- `V_SUBMITTED` — операция отправлена, ответа ещё нет (выполняется). Ведёт себя как
-  V_OPTIONAL во всём, кроме одного: параллельные чтения вправе не показывать результат
-  этой записи, и это не будет маркировать её DEAD.
-- `V_OPTIONAL` — запись отправлена, но не подтверждена, и может быть либо применена
-  (но тогда целиком), либо не применена вообще. Версии попадают в это состояние из
-  V_SUBMITTED и, в режимах с явными fsync, V_ACKED при гибели клиентской сессии до
-  запроса SYNC (это считается маркером, что запись не подтверждена).
-  Наблюдение эффекта OPTIONAL записи в последующем чтении восстанавливает её в ACKED;
-  наблюдение её отсутствия (когда она не видна в своём диапазоне) ⇒ DEAD.
-- `V_ACKED` — запись подтверждена; запоминается `ack_live` (живой набор на момент ack —
-  используется для проверки превышения модели отказов). При immediate_commit=all сразу
-  считается зафиксированной на диске. ACKED запись не должна откатываться, пока сессия
-  клиента жива.
-- `V_SYNCED` — подтверждена и SYNC-нута ⇒ зафиксирована на диске и более не должна
-  откатываться вообще.
-- `V_PARTIAL` — удалена, но, возможно, не со всех OSD (PG осталась в LEFT_ON_DEAD).
-- `V_DEAD` — версия либо точно отказана (-EAGAIN, возможно -ENOSPC в будущем), либо
-  была OPTIONAL, но после этого стало точно известно, что она откачена, так как последующее
-  чтение её уже НЕ наблюдало. Повторное появление DEAD-версии — провал теста.
-
-Содержимое объектов самоидентифицируется версией: паттерн кодирует объект, версию и смещение —
-прочитанное можно декодировать обратно в «какая версия записала эти байты».
-
-Верификация читает ТОЛЬКО через первичный OSD и проверяет, что ответ объясняется слиянием
-ОДНОГО согласованного подмножества записанных версий, где все зафиксированные точно присутствуют,
-все откаченные точно отсутствуют, а содержимое не врёт об авторстве. При неполном удалении
-разрешается INCOMPLETE в EC, а в репликах допускается воскрешение объекта.
-
-Всё выводится из состояний версий и событий, без дополнительных флагов: ACKED не может
-исчезнуть, кроме как через событие, которое сначала ломает гарантию сессии — краш (уносит
-сессию вместе с процессом) или repeer/stop_pg (останавливают dirty-сессии ДО начала пиринга).
-
-Проверки в тесте смотрят ТОЛЬКО на то, как процесс записи видит клиент, и не учитывают
-в модели, например, падения парных OSD сами по себе, так как клиент не может о них знать.
-
-## Симулируемая реальность
-
-- **Симулятор диска**: запросы с задержками; `power_loss()` = потеря питания — незафиксированные
-  записи теряются, зафиксированные остаются. Неизвестный FD/opcode → падение теста (защита заглушки).
-- **Симулятор etcd**: ключ/значение + lease; revoke/TTL эмулируются при крахе (ключи OSD'а
-  удаляются, lease отзывается).
-- **Симулятор парных OSD**: отвечает в соответствии с контрактом вторичного OSD / blockstore.
-  Чтение/запись/удаление, SYNC, LIST. При падении парного OSD не-SYNC-нутые записи гибнут
-  с вероятностью `peer_cache_loss`% (по умолчанию 50 — моделирует волатильный кэш диска),
-  SYNC-нутые выживают всегда. Модель мира на краш пира сама НЕ реагирует: она знает
-  только то, что видит клиент. Краш реперит PG, репер убивает dirty-сессии клиентов
-  (volatile ACKED — это и есть dirty-клиент), сессия гибнет — вот это клиент наблюдает
-  (PEER_STOPPED/EPIPE), и только session_died() делает demote. Если краш откатит
-  что-то при живой сессии — это ненаблюдаемый клиентом откат, проверка 5 его ловит.
-- **Симулятор сети**: service_peers, каждый раунд. Задержка на этапе запроса — применяются случайные
-  `--p-wire`% (по умолчанию 70) очередей subop'ов пира; задержка на этапе ответа — доставляются
-  случайные `p_reply`% применённых ответов. Ответ принадлежит соединению: соединение
-  умерло — недоставленные ответы гибнут с `-EPIPE` (применение остаётся). Однако, при
-  гибели экземпляра OSD мутирующие операции в `sent_ops` копируются в отдельную очередь
-  «сети»; применяются позже (`--p-wire`%) БЕЗ ответа (соединение упало). Немутирующие
-  операции (READ/LIST) отбрасываются (отвечать некому), очередь собранных операций
-  ограничивается 64.
-
-## Операции (step)
-
-На каждом шаге (веса `--p-*`; 0 выключает действие):
-
-- **wait (500)** — ничего не слать, прокрутить симуляцию (событийные окна).
-- **write (60)** — запись случайного диапазона случайного объекта; увеличение версии;
-  дельта запоминается как OPTIONAL; ≤ max_inflight (8) одновременно.
-- **delete (6)** — удаление случайного объекта; версия запоминается как OPTIONAL с
-  нулевой длиной (полное удаление); LEFT_ON_DEAD/ack отслеживаются в account().
-- **read (40)** — чтение объекта, полное с вероятностью `full_read`%, иначе произвольный
-  диапазон; при полном чтении объект проверяется.
-- **sync (30)** — синхронизация кэша (fsync); при подтверждении версии помечаются SYNCED
-  (зафиксированными).
-- **restart (1)** — ЧИСТАЯ перезагрузка OSD: очистка очереди (все inflight закончены), ожидание
-  простоя диска, уничтожение, пересоздание, ожидание подъёма OSD и, с шансом `--p-post`,
-  чтение и проверка содержимого всех объектов («чистый перезапуск, который ничего не потерял»).
-- **takeover (1)** — передача PG: сначала остановка PG через смену первичного OSD на
-  фейковый в заглушке etcd → stop_pg останавливает PG; далее сразу после удаления ключа
-  `/pg/state/1/1` (означающего, что OSD остановил PG) → пересоздание OSD и его переустановка
-  в качестве первичного → перепроверка объектов, как в restart.
-- **crash (2)** — падение первичного OSD: потеря питания диска (потеря не-синкнутых записей),
-  отмена лизы + удаление ключей OSD без отмены операций в полёте к парным OSD, запуск нового
-  экземпляра, перепроверка объектов.
-- **peer_down (3)** — крах парного OSD: выбирается жертва, не ломающая кворум ни
-  одного объекта (проверяются все последние OPTIONAL версии до первой SYNCED), парный OSD
-  останавливается, соединение с ним разрывается.
-- **peer_up (3)** — подъём парного OSD (соединение — если PG или её история его ещё
-  касается); после — settle(): ждать подъёма PG (ACTIVE|INCOMPLETE + покой), затем
-  прогнать recover_pending — отложенные проверки объектов, чьи цепи могли перегласоваться.
-- **set_change (1)** — переезд PG на другой OSD-набор через etcd; с `--recovery` —
-  восстановительные чтения; без — деградация легальна, мир «выживших копий» защищён
-  проверками.
-
-В конце — ожидание завершения: очистка до «PG в порядке и все операции завершены», максимум
-до MAX_STEPS=1e6 шагов → если превышается, то падение с ошибкой «did not settle».
-
-## Учёт результатов операций (account)
-
-- Завершена с ожидаемым кодом возврата → V_ACKED. Удаление без LEFT_ON_DEAD закрывает
-  окно частичного удаления. Удаление с LEFT_ON_DEAD требует повтора после подъёма
-  пиров, а с EC допускает будущий INCOMPLETE, если объект не перезаписан вновь.
-- Завершена с EAGAIN → V_DEAD: версия точно не применена; в тесте сейчас недостижимо,
-  оставлено как заготовка на будущее, также применимо, например, к ENOSPC.
-- Завершена с EPIPE → версия остаётся OPTIONAL: отказ — не ack, но операция могла
-  примениться (а могла и нет). Опциональное удаление также требует повтора после
-  подъёма пиров и тоже разрешает EC INCOMPLETE до следующей перезаписи.
-- Завершена с другой ошибкой → падение теста («failed with retval»).
-
-## Проверки (все точки abort)
-
-**Согласованность данных (world_check):**
-1. world_check — ответ обязан объясняться подмножеством версий: все обязательные
-   (зафиксированные) присутствуют; байты отвечают ровно своей версией
-   («record obj/seq/off mismatch», битмап-гранула «carries data from wrong version»).
-2. «shows %s but the highest…» — прочитана старая версия объекта, последние записи
-   потеряны (монотонность нарушена).
-3. «read shows seq over acked v's granule» — содержимое гранулы (obj/seq/off в record)
-   противоречит версии, которой гранула принадлежит по модели: сборное/порванное
-   содержимое вместо подтверждённых байт.
-4. «DEAD version showed again» — откаченные версии не должны возвращаться.
-5. «committed version read fully absent over floor» / «established version read absent
-   over its granule» — подтверждённые операции не должны исчезать.
-6. read -EIO «OSD deems unrecoverable» — легален ровно в одном окне: незавершённое
-   EC-удаление. Вне его -EIO — падение теста, в т.ч. «чистый» ack удаления + INCOMPLETE
-   после revival (OSD соврал о полноте) и откат цепей при живой сессии без окна удаления;
-   прочие retval ≠ len → падение теста.
-7. bitmap_len ≠ ожидаемому в ответах чтения → падение теста.
-8. «came back INCOMPLETE» после set_change/crash — тот же критерий: окно незавершённого
-   удаления ⇒ INCOMPLETE легален и удаление повторяется; в ином случае → падение теста.
-9. «came back … which no world of the recorded versions explains» — финальная проверка
-   корректности содержимого объектов после завершения теста.
-10. «changed across a clean restart that lost nothing» — неизменность объектов после перезапуска.
-
-**Живость/тайминг:**
-
-11. «did not settle in MAX_STEPS» — операции не завершились за большое число шагов (возможна
-    самоблокировка);
-12. «operations did not complete» — залипшие операции, печатаются в дампе;
-    «a read of object N did not complete in MAX_STEPS ... deadlock?»; «OSD did not load
-    PGs from etcd in MAX_STEPS» / «OSD did not finish peering in MAX_STEPS» — проверка на старте OSD.
-13. «did not release the PG» — OSD не отпустил PG по команде;
-14. «disk did not go idle» — проверка против use-after-free, что все операции диска завершены
-    перед уничтожением OSD;
-15. «PG did not become healthy after reviving every peer» — PG не поднялась после подъёма всех OSD;
-16. «pending recovery/repeated deletion could not complete on a healthy PG» — удаление/восстановление
-    не завершено на живой PG.
-
-**Защиты тестовых заглушек:**
-
-17. Неожиданный SQE (fd/opcode) на заглушке диска; «unexpected secondary opcode» у пира;
-    аргументы `--p-reply ≥ 1`, `--max-inflight ≥ 1` и т.п.
-
-**Проверки покрытия:**
-
-18. `peer_cache_loss<100` + volatile-конфиг + ≥8 смертей пира ⇒ проверка, что хотя бы
-    раз выполнен вариант, в котором парный OSD не потерял журнал.
-19. ≥10 грязных смертей OSD/передач PG ⇒ wire_harvested > 0: хотя бы одна операция
-    первичного OSD, бывшая в полёте, обязана была собраться в «сеть» при остановке —
-    путь «доставки от мёртвого OSD» выполнен.
-
-## Отчёт и воспроизводимость
-
-OK-строка: writes(refused)/deletes/syncs/crashes/restarts/takeovers/peer deaths
-(kept journal)/recoveries/set changes/verified reads/stragglers (applied/captured/lost)/disk ops.
-
-Детерминизм: один seed ⇒ та же конфигурация и то же расписание; `-v` — трасса операций
-диска и парных OSD, захватов операций при падении пиров и изменений etcd.

docs/usage/cli.en.md

@@ -196,7 +196,6 @@ Options can be specified in classic dd style (`key=value`) or like usual (`--key
 | `iimg=<image>`  | Copy from Vitastor image `<image>`                                      |
 | `if=<file>`     | Copy from file `<file>`                                                 |
 | `oimg=<image>`  | Copy to Vitastor image `<image>`                                        |
-| `opool=<pool>`  | Pool to create output image if it doesn't exist                         |
 | `of=<file>`     | Copy to file `<file>`                                                   |
 | `bs=1M`         | Set copy block size                                                     |
 | `count=N`       | Copy only N input blocks. If N ends in B it counts bytes, not blocks    |
@@ -243,16 +242,15 @@ Flatten a layer, i.e. merge data and detach it from parents.
 
 ## rm-data
 
-`vitastor-cli rm-data --pool <id|name> --inode <inode> [--wait-list] [--min-offset <offset>]`
+`vitastor-cli rm-data --pool <pool> --inode <inode> [--wait-list] [--min-offset <offset>]`
 
-Remove inode data without changing metadata. `<inode>` may be decimal or 0x hex.
+Remove inode data without changing metadata.
 
 ```
 --wait-list   Retrieve full objects listings before starting to remove objects.
               Requires more memory, but allows to show correct removal progress.
 --min-offset  Purge only data starting with specified offset.
 --max-offset  Purge only data before specified offset.
-              Offsets are in bytes or with a K/M/G/T unit suffix, decimal or 0x hex.
 --client_wait_up_timeout 16  Timeout for waiting until PGs are up in seconds.
 ```
 
@@ -509,12 +507,15 @@ the Vitastor authorization model.
 
 ## modify-user
 
-`vitastor-cli modify-user --groups <group1,group2,...> <username>`
+`vitastor-cli modify-user --type <type> --groups <group1,group2,...> <username>`
 
 Create or update a user with the given name (matching TLS certificate CN).
 
 Options:
 
+* `--type TYPE` - Set user type: `client` or `admin`. Default is `client`.
+  Administrators (`admin`) have full access to all cluster operations; clients (`client`)
+  have limited access based on ownership and group membership of images.
 * `--groups GROUPS` - Set user's groups (comma-separated). Groups are used together with
   `--owner-group` / `--reader-group` fields of images and `--creator_group` of the pool to
   authorize operations. Pass an empty string to clear the groups.
@@ -533,25 +534,23 @@ Start an HTTP server that handles vitastor-cli commands over a REST API in JSON
 
 Options:
 
-| <!-- -->              | <!-- -->                                                                      |
-|-----------------------|-------------------------------------------------------------------------------|
-| `--bind_address ADDR` | Server IP address(es), separated by space. Default is `127.0.0.1`.            |
-| `--port 8080`         | Server port. Default is `8080`.                                               |
-| `--api_cert FILE`     | Server TLS certificate.                                                       |
-| `--api_pkey FILE`     | Private key for `--api_cert`.                                                 |
-| `--cert FILE`         | Administrator or client certificate used by the server to access the cluster. |
-| `--pkey FILE`         | Private key for `--cert`.                                                     |
+| <!-- -->             | <!-- -->                                                                    |
+|----------------------|-----------------------------------------------------------------------------|
+| `--bind_address ADDR` | Server IP address(es), separated by space. Default is `127.0.0.1`.         |
+| `--port 8080`         | Server port. Default is `8080`.                                            |
+| `--api_cert FILE`     | Path to server TLS certificate file (PEM format). Required for HTTPS.      |
+| `--api_pkey FILE`     | Path to server TLS private key file (PEM format). Required for HTTPS.      |
+| `--client_ca FILE`    | Path to file with TLS CA certificates used to validate client connections. |
 
 When `api_cert`/`api_pkey` are set the server operates in HTTPS mode. If `client_ca`
-is also set (in configuration or on the command line), connecting clients are authenticated by their
-TLS certificates. When [use_perms](../config/security.en.md#use_perms) is enabled, HTTPS with client
-authentication is mandatory and per-user permissions are enforced. For regular clients the server only
-allows operations on images that the user owns or has group access to; all other operations require
-administrator privileges.
+is also set, connecting clients are authenticated by their TLS certificates.
+When [use_perms](../config/security.en.md#use_perms) is enabled, HTTPS with client
+authentication is mandatory and per-user permissions are enforced.
 
 The vitastor-cli process itself must be authenticated as an admin user (its own client
-[cert](../config/security.en.md#cert) must be signed with [admin_ca](../config/security.en.md#admin_ca))
-to be able to serve all requests.
+`cert`/`pkey` must belong to a user of `type=admin`) to be able to serve requests.
+For regular clients the server only allows operations on images that the user owns
+or has group access to; all other operations require administrator rights.
 
 `vitastor-cli serve` exposes a full OpenAPI specification at `/openapi` endpoint.
 Start the server and check it out for the information about available API methods.

docs/usage/cli.ru.md

@@ -203,7 +203,6 @@ vitastor-cli dd [iimg=<image> | if=<file>] [oimg=<image> | of=<file>] [bs=1M] \
 | `iimg=<image>`  | Копировать из образа Vitastor `<image>`                                 |
 | `if=<file>`     | Копировать из файла `<file>`                                            |
 | `oimg=<image>`  | Копировать в образ Vitastor `<image>`                                   |
-| `opool=<pool>`  | Пул для создания выходного образа, если он не существует                |
 | `of=<file>`     | Копировать в файл `<file>`                                              |
 | `bs=1M`         | Задать размер блока копирования                                         |
 | `count=N`       | Копировать не более N блоков. Если N заканчивается на B - то N байт.    |
@@ -252,16 +251,15 @@ vitastor-cli dd [iimg=<image> | if=<file>] [oimg=<image> | of=<file>] [bs=1M] \
 
 ## rm-data
 
-`vitastor-cli rm-data --pool <id|name> --inode <inode> [--wait-list] [--min-offset <offset>]`
+`vitastor-cli rm-data --pool <pool> --inode <inode> [--wait-list] [--min-offset <offset>]`
 
-Удалить данные инода, не меняя метаданные образов. `<inode>` можно задать в десятичном виде или в hex с 0x.
+Удалить данные инода, не меняя метаданные образов.
 
 ```
 --wait-list   Сначала запросить полный листинг объектов, а потом начать удалять.
               Требует больше памяти, но позволяет правильно печатать прогресс удаления.
 --min-offset  Удалять только данные, начиная с заданного смещения.
 --max-offset  Удалять только данные до (исключительно) заданного смещения.
-              Смещения задаются в байтах или с суффиксом K/M/G/T, в десятичном виде или в hex с 0x.
 --client_wait_up_timeout 16  Время ожидания поднятия PG в секундах.
 ```
 
@@ -528,12 +526,15 @@ OSD  PARENT            UP    SIZE  USED%    TAGS          WEIGHT  BLOCK  BITMAP
 
 ## modify-user
 
-`vitastor-cli modify-user --groups <group1,group2,...> <username>`
+`vitastor-cli modify-user --type <type> --groups <group1,group2,...> <username>`
 
 Создать или изменить пользователя с указанным именем (совпадающим с CN TLS-сертификата).
 
 Опции:
 
+* `--type TYPE` - Задать тип пользователя: `client` или `admin`. По умолчанию `client`.
+  Администраторы (`admin`) имеют полный доступ ко всем операциям кластера; клиенты
+  (`client`) имеют ограниченный доступ на основе владения образами и членства в группах.
 * `--groups GROUPS` - Задать группы пользователя (через запятую). Группы используются
   вместе с полями `owner-group` / `reader-group` образов и полем `--creator_group` пулов
   для авторизации операций. Пустая строка очищает список групп.
@@ -552,25 +553,24 @@ OSD  PARENT            UP    SIZE  USED%    TAGS          WEIGHT  BLOCK  BITMAP
 
 Опции:
 
-| <!-- -->              | <!-- -->                                                                         |
-|-----------------------|----------------------------------------------------------------------------------|
-| `--bind_address ADDR` | IP-адрес(а) сервера через пробел. По умолчанию `127.0.0.1`.                      |
-| `--port 8080`         | Порт сервера. По умолчанию `8080`.                                               |
-| `--api_cert FILE`     | TLS-сертификат сервера.                                                          |
-| `--api_pkey FILE`     | Закрытый ключ для сертификата `--api_cert`.                                      |
-| `--cert FILE`         | Сертификат администратора/клиента, используемый сервером для доступа к кластеру. |
-| `--pkey FILE`         | Закрытый ключ для сертификата `--cert`.                                          |
+| <!-- -->              | <!-- -->                                                                     |
+|-----------------------|------------------------------------------------------------------------------|
+| `--bind_address ADDR` | IP-адрес(а) сервера через пробел. По умолчанию `127.0.0.1`.                  |
+| `--port 8080`         | Порт сервера. По умолчанию `8080`.                                           |
+| `--api_cert FILE`     | Путь к файлу TLS-сертификата сервера (в формате PEM). Требуется для HTTPS.   |
+| `--api_pkey FILE`     | Путь к файлу закрытого TLS-ключа сервера (PEM). Требуется для HTTPS.         |
+| `--client_ca FILE`    | Путь к файлу с TLS CA-сертификатами для проверки подключений клиентов.       |
 
 Если заданы `api_cert`/`api_pkey`, сервер работает в режиме HTTPS. Если дополнительно
-задан `client_ca` (в файле конфигурации или в командной строке), подключающиеся клиенты аутентифицируются
-по своим TLS-сертификатам. Если включена опция [use_perms](../config/security.ru.md#use_perms), HTTPS с
-аутентификацией клиентов обязателен, и права доступа проверяются для каждого пользователя. Обычным клиентам
-сервер разрешает только операции с образами, которыми они владеют или к которым имеют доступ через группы;
-остальные операции разрешены только администраторам.
+задан `client_ca`, подключающиеся клиенты аутентифицируются по своим TLS-сертификатам.
+Если включена опция [use_perms](../config/security.ru.md#use_perms), HTTPS с
+аутентификацией клиентов обязателен, и права доступа проверяются для каждого пользователя.
 
-Сам процесс vitastor-cli при этом должен использовать сертификат администратора ([cert](../config/security.en.md#cert),
-подписанный [admin_ca](../config/security.en.md#admin_ca)), чтобы иметь возможность обслуживать
-все виды запросов.
+Сам процесс vitastor-cli при этом должен использовать сертификат (`cert`+`pkey`)
+пользователя с правами администратора (`type=admin`), чтобы иметь возможность
+обслуживать запросы. Обычным клиентам сервер разрешает только операции с образами,
+которыми они владеют или к которым имеют доступ через группы; остальные операции
+разрешены только администраторам.
 
 `vitastor-cli serve` выставляет полную OpenAPI-спецификацию по адресу `/openapi`.
 Запустите сервер и посмотрите её для подробной информации о доступных методах API.

docs/usage/nfs.en.md

@@ -8,27 +8,15 @@
 
 Vitastor has two file system implementations. Both can be used via `vitastor-nfs`.
 
+Commands:
+- [mount](#mount)
+- [start](#start)
+- [upgrade](#upgrade)
+- [defrag](#defrag)
+
 ⚠️ Important: follow the instructions from [Linux NFS write size](#linux-nfs-write-size)
 for optimal Vitastor NFS performance if you use EC and HDD and mount your NFS from Linux.
 
-- [Pseudo-FS](#pseudo-fs)
-- [VitastorFS](#vitastorfs)
-  - [Supported POSIX features](#supported-posix-features)
-  - [Limitations](#limitations)
-- [Linux NFS](#linux-nfs)
-  - [Sync vs async safety](#sync-vs-async-safety)
-  - [Linux NFS readahead](#linux-nfs-readahead)
-  - [Linux NFS write size](#linux-nfs-write-size)
-  - [Horizontal scaling](#horizontal-scaling)
-- [RDMA](#rdma)
-- [Commands](#commands)
-  - [mount](#mount)
-  - [start](#start)
-  - [upgrade](#upgrade)
-  - [defrag](#defrag)
-- [Mounting from /etc/fstab](#mounting-from-etc-fstab)
-- [Common options](#common-options)
-
 ## Pseudo-FS
 
 Simplified pseudo-FS proxy is used for file-based image access emulation. It's not
@@ -54,7 +42,7 @@ vitastor-nfs start --block --etcd_address 192.168.5.10:2379 --portmap 0 --port 2
 To mount the FS exported by this server, run:
 
 ```
-mount server:/ /mnt/ -o port=2050,mountport=2050,nfsvers=3,soft,sync,nolock,tcp
+mount server:/ /mnt/ -o port=2050,mountport=2050,nfsvers=3,soft,nolock,tcp
 ```
 
 ## VitastorFS
@@ -117,57 +105,7 @@ Other notable missing features which should be addressed in the future:
   in the DB. The FS is implemented is such way that this garbage doesn't affect its
   function, but having a tool to clean it up still seems a right thing to do.
 
-## Linux NFS
-
-Linux NFS client has some specific features of behaviour important for Vitastor:
-
-### Sync vs async safety
-
-Linux NFS has 2 options which affect the mount mode: **soft/hard** and **sync/async**.
-
-**⚠️ Warning: soft,async mode is dangerous and may lead to data loss!**
-
-How these options work:
-
-- In the **hard** mode all requests to the server are retried infinitely. If the NFS server becomes
-  unavailable, it becomes impossible to unmount the FS until you reboot the client host.
-- In the **soft** each request is only tried 2 times (default **retrans** option value for TCP connections).
-  In this case, unmounting a dead FS is possible.
-- In the **sync** mode all modification requests are sent to the server immediately. This is slower,
-  but allows the client to see all errors if they happen.
-- In the **async** mode, modification requests are accumulated and sent to the server in the background
-  mode in parallel. It is faster, but it can hide write errors from the client. Instead of receiving an
-  error from a write operation, the client will only get an error when he closes the file (unsuccessful
-  close() call). Another problem is that there's no convenient way to set write parallelism for NFS, so
-  the client can overload the server.
-
-That's exactly why **soft,async** is dangerous. When a large file is copied, the server is easily overwhelmed
-with write requests and starts to respond slowly. Write requests time out, soft mode retries them only 1 time
-and discards them, which results in parts of the file being lost.
-
-So to recap: always use either **soft,sync** mode, or **hard,async**. By default, `vitastor-nfs mount` uses
-**soft,sync** if mode options are not specified.
-
-### Linux NFS readahead
-
-By default, Linux uses rather small readahead value of 128 KB for NFS mounts.
-
-For simple reading/copuing of files it's way too low because it stops VitastorFS from
-using parallel reads.
-
-However, you can increase readahead. You have 3 options:
-
-1. Use the built-in `vitastor-nfs mount --readahead <SIZE>` option (for example, `--readahead 2M`).
-2. Edit `[nfsrahead]` section in `/etc/nfs.conf` - add `nfs=2048` in it (global setting for all NFS 3.0 mounts).
-3. Manually with the following command: \
-   `echo 2048 > /sys/class/bdi/$(findmnt -no MAJ:MIN /mnt/nfs 2>/dev/null | tr -d ' ')/read_ahead_kb`
-
-Here, `2048` is the desired readahead in KB, and `/mnt/nfs` is your FS mountpoint.
-
-You can easily see 2x-3x linear read improvement after you increase the parameter, so we
-really urge you to try increasing it!
-
-### Linux NFS write size
+## Linux NFS write size
 
 Linux NFS client (nfs/nfsv3/nfsv4 kernel modules) has a hard-coded maximum I/O size,
 currently set to 1 MB - see `rsize` and `wsize` in [man 5 nfs](https://linux.die.net/man/5/nfs).
@@ -223,7 +161,7 @@ modprobe nfsv3
 After these (not much complicated 🙂) manipulations NFS begins to be mounted
 with new wsize and rsize by default and it fixes Vitastor-NFS linear write performance.
 
-### Horizontal scaling
+## Horizontal scaling
 
 Linux NFS 3.0 client doesn't support built-in scaling or failover, i.e. you can't
 specify multiple server addresses when mounting the FS.
@@ -262,7 +200,6 @@ Use regular `umount <MOUNTPOINT>` to unmount the FS.
 The server will be automatically stopped when the FS is unmounted.
 
 - `-o|--options <OPT>` - Pass additional NFS mount options (ex.: -o async).
-- `--readahead <SIZE>` - Set Linux FS readahead to `<SIZE>`. Default is 128 KB, larger values increase sequential read speed (for example, try `--readahead 2M`).
 
 ### start
 
@@ -312,31 +249,6 @@ Options:
 | `--include-empty`          | Include old and empty volumes; make sure to restart NFS servers before using it |
 | `--no-rm`                  | Move, but do not delete data                                            |
 
-## Mounting from /etc/fstab
-
-Vitastor packages include mount helpers for `/etc/fstab`: **/usr/sbin/mount.vitastorfs** and **/usr/sbin/mount.vitastorblk**.
-
-Both are simple symbolic links to `vitastor-nfs` and allow to handle `mount` calls correctly.
-
-Example `/etc/fstab` entries:
-
-```
-myfs /S vitastorfs pool=fspool,readahead=2M,--etcd_address=10.0.2.15:2379,soft,sync 0 0
-none /mnt/vitastor-block vitastorblk pool=ecpool 0 0
-```
-
-When you run `mount /S`, the helper is called internally and starts VitastorFS.
-Helper call syntax is:
-
-```
-mount.vitastorfs <FSNAME> <MOUNTPOINT> -o <OPTS>
-```
-
-`<OPTS>` is comma-separated and includes all mount options from fstab. You can include
-both Vitastor and NFS mount options in it. An option is treated as an NFS option if it
-does not start with `--`, `pool=`, `readahead=` or if it doesn't contain `=` or `_`.
-Other options are treated as Vitastor options.
-
 ## Common options
 
 | <!-- -->           | <!-- -->                                                 |

docs/usage/nfs.ru.md

@@ -8,27 +8,15 @@
 
 В Vitastor есть две реализации файловой системы. Обе используются через `vitastor-nfs`.
 
+Команды:
+- [mount](#mount)
+- [start](#start)
+- [upgrade](#upgrade)
+- [defrag](#defrag)
+
 ⚠️ Важно: для оптимальной производительности Vitastor NFS в Linux при использовании
 HDD и EC (erasure кодов) выполните инструкции из раздела [Размер записи Linux NFS](#размер-записи-linux-nfs).
 
-- [Псевдо-ФС](#псевдо-фс)
-- [VitastorFS](#vitastorfs)
-  - [Поддерживаемые функции POSIX](#поддерживаемые-функции-posix)
-  - [Ограничения](#ограничения)
-- [NFS в Linux](#nfs-в-linux)
-  - [Безопасность режимов sync и async](#безопасность-режимов-sync-и-async)
-  - [Упреждающее чтение Linux NFS](#упреждающее-чтение-linux-nfs)
-  - [Размер записи Linux NFS](#размер-записи-linux-nfs)
-  - [Горизонтальное масштабирование](#горизонтальное-масштабирование)
-- [RDMA](#rdma)
-- [Команды](#команды)
-  - [mount](#mount)
-  - [start](#start)
-  - [upgrade](#upgrade)
-  - [defrag](#defrag)
-- [Монтирование из /etc/fstab](#монтирование-из-etc-fstab)
-- [Общие опции](#общие-опции)
-
 ## Псевдо-ФС
 
 Упрощённая реализация псевдо-ФС используется для эмуляции файлового доступа к блочным
@@ -54,7 +42,7 @@ vitastor-nfs start --block --etcd_address 192.168.5.10:2379 --portmap 0 --port 2
 Примонтировать ФС, запущенную с такими опциями, можно следующей командой:
 
 ```
-mount server:/ /mnt/ -o port=2050,mountport=2050,nfsvers=3,soft,sync,nolock,tcp
+mount server:/ /mnt/ -o port=2050,mountport=2050,nfsvers=3,soft,nolock,tcp
 ```
 
 ## VitastorFS
@@ -121,58 +109,7 @@ JSON-формате :-). Для инспекции содержимого БД
   записи. ФС устроена так, что на работу они не влияют, но для порядка и их стоит
   уметь подчищать.
 
-## NFS в Linux
-
-NFS в Linux имеет некоторые важные для Vitastor особенности поведения:
-
-### Безопасность режимов sync и async
-
-В Linux NFS есть 2 опции, влияющие на режим монтирования: **soft/hard** и **sync/async**.
-
-**⚠️ Внимание: сочетание soft,async опасно и может приводить к потере данных!**
-
-Логика работы опций:
-
-- В режиме **hard** все запросы к серверу повторяются бесконечно. Если NFS-сервер становится недоступен,
-  отмонтировать ФС становится невозможно до перезагрузки клиентской машины.
-- В режиме **soft** каждый запрос повторяется только 2 раза (значение опции **retrans** по умолчанию для
-  TCP соединений). В этом случае отмонтировать умершую ФС можно.
-- В режиме **sync** все запросы изменения отправляются на сервер сразу. Это медленнее, но зато любую
-  ошибку записи клиент видит сразу.
-- В режиме **async** запросы изменения накапливаются и отправляются на сервер в фоновом режиме, параллельно.
-  Это быстрее, но ошибки записи могут быть незаметны для клиента и вместо ошибки записи клиент может
-  получить ошибку в момент закрытия файла (вызова close()). Помимо этого, в NFS нет нормального способа
-  регулировать параллелизм записи, поэтому клиент может легко перегружать сервер большим количеством команд.
-
-Именно поэтому режим **soft,async** опасен. При копировании любого большого файла сервер перегружается и
-начинает медленно отвечать. Команды записи заканчиваются таймаутом, soft режим делает 1 повторную попытку,
-после чего забывает команды совсем. Часть блоков файла оказывается просто потерянной.
-
-Резюме: всегда используйте либо режим **soft,sync**, либо **hard,async**. Если опции `vitastor-nfs mount`
-не указаны, режим по умолчанию - **soft,sync**.
-
-### Упреждающее чтение Linux NFS
-
-По умолчанию Linux использует довольно небольшой размер упреждающего чтения для
-файловых систем NFS - 128 КБ.
-
-Для простого чтения/копирования файлов его не достаточно, потому что оно мешает VitastorFS
-нормально утилизировать параллелизм чтения.
-
-Однако, вы можете увеличить эту настройку целыми 3 способами:
-
-1. Встроенной опцией `vitastor-nfs mount --readahead <SIZE>` (например, `--readahead 2M`).
-2. Добавив в секцию `[nfsrahead]` файла конфигурации `/etc/nfs.conf` строку `nfs=2048` (глобальная
-   настройка для всех NFS 3.0 точек монтирования).
-3. Вручную следующей командой: \
-   `echo 2048 > /sys/class/bdi/$(findmnt -no MAJ:MIN /mnt/nfs 2>/dev/null | tr -d ' ')/read_ahead_kb`
-
-Здесь `2048` - размер readahead, а `/mnt/nfs` - путь к вашей ФС.
-
-После увеличения параметра скорость линейного чтения легко может увеличиться в 2-3 раза, так что
-советуем обязательно попробовать!
-
-### Размер записи Linux NFS
+## Размер записи Linux NFS
 
 Клиент Linux NFS (модули ядра nfs/nfsv3/nfsv4) имеет фиксированный в коде максимальный
 размер запроса ввода-вывода, равный 1 МБ - см. `rsize` и `wsize` в [man 5 nfs](https://linux.die.net/man/5/nfs).
@@ -232,7 +169,7 @@ modprobe nfsv3
 монтироваться с новыми wsize и rsize, и производительность линейной записи в Vitastor-NFS
 исправляется.
 
-### Горизонтальное масштабирование
+## Горизонтальное масштабирование
 
 Клиент Linux NFS 3.0 не поддерживает встроенное масштабирование или отказоустойчивость.
 То есть, вы не можете задать несколько адресов серверов при монтировании ФС.
@@ -272,9 +209,6 @@ VitastorFS из GPUDirect.
 Сервер автоматически останавливается при отмонтировании ФС.
 
 - `-o|--options <OPT>` - Передать дополнительные опции монтирования NFS (пример: -o async).
-- `--readahead <SIZE>` - Устанавливает размер упреждающей записи ФС Linux в `<SIZE>`.
-  Значение по умолчанию 128 KB, бОльшие значения значительно улучшают скорость линейного чтения.
-  Попробуйте, например, `--readahead 2M`.
 
 ### start
 
@@ -327,32 +261,6 @@ VitastorFS из GPUDirect.
 | `--include-empty`          | Дефрагментировать старые и пустые тома; обязательно перезапустите NFS-сервера после использования этой опции |
 | `--no-rm`                  | Перемещать, но не удалять данные                                        |
 
-## Монтирование из /etc/fstab
-
-Пакеты Vitastor включают хелперы для монтирования из `/etc/fstab`: **/usr/sbin/mount.vitastorfs** и **/usr/sbin/mount.vitastorblk**.
-
-Оба являются простыми символическими ссылками на `vitastor-nfs` и дают возможность
-корректно обрабатывать вызовы монтирования утилитой `mount`.
-
-Примеры записей `/etc/fstab`:
-
-```
-myfs /S vitastorfs pool=fspool,readahead=2M,--etcd_address=10.0.2.15:2379,soft,sync 0 0
-none /mnt/vitastor-block vitastorblk pool=ecpool 0 0
-```
-
-Когда вы запускаете `mount /S`, внутри вызывается хелпер и запускает VitastorFS.
-Синтаксис вызова самого хелпера:
-
-```
-mount.vitastorfs <FSNAME> <MOUNTPOINT> -o <OPTS>
-```
-
-`<OPTS>` - это набор опций, разделённых запятой, и в нём передаются все опции
-монтирования из fstab. В него можно включать и опции Vitastor, и опции NFS. Опция
-считается опцией NFS, если она не начинается с `--`, `pool=` и `readahead=`, а также
-если она не содержит `=` или `_`. Все прочие опции считаются опциями Vitastor.
-
 ## Общие опции
 
 | <!-- -->           | <!-- -->                                                |

docs/usage/security.en.md

@@ -75,11 +75,13 @@ For a quick setup, use the `/usr/lib/vitastor/mon/make-etcd` script:
 3. Run `/usr/lib/vitastor/mon/make-etcd` without parameters or with the `--antietcd-only`
    parameter if you want to initialize the cluster with Antietcd only, without etcd.
 4. The script will generate all necessary certificates and offer to copy them to the other
-   monitor nodes. Root ssh access to these nodes is required to do it, so agree if it's
-   available.
-5. If you don't have root ssh access from the first node to other nodes, copy configuration
-   manually (vitastor.conf and all certificates/keys, except CA `*_ca.key`) and call
-   `/usr/lib/vitastor/mon/make-etcd` on every node.
+   monitor nodes (agree!).
+5. Log in to all other monitor nodes and repeat the `/usr/lib/vitastor/mon/make-etcd` call there.
+6. If you also have nodes with OSDs only (without monitors), run the following command to
+   copy only the required configuration to these nodes:
+   ```
+   /usr/lib/vitastor/mon/make-etcd --copy-to-osd osdnode1,osdnode2,...
+   ```
 
 After that, you can proceed with OSD initialization.
 
@@ -160,28 +162,26 @@ When OSD transport encryption is enabled (at least for headers), you can enable
 control by turning on the `use_perms=true` option. When this option is enabled, each user
 (and even an OSD or a monitor) can perform only operations allowed for him.
 
-Each user, administrator, OSD or monitor must have their own certificate signed by a specific
-root certificate:
-- `client_ca` for regular users. Common Name of the certificate must be equal to the user's
-  name. Regular users can only read and modify explicitly permitted images and can't modify
-  cluster state.
-- `admin_ca` for administrators. Administrators can read and modify all images, and also
-  adminster the cluster: view overall statistics and status, create and delete OSDs, etc.
-  Common Name is not checked for admin users.
-- `osd_ca` for OSDs and `mon_ca` for monitors (if Antietcd runs separately). OSDs and monitors
-  also have distinct permission sets.
+Each user (or administrator) must have their own certificate signed by a common root
+certificate for clients (`client_ca`), with a Common Name equal to the user name.
+Privilege settings are stored in etcd. OSDs and monitors don't need user accounts;
+they authenticate via separate certificates.
 
-User privileges for regular users are stored in etcd data and currently consist of only
-one property in `/vitastor/config/user/<name>`:
-- `groups` - List of group names the user is a member of.
+User privileges are stored in etcd data under the keys `/vitastor/config/user/<name>`.
+The following is defined per user in this key:
+- Type:
+  - Client (`type=client` or omitted) — can only read and modify explicitly permitted images.
+  - Administrator (`type=admin`) — can read and modify all images, and also administer the
+    cluster: view overall statistics and status, create and delete OSDs, etc.
+- List of group names the user is a member of.
 
 Images have the following properties:
-- Owner (`owner`) — the user name that is allowed to both read and modify the image
-- Owner group (`owner_group`) — the owner group name
-- Reader group (`reader_group`) — the name of the group of users allowed to read the image
+- Owner (owner) — the user name that is allowed to both read and modify the image
+- Owner group (owner_group) — the owner group name
+- Reader group (reader_group) — the name of the group of users allowed to read the image
 
 And there is also a property on the pool:
-- Creator group (`creator_group`) — the name of the group of users allowed to create images in the pool
+- Creator group (creator_group) — the name of the group of users allowed to create images in the pool
 
 For the list of allowed operations on image data on the OSD side, see the
 [OSD data access rights](#osd-data-access-rights) section.
@@ -212,17 +212,15 @@ For the list of allowed operations with etcd data, see the
 
 ### Configuring OSD transport encryption
 
-You need 3 certificates: one for OSDs, one for signing client certificates and one for
-signing administrator certificates. For OSDs, you can use a self-signed certificate (osd_ca.crt)
-or a separate certificate (osd.crt) signed by a trusted osd_ca.crt certificate. For clients,
-you must use separate certificates signed by a common CA certificate (client_ca.crt). One more
-CA certificate (admin_ca.crt) is required for signing administrator certificates.
+You need 2 certificates: one for OSDs and one for signing all client certificates.
+For OSDs, you can use a self-signed certificate (osd_ca.crt) or a separate certificate (osd.crt)
+signed by a trusted osd_ca.crt certificate. For clients, you must use separate certificates
+signed by a common trusted (client_ca.crt).
 
 Add to the Vitastor configuration on OSD servers:
 - use_perms: true
 - osd_ca: osd_ca.crt
 - client_ca: client_ca.crt
-- admin_ca: admin_ca.crt
 - osd_cert: osd_ca.crt
 - osd_pkey: osd_ca.key
 
@@ -237,7 +235,8 @@ The following configuration options are available:
 
 #### Mon with embedded Antietcd
 
-The simplest option. Apart from OSD certificates, you only need 1 certificate for Antietcd (antietcd.crt).
+The simplest option. You need 1 certificate for Antietcd (antietcd.crt), plus root
+certificates for OSDs and clients.
 
 Vitastor settings (`/etc/vitastor/vitastor.conf`):
 - etcd_address: [ "http://mon1:2379", ... ] (addresses of your monitors with port 2379)
@@ -248,7 +247,6 @@ Vitastor settings (`/etc/vitastor/vitastor.conf`):
 - etcd_ca: antietcd.crt
 - osd_ca: osd_ca.crt
 - client_ca: client_ca.crt
-- admin_ca: admin_ca.crt
 
 #### Mon as an Etcd proxy
 
@@ -256,6 +254,7 @@ If you want to enable privileges, but stay on etcd, you can use etcd proxy mode.
 
 You will need 2 separate certificates: one for etcd (etcd.crt) and one for antietcd (antietcd.crt).
 The etcd client port must be different from the standard 2379 — for example, you can pick 2381.
+OSD and client certificates are also needed.
 
