Add detailed security feature documentation

Seems slightly overcomplicated...

.gitea/workflows/test.yml

@@ -234,6 +234,60 @@ jobs:
           echo ""
         done
 
+  test_etcd_fail_https:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 10
+      run: ETCD_SCHEME=https /root/vitastor/tests/test_etcd_fail.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_etcd_fail_https_antietcd:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 10
+      run: ETCD_SCHEME=https ANTIETCD=1 /root/vitastor/tests/test_etcd_fail.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_snapshot_https:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: ETCD_SCHEME=https /root/vitastor/tests/test_snapshot.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
@@ -306,6 +360,78 @@ jobs:
           echo ""
         done
 
+  test_dump_load:
+    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_dump_load.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_dump_load_32k:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: TEST_NAME=32k OSD_ARGS="--data_csum_type crc32c --csum_block_size 32k" OFFSET_ARGS="$OSD_ARGS" /root/vitastor/tests/test_dump_load.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_old_dump_load:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: OLD=1 /root/vitastor/tests/test_dump_load.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_dump_load_old_32k:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: TEST_NAME=old_32k OLD=1 OSD_ARGS="--data_csum_type crc32c --csum_block_size 32k" OFFSET_ARGS="$OSD_ARGS" /root/vitastor/tests/test_dump_load.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_old_interrupted_rebalance:
     runs-on: ubuntu-latest
     needs: build
@@ -648,6 +774,42 @@ jobs:
           echo ""
         done
 
+  test_snapshot_chain_encrypted:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: ENCRYPTED=1 /root/vitastor/tests/test_snapshot_chain.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_snapshot_chain_enc_gcm:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: TEST_NAME=enc_gcm ENCRYPTED=1 VITASTOR_CFG=',"proto_checksums":"gcm"' /root/vitastor/tests/test_snapshot_chain.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_old_snapshot_chain:
     runs-on: ubuntu-latest
     needs: build
@@ -1206,6 +1368,24 @@ jobs:
           echo ""
         done
 
+  test_heal_ec_rdma:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 10
+      run: TEST_NAME=ec_rdma RDMA=1 SCHEME=ec /root/vitastor/tests/test_heal.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_checksum:
     runs-on: ubuntu-latest
     needs: build
@@ -1224,6 +1404,24 @@ jobs:
           echo ""
         done
 
+  test_checksum_xxhash:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: TEST_NAME=xxhash OSD_ARGS="--data_csum_type xxh3_32" /root/vitastor/tests/test_checksum.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_old_checksum:
     runs-on: ubuntu-latest
     needs: build
@@ -1458,6 +1656,24 @@ jobs:
           echo ""
         done
 
+  test_resize_last:
+    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_resize_last.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_resize_auto:
     runs-on: ubuntu-latest
     needs: build
@@ -1494,6 +1710,24 @@ jobs:
           echo ""
         done
 
+  test_old_resize_last:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: OLD=1 /root/vitastor/tests/test_resize_last.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_old_resize_auto:
     runs-on: ubuntu-latest
     needs: build
@@ -2052,3 +2286,39 @@ jobs:
           echo ""
         done
 
+  test_write_encrypted:
+    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_write_encrypted.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_ec:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: SCHEME=ec /root/vitastor/tests/test_write_encrypted.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
+

.gitea/workflows/tests-to-yaml.pl

@@ -38,6 +38,14 @@ for my $line (<>)
             {
                 $test_name .= '_antietcd';
             }
+            elsif ($1 eq 'ETCD_SCHEME' && $2 eq 'https')
+            {
+                $test_name .= '_https';
+            }
+            elsif ($1 eq 'ENCRYPTED')
+            {
+                $test_name .= '_encrypted';
+            }
             elsif ($1 eq 'OLD')
             {
                 $test_name =~ s/^test_/test_old_/s;

.gitignore

@@ -3,3 +3,4 @@
 package-lock.json
 fio
 qemu
+node_modules

CMakeLists.txt

@@ -2,7 +2,7 @@ cmake_minimum_required(VERSION 2.8...3.30)
 
 project(vitastor)
 
-set(VITASTOR_VERSION "3.0.13")
+set(VITASTOR_VERSION "3.0.15")
 
 include(CTest)
 

README-ru.md

@@ -62,6 +62,7 @@ Vitastor поддерживает QEMU-драйвер, протоколы UBLK,
     - [Дисковые параметры OSD](docs/config/layout-osd.ru.md)
     - [Прочие параметры OSD](docs/config/osd.ru.md)
     - [Параметры мониторов](docs/config/monitor.ru.md)
+    - [Безопасность](docs/config/security.ru.md)
   - [Настройки пулов](docs/config/pool.ru.md)
   - [Метаданные образов в etcd](docs/config/inode.ru.md)
 - Использование

README.md

@@ -62,6 +62,7 @@ Read more details in the documentation. You can start from here: [Quick Start](d
     - [OSD Disk Layout](docs/config/layout-osd.en.md)
     - [OSD Runtime Parameters](docs/config/osd.en.md)
     - [Monitor](docs/config/monitor.en.md)
+    - [Security](docs/config/security.en.md)
   - [Pool configuration](docs/config/pool.en.md)
   - [Image metadata in etcd](docs/config/inode.en.md)
 - Usage

csi/Makefile

@@ -1,4 +1,4 @@
-VITASTOR_VERSION ?= v3.0.13
+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:v3.0.13
+          image: vitalif/vitastor-csi:v3.0.15
           args:
             - "--node=$(NODE_ID)"
             - "--endpoint=$(CSI_ENDPOINT)"

csi/deploy/007-csi-provisioner.yaml

@@ -121,7 +121,7 @@ spec:
             privileged: true
             capabilities:
               add: ["SYS_ADMIN"]
-          image: vitalif/vitastor-csi:v3.0.13
+          image: vitalif/vitastor-csi:v3.0.15
           args:
             - "--node=$(NODE_ID)"
             - "--endpoint=$(CSI_ENDPOINT)"

csi/src/config.go

@@ -5,7 +5,7 @@ package vitastor
 
 const (
     vitastorCSIDriverName    = "csi.vitastor.io"
-    vitastorCSIDriverVersion = "3.0.13"
+    vitastorCSIDriverVersion = "3.0.15"
 )
 
 // Config struct fills the parameters of request or user input

debian/changelog

@@ -1,4 +1,4 @@
-vitastor (3.0.13-1) unstable; urgency=medium
+vitastor (3.0.15-1) unstable; urgency=medium
 
   * Bugfixes
 

debian/control

@@ -3,7 +3,7 @@ Section: admin
 Priority: optional
 Maintainer: Vitaliy Filippov <vitalif@yourcmc.ru>
 Build-Depends: debhelper, g++ (>= 8), libstdc++6 (>= 8),
-  linux-libc-dev, libgoogle-perftools-dev, libjerasure-dev, libgf-complete-dev,
+  linux-libc-dev, libgoogle-perftools-dev, libjerasure-dev, libgf-complete-dev, libc-ares-dev,
   libibverbs-dev, librdmacm-dev, libisal-dev, cmake, pkg-config, libnl-3-dev, libnl-genl-3-dev,
   node-bindings <!nocheck>, node-gyp, node-nan
 Standards-Version: 4.5.0

debian/rules

@@ -11,6 +11,7 @@ override_dh_install:
 	cp -v node-binding/package.json node-binding/index.js node-binding/addon.cc node-binding/addon.h node-binding/client.cc node-binding/client.h debian/tmp/usr/lib/x86_64-linux-gnu/nodejs/vitastor
 	cp -v node-binding/build/Release/addon.node debian/tmp/usr/lib/x86_64-linux-gnu/nodejs/vitastor/build/Release
 	dh_install
+	cd debian/vitastor-mon/usr/lib/vitastor/mon && npm install --production
 
 override_dh_installdeb:
 	cat debian/fio_version >> debian/vitastor-fio.substvars

debian/vitastor-build.sh

@@ -37,12 +37,6 @@ rm -rf a b
 
 echo "dep:fio=$FIO" > debian/fio_version
 
-cd /root/vitastor/packages/vitastor-$REL/vitastor-$VER
-mkdir mon/node_modules
-cd mon/node_modules
-curl -s https://git.yourcmc.ru/vitalif/antietcd/archive/master.tar.gz | tar -zx
-curl -s https://git.yourcmc.ru/vitalif/tinyraft/archive/master.tar.gz | tar -zx
-
 cd /root/vitastor/packages/vitastor-$REL
 if [[ ( "$REL" = "trixie" || "$REL" = "resolute" ) && -e ../vitastor-bookworm/vitastor_$VER.orig.tar.xz ]]; then
     # Fucking shit, archives differ between bookworm (xz 5.4.1) and trixie (xz 5.8.1)

debian/vitastor-buildenv.Dockerfile

@@ -12,20 +12,18 @@ 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; \
-        apt-get update; \
-        apt-get -y install wget; \
-        wget https://vitastor.io/debian/pubkey.gpg -O /etc/apt/trusted.gpg.d/vitastor.gpg; \
-        echo "deb https://vitastor.io/debian $REL main" >> /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; \
+    echo "deb https://vitastor.io/debian $REL main" >> /etc/apt/sources.list; \
     grep '^deb ' /etc/apt/sources.list | perl -pe 's/^deb/deb-src/' >> /etc/apt/sources.list; \
     perl -i -pe 's/Types: deb$/Types: deb deb-src/' /etc/apt/sources.list.d/*.sources || true; \
     echo 'APT::Install-Recommends false;' >> /etc/apt/apt.conf; \
     echo 'APT::Install-Suggests false;' >> /etc/apt/apt.conf
 
 RUN apt-get update && \
-    apt-get -y install fio libgoogle-perftools-dev devscripts libjerasure-dev cmake \
+    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 curl nodejs npm node-nan node-bindings && \
     apt-get -y build-dep fio && \
     apt-get --download-only source fio

docker/Makefile

@@ -1,4 +1,4 @@
-VITASTOR_VERSION ?= v3.0.13
+VITASTOR_VERSION ?= v3.0.15
 
 all: build push
 

docker/etc/systemd/system/vitastor-etcd.service

@@ -12,12 +12,7 @@ EnvironmentFile=/etc/vitastor/etcd.conf
 SyslogIdentifier=etcd
 ExecStart=bash -c 'docker run --rm -i -v /var/lib/vitastor/etcd:/data \
     --log-driver none --network host $CONTAINER_OPTIONS --name vitastor-etcd \
-    $ETCD_IMAGE /usr/local/bin/etcd --name "$ETCD_NAME" --data-dir /data \
-    --snapshot-count 10000 --advertise-client-urls http://$ETCD_IP:2379 --listen-client-urls http://$ETCD_IP:2379 \
-    --initial-advertise-peer-urls http://$ETCD_IP:2380 --listen-peer-urls http://$ETCD_IP:2380 \
-    --initial-cluster-token vitastor-etcd-1 --initial-cluster "$ETCD_INITIAL_CLUSTER" \
-    --initial-cluster-state new --max-txn-ops=100000 --max-request-bytes=104857600 \
-    --auto-compaction-retention=10 --auto-compaction-mode=revision'
+    $ETCD_IMAGE /usr/local/bin/etcd --data-dir /data
 ExecStop=docker stop vitastor-etcd
 Restart=always
 StartLimitInterval=0

docker/etc/vitastor/docker.conf

@@ -4,7 +4,7 @@
 #
 
 # Desired Vitastor version
-VITASTOR_VERSION=v3.0.13
+VITASTOR_VERSION=v3.0.15
 
 # Additional arguments for all containers
 # For example, you may want to specify a custom logging driver here

docker/etc/vitastor/etcd.conf

@@ -1,4 +1 @@
 ETCD_IMAGE=quay.io/coreos/etcd:v3.5.18
-ETCD_NAME=""
-ETCD_IP=""
-ETCD_INITIAL_CLUSTER=""

docs/config.en.md

@@ -38,3 +38,4 @@ In the future, additional configuration methods may be added:
 - [OSD Disk Layout](config/layout-osd.en.md)
 - [OSD Runtime Parameters](config/osd.en.md)
 - [Monitor](config/monitor.en.md)
+- [Security Parameters](config/security.en.md)

docs/config.ru.md

@@ -41,3 +41,4 @@
 - [Дисковые параметры OSD](config/layout-osd.ru.md)
 - [Прочие параметры OSD](config/osd.ru.md)
 - [Параметры мониторов](config/monitor.ru.md)
+- [Параметры безопасности](config/security.ru.md)

docs/config/layout-osd.en.md

@@ -198,8 +198,14 @@ put a modified value into etcd key /vitastor/config/global.
 - Type: string
 - Default: none
 
-Data checksum type to use. May be "crc32c" or "none". Set to "crc32c" to
-enable data checksums.
+Data and metadata checksum type to use. May be "crc32c", "xxh3_32" or "none".
+Select crc32c or xxh3_32 and set csum_block_size to enable data checksums.
+
+Both crc32c and xxh3_32 are almost equally fast, xxh3_32 is safer. xxh3_32 is
+the xxhash3 algorithm truncated from 64 to 32 bits (which is still a good hash).
+
+Note that enabled data checksums either increase memory usage or reduce
+performance. Check details in [csum_block_size](#csum_block_size) description.
 
 ## csum_block_size
 

docs/config/layout-osd.ru.md

@@ -209,8 +209,12 @@ journal_block_size и meta_block_size. Однако на данный момен
 - Тип: строка
 - Значение по умолчанию: none
 
-Тип используемых OSD контрольных сумм данных. Может быть "crc32c" или "none".
-Установите в "crc32c", чтобы включить расчёт и проверку контрольных сумм данных.
+Тип используемых OSD контрольных сумм данных и метаданных. Может быть "crc32c",
+"xxh3_32" или "none". Выберите crc32c или xxh3_32 и установите csum_block_size,
+чтобы включить контрольные суммы данных.
+
+И crc32c, и xxh3_32 примерно одинаково быстры, xxh3_32 надёжней. xxh3_32 - это
+алгоритм xxhash3, обрезанный с 64 до 32 бит (это всё равно хороший хеш).
 
 Следует понимать, что контрольные суммы в зависимости от размера блока их
 расчёта либо увеличивают потребление памяти, либо снижают производительность.

docs/config/monitor.en.md

@@ -50,6 +50,9 @@ or antietcd_data_dir options). All other antietcd parameters
 cluster, cluster_key, persist_filter, stale_read can also be set in
 Vitastor configuration with `antietcd_` prefix.
 