 Vitastor settings:
 - etcd_address: [ "http://mon1:2379", ... ] (addresses of your monitors with port 2379)
@@ -275,7 +274,6 @@ Vitastor settings:
 - etcd_ca: antietcd.crt
 - osd_ca: osd_ca.crt
 - client_ca: client_ca.crt
-- admin_ca: admin_ca.crt
 
 etcd command-line options:
 ```
@@ -286,11 +284,11 @@ etcd command-line options:
 
 #### Mon with a separate Antietcd Proxy
 
-If in addition to the previous option you want to offload Vitastor monitor's tasks from Antietcd,
-you can run it separately.
+If in addition to the previous option you want to offload Antietcd from the Vitastor monitor's
+tasks, you can run it separately.
 
 Similar to the previous option, 2 certificates are needed: one for etcd and one for antietcd,
-plus a separate certificate for the monitor.
+plus separate certificates for clients, OSDs, and monitors will be needed.
 
 Vitastor settings:
 - etcd_address: [ "http://mon1:2379", ... ] (addresses of your monitors with port 2379)
@@ -299,18 +297,15 @@ Vitastor settings:
 - etcd_ca: antietcd.crt
 - osd_ca: osd_ca.crt
 - client_ca: client_ca.crt
-- admin_ca: admin_ca.crt
 - mon_etcd_client_cert: mon_ca.crt
 - mon_etcd_client_key: mon_ca.key
 
 Antietcd command-line options:
 ```
---port 2379 --cert /etc/vitastor/antietcd.crt --key /etc/vitastor/antietcd.key \
---client_cert_auth 1 --auth_filter /usr/lib/vitastor/mon/vitastor_auth_filter.js \
---ca /etc/vitastor/client_ca.crt --admin_ca /etc/vitastor/admin_ca.crt \
---osd_ca /etc/vitastor/osd_ca.crt --mon_ca /etc/vitastor/mon_ca.crt \
---etcd_proxy url1,url2,... --etcd_ca /etc/vitastor/etcd.crt \
---etcd_cert /etc/vitastor/antietcd.crt --etcd_key /etc/vitastor/antietcd.key
+--port 2379 \
+--client_cert_auth 1 --auth_filter vitastor_auth_filter.js --etcd_proxy url1,url2,... \
+--cert antietcd.crt --key antietcd.key --ca client_ca.crt --osd_ca osd_ca.crt --mon_ca mon_ca.crt \
+--etcd_cert antietcd.crt --etcd_key antietcd.key --etcd_ca etcd.crt
 ```
 
 etcd command-line options (same as in the previous option):
@@ -331,17 +326,15 @@ Vitastor settings (same as in the previous option):
 - etcd_ca: antietcd.crt
 - osd_ca: osd_ca.crt
 - client_ca: client_ca.crt
-- admin_ca: admin_ca.crt
 - mon_etcd_client_cert: mon_ca.crt
 - mon_etcd_client_key: mon_ca.key
 
 Antietcd command-line options:
 ```
---port 2379 --cert /etc/vitastor/antietcd.crt --key /etc/vitastor/antietcd.key \
---client_cert_auth 1 --auth_filter /usr/lib/vitastor/mon/vitastor_auth_filter.js \
---ca /etc/vitastor/client_ca.crt --admin_ca /etc/vitastor/admin_ca.crt \
---osd_ca /etc/vitastor/osd_ca.crt --mon_ca /etc/vitastor/mon_ca.crt \
---persist_filter /usr/lib/vitastor/mon/vitastor_persist_filter.js
+--port 2379 \
+--client_cert_auth 1 --auth_filter vitastor_auth_filter.js \
+--persist_filter vitastor_persist_filter.js \
+--cert antietcd.crt --key antietcd.key --ca client_ca.crt --osd_ca osd_ca.crt --mon_ca mon_ca.crt
 ```
 
 ### Vault/OpenBao setup
@@ -462,7 +455,7 @@ Step-by-step instructions for setting up a test Vault using OpenBao as an exampl
 
 Below, all key names are given without the common prefix `/vitastor`.
 
-Allowed operations with keys in Antietcd for regular users:
+Allowed operations with keys in Antietcd for clients (`type=client`):
 - Read-only:
   - Always allowed:
     - `/config/global`
@@ -483,7 +476,7 @@ Allowed operations with keys in Antietcd for regular users:
     - `/config/inode/*`
     - `/index/image/*`
 
-Allowed operations with keys in Antietcd for administrators:
+Allowed operations with keys in Antietcd for administrators (`type=admin`):
 - Read:
   - `/stats`
   - `/mon/*`
@@ -495,10 +488,7 @@ Allowed operations with keys in Antietcd for administrators:
   - `/config/*`
   - `/osd/*`
   - `/index/*`
-  - `/pg/config`
   - `/pg/history/*`
-  - `/stats`
-  - `/history/last_clean_pgs`
 
 Allowed operations with keys in etcd for OSDs:
 - Read:
@@ -521,8 +511,6 @@ Allowed operations with keys in etcd for monitors:
   - `/history/last_clean_pgs`
   - `/mon/*`
   - `/pg/history/*`
-  - `/pg/stats/*`
-  - `/pgstats/*`
   - `/inode/stats/*`
   - `/pool/stats/*`
 
@@ -559,21 +547,20 @@ Cluster operations — allowed only for other OSDs:
 
 [vitastor-cli serve](../usage/cli.en.md#serve) also supports client authentication
 via certificates. Only certificates signed by `client_ca` are accepted. A separate
-certificate `api_cert` with the key `api_pkey` is used as the server certificate.
+certificate `server_cert` with the key `server_pkey` is used as the server certificate.
 
-For `vitastor-cli serve` to work correctly, it must use a certificate of an administrator
-(`cert`+`pkey` signed by `admin_ca`) to access Vitastor and serve all requests correctly.
+For `vitastor-cli serve` to work correctly, it itself must use a certificate
+(`cert`+`pkey`) of a user with administrator rights (`type=admin`) to access Vitastor.
 
-Regular users (with `client_ca`-signed certificates), when accessing the API, are only
-allowed operations on images available to them either for reading (for read-only operations)
-or for writing (for modification):
+Regular clients (users with `type=client`), when accessing the API, are only allowed API operations on images
+available to them either for reading (for read-only operations) or for writing (for modification):
 
 - image/list — for images the user can read.
 - image/create — for pools in which the user is allowed to create images, or for
   creating snapshots of images owned by the user.
 - image/delete, image/flatten, image/modify — for images owned by the user.
 
-All other operations are allowed only for administrators.
+All other operations are allowed only for administrators (`type=admin`).
 
 ## Encryption performance
 

docs/usage/security.ru.md

@@ -75,11 +75,13 @@
 3. Запустите `/usr/lib/vitastor/mon/make-etcd` без параметров или с параметром `--antietcd-only`,
    если хотите инициализировать кластер только с Antietcd без etcd.
 4. Скрипт сгенерирует все необходимые сертификаты и предложит скопировать их на остальные узлы
-   мониторов. Для этого нужен root ssh доступ к этим узлам. Если он есть - соглашайтесь, скрипт
-   автоматически скопирует конфигурацию и инициализирует все etcd и мониторы.
-5. Если root ssh доступа с первого узла на остальные узлы мониторов нет - скопируйте на них конфигурацию
-   вручную (vitastor.conf и все сертификаты и ключи, кроме ключей CA `*_ca.key`) и повторите на каждом
-   узле вызов `/usr/lib/vitastor/mon/make-etcd`.
+   мониторов (соглашайтесь!).
+5. Зайдите на все остальные узлы мониторов и повторите там вызов `/usr/lib/vitastor/mon/make-etcd`.
+6. Если у вас будут узлы только с OSD без мониторов, выполните следующую команду, чтобы скопировать
+   только нужную конфигурацию на эти узлы:
+   ```
+   /usr/lib/vitastor/mon/make-etcd --copy-to-osd osdnode1,osdnode2,...
+   ```
 
 После этого можете переходить к инициализации OSD.
 
@@ -129,7 +131,7 @@
 Ключи шифрования образов могут храниться либо в etcd, либо во внешнем Vault. Во втором случае
 в etcd хранятся только ID ключей, а Vitastor вообще не имеет доступа к данным образов. Для
 использования Vault нужно создать образ с опцией `--enc_key vault:ID`, в конфигурации указать
-опции vault_url и vault_ca, создать всем клиентам учётные записи в Vault и дать им доступ
+опции vault_url, и vault_ca, создать всем клиентам учётные записи в Vault и дать им доступ
 к требуемым секретам v1.
 
 Ещё раз повторимся, что если AES-XTS используется с полным шифрованием трафика (`proto_checksums=gcm`),
@@ -160,28 +162,27 @@
 контроль доступа, включив опцию `use_perms=true`. При включённой опции каждый пользователь
 (и даже OSD и монитор) может выполнять только те операции, которые ему разрешены.
 
-Каждый обычный пользователь, администратор, OSD или монитор должен иметь свой сертификат,
-подписанный отдельным корневым сертификатом:
-- Для обычных пользователей - `client_ca`. Common Name сертификата при этом должен быть равен
-  имени пользователя. Обычные пользователи могут только читать и модифицировать явным образом
-  разрешённые им образы и не могут менять общее состояние кластера.
-- Для администраторов - `admin_ca`. Администраторы могут читать и модифицировать все образы,
-  а также администрировать кластер: смотреть общую статистику и состояние, создавать и удалять
-  OSD и так далее.
-- Для OSD - `osd_ca`, для мониторов - `mon_ca` (если Antietcd запускаются отдельно от мониторов).
-  OSD и мониторы тоже имеют свои наборы разрешённых операций.
+Каждый пользователь (или администратор) должен иметь свой сертификат, подписанный общим
+корневым сертификатом для клиентов (`client_ca`), с Common Name, равным имени пользователя.
+Настройки привилегий же хранятся в etcd. Для OSD и мониторов учётные записи не нужны,
+они аутентифицируются по отдельным сертификатам.
 
-Настройки привилегий для обычных пользователей хранятся в etcd и на данный момент состоят
-всего из одного свойства в ключах `/vitastor/config/user/<имя>`:
-- `groups` - Список имён групп, членом которых пользователь является.
+Привилегии пользователей хранятся в данных etcd в ключах `/vitastor/config/user/<имя>`.
+В этом ключе для каждого пользователя задаётся:
+- Тип:
+  - Клиент (`type=client` или не указано) - может читать и модифицировать только явным образом
+    разрешённые образы.
+  - Администратор (`type=admin`) - может читать и модифицировать все образы, а также администрировать
+    кластер: смотреть общую статистику и состояние, создавать и удалять OSD и так далее.
+- Список имён групп, членом которых пользователь является.
 
 У образов есть следующие свойства:
-- Владелец (`owner`) - имя пользователя, которому разрешено и читать, и менять образ
-- Группа владельцев (`owner_group`) - имя группы владельцев
-- Группа читателей (`reader_group`) - имя группы пользователей, которым разрешено читать образ
+- Владелец (owner) - имя пользователя, которому разрешено и читать, и менять образ
+- Группа владельцев (owner_group) - имя группы владельцев
+- Группа читателей (reader_group) - имя группы пользователей, которым разрешено читать образ
 
 И также есть свойство у пула:
-- Группа создателей (`creator_group`) - имя группы пользователей, которым разрешено создавать образы в пуле
+- Группа создателей (creator_group) - имя группы пользователей, которым разрешено создавать образы в пуле
 
 Перечень разрешённых операций с данными образов на стороне OSD смотрите в разделе
 [Права доступа к данным OSD](#права-доступа-к-данным-osd).
@@ -213,23 +214,20 @@
 
 ### Настройка шифрования соединений OSD
 
-Вам нужно 3 сертификата: один для OSD и по одному для подписи сертификатов клиентов и администраторов.
+Вам нужно 2 сертификата: один для OSD и один для подписи сертификатов всех клиентов.
 Для OSD можно использовать самоподписанный сертификат (osd_ca.crt) или отдельный сертификат (osd.crt),
 подписанный доверенным сертификатом osd_ca.crt. Для клиентов нужно использовать отдельные
-сертификаты, подписанные общим доверенным (client_ca.crt), для администраторов - отдельным доверенным
-(admin_ca.crt).
+сертификаты, подписанные общим доверенным (client_ca.crt).
 
 В конфигурацию Vitastor на серверах OSD нужно добавить:
 - use_perms: true
 - osd_ca: osd_ca.crt
 - client_ca: client_ca.crt
-- admin_ca: admin_ca.crt
 - osd_cert: osd_ca.crt
 - osd_pkey: osd_ca.key
 
 На стороне клиентов:
 - use_perms: true
-- osd_ca: osd_ca.crt
 - cert: client.crt
 - pkey: client.key
 
@@ -239,7 +237,8 @@
 
 #### Mon со встроенным Antietcd
 
-Самый простой вариант. Кроме сертификатов OSD вам нужен всего 1 сертификат для Antietcd (antietcd.crt).
+Самый простой вариант. Вам нужен 1 сертификат для Antietcd (antietcd.crt), плюс
+корневые сертификаты для OSD и клиентов.
 
 Настройки Vitastor (`/etc/vitastor/vitastor.conf`):
 - etcd_address: [ "http://mon1:2379", ... ] (адреса ваших мониторов с портом 2379)
@@ -250,7 +249,6 @@
 - etcd_ca: antietcd.crt
 - osd_ca: osd_ca.crt
 - client_ca: client_ca.crt
-- admin_ca: admin_ca.crt
 
 #### Mon в роли Etcd proxy
 
@@ -258,6 +256,7 @@
 
 Вам понадобится 2 отдельных сертификата: один для etcd (etcd.crt) и один для antietcd (antietcd.crt).
 Клиентский порт etcd должен отличаться от стандартного 2379, например, можно выбрать 2381.
+Также нужны сертификаты OSD и клиентов.
 
 Настройки Vitastor:
 - etcd_address: [ "http://mon1:2379", ... ] (адреса ваших мониторов с портом 2379)
@@ -277,7 +276,6 @@
 - etcd_ca: antietcd.crt
 - osd_ca: osd_ca.crt
 - client_ca: client_ca.crt
-- admin_ca: admin_ca.crt
 
 Опции командной строки etcd:
 ```
@@ -291,8 +289,8 @@
 Если в дополнение к предыдущему варианту вы хотите разгрузить Antietcd от задач монитора Vitastor,
 можно запустить его отдельно.
 
-Аналогично предыдущему варианту нужно 2 сертификата: один для etcd и один для antietcd, плюс понадобится
-отдельный сертификат для монитора.
+Аналогично предыдущему варианту нужно 2 сертификата: один для etcd и один для antietcd, плюс понадобятся
+отдельные сертификаты для клиентов, OSD и монитора.
 
 Настройки Vitastor:
 - etcd_address: [ "http://mon1:2379", ... ] (адреса ваших мониторов с портом 2379)
@@ -301,18 +299,15 @@
 - etcd_ca: antietcd.crt
 - osd_ca: osd_ca.crt
 - client_ca: client_ca.crt
-- admin_ca: admin_ca.crt
 - mon_etcd_client_cert: mon_ca.crt
 - mon_etcd_client_key: mon_ca.key
 
 Опции командной строки Antietcd:
 ```
---port 2379 --cert /etc/vitastor/antietcd.crt --key /etc/vitastor/antietcd.key \
---client_cert_auth 1 --auth_filter /usr/lib/vitastor/mon/vitastor_auth_filter.js \
---ca /etc/vitastor/client_ca.crt --admin_ca /etc/vitastor/admin_ca.crt \
---osd_ca /etc/vitastor/osd_ca.crt --mon_ca /etc/vitastor/mon_ca.crt \
---etcd_proxy url1,url2,... --etcd_ca /etc/vitastor/etcd.crt \
---etcd_cert /etc/vitastor/antietcd.crt --etcd_key /etc/vitastor/antietcd.key
+--port 2379 \
+--client_cert_auth 1 --auth_filter vitastor_auth_filter.js --etcd_proxy url1,url2,... \
+--cert antietcd.crt --key antietcd.key --ca client_ca.crt --osd_ca osd_ca.crt --mon_ca mon_ca.crt \
+--etcd_cert antietcd.crt --etcd_key antietcd.key --etcd_ca etcd.crt
 ```
 
 Опции командной строки etcd (не отличаются от предыдущего варианта):
@@ -333,17 +328,15 @@
 - etcd_ca: antietcd.crt
 - osd_ca: osd_ca.crt
 - client_ca: client_ca.crt
-- admin_ca: admin_ca.crt
 - mon_etcd_client_cert: mon_ca.crt
 - mon_etcd_client_key: mon_ca.key
 
 Опции командной строки Antietcd:
 ```
---port 2379 --cert /etc/vitastor/antietcd.crt --key /etc/vitastor/antietcd.key \
---client_cert_auth 1 --auth_filter /usr/lib/vitastor/mon/vitastor_auth_filter.js \
---ca /etc/vitastor/client_ca.crt --admin_ca /etc/vitastor/admin_ca.crt \
---osd_ca /etc/vitastor/osd_ca.crt --mon_ca /etc/vitastor/mon_ca.crt \
---persist_filter /usr/lib/vitastor/mon/vitastor_persist_filter.js
+--port 2379 \
+--client_cert_auth 1 --auth_filter vitastor_auth_filter.js \
+--persist_filter vitastor_persist_filter.js \
+--cert antietcd.crt --key antietcd.key --ca client_ca.crt --osd_ca osd_ca.crt --mon_ca mon_ca.crt
 ```
 
 ### Настройка Vault/OpenBao
@@ -465,7 +458,7 @@
 
 Ниже все названия ключей приведены без общего префикса `/vitastor`.
 
-Разрешённые операции с ключами в Antietcd для обычных пользователей:
+Разрешённые операции с ключами в Antietcd для клиентов (`type=client`):
 - Только чтение:
   - Разрешено всегда:
     - `/config/global`
@@ -486,7 +479,7 @@
     - `/config/inode/*`
     - `/index/image/*`
 
-Разрешённые операции с ключами в Antietcd для администраторов:
+Разрешённые операции с ключами в Antietcd для администраторов (`type=admin`):
 - Чтение:
   - `/stats`
   - `/mon/*`
@@ -498,10 +491,7 @@
   - `/config/*`
   - `/osd/*`
   - `/index/*`
-  - `/pg/config`
   - `/pg/history/*`
-  - `/stats`
-  - `/history/last_clean_pgs`
 
 Разрешённые операции с ключами в etcd для OSD:
 - Чтение:
@@ -524,8 +514,6 @@
   - `/history/last_clean_pgs`
   - `/mon/*`
   - `/pg/history/*`
-  - `/pg/stats/*`
-  - `/pgstats/*`
   - `/inode/stats/*`
   - `/pool/stats/*`
 
@@ -563,22 +551,22 @@
 [vitastor-cli serve](../usage/cli.ru.md#serve) также поддерживает клиентскую
 аутентификацию по сертификатам. Принимаются только сертификаты, подписанные
 `client_ca`. В качестве серверного сертификата используется отдельный сертификат
-`api_cert` с ключом `api_pkey`.
+`server_cert` с ключом `server_pkey`.
 
 При этом для корректной работы `vitastor-cli serve` он сам должен использовать
-для доступа в Vitastor сертификат администратора (`cert`+`pkey`, подписанный `admin_ca`),
-чтобы корректно обрабатывать любые запросы.
+для доступа в Vitastor сертификат (`cert`+`pkey`) пользователя с правами
+администратора (`type=admin`).
 
-Обычным пользователям (с сертификатами, подписанными `client_ca`) при доступе к API
-разрешаются только API-операции с образами, доступными им либо на чтение (для чтения),
-либо на запись (для модификации):
+Обычным клиентам (пользователям с `type=client`) при доступе к API разрешаются
+только API-операции с образами, доступными им либо на чтение (для чтения), либо
+на запись (для модификации):
 
 - image/list - для образов, которые пользователь может читать.
 - image/create - для пулов, в которых пользователю разрешено создавать образы, либо
   для создания снимков образов, которыми пользователь владеет.
 - image/delete, image/flatten, image/modify - для образов, которыми пользователь владеет.
 
-Все остальные операции разрешаются только администраторам.
+Все остальные операции разрешаются только администраторам (`type=admin`).
 
 ## Производительность шифрования
 

mon/antietcd_adapter.js

@@ -85,7 +85,6 @@ class AntiEtcdAdapter
                     antietcd_config.ca = config.client_ca;
                     antietcd_config.osd_ca = config.osd_ca;
                     antietcd_config.mon_ca = config.mon_ca;
-                    antietcd_config.admin_ca = config.admin_ca;
                 }
                 for (const key in config)
                 {

mon/etcd_schema.js

@@ -65,8 +65,6 @@ const etcd_tree = {
             placement_levels: { datacenter: 1, rack: 2, host: 3, osd: 4, ... },
             use_old_pg_combinator: false,
             osd_backfillfull_ratio: 0.99,
-            mon_metrics_path: '/prometheus',
-            mon_healthcheck_path: '/healthcheck',
             // client and osd
             tcp_header_buffer_size: 65536,
             use_sync_send_recv: false,

mon/mon.js

@@ -87,7 +87,7 @@ class Mon
             this.http = await create_http_server(this.config, (req, res) =>
             {
                 const u = new URL(req.url, 'http://'+(req.headers.host || 'localhost'));
-                if (u.pathname.replace(/\/+$/, '') == (this.config.mon_metrics_path||'/metrics'))
+                if (u.pathname.replace(/\/+$/, '') == (this.config.prometheus_path||'/metrics'))
                 {
                     if (!this.watcher_active)
                     {
@@ -100,19 +100,6 @@ class Mon
                         res.write(export_prometheus_metrics(this.state));
                     }
                 }
-                else if (u.pathname.replace(/\/+$/, '') == (this.config.mon_healthcheck_path||'/healthcheck'))
-                {
-                    if (!this.watcher_active)
-                    {
-                        res.writeHead(503);
-                        res.write('Monitor is in standby mode. Please retrieve metrics from master monitor instance\n');
-                    }
-                    else
-                    {
-                        res.setHeader('Content-Type', 'text/plain; version=0.0.4; charset=utf-8');
-                        res.write('Monitor is active');
-                    }
-                }
                 else
                 {
                     res.writeHead(404);

mon/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vitastor-mon",
-  "version": "3.2.2",
+  "version": "3.0.15",
   "description": "Vitastor SDS monitor service",
   "main": "mon-main.js",
   "scripts": {
@@ -9,7 +9,7 @@
   "author": "Vitaliy Filippov",
   "license": "UNLICENSED",
   "dependencies": {
-    "antietcd": "^1.3.3",
+    "antietcd": "^1.3.1",
     "sprintf-js": "^1.1.2",
     "ws": "^7.2.5"
   },

mon/pool_config.js

@@ -3,9 +3,6 @@
 
 const { parse_level_indexes, parse_pg_dsl } = require('./lp_optimizer/dsl_pgs.js');
 
-const MIN_DATA_BLOCK_SIZE = 4096;
-const MAX_DATA_BLOCK_SIZE = 128*1024*1024;
-
 function validate_pool_cfg(pool_id, pool_cfg, placement_levels, warn)
 {
     pool_cfg.pg_size = Math.floor(pool_cfg.pg_size);
@@ -26,8 +23,7 @@ function validate_pool_cfg(pool_id, pool_cfg, placement_levels, warn)
             console.log('Pool '+pool_id+' has invalid coding scheme (one of "xor", "replicated", "ec" and "jerasure" required)');
         return false;
     }
-    // OSDs don't support pg_size > 64 for now
-    if (!pool_cfg.pg_size || pool_cfg.pg_size < 1 || pool_cfg.pg_size > 64 ||
+    if (!pool_cfg.pg_size || pool_cfg.pg_size < 1 || pool_cfg.pg_size > 256 ||
         pool_cfg.scheme !== 'replicated' && pool_cfg.pg_size < 3)
     {
         if (warn)
@@ -54,18 +50,6 @@ function validate_pool_cfg(pool_id, pool_cfg, placement_levels, warn)
             console.log('Pool '+pool_id+' has invalid parity_chunks (must be between 1 and pg_size-2)');
         return false;
     }
-    if (pool_cfg.block_size &&
-        (!Number(pool_cfg.block_size) || (pool_cfg.block_size & (pool_cfg.block_size-1)) ||
-        pool_cfg.block_size < MIN_DATA_BLOCK_SIZE || pool_cfg.block_size > MAX_DATA_BLOCK_SIZE ||
-        pool_cfg.scheme !== 'replicated' && pool_cfg.block_size*pool_cfg.pg_size > 0xFFFFFFFF))
-    {
-        if (warn)
-        {
-            console.log('Pool '+pool_id+' has invalid block_size (must be a power of two between '+
-                MIN_DATA_BLOCK_SIZE+' and '+MAX_DATA_BLOCK_SIZE+'; must not exceed 2^32-1 when multiplied by pg_size)');
-        }
-        return false;
-    }
     if (!pool_cfg.pg_count || pool_cfg.pg_count < 1)
     {
         if (warn)

mon/scripts/make-etcd

@@ -9,6 +9,8 @@ const os = require('os');
 const path = require('path');
 const readline = require('readline');
 
+run().catch(e => { console.error(e); process.exit(1); });
+
 const help_text = `Initialize a Vitastor cluster (etcd, vitastor.conf and TLS certificates)
 (c) Vitaliy Filippov, 2026+ (MIT)
 
@@ -20,8 +22,7 @@ USAGE:
 2) Run: ${process.argv[1]} [./vitastor.conf] [--antietcd-only]
    You can run it on etcd/monitor nodes or on an external node.
    It configures etcd, generates TLS certificates (on the first or external node), copies
-   them to other etcd/monitor nodes using root ssh, and updates vitastor.conf with TLS options.
-   You can also copy configuration and certificates manually.
+   them to other etcd/monitor nodes, and updates vitastor.conf with TLS options.
 3) If you have OSD-only nodes, run:
    ${process.argv[1]} --copy-to-osd-node NODE_NAME ./vitastor.conf
    It copies vitastor.conf and required TLS certificates to that node.
@@ -43,7 +44,7 @@ OPTIONS:
 async function run()
 {
     let config_path = '/etc/vitastor/vitastor.conf';
-    let config_dir = '/etc/vitastor';
+    let config_dir = '/etc/vitastor/';
     let gen_certs = 'auto';
     let antietcd_only = false;
     let copy = 'ask';
@@ -110,7 +111,7 @@ async function run()
         .filter(s => s)
         .map(s => ({
             scheme: s[1],
-            addr: s[2].indexOf(':') >= 0 && s[2][0] != '[' ? '['+s[2]+']' : s[2],
+            addr: s[2].indexOf(':') && s[2][0] != '[' ? '['+s[2]+']' : s[2],
             ip: s[2][0] == '[' ? s[2].substr(1, s[2].length-2) : s[2],
             port: s[3],
         }));
@@ -124,7 +125,7 @@ async function run()
     }
     if (tls)
     {
-        const etcd_ca = config_dir+'/'+path.basename(config.etcd_ca || '/etc/vitastor/etcd_ca.crt');
+        const etcd_ca = config_dir+'/'+path.basename(config.etcd_ca);
         if (gen_certs === true)
         {
             console.log('Certificate generation is requested explicitly, generating');
@@ -157,24 +158,18 @@ async function run()
         if (gen_certs)
         {
             if (copy === 'ask')
-                copy = await ask_copy('Copy certificates and vitastor.conf to other nodes and start them (requires root ssh)? ');
+                copy = await ask_copy('Copy certificates and vitastor.conf to other nodes after generation?');
             copy = (copy === 'y' || copy === 'yes');
-            await make_certs(config_dir, copy, etcds, num, use_perms, antietcd_only);
+            await make_certs(config_dir, copy, etcds, use_perms, antietcd_only);
         }
     }
-    else if (num <= 0)
-    {
-        if (copy === 'ask')
-            copy = await ask_copy('Copy vitastor.conf to other nodes and start them (requires root ssh)? ');
-        if (copy === 'y' || copy === 'yes')
-            await copy_nontls(config_dir, etcds, num);
-    }
     if (num < 0)
     {
         console.log('No matching IPs in etcd_address from '+config_path);
         process.exit(tls && gen_certs ? 0 : 1);
     }
     await configure_etcd(etcds, num, tls, use_perms);
+    await enable_mon();
     process.exit(0);
 }
 
@@ -206,10 +201,10 @@ async function copy_to_osd_nodes(to, dir, use_perms, antietcd_only)
         osd_to_copy.push('osd.crt', 'osd.key', 'client_ca.crt');
     console.warn('Copying configuration to OSD nodes '+to.join(', '));
     for (const node of to)
-        await system("scp "+dir+"/"+osd_to_copy.join(" "+dir+"/")+" root@"+(node.indexOf(':') >= 0 ? "["+node+"]" : node)+":/etc/vitastor/");
+        await system("scp "+dir+osd_to_copy.join(" "+dir)+" root@"+node+":/etc/vitastor/");
 }
 
-async function make_certs(dir, copy, etcds, num, use_perms, antietcd_only)
+async function make_certs(dir, copy, etcds, use_perms, antietcd_only)
 {
     console.log(`-----
 Generating certificates in ${dir}
@@ -219,27 +214,26 @@ Generating certificates in ${dir}
     const osd_to_copy = [ 'vitastor.conf' ];
     if (!antietcd_only)
     {
-        await make_ca("/O=Vitastor etcd CA", dir+"/etcd_ca");
-        await make_signed("/CN=Vitastor etcd", dir+"/etcd", dir+"/etcd_ca", etcds.map(e => "IP:"+e.ip).join(','));
+        await make_ca("/O=Vitastor etcd CA", dir+"etcd_ca");
+        await make_signed("/CN=Vitastor etcd", dir+"etcd", dir+"etcd_ca", etcds.map(e => "IP:"+e.ip).join(','));
         to_copy.push('etcd_ca.crt', 'etcd.crt', 'etcd.key');
         if (!use_perms)
             osd_to_copy.push('etcd_ca.crt');
     }
     if (use_perms || antietcd_only)
     {
-        await make_ca("/O=Vitastor Antietcd CA", dir+"/antietcd_ca");
-        await make_signed("/CN=Vitastor Antietcd", dir+"/antietcd", dir+"/antietcd_ca", etcds.map(e => "IP:"+e.ip).join(','));
+        await make_ca("/O=Vitastor Antietcd CA", dir+"antietcd_ca");
+        await make_signed("/CN=Vitastor Antietcd", dir+"antietcd", dir+"antietcd_ca", etcds.map(e => "IP:"+e.ip).join(','));
         to_copy.push('antietcd_ca.crt', 'antietcd.crt', 'antietcd.key');
         osd_to_copy.push('antietcd_ca.crt');
     }
     if (use_perms)
     {
-        await make_ca("/CN=Vitastor OSD", dir+"/osd");
-        await make_ca("/O=Vitastor Client CA", dir+"/client_ca");
-        await make_ca("/O=Vitastor Admin CA", dir+"/admin_ca");
-        await make_signed("/CN=admin", dir+"/admin", dir+"/admin_ca");
-        to_copy.push('osd.crt', 'osd.key', 'client_ca.crt', 'admin_ca.crt', 'admin.crt', 'admin.key');
-        osd_to_copy.push('osd.crt', 'osd.key', 'client_ca.crt', 'admin_ca.crt', 'admin.crt', 'admin.key');
+        await make_ca("/CN=Vitastor OSD", dir+"osd");
+        await make_ca("/O=Vitastor Client CA", dir+"client_ca");
+        await make_signed("/CN=admin", dir+"admin", dir+"client_ca");
+        to_copy.push('osd.crt', 'osd.key', 'client_ca.crt');
+        osd_to_copy.push('osd.crt', 'osd.key', 'client_ca.crt');
     }
     console.log(`-----
 Certificates generated, commands to copy them:
@@ -249,19 +243,13 @@ Certificates generated, commands to copy them:
   cd ${dir} && scp ${to_copy.filter(f => f != 'osd.key').join(' ')} root@NODE:/etc/vitastor/
 - OSD node:
   cd ${dir} && scp ${osd_to_copy.join(' ')} root@NODE:/etc/vitastor/
-Then don't forget to run:
-chown vitastor:vitastor /etc/vitastor/*.key && /usr/lib/vitastor/mon/make-etcd
 -----
 `);
     if (copy)
     {
-        for (let i = 0; i < etcds.length; i++)
+        for (const node of etcds)
         {
-            if (i == num)
-                continue;
-            const node = etcds[i];
-            await system("scp "+dir+"/"+to_copy.join(" "+dir+"/")+" root@"+(node.ip.indexOf(':') >= 0 ? "["+node.ip+"]" : node.ip)+
-                ":/etc/vitastor/ && ssh root@"+node.ip+" 'chown vitastor:vitastor /etc/vitastor/*.key && /usr/lib/vitastor/mon/make-etcd'");
+            await system("scp "+dir+to_copy.join(" "+dir)+" root@"+node.ip+":/etc/vitastor/");
         }
     }
     else
@@ -270,19 +258,6 @@ chown vitastor:vitastor /etc/vitastor/*.key && /usr/lib/vitastor/mon/make-etcd
     }
 }
 
-async function copy_nontls(dir, etcds, num)
-{
-    const to_copy = [ 'vitastor.conf' ];
-    for (let i = 0; i < etcds.length; i++)
-    {
-        if (i == num)
-            continue;
-        const node = etcds[i];
-        await system("scp "+dir+"/"+to_copy.join(" "+dir+"/")+" root@"+(node.ip.indexOf(':') >= 0 ? "["+node.ip+"]" : node.ip)+
-            ":/etc/vitastor/ && ssh root@"+node.ip+" '/usr/lib/vitastor/mon/make-etcd'");
-    }
-}
-
 async function write_auth_config(config, config_path, etcds, use_perms, antietcd_only)
 {
     const auth = {};
@@ -292,7 +267,7 @@ async function write_auth_config(config, config_path, etcds, use_perms, antietcd
         if (!antietcd_only)
         {
             auth["etcd_proxy"] = {
-                urls: etcds.map(e => (e.ip.indexOf(':') >= 0 ? '['+e.ip+']' : e.ip)+':2381'),
+                urls: etcds.map(e => e.ip+':2381'),
                 cert: "/etc/vitastor/antietcd.crt",
                 key: "/etc/vitastor/antietcd.key",
                 ca: "/etc/vitastor/etcd_ca.crt",
@@ -305,7 +280,6 @@ async function write_auth_config(config, config_path, etcds, use_perms, antietcd
         auth["osd_pkey"] = "/etc/vitastor/osd.key";
         auth["osd_ca"] = "/etc/vitastor/osd.crt";
         auth["client_ca"] = "/etc/vitastor/client_ca.crt";
-        auth["admin_ca"] = "/etc/vitastor/admin_ca.crt";
         auth["cert"] = "/etc/vitastor/admin.crt";
         auth["pkey"] = "/etc/vitastor/admin.key";
     }
@@ -347,10 +321,9 @@ async function configure_etcd(etcds, num, tls, use_perms)
 {
     const in_docker = fs.existsSync("/etc/vitastor/etcd.conf") &&
         fs.existsSync("/etc/vitastor/docker.conf");
-    if (!in_docker && fs.existsSync("/etc/systemd/system/vitastor-etcd.service") &&
-        fs.existsSync("/etc/vitastor/etcd.conf"))
+    if (!in_docker && fs.existsSync("/etc/systemd/system/vitastor-etcd.service"))
     {
-        console.log("/etc/systemd/system/vitastor-etcd.service and /etc/vitastor/etcd.conf already exists. Remove them to reinitialize the cluster");
+        console.log("/etc/systemd/system/vitastor-etcd.service already exists");
         process.exit(1);
     }
     if (!in_docker && fs.existsSync("/etc/systemd/system/etcd.service"))
@@ -381,14 +354,14 @@ async function configure_etcd(etcds, num, tls, use_perms)
         if (use_perms)
         {
             options['client_cert_auth'] = '1';
-            options['trusted_ca_file'] = '/etc/vitastor/antietcd_ca.crt';
+            options['trusted_ca_file'] = '/etc/vitastor/antietcd.crt';
         }
         options['peer_cert_file'] = '/etc/vitastor/etcd.crt';
         options['peer_key_file'] = '/etc/vitastor/etcd.key';
         if (use_perms)
         {
             options['peer_client_cert_auth'] = '1';
-            options['peer_trusted_ca_file'] = '/etc/vitastor/etcd_ca.crt';
+            options['peer_trusted_ca_file'] = '/etc/vitastor/etcd.crt';
         }
     }
     let etcd_conf = fs.existsSync("/etc/vitastor/etcd.conf")
@@ -429,15 +402,14 @@ RestartSec=10
 WantedBy=multi-user.target
 `);
     await system(`useradd etcd`);
-    await system(`chown vitastor:vitastor /etc/vitastor/*.key`);
-    if (options['key_file'])
-    {
-        await system(`chown etcd `+options['key_file']);
-    }
     await system(`systemctl daemon-reload`);
     // Disable distribution etcd unit and enable our one
     await system(`systemctl disable --now etcd`);
     await system(`systemctl enable --now vitastor-etcd`);
+}
+
+async function enable_mon()
+{
     await system(`systemctl enable --now vitastor-mon`);
 }
 
@@ -445,9 +417,7 @@ function replace_env(text, key, value)
 {
     let found = false;
     text = text.replace(new RegExp('^'+key+'\\s*=.*', 'm'), () => { found = true; return key+'='+value; });
-    if (found)
-        return text;
-    return (text != '' ? text.replace(/\s*$/, '\n') : '')+key+'='+value+'\n';
+    return found ? text : text.replace(/\s*$/, '\n')+key+'='+value+'\n';
 }
 
 function select_local_etcd(etcds)
@@ -476,13 +446,11 @@ function select_local_etcd(etcds)
 async function system(cmd)
 {
     console.log('Running '+cmd);
-    process.stdin.pause();
     const cp = child_process.spawn(cmd, { shell: true, stdio: [ 0, 1, 2 ] });
     let finish_cb;
     cp.on('exit', () => finish_cb && finish_cb());
     if (cp.exitCode == null)
         await new Promise(ok => finish_cb = ok);
-    process.stdin.resume();
     return cp.exitCode;
 }
 
@@ -497,9 +465,7 @@ async function make_signed(subj, f, ca, san)
 {
     if (await system(`openssl req -subj '${subj}' ${san ? "-addext 'subjectAltName="+san+"'" : ""} -nodes -new -keyout ${f}.key -out ${f}.csr`))
         process.exit(1);
-    if (await system(`openssl x509 -req -days 3650 -copy_extensions copy -CA ${ca}.crt -CAkey ${ca}.key -CAcreateserial -in ${f}.csr -out ${f}.crt`))
+    if (await system(`openssl x509 -req -days 3650 -CA ${ca}.crt -CAkey ${ca}.key -CAcreateserial -in ${f}.csr -out ${f}.crt`))
         process.exit(1);
     fs.unlinkSync(f+".csr");
 }
-
-run().catch(e => { console.error(e); process.exit(1); });

mon/vitastor_auth_filter.js

@@ -5,7 +5,10 @@
 // Permissions are based on:
 // 1. Users.
 //    Stored in /vitastor/config/user/<username>.
-//    Has 1 property:
+//    Has 2 properties:
+//    - type, one of: admin, client.
+//      admin has full access to all images and also to cluster config.
+//      client has r/w access to owned images and r/o access to images with reader_group.
 //    - groups, a list of group names the user is included in.
 // 2. Images.
 //    Stored in /vitastor/config/inode/<pool>/<inode>. Has the following properties:
@@ -13,9 +16,7 @@
 //    - owner_group (group name)
 //    - reader_group
 // 3. Certificates.
-//    - osd_ca identifies OSDs.
-//    - admin_ca (optional) identifies administrators.
-//    - mon_ca (optional) identifies monitors.
+//    - osd, mon use their own trusted certificates.
 
 const { X509Certificate } = require('node:crypto');
 
@@ -32,11 +33,11 @@ const static_perms = {
         keys: { '/pg/config': true, '/stats': true, '/history/last_clean_pgs': true },
         prefixes: {
             '/config/': false, '/osd/': false, '/mon/': true, '/pg/history/': true,
-            '/pg/stats/': true, '/pgstats/': true, '/inode/stats/': true, '/pool/stats/': true,
+            '/pgstats/': false, '/inode/stats/': true, '/pool/stats/': true,
         },
     },
     admin: {
-        keys: { '/pg/config': true, '/stats': true, '/history/last_clean_pgs': true },
+        keys: { '/stats': false },
         prefixes: {
             '/config/': true, '/osd/': true, '/index/': true, '/pg/history/': true,
             '/mon/': false, '/pg/': false, '/pgstats/': false, '/inode/stats/': false, '/pool/stats/': false,
@@ -70,7 +71,7 @@ class VitastorAuthFilter
         if (!this.cfg.cert || !this.cfg.key || !this.cfg.ca || !this.cfg.osd_ca || !this.cfg.client_cert_auth)
         {
             throw new Error('Authenticated Vitastor setups require enabled client_cert_auth, cert, key'+
-                ' and separate ca (client CA), osd_ca and optionally admin_ca and mon_ca');
+                ' and separate ca (client CA), osd_ca and optionally mon_ca');
         }
         this.osd_ca = await this.antietcd.readPEM(this.cfg.osd_ca);
         this.osd_ca_obj = new X509Certificate(this.osd_ca);
@@ -78,31 +79,27 @@ class VitastorAuthFilter
         if (this.cfg.mon_ca)
         {
             this.mon_ca = await this.antietcd.readPEM(this.cfg.mon_ca);
-            this.mon_ca_obj = new X509Certificate(this.mon_ca);
+            this.mon_ca_obj = new X509Certificate(this.mon_ca_obj);
             this.antietcd.tls.ca.push(this.mon_ca);
         }
-        if (this.cfg.admin_ca)
-        {
-            this.admin_ca = await this.antietcd.readPEM(this.cfg.admin_ca);
-            this.admin_ca_obj = new X509Certificate(this.admin_ca);
-            this.antietcd.tls.ca.push(this.admin_ca);
-        }
     }
 
-    init_context(context, socket)
+    init_context(context, clientCert)
     {
-        let cert = socket.getPeerX509Certificate();
-        if (cert.issuer == this.osd_ca_obj.subject && cert.verify(this.osd_ca_obj.publicKey))
+        let cert = clientCert;
+        while (cert)
         {
-            context.user_type = 'osd';
-        }
-        else if (this.mon_ca_obj && cert.issuer == this.mon_ca_obj.subject && cert.verify(this.mon_ca_obj.publicKey))
-        {
-            context.user_type = 'mon';
-        }
-        else if (this.admin_ca_obj && cert.issuer == this.admin_ca_obj.subject && cert.verify(this.admin_ca_obj.publicKey))
-        {
-            context.user_type = 'admin';
+            if (cert.fingerprint256 == this.osd_ca_obj.fingerprint256)
+            {
+                context.user_type = 'osd';
+                break;
+            }
+            if (this.mon_ca_obj && cert.fingerprint256 == this.mon_ca_obj.fingerprint256)
+            {
+                context.user_type = 'mon';
+                break;
+            }
+            cert = cert.issuerCertificate;
         }
     }
 
@@ -149,7 +146,7 @@ class VitastorAuthFilter
         return cur;
     }
 
-    // userInfo: { name: string, perms: static_perms[type], groups: { [string]: true } }
+    // userInfo: { name: string, type: string, perms: static_perms[type], groups: { [string]: true } }
     _check_compare(check, userInfo, checked)
     {
         let key = String(check.key);
@@ -177,7 +174,7 @@ class VitastorAuthFilter
                 // Allowed to check that a key does not exist
                 if (check.target == 'VERSION' && check.version == 0)
                 {
-                    checked['C'+key] = true;
+                    checked['M'+key] = true;
                     return true;
                 }
                 else if (check.target == 'MOD')
@@ -354,25 +351,6 @@ class VitastorAuthFilter
                 // Allowed to modify known images with CAS checks
                 return true;
             }
-            if (key.substr(0, 14) == '/config/inode/' &&
-                checked['C'+key])
-            {
-                // Allowed to create an image with CAS checks and with read access
-                // to the parent image, if specified
-                const data = this._decode(put.key, put.value);
-                if (data.parent_id)
-                {
-                    const [ cur_pool_id ] = put.key.substr(14).split('/');
-                    const inode = this._get([ ...this.prefix_parts, 'config', 'inode', data.parent_pool || cur_pool_id, data.parent_id ], true);
-                    if (inode && (inode.reader_group && userInfo.groups[inode.reader_group] ||
-                        inode.owner_group && userInfo.groups[inode.owner_group] ||
-                        inode.owner === userInfo.name))
-                    {
-                        return true;
-                    }
-                }
-                return true;
-            }
         }
         return false;
     }
@@ -425,7 +403,7 @@ class VitastorAuthFilter
 
     _get_user(context)
     {
-        if (context.user_type === 'osd' || context.user_type === 'mon' || context.user_type === 'admin')
+        if (context.user_type === 'osd' || context.user_type === 'mon')
         {
             return {
                 name: context.user_type,
@@ -440,7 +418,11 @@ class VitastorAuthFilter
         let userInfo = this._get([ ...this.prefix_parts, 'config', 'user', context.username ], true);
         if (!userInfo)
         {
-            userInfo = {};
+            userInfo = { type: 'client' };
+        }
+        else if (userInfo.type !== 'client' && userInfo.type !== 'admin')
+        {
+            userInfo.type = 'client';
         }
         userInfo.perms = static_perms[userInfo.type] || static_perms['invalid'];
         userInfo.name = context.username;
@@ -458,10 +440,18 @@ class VitastorAuthFilter
     filter_api(context, api/*, data*/)
     {
         let type = 'client';
-        if (context.user_type === 'osd' || context.user_type === 'mon' || context.user_type === 'admin')
+        if (context.user_type === 'osd' || context.user_type === 'mon')
         {
             type = context.user_type;
         }
+        else if (context.username)
+        {
+            const userInfo = this._get([ ...this.prefix_parts, 'config', 'user', context.username ], true);
+            if (userInfo && userInfo.type === 'admin')
+            {
+                type = 'admin';
+            }
+        }
         return api_perms[type] && api_perms[type][api];
     }
 

node-binding/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vitastor",
-  "version": "3.2.2",
+  "version": "3.0.15",
   "description": "Low-level native bindings to Vitastor client library",
   "main": "index.js",
   "keywords": [

patches/cinder-vitastor.py

@@ -50,7 +50,7 @@ from cinder.volume import configuration
 from cinder.volume import driver
 from cinder.volume import volume_utils
 
-VITASTOR_VERSION = '3.2.2'
+VITASTOR_VERSION = '3.0.15'
 
 LOG = logging.getLogger(__name__)
 
@@ -608,7 +608,7 @@ class VitastorDriver(driver.CloneableImageVD,
         """Extend an existing volume."""
         vol_name = utils.convert_str(volume.name)
         size = int(new_size) * units.Gi
-        self._cli('extend volume', 'modify', vol_name, '--resize', size)
+        self._cli('extend volume', 'modify', vol_name, '--resize', new_size)
         LOG.debug(
             "Extend volume from %(old_size)s GB to %(new_size)s GB.",
             {'old_size': volume.size, 'new_size': new_size}

patches/libvirt-11.10-vitastor.patch

@@ -1,699 +0,0 @@
-From 560382fdf86fdbbd08c38187ef602c04e0abece0 Mon Sep 17 00:00:00 2001
-From: Vitaliy Filippov <vitalif@yourcmc.ru>
-Date: Sat, 5 Sep 2026 20:41:38 +0300
-Subject: [PATCH] Add Vitastor support
-
----
- include/libvirt/libvirt-storage.h             |   1 +
- src/conf/domain_conf.c                        |   4 +-
- src/conf/domain_validate.c                    |  11 +-
- src/conf/schemas/domaincommon.rng             |  30 +++++
- src/conf/storage_conf.c                       |  20 ++-
- src/conf/storage_conf.h                       |   2 +
- src/conf/storage_source_conf.c                |   2 +
- src/conf/storage_source_conf.h                |   1 +
- src/conf/virstorageobj.c                      |   3 +
- src/libvirt-storage.c                         |   1 +
- src/libxl/libxl_conf.c                        |   1 +
- src/libxl/xen_xl.c                            |   1 +
- src/qemu/qemu_block.c                         |  45 +++++++
- src/qemu/qemu_domain.c                        |   5 +-
- src/qemu/qemu_nbdkit.c                        |   2 +
- src/qemu/qemu_snapshot.c                      |   2 +
- src/storage/storage_driver.c                  |   1 +
- .../storage_source_backingstore.c             | 123 ++++++++++++++++++
- src/test/test_driver.c                        |   1 +
- .../storagepoolcapsschemadata/poolcaps-fs.xml |   7 +
- .../poolcaps-full.xml                         |   7 +
- tests/storagepoolxml2argvtest.c               |   1 +
- tools/virsh-pool.c                            |   3 +
- 23 files changed, 268 insertions(+), 6 deletions(-)
-
-diff --git a/include/libvirt/libvirt-storage.h b/include/libvirt/libvirt-storage.h
-index aaad4a3da1..5f5daa8341 100644
---- a/include/libvirt/libvirt-storage.h
-+++ b/include/libvirt/libvirt-storage.h
-@@ -326,6 +326,7 @@ typedef enum {
-     VIR_CONNECT_LIST_STORAGE_POOLS_ZFS           = 1 << 17, /* (Since: 1.2.8) */
-     VIR_CONNECT_LIST_STORAGE_POOLS_VSTORAGE      = 1 << 18, /* (Since: 3.1.0) */
-     VIR_CONNECT_LIST_STORAGE_POOLS_ISCSI_DIRECT  = 1 << 19, /* (Since: 5.6.0) */
-+    VIR_CONNECT_LIST_STORAGE_POOLS_VITASTOR      = 1 << 20, /* (Since: 5.0.0) */
- } virConnectListAllStoragePoolsFlags;
- 
- int                     virConnectListAllStoragePools(virConnectPtr conn,
-diff --git a/src/conf/domain_conf.c b/src/conf/domain_conf.c
-index 541dad5bdc..5cb94e1fed 100644
---- a/src/conf/domain_conf.c
-+++ b/src/conf/domain_conf.c
-@@ -7453,7 +7453,8 @@ virDomainDiskSourceNetworkParse(xmlNodePtr node,
-     src->configFile = virXPathString("string(./config/@file)", ctxt);
- 
-     if (src->protocol == VIR_STORAGE_NET_PROTOCOL_HTTP ||
--        src->protocol == VIR_STORAGE_NET_PROTOCOL_HTTPS)
-+        src->protocol == VIR_STORAGE_NET_PROTOCOL_HTTPS ||
-+        src->protocol == VIR_STORAGE_NET_PROTOCOL_VITASTOR)
-         src->query = virXMLPropString(node, "query");
- 
-     if (virDomainStorageNetworkParseHosts(node, ctxt, &src->hosts, &src->nhosts) < 0)
-@@ -32196,6 +32197,7 @@ virDomainStorageSourceTranslateSourcePool(virStorageSource *src,
- 
-     case VIR_STORAGE_POOL_MPATH:
-     case VIR_STORAGE_POOL_RBD:
-+    case VIR_STORAGE_POOL_VITASTOR:
-     case VIR_STORAGE_POOL_SHEEPDOG:
-     case VIR_STORAGE_POOL_GLUSTER:
-     case VIR_STORAGE_POOL_LAST:
-diff --git a/src/conf/domain_validate.c b/src/conf/domain_validate.c
-index 4558e7b210..3c6b2a8ee8 100644
---- a/src/conf/domain_validate.c
-+++ b/src/conf/domain_validate.c
-@@ -520,6 +520,7 @@ virDomainDiskDefValidateSourceChainOne(const virStorageSource *src)
-         case VIR_STORAGE_NET_PROTOCOL_RBD:
-             break;
- 
-+        case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
-         case VIR_STORAGE_NET_PROTOCOL_NBD:
-         case VIR_STORAGE_NET_PROTOCOL_SHEEPDOG:
-         case VIR_STORAGE_NET_PROTOCOL_GLUSTER:
-@@ -592,7 +593,7 @@ virDomainDiskDefValidateSourceChainOne(const virStorageSource *src)
-         }
-     }
- 
--    /* internal snapshots and config files are currently supported only with rbd: */
-+    /* internal snapshots are currently supported only with rbd: */
-     if (virStorageSourceGetActualType(src) != VIR_STORAGE_TYPE_NETWORK &&
-         src->protocol != VIR_STORAGE_NET_PROTOCOL_RBD) {
-         if (src->snapshot) {
-@@ -600,10 +601,14 @@ virDomainDiskDefValidateSourceChainOne(const virStorageSource *src)
-                            _("<snapshot> element is currently supported only with 'rbd' disks"));
-             return -1;
-         }
--
-+    }
-+    /* config files are currently supported only with rbd and vitastor: */
-+    if (virStorageSourceGetActualType(src) != VIR_STORAGE_TYPE_NETWORK &&
-+        src->protocol != VIR_STORAGE_NET_PROTOCOL_RBD &&
-+        src->protocol != VIR_STORAGE_NET_PROTOCOL_VITASTOR) {
-         if (src->configFile) {
-             virReportError(VIR_ERR_XML_ERROR, "%s",
--                           _("<config> element is currently supported only with 'rbd' disks"));
-+                           _("<config> element is currently supported only with 'rbd' and 'vitastor' disks"));
-             return -1;
-         }
-     }
-diff --git a/src/conf/schemas/domaincommon.rng b/src/conf/schemas/domaincommon.rng
-index 1f9ac102a0..51b55ea72b 100644
---- a/src/conf/schemas/domaincommon.rng
-+++ b/src/conf/schemas/domaincommon.rng
-@@ -2092,6 +2092,35 @@
-     </element>
-   </define>
- 
-+  <define name="diskSourceNetworkProtocolVitastor">
-+    <element name="source">
-+      <interleave>
-+        <attribute name="protocol">
-+          <value>vitastor</value>
-+        </attribute>
-+        <ref name="diskSourceCommon"/>
-+        <optional>
-+          <attribute name="name"/>
-+        </optional>
-+        <optional>
-+          <attribute name="query"/>
-+        </optional>
-+        <zeroOrMore>
-+          <ref name="diskSourceNetworkHost"/>
-+        </zeroOrMore>
-+        <optional>
-+          <element name="config">
-+            <attribute name="file">
-+              <ref name="absFilePath"/>
-+            </attribute>
-+            <empty/>
-+          </element>
-+        </optional>
-+        <empty/>
-+      </interleave>
-+    </element>
-+  </define>
-+
-   <define name="diskSourceNetworkProtocolISCSI">
-     <element name="source">
-       <attribute name="protocol">
-@@ -2442,6 +2471,7 @@
-       <ref name="diskSourceNetworkProtocolSimple"/>
-       <ref name="diskSourceNetworkProtocolVxHS"/>
-       <ref name="diskSourceNetworkProtocolNFS"/>
-+      <ref name="diskSourceNetworkProtocolVitastor"/>
-     </choice>
-   </define>
- 
-diff --git a/src/conf/storage_conf.c b/src/conf/storage_conf.c
-index 1dc9365bf2..a8a736be81 100644
---- a/src/conf/storage_conf.c
-+++ b/src/conf/storage_conf.c
-@@ -56,7 +56,7 @@ VIR_ENUM_IMPL(virStoragePool,
-               "logical", "disk", "iscsi",
-               "iscsi-direct", "scsi", "mpath",
-               "rbd", "sheepdog", "gluster",
--              "zfs", "vstorage",
-+              "zfs", "vstorage", "vitastor",
- );
- 
- VIR_ENUM_IMPL(virStoragePoolFormatFileSystem,
-@@ -242,6 +242,18 @@ static virStoragePoolTypeInfo poolTypeInfo[] = {
-           .formatToString = virStorageFileFormatTypeToString,
-       }
-     },
-+    {.poolType = VIR_STORAGE_POOL_VITASTOR,
-+     .poolOptions = {
-+         .flags = (VIR_STORAGE_POOL_SOURCE_HOST |
-+                   VIR_STORAGE_POOL_SOURCE_NETWORK |
-+                   VIR_STORAGE_POOL_SOURCE_NAME),
-+      },
-+      .volOptions = {
-+          .defaultFormat = VIR_STORAGE_FILE_RAW,
-+          .formatFromString = virStorageVolumeFormatFromString,
-+          .formatToString = virStorageFileFormatTypeToString,
-+      }
-+    },
-     {.poolType = VIR_STORAGE_POOL_SHEEPDOG,
-      .poolOptions = {
-          .flags = (VIR_STORAGE_POOL_SOURCE_HOST |
-@@ -538,6 +550,11 @@ virStoragePoolDefParseSource(xmlXPathContextPtr ctxt,
-                        _("element 'name' is mandatory for RBD pool"));
-         return -1;
-     }
-+    if (pool_type == VIR_STORAGE_POOL_VITASTOR && source->name == NULL) {
-+        virReportError(VIR_ERR_XML_ERROR, "%s",
-+                       _("element 'name' is mandatory for Vitastor pool"));
-+        return -1;
-+    }
- 
-     if (options->formatFromString) {
-         g_autofree char *format = NULL;
-@@ -1127,6 +1144,7 @@ virStoragePoolDefFormatBuf(virBuffer *buf,
-     /* RBD, Sheepdog, Gluster and Iscsi-direct devices are not local block devs nor
-      * files, so they don't have a target */
-     if (def->type != VIR_STORAGE_POOL_RBD &&
-+        def->type != VIR_STORAGE_POOL_VITASTOR &&
-         def->type != VIR_STORAGE_POOL_SHEEPDOG &&
-         def->type != VIR_STORAGE_POOL_GLUSTER &&
-         def->type != VIR_STORAGE_POOL_ISCSI_DIRECT) {
-diff --git a/src/conf/storage_conf.h b/src/conf/storage_conf.h
-index fc67957cfe..720c07ef74 100644
---- a/src/conf/storage_conf.h
-+++ b/src/conf/storage_conf.h
-@@ -103,6 +103,7 @@ typedef enum {
-     VIR_STORAGE_POOL_GLUSTER,  /* Gluster device */
-     VIR_STORAGE_POOL_ZFS,      /* ZFS */
-     VIR_STORAGE_POOL_VSTORAGE, /* Virtuozzo Storage */
-+    VIR_STORAGE_POOL_VITASTOR, /* Vitastor */
- 
-     VIR_STORAGE_POOL_LAST,
- } virStoragePoolType;
-@@ -454,6 +455,7 @@ VIR_ENUM_DECL(virStoragePartedFs);
-                  VIR_CONNECT_LIST_STORAGE_POOLS_SCSI     | \
-                  VIR_CONNECT_LIST_STORAGE_POOLS_MPATH    | \
-                  VIR_CONNECT_LIST_STORAGE_POOLS_RBD      | \
-+                 VIR_CONNECT_LIST_STORAGE_POOLS_VITASTOR | \
-                  VIR_CONNECT_LIST_STORAGE_POOLS_SHEEPDOG | \
-                  VIR_CONNECT_LIST_STORAGE_POOLS_GLUSTER  | \
-                  VIR_CONNECT_LIST_STORAGE_POOLS_ZFS      | \
-diff --git a/src/conf/storage_source_conf.c b/src/conf/storage_source_conf.c
-index 087de1eaf2..acc046816b 100644
---- a/src/conf/storage_source_conf.c
-+++ b/src/conf/storage_source_conf.c
-@@ -89,6 +89,7 @@ VIR_ENUM_IMPL(virStorageNetProtocol,
-               "ssh",
-               "vxhs",
-               "nfs",
-+              "vitastor",
- );
- 
- 
-@@ -1313,6 +1314,7 @@ virStorageSourceNetworkDefaultPort(virStorageNetProtocol protocol)
-         case VIR_STORAGE_NET_PROTOCOL_GLUSTER:
-             return 24007;
- 
-+        case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
-         case VIR_STORAGE_NET_PROTOCOL_RBD:
-             /* we don't provide a default for RBD */
-             return 0;
-diff --git a/src/conf/storage_source_conf.h b/src/conf/storage_source_conf.h
-index fc868b31af..1c9fbc02b9 100644
---- a/src/conf/storage_source_conf.h
-+++ b/src/conf/storage_source_conf.h
-@@ -130,6 +130,7 @@ typedef enum {
-     VIR_STORAGE_NET_PROTOCOL_SSH,
-     VIR_STORAGE_NET_PROTOCOL_VXHS,
-     VIR_STORAGE_NET_PROTOCOL_NFS,
-+    VIR_STORAGE_NET_PROTOCOL_VITASTOR,
- 
-     VIR_STORAGE_NET_PROTOCOL_LAST
- } virStorageNetProtocol;
-diff --git a/src/conf/virstorageobj.c b/src/conf/virstorageobj.c
-index 59fa5da372..4739167f5f 100644
---- a/src/conf/virstorageobj.c
-+++ b/src/conf/virstorageobj.c
-@@ -1438,6 +1438,7 @@ virStoragePoolObjSourceFindDuplicateCb(const void *payload,
-             return 1;
-         break;
- 
-+    case VIR_STORAGE_POOL_VITASTOR:
-     case VIR_STORAGE_POOL_ISCSI_DIRECT:
-     case VIR_STORAGE_POOL_RBD:
-     case VIR_STORAGE_POOL_LAST:
-@@ -1921,6 +1922,8 @@ virStoragePoolObjMatch(virStoragePoolObj *obj,
-                (obj->def->type == VIR_STORAGE_POOL_MPATH))   ||
-               (MATCH(VIR_CONNECT_LIST_STORAGE_POOLS_RBD) &&
-                (obj->def->type == VIR_STORAGE_POOL_RBD))     ||
-+              (MATCH(VIR_CONNECT_LIST_STORAGE_POOLS_VITASTOR) &&
-+               (obj->def->type == VIR_STORAGE_POOL_VITASTOR)) ||
-               (MATCH(VIR_CONNECT_LIST_STORAGE_POOLS_SHEEPDOG) &&
-                (obj->def->type == VIR_STORAGE_POOL_SHEEPDOG)) ||
-               (MATCH(VIR_CONNECT_LIST_STORAGE_POOLS_GLUSTER) &&
-diff --git a/src/libvirt-storage.c b/src/libvirt-storage.c
-index db7660aac4..561df34709 100644
---- a/src/libvirt-storage.c
-+++ b/src/libvirt-storage.c
-@@ -94,6 +94,7 @@ virStoragePoolGetConnect(virStoragePoolPtr pool)
-  * VIR_CONNECT_LIST_STORAGE_POOLS_SCSI
-  * VIR_CONNECT_LIST_STORAGE_POOLS_MPATH
-  * VIR_CONNECT_LIST_STORAGE_POOLS_RBD
-+ * VIR_CONNECT_LIST_STORAGE_POOLS_VITASTOR
-  * VIR_CONNECT_LIST_STORAGE_POOLS_SHEEPDOG
-  * VIR_CONNECT_LIST_STORAGE_POOLS_GLUSTER
-  * VIR_CONNECT_LIST_STORAGE_POOLS_ZFS
-diff --git a/src/libxl/libxl_conf.c b/src/libxl/libxl_conf.c
-index 2b988157fa..9d0eb47b25 100644
---- a/src/libxl/libxl_conf.c
-+++ b/src/libxl/libxl_conf.c
-@@ -1069,6 +1069,7 @@ libxlMakeNetworkDiskSrcStr(virStorageSource *src,
-     case VIR_STORAGE_NET_PROTOCOL_SSH:
-     case VIR_STORAGE_NET_PROTOCOL_VXHS:
-     case VIR_STORAGE_NET_PROTOCOL_NFS:
-+    case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
-     case VIR_STORAGE_NET_PROTOCOL_LAST:
-     case VIR_STORAGE_NET_PROTOCOL_NONE:
-         virReportError(VIR_ERR_NO_SUPPORT,
-diff --git a/src/libxl/xen_xl.c b/src/libxl/xen_xl.c
-index b2ff0edcf2..5d9f50c88c 100644
---- a/src/libxl/xen_xl.c
-+++ b/src/libxl/xen_xl.c
-@@ -1461,6 +1461,7 @@ xenFormatXLDiskSrcNet(virStorageSource *src)
-     case VIR_STORAGE_NET_PROTOCOL_SSH:
-     case VIR_STORAGE_NET_PROTOCOL_VXHS:
-     case VIR_STORAGE_NET_PROTOCOL_NFS:
-+    case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
-     case VIR_STORAGE_NET_PROTOCOL_LAST:
-     case VIR_STORAGE_NET_PROTOCOL_NONE:
-         virReportError(VIR_ERR_NO_SUPPORT,
-diff --git a/src/qemu/qemu_block.c b/src/qemu/qemu_block.c
-index a7062d3e96..4f2b6cc0e6 100644
---- a/src/qemu/qemu_block.c
-+++ b/src/qemu/qemu_block.c
-@@ -743,6 +743,38 @@ qemuBlockStorageSourceGetRBDProps(virStorageSource *src,
- }
- 
- 
-+static virJSONValue *
-+qemuBlockStorageSourceGetVitastorProps(virStorageSource *src)
-+{
-+    virJSONValue *ret = NULL;
-+    virStorageNetHostDef *host;
-+    size_t i;
-+    g_auto(virBuffer) buf = VIR_BUFFER_INITIALIZER;
-+    g_autofree char *etcd = NULL;
-+
-+    for (i = 0; i < src->nhosts; i++) {
-+        host = src->hosts + i;
-+        if ((virStorageNetHostTransport)host->transport != VIR_STORAGE_NET_HOST_TRANS_TCP) {
-+            return NULL;
-+        }
-+        virBufferAsprintf(&buf, i > 0 ? ",%s:%u" : "%s:%u", host->name, host->port);
-+    }
-+    if (src->nhosts > 0) {
-+        etcd = virBufferContentAndReset(&buf);
-+    }
-+
-+    if (virJSONValueObjectAdd(&ret,
-+                              "S:etcd-host", etcd,
-+                              "S:etcd-prefix", src->query,
-+                              "S:config-path", src->configFile,
-+                              "s:image", src->path,
-+                              NULL) < 0)
-+        return NULL;
-+
-+    return ret;
-+}
-+
-+
- static virJSONValue *
- qemuBlockStorageSourceGetSshProps(virStorageSource *src)
- {
-@@ -1093,6 +1125,12 @@ qemuBlockStorageSourceGetBackendProps(virStorageSource *src,
-                 return NULL;
-             break;
- 
-+        case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
-+            driver = "vitastor";
-+            if (!(fileprops = qemuBlockStorageSourceGetVitastorProps(src)))
-+                return NULL;
-+            break;
-+
-         case VIR_STORAGE_NET_PROTOCOL_SSH:
-             driver = "ssh";
-             if (!(fileprops = qemuBlockStorageSourceGetSshProps(src)))
-@@ -1996,6 +2034,7 @@ qemuBlockGetBackingStoreString(virStorageSource *src,
- 
-             case VIR_STORAGE_NET_PROTOCOL_SHEEPDOG:
-             case VIR_STORAGE_NET_PROTOCOL_RBD:
-+            case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
-             case VIR_STORAGE_NET_PROTOCOL_VXHS:
-             case VIR_STORAGE_NET_PROTOCOL_NFS:
-             case VIR_STORAGE_NET_PROTOCOL_SSH:
-@@ -2376,6 +2415,12 @@ qemuBlockStorageSourceCreateGetStorageProps(virStorageSource *src,
-                 return -1;
-             break;
- 
-+        case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
-+            driver = "vitastor";
-+            if (!(location = qemuBlockStorageSourceGetVitastorProps(src)))
-+                return -1;
-+            break;
-+
-         case VIR_STORAGE_NET_PROTOCOL_SSH:
-             if (srcPriv->nbdkitProcess) {
-                 /* disk creation not yet supported with nbdkit, and even if it
-diff --git a/src/qemu/qemu_domain.c b/src/qemu/qemu_domain.c
-index ac56fc7cb4..b80b311eea 100644
---- a/src/qemu/qemu_domain.c
-+++ b/src/qemu/qemu_domain.c
-@@ -4677,7 +4677,8 @@ qemuDomainValidateStorageSource(virStorageSource *src,
-     if (src->query &&
-         (actualType != VIR_STORAGE_TYPE_NETWORK ||
-          (src->protocol != VIR_STORAGE_NET_PROTOCOL_HTTPS &&
--          src->protocol != VIR_STORAGE_NET_PROTOCOL_HTTP))) {
-+          src->protocol != VIR_STORAGE_NET_PROTOCOL_HTTP &&
-+          src->protocol != VIR_STORAGE_NET_PROTOCOL_VITASTOR))) {
-         virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s",
-                        _("query is supported only with HTTP(S) protocols"));
-         return -1;
-@@ -4693,6 +4694,7 @@ qemuDomainValidateStorageSource(virStorageSource *src,
-         case VIR_STORAGE_NET_PROTOCOL_ISCSI:
-         case VIR_STORAGE_NET_PROTOCOL_NBD:
-         case VIR_STORAGE_NET_PROTOCOL_RBD:
-+        case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
-         case VIR_STORAGE_NET_PROTOCOL_SSH:
-         case VIR_STORAGE_NET_PROTOCOL_SHEEPDOG:
-             break;
-@@ -9103,6 +9105,7 @@ qemuDomainPrepareStorageSourceTLS(virStorageSource *src,
-         break;
- 
-     case VIR_STORAGE_NET_PROTOCOL_RBD:
-+    case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
-     case VIR_STORAGE_NET_PROTOCOL_SHEEPDOG:
-     case VIR_STORAGE_NET_PROTOCOL_GLUSTER:
-     case VIR_STORAGE_NET_PROTOCOL_ISCSI:
-diff --git a/src/qemu/qemu_nbdkit.c b/src/qemu/qemu_nbdkit.c
-index c1bc6bc363..9d3433bf75 100644
---- a/src/qemu/qemu_nbdkit.c
-+++ b/src/qemu/qemu_nbdkit.c
-@@ -869,6 +869,7 @@ qemuNbdkitInitStorageSource(qemuNbdkitCaps *caps WITHOUT_NBDKIT_UNUSED,
-         case VIR_STORAGE_NET_PROTOCOL_NONE:
-         case VIR_STORAGE_NET_PROTOCOL_NBD:
-         case VIR_STORAGE_NET_PROTOCOL_RBD:
-+        case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
-         case VIR_STORAGE_NET_PROTOCOL_SHEEPDOG:
-         case VIR_STORAGE_NET_PROTOCOL_GLUSTER:
-         case VIR_STORAGE_NET_PROTOCOL_ISCSI:
-@@ -1133,6 +1134,7 @@ qemuNbdkitProcessBuildCommand(qemuNbdkitProcess *proc)
-         case VIR_STORAGE_NET_PROTOCOL_NONE:
-         case VIR_STORAGE_NET_PROTOCOL_NBD:
-         case VIR_STORAGE_NET_PROTOCOL_RBD:
-+        case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
-         case VIR_STORAGE_NET_PROTOCOL_SHEEPDOG:
-         case VIR_STORAGE_NET_PROTOCOL_GLUSTER:
-         case VIR_STORAGE_NET_PROTOCOL_ISCSI:
-diff --git a/src/qemu/qemu_snapshot.c b/src/qemu/qemu_snapshot.c
-index 302775af92..9d8fa0ded2 100644
---- a/src/qemu/qemu_snapshot.c
-+++ b/src/qemu/qemu_snapshot.c
-@@ -665,6 +665,7 @@ qemuSnapshotPrepareDiskExternalInactive(virDomainSnapshotDiskDef *snapdisk,
-         case VIR_STORAGE_NET_PROTOCOL_NONE:
-         case VIR_STORAGE_NET_PROTOCOL_NBD:
-         case VIR_STORAGE_NET_PROTOCOL_RBD:
-+        case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
-         case VIR_STORAGE_NET_PROTOCOL_SHEEPDOG:
-         case VIR_STORAGE_NET_PROTOCOL_GLUSTER:
-         case VIR_STORAGE_NET_PROTOCOL_ISCSI:
-@@ -890,6 +891,7 @@ qemuSnapshotPrepareDiskInternal(virDomainDiskDef *disk,
-         case VIR_STORAGE_NET_PROTOCOL_NONE:
-         case VIR_STORAGE_NET_PROTOCOL_NBD:
-         case VIR_STORAGE_NET_PROTOCOL_RBD:
-+        case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
-         case VIR_STORAGE_NET_PROTOCOL_SHEEPDOG:
-         case VIR_STORAGE_NET_PROTOCOL_GLUSTER:
-         case VIR_STORAGE_NET_PROTOCOL_ISCSI:
-diff --git a/src/storage/storage_driver.c b/src/storage/storage_driver.c
-index e19e032427..59f91f4710 100644
---- a/src/storage/storage_driver.c
-+++ b/src/storage/storage_driver.c
-@@ -1626,6 +1626,7 @@ storageVolLookupByPathCallback(virStoragePoolObj *obj,
- 
-         case VIR_STORAGE_POOL_GLUSTER:
-         case VIR_STORAGE_POOL_RBD:
-+        case VIR_STORAGE_POOL_VITASTOR:
-         case VIR_STORAGE_POOL_SHEEPDOG:
-         case VIR_STORAGE_POOL_ZFS:
-         case VIR_STORAGE_POOL_LAST:
-diff --git a/src/storage_file/storage_source_backingstore.c b/src/storage_file/storage_source_backingstore.c
-index 821378883c..2211f6891b 100644
---- a/src/storage_file/storage_source_backingstore.c
-+++ b/src/storage_file/storage_source_backingstore.c
-@@ -264,6 +264,75 @@ virStorageSourceParseRBDColonString(const char *rbdstr,
- }
- 
- 
-+static int
-+virStorageSourceParseVitastorColonString(const char *colonstr,
-+                                         virStorageSource *src)
-+{
-+    char *p, *e, *next;
-+    g_autofree char *options = NULL;
-+
-+    /* optionally skip the "vitastor:" prefix if provided */
-+    if (STRPREFIX(colonstr, "vitastor:"))
-+        colonstr += strlen("vitastor:");
-+
-+    options = g_strdup(colonstr);
-+
-+    p = options;
-+    while (*p) {
-+        /* find : delimiter or end of string */
-+        for (e = p; *e && *e != ':'; ++e) {
-+            if (*e == '\\') {
-+                e++;
-+                if (*e == '\0')
-+                    break;
-+            }
-+        }
-+        if (*e == '\0') {
-+            next = e;    /* last kv pair */
-+        } else {
-+            next = e + 1;
-+            *e = '\0';
-+        }
-+
-+        if (STRPREFIX(p, "image=")) {
-+            src->path = g_strdup(p + strlen("image="));
-+        } else if (STRPREFIX(p, "etcd-prefix=")) {
-+            src->query = g_strdup(p + strlen("etcd-prefix="));
-+        } else if (STRPREFIX(p, "config-path=")) {
-+            src->configFile = g_strdup(p + strlen("config-path="));
-+        } else if (STRPREFIX(p, "etcd-host=")) {
-+            char *h, *sep;
-+
-+            h = p + strlen("etcd-host=");
-+            while (h < e) {
-+                for (sep = h; sep < e; ++sep) {
-+                    if (*sep == '\\' && (sep[1] == ',' ||
-+                                         sep[1] == ';' ||
-+                                         sep[1] == ' ')) {
-+                        *sep = '\0';
-+                        sep += 2;
-+                        break;
-+                    }
-+                }
-+
-+                if (virStorageSourceRBDAddHost(src, h) < 0)
-+                    return -1;
-+
-+                h = sep;
-+            }
-+        }
-+
-+        p = next;
-+    }
-+
-+    if (!src->path) {
-+        return -1;
-+    }
-+
-+    return 0;
-+}
-+
-+
- static int
- virStorageSourceParseNBDColonString(const char *nbdstr,
-                                     virStorageSource *src)
-@@ -379,6 +448,11 @@ virStorageSourceParseBackingColon(virStorageSource *src,
-             return -1;
-         break;
- 
-+    case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
-+        if (virStorageSourceParseVitastorColonString(path, src) < 0)
-+            return -1;
-+        break;
-+
-     case VIR_STORAGE_NET_PROTOCOL_SHEEPDOG:
-     case VIR_STORAGE_NET_PROTOCOL_LAST:
-     case VIR_STORAGE_NET_PROTOCOL_NONE:
-@@ -953,6 +1027,54 @@ virStorageSourceParseBackingJSONRBD(virStorageSource *src,
-     return 0;
- }
- 
-+static int
-+virStorageSourceParseBackingJSONVitastor(virStorageSource *src,
-+                                         virJSONValue *json,
-+                                         const char *jsonstr G_GNUC_UNUSED,
-+                                         int opaque G_GNUC_UNUSED)
-+{
-+    const char *filename;
-+    const char *image = virJSONValueObjectGetString(json, "image");
-+    const char *conf = virJSONValueObjectGetString(json, "config-path");
-+    const char *etcd_prefix = virJSONValueObjectGetString(json, "etcd-prefix");
-+    virJSONValue *servers = virJSONValueObjectGetArray(json, "server");
-+    size_t nservers;
-+    size_t i;
-+
-+    src->type = VIR_STORAGE_TYPE_NETWORK;
-+    src->protocol = VIR_STORAGE_NET_PROTOCOL_VITASTOR;
-+
-+    /* legacy syntax passed via 'filename' option */
-+    if ((filename = virJSONValueObjectGetString(json, "filename")))
-+        return virStorageSourceParseVitastorColonString(filename, src);
-+
-+    if (!image) {
-+        virReportError(VIR_ERR_INVALID_ARG, "%s",
-+                       _("missing image name in Vitastor backing volume "
-+                         "JSON specification"));
-+        return -1;
-+    }
-+
-+    src->path = g_strdup(image);
-+    src->configFile = g_strdup(conf);
-+    src->query = g_strdup(etcd_prefix);
-+
-+    if (servers) {
-+        nservers = virJSONValueArraySize(servers);
-+
-+        src->hosts = g_new0(virStorageNetHostDef, nservers);
-+        src->nhosts = nservers;
-+
-+        for (i = 0; i < nservers; i++) {
-+            if (virStorageSourceParseBackingJSONInetSocketAddress(src->hosts + i,
-+                                                                  virJSONValueArrayGet(servers, i)) < 0)
-+                return -1;
-+        }
-+    }
-+
-+    return 0;
-+}
-+
- static int
- virStorageSourceParseBackingJSONRaw(virStorageSource *src,
-                                     virJSONValue *json,
-@@ -1130,6 +1252,7 @@ static const struct virStorageSourceJSONDriverParser jsonParsers[] = {
-     {"sheepdog", false, virStorageSourceParseBackingJSONSheepdog, 0},
-     {"ssh", false, virStorageSourceParseBackingJSONSSH, 0},
-     {"rbd", false, virStorageSourceParseBackingJSONRBD, 0},
-+    {"vitastor", false, virStorageSourceParseBackingJSONVitastor, 0},
-     {"raw", true, virStorageSourceParseBackingJSONRaw, 0},
-     {"nfs", false, virStorageSourceParseBackingJSONNFS, 0},
-     {"vxhs", false, virStorageSourceParseBackingJSONVxHS, 0},
-diff --git a/src/test/test_driver.c b/src/test/test_driver.c
-index 1165689de7..bba846351c 100644
---- a/src/test/test_driver.c
-+++ b/src/test/test_driver.c
-@@ -7345,6 +7345,7 @@ testStorageVolumeTypeForPool(int pooltype)
-     case VIR_STORAGE_POOL_ISCSI_DIRECT:
-     case VIR_STORAGE_POOL_GLUSTER:
-     case VIR_STORAGE_POOL_RBD:
-+    case VIR_STORAGE_POOL_VITASTOR:
-         return VIR_STORAGE_VOL_NETWORK;
-     case VIR_STORAGE_POOL_LOGICAL:
-     case VIR_STORAGE_POOL_DISK:
-diff --git a/tests/storagepoolcapsschemadata/poolcaps-fs.xml b/tests/storagepoolcapsschemadata/poolcaps-fs.xml
-index eee75af746..8bd0a57bdd 100644
---- a/tests/storagepoolcapsschemadata/poolcaps-fs.xml
-+++ b/tests/storagepoolcapsschemadata/poolcaps-fs.xml
-@@ -204,4 +204,11 @@
-       </enum>
-     </volOptions>
-   </pool>
-+  <pool type='vitastor' supported='no'>
-+    <volOptions>
-+      <defaultFormat type='raw'/>
-+      <enum name='targetFormatType'>
-+      </enum>
-+    </volOptions>
-+  </pool>
- </storagepoolCapabilities>
-diff --git a/tests/storagepoolcapsschemadata/poolcaps-full.xml b/tests/storagepoolcapsschemadata/poolcaps-full.xml
-index 805950a937..852df0de16 100644
---- a/tests/storagepoolcapsschemadata/poolcaps-full.xml
-+++ b/tests/storagepoolcapsschemadata/poolcaps-full.xml
-@@ -204,4 +204,11 @@
-       </enum>
-     </volOptions>
-   </pool>
-+  <pool type='vitastor' supported='yes'>
-+    <volOptions>
-+      <defaultFormat type='raw'/>
-+      <enum name='targetFormatType'>
-+      </enum>
-+    </volOptions>
-+  </pool>
- </storagepoolCapabilities>
-diff --git a/tests/storagepoolxml2argvtest.c b/tests/storagepoolxml2argvtest.c
-index d5c2531ab8..b19308ac38 100644
---- a/tests/storagepoolxml2argvtest.c
-+++ b/tests/storagepoolxml2argvtest.c
-@@ -57,6 +57,7 @@ testCompareXMLToArgvFiles(bool shouldFail,
-     case VIR_STORAGE_POOL_GLUSTER:
-     case VIR_STORAGE_POOL_ZFS:
-     case VIR_STORAGE_POOL_VSTORAGE:
-+    case VIR_STORAGE_POOL_VITASTOR:
-     case VIR_STORAGE_POOL_LAST:
-     default:
-         VIR_TEST_DEBUG("pool type '%s' has no xml2argv test", defTypeStr);
-diff --git a/tools/virsh-pool.c b/tools/virsh-pool.c
-index 2010ef1356..072e2ff9e8 100644
---- a/tools/virsh-pool.c
-+++ b/tools/virsh-pool.c
-@@ -1187,6 +1187,9 @@ cmdPoolList(vshControl *ctl, const vshCmd *cmd G_GNUC_UNUSED)
-             case VIR_STORAGE_POOL_VSTORAGE:
-                 flags |= VIR_CONNECT_LIST_STORAGE_POOLS_VSTORAGE;
-                 break;
-+            case VIR_STORAGE_POOL_VITASTOR:
-+                flags |= VIR_CONNECT_LIST_STORAGE_POOLS_VITASTOR;
-+                break;
-             case VIR_STORAGE_POOL_LAST:
-                 break;
-             }
--- 
-2.51.0
-

rpm/qemu-kvm-10.1-el10.spec.patch

@@ -1,91 +0,0 @@
---- qemu-kvm.spec.orig	2026-08-24 00:00:00.000000000 +0000
-+++ qemu-kvm.spec	2026-09-05 16:48:22.226624885 +0000
-@@ -102,8 +102,6 @@
- 
- 
- %global target_list %{kvm_target}-softmmu
--%global block_drivers_rw_list qcow2,raw,file,host_device,nbd,iscsi,rbd,blkdebug,luks,null-co,nvme,copy-on-read,throttle,compress,virtio-blk-vhost-vdpa,virtio-blk-vfio-pci,virtio-blk-vhost-user,io_uring,nvme-io_uring
--%global block_drivers_ro_list vdi,vmdk,vhdx,vpc,https
- %define qemudocdir %{_docdir}/%{name}
- %global firmwaredirs "%{_datadir}/qemu-firmware:%{_datadir}/ipxe/qemu:%{_datadir}/seavgabios:%{_datadir}/seabios"
- 
-@@ -134,6 +132,7 @@ Requires: %{name}-block-blkio = %{epoch}
- %if %{have_block_rbd} \
- Requires: %{name}-block-rbd = %{epoch}:%{version}-%{release}     \
- %endif \
-+Requires: %{name}-block-vitastor = %{epoch}:%{version}-%{release}\
- Requires: %{name}-audio-pa = %{epoch}:%{version}-%{release}
- 
- # Since SPICE is removed from RHEL-9, the following Obsoletes:
-@@ -158,7 +157,7 @@ Obsoletes: %{name}-block-ssh <= %{epoch}
- Summary: QEMU is a machine emulator and virtualizer
- Name: qemu-kvm
- Version: 10.1.0
--Release: 16%{?rcrel}%{?dist}%{?cc_suffix}.5.alma.1
-+Release: 16.vitastor%{?rcrel}%{?dist}%{?cc_suffix}.5.alma.1
- # Epoch because we pushed a qemu-1.0 package. AIUI this can't ever be dropped
- # Epoch 15 used for RHEL 8
- # Epoch 17 used for RHEL 9 (due to release versioning offset in RHEL 8.5)
-@@ -181,6 +180,7 @@ Source28: 95-kvm-memlock.conf
- Source30: kvm-s390x.conf
- Source31: kvm-x86.conf
- Source36: README.tests
-+Source37: qemu-vitastor.c
- 
- 
- Patch0004: 0004-Initial-redhat-build.patch
-@@ -471,6 +471,7 @@ Patch149: kvm-file-posix-populate-pwrite
- Patch150: kvm-block-use-pwrite_zeroes_alignment-when-writing-first.patch
- # For RHEL-207389 - qemu-img create/convert fails on target block device with 4k sector size [rhel-10.2.z]
- Patch151: kvm-iotests-add-Linux-loop-device-image-creation-test.patch
-+Patch152: qemu-10.1-vitastor.patch
- 
- # AlmaLinux Patch
- Patch2001: 2001-Add-ppc64-support.patch
-@@ -509,6 +510,7 @@ BuildRequires: libcurl-devel
- %if %{have_block_rbd}
- BuildRequires: librbd-devel
- %endif
-+BuildRequires: vitastor-client-devel
- # We need both because the 'stap' binary is probed for by configure
- BuildRequires: systemtap
- BuildRequires: systemtap-sdt-devel
-@@ -729,6 +731,13 @@ using the rbd protocol.
- %endif
- 
- 
-+%package  block-vitastor
-+Summary: QEMU Vitastor block driver
-+Requires: %{name}-common%{?_isa} = %{epoch}:%{version}-%{release}
-+%description block-vitastor
-+This package provides the additional Vitastor block driver for QEMU.
-+
-+
- %package  audio-pa
- Summary: QEMU PulseAudio audio driver
- Requires: %{name}-common%{?_isa} = %{epoch}:%{version}-%{release}
-@@ -831,6 +840,7 @@ This package provides the additional D-B
- %prep
- %setup -q -n qemu-%{version}%{?rcstr}
- %autopatch -p1
-+cp %{SOURCE37} ./block/vitastor.c
- 
- %global qemu_kvm_build qemu_kvm_build
- mkdir -p %{qemu_kvm_build}
-@@ -1067,6 +1077,7 @@ run_configure \
- %if %{have_block_rbd}
-   --enable-rbd \
- %endif
-+  --enable-vitastor \
- %if %{have_librdma}
-   --enable-rdma \
- %endif
-@@ -1564,6 +1575,8 @@ useradd -r -u 107 -g qemu -G kvm -d / -s
- %files block-rbd
- %{_libdir}/%{name}/block-rbd.so
- %endif
-+%files block-vitastor
-+%{_libdir}/%{name}/block-vitastor.so
- %files audio-pa
- %{_libdir}/%{name}/audio-pa.so
- 

rpm/vitastor-el10.Dockerfile

@@ -11,10 +11,7 @@ WORKDIR /root
 RUN sed -i 's/enabled=0/enabled=1/' /etc/yum.repos.d/*.repo
 RUN dnf -y install epel-release dnf-plugins-core
 RUN dnf -y install https://vitastor.io/rpms/centos/10/vitastor-release-1.0-1.el10.noarch.rpm
-RUN sed -i 's/metalink=/#metalink=/' /etc/yum.repos.d/epel.repo
-RUN sed -i 's!#baseurl=.*epel/!baseurl=https://mirror.yandex.ru/epel/!' /etc/yum.repos.d/epel.repo
-RUN dnf -y install gcc-c++ gperftools-devel fio nodejs rpm-build jerasure-devel isa-l-devel openssl-devel libisa-l_crypto-devel \
-    gf-complete-devel rdma-core-devel cmake libnl3-devel c-ares-devel
+RUN dnf -y install gcc-c++ gperftools-devel fio nodejs rpm-build jerasure-devel isa-l-devel gf-complete-devel rdma-core-devel cmake libnl3-devel c-ares-devel
 RUN dnf download --source fio
 RUN rpm --nomd5 -i fio*.src.rpm
 RUN cd ~/rpmbuild/SPECS && dnf builddep -y --spec fio.spec

rpm/vitastor-el10.spec

@@ -1,18 +1,17 @@
 Name:           vitastor
-Version:        3.2.2
+Version:        3.0.15
 Release:        1%{?dist}
 Summary:        Vitastor, a fast software-defined clustered block storage
 
 License:        Vitastor Network Public License 1.1
 URL:            https://vitastor.io/
-Source0:        vitastor-3.2.2.el10.tar.gz
+Source0:        vitastor-3.0.15.el10.tar.gz
 
 BuildRequires:  gperftools-devel
 BuildRequires:  gcc-c++
 BuildRequires:  nodejs >= 10
 BuildRequires:  jerasure-devel
 BuildRequires:  isa-l-devel
-BuildRequires:  libisa-l_crypto-devel
 BuildRequires:  gf-complete-devel
 BuildRequires:  rdma-core-devel
 BuildRequires:  cmake
@@ -162,8 +161,6 @@ chown vitastor:vitastor /var/lib/vitastor
 %_bindir/vitastor-nbd
 %_bindir/vitastor-ublk
 %_bindir/vitastor-nfs
-%_sbindir/mount.vitastorfs
-%_sbindir/mount.vitastorblk
 %_bindir/vitastor-cli
 %_bindir/vitastor-rm
 %_bindir/vitastor-kv

rpm/vitastor-el7.Dockerfile

@@ -13,11 +13,9 @@ RUN rm -f /etc/yum.repos.d/CentOS-Media.repo
 RUN sed -i 's/^mirrorlist=/#mirrorlist=/; s!#baseurl=http://mirror.centos.org/centos/\$releasever!baseurl=http://vault.centos.org/7.9.2009!' /etc/yum.repos.d/*.repo
 RUN yum -y --enablerepo=extras install centos-release-scl epel-release yum-utils rpm-build
 RUN perl -i -pe 's!mirrorlist=!#mirrorlist=!s; s!#\s*baseurl=http://mirror.centos.org!baseurl=http://vault.centos.org!' /etc/yum.repos.d/CentOS-SCLo-scl*.repo
-RUN yum -y install https://vitastor.io/rpms/centos/7/vitastor-release-1.0-2.el7.noarch.rpm
-RUN sed -i 's/metalink=/#metalink=/' /etc/yum.repos.d/epel.repo
-RUN sed -i 's!#baseurl=.*epel/!baseurl=https://archives.fedoraproject.org/pub/archive/epel/!' /etc/yum.repos.d/epel.repo
+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 gcc make cmake gperftools-devel \
-    fio rh-nodejs12 jerasure-devel libisa-l-devel openssl3-devel libisa-l_crypto-devel gf-complete-devel rdma-core-devel libnl3-devel c-ares-devel
+    fio rh-nodejs12 jerasure-devel libisa-l-devel gf-complete-devel rdma-core-devel libnl3-devel c-ares-devel
 RUN yumdownloader --disablerepo=centos-sclo-rh --source fio
 RUN rpm --nomd5 -i fio*.src.rpm
 RUN rm -f /etc/yum.repos.d/CentOS-Media.repo

rpm/vitastor-el7.spec

@@ -1,25 +1,23 @@
 Name:           vitastor
-Version:        3.2.2
+Version:        3.0.15
 Release:        1%{?dist}
 Summary:        Vitastor, a fast software-defined clustered block storage
 
 License:        Vitastor Network Public License 1.1
 URL:            https://vitastor.io/
-Source0:        vitastor-3.2.2.el7.tar.gz
+Source0:        vitastor-3.0.15.el7.tar.gz
 
 BuildRequires:  gperftools-devel
 BuildRequires:  devtoolset-9-gcc-c++
 BuildRequires:  rh-nodejs12
 BuildRequires:  rh-nodejs12-npm
 BuildRequires:  jerasure-devel
-BuildRequires:  openssl3-devel
 BuildRequires:  libisa-l-devel
-BuildRequires:  libisa-l_crypto-devel
 BuildRequires:  gf-complete-devel
 BuildRequires:  rdma-core-devel
 BuildRequires:  cmake3
 BuildRequires:  libnl3-devel
-BuildRequires:  c-ares-devel >= 1.16
+BuildRequires:  c-ares-devel
 Requires:       vitastor-osd = %{version}-%{release}
 Requires:       vitastor-mon = %{version}-%{release}
 Requires:       vitastor-client = %{version}-%{release}
@@ -107,7 +105,7 @@ Vitastor storage plugin for OpenNebula.
 
 %build
 . /opt/rh/devtoolset-9/enable
-%cmake3 . -DOPENSSL_CRYPTO_LIBRARY=%{_libdir}/openssl3/libcrypto.so -DOPENSSL_SSL_LIBRARY=%{_libdir}/openssl3/libssl.so -DOPENSSL_INCLUDE_DIR=%{_includedir}/openssl3
+%cmake3 .
 %make_build
 
 
@@ -169,8 +167,6 @@ chown vitastor:vitastor /var/lib/vitastor
 %_bindir/vitastor-nbd
 %_bindir/vitastor-ublk
 %_bindir/vitastor-nfs
-%_sbindir/mount.vitastorfs
-%_sbindir/mount.vitastorblk
 %_bindir/vitastor-cli
 %_bindir/vitastor-rm
 %_bindir/vitastor-kv

rpm/vitastor-el8.Dockerfile

@@ -11,11 +11,9 @@ RUN rm -f /etc/yum.repos.d/CentOS-Media.repo
 RUN sed -i 's/^mirrorlist=/#mirrorlist=/; s!#baseurl=http://mirror.centos.org/!baseurl=http://vault.centos.org/!' /etc/yum.repos.d/*.repo
 RUN dnf -y install centos-release-advanced-virtualization epel-release dnf-plugins-core
 RUN sed -i 's/^mirrorlist=/#mirrorlist=/; s!#baseurl=.*!baseurl=http://vault.centos.org/centos/8.4.2105/virt/$basearch/$avdir/!; s!^baseurl=.*Source/.*!baseurl=http://vault.centos.org/centos/8.4.2105/virt/Source/advanced-virtualization/!' /etc/yum.repos.d/CentOS-Advanced-Virtualization.repo
-RUN yum -y install https://vitastor.io/rpms/centos/8/vitastor-release-1.0-2.el8.noarch.rpm
-RUN sed -i 's/metalink=/#metalink=/' /etc/yum.repos.d/epel.repo
-RUN sed -i 's!#baseurl=.*epel/!baseurl=https://mirror.yandex.ru/epel/!' /etc/yum.repos.d/epel.repo
-RUN dnf -y install gcc-toolset-9 gcc-toolset-9-gcc-c++ gperftools-devel fio nodejs rpm-build jerasure-devel \
-    libisa-l-devel openssl3-devel libisa-l_crypto-devel gf-complete-devel libibverbs-devel libarchive cmake libnl3-devel c-ares-devel
+RUN yum -y install https://vitastor.io/rpms/centos/8/vitastor-release-1.0-1.el8.noarch.rpm
+RUN dnf -y install gcc-toolset-9 gcc-toolset-9-gcc-c++ gperftools-devel \
+    fio nodejs rpm-build jerasure-devel libisa-l-devel gf-complete-devel libibverbs-devel libarchive cmake libnl3-devel c-ares-devel
 RUN dnf download --source fio
 RUN rpm --nomd5 -i fio*.src.rpm
 RUN cd ~/rpmbuild/SPECS && dnf builddep -y --enablerepo=powertools --spec fio.spec

rpm/vitastor-el8.spec

@@ -1,24 +1,22 @@
 Name:           vitastor
-Version:        3.2.2
+Version:        3.0.15
 Release:        1%{?dist}
 Summary:        Vitastor, a fast software-defined clustered block storage
 
 License:        Vitastor Network Public License 1.1
 URL:            https://vitastor.io/
-Source0:        vitastor-3.2.2.el8.tar.gz
+Source0:        vitastor-3.0.15.el8.tar.gz
 
 BuildRequires:  gperftools-devel
 BuildRequires:  gcc-toolset-9-gcc-c++
 BuildRequires:  nodejs >= 10
 BuildRequires:  jerasure-devel
-BuildRequires:  openssl3-devel
 BuildRequires:  libisa-l-devel
-BuildRequires:  libisa-l_crypto-devel
 BuildRequires:  gf-complete-devel
 BuildRequires:  rdma-core-devel
 BuildRequires:  cmake
 BuildRequires:  libnl3-devel
-BuildRequires:  c-ares-devel >= 1.16
+BuildRequires:  c-ares-devel
 Requires:       vitastor-osd = %{version}-%{release}
 Requires:       vitastor-mon = %{version}-%{release}
 Requires:       vitastor-client = %{version}-%{release}
@@ -105,7 +103,7 @@ Vitastor storage plugin for OpenNebula.
 