+See also: [antietcd_cert](security.en.md#antietcd_cert),
+[antietcd_key](security.en.md#antietcd_key) and [etcd_proxy](security.en.md#etcd_proxyurls).
+
 You can dump/load data to or from antietcd using Antietcd `anticli` tool:
 
 ```

docs/config/monitor.ru.md

@@ -50,6 +50,9 @@ antietcd_data_file или antietcd_data_dir). Все остальные пара
 node_id, cluster, cluster_key, persist_filter, stale_read также можно задавать
 в конфигурации Vitastor с префиксом `antietcd_`.
 
+Смотрите также настройки [antietcd_cert](security.ru.md#antietcd_cert),
+[antietcd_key](security.ru.md#antietcd_key) и [etcd_proxy](security.ru.md#etcd_proxyurls).
+
 Вы можете выгружать/загружать данные в или из antietcd с помощью его инструмента
 `anticli`:
 

docs/config/security.en.md

@@ -0,0 +1,306 @@
+[Documentation](../../README.md#documentation) → [Configuration](../config.en.md) → Security Parameters
+
+-----
+
+[Читать на русском](security.ru.md)
+
+# Security Parameters
+
+These parameters affect your Vitastor installation security and apply to OSDs, monitors and clients.
+
+Most of them can be set in /etc/vitastor/vitastor.conf and in etcd, but don't support online modification.
+
+All certificate and private key parameters mentioned may contain a path to a PEM file or just
+a PEM string with certificate or a private key. In the latter case, the string must begin with
+"-----BEGIN CERTIFICATE-----" or "-----BEGIN PRIVATE KEY-----".
+
+- [use_perms](#use_perms)
+- [cert](#cert)
+- [pkey](#pkey)
+- [etcd_ca](#etcd_ca)
+- [client_ca](#client_ca)
+- [osd_ca](#osd_ca)
+- [mon_ca](#mon_ca)
+- [antietcd_cert](#antietcd_cert)
+- [antietcd_key](#antietcd_key)
+- [etcd_proxy.urls](#etcd_proxyurls)
+- [etcd_proxy.cert](#etcd_proxycert)
+- [etcd_proxy.key](#etcd_proxykey)
+- [etcd_proxy.ca](#etcd_proxyca)
+- [osd_cert](#osd_cert)
+- [osd_pkey](#osd_pkey)
+- [api_cert](#api_cert)
+- [api_pkey](#api_pkey)
+- [etcd_client_cert](#etcd_client_cert)
+- [etcd_client_key](#etcd_client_key)
+- [osd_etcd_client_cert](#osd_etcd_client_cert)
+- [osd_etcd_client_key](#osd_etcd_client_key)
+- [mon_etcd_client_cert](#mon_etcd_client_cert)
+- [mon_etcd_client_key](#mon_etcd_client_key)
+- [proto_checksums](#proto_checksums)
+- [force_proto_checksums](#force_proto_checksums)
+- [max_cipher_pool_size](#max_cipher_pool_size)
+- [vault_url](#vault_url)
+- [vault_secret_api_path](#vault_secret_api_path)
+- [vault_client_cert](#vault_client_cert)
+- [vault_client_key](#vault_client_key)
+- [vault_ca](#vault_ca)
+- [vault_timeout_ms](#vault_timeout_ms)
+- [vault_error_timeout_sec](#vault_error_timeout_sec)
+- [vault_refresh_leeway_sec](#vault_refresh_leeway_sec)
+
+## use_perms
+
+- Type: boolean
+- Default: false
+
+Enable client permissions in a Vitastor cluster, including Antietcd built into the Monitor.
+Requires configured encryption. Also note that separate Antietcd requires separate configuration
+to use permissions (see [security documentation](../intro/security.en.md) for details).
+
+## cert
+
+- Type: string
+
+Client certificate of the current Vitastor user. Required for Vitastor protocol encryption.
+Must be signed with [client_ca](#client_ca). Also used as the client certificate for etcd/Antietcd
+connections by default.
+
+## pkey
+
+- Type: string
+
+Private key of the current Vitastor user.
+
+## etcd_ca
+
+- Type: string
+
+Trusted TLS CA to verify etcd server certificate. Or just the etcd server's
+certificate itself - it's fine to use it for etcd_ca.
+
+## client_ca
+
+- Type: string
+
+Trusted TLS CA to verify Vitastor client certificates.
+Mandatory for Vitastor protocol encryption.
+
+## osd_ca
+
+- Type: string
+
+Trusted TLS CA to verify Vitastor OSD certificates. Also mandatory for Vitastor protocol
+encryption. Must be different from client_ca. May be equal to osd_cert - different OSDs
+don't require separate certificates at the moment because their permissions don't differ.
+
+## mon_ca
+
+- Type: string
+
+Trusted TLS CA to verify Vitastor Monitor certificates. Used only for separate Antietcd,
+not required when a monitor built-in Antietcd is used. May be equal to mon_client_etcd_cert.
+
+## antietcd_cert
+
+- Type: string
+
+Server TLS certificate for Antietcd built into the Monitor.
+
+## antietcd_key
+
+- Type: string
+
+Private key for antietcd_cert.
+
+## etcd_proxy.urls
+
+- Type: string or array of strings
+
+etcd URLs for Antietcd etcd proxy mode.
+See [Mon as Etcd proxy](../intro/security.en.md#mon-as-etcd-proxy) for details.
+
+## etcd_proxy.cert
+
+- Type: string
+
+Client certificate for Antietcd connections to etcd in proxy mode.
+
+## etcd_proxy.key
+
+- Type: string
+
+Private key for etcd_proxy.cert.
+
+## etcd_proxy.ca
+
+- Type: string
+
+Trusted TLS CA to verify etcd server certificate when connecting to it from Antietcd.
+
+## osd_cert
+
+- Type: string
+
+Vitastor OSD server certificate. Required for Vitastor protocol encryption. May be equal
+to [osd_ca](#osd_ca) - all OSDs share the same permission set for now. Also used as the client
+certificate for connections from OSD to etcd/Antietcd by default.
+
+## osd_pkey
+
+- Type: string
+
+Private key for osd_cert.
+
+## api_cert
+
+- Type: string
+
+Server TLS certificate for [vitastor-cli serve](../usage/cli.en.md#serve) API server.
+
+## api_pkey
+
+- Type: string
+
+Private key for api_cert.
+
+## etcd_client_cert
+
+- Type: string
+
+Client TLS certificate to use for connections from Vitastor clients to etcd/Antietcd if you don't want
+to use the common client certificate [cert](#cert).
+
+## etcd_client_key
+
+- Type: string
+
+Private key for etcd_client_cert.
+
+## osd_etcd_client_cert
+
+- Type: string
+
+Client TLS certificate to use for connections from Vitastor OSDs to etcd/Antietcd if you don't want
+to use the common OSD certificate [osd_cert](#osd_cert).
+
+## osd_etcd_client_key
+
+- Type: string
+
+Private key for osd_etcd_client_cert.
+
+## mon_etcd_client_cert
+
+- Type: string
+
+Client TLS certificate to use for connections from Vitastor Monitors to etcd/Antietcd - required
+if you don't use the built-in Antietcd. In case you use it Monitor has direct access to Antietcd data
+and doesn't require any connection.
+
+## mon_etcd_client_key
+
+- Type: string
+
+Private key for mon_etcd_client_cert.
+
+## proto_checksums
+
+- Type: string
+- Default: payload
+
+One of "full", "payload", "gcm", "none":
+- "full" means calculate and verify transport level checksums from the full message data
+  including the header - recommended for unencrypted setups.
+- "payload" enables checksums only for the actual read/write data, but skips them for message
+  headers - recommended for encrypted setups because headers are already protected by AES-GCM.
+- "gcm" disables checksums and enables AES-GCM encryption of the whole messages including headers
+  and data - AES-GCM already includes MAC which is actually a stronger checksum. This option is
+  slower and is only recommended for untrusted networks.
+- "none" disables transport level checksums at all.
+
+## force_proto_checksums
+
+- Type: string
+
+To allow older clients to connect to a Vitastor cluster with enabled checksums, Vitastor OSDs
+allow clients to downgrade their proto_checksums by default. force_proto_checksums sets the
+minimum security level allowed for connecting clients. When encryption is disabled, default
+force_proto_checksums is none and clients without checksums are allowed. With enabled
+encryption, force_proto_checksums becomes "payload" by default to block unauthenticated data
+on the transport level.
+
+## max_cipher_pool_size
+
+- Type: integer
+- Default: 256
+
+Maximum number of OpenSSL cipher contexts cached in OSD memory, counted separately
+for each cipher and for encryption/decryption. Probably doesn't require modification.
+
+## vault_url
+
+- Type: string
+
+Vault base URL.
+
+Vitastor clients support AES-256-XTS image data encryption with different per-image keys.
+Encryption is performed by the client, OSDs don't have access to decrypted data.
+
+Encryption keys may be stored in etcd or, for the increased security level, in an external
+[HashiCorp Vault](https://developer.hashicorp.com/vault/) or [OpenBao](https://openbao.org/)
+instance.
+
+Vitastor clients use [v1 k/v secrets engine](https://openbao.org/api-docs/secret/kv/kv-v1/)
+and [TLS authentication engine](https://openbao.org/api-docs/auth/cert/) in Vault.
+
+In that case, only key IDs are stored in etcd.
+
+## vault_secret_api_path
+
+- Type: string
+- Default: /v1/secret/
+
+Vault v1 secret API mount path to use.
+
+## vault_client_cert
+
+- Type: string
+
+Client TLS certificate to use for Vault connections if you don't want to use the common Vitastor
+client certificate [cert](#cert) which is also used for Vault connections by default.
+
+## vault_client_key
+
+- Type: string
+
+Private key for the vault_client_cert certificate.
+
+## vault_ca
+
+- Type: string
+
+Trusted TLS CA to verify Vault server certificate. May be path to a file,
+directory or just a PEM string with certificate.
+
+## vault_timeout_ms
+
+- Type: integer
+- Default: 5000
+
+Timeout for Vault requests in milliseconds.
+
+## vault_error_timeout_sec
+
+- Type: integer
+- Default: 60
+
+Time (in seconds) to wait before retrying after receiving an error from Vault.
+
+## vault_refresh_leeway_sec
+
+- Type: integer
+- Default: 60
+
+Extra time (in seconds) before real Vault token lease_timeout to refresh it, just
+in case of system clock drift.

docs/config/security.ru.md

@@ -0,0 +1,312 @@
+[Документация](../../README-ru.md#документация) → [Конфигурация](../config.ru.md) → Параметры безопасности
+
+-----
+
+[Read in English](security.en.md)
+
+# Параметры безопасности
+
+Данные параметры затрагивают безопасность инсталляций Vitastor и используются
+OSD, мониторами и клиентами.
+
+Большая их часть может задаваться в /etc/vitastor/vitastor.conf и в etcd, но не
+поддерживает онлайн-изменение.
+
+Все параметры сертификатов и закрытых ключей могут быть путём к файлу или просто
+строкой с сертификатом в формате PEM. В последнем случае строка должна начинаться с
+"-----BEGIN CERTIFICATE-----" или "-----BEGIN PRIVATE KEY-----".
+
+- [use_perms](#use_perms)
+- [cert](#cert)
+- [pkey](#pkey)
+- [etcd_ca](#etcd_ca)
+- [client_ca](#client_ca)
+- [osd_ca](#osd_ca)
+- [mon_ca](#mon_ca)
+- [antietcd_cert](#antietcd_cert)
+- [antietcd_key](#antietcd_key)
+- [etcd_proxy.urls](#etcd_proxyurls)
+- [etcd_proxy.cert](#etcd_proxycert)
+- [etcd_proxy.key](#etcd_proxykey)
+- [etcd_proxy.ca](#etcd_proxyca)
+- [osd_cert](#osd_cert)
+- [osd_pkey](#osd_pkey)
+- [api_cert](#api_cert)
+- [api_pkey](#api_pkey)
+- [etcd_client_cert](#etcd_client_cert)
+- [etcd_client_key](#etcd_client_key)
+- [osd_etcd_client_cert](#osd_etcd_client_cert)
+- [osd_etcd_client_key](#osd_etcd_client_key)
+- [mon_etcd_client_cert](#mon_etcd_client_cert)
+- [mon_etcd_client_key](#mon_etcd_client_key)
+- [proto_checksums](#proto_checksums)
+- [force_proto_checksums](#force_proto_checksums)
+- [max_cipher_pool_size](#max_cipher_pool_size)
+- [vault_url](#vault_url)
+- [vault_secret_api_path](#vault_secret_api_path)
+- [vault_client_cert](#vault_client_cert)
+- [vault_client_key](#vault_client_key)
+- [vault_ca](#vault_ca)
+- [vault_timeout_ms](#vault_timeout_ms)
+- [vault_error_timeout_sec](#vault_error_timeout_sec)
+- [vault_refresh_leeway_sec](#vault_refresh_leeway_sec)
+
+## use_perms
+
+- Тип: булево (да/нет)
+- Значение по умолчанию: false
+
+Включает клиентские привилегии в кластере Vitastor, в том числе во встроенном в мониторе Antietcd.
+Требует настроенного шифрования протокола. Также обратите внимание, что отдельно установленный Antietcd
+требует отдельной настройки привилегий (подробности смотрите в [документации безопасности](../intro/security.ru.md)).
+
+## cert
+
+- Тип: строка
+
+Клиентский сертификат текущего пользователя Vitastor. Требуется для шифрования протокола Vitastor.
+Должен быть подписан [client_ca](#client_ca). Также по умолчанию используется как клиентский
+сертификат для подключения к etcd/Antietcd и Vault.
+
+## pkey
+
+- Тип: строка
+
+Закрытый ключ текущего пользователя Vitastor.
+
+## etcd_ca
+
+- Тип: строка
+
+Доверенный корневой TLS-сертификат для проверки сертификата сервера etcd.
+Либо же просто сам сертификат сервера etcd - его можно использовать как etcd_ca.
+
+## client_ca
+
+- Тип: строка
+
+Доверенный TLS-сертификат для проверки сертификатов клиентов Vitastor.
+Требуется для шифрования протокола Vitastor.
+
+## osd_ca
+
+- Тип: строка
+
+Доверенный TLS-сертификат для проверки сертификатов OSD Vitastor. Также обязателен
+для шифрования протокола Vitastor. Должен отличаться от client_ca. Может быть равен
+osd_cert - разные OSD не требуют разных сертификатов, потому что на данный момент
+привилегии разных OSD никак не отличаются.
+
+## mon_ca
+
+- Тип: строка
+
+Доверенный TLS-сертификат для проверки сертификатов мониторов Vitastor. Используется
+только отдельно установленным Antietcd, не требуется при использовании встроенного в монитор
+Antietcd. Может быть равен mon_client_etcd_cert.
+
+## antietcd_cert
+
+- Тип: строка
+
+Серверный TLS-сертификат для Antietcd, встроенного в монитор.
+
+## antietcd_key
+
+- Тип: строка
+
+Закрытый ключ для сертификата antietcd_cert.
+
+## etcd_proxy.urls
+
+- Тип: строка или массив строк
+
+Адреса etcd для режима Antietcd etcd-прокси.
+Смотрите подробности в разделе [Mon в роли Etcd proxy](../intro/security.ru.md#mon-в-роли-etcd-proxy).
+
+## etcd_proxy.cert
+
+- Тип: строка
+
+Клиентский сертификат для подключений от Antietcd к etcd в режиме прокси.
+
+## etcd_proxy.key
+
+- Тип: строка
+
+Закрытый ключ для сертификата etcd_proxy.cert.
+
+## etcd_proxy.ca
+
+- Тип: строка
+
+Доверенный TLS-сертификат для проверки сертификата сервера etcd при подключениях от Antietcd.
+
+## osd_cert
+
+- Тип: строка
+
+Сертификат сервера Vitastor OSD. Требуется для шифрования протокола Vitastor. Может быть равен
+[osd_ca](#osd_ca) - все OSD на данный момент имеют одинаковые привилегии. Также по умолчанию
+используется как клиентский сертификат для подключения от OSD к etcd/Antietcd.
+
+## osd_pkey
+
+- Тип: строка
+
+Закрытый ключ для сертификата osd_cert.
+
+## api_cert
+
+- Тип: строка
+
+Серверный TLS-сертификат для API-сервера [vitastor-cli serve](../usage/cli.ru.md#serve).
+
+## api_pkey
+
+- Тип: строка
+
+Закрытый ключ для сертификата api_cert.
+
+## etcd_client_cert
+
+- Тип: строка
+
+Клиентский TLS сертификат для подключений от клиентов Vitastor к etcd/Antietcd, если вы не хотите
+использовать общий клиентский сертификат [cert](#cert).
+
+## etcd_client_key
+
+- Тип: строка
+
+Закрытый ключ для сертификата etcd_client_cert.
+
+## osd_etcd_client_cert
+
+- Тип: строка
+
+Клиентский TLS сертификат для подключений от Vitastor OSD к etcd/Antietcd, если вы не хотите
+использовать общий сертификат OSD [osd_cert](#osd_cert).
+
+## osd_etcd_client_key
+
+- Тип: строка
+
+Закрытый ключ для сертификата osd_etcd_client_cert.
+
+## mon_etcd_client_cert
+
+- Тип: строка
+
+Клиентский TLS сертификат для подключений от мониторов Vitastor к etcd/Antietcd - требуется, если
+вы не используете встроенный в монитор Antietcd. Если вы используете его, то монитор и так имеет
+прямой доступ к данным Antietcd и не требует никаких соединений.
+
+## mon_etcd_client_key
+
+- Тип: строка
+
+Закрытый ключ для сертификата mon_etcd_client_cert.
+
+## proto_checksums
+
+- Тип: строка
+- Значение по умолчанию: payload
+
+Одно из значений "full", "payload", "gcm" и "none":
+- "full" означает расчёт и проверку контрольных сумм на транспортном уровне от полных сообщений,
+  включая их заголовки и данные - рекомендуется для кластеров без шифрования.
+- "payload" включает контрольные суммы только для данных сообщений, но пропускает заголовки -
+  такая настройка рекомендуется для кластеров с включённым шифрованием, потому что в них заголовки
+  и так защищены шифрованием AES-GCM.
+- "gcm" отключает контрольные суммы и включает шифрование полных сообщений включая заголовки и
+  данные - AES-GCM уже включает в себя MAC, который по сути является криптостойкой контрольной
+  суммой. Такая настройка медленнее и рекомендуется только для недоверенных сетей.
+- "none" полностью отключает контрольные суммы на транспортном уровне.
+
+## force_proto_checksums
+
+- Тип: строка
+
+Чтобы старые клиенты Vitastor могли подключаться к кластеру с включёнными контрольными
+суммами, Vitastor OSD по умолчанию разрешают клиентам отключать контрольные суммы
+данных (proto_checksums). Настройка force_proto_checksums задаёт минимальный уровень
+безопасности, разрешённый для подключающихся клиентов. Когда шифрование отключено,
+force_proto_checksums по умолчанию равно none и подключения клиентов без контрольных
+сумм разрешаются. При включённом шифровании значение по умолчанию force_proto_checksums
+становится "payload", чтобы блокировать подключения с неаутентифицированными данными.
+
+## max_cipher_pool_size
+
+- Тип: целое число
+- Значение по умолчанию: 256
+
+Максимальное количество кэшируемых в памяти OSD контекстов шифра OpenSSL, учитываемое
+отдельно для каждого шифра и для шифрования и расшифровки. Вряд ли требует изменения.
+
+## vault_url
+
+- Тип: строка
+
+Базовый адрес Vault.
+
+Клиенты Vitastor поддерживают AES-256-XTS шифрование данных образов с отдельными ключами на
+каждый образ. Данные шифруются клиентами, OSD не имеют доступа к незашифрованным данным.
+
+Ключи шифрования могут храниться в etcd или, для повышенного уровня безопасности, во внешнем
+[HashiCorp Vault](https://developer.hashicorp.com/vault/) или [OpenBao](https://openbao.org/).
+
+Клиенты Vitastor используют [движок секретов v1](https://openbao.org/api-docs/secret/kv/kv-v1/)
+и [TLS-аутентификацию](https://openbao.org/api-docs/auth/cert/) в Vault.
+
+В этом случае, только ID ключей хранятся в etcd.
+
+## vault_secret_api_path
+
+- Тип: строка
+- Значение по умолчанию: /v1/secret/
+
+Путь к API секретов v1 для использования клиентами.
+
+## vault_client_cert
+
+- Тип: строка
+
+Клиентский TLS сертификат для подключений к Vault на тот случай, если вы не хотите использовать
+общий сертификат клиента Vitastor [cert](#cert), используемый для подключений к Vault по умолчанию.
+
+## vault_client_key
+
+- Тип: строка
+
+Закрытый ключ для сертификата vault_client_cert.
+
+## vault_ca
+
+- Тип: строка
+
+Доверенный корневой TLS-сертификат для проверки сертификата сервера Vault.
+Может быть путём к файлу, директории или просто строкой с сертификатом в
+формате PEM.
+
+## vault_timeout_ms
+
+- Тип: целое число
+- Значение по умолчанию: 5000
+
+Максимально время выполнения Vault-запросов в миллисекундах.
+
+## vault_error_timeout_sec
+
+- Тип: целое число
+- Значение по умолчанию: 60
+
+Время (в секундах) для ожидания перед повторной попыткой при получении ошибки от Vault.
+
+## vault_refresh_leeway_sec
+
+- Тип: целое число
+- Значение по умолчанию: 60
+
+Зазор времени (в секундах), чтобы обновлять токены Vault чуть раньше их реального
+lease_timeout, на случай "ухода" системных часов.

docs/config/src/included.en.md

@@ -44,6 +44,8 @@
 
 {{../../config/monitor.en.md|indent=2}}
 
+{{../../config/security.en.md|indent=2}}
+
 {{../../config/pool.en.md|indent=2}}
 
 {{../../config/inode.en.md|indent=2}}

docs/config/src/included.ru.md

@@ -44,6 +44,8 @@
 
 {{../../config/monitor.ru.md|indent=2}}
 
+{{../../config/security.ru.md|indent=2}}
+
 {{../../config/pool.ru.md|indent=2}}
 
 {{../../config/inode.ru.md|indent=2}}

docs/config/src/layout-osd.yml

@@ -233,11 +233,21 @@
   type: string
   default: none
   info: |
-    Data checksum type to use. May be "crc32c" or "none". Set to "crc32c" to
-    enable data checksums.
+    Data and metadata checksum type to use. May be "crc32c", "xxh3_32" or "none".
+    Select crc32c or xxh3_32 and set csum_block_size to enable data checksums.
+
+    Both crc32c and xxh3_32 are almost equally fast, xxh3_32 is safer. xxh3_32 is
+    the xxhash3 algorithm truncated from 64 to 32 bits (which is still a good hash).
+
+    Note that enabled data checksums either increase memory usage or reduce
+    performance. Check details in [csum_block_size](#csum_block_size) description.
   info_ru: |
-    Тип используемых OSD контрольных сумм данных. Может быть "crc32c" или "none".
-    Установите в "crc32c", чтобы включить расчёт и проверку контрольных сумм данных.
+    Тип используемых OSD контрольных сумм данных и метаданных. Может быть "crc32c",
+    "xxh3_32" или "none". Выберите crc32c или xxh3_32 и установите csum_block_size,
+    чтобы включить контрольные суммы данных.
+
+    И crc32c, и xxh3_32 примерно одинаково быстры, xxh3_32 надёжней. xxh3_32 - это
+    алгоритм xxhash3, обрезанный с 64 до 32 бит (это всё равно хороший хеш).
 
     Следует понимать, что контрольные суммы в зависимости от размера блока их
     расчёта либо увеличивают потребление памяти, либо снижают производительность.

docs/config/src/make.js

@@ -64,7 +64,7 @@ for (const file of params_files)
         let out = '\n';
         for (const c of cfg)
         {
-            out += `\n- [${c.name}](#${c.name})`;
+            out += `\n- [${c.name}](#${c.name.replace(/\./g, '')})`;
         }
         for (const c of cfg)
         {

docs/config/src/monitor.yml

@@ -21,6 +21,9 @@
     cluster, cluster_key, persist_filter, stale_read can also be set in
     Vitastor configuration with `antietcd_` prefix.
 
+    See also: [antietcd_cert](security.en.md#antietcd_cert),
+    [antietcd_key](security.en.md#antietcd_key) and [etcd_proxy](security.en.md#etcd_proxyurls).
+
     You can dump/load data to or from antietcd using Antietcd `anticli` tool:
 
     ```
@@ -47,6 +50,9 @@
     node_id, cluster, cluster_key, persist_filter, stale_read также можно задавать
     в конфигурации Vitastor с префиксом `antietcd_`.
 
+    Смотрите также настройки [antietcd_cert](security.ru.md#antietcd_cert),
+    [antietcd_key](security.ru.md#antietcd_key) и [etcd_proxy](security.ru.md#etcd_proxyurls).
+
     Вы можете выгружать/загружать данные в или из antietcd с помощью его инструмента
     `anticli`:
 

docs/config/src/package.json

@@ -0,0 +1,5 @@
+{
+  "dependencies": {
+    "yaml": "^2.8.2"
+  }
+}

docs/config/src/security.en.md

@@ -0,0 +1,9 @@
+# Security Parameters
+
+These parameters affect your Vitastor installation security and apply to OSDs, monitors and clients.
+
+Most of them can be set in /etc/vitastor/vitastor.conf and in etcd, but don't support online modification.
+
+All certificate and private key parameters mentioned may contain a path to a PEM file or just
+a PEM string with certificate or a private key. In the latter case, the string must begin with
+"-----BEGIN CERTIFICATE-----" or "-----BEGIN PRIVATE KEY-----".

docs/config/src/security.ru.md

@@ -0,0 +1,11 @@
+# Параметры безопасности
+
+Данные параметры затрагивают безопасность инсталляций Vitastor и используются
+OSD, мониторами и клиентами.
+
+Большая их часть может задаваться в /etc/vitastor/vitastor.conf и в etcd, но не
+поддерживает онлайн-изменение.
+
+Все параметры сертификатов и закрытых ключей могут быть путём к файлу или просто
+строкой с сертификатом в формате PEM. В последнем случае строка должна начинаться с