 %build
 . /opt/rh/gcc-toolset-9/enable
-%cmake . -DOPENSSL_CRYPTO_LIBRARY=%{_libdir}/openssl3/libcrypto.so -DOPENSSL_SSL_LIBRARY=%{_libdir}/openssl3/libssl.so -DOPENSSL_INCLUDE_DIR=%{_includedir}/openssl3
+%cmake .
 %make_build
 
 
@@ -166,8 +164,6 @@ chown vitastor:vitastor /var/lib/vitastor
 %_bindir/vitastor-nbd
 %_bindir/vitastor-ublk
 %_bindir/vitastor-nfs
-%_sbindir/mount.vitastorfs
-%_sbindir/mount.vitastorblk
 %_bindir/vitastor-cli
 %_bindir/vitastor-rm
 %_bindir/vitastor-kv

rpm/vitastor-el9.Dockerfile

@@ -10,10 +10,7 @@ WORKDIR /root
 RUN sed -i 's/enabled=0/enabled=1/' /etc/yum.repos.d/*.repo
 RUN dnf -y install epel-release dnf-plugins-core
 RUN dnf -y install https://vitastor.io/rpms/centos/9/vitastor-release-1.0-1.el9.noarch.rpm
-RUN sed -i 's/metalink=/#metalink=/' /etc/yum.repos.d/epel.repo
-RUN sed -i 's!#baseurl=.*epel/!baseurl=https://mirror.yandex.ru/epel/!' /etc/yum.repos.d/epel.repo
-RUN dnf -y install gcc-c++ gperftools-devel fio nodejs rpm-build jerasure-devel libisa-l-devel \
-    openssl-devel libisa-l_crypto-devel gf-complete-devel rdma-core-devel libarchive cmake libnl3-devel c-ares-devel
+RUN dnf -y install gcc-c++ gperftools-devel fio nodejs rpm-build jerasure-devel libisa-l-devel gf-complete-devel rdma-core-devel libarchive cmake libnl3-devel c-ares-devel
 RUN dnf download --source fio
 RUN rpm --nomd5 -i fio*.src.rpm
 RUN cd ~/rpmbuild/SPECS && dnf builddep -y --spec fio.spec

rpm/vitastor-el9.spec

@@ -1,24 +1,22 @@
 Name:           vitastor
-Version:        3.2.2
+Version:        3.0.15
 Release:        1%{?dist}
 Summary:        Vitastor, a fast software-defined clustered block storage
 
 License:        Vitastor Network Public License 1.1
 URL:            https://vitastor.io/
-Source0:        vitastor-3.2.2.el9.tar.gz
+Source0:        vitastor-3.0.15.el9.tar.gz
 
 BuildRequires:  gperftools-devel
 BuildRequires:  gcc-c++
 BuildRequires:  nodejs >= 10
 BuildRequires:  jerasure-devel
-BuildRequires:  openssl-devel
 BuildRequires:  libisa-l-devel
-BuildRequires:  libisa-l_crypto-devel
 BuildRequires:  gf-complete-devel
 BuildRequires:  rdma-core-devel
 BuildRequires:  cmake
 BuildRequires:  libnl3-devel
-BuildRequires:  c-ares-devel >= 1.16
+BuildRequires:  c-ares-devel
 Requires:       vitastor-osd = %{version}-%{release}
 Requires:       vitastor-mon = %{version}-%{release}
 Requires:       vitastor-client = %{version}-%{release}
@@ -163,8 +161,6 @@ chown vitastor:vitastor /var/lib/vitastor
 %_bindir/vitastor-nbd
 %_bindir/vitastor-ublk
 %_bindir/vitastor-nfs
-%_sbindir/mount.vitastorfs
-%_sbindir/mount.vitastorblk
 %_bindir/vitastor-cli
 %_bindir/vitastor-rm
 %_bindir/vitastor-kv

src/CMakeLists.txt

@@ -11,7 +11,6 @@ set(WITH_QEMU false CACHE BOOL "Build QEMU driver inside Vitastor source tree")
 set(WITH_FIO true CACHE BOOL "Build FIO driver")
 set(QEMU_PLUGINDIR qemu CACHE STRING "QEMU plugin directory suffix (qemu-kvm on RHEL)")
 set(WITH_ASAN false CACHE BOOL "Build with AddressSanitizer")
-set(WITH_UBSAN false CACHE BOOL "Build with UBSanitizer")
 set(WITH_SYSTEM_LIBURING false CACHE BOOL "Use system liburing")
 if("${CMAKE_INSTALL_PREFIX}" MATCHES "^/usr/local/?$")
 	if(EXISTS "/etc/debian_version")
@@ -21,17 +20,13 @@ if("${CMAKE_INSTALL_PREFIX}" MATCHES "^/usr/local/?$")
 endif()
 set(ENABLE_COVERAGE false CACHE BOOL "Enable code coverage")
 
-add_definitions(-DVITASTOR_VERSION="3.2.2")
+add_definitions(-DVITASTOR_VERSION="3.0.15")
 add_definitions(-D_GNU_SOURCE -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -Wall -Wno-sign-compare -Wno-comment -Wno-parentheses -Wno-pointer-arith -fdiagnostics-color=always -fno-omit-frame-pointer -fvisibility=hidden -I ${CMAKE_SOURCE_DIR}/src)
 add_link_options(-fno-omit-frame-pointer)
 if (${WITH_ASAN})
 	add_definitions(-fsanitize=address)
 	add_link_options(-fsanitize=address -fno-omit-frame-pointer)
 endif (${WITH_ASAN})
-if (${WITH_UBSAN})
-	add_definitions(-fsanitize=undefined -fno-sanitize=alignment)
-	add_link_options(-fsanitize=undefined -fno-sanitize=alignment)
-endif (${WITH_UBSAN})
 set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -fvisibility-inlines-hidden")
 set(CMAKE_CXX_FLAGS_MINSIZEREL "${CMAKE_CXX_FLAGS_MINSIZEREL} -fvisibility-inlines-hidden")
 set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO} -fvisibility-inlines-hidden")
@@ -55,13 +50,9 @@ string(REGEX REPLACE "([\\/\\-]D) *NDEBUG" "" CMAKE_C_FLAGS_RELEASE "${CMAKE_C_F
 string(REGEX REPLACE "([\\/\\-]D) *NDEBUG" "" CMAKE_C_FLAGS_MINSIZEREL "${CMAKE_C_FLAGS_MINSIZEREL}")
 string(REGEX REPLACE "([\\/\\-]D) *NDEBUG" "" CMAKE_C_FLAGS_RELWITHDEBINFO "${CMAKE_C_FLAGS_RELWITHDEBINFO}")
 
-macro(install_symlink _filepath _sympath)
-	get_filename_component(_symname ${_sympath} NAME)
-	get_filename_component(_installdir ${_sympath} PATH)
-	install(CODE "execute_process(COMMAND \"${CMAKE_COMMAND}\" -E create_symlink
-		${_filepath} ${CMAKE_CURRENT_BINARY_DIR}/${_symname}.link)")
-	install(FILES ${CMAKE_CURRENT_BINARY_DIR}/${_symname}.link DESTINATION ${_installdir} RENAME ${_symname})
-	install(CODE "file(REMOVE ${CMAKE_CURRENT_BINARY_DIR}/${_symname}.link)")
+macro(install_symlink filepath sympath)
+	install(CODE "execute_process(COMMAND ${CMAKE_COMMAND} -E create_symlink ${filepath} \$ENV{DESTDIR}${sympath})")
+	install(CODE "message(\"-- Created symlink: ${sympath} -> ${filepath}\")")
 endmacro(install_symlink)
 
 check_include_file("linux/nbd-netlink.h" HAVE_NBD_NETLINK_H)
@@ -87,11 +78,10 @@ if (RDMACM_LIBRARIES)
 	add_definitions(-DWITH_RDMACM)
 endif (RDMACM_LIBRARIES)
 
-find_package(OpenSSL 3.0 REQUIRED)
+find_package(OpenSSL REQUIRED)
 add_definitions(-DWITH_OPENSSL)
-include_directories(${OPENSSL_INCLUDE_DIR})
 
-pkg_check_modules(CARES REQUIRED libcares>=1.16)
+pkg_check_modules(CARES REQUIRED libcares)
 include_directories(${CARES_INCLUDE_DIRS})
 
 if (${WITH_SYSTEM_LIBURING})
@@ -126,7 +116,6 @@ add_subdirectory(kv)
 add_subdirectory(nfs)
 add_subdirectory(osd)
 add_subdirectory(test)
-add_subdirectory(util)
 
 ### Install
 
@@ -134,8 +123,6 @@ install(TARGETS vitastor-osd vitastor-disk vitastor-nbd vitastor-ublk vitastor-n
 install_symlink(vitastor-disk ${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_BINDIR}/vitastor-dump-journal)
 install_symlink(vitastor-cli ${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_BINDIR}/vitastor-rm)
 install_symlink(vitastor-cli ${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_BINDIR}/vita)
-install_symlink(../${CMAKE_INSTALL_BINDIR}/vitastor-nfs ${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_SBINDIR}/mount.vitastorfs)
-install_symlink(../${CMAKE_INSTALL_BINDIR}/vitastor-nfs ${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_SBINDIR}/mount.vitastorblk)
 install(
 	TARGETS vitastor_client vitastor_kv
 	LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}

src/blockstore/CMakeLists.txt

@@ -4,6 +4,7 @@ project(vitastor)
 
 # libvitastor_blk.a
 add_library(vitastor_blk STATIC
+	../util/allocator.cpp ../util/crc32c.c ../util/ringloop.cpp
 	multilist.cpp blockstore_heap.cpp blockstore_disk.cpp
 	blockstore.cpp blockstore_impl.cpp blockstore_init.cpp blockstore_open.cpp
 	blockstore_flush.cpp blockstore_read.cpp blockstore_stable.cpp blockstore_sync.cpp blockstore_write.cpp
@@ -11,8 +12,10 @@ add_library(vitastor_blk STATIC
 )
 target_compile_options(vitastor_blk PUBLIC -fPIC)
 target_link_libraries(vitastor_blk
-	util
-	crc32c
+	${LIBURING_LIBRARIES}
+	${ISAL_LIBRARIES}
+	# for timerfd_manager
+	vitastor_common
 )
 set_target_properties(vitastor_blk PROPERTIES VERSION ${VITASTOR_VERSION} SOVERSION 0)
 
@@ -20,9 +23,9 @@ if (${WITH_FIO})
 	# libfio_vitastor_blk.so
 	add_library(fio_vitastor_blk SHARED
 		fio_engine.cpp
+		../../json11/json11.cpp
 	)
 	target_link_libraries(fio_vitastor_blk
 		vitastor_blk
-		util
 	)
 endif (${WITH_FIO})

src/blockstore/blockstore.h

@@ -77,11 +77,8 @@ Output:
 - retval = number of bytes actually read/written or negative error number
   -EINVAL = invalid input parameters
   -ENOENT = requested object/version does not exist for reads
-  -EEXIST = requested version to write already exists
   -ENOSPC = no space left in the store for writes
-  -EAGAIN = temporary failure, operation should be retried (notably, writes may be retried
-    after stabilizing some journaled entries and freeing up journal space)
-  -EDOM = checksum error
+  -EDOM = checksum error.
 - version = the version actually read or written
 
 ## BS_OP_DELETE
@@ -93,7 +90,7 @@ Input:
 - version = requested version. Treated the same as with BS_OP_WRITE
 
 Output:
-- retval = 0 or negative error number (-EINVAL, -EAGAIN)
+- retval = 0 or negative error number (-EINVAL)
 - version = the version actually written (delete is initially written as an object version)
 
 ## BS_OP_SYNC
@@ -102,7 +99,7 @@ Make sure all previously issued modifications reach physical media.
 
 Input: Nothing except opcode
 Output:
-- retval = 0 or negative error number (-EINVAL, -EAGAIN)
+- retval = 0 or negative error number (-EINVAL)
 
 ## BS_OP_STABLE / BS_OP_ROLLBACK
 
@@ -132,9 +129,7 @@ Input:
 - max_oid = max inode/stripe or 0 to list all objects
 
 Output:
-- retval = total obj_ver_id count or negative error number
-  -ENOMEM = could not allocate memory for the listing
-  -EAGAIN = pool sharding settings are incorrect, please reshard and retry
+- retval = total obj_ver_id count
 - version = stable obj_ver_id count
 - buf = obj_ver_id array allocated by the blockstore. Stable versions come first.
   You must free it yourself after usage with free().
@@ -171,7 +166,7 @@ struct __attribute__ ((visibility("default"))) blockstore_op_t
     };
     uint8_t *buf = NULL;
     uint8_t *bitmap = NULL;
-    int64_t retval = 0;
+    int retval = 0;
 
     uint8_t private_data[BS_OP_PRIVATE_DATA_SIZE];
 };

src/blockstore/blockstore_disk.cpp

@@ -12,9 +12,23 @@
 #include "blockstore_disk.h"
 #include "blockstore_heap.h"
 #include "str_util.h"
-#include "disk_util.h"
 #include "allocator.h"
 
+static uint32_t is_power_of_two(uint64_t value)
+{
+    uint32_t l = 0;
+    while (value > 1)
+    {
+        if (value & 1)
+        {
+            return 64;
+        }
+        value = value >> 1;
+        l++;
+    }
+    return l;
+}
+
 void blockstore_disk_t::parse_config(std::map<std::string, std::string> & config)
 {
     // Parse
@@ -112,7 +126,7 @@ void blockstore_disk_t::parse_config(std::map<std::string, std::string> & config
     {
         data_block_size = (1 << DEFAULT_DATA_BLOCK_ORDER);
     }
-    if ((data_block_size & (data_block_size-1)) || data_block_size < MIN_DATA_BLOCK_SIZE || data_block_size >= MAX_DATA_BLOCK_SIZE)
+    if (is_power_of_two(data_block_size) >= 64 || data_block_size < MIN_DATA_BLOCK_SIZE || data_block_size >= MAX_DATA_BLOCK_SIZE)
     {
         throw std::runtime_error("Bad block size");
     }
@@ -210,32 +224,6 @@ void blockstore_disk_t::parse_config(std::map<std::string, std::string> & config
     }
 }
 
-// A metadata entry must fit into one metadata block; a large block_size with a small
-// csum_block_size otherwise inflates the entry past it and makes the callers divide by zero.
-static uint64_t entries_per_meta_block(uint64_t entry_size, uint64_t meta_block_size,
-    uint64_t data_block_size, uint32_t bitmap_granularity, uint32_t csum_block_size)
-{
-    if (entry_size && entry_size <= meta_block_size)
-    {
-        return meta_block_size / entry_size;
-    }
-    std::string msg = "Metadata entry size ("+std::to_string(entry_size)+" bytes) exceeds meta_block_size ("+
-        std::to_string(meta_block_size)+" bytes) with block_size="+std::to_string(data_block_size)+
-        ", bitmap_granularity="+std::to_string(bitmap_granularity);
-    if (csum_block_size)
-    {
-        msg += ", csum_block_size="+std::to_string(csum_block_size)+" ("+
-            std::to_string(data_block_size/csum_block_size)+" checksums per block)";
-    }
-    msg += ". Decrease block_size";
-    if (csum_block_size)
-    {
-        msg += ", increase csum_block_size";
-    }
-    msg += " or increase meta_block_size";
-    throw std::runtime_error(msg);
-}
-
 void blockstore_disk_t::calc_lengths(bool skip_meta_check)
 {
     // data
@@ -285,20 +273,17 @@ void blockstore_disk_t::calc_lengths(bool skip_meta_check)
     clean_entry_bitmap_size = (data_block_size / bitmap_granularity + 7) / 8;
     clean_dyn_size = clean_entry_bitmap_size*2 + (csum_block_size
         ? data_block_size/csum_block_size*(data_csum_type & 0xFF) : 0);
-    uint64_t entries_per_block = 0;
 recalc:
     if (meta_format == BLOCKSTORE_META_FORMAT_HEAP)
     {
-        entries_per_block = entries_per_meta_block(sizeof(heap_big_write_t) + clean_dyn_size,
-            meta_block_size, data_block_size, bitmap_granularity, csum_block_size);
+        uint32_t entries_per_block = meta_block_size / (sizeof(heap_big_write_t) + clean_dyn_size);
         min_meta_len = (block_count+entries_per_block-1) / entries_per_block * meta_block_size;
     }
     else if (meta_format == BLOCKSTORE_META_FORMAT_V1)
     {
         clean_entry_size = 24 /*sizeof(clean_disk_entry)*/ + 2*clean_entry_bitmap_size;
-        entries_per_block = entries_per_meta_block(clean_entry_size, meta_block_size,
-            data_block_size, bitmap_granularity, 0);
-        min_meta_len = (1 + (block_count - 1 + entries_per_block) / entries_per_block) * meta_block_size;
+        min_meta_len = (1 + (block_count - 1 + meta_block_size / clean_entry_size)
+            / (meta_block_size / clean_entry_size)) * meta_block_size;
         if (!skip_meta_check && meta_area_size < min_meta_len)
         {
 too_small:
@@ -310,9 +295,7 @@ too_small:
     {
         meta_format = BLOCKSTORE_META_FORMAT_V2;
         clean_entry_size = 24 /*sizeof(clean_disk_entry)*/ + clean_dyn_size + 4 /*entry_csum*/;
-        entries_per_block = entries_per_meta_block(clean_entry_size, meta_block_size,
-            data_block_size, bitmap_granularity, csum_block_size);
-        min_meta_len = (1 + (block_count - 1 + entries_per_block) / entries_per_block) * meta_block_size;
+        min_meta_len = (1 + (block_count - 1 + meta_block_size / clean_entry_size) / (meta_block_size / clean_entry_size)) * meta_block_size;
         if (!skip_meta_check && meta_area_size < min_meta_len)
         {
             if (!data_csum_type)
@@ -355,6 +338,41 @@ void blockstore_disk_t::check_lengths()
     }
 }
 
+// FIXME: Move to utils
+static void check_size(int fd, uint64_t *size, uint64_t *sectsize, std::string name)
+{
+    int sect;
+    struct stat st;
+    if (fstat(fd, &st) < 0)
+    {
+        throw std::runtime_error("Failed to stat "+name);
+    }
+    if (S_ISREG(st.st_mode))
+    {
+        *size = st.st_size;
+        if (sectsize)
+        {
+            *sectsize = st.st_blksize;
+        }
+    }
+    else if (S_ISBLK(st.st_mode))
+    {
+        if (ioctl(fd, BLKGETSIZE64, size) < 0 ||
+            ioctl(fd, BLKSSZGET, &sect) < 0)
+        {
+            throw std::runtime_error("Failed to get "+name+" size or block size: "+strerror(errno));
+        }
+        if (sectsize)
+        {
+            *sectsize = sect;
+        }
+    }
+    else
+    {
+        throw std::runtime_error(name+" is neither a file nor a block device");
+    }
+}
+
 static int bs_openmode(const std::string & mode)
 {
     if (mode == "directsync")
@@ -473,7 +491,7 @@ void blockstore_disk_t::open_journal()
         journal_fd = meta_fd;
         journal_device_sect = meta_device_sect;
         journal_device_size = 0;
-        if (journal_offset >= (meta_device_size > 0 ? meta_device_size : data_device_size))
+        if (journal_offset >= data_device_size)
         {
             throw std::runtime_error("journal_offset exceeds device size");
         }

src/blockstore/blockstore_flush.cpp

@@ -80,16 +80,6 @@ int journal_flusher_t::get_syncing_buffer()
     return syncing_buffer;
 }
 
-uint64_t journal_flusher_t::get_flushing_lsn(object_id oid)
-{
-    auto it = flushing.find(oid);
-    if (it == flushing.end())
-    {
-        return 0;
-    }
-    return it->second;
-}
-
 bool journal_flusher_t::is_active()
 {
     return active_flushers > 0;
@@ -106,20 +96,6 @@ void journal_flusher_t::release_trim()
     force_start--;
 }
 
-// Ask for an fsync. Used when a new entry depends on another not-yet-durable entry
-void journal_flusher_t::request_fsync()
-{
-    force_start++;
-    force_fsync++;
-    bs->ringloop->wakeup();
-}
-
-void journal_flusher_t::release_fsync()
-{
-    force_fsync--;
-    force_start--;
-}
-
 void journal_flusher_t::dump_diagnostics()
 {
     printf(
@@ -196,42 +172,30 @@ bool journal_flusher_co::loop()
     else if (wait_state == 17) goto resume_17;
     else if (wait_state == 18) goto resume_18;
     else if (wait_state == 19) goto resume_19;
+    else if (wait_state == 20) goto resume_20;
+    else if (wait_state == 21) goto resume_21;
     else if (wait_state == 22) goto resume_22;
     else if (wait_state == 23) goto resume_23;
     else if (wait_state == 24) goto resume_24;
     else if (wait_state == 25) goto resume_25;
-    else if (wait_state == 26) goto resume_26;
-    else if (wait_state == 27) goto resume_27;
-    else if (wait_state == 28) goto resume_28;
-    else if (wait_state == 29) goto resume_29;
-    else if (wait_state == 30) goto resume_30;
-    else if (wait_state == 31) goto resume_31;
 resume_0:
     wait_state = 0;
     wait_count = 0;
     cur_oid = {};
     res = bs->heap->get_next_compact(cur_oid);
-    if ((bs->intent_write_counter >= bs->journal_trim_interval || flusher->force_fsync > 0) && co_id == 0)
+    if ((bs->intent_write_counter >= bs->journal_trim_interval) && co_id == 0)
     {
+        // Advance fsynced_lsn every <journal_trim_interval> intent writes
+        bs->intent_write_counter = 0;
 resume_17:
 resume_18:
 resume_19:
         if (!fsync_buffer(17))
-        {
             return false;
-        }
-        if (bs->intent_write_counter >= bs->journal_trim_interval)
-        {
-            // Advance fsynced_lsn every <journal_trim_interval> intent writes
-            bs->intent_write_counter = 0;
-resume_29:
-resume_30:
-resume_31:
-            if (!trim_lsn(29))
-            {
-                return false;
-            }
-        }
+resume_20:
+resume_21:
+        if (!trim_lsn(20))
+            return false;
     }
     if (res == ENOENT && flusher->force_start > 0 && co_id == 0 &&
         (!bs->dsk.disable_journal_fsync || !bs->dsk.disable_meta_fsync || !bs->dsk.disable_data_fsync))
@@ -267,6 +231,7 @@ resume_16:
         }
         assert(false);
     }
+    flusher->flushing.insert(cur_oid);
 resume_1:
     wait_state = 1;
     should_repeat = false;
@@ -274,6 +239,7 @@ resume_1:
     if (!cur_obj)
     {
         // Object does not exist
+        flusher->flushing.erase(cur_oid);
         goto resume_0;
     }
     // Scan versions to flush
@@ -306,30 +272,11 @@ resume_1:
     if (!compact_info.compact_lsn)
     {
         // Flushing is aborted
+        flusher->flushing.erase(cur_oid);
         bs->heap->unlock_entry(cur_oid);
         goto resume_0;
     }
-    flusher->flushing[cur_oid] = compact_info.compact_lsn;
     flusher->active_flushers++;
-    if (!compact_info.do_delete && compact_info.clean_wr &&
-        compact_info.clean_wr->type() == BS_HEAP_BIG_INTENT &&
-        (!bs->dsk.csum_block_size || bs->dsk.csum_block_size > bs->dsk.bitmap_granularity))
-    {
-        // Compaction merges newer writes into the data block in place, which changes bytes that
-        // <clean_wr>'s checksums still describe. Reads never take those bytes from it, but the
-        // startup recheck does, and it can only skip the overwritten parts when a checksum block
-        // covers exactly one write granule. With a larger block an overwritten part poisons the
-        // checksum of the whole block, and with checksums disabled a big_intent still carries a
-        // single checksum of its whole extent - it's the only way to tell an unfinished in-place
-        // write from a finished one. Either way see blockstore_heap_t::recheck_verify().
-        // So make sure the entry is out of the recheck's scope by persisting completed_lsn first.
-        // It is below fsynced_lsn already - iterate_compaction() doesn't return anything above it
-resume_26:
-resume_27:
-resume_28:
-        if (!trim_lsn(26))
-            return false;
-    }
     for (i = 0; i < read_vec.size(); i++)
     {
         if ((read_vec[i].copy_flags & COPY_BUF_JOURNAL) &&
@@ -348,13 +295,7 @@ resume_28:
         {
             printf("Compacting %jx:%jx up to l%ju (delete)\n", cur_oid.inode, cur_oid.stripe, compact_info.compact_lsn);
         }
-        // Delete discards everything, so reset to the empty state
         clean_loc = UINT64_MAX;
-        read_vec.clear();
-        csum_copy.clear();
-        copy_count = 0;
-        memset(new_bmp, 0, bs->dsk.clean_entry_bitmap_size);
-        memset(new_ext_bmp, 0, bs->dsk.clean_entry_bitmap_size);
     }
     else
     {
@@ -541,12 +482,12 @@ resume_13:
         printf("Compacted %jx:%jx l%ju (%d writes)\n", cur_oid.inode, cur_oid.stripe, compact_info.compact_lsn, copy_count);
     }
     flusher->active_flushers--;
-    flusher->flushing.erase(cur_oid);
     if (should_repeat)
     {
         // Flush the same object again
         goto resume_1;
     }
+    flusher->flushing.erase(cur_oid);
     // All done
     goto resume_0;
 }
@@ -592,7 +533,7 @@ void journal_flusher_co::fill_partial_checksum_blocks()
         }
         auto & vec = read_vec[read_vec.size()-1];
         read_vec.insert(read_vec.begin()+vec_pos, (copy_buffer_t){
-            .copy_flags = COPY_BUF_DATA|COPY_BUF_COALESCED,
+            .copy_flags = COPY_BUF_JOURNAL|COPY_BUF_COALESCED,
             .offset = hole_start,
             .len = hole_end - hole_start,
             .disk_offset = hole_start,
@@ -767,11 +708,9 @@ bool journal_flusher_co::read_buffered(int wait_base)
     else if (wait_state == wait_base+1)
         goto resume_1;
     wait_count = 0;
-    if (bs->dsk.inmemory_journal && bs->dsk.csum_block_size <= bs->dsk.bitmap_granularity)
+    if (bs->dsk.inmemory_journal && !read_to_fill_incomplete)
     {
         // Happy path: nothing to read :)
-        // We may have to read intent writes to recalculate block checksums, so proceed
-        // even if read_to_fill_incomplete is false, but csum_block_size is > bitmap_granularity
         return true;
     }
     for (i = 0; i < read_vec.size(); i++)
@@ -940,40 +879,27 @@ bool journal_flusher_co::trim_lsn(int wait_base)
 {
     if (wait_state == wait_base)        goto resume_0;
     else if (wait_state == wait_base+1) goto resume_1;
-    else if (wait_state == wait_base+2) goto resume_2;
-resume_0:
-    if (flusher->trimming_lsn)
-    {
-        // Allow only one coroutine to update the superblock at a time to prevent parallel
-        // block updates or the device may in theory apply them in any order.
-        wait_state = wait_base;
-        return false;
-    }
     fsynced_lsn = bs->heap->get_fsynced_lsn();
     if (((blockstore_meta_header_v3_t*)bs->meta_superblock)->completed_lsn == fsynced_lsn)
     {
         return true;
     }
-    flusher->trimming_lsn = true;
     flusher->active_flushers++;
     ((blockstore_meta_header_v3_t*)bs->meta_superblock)->completed_lsn = fsynced_lsn;
     ((blockstore_meta_header_v3_t*)bs->meta_superblock)->set_crc32c();
-    await_sqe(1);
+    await_sqe(0);
     data->iov = (struct iovec){ bs->meta_superblock, (size_t)bs->dsk.meta_block_size };
     data->callback = simple_callback_w;
     io_uring_prep_writev(sqe, bs->dsk.meta_fd, &data->iov, 1, bs->dsk.meta_offset);
     // Update superblock with datasync
     sqe->rw_flags = RWF_DSYNC;
     wait_count++;
-resume_2:
+resume_1:
     if (wait_count > 0)
     {
-        wait_state = wait_base+2;
+        wait_state = wait_base+1;
         return false;
     }
-    // Superblock is updated, resume everyone who was waiting for it
-    flusher->trimming_lsn = false;
-    bs->ringloop->wakeup();
     flusher->active_flushers--;
     return true;
 }

src/blockstore/blockstore_flush.h

@@ -93,30 +93,25 @@ public:
 class journal_flusher_t
 {
     int force_start = 0;
-    int force_fsync = 0;
     int min_flusher_count = 0, max_flusher_count = 0, cur_flusher_count = 0, target_flusher_count = 0;
     journal_flusher_co *co;
     blockstore_impl_t *bs;
     friend class journal_flusher_co;
 
-    robin_hood::unordered_flat_map<object_id, uint64_t> flushing;
+    robin_hood::unordered_flat_set<object_id> flushing;
     int active_flushers = 0;
     std::list<flusher_data_sync_t> data_syncs;
     int wanting_meta_fsync = 0;
     bool fsyncing_meta = false;
     int syncing_buffer = 0;
-    bool trimming_lsn = false;
 
 public:
     journal_flusher_t(blockstore_impl_t *bs);
     ~journal_flusher_t();
     void loop();
     int get_syncing_buffer();
-    uint64_t get_flushing_lsn(object_id oid);
     bool is_active();
     void request_trim();
     void release_trim();
-    void request_fsync();
-    void release_fsync();
     void dump_diagnostics();
 };

src/blockstore/blockstore_heap.cpp

@@ -2,7 +2,6 @@
 // Copyright (c) Vitaliy Filippov, 2025+
 // License: VNPL-1.1 (see README.md for details)
 
-#undef NDEBUG
 #include <assert.h>
 #include <string.h>
 #include <stddef.h>
@@ -197,10 +196,7 @@ uint32_t *heap_entry_t::get_checksum(blockstore_heap_t *heap)
 {
     if (type() == BS_HEAP_SMALL_WRITE || type() == BS_HEAP_INTENT_WRITE)
     {
-        // a zero-length small_write still has a checksum slot reserved by
-        // get_small_entry_size when block checksums are off, and its entry
-        // checksum covers it - so it has to be filled with a defined value
-        if (heap->dsk->csum_block_size)
+        if (heap->dsk->csum_block_size || small().len == 0)
             return NULL;
         return (uint32_t*)((uint8_t*)this + sizeof(heap_small_write_t) + heap->dsk->clean_entry_bitmap_size);
     }
@@ -303,14 +299,6 @@ blockstore_heap_t::~blockstore_heap_t()
 void blockstore_heap_t::start_load(uint64_t completed_lsn)
 {
     this->completed_lsn = completed_lsn;
-    // The persisted completed_lsn may be larger than the lsn of the last entry which survived:
-    // entries above it were still in flight when the power went away. New entries must never
-    // reuse an lsn which is at or below completed_lsn, or on the next load they'd look already
-    // fsynced and escape the rollback of unfinished writes
-    if (completed_lsn > next_lsn)
-    {
-        next_lsn = completed_lsn;
-    }
 }
 
 int blockstore_heap_t::read_blocks(uint64_t disk_offset, uint64_t disk_size, uint8_t *buf, bool allow_corrupted,
@@ -433,7 +421,7 @@ corrupted_object:
             if ((wr->type() == BS_HEAP_BIG_INTENT || wr->type() == BS_HEAP_BIG_WRITE) &&
                 wr->big().block_num >= dsk->block_count)
             {
-                fprintf(stderr, "Error: big_write or big_intent entry %jx:%jx v%ju block_num is too large: %u > %ju. Metadata is incompatible with current parameters. ",
+                fprintf(stderr, "Error: big_write or big_intent entry %jx:%jx v%ju block_num is too large: %u > %lu. Metadata is incompatible with current parameters. ",
                     wr->inode, wr->stripe, wr->version, wr->big_intent().block_num, dsk->block_count);
                 goto corrupted_object;
             }
@@ -592,52 +580,6 @@ void blockstore_heap_t::finish_load()
     }
 }
 
-// Count the newest entries of the object which have to be rechecked after a restart, i.e.
-// the ones whose data may be absent from the disk even though the entry itself is there.
-// Those are buffered small writes - their data is written in parallel with the entry - and
-// the newest intent write if it wasn't fsynced yet.
-// A big-redirect intent may also be hidden under a series of small writes: it is acked as
-// soon as its entry is written, without an explicit data fsync, so a power outage may drop
-// its data from the disk cache while keeping the entry. Its data block belongs to it alone,
-// so rechecking it is always safe. In-place intent writes are only rechecked when they're
-// the newest entry - the block under them may have been rewritten by a later compaction,
-// which would make the recheck fail on perfectly good data.
-// <need_data> is set if any of them requires reading the data device, i.e. if the recheck
-// can't be done synchronously in memory.
-// Commit and rollback entries carry no data of their own, but they're kept in the object's
-// chain, so a stabilize or a rollback right after a write leaves one sitting on top of - or
-// in between - the writes whose data still has to be rechecked. Skip over them.
-heap_entry_t *blockstore_heap_t::skip_commits(heap_entry_t *wr)
-{
-    while (wr && (wr->type() == BS_HEAP_COMMIT || wr->type() == BS_HEAP_ROLLBACK))
-    {
-        wr = prev(wr);
-    }
-    return wr;
-}
-
-int blockstore_heap_t::count_recheck_entries(heap_entry_t *obj, bool & need_data)
-{
-    int count = 0;
-    need_data = false;
-    auto wr = skip_commits(obj);
-    auto newest = wr;
-    for (; wr && wr->type() == BS_HEAP_SMALL_WRITE; wr = skip_commits(prev(wr)))
-    {
-        // Zero-sized writes have nothing to read and verify
-        if (wr->small().len > 0)
-            count++;
-    }
-    if (wr && wr->lsn > completed_lsn &&
-        (wr->type() == BS_HEAP_BIG_INTENT || wr == newest && wr->type() == BS_HEAP_INTENT_WRITE) &&
-        (wr->type() == BS_HEAP_BIG_INTENT ? wr->big_intent().len : wr->small().len) > 0)
-    {
-        count++;
-        need_data = true;
-    }
-    return count;
-}
-
 void blockstore_heap_t::fill_recheck_queue()
 {
     for (auto & pgp: block_index)
@@ -647,8 +589,9 @@ void blockstore_heap_t::fill_recheck_queue()
             inode_map_iterate(ip.second, [&](heap_list_item_t *li)
             {
                 auto obj = &li->entry;
-                bool need_data = false;
-                if (count_recheck_entries(obj, need_data) > 0)
+                // Recheck only the latest intent_write (if after completed_lsn) or a series of small_writes
+                if ((obj->type() == BS_HEAP_INTENT_WRITE || obj->type() == BS_HEAP_BIG_INTENT)
+                    && obj->lsn > completed_lsn || obj->type() == BS_HEAP_SMALL_WRITE)
                 {
                     recheck_queue.push_back(obj);
                 }
@@ -690,9 +633,8 @@ int blockstore_heap_t::mark_used_blocks()
                     li = NULL;
                 }
                 bool overwritten = false;
-                for (; li; li = li->prev)
+                for (; li; li = li->prev, wr = &li->entry)
                 {
-                    auto wr = &li->entry;
                     if (overwritten)
                     {
                         wr->set_garbage();
@@ -721,9 +663,10 @@ int blockstore_heap_t::mark_used_blocks()
                         {
                             if (dsk->skip_double_claim)
                             {
-                                // There was a bug in 3.0.x before 3.0.16 - deletion entries were sometimes
-                                // removed from the disk earlier than previous big_writes. Provide a workaround
-                                // for stores affected by this bug: ignore such objects on start.
+                                // There is a BUG currently:
+                                // Sometimes (under unknown conditions) deletion entries are removed from the disk
+                                // earlier than previous big_writes.
+                                // Until it's fixed, we provide a way to ignore such objects on start.
                                 auto prev_li = used_by[wr->big().block_num];
                                 assert(prev_li);
                                 // Newer LSN must be trusted. Remove the older object.
@@ -912,7 +855,7 @@ void blockstore_heap_t::recheck_full_gc()
                 if (inf.entries[i]->entry.is_garbage())
                 {
                     collected_garbage += inf.entries[i]->entry.size;
-                    unlink_list_item(inf.entries[i]);
+                    remove_list_item(inf.entries[i]);
                 }
                 else
                 {
@@ -933,27 +876,13 @@ void blockstore_heap_t::recheck_full_gc()
     }
 }
 
-// The entries which were hiding under the ones just dropped are the newest ones of the object
-// now, and their data may be unfinished just as well - a redirect intent acked without a data
-// fsync, or a buffered small write, both of which the outage may have left without their data.
-// So look at what became visible and recheck it too
-void blockstore_heap_t::recheck_requeue(object_id oid)
-{
-    auto obj = read_entry(oid);
-    if (obj)
-    {
-        recheck_queue.push_back(obj);
-    }
-}
-
 void blockstore_heap_t::recheck_drop_entries(heap_entry_t *obj, heap_entry_t *bad_wr)
 {
     // write entry is invalid, erase it and all newer entries
     int bad_count = 1;
     for (auto wr = obj; wr && wr != bad_wr; wr = prev(wr))
     {
-        if (wr->type() != BS_HEAP_COMMIT && wr->type() != BS_HEAP_ROLLBACK)
-            bad_count++;
+        bad_count++;
     }
     auto prev_wr = prev(bad_wr);
     if (prev_wr)
@@ -969,94 +898,21 @@ void blockstore_heap_t::recheck_drop_entries(heap_entry_t *obj, heap_entry_t *ba
     while (li && prev_wr != &li->entry)
     {
         auto prev = li->prev;
-        auto type = li->entry.type();
-        if ((type == BS_HEAP_COMMIT || type == BS_HEAP_ROLLBACK) && li->entry.version < bad_wr->version)
-        {
-            // A commit or a rollback of an older version refers to versions which are below
-            // the unfinished write and survive it, so it has to survive too - otherwise a
-            // version the client was told is stable would silently become unstable again
-            li = prev;
-            continue;
-        }
-        // Small writes may be stacked on top of an intent write, so the types may differ
-        assert(type == BS_HEAP_SMALL_WRITE || type == BS_HEAP_INTENT_WRITE ||
-            type == BS_HEAP_BIG_INTENT || type == BS_HEAP_COMMIT || type == BS_HEAP_ROLLBACK);
+        assert(li->entry.type() == bad_wr->type());
         init_erase_bad_entry(li);
         unlink_list_item(li);
         li = prev;
     }
 }
 
-// Verify the data of a rechecked entry, skipping the checksum blocks which newer entries of the
-// same object overwrite. Compaction merges those newer writes into the data block in place: it
-// changes bytes which <wr>'s checksums still describe, without touching <wr> itself. Nothing ever
-// reads those bytes from <wr> - they're shadowed - so a mismatch there means nothing.
-// A shadow only holds while the entry casting it stays, though: the very same recheck may find
-// that newer entry unfinished and drop it, and then the bytes under it become visible again.
-// So entries which this recheck is about to drop cast no shadow. This is why the entry needing
-// a data read is always verified last, see recheck_start_reads() - by then it is already known
-// which of the entries above it survive.
-// This is only exact while a checksum block isn't larger than the write granularity. With larger
-// blocks an overwritten part poisons the checksum of the whole block, and skipping the block
-// would hide a genuinely unfinished write in the rest of it - so there the flusher instead makes
-// sure such an entry is below the persisted completed_lsn and never gets rechecked at all,
-// see the big_intent trim in journal_flusher_co::loop()
-bool blockstore_heap_t::recheck_verify(heap_recheck_state_t *st, heap_entry_t *wr, uint8_t *buf)
+void blockstore_heap_t::recheck_start_reads(heap_recheck_state_t *st)
 {
-    auto obj = st->obj;
-    if (!dsk->csum_block_size || dsk->csum_block_size > dsk->bitmap_granularity ||
-        wr == obj || wr->type() == BS_HEAP_SMALL_WRITE)
-    {
-        // The newest entry isn't shadowed by anything, and a small write has its own copy
-        // of the data in the buffer area which compaction only ever reads
-        return calc_checksums(wr, buf, false);
-    }
-    uint32_t start = 0, end = 0;
-    uint8_t *bitmap = NULL;
-    if (wr->type() == BS_HEAP_BIG_INTENT)
-    {
-        start = wr->big_intent().offset;
-        end = start + wr->big_intent().len;
-        bitmap = wr->get_int_bitmap(this);
-    }
-    else
-    {
-        assert(wr->type() == BS_HEAP_INTENT_WRITE);
-        start = wr->small().offset;
-        end = start + wr->small().len;
-    }
-    uint32_t csum_size = (dsk->data_csum_type & 0xFF);
-    uint8_t *csums = wr->get_checksums(this);
-    if (wr->type() == BS_HEAP_BIG_INTENT)
-        csums += start/dsk->csum_block_size * csum_size;
-    bool ok = true;
-    for (uint32_t pos = start; pos < end; pos += dsk->csum_block_size)
-    {
-        bool shadowed = false;
-        for (auto newer = obj; newer && newer != wr && !shadowed; newer = prev(newer))
-        {
-            if (newer->type() != BS_HEAP_SMALL_WRITE && newer->type() != BS_HEAP_INTENT_WRITE ||
-                st->bad_wr && newer->lsn >= st->bad_wr->lsn)
-                continue;
-            shadowed = newer->small().offset < pos+dsk->csum_block_size &&
-                newer->small().offset+newer->small().len > pos;
-        }
-        if (shadowed)
-            continue;
-        ok = calc_block_checksums((uint32_t*)(csums + (pos-start)/dsk->csum_block_size * csum_size),
-            buf + (pos-start), bitmap, pos, pos+dsk->csum_block_size, false, NULL) && ok;
-    }
-    return ok;
-}
-
-int blockstore_heap_t::recheck_start_reads(heap_recheck_state_t *st)
-{
-    int started = 0;
     if (st->sent_reads >= st->total_reads)
-        return 0;
+        return;
     while (recheck_in_progress < recheck_queue_depth)
     {
-        auto wr = skip_commits(st->next_wr);
+        auto wr = st->next_wr;
+        st->next_wr = prev(st->next_wr);
         uint64_t loc = 0, len = 0;
         bool from_data = false;
         if (wr->type() == BS_HEAP_SMALL_WRITE)
@@ -1091,57 +947,32 @@ int blockstore_heap_t::recheck_start_reads(heap_recheck_state_t *st)
             len = wr->small().len;
             from_data = true;
         }
-        if (!len)
-        {
-            // Zero-sized entries have nothing to read and verify
-            st->next_wr = prev(wr);
-            continue;
-        }
-        if (from_data && st->checked_reads < st->sent_reads)
-        {
-            // Part of this entry's data may be shadowed by the newer entries of the same
-            // object, and whether it really is depends on whether they survive the recheck.
-            // So verify it only when all of them are already verified - see recheck_verify()
-            break;
-        }
-        st->next_wr = prev(wr);
-        st->oldest_wr = wr;
+        uint8_t *buf = (uint8_t*)memalign_or_die(MEM_ALIGNMENT, len);
         st->sent_reads++;
         recheck_in_progress++;
         recheck_pending_reads--;
         bool is_last = st->sent_reads >= st->total_reads;
-        recheck_cb(from_data, loc, len, [this, st, wr](uint8_t *buf)
+        recheck_cb(from_data, loc, len, buf, [this, st, wr, buf]()
         {
             st->checked_reads++;
-            if (!recheck_verify(st, wr, buf))
+            if (!calc_checksums(wr, buf, false))
                 st->bad_wr = !st->bad_wr || st->bad_wr->lsn > wr->lsn ? wr : st->bad_wr;
             if (st->checked_reads >= st->total_reads)
             {
-                auto bad_wr = st->bad_wr;
-                auto obj = st->obj;
-                // Entries below the oldest one we read are unverified, the rest of the set
-                // is already known to be good
-                bool check_more = (bad_wr && bad_wr == st->oldest_wr);
-                recheck_states.erase(obj);
-                if (bad_wr)
-                {
-                    object_id oid = { .inode = obj->inode, .stripe = obj->stripe };
-                    recheck_drop_entries(obj, bad_wr);
-                    if (check_more)
-                        recheck_requeue(oid);
-                }
+                if (st->bad_wr)
+                    recheck_drop_entries(st->obj, st->bad_wr);
+                recheck_states.erase(st->obj);
             }
+            free(buf);
             recheck_in_progress--;
             recheck_small_writes(NULL, 0);
         });
-        started++;
         if (is_last)
             break;
     }
-    return started;
 }
 