+"-----BEGIN CERTIFICATE-----" или "-----BEGIN PRIVATE KEY-----".

docs/config/src/security.yml

@@ -0,0 +1,276 @@
+- name: use_perms
+  type: bool
+  default: false
+  info: |
+    Enable client permissions in a Vitastor cluster, including Antietcd built into the Monitor.
+    Requires configured encryption. Also note that separate Antietcd requires separate configuration
+    to use permissions (see [security documentation](../intro/security.en.md) for details).
+  info_ru: |
+    Включает клиентские привилегии в кластере Vitastor, в том числе во встроенном в мониторе Antietcd.
+    Требует настроенного шифрования протокола. Также обратите внимание, что отдельно установленный Antietcd
+    требует отдельной настройки привилегий (подробности смотрите в [документации безопасности](../intro/security.ru.md)).
+- name: cert
+  type: string
+  info: |
+    Client certificate of the current Vitastor user. Required for Vitastor protocol encryption.
+    Must be signed with [client_ca](#client_ca). Also used as the client certificate for etcd/Antietcd
+    connections by default.
+  info_ru: |
+    Клиентский сертификат текущего пользователя Vitastor. Требуется для шифрования протокола Vitastor.
+    Должен быть подписан [client_ca](#client_ca). Также по умолчанию используется как клиентский
+    сертификат для подключения к etcd/Antietcd и Vault.
+- name: pkey
+  type: string
+  info: Private key of the current Vitastor user.
+  info_ru: Закрытый ключ текущего пользователя Vitastor.
+- name: etcd_ca
+  type: string
+  info: |
+    Trusted TLS CA to verify etcd server certificate. Or just the etcd server's
+    certificate itself - it's fine to use it for etcd_ca.
+  info_ru: |
+    Доверенный корневой TLS-сертификат для проверки сертификата сервера etcd.
+    Либо же просто сам сертификат сервера etcd - его можно использовать как etcd_ca.
+- name: client_ca
+  type: string
+  info: |
+    Trusted TLS CA to verify Vitastor client certificates.
+    Mandatory for Vitastor protocol encryption.
+  info_ru: |
+    Доверенный TLS-сертификат для проверки сертификатов клиентов Vitastor.
+    Требуется для шифрования протокола Vitastor.
+- name: osd_ca
+  type: string
+  info: |
+    Trusted TLS CA to verify Vitastor OSD certificates. Also mandatory for Vitastor protocol
+    encryption. Must be different from client_ca. May be equal to osd_cert - different OSDs
+    don't require separate certificates at the moment because their permissions don't differ.
+  info_ru: |
+    Доверенный TLS-сертификат для проверки сертификатов OSD Vitastor. Также обязателен
+    для шифрования протокола Vitastor. Должен отличаться от client_ca. Может быть равен
+    osd_cert - разные OSD не требуют разных сертификатов, потому что на данный момент
+    привилегии разных OSD никак не отличаются.
+- name: mon_ca
+  type: string
+  info: |
+    Trusted TLS CA to verify Vitastor Monitor certificates. Used only for separate Antietcd,
+    not required when a monitor built-in Antietcd is used. May be equal to mon_client_etcd_cert.
+  info_ru: |
+    Доверенный TLS-сертификат для проверки сертификатов мониторов Vitastor. Используется
+    только отдельно установленным Antietcd, не требуется при использовании встроенного в монитор
+    Antietcd. Может быть равен mon_client_etcd_cert.
+- name: antietcd_cert
+  type: string
+  info: Server TLS certificate for Antietcd built into the Monitor.
+  info_ru: Серверный TLS-сертификат для Antietcd, встроенного в монитор.
+- name: antietcd_key
+  type: string
+  info: Private key for antietcd_cert.
+  info_ru: Закрытый ключ для сертификата antietcd_cert.
+- name: etcd_proxy.urls
+  type: string or array of strings
+  type_ru: строка или массив строк
+  info: |
+    etcd URLs for Antietcd etcd proxy mode.
+    See [Mon as Etcd proxy](../intro/security.en.md#mon-as-etcd-proxy) for details.
+  info_ru: |
+    Адреса etcd для режима Antietcd etcd-прокси.
+    Смотрите подробности в разделе [Mon в роли Etcd proxy](../intro/security.ru.md#mon-в-роли-etcd-proxy).
+- name: etcd_proxy.cert
+  type: string
+  info: Client certificate for Antietcd connections to etcd in proxy mode.
+  info_ru: Клиентский сертификат для подключений от Antietcd к etcd в режиме прокси.
+- name: etcd_proxy.key
+  type: string
+  info: Private key for etcd_proxy.cert.
+  info_ru: Закрытый ключ для сертификата etcd_proxy.cert.
+- name: etcd_proxy.ca
+  type: string
+  info: Trusted TLS CA to verify etcd server certificate when connecting to it from Antietcd.
+  info_ru: Доверенный TLS-сертификат для проверки сертификата сервера etcd при подключениях от Antietcd.
+- name: osd_cert
+  type: string
+  info: |
+    Vitastor OSD server certificate. Required for Vitastor protocol encryption. May be equal
+    to [osd_ca](#osd_ca) - all OSDs share the same permission set for now. Also used as the client
+    certificate for connections from OSD to etcd/Antietcd by default.
+  info_ru: |
+    Сертификат сервера Vitastor OSD. Требуется для шифрования протокола Vitastor. Может быть равен
+    [osd_ca](#osd_ca) - все OSD на данный момент имеют одинаковые привилегии. Также по умолчанию
+    используется как клиентский сертификат для подключения от OSD к etcd/Antietcd.
+- name: osd_pkey
+  type: string
+  info: Private key for osd_cert.
+  info_ru: Закрытый ключ для сертификата osd_cert.
+- name: api_cert
+  type: string
+  info: Server TLS certificate for [vitastor-cli serve](../usage/cli.en.md#serve) API server.
+  info_ru: Серверный TLS-сертификат для API-сервера [vitastor-cli serve](../usage/cli.ru.md#serve).
+- name: api_pkey
+  type: string
+  info: Private key for api_cert.
+  info_ru: Закрытый ключ для сертификата api_cert.
+- name: etcd_client_cert
+  type: string
+  info: |
+    Client TLS certificate to use for connections from Vitastor clients to etcd/Antietcd if you don't want
+    to use the common client certificate [cert](#cert).
+  info_ru: |
+    Клиентский TLS сертификат для подключений от клиентов Vitastor к etcd/Antietcd, если вы не хотите
+    использовать общий клиентский сертификат [cert](#cert).
+- name: etcd_client_key
+  type: string
+  info: Private key for etcd_client_cert.
+  info_ru: Закрытый ключ для сертификата etcd_client_cert.
+- name: osd_etcd_client_cert
+  type: string
+  info: |
+    Client TLS certificate to use for connections from Vitastor OSDs to etcd/Antietcd if you don't want
+    to use the common OSD certificate [osd_cert](#osd_cert).
+  info_ru: |
+    Клиентский TLS сертификат для подключений от Vitastor OSD к etcd/Antietcd, если вы не хотите
+    использовать общий сертификат OSD [osd_cert](#osd_cert).
+- name: osd_etcd_client_key
+  type: string
+  info: Private key for osd_etcd_client_cert.
+  info_ru: Закрытый ключ для сертификата osd_etcd_client_cert.
+- name: mon_etcd_client_cert
+  type: string
+  info: |
+    Client TLS certificate to use for connections from Vitastor Monitors to etcd/Antietcd - required
+    if you don't use the built-in Antietcd. In case you use it Monitor has direct access to Antietcd data
+    and doesn't require any connection.
+  info_ru: |
+    Клиентский TLS сертификат для подключений от мониторов Vitastor к etcd/Antietcd - требуется, если
+    вы не используете встроенный в монитор Antietcd. Если вы используете его, то монитор и так имеет
+    прямой доступ к данным Antietcd и не требует никаких соединений.
+- name: mon_etcd_client_key
+  type: string
+  info: Private key for mon_etcd_client_cert.
+  info_ru: Закрытый ключ для сертификата mon_etcd_client_cert.
+- name: proto_checksums
+  type: string
+  default: payload
+  info: |
+    One of "full", "payload", "gcm", "none":
+    - "full" means calculate and verify transport level checksums from the full message data
+      including the header - recommended for unencrypted setups.
+    - "payload" enables checksums only for the actual read/write data, but skips them for message
+      headers - recommended for encrypted setups because headers are already protected by AES-GCM.
+    - "gcm" disables checksums and enables AES-GCM encryption of the whole messages including headers
+      and data - AES-GCM already includes MAC which is actually a stronger checksum. This option is
+      slower and is only recommended for untrusted networks.
+    - "none" disables transport level checksums at all.
+  info_ru: |
+    Одно из значений "full", "payload", "gcm" и "none":
+    - "full" означает расчёт и проверку контрольных сумм на транспортном уровне от полных сообщений,
+      включая их заголовки и данные - рекомендуется для кластеров без шифрования.
+    - "payload" включает контрольные суммы только для данных сообщений, но пропускает заголовки -
+      такая настройка рекомендуется для кластеров с включённым шифрованием, потому что в них заголовки
+      и так защищены шифрованием AES-GCM.
+    - "gcm" отключает контрольные суммы и включает шифрование полных сообщений включая заголовки и
+      данные - AES-GCM уже включает в себя MAC, который по сути является криптостойкой контрольной
+      суммой. Такая настройка медленнее и рекомендуется только для недоверенных сетей.
+    - "none" полностью отключает контрольные суммы на транспортном уровне.
+- name: force_proto_checksums
+  type: string
+  info: |
+    To allow older clients to connect to a Vitastor cluster with enabled checksums, Vitastor OSDs
+    allow clients to downgrade their proto_checksums by default. force_proto_checksums sets the
+    minimum security level allowed for connecting clients. When encryption is disabled, default
+    force_proto_checksums is none and clients without checksums are allowed. With enabled
+    encryption, force_proto_checksums becomes "payload" by default to block unauthenticated data
+    on the transport level.
+  info_ru: |
+    Чтобы старые клиенты Vitastor могли подключаться к кластеру с включёнными контрольными
+    суммами, Vitastor OSD по умолчанию разрешают клиентам отключать контрольные суммы
+    данных (proto_checksums). Настройка force_proto_checksums задаёт минимальный уровень
+    безопасности, разрешённый для подключающихся клиентов. Когда шифрование отключено,
+    force_proto_checksums по умолчанию равно none и подключения клиентов без контрольных
+    сумм разрешаются. При включённом шифровании значение по умолчанию force_proto_checksums
+    становится "payload", чтобы блокировать подключения с неаутентифицированными данными.
+- name: max_cipher_pool_size
+  type: int
+  default: 256
+  info: |
+    Maximum number of OpenSSL cipher contexts cached in OSD memory, counted separately
+    for each cipher and for encryption/decryption. Probably doesn't require modification.
+  info_ru: |
+    Максимальное количество кэшируемых в памяти OSD контекстов шифра OpenSSL, учитываемое
+    отдельно для каждого шифра и для шифрования и расшифровки. Вряд ли требует изменения.
+- name: vault_url
+  type: string
+  info: |
+    Vault base URL.
+
+    Vitastor clients support AES-256-XTS image data encryption with different per-image keys.
+    Encryption is performed by the client, OSDs don't have access to decrypted data.
+
+    Encryption keys may be stored in etcd or, for the increased security level, in an external
+    [HashiCorp Vault](https://developer.hashicorp.com/vault/) or [OpenBao](https://openbao.org/)
+    instance.
+
+    Vitastor clients use [v1 k/v secrets engine](https://openbao.org/api-docs/secret/kv/kv-v1/)
+    and [TLS authentication engine](https://openbao.org/api-docs/auth/cert/) in Vault.
+
+    In that case, only key IDs are stored in etcd.
+  info_ru: |
+    Базовый адрес Vault.
+
+    Клиенты Vitastor поддерживают AES-256-XTS шифрование данных образов с отдельными ключами на
+    каждый образ. Данные шифруются клиентами, OSD не имеют доступа к незашифрованным данным.
+
+    Ключи шифрования могут храниться в etcd или, для повышенного уровня безопасности, во внешнем
+    [HashiCorp Vault](https://developer.hashicorp.com/vault/) или [OpenBao](https://openbao.org/).
+
+    Клиенты Vitastor используют [движок секретов v1](https://openbao.org/api-docs/secret/kv/kv-v1/)
+    и [TLS-аутентификацию](https://openbao.org/api-docs/auth/cert/) в Vault.
+
+    В этом случае, только ID ключей хранятся в etcd.
+- name: vault_secret_api_path
+  type: string
+  default: /v1/secret/
+  info: Vault v1 secret API mount path to use.
+  info_ru: Путь к API секретов v1 для использования клиентами.
+- name: vault_client_cert
+  type: string
+  info: |
+    Client TLS certificate to use for Vault connections if you don't want to use the common Vitastor
+    client certificate [cert](#cert) which is also used for Vault connections by default.
+  info_ru: |
+    Клиентский TLS сертификат для подключений к Vault на тот случай, если вы не хотите использовать
+    общий сертификат клиента Vitastor [cert](#cert), используемый для подключений к Vault по умолчанию.
+- name: vault_client_key
+  type: string
+  info: Private key for the vault_client_cert certificate.
+  info_ru: Закрытый ключ для сертификата vault_client_cert.
+- name: vault_ca
+  type: string
+  info: |
+    Trusted TLS CA to verify Vault server certificate. May be path to a file,
+    directory or just a PEM string with certificate.
+  info_ru: |
+    Доверенный корневой TLS-сертификат для проверки сертификата сервера Vault.
+    Может быть путём к файлу, директории или просто строкой с сертификатом в
+    формате PEM.
+- name: vault_timeout_ms
+  type: int
+  default: 5000
+  info: Timeout for Vault requests in milliseconds.
+  info_ru: Максимально время выполнения Vault-запросов в миллисекундах.
+- name: vault_error_timeout_sec
+  type: int
+  default: 60
+  info: |
+    Time (in seconds) to wait before retrying after receiving an error from Vault.
+  info_ru: |
+    Время (в секундах) для ожидания перед повторной попыткой при получении ошибки от Vault.
+- name: vault_refresh_leeway_sec
+  type: int
+  default: 60
+  info: |
+    Extra time (in seconds) before real Vault token lease_timeout to refresh it, just
+    in case of system clock drift.
+  info_ru: |
+    Зазор времени (в секундах), чтобы обновлять токены Vault чуть раньше их реального
+    lease_timeout, на случай "ухода" системных часов.

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:v3.0.13`
+   `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:v3.0.13 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:v3.0.13`
+   `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:v3.0.13 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/source.en.md

@@ -15,7 +15,7 @@
 - gcc and g++ 8 or newer, clang 10 or newer, or other compiler with C++11 plus
   designated initializers support from C++20
 - CMake
-- jerasure headers and libraries
+- jerasure, c-ares headers and libraries
 - ISA-L, libibverbs, librdmacm, libnl3 headers and libraries (optional)
 
 ## Basic instructions

docs/installation/source.ru.md

@@ -15,7 +15,7 @@
 - gcc и g++ >= 8, либо clang >= 10, либо другой компилятор с поддержкой C++11 плюс
   назначенных инициализаторов (designated initializers) из C++20
 - CMake
-- Заголовки и библиотеки jerasure
+- Заголовки и библиотеки jerasure, c-ares
 - Опционально - заголовки и библиотеки ISA-L, libibverbs, librdmacm, libnl3
 
 ## Базовая инструкция

docs/intro/features.en.md

@@ -41,6 +41,8 @@
 - [Built-in Prometheus metric exporter](../config/monitor.en.md#enable_prometheus)
 - [NFS RDMA support](../usage/nfs.en.md#rdma) (probably also usable for GPUDirect)
 - [S3](../installation/s3.en.md)
+- [TLS support for etcd connections](../config/security.en.md)
+- [AES-256-XTS image encryption](../usage/cli.en.md#create) and [Vault support](../config/security.en.md#vault_url) for key storage
 
 ## Plugins and tools
 

docs/intro/features.ru.md

@@ -43,6 +43,8 @@
 - [Встроенный Prometheus-экспортер метрик](../config/monitor.ru.md#enable_prometheus)
 - [Поддержка NFS RDMA](../usage/nfs.ru.md#rdma) (вероятно, также подходящая для GPUDirect)
 - [S3](../installation/s3.ru.md)
+- [Поддержка TLS-соединений с etcd](../config/security.ru.md)
+- [AES-256-XTS шифрование данных](../usage/cli.ru.md#create) и [поддержка Vault](../config/security.ru.md#vault_url) для хранения ключей
 
 ## Драйверы и инструменты
 

docs/intro/quickstart.en.md

@@ -41,12 +41,16 @@
 ## Configure monitors
 
 On the monitor hosts:
-- Put identical etcd_address into `/etc/vitastor/vitastor.conf`. Example:
+- Create minimal configuration in `/etc/vitastor/vitastor.conf`:
   ```
   {
-    "etcd_address": ["10.200.1.10:2379","10.200.1.11:2379","10.200.1.12:2379"]
+    "etcd_address": ["http://10.200.1.10:2379","http://10.200.1.11:2379","http://10.200.1.12:2379"],
+    "osd_network": "10.200.1.0/24",
+    "use_perms": false
   }
   ```
+- Note that you can enable encryption by using `https://` and `use_perms` option.
+  [Details](security.en.md#quick-setup) about encryption setup with make-etcd.
 - Create systemd units for etcd by running: `/usr/lib/vitastor/mon/make-etcd`
   Or, if you installed Vitastor in Docker, run `systemctl start vitastor-host; docker exec vitastor make-etcd`.
 - Start etcd and monitors: `systemctl enable --now vitastor-etcd vitastor-mon`

docs/intro/quickstart.ru.md

@@ -41,25 +41,23 @@
 ## Настройте мониторы
 
 На хостах, выделенных под мониторы:
-- Пропишите одинаковые etcd_address в `/etc/vitastor/vitastor.conf`. Например:
+- Создайте минимальную конфигурацию в `/etc/vitastor/vitastor.conf`:
   ```
   {
-    "etcd_address": ["10.200.1.10:2379","10.200.1.11:2379","10.200.1.12:2379"]
+    "etcd_address": ["http://10.200.1.10:2379","http://10.200.1.11:2379","http://10.200.1.12:2379"],
+    "osd_network": "10.200.1.0/24",
+    "use_perms": false
   }
   ```
+- Обратите внимание, что с помощью схемы `https://` и опции `use_perms` можно включить шифрование.
+  [Подробно](security.ru.md#быстрая-настройка) о настройке шифрования через make-etcd.
 - Инициализируйте сервисы etcd, запустив `/usr/lib/vitastor/mon/make-etcd`.\
   Либо, если вы установили Vitastor в Docker, запустите `systemctl start vitastor-host; docker exec vitastor make-etcd`.
 - Запустите etcd и мониторы: `systemctl enable --now vitastor-etcd vitastor-mon`
 
 ## Настройте OSD
 
-- Пропишите etcd_address и [osd_network](../config/network.ru.md#osd_network) в `/etc/vitastor/vitastor.conf`. Например:
-  ```
-  {
-    "etcd_address": ["10.200.1.10:2379","10.200.1.11:2379","10.200.1.12:2379"],
-    "osd_network": "10.200.1.0/24"
-  }
-  ```
+- Создайте/скопируйте с узлов с мониторами файл конфигурации `/etc/vitastor/vitastor.conf`.
 - Инициализуйте OSD:
   - Только SSD или только HDD: `vitastor-disk prepare /dev/sdXXX [/dev/sdYYY ...]`.
     Если вы используете десктопные SSD без конденсаторов, добавьте опцию `--disable_data_fsync off`,

docs/intro/security.en.md

@@ -0,0 +1,657 @@
+[Documentation](../../README.md#documentation) → Introduction → Security in Vitastor
+
+-----
+
+[Читать на русском](security.ru.md)
+
+# Security in Vitastor
+
+- [Overview](#overview)
+- [Quick setup](#quick-setup)
+- Principles of operation
+  - [etcd transport encryption (TLS)](#etcd-transport-encryption-tls)
+  - [OSD transport encryption (AES-GCM)](#osd-transport-encryption-aes-gcm)
+  - [End-to-end image data encryption (AES-XTS)](#end-to-end-image-data-encryption-aes-xts)
+  - [Certificate-based authentication](#certificate-based-authentication)
+  - [Users and access rights](#users-and-access-rights)
+  - [etcd privileges](#etcd-privileges)
+- Manual setup
+  - [Configuring OSD transport encryption](#configuring-osd-transport-encryption)
+  - etcd/Antietcd setup options
+    - [Mon with embedded Antietcd](#mon-with-embedded-antietcd)
+    - [Mon as an Etcd proxy](#mon-as-an-etcd-proxy)
+    - [Mon with a separate Antietcd Proxy](#mon-with-a-separate-antietcd-proxy)
+    - [Standalone Antietcd without etcd](#standalone-antietcd-without-etcd)
+  - [Vault/OpenBao setup](#vaultopenbao-setup)
+    - [Vault setup example](#vault-setup-example)
+- Lists of allowed operations
+  - [etcd data access rights](#etcd-data-access-rights)
+  - [OSD data access rights](#osd-data-access-rights)
+  - [API access rights](#api-access-rights)
+- [Encryption performance](#encryption-performance)
+
+## Overview
+
+Starting from version 3.1.0, Vitastor provides full data protection:
+control plane protection (etcd), data plane protection (OSDs), and end-to-end data encryption.
+
+- Control plane protection:
+  - etcd transport encryption (TLS)
+  - Authentication via client TLS (X.509) certificates
+  - Access control of clients to etcd data
+- Data plane protection:
+  - Full AES-GCM encryption of OSD transport (similar to TLS, but faster)
+  - Alternatively, AES-GCM encryption of just operation headers with data checksums using a secret "salt"
+  - Authentication via client TLS (X.509) certificates
+  - Access control of clients on the OSD side
+- End-to-end encryption:
+  - Data is encrypted using AES-XTS on the client side, the Vitastor cluster has no access to plaintext data
+  - AES-XTS keys can be stored in etcd or in an external Vault/OpenBao
+
+All features are optional and disabled in the simplest configuration. By default, only
+transport-level data checksums ([proto_checksums](../config/security.en.md#proto_checksums)=payload)
+are enabled for clients that support them (>= 3.1.0). For older clients, connections
+without data checksums are allowed by default ([force_proto_checksums](../config/security.en.md#force_proto_checksums) is empty).
+
+For a quick setup, jump to the [Quick setup](#quick-setup) section.
+
+Descriptions of all security-related parameters can be found [here](../config/security.en.md).
+
+## Quick setup
+
+For a quick setup, use the `/usr/lib/vitastor/mon/make-etcd` script:
+
+1. Log in to the node where the first monitor and etcd will be located.
+2. Create `/etc/vitastor/vitastor.conf` with minimal parameters: etcd_address,
+   osd_network and, if you want to enable privileges, use_perms (note `https://`
+   in etcd addresses):
+   ```
+   {
+       "etcd_address": ["https://10.0.0.10:2379","https://10.0.0.11:2379","https://10.0.0.12:2379"],
+       "osd_network": "10.0.0.0/24",
+       "use_perms": true
+   }
+   ```
+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 (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.
+
+If you want to understand the setup in more detail, read the [Principles of operation](#principles-of-operation)
+and [Manual setup](#manual-setup) sections below.
+
+## Principles of operation
+
+### etcd transport encryption (TLS)
+
+Possible setups:
+- Without encryption (http)
+- With encryption (https)
+- With encryption and client certificate authentication. Either the same certificate
+  used for authentication on the OSD side (`cert`+`pkey` / `osd_cert`+`osd_pkey`)
+  is used, or a separately specified certificate (`etcd_client_cert`+`etcd_client_key`).
+
+### OSD transport encryption (AES-GCM)
+
+Possible setups:
+- Unencrypted transport without checksums: `proto_checksums=none`.
+- Unencrypted transport with data checksums: `proto_checksums=payload` (may be omitted,
+  this is the default value). It's allowed to disable checksums on the client side, or
+  use an older client that does not support checksums. If you want to block connections
+  from clients without checksums, use the option `force_proto_checksums=payload`.
+- Header-only encryption with data checksums: activated when the options
+  `cert`, `pkey`, `osd_ca` are set on the client side and `osd_cert`, `osd_pkey`, `osd_ca`, `client_ca`
+  on the OSD side, with `proto_checksums=payload`. In this mode, disabling checksums on the client
+  side is forbidden by default, i.e. `force_proto_checksums=payload` is used.
+- Full transport encryption of all traffic: same as the previous option, but with `proto_checksums=gcm`.
+  In this case, clients are by default allowed to downgrade to checksums only, but this
+  can also be forbidden via `force_proto_checksums=gcm`. This is the slowest setup and
+  it's only recommended for insecure (public) networks. In particular, full traffic
+  encryption together with end-to-end AES-XTS image encryption encrypts data twice.
+
+Encryption uses the AES-256-GCM algorithm and a custom simplified key exchange protocol,
+fully analogous to TLS 1.3 ECDHE.
+
+### End-to-end image data encryption (AES-XTS)
+
+The Vitastor client supports encrypting each image's data with its own key. In this case,
+data is encrypted by the client before sending it to OSDs and OSDs can't see it in plain.
+The encryption key can be changed when cloning/creating image snapshots. For example,
+you can make a base VM image (say, Debian Linux) unencrypted, but have encrypted client VM
+images inheriting from it.
+
+Image encryption keys can be stored in etcd or in an external Vault. In the latter case,
+etcd only stores key IDs and Vitastor cluster can't decrypt the data at all. To use
+Vault, create an image with the `--enc_key vault:ID` option, specify vault_url and vault_ca
+options in the configuration, create accounts for all clients in Vault, and grant them access
+to the required v1 secrets.
+
+Once again, if AES-XTS is used together with full traffic encryption (`proto_checksums=gcm`),
+image data is encrypted twice — first with AES-XTS, and then with AES-GCM. Use it only if
+you are completely paranoid :-).
+
+### Certificate-based authentication
+
+When encryption is enabled, Vitastor clients, OSDs, and monitors authenticate via certificates
+for both etcd (Antietcd) and OSD connections.
+
+Separate certificates must be used for OSDs and monitors — either self-signed, or signed
+by separate CAs (`osd_ca` and `mon_ca`). All OSDs can use the same certificate, and all
+monitors can also use the same certificate, since the privileges of different OSDs or
+different monitors do not differ (theoretically, one could differentiate OSD certificates
+by pool, but there has been no need for this so far).
+
+Also, a monitor certificate may not be needed at all if Antietcd is embedded into the monitor
+itself. In this case, the monitor already has access to all etcd data directly in memory.
+
+### Users and access rights
+
+When transport encryption is disabled, Vitastor operates without access control, i.e.,
+any cluster client has full access to both the management layer and the data layer. This
+option is suitable for dedicated trusted storage networks.
+
+When OSD transport encryption is enabled (at least for headers), you can enable access
+rights by turning on the `use_perms=true` option. When this option is enabled, each user
+can perform only the operations that they are permitted, and even OSDs and monitors are
+also forbidden from performing "unnecessary" operations.
+
+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 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
+
+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
+
+For the list of allowed operations on image data on the OSD side, see the
+[OSD data access rights](#osd-data-access-rights) section.
+
+### etcd privileges
+
+etcd privileges are implemented through Antietcd in all modes of operation.
+
+Built-in etcd privileges are not supported due to numerous inconveniences:
+- Certificate-based authentication does not work at all in etcd's REST interface,
+- Privileges are stored separately from k/v data and cannot participate in transactions,
+- Only the administrator (root) can change privileges,
+- There is no support for filtering range read responses by privileges.
+
+If etcd is used, Antietcd acts as a filtering proxy and can be embedded in the Vitastor
+monitor or run separately. In this case, etcd must allow incoming connections only from
+Antietcd, and all other components must connect to Antietcd.
+
+If Antietcd runs as a part of the Vitastor monitor, it is sufficient to enable the option
+`use_perms=true` and set the required certificates. If Antietcd is run separately, privileges
+have to be enabled separately using Antietcd options. For more details on the setup, see
+the [etcd/Antietcd setup options](#etcdantietcd-setup-options) section.
+
+For the list of allowed operations with etcd data, see the
+[etcd data access rights](#etcd-data-access-rights) section.
+
+## Manual setup
+
+### Configuring OSD transport encryption
+
+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
+- osd_cert: osd_ca.crt
+- osd_pkey: osd_ca.key
+
+On the client side:
+- use_perms: true
+- cert: client.crt
+- pkey: client.key
+
+### etcd/Antietcd setup options
+
+The following configuration options are available:
+
+#### Mon with embedded Antietcd
+
+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)
+- use_perms: true
+- use_antietcd: true
+- antietcd_cert: antietcd.crt
+- antietcd_key: antietcd.key
+- etcd_ca: antietcd.crt
+- osd_ca: osd_ca.crt
+- client_ca: client_ca.crt
+
+#### Mon as an Etcd proxy
+
+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)
+- use_perms: true
+- use_antietcd: true
+- etcd_proxy:
+  ```
+  {
+    "urls": [ "http://mon1:2381", ... ], // addresses of your etcd with port 2381
+    "cert": "antietcd.crt",
+    "key": "antietcd.key",
+    "ca": "etcd.crt"
+  }
+  ```
+- antietcd_cert: antietcd.crt
+- antietcd_key: antietcd.key
+- etcd_ca: antietcd.crt
+- osd_ca: osd_ca.crt
+- client_ca: client_ca.crt
+
+etcd command-line options:
+```
+--advertise-client-urls=https://<ADDRESS>:2381 --listen-client-urls=https://<ADDRESS>:2381 \
+--client-cert-auth --cert-file=etcd.crt --key-file=etcd.key --trusted-ca-file=antietcd.crt \
+--peer-client-cert-auth --peer-cert-file=etcd.crt --peer-key-file=etcd.key --peer-trusted-ca-file=etcd.crt
+```
+
+#### Mon with a separate Antietcd Proxy
+
+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 separate certificates for clients, OSDs, and monitors will be needed.
+
+Vitastor settings:
+- etcd_address: [ "http://mon1:2379", ... ] (addresses of your monitors with port 2379)
+- use_perms: true
+- use_antietcd: false
+- etcd_ca: antietcd.crt
+- osd_ca: osd_ca.crt
+- client_ca: client_ca.crt
+- mon_etcd_client_cert: mon_ca.crt
+- mon_etcd_client_key: mon_ca.key
+
+Antietcd command-line options:
+```
+--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):
+```
+--advertise-client-urls=https://<ADDRESS>:2381 --listen-client-urls=https://<ADDRESS>:2381 \
+--client-cert-auth --cert-file=etcd.crt --key-file=etcd.key --trusted-ca-file=antietcd.crt \
+--peer-client-cert-auth --peer-cert-file=etcd.crt --peer-key-file=etcd.key --peer-trusted-ca-file=etcd.crt
+```
+
+#### Standalone Antietcd without etcd
+
+Same as the previous option, but etcd and its certificate are not needed:
+
+Vitastor settings (same as in the previous option):
+- etcd_address: [ "http://mon1:2379", ... ] (addresses of your monitors with port 2379)
+- use_perms: true
+- use_antietcd: false
+- etcd_ca: antietcd.crt
+- osd_ca: osd_ca.crt
+- client_ca: client_ca.crt
+- mon_etcd_client_cert: mon_ca.crt
+- mon_etcd_client_key: mon_ca.key
+
+Antietcd command-line options:
+```
+--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
+
+To use Vault, each client that needs to get image keys from Vault needs a Vault account.
+Vitastor only supports client certificate-based authentication, so all client certificates
+(`cert`+`pkey`) must be registered in Vault, and they must be granted access to the
+corresponding secrets (v1 secrets API is supported).
+
+The required format of a Vault secret is a single `key` field as a hexadecimal string.
+The AES-256-XTS algorithm is used, so the key length is 64 bytes, i.e., the string must
+consist of 128 hexadecimal digits.
+
+To connect to Vault, set the following settings in Vitastor.conf:
+- `vault_url` — Vault address (e.g., `https://vault:8200`)
+- `vault_ca` — Vault's own certificate
+
+After that, if you create an image (`vitastor-cli create`) with the option `--enc_key vault:<ID>`,
+Vitastor clients will first contact Vault to obtain a token at `/v1/auth/cert/login`,
+and then request the actual secret from Vault at `/v1/secret/<ID>`.
+
+#### Vault setup example
+
+Step-by-step instructions for setting up a test Vault using OpenBao as an example:
+
+1. If TLS is not yet configured, generate a self-signed TLS certificate for Vault:
+   ```
+   openssl req -days 3650 -x509 -addext basicConstraints=critical,CA:TRUE,pathlen:1 --addext subjectAltName=DNS:vault \
+       -new -newkey rsa:4096 -nodes -keyout /etc/openbao/vault.key -out /etc/openbao/vault.crt
+   ```
+   Configure it in `/etc/openbao/openbao.hcl`:
+   ```
+   listener "tcp" {
+       address       = "0.0.0.0:8200"
+       tls_cert_file = "/etc/openbao/vault.crt"
+       tls_key_file  = "/etc/openbao/vault.key"
+   }
+   ```
+   And restart OpenBao (`systemctl restart openbao`).
+2. Copy Vault's TLS certificate for Vitastor:
+   ```
+   cp /etc/openbao/vault.crt /etc/vitastor/vault.crt
+   ```
+   Transfer it to all client nodes and specify it in `/etc/vitastor/vitastor.conf`:
+   ```
+   {
+       ...
+       "vault_url": "http://vault:8200",
+       "vault_ca": "/etc/vitastor/vault.crt"
+   }
+   ```
+3. Check Vault status:
+   ```
+   bao status -ca-cert /etc/openbao/vault.crt -address=https://vault:8200
+   ```
+4. Initialize Vault in test mode from 1 node (with 1 key share):
+   ```
+   bao operator init -n 1 -t 1 -ca-cert /etc/openbao/vault.crt -address=https://vault:8200
+   ```
+5. Unseal Vault:
+   ```
+   bao operator unseal -ca-cert /etc/openbao/vault.crt -address=https://vault:8200
+   ```
+6. Enable certificate-based authentication:
+   ```
+   bao auth enable -ca-cert /etc/openbao/vault.crt -address=https://vault:8200 cert
+   ```
+7. Enable v1 secrets:
+   ```
+   bao secrets enable -ca-cert /etc/openbao/vault.crt -address=https://vault:8200 -path=secret kv-v1
+   ```
+8. Create a test secret:
+   ```
+   bao kv put -ca-cert /etc/openbao/vault.crt -address=https://vault:8200 secret/vitastor/testimg3 key=$(openssl rand -hex 64)
+   ```
+9. Generate a signed certificate for a Vitastor user (on a machine where you have `client_ca.crt` and `client_ca.key`):
+   ```
+   openssl req -subj '/CN=testimg3' -nodes -new -keyout testimg3.key -out testimg3.csr
+   openssl x509 -req -days 3650 -CA client_ca.crt -CAkey client_ca.key -CAcreateserial -in testimg3.csr -out testimg3.crt
+   rm testimg3.csr
+   ```
+10. Create a user in Vault and grant it access to the secret:
+    ```
+    cat >testimg3.policy <<EOF
+    path "/secret/vitastor/testimg3" {
+        capabilities = ["read"]
+    }
+    EOF
+
+    bao policy write -ca-cert /etc/openbao/vault.crt -address=https://vault:8200 testimg3 testimg3.policy
+
+    bao write -ca-cert /etc/openbao/vault.crt -address=https://vault:8200 auth/cert/certs/testimg3 \
+        certificate=@testimg3.crt display_name=testimg3 token_ttl=24h token_policies=testimg3
+    ```
+11. Test access to the secret:
+    ```
+    curl --cacert /etc/vitastor/vault.crt --cert testimg3.crt --key testimg3.key \
+        --json '{}' https://vault:8200/v1/auth/cert/login
+    ```
+    A token will be printed, substitute it into the following request:
+    ```
+    curl --cacert /etc/vitastor/vault.crt --cert testimg3.crt --key testimg3.key \
+        -H 'X-Vault-Token: <RECEIVED TOKEN>' https://vault:8200/v1/secret/vitastor/testimg3
+    ```
+12. Create an image in Vitastor with the given secret (as an administrator or someone who
+    has the right to create images in your pool):
+    ```
+    vitastor-cli create -s 100G --enc_key vault:vitastor/testimg3 --owner testimg3 testimg3
+    ```
+13. Test access to the image as user testimg3:
+    ```
+    vitastor-cli --cert testimg3.crt --pkey testimg3.key dd if=/dev/urandom oimg=testimg3 bs=1M count=100
+    ```
+
+## Lists of allowed operations
+
+### etcd data access rights
+
+Below, all key names are given without the common prefix `/vitastor`.
+
+Allowed operations with keys in Antietcd for clients (`type=client`):
+- Read-only:
+  - Always allowed:
+    - `/config/global`
+    - `/config/node_placement`
+    - `/config/pools`
+    - `/pg/config`
+    - `/osd/state/*`
+    - `/pg/state/*`
+    - `/index/maxid/*`
+  - For images [readable by the user](#users-and-access-rights):
+    - `/config/inode/*`
+    - `/index/image/*`
+    - `/inode/stats/*`
+- Read and write:
+  - For pools in which the user can create images:
+    - `/index/maxid/*`
+  - For images owned by the user:
+    - `/config/inode/*`
+    - `/index/image/*`
+
+Allowed operations with keys in Antietcd for administrators (`type=admin`):
+- Read:
+  - `/stats`
+  - `/mon/*`
+  - `/pg/*`
+  - `/pgstats/*`
+  - `/inode/stats/*`
+  - `/pool/stats/*`
+- Read and write:
+  - `/config/*`
+  - `/osd/*`
+  - `/index/*`
+  - `/pg/history/*`
+
+Allowed operations with keys in etcd for OSDs:
+- Read:
+  - `/pg/config`
+  - `/config/*`
+- Read and write:
+  - `/osd/*`
+  - `/pg/state/*`
+  - `/pg/history/*`
+  - `/pgstats/*`
+
+Allowed operations with keys in etcd for monitors:
+- Read:
+  - `/config/*`
+  - `/osd/*`
+  - `/pgstats/*`
+- Read and write:
+  - `/pg/config`
+  - `/stats`
+  - `/history/last_clean_pgs`
+  - `/mon/*`
+  - `/pg/history/*`
+  - `/inode/stats/*`
+  - `/pool/stats/*`
+
+### OSD data access rights
+
+When the `use_perms` option and encryption are enabled, OSDs authenticate clients via
+certificates and allow each client only what is allowed by the access control model.
+
+Client operations:
+- READ — allowed for images the user has read access to.
+- WRITE, DELETE, SCRUB — allowed for images the user has write access to.
+- SYNC — the operation is not tied to an image and is always allowed.
+- DESCRIBE — the operation is allowed only for administrators (used by the commands
+  `vitastor-cli describe` and `fix`).
+- PING — the operation is always allowed.
+- SHOW_CONFIG — the operation is always allowed, however, if the client presents
+  itself as an OSD in it, then it is verified that it uses a certificate signed by `osd_ca`.
+- SEC_LIST (listing) — allowed for other OSDs and administrators with any parameters,
+  and for regular clients only allowed for requests limited to an image the user has
+  read access to.
+
+Cluster operations — allowed only for other OSDs:
+- SEC_READ
+- SEC_WRITE
+- SEC_WRITE_STABLE
+- SEC_SYNC
+- SEC_STABILIZE
+- SEC_ROLLBACK
+- SEC_DELETE
+- SEC_READ_BMP
+- SEC_LOCK
+
+### API access rights
+
+[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 `server_cert` with the key `server_pkey` is used as the server certificate.
+
+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 clients, when accessing the API, are only allowed API operations on images
+available to them either for reading (for reads) or for writing (for modification).
+All other API calls are allowed only for administrators.
+
+List of allowed API operations:
+
+Clients (users with `type=client`) are allowed the following operations:
+- 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 (`type=admin`).
+
+## Encryption performance
+
+You may wonder — how fast is all this wonderful encryption?
+
+The answer is — it depends heavily on the CPU. On modern processors (with AVX512 with VAES
+support) it is very fast — AES encryption speed can reach 10-20 GB/s and above. This
+primarily concerns the CPU of client machines, because end-to-end encryption is performed
+entirely on the client, and client uses its signle thread for transport encryption too,
+while there are many OSDs on the server side, and it is easier to add resources there.
+
+On older processors, the speed is noticeably worse — for example, on a Xeon E5 v4 it is
+only 3 GB/s.
+
+You can evaluate the performance of your processors using the `vitastor-cli cpubench` command.
+
+Example output (💪 AMD EPYC 9575F):
+
+```
+$ vitastor-cli cpubench
+Vitastor transport encryption benchmark (AES-256-GCM, AES-256-XTS and xxhash3)
+
+Warmup...
+
+No transport encryption, data checksums enabled, e2e unencrypted image
+xxhash3 1 M block... 209000 iterations in 2001 ms = 104447.78 MB/s
+xxhash3 4 K block... 37000000 iterations in 2022 ms = 71479.35 MB/s
+
+Header encryption with payload checksums, e2e unencrypted image
+AES-256-GCM encrypt header + xxhash3 1 M block... 210000 iterations in 2015 ms = 104218.36 MB/s
+AES-256-GCM encrypt header + xxhash3 4 K block... 26000000 iterations in 2073 ms = 48993.01 MB/s
+
+Full transport encryption, e2e unencrypted image
+AES-256-GCM encrypt header and 1 M block... 54000 iterations in 2000 ms = 27000.00 MB/s
+AES-256-GCM encrypt header and 4 K block... 11700000 iterations in 2014 ms = 22692.71 MB/s
+
+No transport encryption, no checksums, e2e encrypted image
+AES-256-XTS encrypt 1 M block... 50000 iterations in 2039 ms = 24521.82 MB/s
+AES-256-XTS encrypt 4 K block... 12600000 iterations in 2009 ms = 24499.13 MB/s
+
+No transport encryption, e2e encrypted image, data checksums enabled
+AES-256-XTS encrypt + xxhash3 1 M block... 40000 iterations in 2013 ms = 19870.84 MB/s
+AES-256-XTS encrypt + xxhash3 4 K block... 10200000 iterations in 2011 ms = 19812.90 MB/s
+
+Header encryption with payload checksums, e2e encrypted image
+AES-256-GCM encrypt header + AES-256-XTS encrypt + xxhash3 1 M block... 40000 iterations in 2014 ms = 19860.97 MB/s
+AES-256-GCM encrypt header + AES-256-XTS encrypt + xxhash3 4 K block... 8700000 iterations in 2011 ms = 16899.24 MB/s
+
+Full transport encryption, e2e encrypted image
+AES-256-XTS + AES-256-GCM encrypt 1 M block... 26000 iterations in 2062 ms = 12609.12 MB/s
+AES-256-XTS + AES-256-GCM encrypt 4 K block... 6300000 iterations in 2006 ms = 12267.88 MB/s
+```
+
+And here is Xeon E5-2680v4:
+
+```
+$ vitastor-cli cpubench
+Vitastor transport encryption benchmark (AES-256-GCM, AES-256-XTS and xxhash3)
+
+Warmup...
+
+No transport encryption, data checksums enabled, e2e unencrypted image
+xxhash3 1 M block... 62000 iterations in 2021 ms = 30677.88 MB/s
+xxhash3 4 K block... 12400000 iterations in 2006 ms = 24146.31 MB/s
+
+Header encryption with payload checksums, e2e unencrypted image
+AES-256-GCM encrypt header + xxhash3 1 M block... 62000 iterations in 2027 ms = 30587.07 MB/s
+AES-256-GCM encrypt header + xxhash3 4 K block... 6800000 iterations in 2011 ms = 13208.60 MB/s
+
+Full transport encryption, e2e unencrypted image
+AES-256-GCM encrypt header and 1 M block... 7000 iterations in 2317 ms = 3021.15 MB/s
+AES-256-GCM encrypt header and 4 K block... 1500000 iterations in 2102 ms = 2787.52 MB/s
+
+No transport encryption, no checksums, e2e encrypted image
+AES-256-XTS encrypt 1 M block... 7000 iterations in 2317 ms = 3021.15 MB/s
+AES-256-XTS encrypt 4 K block... 1600000 iterations in 2088 ms = 2993.30 MB/s
+
+No transport encryption, e2e encrypted image, data checksums enabled
+AES-256-XTS encrypt + xxhash3 1 M block... 6000 iterations in 2188 ms = 2742.23 MB/s
+AES-256-XTS encrypt + xxhash3 4 K block... 1400000 iterations in 2053 ms = 2663.78 MB/s
+
+Header encryption with payload checksums, e2e encrypted image
+AES-256-GCM encrypt header + AES-256-XTS encrypt + xxhash3 1 M block... 6000 iterations in 2190 ms = 2739.73 MB/s
+AES-256-GCM encrypt header + AES-256-XTS encrypt + xxhash3 4 K block... 1300000 iterations in 2101 ms = 2417.00 MB/s
+
+Full transport encryption, e2e encrypted image
+AES-256-XTS + AES-256-GCM encrypt 1 M block... 4000 iterations in 2666 ms = 1500.38 MB/s
+AES-256-XTS + AES-256-GCM encrypt 4 K block... 800000 iterations in 2113 ms = 1478.94 MB/s
+```

docs/intro/security.ru.md

@@ -0,0 +1,662 @@
+[Документация](../../README-ru.md#документация) → Введение → Безопасность в Vitastor
+
+-----
+
+[Read in English](security.en.md)
+
+# Безопасность в Vitastor
+
+- [Обзор](#обзор)
+- [Быстрая настройка](#быстрая-настройка)
+- Принципы работы
+  - [Шифрование соединений с etcd (TLS)](#шифрование-соединений-с-etcd-tls)
+  - [Шифрование соединений с OSD (AES-GCM)](#шифрование-соединений-с-osd-aes-gcm)
+  - [Сквозное шифрование данных образов (AES-XTS)](#сквозное-шифрование-данных-образов-aes-xts)
+  - [Аутентификация по сертификатам](#аутентификация-по-сертификатам)
+  - [Пользователи и права доступа](#пользователи-и-права-доступа)
+  - [Привилегии etcd](#привилегии-etcd)
+- Ручная настройка
+  - [Настройка шифрования соединений OSD](#настройка-шифрования-соединений-osd)
+  - Варианты настройки etcd/Antietcd
+    - [Mon со встроенным Antietcd](#mon-со-встроенным-antietcd)
+    - [Mon в роли Etcd proxy](#mon-в-роли-etcd-proxy)
+    - [Mon с отдельным Antietcd Proxy](#mon-с-отдельным-antietcd-proxy)
+    - [Отдельный Antietcd без etcd](#отдельный-antietcd-без-etcd)
+  - [Настройка Vault/OpenBao](#настройка-vaultopenbao)
+    - [Пример настройки Vault](#пример-настройки-vault)
+- Списки разрешённых операций
+  - [Права доступа к данным etcd](#права-доступа-к-данным-etcd)
+  - [Права доступа к данным OSD](#права-доступа-к-данным-osd)
+  - [Права доступа к API](#права-доступа-к-api)
+- [Производительность шифрования](#производительность-шифрования)
+
+## Обзор
+
+Начиная с версии 3.1.0, Vitastor предоставляет полную защиту данных: защиту слоя
+управления (etcd), защиту слоя данных (OSD) и сквозное шифрование данных.
+
+- Защита слоя управления:
+  - Шифрование соединений с etcd (TLS)
+  - Аутентификация по клиентским TLS (X.509) сертификатам
+  - Разграничение прав доступа клиентов к данным etcd
+- Защита слоя данных:
+  - Либо полное AES-GCM шифрование соединений с OSD (аналогично TLS, но быстрее)
+  - Либо шифрование AES-GCM только заголовков команд с контрольными суммами данных с секретной "солью"
+  - Аутентификация по клиентским TLS (X.509) сертификатам
+  - Разграничение прав доступа клиентов на стороне OSD
+- Сквозное шифрование:
+  - Данные шифруются AES-XTS на стороне клиента, кластер Vitastor не имеет доступа к открытым данным
+  - Ключи AES-XTS могут храниться в etcd или во внешнем Vault/OpenBao
+
+Все функции опциональны и в простейшем варианте настройки выключены. По умолчанию включены
+только контрольные суммы данных на транспортном уровне ([proto_checksums](../config/security.ru.md#proto_checksums)=payload) для
+поддерживающих их клиентов (>= 3.1.0). Для более старых клиентов по умолчанию разрешены
+соединения без контрольных сумм данных ([force_proto_checksums](../config/security.ru.md#force_proto_checksums) пусто).
+
+Для быстрой настройки перейдите к разделу [Быстрая настройка](#быстрая-настройка).
+
+Описания всех параметров, связанных с безопасностью, читайте [здесь](../config/security.ru.md).
+
+## Быстрая настройка
+