-bool blockstore_heap_t::recheck_small_writes(std::function<void(bool is_data, uint64_t offset, uint64_t len, std::function<void(uint8_t* buf)>)> read_buffer, int queue_depth)
+bool blockstore_heap_t::recheck_small_writes(std::function<void(bool is_data, uint64_t offset, uint64_t len, uint8_t* buf, std::function<void()>)> read_buffer, int queue_depth)
 {
     if (in_recheck)
     {
@@ -1162,52 +993,26 @@ bool blockstore_heap_t::recheck_small_writes(std::function<void(bool is_data, ui
     in_recheck = true;
     while (recheck_pending_reads > 0 && recheck_in_progress < recheck_queue_depth)
     {
-        int started = 0;
         for (auto & sp: recheck_states)
-            started += recheck_start_reads(&sp.second);
-        if (!started)
-        {
-            // Everything left is postponed until the reads in flight complete
-            break;
-        }
+            recheck_start_reads(&sp.second);
     }
     while (recheck_queue.size() > 0 && recheck_in_progress < recheck_queue_depth)
     {
         heap_entry_t *obj = recheck_queue.front();
         recheck_queue.pop_front();
-        bool need_data = false;
-        int recheck_count = count_recheck_entries(obj, need_data);
-        if (!recheck_count)
-        {
-            // Nothing left to recheck - the object was requeued after a drop which exposed
-            // entries that are already known to be good
-            continue;
-        }
-        if (!need_data && buffer_area)
+        if (obj->type() == BS_HEAP_SMALL_WRITE && buffer_area)
         {
             // Check this object synchronously
-            heap_entry_t *bad_wr = NULL, *oldest_wr = NULL;
-            int checked = 0;
-            auto wr = skip_commits(obj);
-            for (; wr && wr->type() == BS_HEAP_SMALL_WRITE && checked < recheck_count; wr = skip_commits(prev(wr)))
+            heap_entry_t *bad_wr = NULL;
+            for (auto wr = obj; wr && wr->type() == BS_HEAP_SMALL_WRITE; wr = prev(wr))
             {
-                if (!wr->small().len)
-                    continue;
                 fprintf(stderr, "Notice: rechecking %jx:%jx l%ju - %u bytes at %ju in buffer area\n",
                     wr->inode, wr->stripe, wr->lsn, wr->small().len, wr->small().location);
-                oldest_wr = wr;
-                checked++;
                 if (!calc_checksums(wr, buffer_area + wr->small().location, false))
                     bad_wr = wr;
             }
             if (bad_wr)
-            {
-                object_id oid = { .inode = obj->inode, .stripe = obj->stripe };
-                bool check_more = (bad_wr == oldest_wr);
                 recheck_drop_entries(obj, bad_wr);
-                if (check_more)
-                    recheck_requeue(oid);
-            }
         }
         else
         {
@@ -1215,7 +1020,10 @@ bool blockstore_heap_t::recheck_small_writes(std::function<void(bool is_data, ui
             auto & st = recheck_states[obj];
             st.obj = obj;
             st.next_wr = obj;
-            st.total_reads = recheck_count;
+            st.total_reads = 1;
+            if (obj->type() == BS_HEAP_SMALL_WRITE)
+                for (auto wr = prev(obj); wr && wr->type() == BS_HEAP_SMALL_WRITE; wr = prev(wr))
+                    st.total_reads++;
             recheck_pending_reads += st.total_reads;
             recheck_start_reads(&st);
         }
@@ -1228,7 +1036,7 @@ bool blockstore_heap_t::recheck_small_writes(std::function<void(bool is_data, ui
         recheck_queue_depth = 0;
         if (cb)
         {
-            cb(false, 0, 0, NULL);
+            cb(false, 0, 0, NULL, NULL);
         }
         return true;
     }
@@ -1256,8 +1064,6 @@ int blockstore_heap_t::finish_recheck()
     }
     completed_lsn = next_lsn;
     first_inflight_lsn = next_lsn+1;
-    // Everything we just loaded from disk is durable by definition
-    fsynced_lsn = next_lsn;
     std::sort(compact_queue.begin(), compact_queue.end(), [this](const object_id & a, const object_id & b)
     {
         auto ao = read_entry(a);
@@ -1287,12 +1093,6 @@ bool blockstore_heap_t::calc_checksums(heap_entry_t *wr, uint8_t *data, bool set
             len = wr->big_intent().len;
         else
             assert(0);
-        if (!len)
-        {
-            if (set)
-                *wr_csum = 0;
-            return true;
-        }
         uint32_t real_csum = 0;
         if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32)
             real_csum = (uint32_t)XXH3_64bits(data, len);
@@ -1370,9 +1170,9 @@ bool blockstore_heap_t::calc_block_checksums(uint32_t *block_csums, uint8_t *bit
     uint32_t pos = start;
     uint32_t block_end = (start/dsk->csum_block_size + 1)*dsk->csum_block_size;
     uint32_t block_crc = 0;
+    bool isset = false;
     while (pos < end)
     {
-        bool isset = false;
         uint32_t blk_start = pos;
         if (bitmap)
         {
@@ -1671,24 +1471,8 @@ void blockstore_heap_t::gc_block(heap_block_info_t & inf)
     }
 }
 
-// When the previous object's entry is not fsynced yet, we can't write a new one EXCEPT into
-// the same block which is passed here as <prefer_block>. If prefer_block is set but we can't
-// write into it right now, we return EAGAIN.
-int blockstore_heap_t::allocate_entry(uint32_t entry_size, uint32_t *block_num, bool allow_last_free, uint32_t prefer_block)
+int blockstore_heap_t::allocate_entry(uint32_t entry_size, uint32_t *block_num, bool allow_last_free)
 {
-    if (prefer_block != UINT32_MAX)
-    {
-        auto & inf = block_info.at(prefer_block);
-        // A block accumulates a consecutive LSN range only, see modify_alloc() below
-        if (inf.is_writing ||
-            inf.mod_lsn_to && inf.mod_lsn_to != next_lsn ||
-            inf.used_space - inf.garbage_space + entry_size > dsk->meta_block_size)
-        {
-            return EAGAIN;
-        }
-        last_allocated_block = prefer_block;
-        goto allocated;
-    }
     if (last_allocated_block != UINT32_MAX)
     {
         // First try to write into the same block as the previous time
@@ -1751,7 +1535,6 @@ int blockstore_heap_t::allocate_entry(uint32_t entry_size, uint32_t *block_num,
             return ENOSPC;
         }
     }
-allocated:
     // Write into the same block
     *block_num = last_allocated_block;
     modify_alloc(last_allocated_block, [&](heap_block_info_t & inf)
@@ -1818,15 +1601,11 @@ void blockstore_heap_t::insert_list_items(heap_list_item_t** v, size_t count, bo
     }
 }
 
-// <prev_obj> is the newest entry of the object the new entry is added to, or NULL if there is
-// none. It is used to keep all not-yet-durable entries of one object in the same metadata block,
-// see allocate_entry(). May return EAGAIN or EBUSY, meaning "retry later" - see there.
-int blockstore_heap_t::add_entry(uint32_t wr_size, heap_entry_t *prev_obj, uint32_t *modified_block, bool prefer_same,
+int blockstore_heap_t::add_entry(uint32_t wr_size, uint32_t *modified_block,
     bool allow_last_free, bool explicit_complete, std::function<void(heap_entry_t *wr)> fill_entry)
 {
     uint32_t block_num;
-    int res = allocate_entry(wr_size, &block_num, allow_last_free, prefer_same && prev_obj
-        && prev_obj->lsn > get_fsynced_lsn() ? list_item(prev_obj)->block_num : UINT32_MAX);
+    int res = allocate_entry(wr_size, &block_num, allow_last_free);
     if (res != 0)
     {
         return res;
@@ -1871,7 +1650,7 @@ int blockstore_heap_t::add_small_write(object_id oid, heap_entry_t **obj_ptr, ui
     }
     uint32_t wr_size = get_small_entry_size(offset, len);
     // Small writes are written in parallel with buffered data so they require explicit_complete
-    return add_entry(wr_size, obj, modified_block, true, false, true, [&](heap_entry_t *wr)
+    return add_entry(wr_size, modified_block, false, true, [&](heap_entry_t *wr)
     {
         wr->entry_type = type;
         wr->inode = oid.inode;
@@ -1912,7 +1691,7 @@ int blockstore_heap_t::add_big_write(object_id oid, heap_entry_t *old_head, bool
     }
     uint32_t wr_size = get_big_entry_size();
     // Big writes are written after writing data so they don't require explicit_complete
-    return add_entry(wr_size, old_head, modified_block, false, false, false, [&](heap_entry_t *wr)
+    return add_entry(wr_size, modified_block, false, false, [&](heap_entry_t *wr)
     {
         wr->entry_type = BS_HEAP_BIG_WRITE | (stable ? BS_HEAP_STABLE : 0);
         wr->inode = oid.inode;
@@ -1938,7 +1717,7 @@ int blockstore_heap_t::add_redirect_intent(object_id oid, heap_entry_t **obj_ptr
 {
     uint32_t wr_size = get_big_intent_entry_size();
     // Big-redirect intents, just like regular big writes, are written after writing data so they don't require explicit_complete
-    return add_entry(wr_size, *obj_ptr, modified_block, false, false, false, [&](heap_entry_t *wr)
+    return add_entry(wr_size, modified_block, false, false, [&](heap_entry_t *wr)
     {
         wr->entry_type = BS_HEAP_BIG_INTENT|BS_HEAP_STABLE;
         wr->inode = oid.inode;
@@ -1974,7 +1753,7 @@ int blockstore_heap_t::add_big_intent(object_id oid, heap_entry_t **obj_ptr, uin
     }
     uint32_t wr_size = get_big_intent_entry_size();
     // Big intents are written before writing data so they require explicit_complete
-    return add_entry(wr_size, obj, modified_block, false, false, true, [&](heap_entry_t *wr)
+    return add_entry(wr_size, modified_block, false, true, [&](heap_entry_t *wr)
     {
         wr->entry_type = BS_HEAP_BIG_INTENT | BS_HEAP_STABLE;
         wr->inode = oid.inode;
@@ -2018,9 +1797,7 @@ int blockstore_heap_t::add_compact(heap_entry_t *obj, uint64_t compact_version,
     }
     if (do_delete)
     {
-        // Compaction only ever touches durable entries and reuses their LSNs, so it doesn't
-        // need the "same block as the previous entry of the object" placement
-        return add_entry(get_simple_entry_size(), NULL, modified_block, false, false, false, [&](heap_entry_t *wr)
+        return add_entry(get_simple_entry_size(), modified_block, false, false, [&](heap_entry_t *wr)
         {
             wr->entry_type = BS_HEAP_DELETE|BS_HEAP_STABLE;
             wr->inode = obj->inode;
@@ -2031,7 +1808,7 @@ int blockstore_heap_t::add_compact(heap_entry_t *obj, uint64_t compact_version,
     }
     uint32_t wr_size = get_big_entry_size();
     // Compaction entry is added after copying data so it doesn't require explicit_complete
-    return add_entry(wr_size, NULL, modified_block, false, true, false, [&](heap_entry_t *new_wr)
+    return add_entry(wr_size, modified_block, true, false, [&](heap_entry_t *new_wr)
     {
         new_wr->entry_type = BS_HEAP_BIG_WRITE|BS_HEAP_STABLE;
         new_wr->inode = obj->inode;
@@ -2064,11 +1841,11 @@ int blockstore_heap_t::punch_holes(heap_entry_t *wr, uint8_t *new_bitmap, uint8_
     return 0;
 }
 
-int blockstore_heap_t::add_simple(heap_entry_t *obj, uint64_t version, uint32_t *modified_block, uint32_t entry_type, bool prefer_same)
+int blockstore_heap_t::add_simple(heap_entry_t *obj, uint64_t version, uint32_t *modified_block, uint32_t entry_type)
 {
     uint32_t wr_size = get_simple_entry_size();
     // Simple entries don't have data so they don't require explicit_complete
-    return add_entry(wr_size, obj, modified_block, prefer_same, false, false, [&](heap_entry_t *wr)
+    return add_entry(wr_size, modified_block, false, false, [&](heap_entry_t *wr)
     {
         wr->entry_type = entry_type;
         wr->inode = obj->inode;
@@ -2088,7 +1865,7 @@ int blockstore_heap_t::add_commit(heap_entry_t *obj, uint64_t version, uint32_t
         {
             auto rollback_version = wr->version;
             wr = prev(wr);
-            while (wr && wr->version > rollback_version)
+            while (wr->version > rollback_version)
             {
                 assert(!(wr->entry_type & BS_HEAP_STABLE));
                 wr = prev(wr);
@@ -2098,19 +1875,13 @@ int blockstore_heap_t::add_commit(heap_entry_t *obj, uint64_t version, uint32_t
         if (wr->type() == BS_HEAP_COMMIT)
         {
             if (commit_version < wr->version)
-            {
                 commit_version = wr->version;
-            }
             wr = prev(wr);
             continue;
         }
-        if (wr->version >= version)
+        if (wr->version == version)
         {
-            // A version may be already committed, flushed, and removed, and it's not a bug
             found = true;
-        }
-        if (wr->version <= version)
-        {
             if (!(wr->entry_type & BS_HEAP_STABLE) && wr->version > commit_version)
             {
                 uncommitted = true;
@@ -2131,7 +1902,7 @@ int blockstore_heap_t::add_commit(heap_entry_t *obj, uint64_t version, uint32_t
     {
         return 0;
     }
-    return add_simple(obj, version, modified_block, BS_HEAP_COMMIT, true);
+    return add_simple(obj, version, modified_block, BS_HEAP_COMMIT);
 }
 
 int blockstore_heap_t::add_rollback(heap_entry_t *obj, uint64_t version, uint32_t *modified_block)
@@ -2190,13 +1961,13 @@ int blockstore_heap_t::add_rollback(heap_entry_t *obj, uint64_t version, uint32_
     {
         return 0;
     }
-    return add_simple(obj, version, modified_block, BS_HEAP_ROLLBACK, true);
+    return add_simple(obj, version, modified_block, BS_HEAP_ROLLBACK);
 }
 
 int blockstore_heap_t::add_delete(heap_entry_t *obj, uint32_t *modified_block)
 {
     assert(obj);
-    return add_simple(obj, 0, modified_block, BS_HEAP_DELETE|BS_HEAP_STABLE, false);
+    return add_simple(obj, 0, modified_block, BS_HEAP_DELETE|BS_HEAP_STABLE);
 }
 
 // 2nd step: mark the block as being written (to prevent further in-memory updates to it),
@@ -2307,19 +2078,12 @@ void blockstore_heap_t::mark_garbage_up_to(heap_entry_t *wr)
         return;
     }
     assert(wr->is_overwrite());
-    // <wr> survives, so its data block must not be freed no matter how far down the chain
-    // another entry referring to it turns out to be: compaction merges newer writes into the
-    // base block in place and gives its location to the entry it adds, and a rolled back big
-    // write of its own may well sit in between the two.
-    // <used_big> additionally keeps the walk from freeing the same block twice when several
-    // entries in a row share it
-    uint32_t keep_big = (wr->type() == BS_HEAP_BIG_WRITE || wr->type() == BS_HEAP_BIG_INTENT ? wr->big().block_num : UINT32_MAX);
-    uint32_t used_big = keep_big;
+    uint32_t used_big = (wr->type() == BS_HEAP_BIG_WRITE || wr->type() == BS_HEAP_BIG_INTENT ? wr->big().block_num : UINT32_MAX);
     wr = prev(wr);
     while (wr && !wr->is_garbage())
     {
         auto prev_wr = prev(wr);
-        mark_garbage(list_item(wr)->block_num, wr, used_big, keep_big);
+        mark_garbage(list_item(wr)->block_num, wr, used_big);
         if (wr->type() == BS_HEAP_BIG_WRITE || wr->type() == BS_HEAP_BIG_INTENT)
         {
             used_big = wr->big().block_num;
@@ -2328,7 +2092,7 @@ void blockstore_heap_t::mark_garbage_up_to(heap_entry_t *wr)
     }
 }
 
-void blockstore_heap_t::mark_garbage(uint32_t block_num, heap_entry_t *prev_wr, uint32_t used_big, uint32_t keep_big)
+void blockstore_heap_t::mark_garbage(uint32_t block_num, heap_entry_t *prev_wr, uint32_t used_big)
 {
     prev_wr->set_garbage();
     garbage_entries++;
@@ -2338,8 +2102,7 @@ void blockstore_heap_t::mark_garbage(uint32_t block_num, heap_entry_t *prev_wr,
     {
         free_buffer_area(prev_wr->inode, prev_wr->small().location, prev_wr->small().len);
     }
-    else if ((prev_wr->type() == BS_HEAP_BIG_WRITE || prev_wr->type() == BS_HEAP_BIG_INTENT) &&
-        prev_wr->big().block_num != used_big && prev_wr->big().block_num != keep_big)
+    else if ((prev_wr->type() == BS_HEAP_BIG_WRITE || prev_wr->type() == BS_HEAP_BIG_INTENT) && prev_wr->big().block_num != used_big)
     {
         free_data(prev_wr->inode, prev_wr->big_location(this));
     }
@@ -2364,17 +2127,6 @@ int blockstore_heap_t::get_next_compact(object_id & oid)
     return 0;
 }
 
-bool blockstore_heap_t::is_absent(heap_entry_t *obj)
-{
-    bool present = false;
-    iterate_with_stable(obj, UINT64_MAX, [&](heap_entry_t *wr, bool stable)
-    {
-        present = (wr->type() != BS_HEAP_DELETE);
-        return false;
-    });
-    return !present;
-}
-
 void blockstore_heap_t::iterate_with_stable(heap_entry_t *obj, uint64_t max_lsn, std::function<bool(heap_entry_t*, bool)> cb)
 {
     auto old_wr = obj;
@@ -2438,11 +2190,10 @@ heap_compact_t blockstore_heap_t::iterate_compaction(heap_entry_t *obj, uint64_t
     {
         if (wr->type() == BS_HEAP_ROLLBACK && wr->lsn <= fsynced_lsn)
         {
-            // compact_version in commit/rollback is taken from actual write entries
-            // to handle lsn=0 reset correctly
             if (!res.compact_lsn)
             {
                 res.compact_lsn = wr->lsn;
+                res.compact_version = wr->version;
             }
             if (rollback_version > wr->version)
             {
@@ -2452,15 +2203,14 @@ heap_compact_t blockstore_heap_t::iterate_compaction(heap_entry_t *obj, uint64_t
         }
         if (wr->type() == BS_HEAP_COMMIT && wr->lsn <= fsynced_lsn)
         {
-            res.do_delete = false;
             if (!res.compact_lsn)
             {
                 res.compact_lsn = wr->lsn;
+                res.compact_version = wr->version;
             }
+            res.do_delete = false;
             if (commit_version < wr->version)
-            {
                 commit_version = wr->version;
-            }
             continue;
         }
         bool rolled_back = (wr->version > rollback_version);
@@ -2473,15 +2223,12 @@ heap_compact_t blockstore_heap_t::iterate_compaction(heap_entry_t *obj, uint64_t
         if (!stable && !committed || wr->lsn > fsynced_lsn)
         {
             // Unstable and non-fsynced writes can't be compacted yet
-            // Reset newer compact_lsn to place compact entry correctly
             res.do_delete = false;
             res.compact_lsn = 0;
             res.compact_version = 0;
-            if ((!under_pressure || stable) && (wr->type() == BS_HEAP_BIG_WRITE || wr->type() == BS_HEAP_BIG_INTENT || wr->type() == BS_HEAP_DELETE))
+            if (!under_pressure && (wr->type() == BS_HEAP_BIG_WRITE || wr->type() == BS_HEAP_DELETE))
             {
                 // We may postpone compaction if we have an unstable overwrite when not under pressure
-                // We MUST postpone if the overwrite is STABLE (even if under pressure), 
-                // because add_compact will reject STABLE overwrites with EBUSY.
                 return res;
             }
             continue;
@@ -2492,9 +2239,6 @@ heap_compact_t blockstore_heap_t::iterate_compaction(heap_entry_t *obj, uint64_t
             if (!stable && !res.compact_lsn)
             {
                 res.compact_lsn = wr->lsn;
-            }
-            if (res.compact_version < wr->version)
-            {
                 res.compact_version = wr->version;
             }
             res.clean_wr = wr;
@@ -2505,45 +2249,18 @@ heap_compact_t blockstore_heap_t::iterate_compaction(heap_entry_t *obj, uint64_t
         {
             // Object is deleted
             assert(!has_small && stable); // unstable deletes are not supported
-            if (!res.do_delete && res.compact_lsn)
-            {
-                // A COMMIT above the delete reset do_delete, but all writes between
-                // them are lost (incorrect chain / no fsync). Restore do_delete.
-                res.do_delete = true;
-                res.compact_version = 0;
-            }
             return res;
         }
         assert(wr->type() == BS_HEAP_SMALL_WRITE || wr->type() == BS_HEAP_INTENT_WRITE);
         if (!res.compact_lsn)
         {
             res.compact_lsn = wr->lsn;
-        }
-        if (res.compact_version < wr->version)
-        {
             res.compact_version = wr->version;
         }
         res.do_delete = false;
         has_small = true;
         small_wr_cb(wr);
     }
-    if (!res.clean_wr && !res.do_delete && res.compact_lsn)
-    {
-        if (has_small)
-        {
-            // The chain holds no big entry surviving the rollbacks and no executed
-            // delete: neither a merge nor a delete is possible. The chain is invalid
-            // but we can't just remove it, so abort compaction and leave it as is.
-            res.compact_lsn = 0;
-        }
-        else
-        {
-            // Otherwise it means that there are no entries: it's either a bare COMMIT
-            // or a COMMIT over a ROLLBACK which masks all previous entries. Remove everything.
-            res.do_delete = true;
-        }
-        res.compact_version = 0;
-    }
     return res;
 }
 
@@ -2921,21 +2638,26 @@ void blockstore_heap_t::apply_inflight(heap_inflight_lsn_t & inflight)
     {
         // Remove entry
         auto li = list_item(wr);
-        if (!li->next)
-        {
-            // The last freed entry must be a deletion
-            assert(!li->prev);
-            assert((li->entry.entry_type & ~BS_HEAP_GARBAGE) == (BS_HEAP_DELETE|BS_HEAP_STABLE));
-        }
-        else if (!li->prev && li->next->entry.entry_type == (BS_HEAP_DELETE|BS_HEAP_STABLE))
-        {
-            // free BS_HEAP_DELETEs when all previous entries are also freed
-            mark_garbage(li->next->block_num, &li->next->entry, UINT32_MAX, UINT32_MAX);
-        }
-        unlink_list_item(li);
+        remove_list_item(li);
     }
 }
 
+void blockstore_heap_t::remove_list_item(heap_list_item_t *li)
+{
+    if (!li->next)
+    {
+        // The last freed entry must be a deletion
+        assert(!li->prev);
+        assert((li->entry.entry_type & ~BS_HEAP_GARBAGE) == (BS_HEAP_DELETE|BS_HEAP_STABLE));
+    }
+    else if (!li->prev && li->next->entry.entry_type == (BS_HEAP_DELETE|BS_HEAP_STABLE))
+    {
+        // free BS_HEAP_DELETEs when all previous entries are also freed
+        mark_garbage(li->next->block_num, &li->next->entry, UINT32_MAX);
+    }
+    unlink_list_item(li);
+}
+
 void blockstore_heap_t::unlink_list_item(heap_list_item_t *li)
 {
     auto prev = li->prev;

src/blockstore/blockstore_heap.h

@@ -167,9 +167,6 @@ struct heap_recheck_state_t
     size_t sent_reads = 0;
     size_t checked_reads = 0;
     heap_entry_t *bad_wr = NULL;
-    // The oldest entry a read was issued for. Everything below it is unverified, so if it
-    // turns out to be bad the object has to be looked at again after dropping it
-    heap_entry_t *oldest_wr = NULL;
 };
 
 using i64hash_t = robin_hood::hash<uint64_t>;
@@ -230,15 +227,14 @@ class blockstore_heap_t
     size_t recheck_pending_reads = 0;
     int recheck_in_progress = 0;
     bool in_recheck = false;
-    std::function<void(bool is_data, uint64_t offset, uint64_t len, std::function<void(uint8_t*)>)> recheck_cb;
+    std::function<void(bool is_data, uint64_t offset, uint64_t len, uint8_t* buf, std::function<void()>)> recheck_cb;
     int recheck_queue_depth = 0;
 
     uint64_t get_pg_id(inode_t inode, uint64_t stripe);
     bool validate_object(heap_entry_t *obj);
     void fill_recheck_queue();
-    void recheck_requeue(object_id oid);
     void recheck_drop_entries(heap_entry_t *obj, heap_entry_t *bad_wr);
-    int recheck_start_reads(heap_recheck_state_t *st);
+    void recheck_start_reads(heap_recheck_state_t *st);
     int mark_used_blocks();
     void init_free_bad_entry(heap_entry_t *wr);
     void init_erase_bad_entry(heap_list_item_t *li);
@@ -249,21 +245,19 @@ class blockstore_heap_t
     void reshard_add(heap_reshard_state_t *st, heap_list_item_t *li);
 
     void gc_block(heap_block_info_t & inf);
-    int allocate_entry(uint32_t entry_size, uint32_t *block_num, bool allow_last_free, uint32_t prefer_block);
+    int allocate_entry(uint32_t entry_size, uint32_t *block_num, bool allow_last_free);
     void insert_list_items(heap_list_item_t** v, size_t count, bool postpone);
+    void remove_list_item(heap_list_item_t *li);
     void unlink_list_item(heap_list_item_t *li);
-    int add_entry(uint32_t wr_size, heap_entry_t *prev_obj, uint32_t *modified_block, bool prefer_same,
-        bool allow_last_free, bool explicit_complete, std::function<void(heap_entry_t *wr)> fill_entry);
-    int add_simple(heap_entry_t *obj, uint64_t version, uint32_t *modified_block, uint32_t entry_type, bool prefer_same);
+    int add_entry(uint32_t wr_size, uint32_t *modified_block, bool allow_last_free,
+        bool explicit_complete, std::function<void(heap_entry_t *wr)> fill_entry);
+    int add_simple(heap_entry_t *obj, uint64_t version, uint32_t *modified_block, uint32_t entry_type);
     uint32_t meta_alloc_pos(const heap_block_info_t & inf);
     void modify_alloc(uint32_t block_num, std::function<void(heap_block_info_t &)> change_cb);
     void mark_garbage_up_to(heap_entry_t *wr);
-    void mark_garbage(uint32_t block_num, heap_entry_t *prev_wr, uint32_t used_big, uint32_t keep_big);
+    void mark_garbage(uint32_t block_num, heap_entry_t *prev_wr, uint32_t used_big);
     void push_inflight_lsn(uint64_t lsn, heap_entry_t *wr, uint64_t flags);
     void mark_completed_lsns(uint64_t mod_lsn);
-    heap_entry_t *skip_commits(heap_entry_t *wr);
-    int count_recheck_entries(heap_entry_t *obj, bool & need_data);
-    bool recheck_verify(heap_recheck_state_t *st, heap_entry_t *wr, uint8_t *buf);
     void apply_inflight(heap_inflight_lsn_t & inflight);
 public:
     blockstore_heap_t(blockstore_disk_t *dsk, uint8_t *buffer_area, int log_level = 0);
@@ -281,7 +275,7 @@ public:
     // before finishing initialization if not R/O
     std::vector<uint32_t> get_recheck_modified_blocks();
     // recheck small write data after reading the database from disk
-    bool recheck_small_writes(std::function<void(bool is_data, uint64_t offset, uint64_t len, std::function<void(uint8_t* buf)>)> read_buffer, int queue_depth);
+    bool recheck_small_writes(std::function<void(bool is_data, uint64_t offset, uint64_t len, uint8_t* buf, std::function<void()>)> read_buffer, int queue_depth);
     int finish_recheck();
     // reshard database according to the pool's PG count
     void* reshard_start(pool_id_t pool, uint32_t pg_count, uint32_t pg_stripe_size, uint64_t chunk_limit);
@@ -294,9 +288,6 @@ public:
     heap_entry_t *lock_and_read_entry(object_id oid);
     // read an object entry without locking it
     heap_entry_t *read_entry(object_id oid);
-    // true if a read of the chain would return ENOENT: the head is a DELETE entry
-    // or all entries are cut off by a ROLLBACK to version 0
-    bool is_absent(heap_entry_t *obj);
     // unlock an entry
     bool unlock_entry(object_id oid);
     // set or verify checksums in a write request

src/blockstore/blockstore_impl.cpp

@@ -43,24 +43,6 @@ blockstore_impl_t::blockstore_impl_t(blockstore_config_t & config, ring_loop_i *
 
 blockstore_impl_t::~blockstore_impl_t()
 {
-    // Operations which were queued or in flight when we were destroyed still have their
-    // private state constructed, and a read also owns its read vector. Nobody is going to
-    // complete them now, so release it here
-    for (auto op: submit_queue)
-        if (op)
-            PRIV(op)->~blockstore_op_private_t();
-    submit_queue.clear();
-    for (auto op: reads_in_flight)
-    {
-        free_read_buffers(PRIV(op)->read_vec);
-        PRIV(op)->~blockstore_op_private_t();
-    }
-    reads_in_flight.clear();
-    // Metadata block buffers are normally freed by the write completion callback,
-    // so anything still here was queued or in flight when we were destroyed
-    for (auto & mb: modified_blocks)
-        free(mb.second.buf);
-    modified_blocks.clear();
     if (flusher)
         delete flusher;
     if (heap)
@@ -286,19 +268,6 @@ void blockstore_impl_t::check_wait(blockstore_op_t *op)
         flusher->release_trim();
         PRIV(op)->wait_for = 0;
     }
-    else if (PRIV(op)->wait_for == WAIT_META_FSYNC)
-    {
-        if (heap->get_fsynced_lsn() < PRIV(op)->wait_detail)
-        {
-            // do not submit
-#ifdef BLOCKSTORE_DEBUG
-            printf("Still waiting for LSN %ju to become durable\n", PRIV(op)->wait_detail);
-#endif
-            return;
-        }
-        flusher->release_fsync();
-        PRIV(op)->wait_for = 0;
-    }
     else
     {
         throw std::runtime_error("BUG: op->wait_for value is unexpected");
@@ -376,7 +345,7 @@ void blockstore_impl_t::process_list(blockstore_op_t *op)
         std::sort(result, result + stable_count);
         if (stable_count > op->list_stable_limit)
         {
-            memmove(result + op->list_stable_limit, result + stable_count, unstable_count*sizeof(obj_ver_id));
+            memmove(result + op->list_stable_limit, result + stable_count, unstable_count);
             stable_count = op->list_stable_limit;
         }
     }

src/blockstore/blockstore_impl.h

@@ -9,9 +9,12 @@
 #include "ondisk_formats.h"
 
 #include <sys/types.h>
+#include <sys/ioctl.h>
 #include <sys/stat.h>
+#include <fcntl.h>
 #include <time.h>
 #include <unistd.h>
+#include <linux/fs.h>
 
 #include <vector>
 #include <list>
@@ -39,7 +42,7 @@ struct blockstore_op_private_t
     int op_state;
 
     // Write, sync, stabilize
-    uint32_t modified_block;
+    uint32_t modified_block, modified_block2;
 
     // Read
     std::vector<copy_buffer_t> read_vec;
@@ -52,9 +55,7 @@ struct blockstore_op_private_t
     uint32_t write_type;
 
     // Stabilize, rollback
-    heap_entry_t *obj;
-    int stab_pos, res;
-    uint32_t modified_block2;
+    int stab_pos;
 
     // Write
     timespec tv_begin;
@@ -108,9 +109,6 @@ public:
     uint8_t* meta_superblock = NULL;
     uint8_t *buffer_area = NULL;
     std::vector<blockstore_op_t*> submit_queue;
-    // Reads waiting for their disk requests. They're out of submit_queue already, but we
-    // still own their private state, including the read vector - see the destructor
-    robin_hood::unordered_flat_set<blockstore_op_t*> reads_in_flight;
     int unsynced_data_write_count = 0, unsynced_buffer_write_count = 0, unsynced_meta_write_count = 0;
     int unsynced_queued_ops = 0;
 

src/blockstore/blockstore_init.cpp

@@ -17,20 +17,6 @@
         throw std::runtime_error("io_uring is full during initialization");\
     data = ((ring_data_t*)sqe->user_data)
 
-struct blockstore_recheck_io_t
-{
-    bool is_data;
-    uint64_t offset;
-    uint64_t len;
-    uint8_t *buf;
-    std::function<void(uint8_t *buf)> cb;
-
-    blockstore_recheck_io_t(bool is_data, uint64_t offset, uint64_t len, std::function<void(uint8_t *buf)>&& cb):
-        is_data(is_data), offset(offset), len(len), buf(NULL), cb(std::move(cb))
-    {
-    }
-};
-
 blockstore_init_meta::blockstore_init_meta(blockstore_impl_t *bs)
 {
     this->bs = bs;
@@ -253,41 +239,29 @@ resume_4:
     // asynchronous recheck
 resume_6:
     wait_state = 6;
-    bs->heap->recheck_small_writes([this](bool is_data, uint64_t offset, uint64_t len, std::function<void(uint8_t*)> cb)
+    bs->heap->recheck_small_writes([this](bool is_data, uint64_t offset, uint64_t len, uint8_t *buf, std::function<void()> cb)
     {
-        if (!len)
+        if (!buf)
         {
             wait_state = 7;
             bs->ringloop->wakeup();
             return;
         }
-        recheck_ios.push_back(new blockstore_recheck_io_t(is_data, offset, len, std::move(cb)));
-    }, bs->meta_write_recheck_parallelism);
-    while (recheck_ios.size())
-    {
-        blockstore_recheck_io_t* rc = recheck_ios.back();
-        sqe = bs->get_sqe();
-        if (!sqe)
-            break;
-        if (!rc->buf)
-            rc->buf = (uint8_t*)memalign_or_die(MEM_ALIGNMENT, rc->len);
-        data = ((ring_data_t*)sqe->user_data);
-        data->iov = (iovec){ rc->buf, rc->len };
-        data->callback = [this, rc](ring_data_t *data)
+        GET_SQE();
+        data->iov = (iovec){ buf, len };
+        data->callback = [offset, cb](ring_data_t *data)
         {
             if (data->res < 0)
             {
-                fprintf(stderr, "Buffer area read failed at offset %ju: %d\n", rc->offset, data->res);
+                fprintf(stderr, "Buffer area read failed at offset %ju: %d\n", offset, data->res);
                 exit(1);
             }
-            rc->cb(rc->buf);
-            free(rc->buf);
-            delete rc;
+            cb();
         };
-        io_uring_prep_readv(sqe, (rc->is_data ? bs->dsk.data_fd : bs->dsk.journal_fd), &data->iov, 1,
-            (rc->is_data ? bs->dsk.data_offset : bs->dsk.journal_offset) + rc->offset);
-        recheck_ios.pop_back();
-    }
+        io_uring_prep_readv(sqe, (is_data ? bs->dsk.data_fd : bs->dsk.journal_fd), &data->iov, 1,
+            (is_data ? bs->dsk.data_offset : bs->dsk.journal_offset) + offset);
+        bs->ringloop->submit();
+    }, bs->meta_write_recheck_parallelism);
     return 1;
 resume_7:
     if (bs->heap->finish_recheck() != 0)

src/blockstore/blockstore_init.h

@@ -11,8 +11,6 @@ struct blockstore_init_meta_buf
     int state = 0;
 };
 
-struct blockstore_recheck_io_t;
-
 class blockstore_init_meta
 {
     blockstore_impl_t *bs;
@@ -30,7 +28,6 @@ class blockstore_init_meta
     uint64_t last_read_offset = 0;
     uint64_t entries_loaded = 0;
     std::vector<uint32_t> recheck_mod;
-    std::vector<blockstore_recheck_io_t*> recheck_ios;
     int i = 0, j = 0;
     bool handle_meta_block(uint8_t *buf, uint64_t count, uint64_t done_cnt);
     void handle_event(ring_data_t *data, int buf_num, const char *op);

src/blockstore/blockstore_internal.h

@@ -43,20 +43,6 @@
 #define WAIT_SQE 1
 // Suspend operation until there are <wait_detail> bytes of free space in the journal on disk
 #define WAIT_COMPACTION 2
-// Suspend operation until LSN <wait_detail> becomes durable
-#define WAIT_META_FSYNC 3
-
-// In non-immediate_commit setups bs_heap may require us to write into the same block as the
-// previous object's entry block. EAGAIN is returned if it's impossible, in that case we
-// request an fsync and fsync lifts the placement restriction entirely.
-#define BS_SUBMIT_CHECK_PLACEMENT(res, prev_lsn) \
-    if ((res) == EAGAIN)\
-    {\
-        PRIV(op)->wait_for = WAIT_META_FSYNC;\
-        PRIV(op)->wait_detail = (prev_lsn);\
-        flusher->request_fsync();\
-        return 0;\
-    }
 
 #define COPY_BUF_JOURNAL    0x01
 #define COPY_BUF_DATA       0x02

src/blockstore/blockstore_open.cpp

@@ -105,12 +105,12 @@ void blockstore_impl_t::parse_config(blockstore_config_t & config, bool init)
     {
         meta_write_recheck_parallelism = 16;
     }
-    if (immediate_commit != IMMEDIATE_NONE && (!dsk.disable_journal_fsync || !dsk.disable_meta_fsync))
+    if (immediate_commit != IMMEDIATE_NONE && !dsk.disable_journal_fsync)
     {
-        throw std::runtime_error("immediate_commit requires disable_journal_fsync and disable_meta_fsync");
+        throw std::runtime_error("immediate_commit requires disable_journal_fsync");
     }
     if (immediate_commit == IMMEDIATE_ALL && !dsk.disable_data_fsync)
     {
-        throw std::runtime_error("immediate_commit=all requires disable_data_fsync");
+        throw std::runtime_error("immediate_commit=all requires disable_journal_fsync and disable_data_fsync");
     }
 }

src/blockstore/blockstore_read.cpp

@@ -75,12 +75,10 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *op)
                 return false;
             }
         }
-        if (need_skip && (wr->type() == BS_HEAP_SMALL_WRITE || wr->type() == BS_HEAP_INTENT_WRITE) &&
+        if (need_skip && wr->type() == BS_HEAP_SMALL_WRITE &&
             wr->small().offset < blk_end && wr->small().offset+wr->small().len > blk_start)
         {
-            // Small write may mutate big write checksums during flush, and an intent write
-            // overwrites part of the data block in place right away. Either way the checksum
-            // of a checksum block they only partially cover no longer matches the disk
+            // Small write may mutate big write checksums during flush
             skip_csum = COPY_BUF_SKIP_CSUM;
         }
         return true;
@@ -119,7 +117,6 @@ undo_wait:
         return 2;
     }
     op->retval = 0;
-    reads_in_flight.insert(op);
     return 2;
 }
 
@@ -137,8 +134,7 @@ int blockstore_impl_t::fulfill_read(blockstore_op_t *op)
         }
         else if ((vec.copy_flags & COPY_BUF_JOURNAL) && dsk.inmemory_journal)
         {
-            // disk_loc may be negative - make ubsan happy
-            memcpy(op->buf + vec.offset - op->offset, buffer_area + (vec.disk_loc + vec.disk_offset), vec.len);
+            memcpy(op->buf + vec.offset - op->offset, buffer_area + vec.disk_loc + vec.disk_offset, vec.len);
         }
         else
         {
@@ -259,7 +255,7 @@ uint32_t blockstore_impl_t::prepare_read_simple(std::vector<copy_buffer_t> & rea
                 blk_end = blk_end > wr->small().offset+wr->small().len ? wr->small().offset+wr->small().len : blk_end;
             }
             if ((blk_end-1)/dsk.csum_block_size == blk_start/dsk.csum_block_size ||
-                (blk_end-1)/dsk.csum_block_size == blk_start/dsk.csum_block_size+1 && blk_end != end && blk_start != start ||
+                blk_end/dsk.csum_block_size == blk_start/dsk.csum_block_size+1 && blk_end != end && blk_start != start ||
                 blk_end == end && blk_start == start)
             {
                 // single block, exactly two partial blocks, or any number of full blocks
@@ -408,7 +404,6 @@ void blockstore_impl_t::handle_read_event(ring_data_t *data, blockstore_op_t *op
             op->retval = op->len;
         heap->unlock_entry(op->oid);
         free_read_buffers(PRIV(op)->read_vec);
-        reads_in_flight.erase(op);
         FINISH_OP(op);
     }
 }

src/blockstore/blockstore_stable.cpp

@@ -14,8 +14,6 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
     else if (priv->op_state == 3) goto resume_3;
     else if (priv->op_state == 4) goto resume_4;
     else if (priv->op_state == 5) goto resume_5;
-    else if (priv->op_state == 6) goto resume_6;
-    else if (priv->op_state == 7) goto resume_7;
     assert(!priv->op_state);
     op->retval = 0;
     PRIV(op)->lsn = 0;
@@ -23,50 +21,30 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
     for (priv->stab_pos = 0; priv->stab_pos < op->len; priv->stab_pos++)
     {
         {
-            priv->obj = heap->read_entry(v[priv->stab_pos].oid);
-            if (!priv->obj)
+            auto obj = heap->read_entry(v[priv->stab_pos].oid);
+            if (!obj)
             {
                 op->retval = -ENOENT;
                 FINISH_OP(op);
                 return 2;
             }
             priv->modified_block2 = UINT32_MAX;
-            priv->res = op->opcode == BS_OP_STABLE
-                ? heap->add_commit(priv->obj, v[priv->stab_pos].version, &priv->modified_block2)
-                : heap->add_rollback(priv->obj, v[priv->stab_pos].version, &priv->modified_block2);
-            if (priv->res != 0)
-            {
-                // Flush the previous modified block on error or nobody may flush it at all
-resume_6:
-                if (priv->modified_block != UINT32_MAX)
-                {
-                    priv->op_state = 6;
-                    BS_SUBMIT_CHECK_SQES(1);
-                    prepare_meta_block_write(priv->modified_block);
-resume_7:
-                    if (meta_block_is_pending(priv->modified_block))
-                    {
-                        priv->op_state = 7;
-                        return 1;
-                    }
-                    priv->modified_block = UINT32_MAX;
-                    priv->op_state = 0;
-                }
-            }
-            BS_SUBMIT_CHECK_PLACEMENT(priv->res, priv->obj->lsn);
-            if (priv->res == EBUSY)
+            int res = op->opcode == BS_OP_STABLE
+                ? heap->add_commit(obj, v[priv->stab_pos].version, &priv->modified_block2)
+                : heap->add_rollback(obj, v[priv->stab_pos].version, &priv->modified_block2);
+            if (res == EBUSY)
             {
                 op->retval = -EBUSY;
                 FINISH_OP(op);
                 return 2;
             }
-            if (priv->res == ENOENT)
+            if (res == ENOENT)
             {
                 op->retval = -ENOENT;
                 FINISH_OP(op);
                 return 2;
             }
-            if (priv->res == ENOSPC)
+            if (res == ENOSPC)
             {
                 if (!heap->get_to_compact_count())
                 {
@@ -80,12 +58,11 @@ resume_7:
                 flusher->request_trim();
                 return 0;
             }
-            assert(priv->res == 0);
+            assert(res == 0);
         }
 resume_1:
         if (priv->modified_block != UINT32_MAX && priv->modified_block2 != priv->modified_block)
         {
-            priv->op_state = 1;
             BS_SUBMIT_CHECK_SQES(1);
             prepare_meta_block_write(priv->modified_block);
 resume_2:

src/blockstore/blockstore_write.cpp

@@ -34,7 +34,6 @@ void blockstore_impl_t::prepare_meta_block_write(uint32_t modified_block)
         assert(it != modified_blocks.end());
         free(it->second.buf);
         modified_blocks.erase(it);
-        unsynced_meta_write_count++;
         heap->complete_block_write(modified_block);
         ringloop->wakeup();
     };
@@ -42,6 +41,7 @@ void blockstore_impl_t::prepare_meta_block_write(uint32_t modified_block)
     io_uring_prep_writev(
         sqe, dsk.meta_fd, &data->iov, 1, dsk.meta_offset + ((uint64_t)modified_block+1)*dsk.meta_block_size
     );
+    unsynced_meta_write_count++;
     pending_modified_blocks.push_back(modified_block);
     modified_blocks[modified_block] = { .sent = false, .buf = buf };
 }
@@ -77,36 +77,34 @@ bool blockstore_impl_t::intent_write_allowed(blockstore_op_t *op, heap_entry_t *
     {
         return false;
     }
-    bool ok = false;
+    bool ok = true;
     heap->iterate_with_stable(obj, obj->lsn, [&](heap_entry_t *wr, bool stable)
     {
+        // Intent writes are not allowed over buffered writes
         auto t = wr->type();
-        if (!stable)
+        if (t == BS_HEAP_SMALL_WRITE)
         {
-            // Intent writes are not allowed over unstable writes
+            ok = false;
             return false;
         }
-        else if (t == BS_HEAP_INTENT_WRITE || t == BS_HEAP_BIG_INTENT)
+        // Intent writes are not allowed over unstable writes
+        if (!stable)
         {
-            // Intent writes are not allowed over unfinished intent writes
-            ok = (wr->lsn <= heap->get_fsynced_lsn());
+            ok = false;
+            return false;
         }
-        // Intent writes are allowed over BIG_WRITEs even with enable data_fsync
-        // because BIG_WRITEs are always fsynced before adding the entry
-        else if (t == BS_HEAP_BIG_WRITE)
+        // Intent writes are not allowed over unfinished intent writes
+        if ((t == BS_HEAP_INTENT_WRITE || t == BS_HEAP_BIG_INTENT) && wr->lsn > heap->get_fsynced_lsn())
         {
-            // But sadly with one exception: a BIG_WRITE may be a result of compaction, with a
-            // reused older LSN, still in progress of being written. In that case it can still
-            // disappear after a reboot, and if it disappears, an older small_write may be replayed
-            // again over our intent write data if it's there. So we have to check the previous write too...
-            if (wr->lsn == flusher->get_flushing_lsn({ .inode = wr->inode, .stripe = wr->stripe }))
-            {
-                return true;
-            }
-            ok = true;
+            ok = false;
+            return false;
         }
-        // Intent writes are not allowed over buffered writes and other write types
-        return false;
+        // Intent writes are allowed over BIG_WRITEs even with fsyncs because BIG_WRITE is always counted as fsynced
+        if (t == BS_HEAP_BIG_WRITE || t == BS_HEAP_BIG_INTENT)
+        {
+            return false;
+        }
+        return true;
     });
     return ok;
 }
@@ -127,8 +125,7 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
     if (op->opcode == BS_OP_DELETE)
     {
         // Delete
-        // An absent chain (deleted or fully rolled back) is already "not there"
-        if (!obj || heap->is_absent(obj))
+        if (!obj || obj->type() == BS_HEAP_DELETE)
         {
             // Already deleted
             op->retval = 0;
@@ -138,7 +135,6 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
         PRIV(op)->write_type = BS_HEAP_DELETE;
         BS_SUBMIT_CHECK_SQES(1);
         int res = heap->add_delete(obj, &PRIV(op)->modified_block);
-        assert(res != EAGAIN);
         if (res == ENOSPC)
             goto enospc;
         assert(res == 0);
@@ -149,7 +145,7 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
     }
     // FIXME: Allow to do initial writes as buffered, not redirected
     // FIXME: Allow to do direct writes over holes
-    else if (!obj || heap->is_absent(obj) || op->offset == 0 && op->len == dsk.data_block_size)
+    else if (!obj || obj->type() == BS_HEAP_DELETE || op->offset == 0 && op->len == dsk.data_block_size)
     {
         // Big (redirect) write
         PRIV(op)->write_type = dsk.disable_data_fsync || op->opcode != BS_OP_WRITE_STABLE ? BS_HEAP_BIG_WRITE : _REDIRECT_INTENT;
@@ -184,8 +180,7 @@ enospc:
         data->callback = [this, op](ring_data_t *data) { handle_write_event(data, op); };
         assert(loc+op->offset+op->len <= dsk.block_count*dsk.data_block_size);
         io_uring_prep_writev(sqe, dsk.data_fd, &data->iov, 1, dsk.data_offset + loc + op->offset);
-        // redirect_intent writes skip the data fsync (resume_2), so they must not take a slot
-        if (!dsk.disable_data_fsync && PRIV(op)->write_type == BS_HEAP_BIG_WRITE)
+        if (!dsk.disable_data_fsync)
         {
             // use PRIV->lsn for fsync_data_id
             PRIV(op)->lsn = ++data_fsync_next;
@@ -213,7 +208,6 @@ enospc:
             PRIV(op)->location = obj->big_location(heap);
             res = heap->add_big_intent(op->oid, &obj, op->version, op->offset, op->len, op->bitmap,
                 (uint8_t*)op->buf, NULL, &PRIV(op)->modified_block);
-            assert(res != EAGAIN);
             if (res == ENOSPC)
                 goto enospc;
             assert(res == 0);
@@ -231,7 +225,6 @@ enospc:
             PRIV(op)->location = wr->big_location(heap);
             res = heap->add_small_write(op->oid, &obj, (BS_HEAP_INTENT_WRITE | (op->opcode == BS_OP_WRITE_STABLE ? BS_HEAP_STABLE : 0)),
                 op->version, op->offset, op->len, 0, op->bitmap, (uint8_t*)op->buf, &PRIV(op)->modified_block);
-            BS_SUBMIT_CHECK_PLACEMENT(res, obj->lsn);
             if (res == ENOSPC)
                 goto enospc;
             assert(res == 0);
@@ -250,12 +243,6 @@ enospc:
         uint64_t loc = !op->len ? 0 : heap->find_free_buffer_area(op->len);
         if (loc == UINT64_MAX)
         {
-            if (!heap->get_to_compact_count())
-            {
-                op->retval = -EAGAIN;
-                FINISH_OP(op);
-                return 2;
-            }
             PRIV(op)->wait_for = WAIT_COMPACTION;
             PRIV(op)->wait_detail = heap->get_compacted_count();
             flusher->request_trim();
@@ -265,7 +252,6 @@ enospc:
         BS_SUBMIT_CHECK_SQES(1 + (op->len > 0 ? 1 : 0));
         int res = heap->add_small_write(op->oid, &obj, (BS_HEAP_SMALL_WRITE | (op->opcode == BS_OP_WRITE_STABLE ? BS_HEAP_STABLE : 0)),
             op->version, op->offset, op->len, loc, op->bitmap, (uint8_t*)op->buf, &PRIV(op)->modified_block);
-        BS_SUBMIT_CHECK_PLACEMENT(res, obj->lsn);
         if (res == ENOSPC)
             goto enospc;
         assert(res == 0);
@@ -357,9 +343,6 @@ resume_11:
             data->iov = { 0 };
             data->callback = [this, op, fs = data_fsync_cur](ring_data_t *data)
             {
-                live = true;
-                if (data->res != 0)
-                    disk_error_abort("sync data", data->res, 0);
                 if (fs > data_fsync_done)
                 {
                     data_fsync_done = fs;
@@ -389,7 +372,6 @@ resume_4:
             res = heap->add_big_write(op->oid, obj, op->opcode == BS_OP_WRITE_STABLE,
                 op->version, op->offset, op->len, PRIV(op)->location, op->bitmap, (uint8_t*)op->buf, &PRIV(op)->modified_block);
         }
-        assert(res != EAGAIN);
         if (res == ENOSPC)
         {
             if (!heap->get_to_compact_count())

src/blockstore/fio_engine.cpp

@@ -262,7 +262,7 @@ static enum fio_q_status _bs_queue(struct thread_data *td, struct io_u *io, bool
             bsd->inflight--;
             bsd->completed.push_back(io);
             if (bsd->trace)
-                printf("--- OP_READ %zx n=%d retval=%jd\n", (size_t)op, n, op->retval);
+                printf("--- OP_READ %zx n=%d retval=%d\n", (size_t)op, n, op->retval);
             delete op;
         };
         break;
@@ -283,7 +283,7 @@ static enum fio_q_status _bs_queue(struct thread_data *td, struct io_u *io, bool
             {
                 bs_data *bsd = ((bs_data*)((thread_data*)io->engine_data)->io_ops_data);
                 if (bsd->trace)
-                    printf("--- OP_WRITE %zx n=%d retval=%jd\n", (size_t)op, n, op->retval);
+                    printf("--- OP_WRITE %zx n=%d retval=%d\n", (size_t)op, n, op->retval);
                 if (op->retval < 0)
                 {
                     io->error = op->retval < 0 ? -op->retval : 0;
@@ -305,7 +305,7 @@ static enum fio_q_status _bs_queue(struct thread_data *td, struct io_u *io, bool
                     {
                         bs_data *bsd = ((bs_data*)((thread_data*)io->engine_data)->io_ops_data);
                         if (bsd->trace)
-                            printf("--- OP_STABLE %zx n=%d retval=%jd\n", (size_t)op, n, op->retval);
+                            printf("--- OP_STABLE %zx n=%d retval=%d\n", (size_t)op, n, op->retval);
                         io->error = op->retval < 0 ? -op->retval : 0;
                         bsd->inflight--;
                         bsd->completed.push_back(io);
@@ -323,7 +323,7 @@ static enum fio_q_status _bs_queue(struct thread_data *td, struct io_u *io, bool
             {
                 bs_data *bsd = ((bs_data*)((thread_data*)io->engine_data)->io_ops_data);
                 if (bsd->trace)
-                    printf("--- OP_WRITE_STABLE %zx n=%d retval=%jd\n", (size_t)op, n, op->retval);
+                    printf("--- OP_WRITE_STABLE %zx n=%d retval=%d\n", (size_t)op, n, op->retval);
                 io->error = op->retval < 0 ? -op->retval : 0;
                 bsd->inflight--;
                 bsd->completed.push_back(io);
@@ -342,7 +342,7 @@ static enum fio_q_status _bs_queue(struct thread_data *td, struct io_u *io, bool
             bsd->completed.push_back(io);
             bsd->inflight--;
             if (bsd->trace)
-                printf("--- OP_SYNC %zx n=%d retval=%jd\n", (size_t)op, n, op->retval);
+                printf("--- OP_SYNC %zx n=%d retval=%d\n", (size_t)op, n, op->retval);
             delete op;
         };
         bsd->last_sync = true;

src/blockstore/multilist.cpp

@@ -2,7 +2,6 @@
 // Copyright (c) Vitaliy Filippov, 2025+
 // License: VNPL-1.1 (see README.md for details)
 
-#undef NDEBUG
 #include <assert.h>
 #include <stdio.h>
 #include <stdlib.h>

src/blockstore/v1/flush.cpp

@@ -35,6 +35,7 @@ journal_flusher_t::journal_flusher_t(blockstore_impl_t *bs)
 
 journal_flusher_co::journal_flusher_co()
 {
+    wait_state = 0;
     simple_callback_r = [this](ring_data_t* data)
     {
         bs->live = true;
@@ -58,24 +59,6 @@ journal_flusher_co::journal_flusher_co()
     };
 }
 
-journal_flusher_co::~journal_flusher_co()
-{
-    // A flush interrupted by an emergency shutdown still holds its read buffers and
-    // references to entry bitmaps/checksums. free_buffers() can't be used here because
-    // it also touches the shared metadata sector map, which may already be gone.
-    for (auto & vi: v)
-    {
-        if (vi.buf && (vi.copy_flags == COPY_BUF_JOURNAL || (vi.copy_flags & COPY_BUF_CSUM_FILL)) &&
-            (!bs->journal.inmemory || vi.buf < bs->journal.buffer ||
-            vi.buf >= (uint8_t*)bs->journal.buffer + bs->journal.len))
-        {
-            free(vi.buf);
-        }
-        bs->free_dirty_dyn_data(vi);
-    }
-    v.clear();
-}
-
 journal_flusher_t::~journal_flusher_t()
 {
     if (!bs->journal.inmemory)
@@ -93,16 +76,6 @@ size_t journal_flusher_t::get_queue_size()
     return flush_queue.size();
 }
 
-bool journal_flusher_t::may_advance()
-{
-    return flush_queue.size() || active_flushers > 0 || trim_possible;
-}
-
-bool journal_flusher_t::may_advance_except_one()
-{
-    return flush_queue.size() || active_flushers > 1 || trim_possible;
-}
-
 void journal_flusher_t::loop()
 {
     target_flusher_count = bs->write_iodepth*2;
@@ -217,7 +190,6 @@ void journal_flusher_t::request_trim()
 
 void journal_flusher_t::mark_trim_possible()
 {
-    trim_possible = true;
     if (trim_wanted > 0)
     {
         dequeuing = true;
@@ -680,10 +652,9 @@ resume_2:
         flusher->sync_to_repeat.erase(repeat_it);
     trim_journal:
         // Clear unused part of the journal every <journal_trim_interval> flushes
-        if ((bs->journal_trim_interval && !((++flusher->journal_trim_counter) % bs->journal_trim_interval) ||
-            flusher->trim_wanted > 0) && flusher->trim_possible)
+        if (bs->journal_trim_interval && !((++flusher->journal_trim_counter) % bs->journal_trim_interval) ||
+            flusher->trim_wanted > 0)
         {
-            flusher->trim_possible = false;
     resume_30:
     resume_31:
     resume_32:
@@ -800,8 +771,6 @@ void journal_flusher_co::free_buffers()
         {
             free(it->buf);
         }
-        // prepare_read() takes a reference to the entry's bitmap/checksums, drop it
-        bs->free_dirty_dyn_data(*it);
     }
     v.clear();
 }
@@ -1527,19 +1496,10 @@ bool journal_flusher_co::trim_journal(int wait_base)
                 printf("Journal flushed\n");
                 exit(0);
             }
-            bs->wakeup_wait_journal();
-        }
-        else if (!flusher->may_advance_except_one())
-        {
-            bs->wakeup_wait_journal();
         }
         flusher->journal_trim_counter = 0;
         flusher->trimming = false;
     }
-    else if (!flusher->may_advance_except_one())
-    {
-        bs->wakeup_wait_journal();
-    }
     return true;
 }
 

src/blockstore/v1/flush.h

@@ -3,34 +3,34 @@
 
 struct copy_buffer_t
 {
-    int copy_flags = 0;
-    uint64_t offset = 0, len = 0, disk_offset = 0;
-    uint64_t journal_sector = 0; // only for reads: sector+1 if used and !journal.inmemory, otherwise 0
-    void *buf = NULL;
-    uint8_t *csum_buf = NULL;
-    int *dyn_data = NULL;
+    int copy_flags;
+    uint64_t offset, len, disk_offset;
+    uint64_t journal_sector; // only for reads: sector+1 if used and !journal.inmemory, otherwise 0
+    void *buf;
+    uint8_t *csum_buf;
+    int *dyn_data;
 };
 
 struct meta_sector_t
 {
-    uint64_t offset = 0, len = 0;
-    int state = 0;
-    void *buf = NULL;
-    int usage_count = 0;
+    uint64_t offset, len;
+    int state;
+    void *buf;
+    int usage_count;
 };
 
 struct flusher_sync_t
 {
-    bool fsync_meta = false;
-    int ready_count = 0;
-    int state = 0;
+    bool fsync_meta;
+    int ready_count;
+    int state;
 };
 
 struct flusher_meta_write_t
 {
-    uint64_t sector = 0, pos = 0;
-    bool submitted = 0;
-    void *buf = NULL;
+    uint64_t sector, pos;
+    bool submitted;
+    void *buf;
     std::map<uint64_t, meta_sector_t>::iterator it;
 };
 
@@ -39,37 +39,36 @@ class journal_flusher_t;
 // Journal flusher coroutine
 class journal_flusher_co
 {
-    blockstore_impl_t *bs = NULL;
-    journal_flusher_t *flusher = NULL;
-    int wait_state = 0, wait_count = 0, wait_journal_count = 0;
-    struct io_uring_sqe *sqe = NULL;
-    struct ring_data_t *data = NULL;
+    blockstore_impl_t *bs;
+    journal_flusher_t *flusher;
+    int wait_state, wait_count, wait_journal_count;
+    struct io_uring_sqe *sqe;
+    struct ring_data_t *data;
 
     std::list<flusher_sync_t>::iterator cur_sync;
 
-    obj_ver_id cur = {};
+    obj_ver_id cur;
     std::map<obj_ver_id, dirty_entry>::iterator dirty_it, dirty_start, dirty_end;
     std::map<object_id, uint64_t>::iterator repeat_it;
     std::function<void(ring_data_t*)> simple_callback_r, simple_callback_rj, simple_callback_w;
 
     bool try_trim = false;
-    bool skip_copy = false, has_delete = false, has_writes = false;
-    ~journal_flusher_co();
+    bool skip_copy, has_delete, has_writes;
     std::vector<copy_buffer_t> v;
     std::vector<copy_buffer_t>::iterator it;
-    int i = 0;
-    bool fill_incomplete = false, cleared_incomplete = false;
-    int read_to_fill_incomplete = 0;
-    int copy_count = 0;
-    uint64_t clean_loc = 0, clean_ver = 0, old_clean_loc = 0, old_clean_ver = 0;
+    int i;
+    bool fill_incomplete, cleared_incomplete;
+    int read_to_fill_incomplete;
+    int copy_count;
+    uint64_t clean_loc, clean_ver, old_clean_loc, old_clean_ver;
     flusher_meta_write_t meta_old, meta_new;
-    bool clean_init_bitmap = false;
-    uint64_t clean_bitmap_offset = 0, clean_bitmap_len = 0;
-    uint8_t *clean_init_dyn_ptr = NULL;
-    uint8_t *new_clean_bitmap = NULL;
+    bool clean_init_bitmap;
+    uint64_t clean_bitmap_offset, clean_bitmap_len;
+    uint8_t *clean_init_dyn_ptr;
+    uint8_t *new_clean_bitmap;
     std::unordered_set<uint32_t> mangle_csum_blocks;
 
-    uint64_t new_trim_pos = 0;
+    uint64_t new_trim_pos;
 
     friend class journal_flusher_t;
     void scan_dirty();
@@ -95,20 +94,19 @@ public:
 class journal_flusher_t
 {
     int trim_wanted = 0;
-    bool trim_possible = false;
-    bool dequeuing = false;
-    int min_flusher_count = 0, max_flusher_count = 0, cur_flusher_count = 0, target_flusher_count = 0;
-    int flusher_start_threshold = 0;
-    journal_flusher_co *co = NULL;
-    blockstore_impl_t *bs = NULL;
+    bool dequeuing;
+    int min_flusher_count, max_flusher_count, cur_flusher_count, target_flusher_count;
+    int flusher_start_threshold;
+    journal_flusher_co *co;
+    blockstore_impl_t *bs;
     friend class journal_flusher_co;
 
-    int journal_trim_counter = 0;
-    bool trimming = false;
-    void* journal_superblock = NULL;
+    int journal_trim_counter;
+    bool trimming;
+    void* journal_superblock;
 
-    int active_flushers = 0;
-    int syncing_flushers = 0;
+    int active_flushers;
+    int syncing_flushers;
     std::list<flusher_sync_t> syncs;
     std::map<object_id, uint64_t> sync_to_repeat;
 
@@ -119,14 +117,12 @@ class journal_flusher_t
 
     bool try_find_older(std::map<obj_ver_id, dirty_entry>::iterator & dirty_end, obj_ver_id & cur);
     bool try_find_other(std::map<obj_ver_id, dirty_entry>::iterator & dirty_end, obj_ver_id & cur);
-    bool may_advance_except_one();
 
 public:
     journal_flusher_t(blockstore_impl_t *bs);
     ~journal_flusher_t();
     void loop();
     bool is_trim_wanted() { return trim_wanted; }
-    bool may_advance();
     bool is_active();
     size_t get_queue_size();
     void mark_trim_possible();

src/blockstore/v1/impl.cpp

@@ -24,7 +24,6 @@ blockstore_impl_t::blockstore_impl_t(blockstore_config_t & config, ring_loop_i *
         calc_lengths();
         alloc_dyn_data = dsk.clean_dyn_size > sizeof(void*) || dsk.csum_block_size > 0;
         zero_object = (uint8_t*)memalign_or_die(MEM_ALIGNMENT, dsk.data_block_size);
-        memset(zero_object, 0, dsk.data_block_size);
         data_alloc = new allocator_t(dsk.block_count);
     }
     catch (std::exception & e)
@@ -35,47 +34,13 @@ blockstore_impl_t::blockstore_impl_t(blockstore_config_t & config, ring_loop_i *
     flusher = new journal_flusher_t(this);
 }
 
-// Submit an operation the blockstore created for itself, and remember it so that it can
-// be cleaned up if we're destroyed before it completes
-void blockstore_impl_t::enqueue_own_op(blockstore_op_t *op)
-{
-    own_ops.insert(op);
-    auto cb = op->callback;
-    op->callback = [this, cb](blockstore_op_t *op)
-    {
-        own_ops.erase(op);
-        cb(op);
-    };
-    enqueue_op(op);
-}
-
 blockstore_impl_t::~blockstore_impl_t()
 {
-    // Destroy private state of operations still in the queue
-    for (auto op: submit_queue)
-    {
-        if (op && !own_ops.count(op))
-            PRIV(op)->~blockstore_op_private_t();
-    }
-    submit_queue.clear();
-    // Reads which were waiting for the disk own their read vector, and nobody is going to
-    // complete them now
-    for (auto op: reads_in_flight)
-    {
-        release_read_vec(op);
-        PRIV(op)->~blockstore_op_private_t();
-    }
-    reads_in_flight.clear();
-    // Destroy our own operations
-    for (auto op: own_ops)
-    {
-        PRIV(op)->~blockstore_op_private_t();
-        delete op;
-    }
-    own_ops.clear();
-    // Clear dyn_data in dirty_db
     for (auto& obj: dirty_db)
-        free_dirty_dyn_data(obj.second);
+    {
+        if (obj.second.dyn_data)
+            free(obj.second.dyn_data);
+    }
     delete data_alloc;
     delete flusher;
     if (zero_object)
@@ -160,7 +125,6 @@ void blockstore_impl_t::loop()
         // has_writes == 1 - some writes in progress
         // has_writes == 2 - tried to submit some writes, but failed
         int has_writes = 0, op_idx = 0, new_idx = 0;
-        bool has_unfinished_sync = false;
         for (; op_idx < submit_queue.size(); op_idx++, new_idx++)
         {
             auto op = submit_queue[op_idx];
@@ -218,22 +182,7 @@ void blockstore_impl_t::loop()
                 // sync only completed writes?
                 // wait for the data device fsync to complete, then submit journal writes for big writes
                 // then submit an fsync operation
-                //
-                // Only one sync at a time: continue_sync() takes the unsynced writes over into
-                // its own state, so a second sync started while the first one is still in flight
-                // would find nothing left to sync and report success without anything having
-                // reached the disk - while the writes it was supposed to cover are still in the
-                // first sync's hands.
-                // A deletion takes them over in the very same way, see
-                // forget_unstable_before_delete() - so wait for it to be journaled and acked,
-                // and it lands in the unsynced lists itself by then
-                if (!has_unfinished_sync && !unsynced_forgotten)
-                {
-                    wr_st = continue_sync(op);
-                    has_unfinished_sync = (wr_st != 2);
-                }
-                else
-                    wr_st = 0;
+                wr_st = continue_sync(op);
             }
             else if (op->opcode == BS_OP_STABLE)
             {
@@ -262,6 +211,10 @@ void blockstore_impl_t::loop()
                     // ring is full, stop submission
                     break;
                 }
+                else if (PRIV(op)->wait_for == WAIT_JOURNAL)
+                {
+                    PRIV(op)->wait_detail2 = (unstable_writes.size()+unstable_unsynced);
+                }
             }
         }
         if (op_idx != new_idx)
@@ -318,7 +271,7 @@ bool blockstore_impl_t::is_safe_to_stop()
             {
                 delete op;
             };
-            enqueue_own_op(op);
+            enqueue_op(op);
             stop_sync_submitted = true;
         }
         return false;
@@ -326,12 +279,6 @@ bool blockstore_impl_t::is_safe_to_stop()
     return true;
 }
 
-void blockstore_impl_t::wakeup_wait_journal()
-{
-    wait_journal_counter++;
-    ringloop->wakeup();
-}
-
 void blockstore_impl_t::check_wait(blockstore_op_t *op)
 {
     if (PRIV(op)->wait_for == WAIT_SQE)
@@ -348,11 +295,12 @@ void blockstore_impl_t::check_wait(blockstore_op_t *op)
     }
     else if (PRIV(op)->wait_for == WAIT_JOURNAL)
     {
-        if (wait_journal_counter == PRIV(op)->wait_detail)
+        if (journal.used_start == PRIV(op)->wait_detail &&
+            (unstable_writes.size()+unstable_unsynced) == PRIV(op)->wait_detail2)
         {
             // do not submit
 #ifdef BLOCKSTORE_DEBUG
-            printf("Still waiting for flush journal counter %ju\n", PRIV(op)->wait_detail);
+            printf("Still waiting to flush journal offset %08jx\n", PRIV(op)->wait_detail);
 #endif
             return;
         }
@@ -428,7 +376,6 @@ void blockstore_impl_t::init_op(blockstore_op_t *op)
     PRIV(op)->wait_for = 0;
     PRIV(op)->op_state = 0;
     PRIV(op)->pending_ops = 0;
-    PRIV(op)->forgot_unsynced = false;
 }
 
 static bool replace_stable(object_id oid, uint64_t version, int search_start, int search_end, obj_ver_id* list)

src/blockstore/v1/impl.h

@@ -3,8 +3,6 @@
 
 #pragma once
 
-#include <set>
-
 #include "blockstore.h"
 #include "blockstore_disk.h"
 #include "ondisk_formats.h"
@@ -95,7 +93,7 @@ struct blockstore_op_private_t
 {
     // Wait status
     int wait_for;
-    uint64_t wait_detail;
+    uint64_t wait_detail, wait_detail2;
     int pending_ops;
     int op_state;
 
@@ -111,9 +109,6 @@ struct blockstore_op_private_t
     // 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;
     timespec tv_begin;
-    // Set on a deletion which dropped preceding unsynced writes of the same object,
-    // see forget_unstable_before_delete()
-    bool forgot_unsynced;
 
     // Sync
     std::vector<obj_ver_id> sync_big_writes, sync_small_writes;
@@ -168,15 +163,8 @@ class blockstore_impl_t: public blockstore_i
     uint8_t *clean_bitmaps = NULL;
     blockstore_dirty_db_t dirty_db;
     std::vector<blockstore_op_t*> submit_queue;
-    // Reads waiting for their disk requests. They're out of submit_queue already, but we
-    // still own their private state, including the read vector - see the destructor
-    std::set<blockstore_op_t*> reads_in_flight;
-    std::set<blockstore_op_t*> own_ops;
     std::vector<obj_ver_id> unsynced_big_writes, unsynced_small_writes;
     int unsynced_big_write_count = 0, unstable_unsynced = 0;
-    // Number of deletions which took over the durability of preceding unsynced writes but
-    // aren't journaled themselves yet, see forget_unstable_before_delete()
-    int unsynced_forgotten = 0;
     int unsynced_queued_ops = 0;
     allocator_t *data_alloc = NULL;
     uint64_t used_blocks = 0;
@@ -211,6 +199,9 @@ class blockstore_impl_t: public blockstore_i
     friend class journal_flusher_co;
 
     void calc_lengths();
+    void open_data();
+    void open_meta();
+    void open_journal();
     uint8_t* get_clean_entry_bitmap(uint64_t block_loc, int offset);
 
     blockstore_clean_db_t& clean_db_shard(object_id oid);
@@ -218,7 +209,6 @@ class blockstore_impl_t: public blockstore_i
 
     // Journaling
     void prepare_journal_sector_write(int sector, blockstore_op_t *op);
-    void mark_not_dirty(int journal_sector);
     void handle_journal_write(ring_data_t *data, uint64_t flush_id);
     void disk_error_abort(const char *op, int retval, int expected);
 
@@ -228,15 +218,12 @@ class blockstore_impl_t: public blockstore_i
     blockstore_init_meta* metadata_init_reader;
     blockstore_init_journal* journal_init_reader;
 
-    uint64_t wait_journal_counter = 0;
-    void wakeup_wait_journal();
     void check_wait(blockstore_op_t *op);
     void init_op(blockstore_op_t *op);
 
     // Read
     int dequeue_read(blockstore_op_t *read_op);
     void release_clean(blockstore_op_t *op);
-    void release_read_vec(blockstore_op_t *read_op);
     void find_holes(std::vector<copy_buffer_t> & read_vec, uint32_t item_start, uint32_t item_end,
         std::function<int(int, bool, uint32_t, uint32_t)> callback);
     int fulfill_read(blockstore_op_t *read_op,
@@ -254,7 +241,7 @@ class blockstore_impl_t: public blockstore_i
         uint64_t offset, uint64_t submit_len, uint64_t & blk_begin, uint64_t & blk_end, uint8_t* & blk_buf);
     bool read_range_fulfilled(std::vector<copy_buffer_t> & rv, uint64_t & fulfilled, uint8_t *read_buf,
         uint8_t *clean_entry_bitmap, uint32_t item_start, uint32_t item_end);
-    bool read_checksum_block(blockstore_op_t *op, int rv_pos, uint64_t &fulfilled, uint64_t clean_loc, bool padded);
+    bool read_checksum_block(blockstore_op_t *op, int rv_pos, uint64_t &fulfilled, uint64_t clean_loc);
     uint8_t* read_clean_meta_block(blockstore_op_t *read_op, uint64_t clean_loc, int rv_pos);
     bool verify_padded_checksums(uint8_t *clean_entry_bitmap, uint8_t *csum_buf, uint32_t offset,
         iovec *iov, int n_iov, std::function<void(uint32_t, uint32_t, uint32_t)> bad_block_cb);
@@ -267,16 +254,13 @@ class blockstore_impl_t: public blockstore_i
     void handle_read_event(ring_data_t *data, blockstore_op_t *op);
 
     // Write
-    bool forget_unstable_before_delete(blockstore_op_t *op, blockstore_dirty_db_t::iterator dirty_it);
     bool enqueue_write(blockstore_op_t *op);
-    void unshift_stable_flush(obj_ver_id ov);
     void cancel_all_writes(blockstore_op_t *op, blockstore_dirty_db_t::iterator dirty_it, int retval);
     int dequeue_write(blockstore_op_t *op);
     int dequeue_del(blockstore_op_t *op);
     int continue_write(blockstore_op_t *op);
     void release_journal_sectors(blockstore_op_t *op);
     void handle_write_event(ring_data_t *data, blockstore_op_t *op);
-    void enqueue_own_op(blockstore_op_t *op);
 
     // Sync
     int continue_sync(blockstore_op_t *op);
@@ -296,7 +280,6 @@ class blockstore_impl_t: public blockstore_i
     void mark_rolled_back(const obj_ver_id & ov);
     void erase_dirty(blockstore_dirty_db_t::iterator dirty_start, blockstore_dirty_db_t::iterator dirty_end, uint64_t clean_loc);
     void free_dirty_dyn_data(dirty_entry & e);
-    void free_dirty_dyn_data(copy_buffer_t & e);
 
     // List
     void process_list(blockstore_op_t *op);

src/blockstore/v1/init.cpp

@@ -510,8 +510,6 @@ int blockstore_init_journal::loop()
         goto resume_6;
     else if (wait_state == 7)
         goto resume_7;
-    else if (wait_state == 8)
-        goto resume_8;
     printf("Reading blockstore journal\n");
     if (!bs->journal.inmemory)
         submitted_buf = memalign_or_die(MEM_ALIGNMENT, 2*bs->journal.block_size);
@@ -677,7 +675,6 @@ resume_1:
                     // zero out corrupted entry, if required
                     if (init_write_buf && !bs->readonly)
                     {
-                        assert(init_write_sector > 0);
                         GET_SQE();
                         data->iov = { init_write_buf, (size_t)bs->journal.block_size };
                         data->callback = simple_callback;
@@ -755,43 +752,6 @@ resume_1:
     }
     bs->flusher->mark_trim_possible();
     bs->journal.dirty_start = bs->journal.next_free;
-    if (!bs->readonly)
-    {
-        // Everything we just read may still be sitting in the volatile write cache of the device:
-        // the instance which wrote it is gone, so nothing is going to flush it any more, and a
-        // sync would find nothing of its own to sync and report success while a power outage
-        // could still take the recovered state away. Make it durable right here instead
-        {
-            bool fsync_data = !bs->disable_data_fsync;
-            bool fsync_meta = !bs->disable_meta_fsync;
-            bool fsync_journal = !bs->disable_journal_fsync;
-            // The same device may serve several roles - fsync each distinct one exactly once
-            if (bs->dsk.meta_fd == bs->dsk.data_fd)
-                fsync_data = fsync_data || fsync_meta, fsync_meta = false;
-            if (bs->dsk.journal_fd == bs->dsk.data_fd)
-                fsync_data = fsync_data || fsync_journal, fsync_journal = false;
-            else if (bs->dsk.journal_fd == bs->dsk.meta_fd)
-                fsync_meta = fsync_meta || fsync_journal, fsync_journal = false;
-            for (int i = 0; i < 3; i++)
-            {
-                if (i == 0 ? !fsync_data : (i == 1 ? !fsync_meta : !fsync_journal))
-                    continue;
-                GET_SQE();
-                data->iov = { 0 };
-                data->callback = simple_callback;
-                io_uring_prep_fsync(sqe, i == 0 ? bs->dsk.data_fd : (i == 1 ? bs->dsk.meta_fd : bs->dsk.journal_fd), IORING_FSYNC_DATASYNC);
-                wait_count++;
-            }
-            if (wait_count > 0)
-                bs->ringloop->submit();
-        }
-    resume_8:
-        if (wait_count > 0)
-        {
-            wait_state = 8;
-            return 1;
-        }
-    }
     printf(
         "Journal entries loaded: %ju, free journal space: %ju bytes (%08jx..%08jx is used), free blocks: %ju / %ju\n",
         entries_loaded,
@@ -829,8 +789,7 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
         {
             auto buf_pos = proc_pos - done_pos + pos;
             journal_entry *je = (journal_entry*)((uint8_t*)buf + buf_pos);
-            if ((bs->journal.block_size-pos) < 12 || je->size < 12 /*crc,magic,type,size*/ ||
-                je->magic != JOURNAL_MAGIC || buf_pos+je->size > len || je_crc32(je) != je->crc32 ||
+            if (je->magic != JOURNAL_MAGIC || buf_pos+je->size > len || je_crc32(je) != je->crc32 ||
                 je->type < JE_MIN || je->type > JE_MAX || started && je->crc32_prev != crc32_last)
             {
                 if (pos == 0)
@@ -988,10 +947,10 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
                             if (block_crc32 != *block_csums)
                             {
                                 printf(
-                                    "Journal entry data is corrupt for small_write%s oid=%jx:%jx ver=%ju offset=%u len=%u data_offset=%jx - block %u crc32 %x != %x\n",
+                                    "Journal entry data is corrupt for small_write%s oid=%jx:%jx ver=%ju offset=%u len=%u - block %u crc32 %x != %x\n",
                                     je->type == JE_SMALL_WRITE_INSTANT ? "_instant" : "",
                                     je->small_write.oid.inode, je->small_write.oid.stripe, je->small_write.version,
-                                    je->small_write.offset, je->small_write.len, (uint64_t)je->small_write.data_offset,
+                                    je->small_write.offset, je->small_write.len,
                                     pos, block_crc32, *block_csums
                                 );
                                 data_csum_valid = false;
@@ -1006,12 +965,7 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
                     // interesting thing is that we must clear the corrupt entry if we're not readonly,
                     // because we don't write next entries in the same journal block
                     memset((uint8_t*)buf + proc_pos - done_pos + pos, 0, bs->journal.block_size - pos);
-                    // If the corrupt entry was the first one in its block then clearing it
-                    // empties the whole block, and an empty block is exactly what marks the
-                    // end of the journal. Resuming after such a block would turn it into a
-                    // permanent barrier: every later replay would stop at it again and
-                    // discard everything written after it. So reuse the block instead.
-                    bs->journal.next_free = pos == 0 ? proc_pos : prev_free;
+                    bs->journal.next_free = prev_free;
                     init_write_buf = (uint8_t*)buf + proc_pos - done_pos;
                     init_write_sector = proc_pos;
                     return 0;
@@ -1045,7 +999,7 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
                         memcpy((uint8_t*)dyn+sizeof(int), dyn_from, dyn_size);
                     }
                     bs->dirty_db.emplace(ov, (dirty_entry){
-                        .state = (uint32_t)(BS_ST_SMALL_WRITE | BS_ST_SYNCED | (je->type == JE_SMALL_WRITE_INSTANT ? BS_ST_INSTANT : 0)),
+                        .state = (BS_ST_SMALL_WRITE | BS_ST_SYNCED),
                         .flags = 0,
                         .location = location,
                         .offset = je->small_write.offset,
@@ -1129,7 +1083,7 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
                         memcpy((uint8_t*)dyn+sizeof(int), dyn_from, dyn_size);
                     }
                     auto dirty_it = bs->dirty_db.emplace(ov, (dirty_entry){
-                        .state = (uint32_t)(BS_ST_BIG_WRITE | BS_ST_SYNCED | (je->type == JE_BIG_WRITE_INSTANT ? BS_ST_INSTANT : 0)),
+                        .state = (BS_ST_BIG_WRITE | BS_ST_SYNCED),
                         .flags = 0,
                         .location = je->big_write.location,
                         .offset = je->big_write.offset,
@@ -1216,12 +1170,6 @@ int blockstore_init_journal::handle_journal_part(void *buf, uint64_t done_pos, u
                     clean_it->second.version < je->del.version);
                 if (!clean_exists && dirty_exists)
                 {
-                    // Stabilize previous entries to fix inode_space_stats and used_blocks (FIXME - remove it)
-                    obj_ver_id ov = {
-                        .oid = je->del.oid,
-                        .version = dirty_it->first.version,
-                    };
-                    bs->mark_stable(ov, true);
                     // Clean entry doesn't exist. This means that the delete is already flushed.
                     // So we must not flush this object anymore.
                     erase_dirty_object(dirty_it);

src/blockstore/v1/journal.cpp

@@ -15,7 +15,6 @@ blockstore_journal_check_t::blockstore_journal_check_t(blockstore_impl_t *bs)
     first_sector = -1;
     next_in_pos = bs->journal.in_sector_pos;
     right_dir = next_pos >= bs->journal.used_start;
-    give_up = false;
 }
 
 // Check if we can write <required> entries of <size> bytes and <data_after> data bytes after them to the journal
@@ -25,9 +24,7 @@ int blockstore_journal_check_t::check_available(blockstore_op_t *op, int entries
     int required = entries_required;
     while (1)
     {
-        // Must match journal_t::entry_fits()
-        int fits = next_pos == bs->journal.next_free && bs->journal.sector_info[next_sector].written &&
-            (bs->journal.no_same_sector_overwrites || bs->journal.sector_info[next_sector].flush_count > 0)
+        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)
@@ -122,20 +119,13 @@ int blockstore_journal_check_t::check_available(blockstore_op_t *op, int entries
         if (bs->log_level > 5)
         {
             printf(
-                "Ran out of journal space (used_start=%08jx, next_free=%08jx, dirty_start=%08jx, queued %zu)\n",
-                bs->journal.used_start, bs->journal.next_free, bs->journal.dirty_start, bs->flusher->get_queue_size()
+                "Ran out of journal space (used_start=%08jx, next_free=%08jx, dirty_start=%08jx)\n",
+                bs->journal.used_start, bs->journal.next_free, bs->journal.dirty_start
             );
         }
-        if (!bs->flusher->may_advance())
-        {
-            give_up = true;
-        }
-        else
-        {
-            PRIV(op)->wait_for = WAIT_JOURNAL;
-            PRIV(op)->wait_detail = bs->wait_journal_counter;
-            bs->flusher->request_trim();
-        }
+        PRIV(op)->wait_for = WAIT_JOURNAL;
+        bs->flusher->request_trim();
+        PRIV(op)->wait_detail = bs->journal.used_start;
         return 0;
     }
     return 1;
@@ -185,8 +175,6 @@ journal_entry* prefill_single_journal_entry(journal_t & journal, uint16_t type,
 void blockstore_impl_t::prepare_journal_sector_write(int cur_sector, blockstore_op_t *op)
 {
     // Don't submit the same sector twice in the same batch
-    assert(journal.sector_info[cur_sector].dirty);
-    bool submitted_now = false;
     if (!journal.sector_info[cur_sector].submit_id)
     {
         io_uring_sqe *sqe = get_sqe();
@@ -205,24 +193,16 @@ void blockstore_impl_t::prepare_journal_sector_write(int cur_sector, blockstore_
             (size_t)journal.block_size
         };
         data->callback = [this, flush_id = journal.submit_id](ring_data_t *data) { handle_journal_write(data, flush_id); };
-        assert(journal.sector_info[cur_sector].offset > 0);
         assert(journal.sector_info[cur_sector].offset+journal.block_size <= dsk.journal_len);
         io_uring_prep_writev(
             sqe, dsk.journal_fd, &data->iov, 1, journal.offset + journal.sector_info[cur_sector].offset
         );
-        submitted_now = true;
     }
     journal.sector_info[cur_sector].dirty = false;
-    // But always remember that this operation has to wait until this exact journal write is finished.
-    // An operation which appends to a sector another one has already queued in this same batch also
-    // has to wait for everything queued in between - the journal data write of that other operation
-    // in particular. That data lies behind our entry on the disk, and recovery stops at the first
-    // hole in the journal, so completing before it is durable would let an acknowledged operation
-    // disappear together with the unfinished write in front of it
-    journal.flushing_ops.emplace(journal.submit_id, (pending_journaling_t){
+    // But always remember that this operation has to wait until this exact journal write is finished
+    journal.flushing_ops.emplace(journal.sector_info[cur_sector].submit_id, (pending_journaling_t){
         .pending = 1,
-        // Only the operation which queued the sector write itself releases its flush_count
-        .sector = submitted_now ? cur_sector : -1,
+        .sector = cur_sector,
         .op = op,
     });
     auto priv = PRIV(op);
@@ -246,33 +226,38 @@ void blockstore_impl_t::handle_journal_write(ring_data_t *data, uint64_t flush_i
     {
         journal.sector_info[fl_it->second.sector].flush_count--;
     }
-    // Mark every operation waiting for this exact write as no longer pending
-    for (auto it = fl_it; it != journal.flushing_ops.end() && it->first == flush_id; it++)
+    auto is_first = fl_it == journal.flushing_ops.begin();
+    while (fl_it != journal.flushing_ops.end())
     {
-        it->second.pending = 0;
-    }
-    // Do not complete an operation while an earlier journal write is still in flight:
-    // a power outage would then leave a hole in the journal in front of it, and recovery
-    // stops at the first hole, so an already acknowledged operation would be lost.
-    // Only release operations when the write that just finished is the oldest unfinished
-    // one, and then cascade over the following writes that are already finished, stopping
-    // at the first one which isn't.
-    if (fl_it != journal.flushing_ops.begin())
-    {
-        return;
-    }
-    while (fl_it != journal.flushing_ops.end() && !fl_it->second.pending)
-    {
-        auto priv = PRIV(fl_it->second.op);
-        priv->pending_ops--;
-        assert(priv->pending_ops >= 0);
-        if (priv->pending_ops == 0)
+        bool del = false;
+        if (fl_it->first == flush_id)
         {
-            release_journal_sectors(fl_it->second.op);
-            priv->op_state++;
-            ringloop->wakeup();
+            fl_it->second.pending = 0;
+            del = is_first;
+        }
+        else
+        {
+            del = !fl_it->second.pending;
+        }
+        if (del)
+        {
+            // Do not complete this operation if previous writes are unfinished
+            // Otherwise also complete following operations waiting for this one
+            auto priv = PRIV(fl_it->second.op);
+            priv->pending_ops--;
+            assert(priv->pending_ops >= 0);
+            if (priv->pending_ops == 0)
+            {
+                release_journal_sectors(fl_it->second.op);
+                priv->op_state++;
+                ringloop->wakeup();
+            }
+            journal.flushing_ops.erase(fl_it++);
+        }
+        else
+        {
+            fl_it++;
         }
-        journal.flushing_ops.erase(fl_it++);
     }
 }
 

src/blockstore/v1/journal.h

@@ -23,12 +23,12 @@ struct pending_journaling_t
 
 struct journal_t
 {
-    int fd = -1;
+    int fd;
     bool inmemory = false;
     bool flush_journal = false;
     void *buffer = NULL;
 
-    uint64_t block_size = 0;
+    uint64_t block_size;
     uint64_t offset, len;
     // Next free block offset
     uint64_t next_free = 0;
@@ -41,7 +41,7 @@ struct journal_t
     // Current sector(s) used for writing
     void *sector_buf = NULL;
     journal_sector_info_t *sector_info = NULL;
-    uint64_t sector_count = 0;
+    uint64_t sector_count;
     bool no_same_sector_overwrites = false;
     int cur_sector = 0;
     int in_sector_pos = 0;
@@ -59,10 +59,8 @@ struct journal_t
     void dump_diagnostics();
     inline bool entry_fits(int size)
     {
-        // Never append to a sector which is still being written: re-submitting it is probably UB
         return !(block_size - in_sector_pos < size ||
-            sector_info[cur_sector].written &&
-                (no_same_sector_overwrites || sector_info[cur_sector].flush_count > 0));
+            no_same_sector_overwrites && sector_info[cur_sector].written);
     }
 };
 
@@ -72,7 +70,6 @@ struct blockstore_journal_check_t
     uint64_t next_pos, next_sector, next_in_pos;
     int sectors_to_write, first_sector;
     bool right_dir; // writing to the end or the beginning of the ring buffer
-    bool give_up;
 
     blockstore_journal_check_t(blockstore_impl_t *bs);
     int check_available(blockstore_op_t *op, int required, int size, int data_after);

src/blockstore/v1/read.cpp

@@ -110,7 +110,7 @@ int blockstore_impl_t::fulfill_read(blockstore_op_t *read_op,
                     (*dyn_data)++;
                 }
                 // Submit the journal checksum block read
-                if (!read_checksum_block(read_op, 1, fulfilled, item_location - item_start, false))
+                if (!read_checksum_block(read_op, 1, fulfilled, item_location - item_start))
                 {
                     r = 0;
                 }
@@ -173,8 +173,7 @@ int blockstore_impl_t::fill_partial_checksum_blocks(std::vector<copy_buffer_t> &
     if (read_end == read_offset)
         return 0;
     int required = 0;
-    if (read_buf)
-        read_buf -= read_offset;
+    read_buf -= read_offset;
     uint32_t last_block = (read_end-1)/dsk.csum_block_size;
     uint32_t start_block = read_offset/dsk.csum_block_size;
     uint32_t item_start_block = item_start/dsk.csum_block_size;
@@ -190,20 +189,20 @@ int blockstore_impl_t::fill_partial_checksum_blocks(std::vector<copy_buffer_t> &
         };
         rv.insert(rv.begin() + pos, el);
         if (read_buf)
-            memset(read_buf + el.offset, 0, el.len);
+            memset(read_buf + el.offset - read_offset, 0, el.len);
         fulfilled += el.len;
         return 1;
     };
     if (read_offset < item_start)
     {
         // Zero-fill the beginning
-        find_holes(rv, read_offset, item_start < read_end ? item_start : read_end, zero_range);
+        find_holes(rv, read_offset, item_start, zero_range);
         read_offset = item_start;
     }
     if (read_end > item_end)
     {
         // Zero-fill the end
-        find_holes(rv, read_offset < item_end ? item_end : read_offset, read_end, zero_range);
+        find_holes(rv, item_end, read_end, zero_range);
         read_end = item_end;
     }
     while (start_block <= last_block)
@@ -228,12 +227,10 @@ int blockstore_impl_t::fill_partial_checksum_blocks(std::vector<copy_buffer_t> &
             }
             end_block++;
             // OK, mark this range as required
-            uint32_t rv_start = start_block*dsk.csum_block_size < item_start ? item_start : start_block*dsk.csum_block_size;
-            uint32_t rv_end = end_block*dsk.csum_block_size > item_end ? item_end : end_block*dsk.csum_block_size;
             rv.push_back((copy_buffer_t){
                 .copy_flags = COPY_BUF_CSUM_FILL | (from_journal ? COPY_BUF_JOURNALED_BIG : 0),
-                .offset = rv_start,
-                .len = rv_end - rv_start,
+                .offset = start_block*dsk.csum_block_size,
+                .len = (end_block-start_block)*dsk.csum_block_size,
                 // save checksum reference if we're reading clean data from the journal
                 .csum_buf = from_journal
                     ? clean_entry_bitmap + dsk.clean_entry_bitmap_size + (start_block-item_start_block)*(dsk.data_csum_type & 0xFF)
@@ -281,7 +278,7 @@ bool blockstore_impl_t::read_range_fulfilled(std::vector<copy_buffer_t> & rv, ui
                     .offset = bmp_start*dsk.bitmap_granularity,
                     .len = (bmp_pos-bmp_start)*dsk.bitmap_granularity,
                 };
-                rv.insert(rv.begin() + pos + diff, el);
+                rv.insert(rv.begin() + pos, el);
                 if (read_buf)
                     memset(read_buf + el.offset, 0, el.len);
                 fulfilled += el.len;
@@ -302,7 +299,7 @@ bool blockstore_impl_t::read_range_fulfilled(std::vector<copy_buffer_t> & rv, ui
     return all_done;
 }
 