+Для быстрой настройки используйте скрипт `/usr/lib/vitastor/mon/make-etcd`:
+
+1. Зайдите на узел, на котором будет располагаться первый монитор и etcd.
+2. Создайте там минимальный `/etc/vitastor/vitastor.conf` с параметрами etcd_address,
+   osd_network и, если хотите включить привилегии - use_perms (обратите внимание на `https://`
+   в адресах etcd):
+   ```
+   {
+       "etcd_address": ["https://10.0.0.10:2379","https://10.0.0.11:2379","https://10.0.0.12:2379"],
+       "osd_network": "10.0.0.0/24",
+       "use_perms": true
+   }
+   ```
+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 (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.
+
+Если хотите разобраться в настройке подробнее, читайте далее разделы [Принципы работы](#принципы-работы)
+и [Ручная настройка](#ручная-настройка).
+
+## Принципы работы
+
+### Шифрование соединений с etcd (TLS)
+
+Варианты настройки:
+- Без шифрования (http)
+- С шифрованием (https)
+- С шифрованием и аутентификацией по клиентским сертификатам. Используется либо тот
+  же сертификат, что используется для аутентификации на стороне OSD (`cert`+`pkey` / `osd_cert`+`osd_pkey`),
+  либо отдельно указанный сертификат (`etcd_client_cert`+`etcd_client_key`)
+
+### Шифрование соединений с OSD (AES-GCM)
+
+Варианты настройки:
+- Без шифрования и без контрольных сумм: `proto_checksums=none`.
+- Без шифрования, с контрольными суммами данных: `proto_checksums=payload` (можно не указывать,
+  т.к. это значение по умолчанию). При этом контрольные суммы можно отключить на стороне
+  клиента либо использовать более старые версии клиента, не поддерживающие контрольные суммы.
+  Если нужно запретить подключение клиентов без контрольных сумм, можно использовать опцию
+  `force_proto_checksums=payload`.
+- С шифрованием заголовков и контрольными суммами данных: активируется при установленных опциях
+  `cert`, `pkey`, `osd_ca` на стороне клиента и `osd_cert`, `osd_pkey`, `osd_ca`, `client_ca`
+  на стороне OSD, при `proto_checksums=payload`. При этом по умолчанию запрещается
+  отключение контрольных сумм на уровне клиента, то есть используется `force_proto_checksums=payload`.
+- С полным шифрованием всего трафика: аналогично прошлому варианту, но с `proto_checksums=gcm`.
+  Клиенту при этом по умолчанию разрешается понизить уровень защиты до контрольных сумм, но
+  это тоже можно запретить через `force_proto_checksums=gcm`. Данный вариант самый медленный и
+  рекомендуется только для небезопасных (публичных) сетей. В том числе потому, что при использовании
+  и полного шифрования трафика, и сквозного шифрования образов AES-XTS, данные шифруются дважды.
+
+Для шифрования используется алгоритм AES-256-GCM и собственный упрощённый протокол согласования
+ключей, полностью аналогичный TLS 1.3 ECDHE.
+
+### Сквозное шифрование данных образов (AES-XTS)
+
+Клиент Vitastor поддерживает шифрование данных каждого образа своим ключом. В этом случае на OSD
+уходят уже зашифрованные данные и сами OSD не видят исходные данные клиента. При этом ключ можно
+менять при клонировании/создании снимков образов. Например, можно сделать базовый образ ВМ
+(условный Debian Linux) нешифрованным, но наследовать от него шифрованные образы клиентских ВМ.
+
+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
+к требуемым секретам v1.
+
+Ещё раз повторимся, что если AES-XTS используется с полным шифрованием трафика (`proto_checksums=gcm`),
+то данные образов шифруются дважды - сначала AES-XTS, а потом AES-GCM. Можете использовать,
+только если вы совсем параноик :-).
+
+### Аутентификация по сертификатам
+
+При включённом шифровании клиенты, OSD и мониторы Vitastor аутентифицируются по сертификатам
+как при соединениях с etcd (Antietcd), так и с OSD.
+
+Для OSD и мониторов должны использоваться отдельные сертификаты - либо самоподписанные, либо
+подписанные отдельными CA (`osd_ca` и `mon_ca`). При этом все OSD могут использовать один и
+тот же сертификат и все мониторы тоже могут использовать один и тот же сертификат, так как
+привилегии разных OSD или разных мониторов ничем не отличаются (теоретически можно было бы
+сделать разграничение сертификатов OSD по пулам, но пока что такой необходимости не было).
+
+Также сертификат монитора может быть вообще не нужен, если Antietcd встраивается в сам монитор.
+В этом случае монитор и так имеет доступ ко всем данным etcd прямо в памяти.
+
+### Пользователи и права доступа
+
+При отключённом шифровании трафика Vitastor работает без разграничения прав доступа, то есть,
+любой клиент кластера имеет полный доступ как к слою управлению, так и к слою данных. Такой
+вариант подходит для выделенных доверенных сетей хранения.
+
+При включённом шифровании трафика OSD (хотя бы заголовков) есть возможность задействовать
+права доступа, включив опцию `use_perms=true`. При включённой опции каждый пользователь может
+выполнять только те операции, которые ему разрешены, и даже OSD и мониторам также запрещены
+"лишние" операции.
+
+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.
+
+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
+
+И также есть свойство у пула:
+- 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).
+
+### Привилегии etcd
+
+Привилегии etcd реализуются через Antietcd во всех режимах работы.
+
+Встроенные привилегии etcd не поддерживаются по причине их многочисленных неудобств:
+- Аутентификация по сертификатам вообще не работает в REST интерфейсе etcd,
+- Привилегии хранятся отдельно от k/v данных и не могут участвовать в транзакциях,
+- Менять привилегии может только администратор (root),
+- Нет поддержки фильтрации диапазонных ответов чтения по привилегиям.
+
+Если используется etcd, то Antietcd выступает в роли фильтрующего прокси, при этом он
+может быть встроен в монитор Vitastor или запущен отдельно. В этом случае etcd должен
+разрешать входящие подключения только от Antietcd, а все остальные компоненты должны
+соединяться с Antietcd.
+
+Если Antietcd запускается в составе монитора Vitastor, то достаточно включить опцию
+`use_perms=true` и задать нужные сертификаты. Если Antietcd запускается отдельно, то
+привилегии нужно включать отдельно опциями Antietcd. Подробнее о настройке смотрите
+раздел [Варианты настройки etcd/Antietcd](#варианты-настройки-etcdantietcd).
+
+Перечень разрешённых операций с данными etcd смотрите в разделе
+[Права доступа к данным etcd](#права-доступа-к-данным-etcd).
+
+## Ручная настройка
+
+### Настройка шифрования соединений OSD
+
+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).
+
+В конфигурацию Vitastor на серверах OSD нужно добавить:
+- use_perms: true
+- osd_ca: osd_ca.crt
+- client_ca: client_ca.crt
+- osd_cert: osd_ca.crt
+- osd_pkey: osd_ca.key
+
+На стороне клиентов:
+- use_perms: true
+- cert: client.crt
+- pkey: client.key
+
+### Варианты настройки etcd/Antietcd
+
+Доступны следующие варианты настройки:
+
+#### Mon со встроенным Antietcd
+
+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)
+- use_perms: true
+- use_antietcd: true
+- antietcd_cert: antietcd.crt
+- antietcd_key: antietcd.key
+- etcd_ca: antietcd.crt
+- osd_ca: osd_ca.crt
+- client_ca: client_ca.crt
+
+#### Mon в роли Etcd proxy
+
+Если вы хотите включить привилегии, но остаться на etcd, можно задействовать режим etcd proxy.
+
+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)
+- use_perms: true
+- use_antietcd: true
+- etcd_proxy:
+  ```
+  {
+    "urls": [ "http://mon1:2381", ... ], // адреса ваших etcd с портом 2381
+    "cert": "antietcd.crt",
+    "key": "antietcd.key",
+    "ca": "etcd.crt"
+  }
+  ```
+- antietcd_cert: antietcd.crt
+- antietcd_key: antietcd.key
+- etcd_ca: antietcd.crt
+- osd_ca: osd_ca.crt
+- client_ca: client_ca.crt
+
+Опции командной строки etcd:
+```
+--advertise-client-urls=https://<АДРЕС>:2381 --listen-client-urls=https://<АДРЕС>:2381 \
+--client-cert-auth --cert-file=etcd.crt --key-file=etcd.key --trusted-ca-file=antietcd.crt \
+--peer-client-cert-auth --peer-cert-file=etcd.crt --peer-key-file=etcd.key --peer-trusted-ca-file=etcd.crt
+```
+
+#### Mon с отдельным Antietcd Proxy
+
+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 понадобятся
+отдельные сертификаты для клиентов, OSD и монитора.
+
+Настройки Vitastor:
+- etcd_address: [ "http://mon1:2379", ... ] (адреса ваших мониторов с портом 2379)
+- use_perms: true
+- use_antietcd: false
+- etcd_ca: antietcd.crt
+- osd_ca: osd_ca.crt
+- client_ca: client_ca.crt
+- mon_etcd_client_cert: mon_ca.crt
+- mon_etcd_client_key: mon_ca.key
+
+Опции командной строки Antietcd:
+```
+--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):
+```
+--advertise-client-urls=https://<АДРЕС>:2381 --listen-client-urls=https://<АДРЕС>:2381 \
+--client-cert-auth --cert-file=etcd.crt --key-file=etcd.key --trusted-ca-file=antietcd.crt \
+--peer-client-cert-auth --peer-cert-file=etcd.crt --peer-key-file=etcd.key --peer-trusted-ca-file=etcd.crt
+```
+
+#### Отдельный Antietcd без etcd
+
+Аналогично предыдущему варианту, но etcd и его сертификат не нужны:
+
+Настройки Vitastor (не отличаются от предыдущего варианта):
+- etcd_address: [ "http://mon1:2379", ... ] (адреса ваших мониторов с портом 2379)
+- use_perms: true
+- use_antietcd: false
+- etcd_ca: antietcd.crt
+- osd_ca: osd_ca.crt
+- client_ca: client_ca.crt
+- mon_etcd_client_cert: mon_ca.crt
+- mon_etcd_client_key: mon_ca.key
+
+Опции командной строки Antietcd:
+```
+--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
+
+Для использования Vault каждому клиенту, который будет получать из Vault ключи
+образов, нужна учётная запись в Vault. Vitastor поддерживает только аутентификацию
+по клиентским сертификатам, так что все сертификаты клиентов (`cert`+`pkey`) должны
+быть зарегистрированы в Vault и им должен быть дан доступ к соответствующим секретам
+(поддерживается API секретов v1).
+
+Требуемый формат секрета Vault - одно поле `key` в формате шестнадцатеричной строки.
+Используется алгоритм AES-256-XTS, так что длина ключа - 64 байта, то есть строка
+должна состоять из 128 шестнадцатеричных цифр.
+
+Для подключения Vault включите следующие настройки в Vitastor.conf:
+- `vault_url` - адрес Vault (например, `https://vault:8200`)
+- `vault_ca` - сертификат самого Vault
+
+После этого, если создать образ (`vitastor-cli create`) с опцией `--enc_key vault:<ID>`,
+то для получения ключа клиенты Vitastor сначала обратятся к Vault для получения токена
+по адресу `/v1/auth/cert/login`, а потом запросят из Vault сам секрет по адресу `/v1/secret/<ID>`.
+
+#### Пример настройки Vault
+
+Пошаговая инструкция для настройки тестового Vault на примере OpenBao:
+
+1. Если ещё не настроен TLS, генерируем самоподписанный TLS сертификат для Vault:
+   ```
+   openssl req -days 3650 -x509 -addext basicConstraints=critical,CA:TRUE,pathlen:1 --addext subjectAltName=DNS:vault \
+       -new -newkey rsa:4096 -nodes -keyout /etc/openbao/vault.key -out /etc/openbao/vault.crt
+   ```
+   Настраиваем его в `/etc/openbao/openbao.hcl`:
+   ```
+   listener "tcp" {
+       address       = "0.0.0.0:8200"
+       tls_cert_file = "/etc/openbao/vault.crt"
+       tls_key_file  = "/etc/openbao/vault.key"
+   }
+   ```
+   И перезапускаем OpenBao (`systemctl restart openbao`).
+2. Копируем TLS сертификат Vault для Vitastor:
+   ```
+   cp /etc/openbao/vault.crt /etc/vitastor/vault.crt
+   ```
+   Переносим его на все клиентские ноды и прописываем в `/etc/vitastor/vitastor.conf`:
+   ```
+   {
+       ...
+       "vault_url": "http://vault:8200",
+       "vault_ca": "/etc/vitastor/vault.crt"
+   }
+   ```
+3. Проверяем статус Vault:
+   ```
+   bao status -ca-cert /etc/openbao/vault.crt -address=https://vault:8200
+   ```
+4. Инициализируем Vault в тестовом режиме из 1 ноды (с 1 частью ключа):
+   ```
+   bao operator init -n 1 -t 1 -ca-cert /etc/openbao/vault.crt -address=https://vault:8200
+   ```
+5. Разблокируем Vault:
+   ```
+   bao operator unseal -ca-cert /etc/openbao/vault.crt -address=https://vault:8200
+   ```
+6. Включаем аутентификацию по сертификатам:
+   ```
+   bao auth enable -ca-cert /etc/openbao/vault.crt -address=https://vault:8200 cert
+   ```
+7. Включаем секреты v1:
+   ```
+   bao secrets enable -ca-cert /etc/openbao/vault.crt -address=https://vault:8200 -path=secret kv-v1
+   ```
+8. Создаём тестовый секрет:
+   ```
+   bao kv put -ca-cert /etc/openbao/vault.crt -address=https://vault:8200 secret/vitastor/testimg3 key=$(openssl rand -hex 64)
+   ```
+9. Генерируем подписанный сертификат для пользователя Vitastor (там, где у вас есть `client_ca.crt` и `client_ca.key`):
+   ```
+   openssl req -subj '/CN=testimg3' -nodes -new -keyout testimg3.key -out testimg3.csr
+   openssl x509 -req -days 3650 -CA client_ca.crt -CAkey client_ca.key -CAcreateserial -in testimg3.csr -out testimg3.crt
+   rm testimg3.csr
+   ```
+10. Создаём пользователя в Vault и даём ему доступ к секрету:
+    ```
+    cat >testimg3.policy <<EOF
+    path "/secret/vitastor/testimg3" {
+        capabilities = ["read"]
+    }
+    EOF
+
+    bao policy write -ca-cert /etc/openbao/vault.crt -address=https://vault:8200 testimg3 testimg3.policy
+
+    bao write -ca-cert /etc/openbao/vault.crt -address=https://vault:8200 auth/cert/certs/testimg3 \
+        certificate=@testimg3.crt display_name=testimg3 token_ttl=24h token_policies=testimg3
+    ```
+11. Тестируем доступ к секрету:
+    ```
+    curl --cacert /etc/vitastor/vault.crt --cert testimg3.crt --key testimg3.key \
+        --json '{}' https://vault:8200/v1/auth/cert/login
+    ```
+    Будет выведен токен, подставляем его в следующий запрос:
+    ```
+    curl --cacert /etc/vitastor/vault.crt --cert testimg3.crt --key testimg3.key \
+        -H 'X-Vault-Token: <ПОЛУЧЕННЫЙ ТОКЕН>' https://vault:8200/v1/secret/vitastor/testimg3
+    ```
+12. Создаём образ в Vitastor с заданным секретом (от имени администратора или того, кто имеет
+    право создавать образы в вашем пуле):
+    ```
+    vitastor-cli create -s 100G --enc_key vault:vitastor/testimg3 --owner testimg3 testimg3
+    ```
+13. Тестируем доступ к образу от имени пользователя testimg3:
+    ```
+    vitastor-cli --cert testimg3.crt --pkey testimg3.key dd if=/dev/urandom oimg=testimg3 bs=1M count=100
+    ```
+
+## Списки разрешённых операций
+
+### Права доступа к данным etcd
+
+Ниже все названия ключей приведены без общего префикса `/vitastor`.
+
+Разрешённые операции с ключами в Antietcd для клиентов (`type=client`):
+- Только чтение:
+  - Разрешено всегда:
+    - `/config/global`
+    - `/config/node_placement`
+    - `/config/pools`
+    - `/pg/config`
+    - `/osd/state/*`
+    - `/pg/state/*`
+    - `/index/maxid/*`
+  - Для образов, которые [может читать пользователь](#пользователи-и-права-доступа):
+    - `/config/inode/*`
+    - `/index/image/*`
+    - `/inode/stats/*`
+- Чтение и запись:
+  - Для пулов, в которых может создавать образы пользователь:
+    - `/index/maxid/*`
+  - Для образов, которыми владеет пользователь:
+    - `/config/inode/*`
+    - `/index/image/*`
+
+Разрешённые операции с ключами в Antietcd для администраторов (`type=admin`):
+- Чтение:
+  - `/stats`
+  - `/mon/*`
+  - `/pg/*`
+  - `/pgstats/*`
+  - `/inode/stats/*`
+  - `/pool/stats/*`
+- Чтение и запись:
+  - `/config/*`
+  - `/osd/*`
+  - `/index/*`
+  - `/pg/history/*`
+
+Разрешённые операции с ключами в etcd для OSD:
+- Чтение:
+  - `/pg/config`
+  - `/config/*`
+- Чтение и запись:
+  - `/osd/*`
+  - `/pg/state/*`
+  - `/pg/history/*`
+  - `/pgstats/*`
+
+Разрешённые операции с ключами в etcd для мониторов:
+- Чтение:
+  - `/config/*`
+  - `/osd/*`
+  - `/pgstats/*`
+- Чтение и запись:
+  - `/pg/config`
+  - `/stats`
+  - `/history/last_clean_pgs`
+  - `/mon/*`
+  - `/pg/history/*`
+  - `/inode/stats/*`
+  - `/pool/stats/*`
+
+### Права доступа к данным OSD
+
+При включённой опции `use_perms` и шифровании OSD аутентифицирует клиентов по сертификатам
+и разрешает каждому клиенту только то, что ему разрешено согласно модели прав доступа.
+
+Клиентские операции:
+- READ - разрешено для образов, доступных пользователю на чтение.
+- WRITE, DELETE, SCRUB - разрешены для образов, доступных пользователю на запись.
+- SYNC - операция не связана с образом и разрешена всегда.
+- DESCRIBE - операция разрешена только для администраторов (используются командами
+  `vitastor-cli describe` и `fix`).
+- PING - операция разрешена всегда.
+- SHOW_CONFIG - операция разрешена всегда, однако если в ней клиент представляется
+  как OSD, то проверяется, что он использует сертификат, подписанный `osd_ca`.
+- SEC_LIST (листинг) - разрешена другим OSD и администраторам с любыми параметрами,
+  а обычным клиентам разрешена только для запросов, ограниченных образом, доступным
+  пользователю на чтение.
+
+Кластерные операции - разрешаются только другим OSD:
+- SEC_READ
+- SEC_WRITE
+- SEC_WRITE_STABLE
+- SEC_SYNC
+- SEC_STABILIZE
+- SEC_ROLLBACK
+- SEC_DELETE
+- SEC_READ_BMP
+- SEC_LOCK
+
+### Права доступа к API
+
+[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
+`server_cert` с ключом `server_pkey`.
+
+При этом для корректной работы `vitastor-cli serve` он сам должен использовать
+для доступа в Vitastor сертификат (`cert`+`pkey`) пользователя с правами
+администратора (`type=admin`).
+
+Обычным клиентам при доступе к API разрешаются только API-операции с образами,
+доступными им либо на чтение (для чтения), либо на запись (для модификации).
+Все остальные API-вызовы разрешаются только для администраторов.
+
+Список разрешённых операций API:
+
+Клиентам (пользователям с `type=client`) разрешаются операции:
+- 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 (`type=admin`).
+
+## Производительность шифрования
+
+У вас может возникнуть вопрос - а как быстро всё это прекрасное шифрование работает?
+
+Ответ - сильно зависит от процессора. На современных процессорах (при наличии AVX512 с VAES)
+очень быстро - скорость шифрования AES может составлять 10-20 Гбайт/с и выше. В первую очередь
+подразумевается CPU клиентских машин, потому что сквозное шифрование выполняется целиком на
+клиенте, а транспортное хоть также и затрагивает OSD, но у клиента поток один, а OSD на стороне
+сервера много и добавить там ресурсов легче.
+
+На более старых процессорах скорость заметно хуже, например, на Xeon E5 v4 она составляет
+буквально 3 Гбайт/с.
+
+Вы можете оценить производительность своих процессоров с помощью команды `vitastor-cli cpubench`.
+
+Пример вывода (💪 AMD EPYC 9575F):
+
+```
+$ vitastor-cli cpubench
+Vitastor transport encryption benchmark (AES-256-GCM, AES-256-XTS and xxhash3)
+
+Warmup...
+
+No transport encryption, data checksums enabled, e2e unencrypted image
+xxhash3 1 M block... 209000 iterations in 2001 ms = 104447.78 MB/s
+xxhash3 4 K block... 37000000 iterations in 2022 ms = 71479.35 MB/s
+
+Header encryption with payload checksums, e2e unencrypted image
+AES-256-GCM encrypt header + xxhash3 1 M block... 210000 iterations in 2015 ms = 104218.36 MB/s
+AES-256-GCM encrypt header + xxhash3 4 K block... 26000000 iterations in 2073 ms = 48993.01 MB/s
+
+Full transport encryption, e2e unencrypted image
+AES-256-GCM encrypt header and 1 M block... 54000 iterations in 2000 ms = 27000.00 MB/s
+AES-256-GCM encrypt header and 4 K block... 11700000 iterations in 2014 ms = 22692.71 MB/s
+
+No transport encryption, no checksums, e2e encrypted image
+AES-256-XTS encrypt 1 M block... 50000 iterations in 2039 ms = 24521.82 MB/s
+AES-256-XTS encrypt 4 K block... 12600000 iterations in 2009 ms = 24499.13 MB/s
+
+No transport encryption, e2e encrypted image, data checksums enabled
+AES-256-XTS encrypt + xxhash3 1 M block... 40000 iterations in 2013 ms = 19870.84 MB/s
+AES-256-XTS encrypt + xxhash3 4 K block... 10200000 iterations in 2011 ms = 19812.90 MB/s
+
+Header encryption with payload checksums, e2e encrypted image
+AES-256-GCM encrypt header + AES-256-XTS encrypt + xxhash3 1 M block... 40000 iterations in 2014 ms = 19860.97 MB/s
+AES-256-GCM encrypt header + AES-256-XTS encrypt + xxhash3 4 K block... 8700000 iterations in 2011 ms = 16899.24 MB/s
+
+Full transport encryption, e2e encrypted image
+AES-256-XTS + AES-256-GCM encrypt 1 M block... 26000 iterations in 2062 ms = 12609.12 MB/s
+AES-256-XTS + AES-256-GCM encrypt 4 K block... 6300000 iterations in 2006 ms = 12267.88 MB/s
+```
+
+А вот Xeon E5-2680v4:
+
+```
+$ vitastor-cli cpubench
+Vitastor transport encryption benchmark (AES-256-GCM, AES-256-XTS and xxhash3)
+
+Warmup...
+
+No transport encryption, data checksums enabled, e2e unencrypted image
+xxhash3 1 M block... 62000 iterations in 2021 ms = 30677.88 MB/s
+xxhash3 4 K block... 12400000 iterations in 2006 ms = 24146.31 MB/s
+
+Header encryption with payload checksums, e2e unencrypted image
+AES-256-GCM encrypt header + xxhash3 1 M block... 62000 iterations in 2027 ms = 30587.07 MB/s
+AES-256-GCM encrypt header + xxhash3 4 K block... 6800000 iterations in 2011 ms = 13208.60 MB/s
+
+Full transport encryption, e2e unencrypted image
+AES-256-GCM encrypt header and 1 M block... 7000 iterations in 2317 ms = 3021.15 MB/s
+AES-256-GCM encrypt header and 4 K block... 1500000 iterations in 2102 ms = 2787.52 MB/s
+
+No transport encryption, no checksums, e2e encrypted image
+AES-256-XTS encrypt 1 M block... 7000 iterations in 2317 ms = 3021.15 MB/s
+AES-256-XTS encrypt 4 K block... 1600000 iterations in 2088 ms = 2993.30 MB/s
+
+No transport encryption, e2e encrypted image, data checksums enabled
+AES-256-XTS encrypt + xxhash3 1 M block... 6000 iterations in 2188 ms = 2742.23 MB/s
+AES-256-XTS encrypt + xxhash3 4 K block... 1400000 iterations in 2053 ms = 2663.78 MB/s
+
+Header encryption with payload checksums, e2e encrypted image
+AES-256-GCM encrypt header + AES-256-XTS encrypt + xxhash3 1 M block... 6000 iterations in 2190 ms = 2739.73 MB/s
+AES-256-GCM encrypt header + AES-256-XTS encrypt + xxhash3 4 K block... 1300000 iterations in 2101 ms = 2417.00 MB/s
+
+Full transport encryption, e2e encrypted image
+AES-256-XTS + AES-256-GCM encrypt 1 M block... 4000 iterations in 2666 ms = 1500.38 MB/s
+AES-256-XTS + AES-256-GCM encrypt 4 K block... 800000 iterations in 2113 ms = 1478.94 MB/s
+```

docs/usage/cli.en.md

@@ -125,18 +125,31 @@ bench-kaveri         kaveri    10 G  10 G    0 B/s  0     0      0 us  0 B/s  0
 
 ## create
 
-`vitastor-cli create -s|--size <size> [-p|--pool <id|name>] [--parent <parent_name>[@<snapshot>]] <name>`
+`vitastor-cli create -s|--size SIZE [OPTIONS] <name>`
 
-Create an image. You may use K/M/G/T suffixes for `<size>`. If `--parent` is specified,
-a copy-on-write image clone is created. Parent must be a snapshot (readonly image).
-Pool must be specified if there is more than one pool.
+Create an image. Options:
+
+* `-s|--size SIZE` - New image size in bytes or with a K/M/G/T unit suffix.
+* `-p|--pool POOL` - Specify pool for the new image (may be omitted if there is only 1 pool).
+* `--parent PARENT` - Create a copy-on-write image clone based on PARENT (or PARENT@SNAPSHOT).
+  If parent is not a snapshot, it must be a read-only image.
+* `--enc-key random` - Generate a new random AES-256-XTS encryption key for the new image.
+* `--enc-key HEX` - Set a specified AES-256-XTS key (64 bytes in hex) for the new image.
+* `--enc-key vault:ID` - Use an encryption key from an external Vault secret with specified ID.
 
 ```
-vitastor-cli create --snapshot <snapshot> [-p|--pool <id|name>] <image>
-vitastor-cli snap-create [-p|--pool <id|name>] <image>@<snapshot>
+vitastor-cli create --snapshot <snapshot> [OPTIONS] <image>
+vitastor-cli snap-create [OPTIONS] <image>@<snapshot>
 ```
 
-Create a snapshot of image `<name>` (either form can be used). May be used live if only a single writer is active.
+Create a snapshot of image `<image>`. May be used live if only a single writer is active.
+
+Options:
+
+* `-p|--pool POOL` - Move image to pool POOL, leaving the snapshot in the old pool.
+* `--enc-key random` - Change image encryption key to a new random AES-256-XTS key.
+* `--enc-key KEY` - Change image encryption key to a specified key, Vault key or to an empty key.
+  By default, the image retains its old encryption key when taking a snapshot.
 
 See also about [how to export snapshots](qemu.en.md#exporting-snapshots).
 
@@ -151,6 +164,7 @@ You should resize file system in the image, if present, before shrinking it.
 * `--deleted 1|0` - Set/clear 'deleted image' flag (set automatically during unfinished deletes).
 * `-f|--force` - Proceed with shrinking or setting readwrite flag even if the image has children.
 * `--down-ok` - Proceed with shrinking even if some data will be left on unavailable OSDs.
+* `--enc-key HEX` - Change image encryption key (allowed only with `--force`).
 
 ## dd
 

docs/usage/cli.ru.md

@@ -127,19 +127,32 @@ bench-kaveri         kaveri    10 G  10 G    0 B/s  0     0      0 us  0 B/s  0
 
 ## create
 
-`vitastor-cli create -s|--size <size> [-p|--pool <id|name>] [--parent <parent_name>[@<snapshot>]] <name>`
+`vitastor-cli create -s|--size SIZE [ОПЦИИ] <name>`
 
-Создать образ. Для размера `<size>` можно использовать суффиксы K/M/G/T (килобайт-мегабайт-гигабайт-терабайт).
-Если указана опция `--parent`, создаётся клон образа. Родитель `<parent_name>[@<snapshot>]` должен быть
-снимком (или просто немодифицируемым образом). Пул обязательно указывать, если в кластере больше одного пула.
+Создать образ. Опции:
+
+* `-s|--size SIZE` - Размер нового образа в байтах или с суффиксом K/M/G/T (кило/мега/гига/терабайт).
+* `-p|--pool POOL` - Создать образ в заданном пуле (можно не указывать, если пул всего один).
+* `--parent PARENT` - Создать легковесный клон на основе образа `PARENT` или снимка `PARENT@SNAP`.
+  Если `PARENT` - не снимок, он должен быть помечен как образ только для чтения.
+* `--enc-key random` - Сгенерировать случайный ключ шифрования AES-256-XTS для нового образа.
+* `--enc-key HEX` - Установить заданный ключ AES-256-XTS (64 байта в hex) для нового образа.
+* `--enc-key vault:ID` - Использовать ключ из внешнего секрета с заданным ID из Vault.
 