-bool blockstore_impl_t::read_checksum_block(blockstore_op_t *op, int rv_pos, uint64_t &fulfilled, uint64_t clean_loc, bool padded)
+bool blockstore_impl_t::read_checksum_block(blockstore_op_t *op, int rv_pos, uint64_t &fulfilled, uint64_t clean_loc)
 {
     auto & rv = PRIV(op)->read_vec;
     auto *vi = &rv[rv.size()-rv_pos];
@@ -337,22 +334,10 @@ bool blockstore_impl_t::read_checksum_block(blockstore_op_t *op, int rv_pos, uin
         }
         return 0;
     });
-    if (padded && (item_start % dsk.csum_block_size))
-    {
-        n_iov++;
-    }
-    if (padded && (item_end % dsk.csum_block_size))
-    {
-        n_iov++;
-    }
     void *buf = memalign_or_die(MEM_ALIGNMENT, fill_size + n_iov*sizeof(struct iovec));
     iovec *iov = (struct iovec*)((uint8_t*)buf+fill_size);
     n_iov = 0;
     fill_size = 0;
-    if (padded && (item_start % dsk.csum_block_size))
-    {
-        iov[n_iov++] = (struct iovec){ zero_object, item_start % dsk.csum_block_size };
-    }
     find_holes(rv, item_start, item_end, [&](int pos, bool alloc, uint32_t cur_start, uint32_t cur_end)
     {
         int res = 0;
@@ -390,31 +375,18 @@ bool blockstore_impl_t::read_checksum_block(blockstore_op_t *op, int rv_pos, uin
         }
         return res;
     });
-    if (padded && (item_end % dsk.csum_block_size))
-    {
-        iov[n_iov++] = (struct iovec){ zero_object, dsk.csum_block_size - (item_end % dsk.csum_block_size) };
-    }
     vi = &rv[rv.size()-rv_pos];
     // Save buf into read_vec too but in a creepy way
     // FIXME: Shit, something else should be invented %)
     *vi = (copy_buffer_t){
         .copy_flags = vi->copy_flags,
-        .offset = (padded ? vi->offset - vi->offset % dsk.csum_block_size : vi->offset),
+        .offset = vi->offset,
         .len = ((uint64_t)n_iov << 32) | fill_size,
-        .disk_offset = clean_loc + (padded ? vi->offset - vi->offset % dsk.csum_block_size : vi->offset),
+        .disk_offset = clean_loc + item_start,
         .buf = (uint8_t*)buf,
         .csum_buf = vi->csum_buf,
         .dyn_data = vi->dyn_data,
     };
-    if (padded && (item_start % dsk.csum_block_size))
-    {
-        iov++;
-        n_iov--;
-    }
-    if (padded && (item_end % dsk.csum_block_size))
-    {
-        n_iov--;
-    }
     int submit_fd = (vi->copy_flags & COPY_BUF_JOURNAL ? dsk.journal_fd : dsk.data_fd);
     uint64_t submit_offset = (vi->copy_flags & COPY_BUF_JOURNAL ? journal.offset : dsk.data_offset);
     uint32_t d_pos = 0;
@@ -511,7 +483,7 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
                     // Copy from memory or read from journal, possibly checking checksums
                     if (!fulfill_read(read_op, fulfilled, dirty.offset, dirty.offset + dirty.len,
                         dirty.state, dirty_it->first.version, dirty.location, dirty.journal_sector+1,
-                        journal.inmemory || !dsk.csum_block_size ? NULL : bmp_ptr+dsk.clean_entry_bitmap_size, dyn_data))
+                        journal.inmemory ? NULL : bmp_ptr+dsk.clean_entry_bitmap_size, dyn_data))
                     {
                         goto undo_read;
                     }
@@ -549,7 +521,6 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
         // May happen if there are entries in dirty_db but all of them are !version_ok
         read_op->version = 0;
         read_op->retval = -ENOENT;
-        release_read_vec(read_op);
         FINISH_OP(read_op);
         return 2;
     }
@@ -563,9 +534,6 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
             // region is not allocated - return zeroes
             memset(read_op->buf, 0, read_op->len);
         }
-        // Nothing was submitted, so no completion handler will run - release the
-        // references the read vector holds on entry bitmaps and checksums
-        release_read_vec(read_op);
         read_op->retval = read_op->len;
         FINISH_OP(read_op);
         return 2;
@@ -580,29 +548,25 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
         }
     }
     read_op->retval = 0;
-    reads_in_flight.insert(read_op);
     return 2;
 undo_read:
     // need to wait. undo added requests, don't dequeue op
     release_clean(read_op);
-    release_read_vec(read_op);
-    rv.clear();
-    return 0;
-}
-
-// Free the buffers and drop the references that a read vector holds on the bitmaps and
-// checksums of dirty entries. Safe to call more than once.
-void blockstore_impl_t::release_read_vec(blockstore_op_t *read_op)
-{
-    for (auto & vec: PRIV(read_op)->read_vec)
+    for (auto & vec: rv)
     {
         if ((vec.copy_flags & COPY_BUF_CSUM_FILL) && vec.buf)
         {
             free(vec.buf);
             vec.buf = NULL;
         }
-        free_dirty_dyn_data(vec);
+        if (vec.dyn_data && --(*vec.dyn_data) == 0) // refcount
+        {
+            free(vec.dyn_data);
+            vec.dyn_data = NULL;
+        }
     }
+    rv.clear();
+    return 0;
 }
 
 void blockstore_impl_t::release_clean(blockstore_op_t *op)
@@ -710,7 +674,7 @@ bool blockstore_impl_t::fulfill_clean_read(blockstore_op_t *read_op, uint64_t &
         }
         for (int i = req; i > 0; i--)
         {
-            if (!read_checksum_block(read_op, i, fulfilled, clean_loc, true))
+            if (!read_checksum_block(read_op, i, fulfilled, clean_loc))
             {
                 return false;
             }
@@ -947,35 +911,27 @@ void blockstore_impl_t::handle_read_event(ring_data_t *data, blockstore_op_t *op
             {
                 for (int i = 0; i < rv.size(); i++)
                 {
-                    // Every entry holds a reference to the bitmap/checksums of its dirty
-                    // entry, so it has to be released even when the entry itself is skipped
-                    bool skip = false;
                     if (rv[i].copy_flags & COPY_BUF_META_BLOCK)
                     {
                         // Metadata read. Skip
                         assert(!meta_block);
                         meta_block = rv[i].buf;
                         rv[i].buf = NULL;
-                        skip = true;
+                        continue;
                     }
-                    else if (rv[i].copy_flags & COPY_BUF_ZERO)
+                    if (rv[i].copy_flags & COPY_BUF_ZERO)
                     {
                         // Zero read
-                        skip = true;
+                        continue;
                     }
-                    else if (rv[i].copy_flags & COPY_BUF_COALESCED)
+                    if (rv[i].copy_flags & COPY_BUF_COALESCED)
                     {
                         // Sub-block shared with another read. Skip
-                        skip = true;
+                        continue;
                     }
-                    else if ((rv[i].copy_flags & COPY_BUF_JOURNAL) && journal.inmemory)
+                    if ((rv[i].copy_flags & COPY_BUF_JOURNAL) && journal.inmemory)
                     {
                         // Do not check journal checksums in-memory
-                        skip = true;
-                    }
-                    if (skip)
-                    {
-                        free_dirty_dyn_data(rv[i]);
                         continue;
                     }
                     iovec single_iov = {};
@@ -1043,7 +999,11 @@ void blockstore_impl_t::handle_read_event(ring_data_t *data, blockstore_op_t *op
                     }
                     free(rv[i].buf);
                     rv[i].buf = NULL;
-                    free_dirty_dyn_data(rv[i]);
+                    if (rv[i].dyn_data && --(*rv[i].dyn_data) == 0) // refcount
+                    {
+                        free(rv[i].dyn_data);
+                        rv[i].dyn_data = NULL;
+                    }
                 }
             }
             else
@@ -1052,11 +1012,10 @@ void blockstore_impl_t::handle_read_event(ring_data_t *data, blockstore_op_t *op
                 {
                     if (vec.copy_flags & COPY_BUF_META_BLOCK)
                     {
-                        // Metadata read. Skip, but still release the entry reference
+                        // Metadata read. Skip
                         assert(!meta_block);
                         meta_block = vec.buf;
                         vec.buf = NULL;
-                        free_dirty_dyn_data(vec);
                         continue;
                     }
                     if (vec.csum_buf)
@@ -1078,7 +1037,11 @@ void blockstore_impl_t::handle_read_event(ring_data_t *data, blockstore_op_t *op
                             }
                         }
                     }
-                    free_dirty_dyn_data(vec);
+                    if (vec.dyn_data && --(*vec.dyn_data) == 0) // refcount
+                    {
+                        free(vec.dyn_data);
+                        vec.dyn_data = NULL;
+                    }
                 }
             }
             if (meta_block)
@@ -1107,7 +1070,6 @@ void blockstore_impl_t::handle_read_event(ring_data_t *data, blockstore_op_t *op
         }
         if (op->retval == 0)
             op->retval = op->len;
-        reads_in_flight.erase(op);
         FINISH_OP(op);
     }
 }
@@ -1127,14 +1089,6 @@ int blockstore_impl_t::read_bitmap(object_id oid, uint64_t target_version, void
             // Condition has to be the same as in dequeue_read()
             if (!IS_IN_FLIGHT(dirty_it->second.state) && target_version >= dirty_it->first.version)
             {
-                if (IS_DELETE(dirty_it->second.state))
-                {
-                    if (result_version)
-                        *result_version = 0;
-                    if (bitmap)
-                        memset(bitmap, 0, dsk.clean_entry_bitmap_size);
-                    return -ENOENT;
-                }
                 if (result_version)
                     *result_version = dirty_it->first.version;
                 if (bitmap)

src/blockstore/v1/rollback.cpp

@@ -28,7 +28,7 @@ int blockstore_impl_t::dequeue_rollback(blockstore_op_t *op)
         else
         {
             dirty_it--;
-            if (dirty_it->first.oid != ov.oid || dirty_it->first.version <= ov.version)
+            if (dirty_it->first.oid != ov.oid || dirty_it->first.version < ov.version)
             {
                 // Already rolled back, skip this object version
                 return STAB_SPLIT_DONE;
@@ -40,12 +40,8 @@ int blockstore_impl_t::dequeue_rollback(blockstore_op_t *op)
                     // Object write is still in progress. Wait until the write request completes
                     return STAB_SPLIT_WAIT;
                 }
-                else if (IS_STABLE(dirty_it->second.state))
-                {
-                    // Can't jump back over a stable version
-                    return -EBUSY;
-                }
-                else if (!IS_SYNCED(dirty_it->second.state))
+                else if (!IS_SYNCED(dirty_it->second.state) ||
+                    IS_STABLE(dirty_it->second.state))
                 {
                     // Sync the object
                     return STAB_SPLIT_SYNC;
@@ -67,12 +63,6 @@ int blockstore_impl_t::dequeue_rollback(blockstore_op_t *op)
     blockstore_journal_check_t space_check(this);
     if (!space_check.check_available(op, op->len, sizeof(journal_entry_rollback), 0))
     {
-        if (space_check.give_up)
-        {
-            op->retval = -EAGAIN;
-            FINISH_OP(op);
-            return 2;
-        }
         return 0;
     }
     // There is sufficient space. Check SQEs
@@ -172,10 +162,7 @@ void blockstore_impl_t::mark_rolled_back(const obj_ver_id & ov)
             if (unstab_it != unstable_writes.end())
             {
                 if (max_unstable == 0)
-                {
                     unstable_writes.erase(unstab_it);
-                    wakeup_wait_journal();
-                }
                 else
                     unstab_it->second = max_unstable;
             }
@@ -205,8 +192,6 @@ void blockstore_impl_t::erase_dirty(blockstore_dirty_db_t::iterator dirty_start,
             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;
-                // Entering WAIT_DEL -> request_trim, exiting -> release_trim
-                flusher->release_trim();
                 if (IS_BIG_WRITE(dirty_it->second.state))
                 {
                     next_state = BS_ST_WAIT_BIG;
@@ -233,28 +218,22 @@ void blockstore_impl_t::erase_dirty(blockstore_dirty_db_t::iterator dirty_start,
 #endif
             data_alloc->set(dirty_it->second.location / dsk.data_block_size, false);
         }
-        // A big write is only journaled when it gets synced, so an entry erased before that
-        // holds no journal sector reference at all
-        if (dirty_it->second.journal_sector != 0)
-        {
-            auto used_it = journal.used_sectors.find(dirty_it->second.journal_sector);
-            uint64_t used = used_it == journal.used_sectors.end() ? UINT64_MAX : --used_it->second;
+        auto used = --journal.used_sectors.at(dirty_it->second.journal_sector);
 #ifdef BLOCKSTORE_DEBUG
-            printf(
-                "remove usage of journal offset %08jx by %jx:%jx v%ju (%ju refs)\n", dirty_it->second.journal_sector,
-                dirty_it->first.oid.inode, dirty_it->first.oid.stripe, dirty_it->first.version, used
-            );
+        printf(
+            "remove usage of journal offset %08jx by %jx:%jx v%ju (%ju refs)\n", dirty_it->second.journal_sector,
+            dirty_it->first.oid.inode, dirty_it->first.oid.stripe, dirty_it->first.version, used
+        );
 #endif
-            if (used == 0)
+        if (used == 0)
+        {
+            journal.used_sectors.erase(dirty_it->second.journal_sector);
+            if (dirty_it->second.journal_sector == journal.sector_info[journal.cur_sector].offset)
             {
-                journal.used_sectors.erase(dirty_it->second.journal_sector);
-                if (dirty_it->second.journal_sector == journal.sector_info[journal.cur_sector].offset)
-                {
-                    // Mark current sector as "full" to select the new one
-                    journal.in_sector_pos = dsk.journal_block_size;
-                }
-                flusher->mark_trim_possible();
+                // Mark current sector as "full" to select the new one
+                journal.in_sector_pos = dsk.journal_block_size;
             }
+            flusher->mark_trim_possible();
         }
         free_dirty_dyn_data(dirty_it->second);
         if (dirty_it == dirty_start)
@@ -281,17 +260,4 @@ void blockstore_impl_t::free_dirty_dyn_data(dirty_entry & e)
     }
 }
 
-void blockstore_impl_t::free_dirty_dyn_data(copy_buffer_t & e)
-{
-    if (e.dyn_data)
-    {
-        assert(alloc_dyn_data);
-        if (--*e.dyn_data == 0) // refcount
-        {
-            free(e.dyn_data);
-        }
-        e.dyn_data = NULL;
-    }
-}
-
 } // namespace v1

src/blockstore/v1/stable.cpp

@@ -73,11 +73,7 @@ static void append_version(ver_vector_t & vec, obj_ver_id ov)
     vec.items[vec.size++] = ov;
 }
 
-// Move the unsynced versions of <ov.oid> which the operation has to sync first from <check>
-// into <to>. Stabilization needs the versions up to <ov.version> synced, and a rollback needs
-// the versions above it - those are the ones it discards, and a rollback can only be journaled
-// after they reach the journal themselves
-static bool check_unsynced(std::vector<obj_ver_id> & check, obj_ver_id ov, std::vector<obj_ver_id> & to, int *count, bool above)
+static bool check_unsynced(std::vector<obj_ver_id> & check, obj_ver_id ov, std::vector<obj_ver_id> & to, int *count)
 {
     bool found = false;
     int j = 0, k = 0;
@@ -85,7 +81,7 @@ static bool check_unsynced(std::vector<obj_ver_id> & check, obj_ver_id ov, std::
     {
         if (check[j] == ov)
             found = true;
-        if (check[j].oid == ov.oid && (above ? check[j].version > ov.version : check[j].version <= ov.version))
+        if (check[j].oid == ov.oid && check[j].version <= ov.version)
         {
             to.push_back(check[j++]);
             if (count)
@@ -191,9 +187,8 @@ int blockstore_impl_t::split_stab_op(blockstore_op_t *op, std::function<int(obj_
                 PRIV(op)->sync_small_writes.clear();
                 add_sync = true;
             }
-            bool above = op->opcode == BS_OP_ROLLBACK;
-            check_unsynced(unsynced_small_writes, *v, PRIV(op)->sync_small_writes, NULL, above);
-            check_unsynced(unsynced_big_writes, *v, PRIV(op)->sync_big_writes, &unsynced_big_write_count, above);
+            check_unsynced(unsynced_small_writes, *v, PRIV(op)->sync_small_writes, NULL);
+            check_unsynced(unsynced_big_writes, *v, PRIV(op)->sync_big_writes, &unsynced_big_write_count);
         }
         else /* if (action == STAB_SPLIT_TODO) */
         {
@@ -215,8 +210,6 @@ int blockstore_impl_t::split_stab_op(blockstore_op_t *op, std::function<int(obj_
     if (!todo && !bad_vers.size)
     {
         // Already stable
-        free(good_vers.items);
-        free(bad_vers.items);
         op->retval = 0;
         FINISH_OP(op);
         return 2;
@@ -224,10 +217,7 @@ int blockstore_impl_t::split_stab_op(blockstore_op_t *op, std::function<int(obj_
     op->retval = 0;
     if (!todo && !add_sync)
     {
-        // Only wait for inflight writes or current in-progress syncs.
-        // The operation is retried from scratch, so the version vectors go with it
-        free(good_vers.items);
-        free(bad_vers.items);
+        // Only wait for inflight writes or current in-progress syncs
         return 0;
     }
     blockstore_op_t *sync_op = NULL, *split_stab_op = NULL;
@@ -372,12 +362,6 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
     blockstore_journal_check_t space_check(this);
     if (!space_check.check_available(op, op->len, sizeof(journal_entry_stable), 0))
     {
-        if (space_check.give_up)
-        {
-            op->retval = -EAGAIN;
-            FINISH_OP(op);
-            return 2;
-        }
         return 0;
     }
     // There is sufficient space. Check SQEs
@@ -461,9 +445,9 @@ void blockstore_impl_t::mark_stable(obj_ver_id v, bool forget_dirty)
                 {
                     break;
                 }
-                if (!IS_SYNCED(back_it->second.state))
+                if (!IS_STABLE(back_it->second.state))
                 {
-                    // There are preceding unsynced versions, can't flush <v>
+                    // There are preceding unstable versions, can't flush <v>
                     return;
                 }
             }
@@ -479,7 +463,6 @@ void blockstore_impl_t::mark_stable(obj_ver_id v, bool forget_dirty)
                 v.version = dirty_it->first.version;
             }
         }
-        auto erase_end = dirty_db.end();
         while (1)
         {
             bool was_stable = IS_STABLE(dirty_it->second.state);
@@ -487,7 +470,6 @@ void blockstore_impl_t::mark_stable(obj_ver_id v, bool forget_dirty)
             {
                 dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | BS_ST_STABLE;
                 // Allocations and deletions are counted when they're stabilized
-                // FIXME: Count-on-stabilize logic is actually really strange, reconsider it
                 if (IS_BIG_WRITE(dirty_it->second.state))
                 {
                     int exists = -1;
@@ -541,45 +523,43 @@ void blockstore_impl_t::mark_stable(obj_ver_id v, bool forget_dirty)
                 exit(1);
             }
             if (forget_dirty && (IS_BIG_WRITE(dirty_it->second.state) ||
-                IS_DELETE(dirty_it->second.state)) && erase_end == dirty_db.end())
+                IS_DELETE(dirty_it->second.state)))
             {
                 // Big write overrides all previous dirty entries
-                erase_end = dirty_it;
+                auto erase_end = dirty_it;
+                while (dirty_it != dirty_db.begin())
+                {
+                    dirty_it--;
+                    if (dirty_it->first.oid != v.oid)
+                    {
+                        dirty_it++;
+                        break;
+                    }
+                }
+                auto & clean_db = clean_db_shard(v.oid);
+                auto clean_it = clean_db.find(v.oid);
+                uint64_t clean_loc = clean_it != clean_db.end()
+                    ? clean_it->second.location : UINT64_MAX;
+                erase_dirty(dirty_it, erase_end, clean_loc);
+                break;
             }
-            // Stopping at an already stable entry is only correct when we aren't erasing anything:
-            // everything below a big write or a deletion is dead and has to be erased even if it
-            // was stabilized earlier. Entries are stabilized one by one during the journal replay,
-            // so a small write in between used to hide an older big write from this walk forever,
-            // leaving its data block claimed by an entry which isn't current any more.
-            // <forget_dirty> is only set by the journal replay, so the walk stays short at runtime
-            if (!forget_dirty && was_stable || dirty_it == dirty_db.begin())
+            if (was_stable || dirty_it == dirty_db.begin())
             {
                 break;
             }
             dirty_it--;
             if (dirty_it->first.oid != v.oid)
             {
-                dirty_it++;
                 break;
             }
         }
         flusher->enqueue_flush(v);
-        if (erase_end != dirty_db.end())
-        {
-            // Erase only after marking previous entries stable to not skip used_blocks++
-            auto & clean_db = clean_db_shard(v.oid);
-            auto clean_it = clean_db.find(v.oid);
-            uint64_t clean_loc = clean_it != clean_db.end()
-                ? clean_it->second.location : UINT64_MAX;
-            erase_dirty(dirty_it, erase_end, clean_loc);
-        }
     }
     auto unstab_it = unstable_writes.find(v.oid);
     if (unstab_it != unstable_writes.end() &&
         unstab_it->second <= v.version)
     {
         unstable_writes.erase(unstab_it);
-        wakeup_wait_journal();
     }
 }
 

src/blockstore/v1/sync.cpp

@@ -79,7 +79,6 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op)
         // 2nd step: Data device is synced, prepare & write journal entries
         // Check space in the journal and journal memory buffers
         blockstore_journal_check_t space_check(this);
-        bool space_ok = true;
         if (dsk.csum_block_size)
         {
             // More complex check because all journal entries have different lengths
@@ -91,34 +90,13 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op)
                 uint64_t dyn_size = dsk.dirty_dyn_size(dirty_entry.offset, dirty_entry.len);
                 if (!space_check.check_available(op, 1, sizeof(journal_entry_big_write) + dyn_size, 0))
                 {
-                    space_ok = false;
-                    break;
+                    return 0;
                 }
             }
         }
         else if (!space_check.check_available(op, PRIV(op)->sync_big_writes.size(),
             sizeof(journal_entry_big_write) + dsk.clean_entry_bitmap_size, 0))
         {
-            space_ok = false;
-        }
-        if (!space_ok)
-        {
-            if (space_check.give_up)
-            {
-                // Return everything back
-                unsynced_big_write_count += PRIV(op)->sync_big_writes.size();
-                PRIV(op)->sync_big_writes.insert(PRIV(op)->sync_big_writes.end(),
-                    unsynced_big_writes.begin(), unsynced_big_writes.end());
-                PRIV(op)->sync_big_writes.swap(unsynced_big_writes);
-                PRIV(op)->sync_big_writes.clear();
-                PRIV(op)->sync_small_writes.insert(PRIV(op)->sync_small_writes.end(),
-                    unsynced_small_writes.begin(), unsynced_small_writes.end());
-                PRIV(op)->sync_small_writes.swap(unsynced_small_writes);
-                PRIV(op)->sync_small_writes.clear();
-                op->retval = -EAGAIN;
-                FINISH_OP(op);
-                return 2;
-            }
             return 0;
         }
         // Check SQEs. Don't bother about merging, submit each journal sector as a separate request
@@ -198,54 +176,60 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op)
 void blockstore_impl_t::ack_sync(blockstore_op_t *op)
 {
     // Handle states
-    auto mark_sync = [&](obj_ver_id & ov)
+    for (auto it = PRIV(op)->sync_big_writes.begin(); it != PRIV(op)->sync_big_writes.end(); it++)
     {
-        auto dirty_it = dirty_db.find(ov);
-        bool is_big = ((dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_BIG_WRITE);
+#ifdef BLOCKSTORE_DEBUG
+        printf("Ack sync big %jx:%jx v%ju\n", it->oid.inode, it->oid.stripe, it->version);
+#endif
+        auto & unstab = unstable_writes[it->oid];
+        unstab = unstab < it->version ? it->version : unstab;
+        auto dirty_it = dirty_db.find(*it);
         dirty_it->second.state = ((dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | BS_ST_SYNCED);
-        uint64_t stable_ver = 0;
-        if (IS_INSTANT(dirty_it->second.state))
+        if (dirty_it->second.state & BS_ST_INSTANT)
         {
-            stable_ver = ov.version;
+            mark_stable(dirty_it->first);
         }
         else
         {
             unstable_unsynced--;
             assert(unstable_unsynced >= 0);
-            wakeup_wait_journal();
         }
         dirty_it++;
-        while (dirty_it != dirty_db.end() && dirty_it->first.oid == ov.oid)
+        while (dirty_it != dirty_db.end() && dirty_it->first.oid == it->oid)
         {
-            if (is_big && (dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_WAIT_BIG)
+            if ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_WAIT_BIG)
             {
                 dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | BS_ST_IN_FLIGHT;
             }
-            else if ((dirty_it->second.state & BS_ST_WORKFLOW_MASK) == BS_ST_SYNCED &&
-                IS_INSTANT(dirty_it->second.state))
-            {
-                stable_ver = dirty_it->first.version;
-            }
             dirty_it++;
         }
-        if (stable_ver != 0)
-        {
-            mark_stable((obj_ver_id){ .oid = ov.oid, .version = stable_ver });
-        }
-    };
-    for (auto & ov: PRIV(op)->sync_big_writes)
-    {
-#ifdef BLOCKSTORE_DEBUG
-        printf("Ack sync big %jx:%jx v%ju\n", ov.oid.inode, ov.oid.stripe, ov.version);
-#endif
-        mark_sync(ov);
     }
-    for (auto & ov: PRIV(op)->sync_small_writes)
+    for (auto it = PRIV(op)->sync_small_writes.begin(); it != PRIV(op)->sync_small_writes.end(); it++)
     {
 #ifdef BLOCKSTORE_DEBUG
-        printf("Ack sync small %jx:%jx v%ju\n", ov.oid.inode, ov.oid.stripe, ov.version);
+        printf("Ack sync small %jx:%jx v%ju\n", it->oid.inode, it->oid.stripe, it->version);
 #endif
-        mark_sync(ov);
+        auto & unstab = unstable_writes[it->oid];
+        unstab = unstab < it->version ? it->version : unstab;
+        if (dirty_db[*it].state == (BS_ST_DELETE | BS_ST_WRITTEN))
+        {
+            dirty_db[*it].state = (BS_ST_DELETE | BS_ST_SYNCED);
+            // Deletions are treated as immediately stable
+            mark_stable(*it);
+        }
+        else /* (BS_ST_INSTANT?) | BS_ST_SMALL_WRITE | BS_ST_WRITTEN */
+        {
+            dirty_db[*it].state = (dirty_db[*it].state & ~BS_ST_WORKFLOW_MASK) | BS_ST_SYNCED;
+            if (dirty_db[*it].state & BS_ST_INSTANT)
+            {
+                mark_stable(*it);
+            }
+            else
+            {
+                unstable_unsynced--;
+                assert(unstable_unsynced >= 0);
+            }
+        }
     }
     op->retval = 0;
     FINISH_OP(op);

src/blockstore/v1/write.cpp

@@ -6,121 +6,25 @@
 
 namespace v1 {
 
-// A deletion supersedes everything written before it and is stable as soon as it's written,
-// so versions preceding it that aren't in the journal yet are free to forget. And they have to
-// be forgotten: a big write is only journaled when it gets synced, so leaving one behind would
-// put its journal entry after the deletion and recovery would replay them in the wrong
-// order, resurrecting the object. Returns false if a preceding write is still running -
-// its completion refers to the entry, so we have to let it finish first.
-//
-// Entries which are already synced must NOT be forgotten. Their journal entries are on the disk
-// and can't be taken back, while the deletion itself only reaches the journal on the next sync.
-// Forgetting them drops their journal references, and the journal may then be trimmed past the
-// object's big write while its small writes still sit in later sectors kept alive by other
-// objects - recovery would then replay those small writes with no base under them.
-// An entry which isn't in the unsynced lists any more has been taken over by a sync which is
-// still in flight: continue_sync() swaps those lists over to itself, but the entries only get
-// the SYNCED state later, in ack_sync()
-static bool is_still_unsynced(const std::vector<obj_ver_id> & unsynced, obj_ver_id ov)
-{
-    for (auto & u: unsynced)
-        if (u == ov)
-            return true;
-    return false;
-}
-
-bool blockstore_impl_t::forget_unstable_before_delete(blockstore_op_t *op, blockstore_dirty_db_t::iterator dirty_it)
-{
-    if (immediate_commit == IMMEDIATE_ALL)
-    {
-        // Everything is journaled as soon as it's written here, so nothing can end up
-        // behind the deletion in the journal and there is nothing to forget
-        return true;
-    }
-    auto first_it = dirty_it;
-    while (first_it != dirty_db.begin())
-    {
-        auto prev_it = first_it;
-        prev_it--;
-        if (prev_it->first.oid != op->oid || IS_SYNCED(prev_it->second.state))
-            break;
-        if (IS_IN_FLIGHT(prev_it->second.state))
-            return false;
-        if (!is_still_unsynced(IS_BIG_WRITE(prev_it->second.state) ? unsynced_big_writes : unsynced_small_writes, prev_it->first))
-        {
-            // A sync in flight already owns it, so we can't erase it here: that would leave
-            // the sync with a dangling dirty_db entry in ack_sync() and would decrement
-            // unstable_unsynced a second time.
-            // A small write is journaled as soon as it's dequeued, so it is already in front
-            // of our deletion. A big write is not: its journal entry is only written when the
-            // sync reaches its journaling phase, which may well be after our deletion is
-            // journaled. Journal replay would then take it for a new write over the deletion
-            // and resurrect the object, so wait for that sync to finish instead
-            if (IS_BIG_WRITE(prev_it->second.state))
-                return false;
-            break;
-        }
-        first_it = prev_it;
-    }
-    if (first_it == dirty_it)
-        return true;
-    // Their durability becomes ours: they're acknowledged, but from now on the only thing
-    // on the disk which accounts for them is our deletion. A sync started before we're
-    // journaled and acked would otherwise find nothing to do and report success
-    unsynced_forgotten++;
-    PRIV(op)->forgot_unsynced = true;
-    for (auto it = first_it; it != dirty_it; it++)
-    {
-        // Drop it from the unsynced lists so that a later sync doesn't journal it
-        bool is_big = IS_BIG_WRITE(it->second.state);
-        auto & unsynced = is_big ? unsynced_big_writes : unsynced_small_writes;
-        for (size_t i = 0; i < unsynced.size(); i++)
-        {
-            if (unsynced[i].oid == it->first.oid && unsynced[i].version == it->first.version)
-            {
-                unsynced.erase(unsynced.begin()+i, unsynced.begin()+i+1);
-                if (is_big && immediate_commit != IMMEDIATE_ALL)
-                    unsynced_big_write_count--;
-                break;
-            }
-        }
-        if (!(it->second.state & BS_ST_INSTANT) && !IS_SYNCED(it->second.state))
-        {
-            unstable_unsynced--;
-            assert(unstable_unsynced >= 0);
-        }
-    }
-    auto & clean_db = clean_db_shard(op->oid);
-    auto clean_it = clean_db.find(op->oid);
-    uint64_t clean_loc = clean_it != clean_db.end() ? clean_it->second.location : UINT64_MAX;
-    erase_dirty(first_it, dirty_it, clean_loc);
-    wakeup_wait_journal();
-    return true;
-}
-
 bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
 {
     // Check or assign version number
     bool found = false, deleted = false, unsynced = false, is_del = (op->opcode == BS_OP_DELETE);
-    bool wait_big = false, wait_del = false, in_open_sector = false;
+    bool wait_big = false, wait_del = false;
     void *dyn = NULL;
-    uint8_t *dyn_ptr = NULL;
     if (is_del)
     {
         op->len = 0;
     }
-    else
+    size_t dyn_size = dsk.dirty_dyn_size(op->offset, op->len);
+    if (!is_del && alloc_dyn_data)
     {
-        if (alloc_dyn_data)
-        {
-            // FIXME: Working with `dyn_data` has to be refactored somehow but I first have to decide how :)
-            // +sizeof(int) = refcount
-            size_t dyn_size = dsk.dirty_dyn_size(op->offset, op->len);
-            dyn = calloc_or_die(1, dyn_size+sizeof(int));
-            *((int*)dyn) = 1;
-        }
-        dyn_ptr = (alloc_dyn_data ? (uint8_t*)dyn+sizeof(int) : (uint8_t*)&dyn);
+        // FIXME: Working with `dyn_data` has to be refactored somehow but I first have to decide how :)
+        // +sizeof(int) = refcount
+        dyn = calloc_or_die(1, dyn_size+sizeof(int));
+        *((int*)dyn) = 1;
     }
+    uint8_t *dyn_ptr = (alloc_dyn_data ? (uint8_t*)dyn+sizeof(int) : (uint8_t*)&dyn);
     uint64_t version = 1;
     if (dirty_db.size() > 0)
     {
@@ -131,15 +35,6 @@ bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
         dirty_it--; // segfaults when dirty_db is empty
         if (dirty_it != dirty_db.end() && dirty_it->first.oid == op->oid)
         {
-            if (op->opcode == BS_OP_WRITE_STABLE && !IS_INSTANT(dirty_it->second.state))
-            {
-                op->retval = -EINVAL;
-                if (!is_del && alloc_dyn_data)
-                {
-                    free(dyn);
-                }
-                return false;
-            }
             found = true;
             version = dirty_it->first.version + 1;
             deleted = IS_DELETE(dirty_it->second.state);
@@ -148,7 +43,6 @@ bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
             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);
-            in_open_sector = dirty_it->second.journal_sector == journal.sector_info[journal.cur_sector].offset;
             if (!is_del && !deleted)
             {
                 void *dyn_from = alloc_dyn_data
@@ -199,26 +93,6 @@ bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
             wait_del = true;
             PRIV(op)->real_version = op->version;
             op->version = version;
-            if (!unsynced && in_open_sector &&
-                journal.sector_info[journal.cur_sector].flush_count == 0)
-            {
-                // In exotic configurations without no_same_sector_overwrites, the deletion may reside
-                // in the current sector. In that case nothing advances dirty_start and the deletion
-                // becomes unflushable... Example is test_blockstore_chaos --seed 10901878
-                if (!journal.sector_info[journal.cur_sector].dirty)
-                {
-                    // If it's unmodified we can just move to the next sector
-                    // Probably unreachable currently due to "Forcibly move to the next sector" logic below
-                    journal.in_sector_pos = journal.block_size;
-                    mark_not_dirty(journal.cur_sector);
-                }
-                else
-                {
-                    // If it's modified we may add a SYNC so it writes the sector out
-                    assert(immediate_commit != IMMEDIATE_ALL);
-                    unsynced = true;
-                }
-            }
             if (unsynced)
             {
                 // Issue an additional sync so the delete reaches the journal
@@ -228,20 +102,20 @@ bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
                 sync_op->version = op->version;
                 sync_op->callback = [this](blockstore_op_t *sync_op)
                 {
-                    unshift_stable_flush((obj_ver_id){
+                    flusher->unshift_flush((obj_ver_id){
                         .oid = sync_op->oid,
                         .version = sync_op->version-1,
-                    });
+                    }, true);
                     delete sync_op;
                 };
-                enqueue_own_op(sync_op);
+                enqueue_op(sync_op);
             }
             else
             {
-                unshift_stable_flush((obj_ver_id){
+                flusher->unshift_flush((obj_ver_id){
                     .oid = op->oid,
                     .version = version-1,
-                });
+                }, true);
             }
         }
         else
@@ -269,7 +143,7 @@ bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
         {
             delete sync_op;
         };
-        enqueue_own_op(sync_op);
+        enqueue_op(sync_op);
     }
     else if (!imm)
         unsynced_queued_ops++;
@@ -300,11 +174,6 @@ bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
         if (op->bitmap)
             memcpy(dyn_ptr, op->bitmap, dsk.clean_entry_bitmap_size);
     }
-    if (wait_del)
-    {
-        // Entering WAIT_DEL -> request_trim, exiting -> release_trim
-        flusher->request_trim();
-    }
     // Calculate checksums
     // FIXME: Allow to receive checksums from outside?
     if (!is_del && dsk.data_csum_type && op->len > 0)
@@ -345,27 +214,8 @@ bool blockstore_impl_t::enqueue_write(blockstore_op_t *op)
     return true;
 }
 
-// Ask the flusher to flush <ov> right now, but only if it's already stable.
-// A sync we submit ourselves may complete without syncing anything, because a sync submitted
-// before it has already taken those writes over and is still in flight. In that case the entry
-// becomes stable - and is queued for flushing by mark_stable() - when that sync completes,
-// and forcing a flush of a still unstable entry would abort the flusher
-void blockstore_impl_t::unshift_stable_flush(obj_ver_id ov)
-{
-    auto dirty_it = dirty_db.find(ov);
-    if (dirty_it != dirty_db.end() && IS_STABLE(dirty_it->second.state))
-    {
-        flusher->unshift_flush(ov, true);
-    }
-}
-
 void blockstore_impl_t::cancel_all_writes(blockstore_op_t *op, blockstore_dirty_db_t::iterator dirty_it, int retval)
 {
-    if (PRIV(op)->forgot_unsynced)
-    {
-        PRIV(op)->forgot_unsynced = false;
-        unsynced_forgotten--;
-    }
     while (dirty_it != dirty_db.end() && dirty_it->first.oid == op->oid)
     {
         free_dirty_dyn_data(dirty_it->second);
@@ -458,11 +308,6 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
             sizeof(journal_entry_big_write) + dsk.clean_dyn_size,
             (unstable_writes.size()+unstable_unsynced+((dirty_it->second.state & BS_ST_INSTANT) ? 0 : 1))*journal.block_size))
         {
-            if (space_check.give_up)
-            {
-                cancel_all_writes(op, dirty_it, -EAGAIN);
-                return 2;
-            }
             return 0;
         }
         // Big (redirect) write
@@ -555,11 +400,6 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
                 sizeof(journal_entry_small_write) + dyn_size,
                 op->len + (unstable_writes.size()+unstable_unsynced+((dirty_it->second.state & BS_ST_INSTANT) ? 0 : 1))*journal.block_size))
         {
-            if (space_check.give_up)
-            {
-                cancel_all_writes(op, dirty_it, -EAGAIN);
-                return 2;
-            }
             return 0;
         }
         // There is sufficient space. Check SQE(s)
@@ -572,8 +412,7 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
         write_iodepth++;
         // Got SQEs. Prepare previous journal sector write if required
         if (immediate_commit == IMMEDIATE_NONE &&
-            !journal.entry_fits(sizeof(journal_entry_small_write) + dyn_size) &&
-            journal.sector_info[journal.cur_sector].dirty)
+            !journal.entry_fits(sizeof(journal_entry_small_write) + dyn_size))
         {
             prepare_journal_sector_write(journal.cur_sector, op);
         }
@@ -633,13 +472,7 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
             ? (uint8_t*)dirty_it->second.dyn_data+sizeof(int) : (uint8_t*)&dirty_it->second.dyn_data), dyn_size);
         je->crc32 = je_crc32((journal_entry*)je);
         journal.crc32_last = je->crc32;
-        // With immediate_commit=none we normally leave the sector to be written out by a later
-        // sync. But if it is already queued for writing in this same batch, our entry goes to
-        // the disk with that write anyway - so join it instead of leaving the sector dirty,
-        // otherwise the next operation to write the sector out would queue a second write of
-        // it while the first one is still in flight. It costs no extra I/O: the write is
-        // already prepared, so prepare_journal_sector_write() only subscribes us to it
-        if (immediate_commit != IMMEDIATE_NONE || journal.sector_info[journal.cur_sector].submit_id)
+        if (immediate_commit != IMMEDIATE_NONE)
         {
             prepare_journal_sector_write(journal.cur_sector, op);
         }
@@ -662,7 +495,6 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
                 .op = op,
             });
             data2->callback = [this, flush_id = journal.submit_id](ring_data_t *data) { handle_journal_write(data, flush_id); };
-            assert(journal.next_free > 0);
             assert(journal.next_free+op->len <= dsk.journal_len);
             io_uring_prep_writev(sqe2, dsk.journal_fd, &data2->iov, 1, journal.offset + journal.next_free);
             PRIV(op)->pending_ops++;
@@ -723,25 +555,6 @@ resume_2:
         if (!space_check.check_available(op, 1, sizeof(journal_entry_big_write) + dyn_size,
             (unstable_writes.size()+unstable_unsynced+((dirty_it->second.state & BS_ST_INSTANT) ? 0 : 1))*journal.block_size))
         {
-            if (space_check.give_up)
-            {
-                auto uo_it = used_clean_objects.find(dirty_it->second.location);
-                if (uo_it != used_clean_objects.end())
-                    uo_it->second.was_freed = true;
-                else
-                    data_alloc->set(dirty_it->second.location / dsk.data_block_size, false);
-                if (!(dirty_it->second.state & BS_ST_INSTANT))
-                {
-                    unstable_unsynced--;
-                    assert(unstable_unsynced >= 0);
-                    wakeup_wait_journal();
-                }
-                if (immediate_commit != IMMEDIATE_ALL)
-                    unsynced_big_write_count--;
-                write_iodepth--;
-                cancel_all_writes(op, dirty_it, -EAGAIN);
-                return 2;
-            }
             return 0;
         }
         BS_SUBMIT_CHECK_SQES(1);
@@ -805,7 +618,6 @@ resume_4:
             {
                 unstable_unsynced--;
                 assert(unstable_unsynced >= 0);
-                wakeup_wait_journal();
             }
         }
         dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK)
@@ -834,12 +646,6 @@ resume_4:
                 });
             }
         }
-        if (PRIV(op)->forgot_unsynced)
-        {
-            // We're in the unsynced lists ourselves now, so a sync may proceed again
-            PRIV(op)->forgot_unsynced = false;
-            unsynced_forgotten--;
-        }
         if (imm && (dirty_it->second.state & BS_ST_TYPE_MASK) == BS_ST_BIG_WRITE)
         {
             // Unblock small writes
@@ -914,20 +720,6 @@ void blockstore_impl_t::handle_write_event(ring_data_t *data, blockstore_op_t *o
     }
 }
 
-void blockstore_impl_t::mark_not_dirty(int journal_sector)
-{
-    uint64_t new_ds = journal.sector_info[journal_sector].offset + journal.block_size;
-    if (new_ds >= journal.len)
-    {
-        new_ds = journal.block_size;
-    }
-    if ((journal.dirty_start + (journal.dirty_start >= journal.used_start ? 0 : journal.len)) <
-        (new_ds + (new_ds >= journal.used_start ? 0 : journal.len)))
-    {
-        journal.dirty_start = new_ds;
-    }
-}
-
 void blockstore_impl_t::release_journal_sectors(blockstore_op_t *op)
 {
     // Release flushed journal sectors
@@ -942,11 +734,19 @@ void blockstore_impl_t::release_journal_sectors(blockstore_op_t *op)
                 if (s == (1+journal.cur_sector))
                 {
                     // Forcibly move to the next sector and move dirty position
-                    // FIXME: This is the target behaviour, ok, but it seems to duplicate no_same_sector_overwrites...
                     journal.in_sector_pos = journal.block_size;
                 }
                 // We know for sure that we won't write into this sector anymore
-                mark_not_dirty(s-1);
+                uint64_t new_ds = journal.sector_info[s-1].offset + journal.block_size;
+                if (new_ds >= journal.len)
+                {
+                    new_ds = journal.block_size;
+                }
+                if ((journal.dirty_start + (journal.dirty_start >= journal.used_start ? 0 : journal.len)) <
+                    (new_ds + (new_ds >= journal.used_start ? 0 : journal.len)))
+                {
+                    journal.dirty_start = new_ds;
+                }
             }
             if (s == PRIV(op)->max_flushed_journal_sector)
                 break;
@@ -967,25 +767,16 @@ int blockstore_impl_t::dequeue_del(blockstore_op_t *op)
         .version = op->version,
     });
     assert(dirty_it != dirty_db.end());
-    if (!forget_unstable_before_delete(op, dirty_it))
-    {
-        // A preceding write is still running, retry when it finishes
-        return 0;
-    }
     blockstore_journal_check_t space_check(this);
     if (!space_check.check_available(op, 1, sizeof(journal_entry_del), (unstable_writes.size()+unstable_unsynced)*journal.block_size))
     {
-        if (space_check.give_up)
-        {
-            cancel_all_writes(op, dirty_it, -EAGAIN);
-            return 2;
-        }
         return 0;
     }
-    // Write current journal sector only if it's dirty and full, or in the immediate_commit mode.
+    // Write current journal sector only if it's dirty and full, or in the immediate_commit mode
     BS_SUBMIT_CHECK_SQES(
         (immediate_commit != IMMEDIATE_NONE ||
-            !journal.entry_fits(sizeof(journal_entry_del)) ? 1 : 0)
+            (dsk.journal_block_size - journal.in_sector_pos) < sizeof(journal_entry_del) &&
+            journal.sector_info[journal.cur_sector].dirty) ? 1 : 0
     );
     if (write_iodepth >= max_write_iodepth)
     {
@@ -994,7 +785,7 @@ int blockstore_impl_t::dequeue_del(blockstore_op_t *op)
     write_iodepth++;
     // Prepare journal sector write
     if (immediate_commit == IMMEDIATE_NONE &&
-        !journal.entry_fits(sizeof(journal_entry_del)) &&
+        (dsk.journal_block_size - journal.in_sector_pos) < sizeof(journal_entry_del) &&
         journal.sector_info[journal.cur_sector].dirty)
     {
         prepare_journal_sector_write(journal.cur_sector, op);
@@ -1017,8 +808,7 @@ int blockstore_impl_t::dequeue_del(blockstore_op_t *op)
     je->crc32 = je_crc32((journal_entry*)je);
     journal.crc32_last = je->crc32;
     dirty_it->second.state = BS_ST_DELETE | BS_ST_SUBMITTED;
-    // Join a sector write which is already queued in this same batch, see dequeue_write()
-    if (immediate_commit != IMMEDIATE_NONE || journal.sector_info[journal.cur_sector].submit_id)
+    if (immediate_commit != IMMEDIATE_NONE)
     {
         prepare_journal_sector_write(journal.cur_sector, op);
     }

src/client/CMakeLists.txt

@@ -7,13 +7,21 @@ add_library(vitastor_common STATIC
 	etcd_state_client.cpp
 	msgr_stop.cpp
 	msgr_op.cpp
+	../../json11/json11.cpp
 	osd_ops.cpp
 	pg_states.cpp
 	msgr_encrypt.cpp
 	msgr_handshake.cpp
+	../util/allocator.cpp
+	../util/addr_util.cpp
+	../util/timerfd_manager.cpp
+	../util/str_util.cpp
+	../util/json_util.cpp
+	../util/xxh_x86dispatch.c
+	 ../util/openssl_util.cpp
 )
 target_compile_options(vitastor_common PUBLIC -fPIC)
-target_link_libraries(vitastor_common ${OPENSSL_LIBRARIES} ${ISAL_CRYPTO_LIBRARIES} util)
+target_link_libraries(vitastor_common ${OPENSSL_LIBRARIES} ${ISAL_CRYPTO_LIBRARIES})
 