 ```
-vitastor-cli create --snapshot <snapshot> [-p|--pool <id|name>] <image>
-vitastor-cli snap-create [-p|--pool <id|name>] <image>@<snapshot>
+vitastor-cli create --snapshot <snapshot> [ОПЦИИ] <image>
+vitastor-cli snap-create [ОПЦИИ] <image>@<snapshot>
 ```
 
-Создать снимок образа `<name>` (можно использовать любую форму команды). Снимок можно создавать без остановки
-клиентов, если пишущий клиент максимум 1.
+Создать снимок образа `<image>` (можно использовать любую форму команды).
+Снимок можно создавать без остановки клиентов, если пишущих клиентов не больше одного.
+
+Опции:
+
+* `-p|--pool POOL` - Переместить образ в пул POOL, оставив снимок в старом пуле.
+* `--enc-key random` - Изменить ключ шифрования образа на новый случайный ключ AES-256-XTS.
+* `--enc-key KEY` - Изменить ключ шифрования образа на заданный ключ, ключ из Vault или пустой ключ.
+  По умолчанию шифрованные образы сохраняют старый ключ при снятии снимка.
 
 Смотрите также информацию о том, [как экспортировать снимки](qemu.ru.md#экспорт-снимков).
 
@@ -156,6 +169,7 @@ vitastor-cli snap-create [-p|--pool <id|name>] <image>@<snapshot>
 * `--deleted 1|0` - Установить/снять флаг "образ удалён" (устанавливается при незавершённом удалении).
 * `-f|--force` - Разрешить уменьшение или перевод в чтение-запись образа, у которого есть клоны.
 * `--down-ok` - Разрешить уменьшение, даже если часть данных останется неудалённой на недоступных OSD.
+* `--enc-key HEX` - Изменить ключ шифрования образа (разрешено только с `--force`).
 
 ## dd
 

mon/antietcd_adapter.js

@@ -3,6 +3,7 @@
 
 const AntiEtcd = require('antietcd');
 
+const vitastor_auth_filter = require('./vitastor_auth_filter.js');
 const vitastor_persist_filter = require('./vitastor_persist_filter.js');
 const { b64, local_ips } = require('./utils.js');
 
@@ -18,7 +19,7 @@ class AntiEtcdAdapter
                 cluster = cluster ? (''+(cluster||'')).split(/,+/) : [];
             cluster = Object.keys(cluster.reduce((a, url) =>
             {
-                a[url.toLowerCase().replace(/^(https?:\/\/)/, '').replace(/\/.*$/, '')] = true;
+                a[url.toLowerCase().replace(/^(https?:\/\/)?(.*?)(\/.*)?$/, (m, m1, m2) => (m1||'http://')+m2)] = true;
                 return a;
             }, {}));
             const cfg_port = config.antietcd_port;
@@ -26,7 +27,18 @@ class AntiEtcdAdapter
             is_local['0.0.0.0'] = true;
             is_local['::'] = true;
             is_local[''] = true;
-            const selected = cluster.map(s => s.split(':', 2)).filter(ip => is_local[ip[0]] && (!cfg_port || ip[1] == cfg_port));
+            // split :, 3 -> <schema>:<//ip>:<port>
+            const selected = [];
+            for (let i = 0; i < cluster.length; i++)
+            {
+                const m = /^(https?:\/\/)?(?:\[(.*)\]|([^\[\:]+))(?::(\d+))?$/.exec(cluster[i]);
+                if (!m)
+                    continue;
+                const ip = m[3] || m[2];
+                const port = m[4] || 2379;
+                if (is_local[ip] && (!cfg_port || port == cfg_port))
+                    selected.push({ idx: i, ip, port });
+            }
             if (selected.length > 1)
             {
                 console.error('More than 1 etcd_address matches local IPs, please specify port');
@@ -35,16 +47,45 @@ class AntiEtcdAdapter
             else if (selected.length == 1)
             {
                 const antietcd_config = {
-                    ip: selected[0][0],
-                    port: selected[0][1],
-                    data: config.antietcd_data_file || ((config.antietcd_data_dir || '/var/lib/vitastor') + '/mon_'+selected[0][1]+'.json.gz'),
+                    ip: selected[0].ip,
+                    port: selected[0].port,
+                    cert: config.antietcd_cert,
+                    key: config.antietcd_key,
+                    ca: config.client_ca,
+                    data: config.antietcd_data_file || ((config.antietcd_data_dir || '/var/lib/vitastor') + '/mon_'+selected[0].port+'.json.gz'),
                     persist_filter: vitastor_persist_filter({ vitastor_prefix: config.etcd_prefix || '/vitastor' }),
-                    node_id: selected[0][0]+':'+selected[0][1], // node_id = ip:port
-                    cluster: (cluster.length == 1 ? null : cluster.reduce((a, c) => { a[c] = "http://"+c; return a; }, {})),
+                    node_id: cluster[selected[0].idx].replace(/^(https?:\/\/)/, ''), // same as in <cluster> below
+                    cluster: (cluster.length == 1 ? null : cluster.reduce((a, c) => { a[c.replace(/^(https?:\/\/)/, '')] = c; return a; }, {})),
                     cluster_key: (config.etcd_prefix || '/vitastor'),
                     stale_read: 1,
                     log_level: 1,
+                    logs: { cluster: true },
                 };
+                if (config.etcd_proxy)
+                {
+                    // Monitor may use the builtin etcd_proxy mode
+                    if (!config.etcd_proxy.urls)
+                    {
+                        console.error('etcd_proxy.urls are empty');
+                        process.exit(1);
+                    }
+                    antietcd_config.etcd_proxy = config.etcd_proxy.urls;
+                    antietcd_config.etcd_cert = config.etcd_proxy.cert;
+                    antietcd_config.etcd_key = config.etcd_proxy.key;
+                    antietcd_config.etcd_ca = config.etcd_proxy.ca;
+                    delete antietcd_config.data;
+                    delete antietcd_config.persist_filter;
+                    delete antietcd_config.cluster;
+                    delete antietcd_config.cluster_key;
+                }
+                if (config.use_perms)
+                {
+                    antietcd_config.client_cert_auth = true;
+                    antietcd_config.auth_filter = vitastor_auth_filter;
+                    antietcd_config.ca = config.client_ca;
+                    antietcd_config.osd_ca = config.osd_ca;
+                    antietcd_config.mon_ca = config.mon_ca;
+                }
                 for (const key in config)
                 {
                     if (key.substr(0, 9) === 'antietcd_')
@@ -169,7 +210,7 @@ class AntiEtcdAdapter
                     await new Promise(ok => setTimeout(ok, timeout-(Date.now()-prev)));
                 }
                 prev = Date.now();
-                const res = await this.antietcd.api(path.replace(/^\/+/, '').replace(/\/+$/, '').replace(/\/+/g, '_'), body);
+                const res = await this.antietcd.api(path.replace(/^\/+/, '').replace(/\/+$/, '').replace(/\/+/g, '_'), body, { user_type: 'mon' });
                 if (res.error)
                 {
                     console.error('Failed to query antietcd '+path+' (retry '+retry+'/'+retries+'): '+res.error);

mon/etcd_adapter.js

@@ -1,7 +1,9 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 (see README.md for details)
 
+const fs = require('fs');
 const http = require('http');
+const https = require('https');
 const WebSocket = require('ws');
 const { b64, local_ips } = require('./utils.js');
 
@@ -15,11 +17,30 @@ class EtcdAdapter
         this.ws = null;
         this.ws_alive = false;
         this.ws_keepalive_timer = null;
+        this.opts = {};
     }
 
     parse_config(config)
     {
         this.parse_etcd_addresses(config.etcd_address||config.etcd_url);
+        if (config.mon_etcd_client_cert || config.etcd_client_cert)
+        {
+            this.opts.cert = config.mon_etcd_client_cert || config.etcd_client_cert;
+            if (this.opts.cert.substr(0, 5) != '-----')
+                this.opts.cert = fs.readFileSync(this.opts.cert, { encoding: 'utf-8' });
+        }
+        if (config.mon_etcd_client_key || config.etcd_client_key)
+        {
+            this.opts.key = config.mon_etcd_client_key || config.etcd_client_key;
+            if (this.opts.key.substr(0, 5) != '-----')
+                this.opts.key = fs.readFileSync(this.opts.key, { encoding: 'utf-8' });
+        }
+        if (config.etcd_ca)
+        {
+            this.opts.ca = config.etcd_ca;
+            if (this.opts.ca.substr(0, 5) != '-----')
+                this.opts.ca = fs.readFileSync(this.opts.ca, { encoding: 'utf-8' });
+        }
     }
 
     parse_etcd_addresses(addrs)
@@ -39,7 +60,7 @@ class EtcdAdapter
         for (let url of addrs)
         {
             let scheme = 'http';
-            url = url.trim().replace(/^(https?):\/\//, (m, m1) => { scheme = m1; return ''; });
+            url = url.trim().replace(/^(https?):\/\//i, (m, m1) => { scheme = m1.toLowerCase(); return ''; });
             const slash = url.indexOf('/');
             const colon = url.indexOf(':');
             const is_local = is_local_ip[colon >= 0 ? url.substr(0, colon) : (slash >= 0 ? url.substr(0, slash) : url)];
@@ -130,7 +151,7 @@ class EtcdAdapter
                     }
                     ok(false);
                 }, this.mon.config.etcd_mon_timeout);
-                this.ws = new WebSocket(base+'/watch');
+                this.ws = new WebSocket(base+'/watch', this.opts);
                 this.ws_used_url = cur_addr;
                 const fail = () =>
                 {
@@ -272,7 +293,7 @@ class EtcdAdapter
             {
                 throw new Error(MON_STOPPED);
             }
-            const res = await POST(base+path, body, timeout);
+            const res = await POST(base+path, body, timeout, this.opts);
             if (this.mon.stopped)
             {
                 throw new Error(MON_STOPPED);
@@ -298,7 +319,7 @@ class EtcdAdapter
     }
 }
 
-function POST(url, body, timeout)
+function POST(url, body, timeout, opts)
 {
     return new Promise(ok =>
     {
@@ -310,10 +331,10 @@ function POST(url, body, timeout)
             req = null;
             ok({ error: 'timeout' });
         }, timeout) : null;
-        let req = http.request(url, { method: 'POST', headers: {
+        let req = (url.substr(0, 5) == 'https' ? https : http).request(url, { method: 'POST', headers: {
             'Content-Type': 'application/json',
             'Content-Length': body_text.length,
-        } }, (res) =>
+        }, ...(opts||{}) }, (res) =>
         {
             if (!req)
             {

mon/etcd_schema.js

@@ -16,6 +16,7 @@ const etcd_allow = new RegExp('^'+[
     'config/pools',
     'config/osd/[1-9]\\d*',
     'config/pgs', // old name
+    'config/user/.*',
     'pg/config',
     'config/inode/[1-9]\\d*/[1-9]\\d*',
     'osd/state/[1-9]\\d*',
@@ -45,7 +46,14 @@ const etcd_tree = {
             config_path: "/etc/vitastor/vitastor.conf",
             etcd_prefix: "/vitastor",
             // etcd connection - configurable online
-            etcd_address: "10.0.115.10:2379/v3",
+            etcd_address: "http://10.0.115.10:2379/v3",
+            etcd_client_cert: "",
+            etcd_client_key: "",
+            osd_etcd_client_cert: "",
+            osd_etcd_client_key: "",
+            mon_etcd_client_cert: "",
+            mon_etcd_client_key: "",
+            etcd_ca: "",
             // mon
             etcd_mon_ttl: 5, // min: 1
             etcd_mon_timeout: 1000, // ms. min: 0
@@ -201,6 +209,8 @@ const etcd_tree = {
                 primary_affinity_tags?: 'nvme' | [ 'nvme', ... ],
                 // scrub interval
                 scrub_interval?: '30d',
+                // users allowed to create images in this pool
+                creator_group?: '',
             },
             ...
         }, */
@@ -217,10 +227,21 @@ const etcd_tree = {
                     parent_id?: <inode_t>,
                     readonly?: boolean,
                     deleted?: boolean,
+                    enc_key?: string,
+                    owner?: string,
+                    owner_group?: string,
+                    reader_group?: string,
                 }
             }
         }, */
         inode: {},
+        /* user: {
+            <username>: {
+                type: 'osd'|'mon'|'admin'|'client',
+                groups: string[],
+            },
+        }, */
+        user: {},
     },
     osd: {
         state: {

mon/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vitastor-mon",
-  "version": "3.0.13",
+  "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": "^1.2.4",
+    "antietcd": "^1.3.1",
     "sprintf-js": "^1.1.2",
     "ws": "^7.2.5"
   },

mon/scripts/make-etcd

@@ -1,38 +1,324 @@
 #!/usr/bin/node
-// Simple systemd unit generator for etcd
+// Simple Vitastor etcd / antietcd / TLS configurator
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: MIT
 
-// USAGE:
-// 1) Put the same etcd_address into /etc/vitastor/vitastor.conf on all monitor nodes
-// 2) Run ./make-etcd.js. It will create the etcd service on one of specified IPs
-
 const child_process = require('child_process');
 const fs = require('fs');
 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)
+
+USAGE:
+1) Create a minimal vitastor.conf with etcd_address, osd_network and (optionally) use_perms.
+   Non-encrypted: {"etcd_address":["http://10.0.0.10:2379","http://10.0.0.11:2379","http://10.0.0.12:2379"],"osd_network":"10.0.0.0/24"}
+   Encrypted: {"etcd_address":["https://10.0.0.10:2379","https://10.0.0.11:2379","https://10.0.0.12:2379"],"use_perms":true,"osd_network":"10.0.0.0/24"}
+   (Note https:// etcd URLs!)
+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, 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.
+
+OPTIONS:
+--antietcd-only
+    disable etcd (proxy or direct mode), use only antietcd
+--gen-certs
+    force certificate generation even if it's not the first node
+--no-certs
+    disable certificate generation
+--copy yes|no|ask
+    copy vitastor.conf and TLS certificates for monitor&etcd to monitor nodes using scp
+    (default is ask)
+--copy-to-osd-node NODE[,NODE2,...]
+    copy vitastor.conf and TLS certificates for OSDs to NODES using scp
+`;
+
 async function run()
 {
-    const config_path = process.argv[2] || '/etc/vitastor/vitastor.conf';
-    if (config_path == '-h' || config_path == '--help')
+    let config_path = '/etc/vitastor/vitastor.conf';
+    let config_dir = '/etc/vitastor/';
+    let gen_certs = 'auto';
+    let antietcd_only = false;
+    let copy = 'ask';
+    let copy_to_osd = null;
+    for (let i = 2; i < process.argv.length; i++)
     {
-        console.log(
-            'Initialize systemd etcd service for Vitastor\n'+
-            '(c) Vitaliy Filippov, 2019+ (MIT)\n'+
-            '\n'+
-            'USAGE:\n'+
-            '1) Put the same etcd_address into /etc/vitastor/vitastor.conf on all monitor nodes\n'+
-            '2) Run '+process.argv[1]+' [config_path]\n'
-        );
-        process.exit(0);
+        const arg = process.argv[i];
+        if (arg == '-h' || arg == '--help')
+        {
+            console.log(help_text);
+            process.exit(0);
+        }
+        else if (arg == '--gen-certs')
+        {
+            gen_certs = true;
+        }
+        else if (arg == '--no-certs')
+        {
+            gen_certs = false;
+        }
+        else if (arg == '--antietcd-only')
+        {
+            antietcd_only = true;
+        }
+        else if (arg == '--copy-to-osd-node' && i < process.argv.length-1)
+        {
+            i++;
+            copy_to_osd = process.argv[i].split(/,/);
+        }
+        else if (arg == '--copy' && i < process.argv.length-1)
+        {
+            i++;
+            copy = process.argv[i];
+            if (copy !== 'ask' && copy !== 'yes' && copy !== 'no')
+            {
+                console.error('--copy should be "ask", "yes" or "no"');
+                process.exit(1);
+            }
+        }
+        else if (arg[0] == '-')
+        {
+            console.error('Unknown option: '+arg[0]);
+            process.exit(1);
+        }
+        else
+        {
+            config_path = arg;
+            config_dir = path.dirname(arg);
+        }
     }
     if (!fs.existsSync(config_path))
     {
         console.log(config_path+' is missing');
         process.exit(1);
     }
+    const config = JSON.parse(fs.readFileSync(config_path, { encoding: 'utf-8' }));
+    if (!config.etcd_address)
+    {
+        console.log("etcd_address is missing in "+config_path);
+        process.exit(1);
+    }
+    const etcds = (config.etcd_address instanceof Array ? config.etcd_address : (''+config.etcd_address).split(/,/))
+        .map(s => /^(https?):\/\/(\[[^\]]+\]|[^\[\]\:\/]+)(?::(\d+))?/.exec(s.toLowerCase()))
+        .filter(s => s)
+        .map(s => ({
+            scheme: s[1],
+            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],
+        }));
+    const tls = etcds.filter(e => e.scheme === 'https').length > 0;
+    const use_perms = tls && config.use_perms;
+    const num = select_local_etcd(etcds);
+    if (copy_to_osd)
+    {
+        copy_to_osd_nodes(copy_to_osd, config_dir, use_perms, antietcd_only);
+        process.exit(0);
+    }
+    if (tls)
+    {
+        const etcd_ca = config_dir+'/'+path.basename(config.etcd_ca);
+        if (gen_certs === true)
+        {
+            console.log('Certificate generation is requested explicitly, generating');
+        }
+        else if (gen_certs === false)
+        {
+            console.log('Certificate generation is disabled explicitly, skipping');
+        }
+        else if (num < 0)
+        {
+            gen_certs = true;
+            console.log('No matching IPs in etcd_address from '+config_path+', only generating certificates');
+        }
+        else if (config.etcd_ca && fs.existsSync(etcd_ca))
+        {
+            gen_certs = false;
+            console.log(etcd_ca+' already exists, assuming certificates are already generated');
+        }
+        else if (num === 0)
+        {
+            gen_certs = true;
+            console.log('This is monitor node 1, generating certificates');
+        }
+        else
+        {
+            console.log('This is monitor node '+(num+1)+', '+etcd_ca+' does not exist, please copy certificates to this node');
+            process.exit(1);
+        }
+        await write_auth_config(config, config_path, etcds, use_perms, antietcd_only);
+        if (gen_certs)
+        {
+            if (copy === 'ask')
+                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, use_perms, antietcd_only);
+        }
+    }
+    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);
+}
+
+async function ask_copy(question)
+{
+    const rl = readline.createInterface({
+        input: process.stdin,
+        output: process.stdout,
+        prompt: '> ',
+    });
+    let copy;
+    while (copy != 'y' && copy != 'n' && copy != 'yes' && copy != 'no')
+    {
+        if (copy)
+            console.log('Please type "yes" or "no"');
+        copy = await new Promise(ok => rl.question(question, ok));
+    }
+    return copy;
+}
+
+async function copy_to_osd_nodes(to, dir, use_perms, antietcd_only)
+{
+    const osd_to_copy = [ 'vitastor.conf' ];
+    if (!antietcd_only && !use_perms)
+        osd_to_copy.push('etcd_ca.crt');
+    else
+        osd_to_copy.push('antietcd_ca.crt');
+    if (use_perms)
+        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+":/etc/vitastor/");
+}
+
+async function make_certs(dir, copy, etcds, use_perms, antietcd_only)
+{
+    console.log(`-----
+Generating certificates in ${dir}
+-----
+`);
+    const to_copy = [ 'vitastor.conf' ];
+    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(','));
+        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(','));
+        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_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:
+- Monitor+OSD node:
+  cd ${dir} && scp ${to_copy.join(' ')} root@NODE:/etc/vitastor/
+- Monitor node:
+  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/
+-----
+`);
+    if (copy)
+    {
+        for (const node of etcds)
+        {
+            await system("scp "+dir+to_copy.join(" "+dir)+" root@"+node.ip+":/etc/vitastor/");
+        }
+    }
+    else
+    {
+        console.warn('Certificates generated in '+dir+', please copy them to other nodes');
+    }
+}
+
+async function write_auth_config(config, config_path, etcds, use_perms, antietcd_only)
+{
+    const auth = {};
+    if (use_perms)
+    {
+        auth["use_antietcd"] = true;
+        if (!antietcd_only)
+        {
+            auth["etcd_proxy"] = {
+                urls: etcds.map(e => e.ip+':2381'),
+                cert: "/etc/vitastor/antietcd.crt",
+                key: "/etc/vitastor/antietcd.key",
+                ca: "/etc/vitastor/etcd_ca.crt",
+            };
+        }
+        auth["antietcd_cert"] = "/etc/vitastor/antietcd.crt";
+        auth["antietcd_key"] = "/etc/vitastor/antietcd.key";
+        auth["etcd_ca"] = "/etc/vitastor/antietcd_ca.crt";
+        auth["osd_cert"] = "/etc/vitastor/osd.crt";
+        auth["osd_pkey"] = "/etc/vitastor/osd.key";
+        auth["osd_ca"] = "/etc/vitastor/osd.crt";
+        auth["client_ca"] = "/etc/vitastor/client_ca.crt";
+        auth["cert"] = "/etc/vitastor/admin.crt";
+        auth["pkey"] = "/etc/vitastor/admin.key";
+    }
+    else
+    {
+        if (antietcd_only)
+        {
+            auth["use_antietcd"] = true;
+            auth["antietcd_cert"] = "/etc/vitastor/antietcd.crt";
+            auth["antietcd_key"] = "/etc/vitastor/antietcd.key";
+            auth["etcd_ca"] = "/etc/vitastor/antietcd_ca.crt";
+        }
+        else
+        {
+            auth["etcd_ca"] = "/etc/vitastor/etcd.crt";
+        }
+    }
+    for (const k in auth)
+    {
+        if ((k in config) && JSON.stringify(auth[k]) != JSON.stringify(config[k]))
+        {
+            // Auth options already overridden with non-default
+            console.log(k+" is already overridden in "+config_path+", skipping config update");
+            return;
+        }
+    }
+    for (const k in auth)
+    {
+        config[k] = auth[k];
+    }
+    console.log(`-----
+Updating ${config_path}
+-----
+`);
+    fs.writeFileSync(config_path, JSON.stringify(config, 0, 4));
+}
+
+async function configure_etcd(etcds, num, tls, use_perms)
+{
     const in_docker = fs.existsSync("/etc/vitastor/etcd.conf") &&
         fs.existsSync("/etc/vitastor/docker.conf");
     if (!in_docker && fs.existsSync("/etc/systemd/system/vitastor-etcd.service"))
@@ -45,37 +331,55 @@ async function run()
         console.log("/etc/systemd/system/etcd.service already exists");
         process.exit(1);
     }
-    const config = JSON.parse(fs.readFileSync(config_path, { encoding: 'utf-8' }));
-    if (!config.etcd_address)
+    const etcd_url = etcds[num].scheme + '://' + etcds[num].addr;
+    const options = {
+        name: 'etcd'+etcds[num].ip.replace(/[^0-9a-z_]/ig, '_'),
+        advertise_client_urls: etcd_url+':'+(use_perms ? 2381 : 2379),
+        listen_client_urls: etcd_url+':'+(use_perms ? 2381 : 2379),
+        initial_advertise_peer_urls: etcd_url+':2380',
+        listen_peer_urls: etcd_url+':2380',
+        initial_cluster_token: 'vitastor-etcd-1',
+        initial_cluster_state: 'new',
+        initial_cluster: etcds.map(e => `etcd${e.ip.replace(/[^0-9a-z_]/ig, '_')}=${e.scheme}://${e.addr}:2380`).join(','),