 # libvitastor_net.a
 set(MSGR_RDMA "")
@@ -25,12 +33,14 @@ if (RDMACM_LIBRARIES)
 	set(MSGR_RDMACM "msgr_rdmacm.cpp")
 endif (RDMACM_LIBRARIES)
 add_library(vitastor_net STATIC
+	../util/epoll_manager.cpp
 	etcd_state_client_http.cpp
 	messenger.cpp
 	msgr_iothread.cpp
 	msgr_send.cpp
 	msgr_receive.cpp
 	msgr_encrypt.cpp
+	../util/ringloop.cpp
 	http_client.cpp
 	${MSGR_RDMA}
 	${MSGR_RDMACM}
@@ -80,9 +90,8 @@ if (${WITH_FIO})
 	# libfio_vitastor_sec.so
 	add_library(fio_vitastor_sec SHARED
 		fio_sec_osd.cpp
-	)
-	target_link_libraries(fio_vitastor_sec
-		util
+		../util/rw_blocking.cpp
+		../util/addr_util.cpp
 	)
 endif (${WITH_FIO})
 

src/client/cluster_client.cpp

@@ -2,7 +2,6 @@
 // License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
 #include <stdexcept>
-#undef NDEBUG
 #include <assert.h>
 #include "pg_states.h"
 #include "cluster_client_impl.h"
@@ -39,7 +38,7 @@ cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd
         {
             // peer_osd just dropped connection
             // determine WHICH dirty_buffers are now obsolete and repeat them
-            if (wb->repeat_ops_for(this, peer_osd) > 0)
+            if (wb->repeat_ops_for(this, peer_osd, 0, 0) > 0)
             {
                 continue_ops();
             }
@@ -48,7 +47,7 @@ cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd
     msgr.exec_op = [this](osd_op_t *op)
     {
         // Garbage in
-        fprintf(stderr, "Can't handle incoming operation from client %ju\n", op->client_id);
+        fprintf(stderr, "Can't handle incoming operation from client %lu\n", op->client_id);
         msgr.stop_client(op->client_id);
         delete op;
     };
@@ -56,7 +55,7 @@ cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd
 
     st_cli = std::move(st_cli_ptr);
     st_cli->on_load_config_hook = [this](json11::Json::object & cfg) { on_load_config_hook(cfg); };
-    st_cli->on_change_osd_state_hook = [this](osd_num_t peer_osd) { on_change_osd_state_hook(peer_osd); };
+    st_cli->on_change_osd_state_hook = [this](uint64_t peer_osd) { on_change_osd_state_hook(peer_osd); };
     st_cli->on_change_pool_config_hook = [this]() { on_change_pool_config_hook(); };
     st_cli->on_change_pg_config_hook = [this]() { on_change_pool_config_hook(); };
     st_cli->on_change_pg_state_hook = [this](pool_id_t pool_id, pg_num_t pg_num, osd_num_t prev_primary) { on_change_pg_state_hook(pool_id, pg_num, prev_primary); };
@@ -591,7 +590,7 @@ void cluster_client_t::on_change_pool_config_hook()
         {
             if (log_level > 2 && pg_counts[pool_item.first])
             {
-                fprintf(stderr, "Pool %u (%s) PG count changed from %ju to %ju\n", pool_item.first, pool_item.second.name.c_str(),
+                fprintf(stderr, "Pool %u (%s) PG count changed from %lu to %lu\n", pool_item.first, pool_item.second.name.c_str(),
                     pg_counts[pool_item.first], pool_item.second.real_pg_count);
             }
             // At this point, all pool operations should have been suspended
@@ -610,11 +609,18 @@ void cluster_client_t::on_change_pool_config_hook()
     }
     inode_cache.clear();
     inode_cache_children.clear();
+    vault_keys.clear();
     continue_ops();
 }
 
 void cluster_client_t::on_change_pg_state_hook(pool_id_t pool_id, pg_num_t pg_num, osd_num_t prev_primary)
 {
+    auto & pg_cfg = st_cli->pool_config[pool_id].pg_config[pg_num];
+    if (pg_cfg.cur_primary != prev_primary)
+    {
+        // Repeat this PG operations because an OSD which stopped being primary may not fsync operations
+        wb->repeat_ops_for(this, 0, pool_id, pg_num);
+    }
     // Always continue to resume operations hung because of lack of the primary OSD
     continue_ops();
     continue_lists();
@@ -633,7 +639,7 @@ bool cluster_client_t::get_immediate_commit(uint64_t inode)
     return pool_it->second.immediate_commit == IMMEDIATE_ALL;
 }
 
-void cluster_client_t::on_change_osd_state_hook(osd_num_t peer_osd)
+void cluster_client_t::on_change_osd_state_hook(uint64_t peer_osd)
 {
     osd_tree_metrics.erase(peer_osd);
     if (msgr.wanted_peers.find(peer_osd) != msgr.wanted_peers.end())
@@ -745,7 +751,7 @@ void cluster_client_t::execute(cluster_op_t *op)
         offline_ops.push_back(op);
         return;
     }
-    op->flags = op->flags & (OSD_OP_IGNORE_READONLY | OSD_OP_WAIT_UP_TIMEOUT | OSD_OP_IGNORE_WRITEBACK); // allowed client flags
+    op->flags = op->flags & (OSD_OP_IGNORE_READONLY | OSD_OP_WAIT_UP_TIMEOUT); // allowed client flags
     execute_internal(op);
 }
 
@@ -853,8 +859,10 @@ void cluster_client_t::execute_internal(cluster_op_t *op)
     }
 }
 
-void cluster_client_t::execute_cas(cluster_op_t *op, bool nosync)
+void cluster_client_t::execute_cas(cluster_op_t *op)
 {
+    slice_rw(op);
+    op->needs_reslice = false;
     if ((op->opcode == OSD_OP_WRITE || op->opcode == OSD_OP_DELETE) && op->version && op->parts.size() > 1)
     {
         // Atomic writes to multiple stripes are unsupported
@@ -863,28 +871,6 @@ void cluster_client_t::execute_cas(cluster_op_t *op, bool nosync)
         cb(op);
         return;
     }
-    if (!(op->flags & OP_IMMEDIATE_COMMIT) && !nosync &&
-        wb->has_dirty(op->inode, op->offset, op->len))
-    {
-        cluster_op_t *sync_op = new cluster_op_t;
-        sync_op->opcode = OSD_OP_SYNC;
-        sync_op->flags = 0;
-        sync_op->callback = [this, op](cluster_op_t* sync_op)
-        {
-            execute_cas(op, true);
-            delete sync_op;
-        };
-        execute_internal(sync_op);
-        return;
-    }
-    if (!slice_rw(op))
-    {
-        op->retval = -EINVAL;
-        auto cb = std::move(op->callback);
-        cb(op);
-        return;
-    }
-    op->needs_reslice = false;
     int res = try_send(op, 0, [this, op](osd_op_t *part)
     {
         int expected = part->req.hdr.opcode == OSD_OP_DELETE ? 0 : part->req.rw.len;
@@ -893,7 +879,7 @@ void cluster_client_t::execute_cas(cluster_op_t *op, bool nosync)
             op->retval = -EIO;
         op->retval = op->retval == -EPIPE ? -EINTR : op->retval;
         auto peer_it = msgr.osd_peers.find(op->parts[0].osd_num);
-        if (op->retval != expected || (op->flags & OP_IMMEDIATE_COMMIT))
+        if (op->retval != 0 || (op->flags & OP_IMMEDIATE_COMMIT))
         {
             auto cb = std::move(op->callback);
             cb(op);
@@ -1032,7 +1018,7 @@ bool cluster_client_t::check_rw(cluster_op_t *op)
         {
             for (auto & parent: icache->chain)
             {
-                if (INODE_POOL(parent) == INODE_POOL(op->inode) && wb->has_dirty(parent, op->offset, op->len))
+                if (INODE_POOL(parent) == INODE_POOL(op->inode) && wb->has_inode(parent))
                 {
                     // Deoptimise reads - we have dirty data for one of the parent layer(s).
                     op->deoptimise_snapshot = true;
@@ -1086,12 +1072,7 @@ int cluster_client_t::continue_rw(cluster_op_t *op)
         goto resume_2;
 resume_0:
     // Slice the operation into parts
-    if (!slice_rw(op))
-    {
-        op->retval = -EINVAL;
-        erase_op(op);
-        return 1;
-    }
+    slice_rw(op);
     op->needs_reslice = false;
 resume_1:
     // Send unsent parts, if they're not subject to change
@@ -1225,16 +1206,8 @@ resume_2:
         op->retval = op->len;
         if (op->opcode == OSD_OP_READ_BITMAP || op->opcode == OSD_OP_READ_CHAIN_BITMAP)
         {
-            auto pool_it = st_cli->pool_config.find(INODE_POOL(op->inode));
-            if (pool_it == st_cli->pool_config.end())
-            {
-                op->retval = -EINVAL;
-            }
-            else
-            {
-                auto & pool_cfg = pool_it->second;
-                op->retval = op->len / pool_cfg.bitmap_granularity;
-            }
+            auto & pool_cfg = st_cli->pool_config.at(INODE_POOL(op->inode));
+            op->retval = op->len / pool_cfg.bitmap_granularity;
         }
         if (op->flush_id)
         {
@@ -1244,11 +1217,9 @@ resume_2:
         return 1;
     }
     else if (op->retval != 0 && op->opcode != OSD_OP_SYNC && !(op->flags & OP_FLUSH_BUFFER) &&
-        op->retval != -EPIPE && op->retval != -EAGAIN &&
-        (op->retval != -EIO || !client_eio_retry_interval) &&
-        (op->retval != -ENOSPC || !client_retry_enospc))
+        op->retval != -EPIPE && (op->retval != -EIO || !client_eio_retry_interval) && (op->retval != -ENOSPC || !client_retry_enospc))
     {
-        // Fatal error (neither -EPIPE, -EAGAIN, -EIO nor -ENOSPC)
+        // Fatal error (neither -EPIPE, -EIO nor -ENOSPC)
         erase_op(op);
         return 1;
     }
@@ -1311,17 +1282,11 @@ static void add_iov(int size, int skip, cluster_op_t *op, int &iov_idx, size_t &
     }
 }
 
-bool cluster_client_t::slice_rw(cluster_op_t *op)
+void cluster_client_t::slice_rw(cluster_op_t *op)
 {
     // Slice the request into individual object stripe requests
     // Primary OSDs still operate individual stripes, but their size is multiplied by PG minsize in case of EC
-    auto pool_it = st_cli->pool_config.find(INODE_POOL(op->cur_inode));
-    if (pool_it == st_cli->pool_config.end() ||
-        !pool_it->second.real_pg_count)
-    {
-        return false;
-    }
-    auto & pool_cfg = pool_it->second;
+    auto & pool_cfg = st_cli->pool_config.at(INODE_POOL(op->cur_inode));
     uint32_t pg_data_size = (pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks);
     uint64_t pg_block_size = pool_cfg.data_block_size * pg_data_size;
     uint64_t first_stripe = (op->offset / pg_block_size) * pg_block_size;
@@ -1416,7 +1381,48 @@ bool cluster_client_t::slice_rw(cluster_op_t *op)
         op->parts[i].osd_num = 0;
         i++;
     }
-    return true;
+}
+
+bool cluster_client_t::affects_pg(uint64_t inode, uint64_t offset, uint64_t len, pool_id_t pool_id, pg_num_t pg_num)
+{
+    if (INODE_POOL(inode) != pool_id)
+    {
+        return false;
+    }
+    auto & pool_cfg = st_cli->pool_config.at(INODE_POOL(inode));
+    uint32_t pg_data_size = (pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks);
+    uint64_t pg_block_size = pool_cfg.data_block_size * pg_data_size;
+    uint64_t first_stripe = (offset / pg_block_size) * pg_block_size;
+    uint64_t last_stripe = len > 0 ? ((offset + len - 1) / pg_block_size) * pg_block_size : first_stripe;
+    if ((last_stripe/pool_cfg.pg_stripe_size) - (first_stripe/pool_cfg.pg_stripe_size) + 1 >= pool_cfg.real_pg_count)
+    {
+        // All PGs are affected
+        return true;
+    }
+    pg_num_t first_pg_num = (first_stripe/pool_cfg.pg_stripe_size) % pool_cfg.real_pg_count + 1; // like map_to_pg()
+    pg_num_t last_pg_num = (last_stripe/pool_cfg.pg_stripe_size) % pool_cfg.real_pg_count + 1; // like map_to_pg()
+    return (first_pg_num <= last_pg_num
+        ? (pg_num >= first_pg_num && pg_num <= last_pg_num)
+        : (pg_num >= first_pg_num || pg_num <= last_pg_num));
+}
+
+bool cluster_client_t::affects_osd(uint64_t inode, uint64_t offset, uint64_t len, osd_num_t osd)
+{
+    auto & pool_cfg = st_cli->pool_config.at(INODE_POOL(inode));
+    uint32_t pg_data_size = (pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks);
+    uint64_t pg_block_size = pool_cfg.data_block_size * pg_data_size;
+    uint64_t first_stripe = (offset / pg_block_size) * pg_block_size;
+    uint64_t last_stripe = len > 0 ? ((offset + len - 1) / pg_block_size) * pg_block_size : first_stripe;
+    for (uint64_t stripe = first_stripe; stripe <= last_stripe; stripe += pg_block_size)
+    {
+        pg_num_t pg_num = (stripe/pool_cfg.pg_stripe_size) % pool_cfg.real_pg_count + 1; // like map_to_pg()
+        auto pg_it = pool_cfg.pg_config.find(pg_num);
+        if (pg_it != pool_cfg.pg_config.end() && pg_it->second.cur_primary == osd)
+        {
+            return true;
+        }
+    }
+    return false;
 }
 
 int cluster_client_t::try_send(cluster_op_t *op, int i, std::function<void(osd_op_t *op_part)> cb)
@@ -1426,12 +1432,7 @@ int cluster_client_t::try_send(cluster_op_t *op, int i, std::function<void(osd_o
         init_msgr();
     }
     auto part = &op->parts[i];
-    auto pool_it = st_cli->pool_config.find(INODE_POOL(op->cur_inode));
-    if (pool_it == st_cli->pool_config.end())
-    {
-        return TRY_SEND_OFFLINE;
-    }
-    auto & pool_cfg = pool_it->second;
+    auto & pool_cfg = st_cli->pool_config.at(INODE_POOL(op->cur_inode));
     auto pg_it = pool_cfg.pg_config.find(part->pg_num);
     if (pg_it != pool_cfg.pg_config.end() &&
         !pg_it->second.pause && pg_it->second.cur_primary &&
@@ -1554,10 +1555,8 @@ resume_1:
     }
     if (op->retval != 0)
     {
-        for (int i = 0; i < op->parts.size(); i++)
-            dirty_osds.insert(op->parts[i].osd_num);
         wb->fsync_error();
-        if (op->retval == -EPIPE || op->retval == -EAGAIN || op->retval == -EIO || op->retval == -ENOSPC)
+        if (op->retval == -EPIPE || op->retval == -EIO || op->retval == -ENOSPC)
         {
             // Retry later
             op->parts.clear();
@@ -1608,6 +1607,9 @@ static inline void mem_or(void *res, const void *r2, unsigned int len)
     }
 }
 
+// Error priority: others > EPERM > EIO > ENOSPC > ETIMEDOUT > EPIPE
+#define ERR_PRIO(e) (((e) == -EPERM ? 5 : ((e) == -EIO ? 4 : ((e) == -ENOSPC ? 3 : ((e) == -ETIMEDOUT ? 2 : ((e) == -EPIPE ? 1 : (!(e) ? 0 : 10)))))))
+
 void cluster_client_t::handle_op_part(cluster_op_part_t *part)
 {
     cluster_op_t *op = part->parent;
@@ -1619,7 +1621,7 @@ void cluster_client_t::handle_op_part(cluster_op_part_t *part)
         if (ERR_PRIO(part->op.reply.hdr.retval) > ERR_PRIO(op->retval))
             op->retval = part->op.reply.hdr.retval;
         uint64_t stop_client_id = 0;
-        if (op->retval != -EINTR && op->retval != -EIO && op->retval != -ENOSPC && op->retval != -EPERM && op->retval != -EAGAIN)
+        if (op->retval != -EINTR && op->retval != -EIO && op->retval != -ENOSPC && op->retval != -EPERM)
         {
             stop_client_id = part->op.client_id;
             if (op->retval != -EPIPE || log_level > 0)
@@ -1641,11 +1643,11 @@ void cluster_client_t::handle_op_part(cluster_op_part_t *part)
         // So do all these things after modifying operation state, otherwise we may hit reenterability bugs
         // FIXME postpone such things to set_immediate here to avoid bugs
         // Set op->retry_after to retry operation after a short pause (not immediately)
-        if (!op->retry_after && (op->retval == -EPIPE || op->retval == -EAGAIN ||
+        if (!op->retry_after && (op->retval == -EPIPE ||
             op->retval == -EIO && client_eio_retry_interval ||
             op->retval == -ENOSPC && client_retry_enospc))
         {
-            op->retry_after = op->retval != -EPIPE && op->retval != -EAGAIN ? client_eio_retry_interval : client_retry_interval;
+            op->retry_after = op->retval != -EPIPE ? client_eio_retry_interval : client_retry_interval;
         }
         reset_retry_timer(op->retry_after);
         if (stop_client_id)
@@ -1666,10 +1668,7 @@ void cluster_client_t::handle_op_part(cluster_op_part_t *part)
         // OK
         op->inflight_count--;
         if ((op->opcode == OSD_OP_WRITE || op->opcode == OSD_OP_DELETE) && !(op->flags & OP_IMMEDIATE_COMMIT))
-        {
-            wb->mark_flush_written_to(op->inode, part->op.req.rw.offset, part->op.req.rw.len, op->flush_id, part->osd_num);
             dirty_osds.insert(part->osd_num);
-        }
         part->flags |= PART_DONE;
         op->done_count++;
         if ((op->opcode == OSD_OP_READ || op->opcode == OSD_OP_READ_BITMAP || op->opcode == OSD_OP_READ_CHAIN_BITMAP)
@@ -1695,12 +1694,7 @@ void cluster_client_t::handle_op_part(cluster_op_part_t *part)
 void cluster_client_t::copy_part_bitmap(cluster_op_t *op, cluster_op_part_t *part)
 {
     // Copy (OR) bitmap
-    auto pool_it = st_cli->pool_config.find(INODE_POOL(op->cur_inode));
-    if (pool_it == st_cli->pool_config.end())
-    {
-        return;
-    }
-    auto & pool_cfg = pool_it->second;
+    auto & pool_cfg = st_cli->pool_config.at(INODE_POOL(op->cur_inode));
     uint32_t pg_block_size = pool_cfg.data_block_size * (
         pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks
     );

src/client/cluster_client.h

@@ -17,7 +17,6 @@
 
 #define OSD_OP_IGNORE_READONLY 0x08
 #define OSD_OP_WAIT_UP_TIMEOUT 0x10
-#define OSD_OP_IGNORE_WRITEBACK 0x20
 
 struct cluster_op_t;
 
@@ -29,7 +28,7 @@ struct cluster_op_part_t
     pg_num_t pg_num;
     osd_num_t osd_num;
     osd_op_buf_list_t iov;
-    uint32_t flags;
+    unsigned flags;
     osd_op_t op;
 };
 
@@ -40,11 +39,10 @@ struct __attribute__((visibility("default"))) cluster_op_t
     uint64_t offset;
     uint64_t len;
     // for reads and writes within a single object (stripe),
-    // non-cached reads return current version and writes can use "CAS" semantics
+    // reads can return current version and writes can use "CAS" semantics
     uint64_t version = 0;
     // flags: OSD_OP_IGNORE_READONLY - ignore inode readonly flag
     // OSD_OP_WAIT_UP_TIMEOUT - do not retry the operation infinitely if PG is inactive, only for for <wait_up_timeout>
-    // OSD_OP_IGNORE_WRITEBACK - do not read from write-back/write-repeat cache (it doesn't return version number)
     uint64_t flags = 0;
     // negative retval is an error number
     // write and read return len on success
@@ -93,7 +91,6 @@ struct inode_cache_t
     bool has_parent_loop = false;
     inode_t other_pool_parent_id = 0;
     int err_code = 0;
-    timespec err_timeout = {};
 
     ~inode_cache_t();
 };
@@ -101,7 +98,6 @@ struct inode_cache_t
 struct vault_load_key_t
 {
     int key_state = 0;
-    timespec load_time = {};
     std::string key;
 };
 
@@ -212,20 +208,23 @@ protected:
     void vault_destroy();
     void vault_parse_secret(const std::string & key_id, const std::string & err, json11::Json data);
 
+    bool affects_osd(uint64_t inode, uint64_t offset, uint64_t len, osd_num_t osd);
+    bool affects_pg(uint64_t inode, uint64_t offset, uint64_t len, pool_id_t pool_id, pg_num_t pg_num);
+
     void on_load_config_hook(json11::Json::object & config);
     void on_load_pgs_hook(bool success);
     void on_change_pool_config_hook();
     void on_change_pg_state_hook(pool_id_t pool_id, pg_num_t pg_num, osd_num_t prev_primary);
-    void on_change_osd_state_hook(osd_num_t peer_osd);
+    void on_change_osd_state_hook(uint64_t peer_osd);
     void on_change_node_placement_hook();
     void on_change_inode_hook(uint64_t inode, bool removed);
 
     void execute_internal(cluster_op_t *op);
-    void execute_cas(cluster_op_t *op, bool nosync = false);
+    void execute_cas(cluster_op_t *op);
     void unshift_op(cluster_op_t *op);
     int continue_rw(cluster_op_t *op);
     bool check_rw(cluster_op_t *op);
-    bool slice_rw(cluster_op_t *op);
+    void slice_rw(cluster_op_t *op);
     void reset_retry_timer(int new_duration);
     int try_send(cluster_op_t *op, int i, std::function<void(osd_op_t *op_part)> cb = nullptr);
     int continue_sync(cluster_op_t *op);

src/client/cluster_client_icache.cpp

@@ -1,9 +1,8 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
-#undef NDEBUG
-#include <assert.h>
 #include <stdexcept>
+#include <assert.h>
 #include "cluster_client_impl.h"
 #include "str_util.h"
 
@@ -74,18 +73,7 @@ std::shared_ptr<inode_cache_t> cluster_client_t::inode_cache_get(inode_t ino)
     auto icache_it = inode_cache.find(ino);
     if (icache_it != inode_cache.end())
     {
-        if (icache_it->second &&
-            icache_it->second->err_code &&
-            icache_it->second->err_timeout.tv_sec)
-        {
-            timespec tv = {};
-            clock_gettime(CLOCK_REALTIME, &tv);
-            if (icache_it->second->err_timeout.tv_sec > tv.tv_sec)
-                return icache_it->second;
-            printf("%lu\n", icache_it->second ? icache_it->second->err_timeout.tv_sec : 0);
-        }
-        else
-            return icache_it->second;
+        return icache_it->second;
     }
     // Fill inode cache
     auto ino_it = st_cli->inode_config.find(ino);
@@ -150,25 +138,16 @@ std::shared_ptr<inode_cache_t> cluster_client_t::inode_cache_get(inode_t ino)
             inode_config_t *cfg = !i ? &inode_cfg : chain_cfg[i-1];
             if (cfg && cfg->enc_key.substr(0, strlen(VAULT_KEY_PREFIX)) == VAULT_KEY_PREFIX)
             {
-                auto & ik = vault_keys[cfg->enc_key];
-                if (ik.key_state == VAULT_KEY_ERROR)
-                {
-                    timespec tv = {};
-                    clock_gettime(CLOCK_REALTIME, &tv);
-                    if (ik.load_time.tv_sec + vault_error_timeout_sec <= tv.tv_sec)
-                        ik.key_state = VAULT_KEY_NOT_LOADED;
-                }
-                if (vault_url.empty() || ik.key_state == VAULT_KEY_ERROR)
+                auto & ik = vault_keys[inode_cfg.enc_key];
+                if (ik.key_state == VAULT_KEY_ERROR || vault_url.empty())
                 {
                     icache->err_code = EPERM;
-                    icache->err_timeout = ik.load_time;
-                    icache->err_timeout.tv_sec += vault_error_timeout_sec;
                     enc_key_count = 0;
                 }
                 else if (ik.key_state == VAULT_KEY_NOT_LOADED)
                 {
                     ik.key_state = VAULT_KEY_LOADING;
-                    vault_key_load_queue.push_back(cfg->enc_key);
+                    vault_key_load_queue.push_back(inode_cfg.enc_key);
                     vault_load_keys();
                     icache->err_code = EAGAIN;
                     enc_key_count = 0;
@@ -233,7 +212,6 @@ void cluster_client_t::vault_parse_secret(const std::string & key_id, const std:
 {
     vault_loading = false;
     auto & k = vault_keys[key_id];
-    clock_gettime(CLOCK_REALTIME, &k.load_time);
     if (err != "")
     {
         k.key_state = VAULT_KEY_ERROR;

src/client/cluster_client_impl.h

@@ -4,7 +4,6 @@
 #pragma once
 
 #include "cluster_client.h"
-#include "small_vector.h"
 
 #define PART_SENT 1
 #define PART_DONE 2
@@ -30,11 +29,6 @@ struct cluster_buffer_t
     int state;
     uint64_t flush_id;
     uint64_t *refcnt;
-    // Funny thing is that a buffer may be CACHE_REPEATING or even CACHE_DIRTY
-    // which means it's NOT written yet and have written_to_osd at the same time,
-    // because it may be left from the PREVIOUS write attempt which we MUST remember
-    // to correctly handle OSD failure retries.
-    small_vector_t<osd_num_t, 4> written_to_osd;
 };
 
 typedef std::map<object_id, cluster_buffer_t>::iterator dirty_buf_it_t;
@@ -53,17 +47,16 @@ public:
     std::multimap<uint64_t, uint64_t*> flushed_buffers; // flush_id => refcnt
 
     ~writeback_cache_t();
-    bool has_dirty(uint64_t inode, uint64_t offset, uint64_t len);
+    bool has_inode(uint64_t inode);
     dirty_buf_it_t find_dirty(uint64_t inode, uint64_t offset);
     bool is_left_merged(dirty_buf_it_t dirty_it);
     bool is_right_merged(dirty_buf_it_t dirty_it);
     void copy_write(cluster_op_t *op, int state, uint64_t new_flush_id = 0);
-    int repeat_ops_for(cluster_client_t *cli, osd_num_t peer_osd);
+    int repeat_ops_for(cluster_client_t *cli, osd_num_t peer_osd, pool_id_t pool_id, pg_num_t pg_num);
     void start_writebacks(cluster_client_t *cli, int count);
     bool read_from_cache(cluster_op_t *op, uint32_t bitmap_granularity);
     void flush_buffers(cluster_client_t *cli, dirty_buf_it_t from_it, dirty_buf_it_t to_it);
     void mark_flush_written(uint64_t inode, uint64_t offset, uint64_t len, uint64_t flush_id);
-    void mark_flush_written_to(uint64_t inode, uint64_t offset, uint64_t len, uint64_t flush_id, osd_num_t osd_num);
     void delete_flush(uint64_t inode, uint64_t offset, uint64_t len, uint64_t flush_id);
     void fsync_start();
     void fsync_error();

src/client/cluster_client_list.cpp

@@ -63,15 +63,14 @@ void cluster_client_t::list_inode(inode_t inode, uint64_t min_offset, uint64_t m
 {
     init_msgr();
     pool_id_t pool_id = INODE_POOL(inode);
-    auto pool_it = st_cli->pool_config.find(pool_id);
-    if (!pool_id || pool_it == st_cli->pool_config.end())
+    if (!pool_id || st_cli->pool_config.find(pool_id) == st_cli->pool_config.end())
     {
         if (log_level > 0)
             fprintf(stderr, "Pool %u does not exist\n", pool_id);
         pg_callback(-EINVAL, 0, 0, std::set<object_id>());
         return;
     }
-    auto pg_stripe_size = pool_it->second.pg_stripe_size;
+    auto pg_stripe_size = st_cli->pool_config.at(pool_id).pg_stripe_size;
     if (min_offset)
         min_offset = (min_offset/pg_stripe_size) * pg_stripe_size;
     inode_list_t *lst = new inode_list_t();
@@ -249,12 +248,7 @@ void cluster_client_t::set_list_retry_timeout(int ms, timespec new_time)
 
 int cluster_client_t::start_pg_listing(inode_list_pg_t *pg)
 {
-    auto pool_it = st_cli->pool_config.find(pg->lst->pool_id);
-    if (pool_it == st_cli->pool_config.end())
-    {
-        return LIST_PG_WAIT_ACTIVE;
-    }
-    auto & pool_cfg = pool_it->second;
+    auto & pool_cfg = st_cli->pool_config.at(pg->lst->pool_id);
     auto pg_it = pool_cfg.pg_config.find(pg->pg_num);
     assert(pg->lst->real_pg_count == pool_cfg.real_pg_count);
     if (pg_it == pool_cfg.pg_config.end() ||
@@ -370,7 +364,8 @@ void cluster_client_t::send_list(inode_list_osd_t *cur_list)
             fprintf(stderr, "Failed to get PG %u/%u object list from OSD %ju (retval=%jd), skipping\n",
                 cur_list->pg->lst->pool_id, cur_list->pg->pg_num, cur_list->osd_num, op->reply.hdr.retval);
             if (!cur_list->pg->errcode ||
-                ERR_PRIO(op->reply.hdr.retval) > ERR_PRIO(cur_list->pg->errcode))
+                cur_list->pg->errcode == -EPIPE ||
+                op->reply.hdr.retval != -EPIPE)
             {
                 cur_list->pg->errcode = op->reply.hdr.retval;
             }
@@ -433,7 +428,7 @@ void cluster_client_t::finish_list_pg(inode_list_pg_t *pg, bool retry_epipe)
     auto lst = pg->lst;
     if (pg->inflight_ops == 0)
     {
-        if ((pg->errcode == -EPIPE || pg->errcode == -EAGAIN) && retry_epipe)
+        if (pg->errcode == -EPIPE && retry_epipe)
         {
             // Retry listing after <client_retry_interval> ms on EPIPE
             pg->state = LIST_PG_WAIT_RETRY;

src/client/cluster_client_wb.cpp

@@ -1,8 +1,7 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
-#undef NDEBUG
-#include <assert.h>
+#include <cassert>
 
 #include "cluster_client_impl.h"
 
@@ -18,10 +17,13 @@ writeback_cache_t::~writeback_cache_t()
     dirty_buffers.clear();
 }
 
-bool writeback_cache_t::has_dirty(uint64_t inode, uint64_t offset, uint64_t len)
+bool writeback_cache_t::has_inode(uint64_t inode)
 {
-    auto dirty_it = find_dirty(inode, offset);
-    return dirty_it != dirty_buffers.end() && dirty_it->first.stripe < offset+len;
+    auto dirty_it = dirty_buffers.lower_bound((object_id){
+        .inode = inode,
+        .stripe = 0,
+    });
+    return dirty_it != dirty_buffers.end() && dirty_it->first.inode == inode;
 }
 
 dirty_buf_it_t writeback_cache_t::find_dirty(uint64_t inode, uint64_t offset)
@@ -93,7 +95,6 @@ void writeback_cache_t::copy_write(cluster_op_t *op, int state, uint64_t new_flu
         // approaches and fix everything accordingly when I decide to implement TRIM.
         return;
     }
-    small_vector_t<osd_num_t, 4> was_written_to;
     auto dirty_it = find_dirty(op->inode, op->offset);
     auto new_end = op->offset + op->len;
     while (dirty_it != dirty_buffers.end() &&
@@ -101,11 +102,6 @@ void writeback_cache_t::copy_write(cluster_op_t *op, int state, uint64_t new_flu
         dirty_it->first.stripe < op->offset+op->len)
     {
         assert(dirty_it->first.stripe + dirty_it->second.len > op->offset);
-        for (auto osd_num: dirty_it->second.written_to_osd)
-        {
-            if (std::find(was_written_to.begin(), was_written_to.end(), osd_num) == was_written_to.end())
-                was_written_to.push_back(osd_num);
-        }
         // Remove overlapping part(s) of buffers
         auto old_end = dirty_it->first.stripe + dirty_it->second.len;
         if (dirty_it->first.stripe < op->offset)
@@ -123,7 +119,6 @@ void writeback_cache_t::copy_write(cluster_op_t *op, int state, uint64_t new_flu
                     .state = dirty_it->second.state,
                     .flush_id = dirty_it->second.flush_id,
                     .refcnt = dirty_it->second.refcnt,
-                    .written_to_osd = dirty_it->second.written_to_osd,
                 });
                 if (dirty_it->second.buf)
                 {
@@ -182,7 +177,6 @@ void writeback_cache_t::copy_write(cluster_op_t *op, int state, uint64_t new_flu
                 .state = dirty_it->second.state,
                 .flush_id = dirty_it->second.flush_id,
                 .refcnt = dirty_it->second.refcnt,
-                .written_to_osd = dirty_it->second.written_to_osd,
             });
             dirty_buffers.erase(dirty_it);
             dirty_it = new_dirty_it;
@@ -233,7 +227,6 @@ void writeback_cache_t::copy_write(cluster_op_t *op, int state, uint64_t new_flu
         .state = state,
         .flush_id = new_flush_id,
         .refcnt = refcnt,
-        .written_to_osd = std::move(was_written_to),
     });
     if (state == CACHE_DIRTY)
     {
@@ -272,7 +265,7 @@ void writeback_cache_t::copy_write(cluster_op_t *op, int state, uint64_t new_flu
     }
 }
 
-int writeback_cache_t::repeat_ops_for(cluster_client_t *cli, osd_num_t peer_osd)
+int writeback_cache_t::repeat_ops_for(cluster_client_t *cli, osd_num_t peer_osd, pool_id_t pool_id, pg_num_t pg_num)
 {
     int repeated = 0;
     if (dirty_buffers.size())
@@ -282,10 +275,11 @@ int writeback_cache_t::repeat_ops_for(cluster_client_t *cli, osd_num_t peer_osd)
         for (auto wr_it = dirty_buffers.begin(), flush_it = wr_it, last_it = wr_it; ; )
         {
             bool end = wr_it == dirty_buffers.end();
-            auto & wr_to = wr_it->second.written_to_osd;
-            bool flush_this = std::find(wr_to.begin(), wr_to.end(), peer_osd) != wr_to.end();
-            if (flush_this)
-                wr_to.clear();
+            bool flush_this = !end && wr_it->second.state != CACHE_REPEATING && wr_it->second.state != CACHE_DIRTY;
+            if (peer_osd)
+                flush_this = flush_this && cli->affects_osd(wr_it->first.inode, wr_it->first.stripe, wr_it->second.len, peer_osd);
+            if (pool_id && pg_num)
+                flush_this = flush_this && cli->affects_pg(wr_it->first.inode, wr_it->first.stripe, wr_it->second.len, pool_id, pg_num);
             if (flush_it != wr_it && (end || !flush_this ||
                 wr_it->first.inode != flush_it->first.inode ||
                 wr_it->first.stripe != last_it->first.stripe+last_it->second.len))
@@ -317,28 +311,28 @@ void writeback_cache_t::flush_buffers(cluster_client_t *cli, dirty_buf_it_t from
     op->offset = from_it->first.stripe;
     op->len = prev_it->first.stripe + prev_it->second.len - from_it->first.stripe;
     uint32_t calc_len = 0;
-    op->flush_id = ++last_flush_id;
+    uint64_t flush_id = ++last_flush_id;
     for (auto it = from_it; it != to_it; it++)
     {
         it->second.state = CACHE_REPEATING;
-        it->second.flush_id = op->flush_id;
+        it->second.flush_id = flush_id;
         if (it->second.buf)
         {
             (*it->second.refcnt)++;
-            flushed_buffers.emplace(op->flush_id, it->second.refcnt);
+            flushed_buffers.emplace(flush_id, it->second.refcnt);
             op->iov.push_back(it->second.buf, it->second.len);
         }
         calc_len += it->second.len;
     }
     assert(calc_len == op->len);
     writebacks_active++;
-    op->callback = [this](cluster_op_t* op)
+    op->callback = [this, flush_id](cluster_op_t* op)
     {
         // Buffer flushes are always retried, regardless of the error,
         // so they should never result in an error here
         assert(op->retval == op->len);
-        for (auto fl_it = flushed_buffers.lower_bound(op->flush_id);
-            fl_it != flushed_buffers.end() && fl_it->first == op->flush_id; )
+        for (auto fl_it = flushed_buffers.find(flush_id);
+            fl_it != flushed_buffers.end() && fl_it->first == flush_id; )
         {
             if (!--(*fl_it->second)) // refcnt
             {
@@ -348,7 +342,11 @@ void writeback_cache_t::flush_buffers(cluster_client_t *cli, dirty_buf_it_t from
         }
         if (op->flags & OP_IMMEDIATE_COMMIT)
         {
-            delete_flush(op->inode, op->offset, op->len, op->flush_id);
+            delete_flush(op->inode, op->offset, op->len, flush_id);
+        }
+        else
+        {
+            mark_flush_written(op->inode, op->offset, op->len, flush_id);
         }
         delete op;
         writebacks_active--;
@@ -400,21 +398,6 @@ void writeback_cache_t::mark_flush_written(uint64_t inode, uint64_t offset, uint
     }
 }
 
-void writeback_cache_t::mark_flush_written_to(uint64_t inode, uint64_t offset, uint64_t len, uint64_t flush_id, osd_num_t osd_num)
-{
-    for (auto dirty_it = find_dirty(inode, offset);
-        dirty_it != dirty_buffers.end() && dirty_it->first.inode == inode &&
-        dirty_it->first.stripe < offset+len; dirty_it++)
-    {
-        if (dirty_it->second.flush_id == flush_id && dirty_it->second.state == CACHE_REPEATING)
-        {
-            auto & wr_to = dirty_it->second.written_to_osd;
-            if (std::find(wr_to.begin(), wr_to.end(), osd_num) == wr_to.end())
-                wr_to.push_back(osd_num);
-        }
-    }
-}
-
 void writeback_cache_t::start_writebacks(cluster_client_t *cli, int count)
 {
     if (!writeback_queue.size())
@@ -540,8 +523,7 @@ bool writeback_cache_t::read_from_cache(cluster_op_t *op, uint32_t bitmap_granul
 {
     bool dirty_copied = false;
     if (dirty_buffers.size() && (op->opcode == OSD_OP_READ ||
-        op->opcode == OSD_OP_READ_BITMAP || op->opcode == OSD_OP_READ_CHAIN_BITMAP) &&
-        !(op->flags & OSD_OP_IGNORE_WRITEBACK))
+        op->opcode == OSD_OP_READ_BITMAP || op->opcode == OSD_OP_READ_CHAIN_BITMAP))
     {
         // We also have to return reads from CACHE_REPEATING buffers - they are not
         // guaranteed to be present on target OSDs at the moment of repeating

src/client/etcd_state_client.cpp

@@ -66,6 +66,10 @@ std::shared_ptr<user_info_t> etcd_state_client_t::get_user(const std::string & u
 
 bool etcd_state_client_t::check_image_perm(const std::shared_ptr<user_info_t> & user_info, inode_t inode_num, bool write)
 {
+    if (user_info->type == user_type_t::ADMIN)
+    {
+        return true;
+    }
     auto cache_it = user_info->perm_cache.find(inode_num);
     if (cache_it != user_info->perm_cache.end() &&
         cache_it->second.mod_revision == user_perm_cache_revision)
@@ -410,18 +414,15 @@ void etcd_state_client_t::clean_nonexistent_pgs()
             }
         }
     }
-    std::vector<osd_num_t> stale_peers;
     for (auto & peer_item: peer_states)
     {
         if (seen_peers.find(peer_item.first) == seen_peers.end())
-            stale_peers.push_back(peer_item.first);
-    }
-    for (auto & stale_peer: stale_peers)
-    {
-        fprintf(stderr, "OSD %ju state disappeared after reload, forgetting it\n", stale_peer);
-        parse_state((etcd_kv_t){
-            .key = etcd_prefix+"/osd/state/"+std::to_string(stale_peer),
-        });
+        {
+            fprintf(stderr, "OSD %ju state disappeared after reload, forgetting it\n", peer_item.first);
+            parse_state((etcd_kv_t){
+                .key = etcd_prefix+"/osd/state/"+std::to_string(peer_item.first),
+            });
+        }
     }
     seen_peers.clear();
 }
@@ -468,9 +469,10 @@ void etcd_state_client_t::parse_state(const etcd_kv_t & kv)
             // PG Size
             pc.pg_size = pool_item.second["pg_size"].uint64_value();
             if (pc.pg_size < 1 ||
-                pc.pg_size < 3 && (pc.scheme == POOL_SCHEME_XOR || pc.scheme == POOL_SCHEME_EC) ||
+                pool_item.second["pg_size"].uint64_value() < 3 &&
+                (pc.scheme == POOL_SCHEME_XOR || pc.scheme == POOL_SCHEME_EC) ||
                 // limit is 64 because osd_peering_pg.cpp uses a 64-bit mask for has_roles
-                pc.pg_size > 64)
+                pool_item.second["pg_size"].uint64_value() > 64)
             {
                 fprintf(stderr, "Pool %u has invalid pg_size, skipping pool\n", pool_id);
                 continue;
@@ -518,18 +520,16 @@ void etcd_state_client_t::parse_state(const etcd_kv_t & kv)
                 continue;
             }
             // Data Block Size
-            uint64_t data_block_size = pool_item.second["block_size"].uint64_value();
-            if (!data_block_size)
-                data_block_size = global_block_size;
-            if ((data_block_size & (data_block_size-1)) ||
-                data_block_size < MIN_DATA_BLOCK_SIZE || data_block_size > MAX_DATA_BLOCK_SIZE ||
-                pc.scheme != POOL_SCHEME_REPLICATED && data_block_size*pc.pg_size > UINT32_MAX)
+            pc.data_block_size = pool_item.second["block_size"].uint64_value();
+            if (!pc.data_block_size)
+                pc.data_block_size = global_block_size;
+            if ((pc.data_block_size & (pc.data_block_size-1)) ||
+                pc.data_block_size < MIN_DATA_BLOCK_SIZE || pc.data_block_size > MAX_DATA_BLOCK_SIZE)
             {
-                fprintf(stderr, "Pool %u has invalid block_size (must be a power of two between %u and %u; must not exceed 2^32-1 when multiplied by pg_size), skipping pool\n",
+                fprintf(stderr, "Pool %u has invalid block_size (must be a power of two between %u and %u), skipping pool\n",
                     pool_id, MIN_DATA_BLOCK_SIZE, MAX_DATA_BLOCK_SIZE);
                 continue;
             }
-            pc.data_block_size = data_block_size;
             // Bitmap Granularity
             pc.bitmap_granularity = pool_item.second["bitmap_granularity"].uint64_value();
             if (!pc.bitmap_granularity)
@@ -902,6 +902,7 @@ void etcd_state_client_t::parse_state(const etcd_kv_t & kv)
         {
             if (inf)
             {
+                inf->type = user_type_t::CLIENT;
                 inf->groups.clear();
                 inf->perm_cache.clear();
             }
@@ -914,6 +915,7 @@ void etcd_state_client_t::parse_state(const etcd_kv_t & kv)
                 inf = std::make_shared<user_info_t>();
                 inf->name = name;
             }
+            inf->type = value["type"] == "admin" ? user_type_t::ADMIN : user_type_t::CLIENT;
             inf->groups.clear();
             for (auto & group: value["groups"].array_items())
             {
@@ -959,8 +961,6 @@ void etcd_state_client_t::insert_inode_config(const inode_config_t & cfg)
             if (w->name == cfg.name)
             {
                 w->cfg = cfg;
-                if (w->callback)
-                    w->callback(w);
             }
         }
     }

src/client/etcd_state_client.h

@@ -101,7 +101,6 @@ struct inode_watch_t
 {
     std::string name;
     inode_config_t cfg = {};
-    std::function<void(inode_watch_t*)> callback;
 };
 
 struct http_url_t
@@ -112,6 +111,12 @@ struct http_url_t
     std::string path;
 };
 
+enum class user_type_t
+{
+    CLIENT = 0,
+    ADMIN = 1,
+};
+
 struct user_perm_t
 {
     enum class perm_type_t: uint8_t;
@@ -125,6 +130,7 @@ struct user_perm_t
 struct user_info_t
 {
     std::string name;
+    user_type_t type;
     robin_hood::unordered_flat_set<std::string> groups;
     robin_hood::unordered_flat_map<inode_t, user_perm_t> perm_cache;
 };
@@ -159,7 +165,7 @@ public:
     uint32_t global_bitmap_granularity = DEFAULT_BITMAP_GRANULARITY;
     uint32_t global_immediate_commit = IMMEDIATE_NONE;
 
-    osd_num_t osd_num = 0;
+    uint64_t osd_num = 0;
     std::string etcd_prefix;
     std::string etcd_client_cert;
     std::string etcd_client_key;