+        snapshot_count: 10000,
+        max_txn_ops: 100000,
+        max_request_bytes: 104857600,
+        auto_compaction_retention: 10,
+        auto_compaction_mode: 'revision',
+    };
+    if (tls)
     {
-        console.log("etcd_address is missing in "+config_path);
-        process.exit(1);
+        options['cert_file'] = '/etc/vitastor/etcd.crt';
+        options['key_file'] = '/etc/vitastor/etcd.key';
+        if (use_perms)
+        {
+            options['client_cert_auth'] = '1';
+            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.crt';
+        }
     }
-    const etcds = (config.etcd_address instanceof Array ? config.etcd_address : (''+config.etcd_address).split(/,/))
-        .map(s => (''+s).replace(/^https?:\/\/|(:\d+)?(\/.*)?$/g, '').replace(/^\[(.*)\]$/, '$1').toLowerCase());
-    const num = select_local_etcd(etcds);
-    if (num < 0)
+    let etcd_conf = fs.existsSync("/etc/vitastor/etcd.conf")
+        ? fs.readFileSync("/etc/vitastor/etcd.conf", { encoding: 'utf-8' })
+        : "";
+    for (const k in options)
     {
-        console.log('No matching IPs in etcd_address from '+config_path);
-        process.exit(0);
+        etcd_conf = replace_env(etcd_conf, 'ETCD_'+k.toUpperCase().replace(/-/, '_'), options[k]);
     }
-    const etcd_url = 'http://' + (etcds[num].indexOf(':') >= 0 ? '['+etcds[num]+']' : etcds[num]);
-    const etcd_name = 'etcd'+etcds[num].replace(/[^0-9a-z_]/ig, '_');
-    const etcd_cluster = etcds.map(e => `etcd${e.replace(/[^0-9a-z_]/ig, '_')}=http://${e.indexOf(':') >= 0 ? '['+e+']' : e}:2380`).join(',');
+    fs.writeFileSync("/etc/vitastor/etcd.conf", etcd_conf);
     if (in_docker)
     {
-        let etcd_conf = fs.readFileSync("/etc/vitastor/etcd.conf", { encoding: 'utf-8' });
-        etcd_conf = replace_env(etcd_conf, 'ETCD_NAME', etcd_name);
-        etcd_conf = replace_env(etcd_conf, 'ETCD_IP', etcds[num]);
-        etcd_conf = replace_env(etcd_conf, 'ETCD_INITIAL_CLUSTER', etcd_cluster);
-        fs.writeFileSync("/etc/vitastor/etcd.conf", etcd_conf);
         console.log('etcd for Vitastor configured. Run `systemctl enable --now vitastor-etcd` to start etcd');
         process.exit(0);
     }
     await system(`mkdir -p /var/lib/etcd/vitastor`);
     fs.writeFileSync(
-        "/etc/systemd/system/vitastor-etcd.service",
-`[Unit]
+        "/etc/systemd/system/vitastor-etcd.service", `[Unit]
 Description=etcd for vitastor
 After=network-online.target local-fs.target time-sync.target
 Wants=network-online.target local-fs.target time-sync.target
@@ -83,12 +387,8 @@ Wants=network-online.target local-fs.target time-sync.target
 [Service]
 Restart=always
 Environment=GOGC=50
-ExecStart=etcd --name ${etcd_name} --data-dir /var/lib/etcd/vitastor \\
-    --snapshot-count 10000 --advertise-client-urls ${etcd_url}:2379 --listen-client-urls ${etcd_url}:2379 \\
-    --initial-advertise-peer-urls ${etcd_url}:2380 --listen-peer-urls ${etcd_url}:2380 \\
-    --initial-cluster-token vitastor-etcd-1 --initial-cluster ${etcd_cluster} \\
-    --initial-cluster-state new --max-txn-ops=100000 --max-request-bytes=104857600 \\
-    --auto-compaction-retention=10 --auto-compaction-mode=revision
+EnvironmentFile=/etc/vitastor/etcd.conf
+ExecStart=etcd --data-dir /var/lib/etcd/vitastor
 WorkingDirectory=/var/lib/etcd/vitastor
 ExecStartPre=+chown -R etcd /var/lib/etcd/vitastor
 User=etcd
@@ -106,7 +406,11 @@ WantedBy=multi-user.target
     // Disable distribution etcd unit and enable our one
     await system(`systemctl disable --now etcd`);
     await system(`systemctl enable --now vitastor-etcd`);
-    process.exit(0);
+}
+
+async function enable_mon()
+{
+    await system(`systemctl enable --now vitastor-mon`);
 }
 
 function replace_env(text, key, value)
@@ -119,16 +423,29 @@ function replace_env(text, key, value)
 function select_local_etcd(etcds)
 {
     const ifaces = os.networkInterfaces();
+    const local = {};
     for (const ifname in ifaces)
+    {
         for (const iface of ifaces[ifname])
-            for (let i = 0; i < etcds.length; i++)
-                if (etcds[i] == iface.address.toLowerCase())
-                    return i;
+        {
+            const addr = iface.address;
+            if (iface.family == 'IPv6')
+                local[addr.toLowerCase()] = local['['+addr.toLowerCase()+']'] = true;
+            else
+                local[addr] = true;
+        }
+    }
+    for (let i = 0; i < etcds.length; i++)
+    {
+        if (local[etcds[i].addr])
+            return i;
+    }
     return -1;
 }
 
 async function system(cmd)
 {
+    console.log('Running '+cmd);
     const cp = child_process.spawn(cmd, { shell: true, stdio: [ 0, 1, 2 ] });
     let finish_cb;
     cp.on('exit', () => finish_cb && finish_cb());
@@ -136,3 +453,19 @@ async function system(cmd)
         await new Promise(ok => finish_cb = ok);
     return cp.exitCode;
 }
+
+async function make_ca(subj, filename)
+{
+    if (await system("openssl req -days 3650 -x509 -subj '"+subj+"' -addext basicConstraints=critical,CA:TRUE,pathlen:1"+
+        " -new -newkey rsa:4096 -nodes -keyout "+filename+".key -out "+filename+".crt"))
+        process.exit(1);
+}
+
+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 -CA ${ca}.crt -CAkey ${ca}.key -CAcreateserial -in ${f}.csr -out ${f}.crt`))
+        process.exit(1);
+    fs.unlinkSync(f+".csr");
+}

mon/vitastor_auth_filter.js

@@ -0,0 +1,539 @@
+// AntiEtcd authentication filter for Vitastor
+// (c) Vitaliy Filippov, 2026
+// License: Mozilla Public License 2.0 or Vitastor Network Public License 1.1
+
+// Permissions are based on:
+// 1. Users.
+//    Stored in /vitastor/config/user/<username>.
+//    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:
+//    - owner (user name)
+//    - owner_group (group name)
+//    - reader_group
+// 3. Certificates.
+//    - osd, mon use their own trusted certificates.
+
+const { X509Certificate } = require('node:crypto');
+
+const static_perms = {
+    invalid: {
+        keys: {},
+        prefixes: {},
+    },
+    osd: {
+        keys: { '/pg/config': false },
+        prefixes: { '/config/': false, '/osd/': true, '/pg/state/': true, '/pg/history/': true, '/pgstats/': true },
+    },
+    mon: {
+        keys: { '/pg/config': true, '/stats': true, '/history/last_clean_pgs': true },
+        prefixes: {
+            '/config/': false, '/osd/': false, '/mon/': true, '/pg/history/': true,
+            '/pgstats/': false, '/inode/stats/': true, '/pool/stats/': true,
+        },
+    },
+    admin: {
+        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,
+        },
+    },
+    client: {
+        keys: { '/config/global': false, '/config/node_placement': false, '/config/pools': false, '/pg/config': false },
+        prefixes: { '/osd/state/': false, '/pg/state/': false, '/index/maxid/': false },
+    },
+};
+
+const api_perms = {
+    osd: { lease_grant: true, lease_revoke: true, lease_keepalive: true, maintenance_status: true },
+    mon: { lease_grant: true, lease_revoke: true, lease_keepalive: true, maintenance_status: true },
+    admin: { maintenance_status: true },
+    client: { maintenance_status: true },
+};
+
+class VitastorAuthFilter
+{
+    constructor(antietcd)
+    {
+        this.cfg = antietcd.cfg;
+        this.antietcd = antietcd;
+        this.prefix = this.cfg.vitastor_prefix || '/vitastor';
+        this.prefix_parts = this.prefix.split('/');
+    }
+
+    async init()
+    {
+        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 mon_ca');
+        }
+        this.osd_ca = await this.antietcd.readPEM(this.cfg.osd_ca);
+        this.osd_ca_obj = new X509Certificate(this.osd_ca);
+        this.antietcd.tls.ca.push(this.osd_ca);
+        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_obj);
+            this.antietcd.tls.ca.push(this.mon_ca);
+        }
+    }
+
+    init_context(context, clientCert)
+    {
+        let cert = clientCert;
+        while (cert)
+        {
+            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;
+        }
+    }
+
+    _get(path, decode)
+    {
+        let cur = this.antietcd.etctree.state;
+        path = path instanceof Array ? path : path.split('/');
+        for (const p of path)
+        {
+            if (!cur.children)
+            {
+                return null;
+            }
+            cur = cur.children[p];
+            if (!cur)
+            {
+                return null;
+            }
+        }
+        if (decode)
+        {
+            return this._decode(path, cur.value);
+        }
+        return cur;
+    }
+
+    _decode(path, cur)
+    {
+        if (!cur)
+        {
+            return null;
+        }
+        if (cur)
+        {
+            try
+            {
+                cur = JSON.parse(cur);
+            }
+            catch (e)
+            {
+                console.warn('Invalid JSON in '+(path instanceof Array ? path.join('/') : path)+': '+e);
+            }
+        }
+        return cur;
+    }
+
+    // userInfo: { name: string, type: string, perms: static_perms[type], groups: { [string]: true } }
+    _check_compare(check, userInfo, checked)
+    {
+        let key = String(check.key);
+        if (key.substr(0, this.prefix.length) !== this.prefix)
+        {
+            return false;
+        }
+        key = key.substr(this.prefix.length);
+        if (key in userInfo.perms.keys)
+        {
+            return true;
+        }
+        for (const pfx in userInfo.perms.prefixes)
+        {
+            if (key.substr(0, pfx.length) == pfx)
+            {
+                return true;
+            }
+        }
+        if (userInfo.type == 'client')
+        {
+            // Image permissions
+            if (key.substr(0, 14) == '/config/inode/')
+            {
+                // Allowed to check that a key does not exist
+                if (check.target == 'VERSION' && check.version == 0)
+                {
+                    checked['M'+key] = true;
+                    return true;
+                }
+                else if (check.target == 'MOD')
+                {
+                    const data = this._get(check.key);
+                    if (!data || data.mod_revision != check.mod_revision)
+                    {
+                        // Break check to trigger CAS failure
+                        check.mod_revision = '18446744073709551615'; // UINT64_MAX
+                        return true;
+                    }
+                    const inode = this._decode(check.key, data.value);
+                    if (inode && (inode.owner_group && userInfo.groups[inode.owner_group] ||
+                        inode.owner === userInfo.name))
+                    {
+                        checked['M'+key] = true;
+                        return true;
+                    }
+                }
+                return false;
+            }
+            if (key.substr(0, 13) == '/index/image/')
+            {
+                // Allowed to check that a key does not exist
+                if (check.target == 'VERSION' && check.version == 0)
+                {
+                    checked['M'+key] = true;
+                    return true;
+                }
+                else if (check.target == 'MOD')
+                {
+                    let data = this._get(check.key);
+                    if (!data || data.mod_revision != check.mod_revision)
+                    {
+                        // Break check to trigger CAS failure
+                        check.mod_revision = '18446744073709551615'; // UINT64_MAX
+                        return true;
+                    }
+                    data = this._decode(check.key, data.value);
+                    if (data)
+                    {
+                        const inode = this._get([ ...this.prefix_parts, 'config', 'inode', data.pool_id, data.id ], true);
+                        if (inode && (inode.owner_group && userInfo.groups[inode.owner_group] ||
+                            inode.owner === userInfo.name))
+                        {
+                            checked['M'+key] = true;
+                            return true;
+                        }
+                    }
+                }
+                return false;
+            }
+            if (key.substr(0, 13) == '/index/maxid/')
+            {
+                const pool_id = key.substr(13);
+                const pool_cfg = this._get([ ...this.prefix_parts, 'config', 'pools' ], true);
+                if (!pool_cfg || !pool_cfg[pool_id] || !pool_cfg[pool_id].creator_group || !userInfo.groups[pool_cfg[pool_id].creator_group])
+                {
+                    return false;
+                }
+                if (check.target == 'VERSION' && check.version == 0)
+                {
+                    checked['I'+parseInt(key.substr(13))+'_0'] = true;
+                    return true;
+                }
+                else if (check.target == 'MOD')
+                {
+                    const data = this._get(check.key);
+                    if (!data || data.mod_revision != check.mod_revision)
+                    {
+                        // Break check to trigger CAS failure
+                        check.mod_revision = '18446744073709551615'; // UINT64_MAX
+                        return true;
+                    }
+                    checked['I'+parseInt(key.substr(13))+'_'+data.value] = true;
+                    return true;
+                }
+                return false;
+            }
+        }
+        return false;
+    }
+
+    _check_read(kv, userInfo)
+    {
+        let key = String(kv.key);
+        if (key.substr(0, this.prefix.length) !== this.prefix)
+        {
+            return false;
+        }
+        key = key.substr(this.prefix.length);
+        if (key in userInfo.perms.keys)
+        {
+            return true;
+        }
+        for (const pfx in userInfo.perms.prefixes)
+        {
+            if (key.substr(0, pfx.length) == pfx)
+            {
+                return true;
+            }
+        }
+        if (userInfo.type == 'client')
+        {
+            // Image permissions
+            if (key.substr(0, 14) == '/config/inode/')
+            {
+                const inode = this._decode(kv.key, kv.value);
+                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 false;
+            }
+            if (key.substr(0, 13) == '/index/image/')
+            {
+                const data = this._decode(kv.key, kv.value);
+                const inode = this._get([ ...this.prefix_parts, 'config', 'inode', data.pool_id, data.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 false;
+            }
+            if (key.substr(0, 13) == '/inode/stats/')
+            {
+                const [ pool_id, id ] = key.substr(13).split('/');
+                const inode = this._get([ ...this.prefix_parts, 'config', 'inode', pool_id, 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 false;
+            }
+        }
+        return false;
+    }
+
+    _check_write(put, userInfo, checked)
+    {
+        let key = String(put.key);
+        if (key.substr(0, this.prefix.length) !== this.prefix)
+        {
+            return false;
+        }
+        key = key.substr(this.prefix.length);
+        if (userInfo.perms.keys[key])
+        {
+            return true;
+        }
+        for (const pfx in userInfo.perms.prefixes)
+        {
+            if (userInfo.perms.prefixes[pfx] && key.substr(0, pfx.length) == pfx)
+            {
+                return true;
+            }
+        }
+        if (checked && userInfo.type == 'client')
+        {
+            if (key.substr(0, 13) == '/index/maxid/' &&
+                checked['I'+parseInt(key.substr(13))+'_'+(put.value-1)])
+            {
+                // Allowed to increment maxid
+                return true;
+            }
+            if (checked['M'+key])
+            {
+                // Allowed to modify known images with CAS checks
+                return true;
+            }
+        }
+        return false;
+    }
+
+    _check_req(req, userInfo, checked)
+    {
+        let r;
+        if ((r = (req.request_range || req.requestRange)))
+        {
+            // All range queries are allowed, but responses are filtered - it's simpler
+        }
+        else if ((r = (req.request_put || req.requestPut)))
+        {
+            if (!this._check_write(r, userInfo, checked))
+                return false;
+        }
+        else if ((r = (req.request_delete_range || req.requestDeleteRange)))
+        {
+            if (!r.range_end || r.range_end === r.key)
+            {
+                if (!this._check_write({ key: r.key }, userInfo))
+                    return false;
+            }
+            else
+            {
+                // All keys in range must satisfy prefix
+                r.range_end = String(r.range_end);
+                if (r.key.length != r.range_end.length ||
+                    r.key[r.key.length-1] != '/' ||
+                    r.range_end[r.range_end.length-1] != '0')
+                {
+                    return false;
+                }
+                let key = r.key.substr(this.prefix.length);
+                let found = false;
+                for (const pfx in userInfo.perms.prefixes)
+                {
+                    if (userInfo.perms.prefixes[pfx] && key.substr(0, pfx.length) == pfx)
+                    {
+                        found = true;
+                        break;
+                    }
+                }
+                if (!found)
+                    return false;
+            }
+        }
+        return true;
+    }
+
+    _get_user(context)
+    {
+        if (context.user_type === 'osd' || context.user_type === 'mon')
+        {
+            return {
+                name: context.user_type,
+                type: context.user_type,
+                perms: static_perms[context.user_type],
+            };
+        }
+        if (!context.username)
+        {
+            return {};
+        }
+        let userInfo = this._get([ ...this.prefix_parts, 'config', 'user', context.username ], true);
+        if (!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;
+        if (userInfo.groups instanceof Array)
+        {
+            userInfo.groups = userInfo.groups.reduce((a, c) => { a[c] = true; return a; }, {});
+        }
+        else
+        {
+            userInfo.groups = {};
+        }
+        return userInfo;
+    }
+
+    filter_api(context, api/*, data*/)
+    {
+        let type = 'client';
+        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];
+    }
+
+    filter_txn(context, txn)
+    {
+        const userInfo = this._get_user(context);
+        if (!userInfo)
+        {
+            return null;
+        }
+        const checked = {};
+        if (txn.compare)
+        {
+            for (const check of txn.compare)
+            {
+                if (!this._check_compare(check, userInfo, checked))
+                    return null;
+            }
+        }
+        // Special transactions:
+        // 1. create image: create config/inode and index/image, increment index/maxid/<pool> (with CAS)
+        // 2. create snapshot: same as create image but also rename previous to @snap
+        if (txn.success)
+        {
+            for (const req of txn.success)
+            {
+                if (!this._check_req(req, userInfo, checked))
+                    return null;
+            }
+        }
+        if (txn.failure)
+        {
+            for (const req of txn.failure)
+            {
+                if (!this._check_req(req, userInfo, null))
+                    return null;
+            }
+        }
+        return txn;
+    }
+
+    filter_txn_response(context, txn, res)
+    {
+        if (!res.responses)
+        {
+            return;
+        }
+        const userInfo = this._get_user(context);
+        if (!userInfo)
+        {
+            for (const resp of res.responses)
+            {
+                if (resp.response_range && resp.response_range.kvs)
+                {
+                    resp.response_range.kvs = [];
+                }
+            }
+            return;
+        }
+        for (const resp of res.responses)
+        {
+            if (resp.response_range && resp.response_range.kvs)
+            {
+                resp.response_range.kvs = resp.response_range.kvs.filter(kv => this._check_read(kv, userInfo));
+            }
+        }
+    }
+
+    filter_watch_message(context, msg)
+    {
+        if (!msg.result || !msg.result.events)
+        {
+            return;
+        }
+        const userInfo = this._get_user(context);
+        if (!userInfo)
+        {
+            msg.result.events = [];
+            return;
+        }
+        msg.result.events = msg.result.events.filter(ev => this._check_read(ev.kv, userInfo));
+    }
+}
+
+module.exports = VitastorAuthFilter;

node-binding/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vitastor",
-  "version": "3.0.13",
+  "version": "3.0.15",
   "description": "Low-level native bindings to Vitastor client library",
   "main": "index.js",
   "keywords": [

patches/VitastorPlugin.pm

@@ -366,15 +366,38 @@ sub map_volume
     my $prefix = defined $scfg->{vitastor_prefix} ? $scfg->{vitastor_prefix} : 'pve/';
 
     my ($vtype, $img_name, $vmid) = $class->parse_volname($volname);
-    my $name = $img_name;
+    my $name = $prefix.$img_name;
     $name .= '@'.$snapname if $snapname;
 
     my $mapped = run_cli($scfg, [ 'ls' ], binary => '/usr/bin/vitastor-nbd');
-    my ($kerneldev) = grep { $mapped->{$_}->{image} eq $prefix.$name } keys %$mapped;
-    return $kerneldev if $kerneldev && -b $kerneldev; # already mapped
+    my ($kerneldev) = grep {
+        $mapped->{$_} && $mapped->{$_}->{image} && $mapped->{$_}->{image} eq $name
+    } keys %$mapped;
 
-    $kerneldev = run_cli($scfg, [ 'map', '--image', $prefix.$name ], binary => '/usr/bin/vitastor-nbd', json => 0);
-    return $kerneldev;
+    if ($kerneldev && -b $kerneldev)
+    {
+        my $size = `/usr/sbin/blockdev --getsize64 $kerneldev`;
+        return $kerneldev if $size && $size > 0;
+    }
+
+    my $map_out = run_cli($scfg, [ 'map', '--image', $name ], binary => '/usr/bin/vitastor-nbd', json => 0);
+    $map_out =~ s/^\s+|\s+$//gso;
+
+    # Wait until the device is started
+    for (my $i = 0; $i < 100; $i++)
+    {
+        $mapped = run_cli($scfg, [ 'ls' ], binary => '/usr/bin/vitastor-nbd');
+        ($kerneldev) = grep { $mapped->{$_} && $mapped->{$_}->{image} && $mapped->{$_}->{image} eq $name } keys %$mapped;
+        if ($kerneldev && -b $kerneldev)
+        {
+            my $size = `/usr/sbin/blockdev --getsize64 $kerneldev`;
+            return $kerneldev if $size && $size > 0;
+        }
+        select(undef, undef, undef, 0.1);
+    }
+
+    die "Failed to map Vitastor image $name via NBD".
+        ($map_out ? ", vitastor-nbd map returned '$map_out'" : "")."\n";
 }
 
 sub unmap_volume
@@ -383,13 +406,19 @@ sub unmap_volume
     my $prefix = defined $scfg->{vitastor_prefix} ? $scfg->{vitastor_prefix} : 'pve/';
 
     my ($vtype, $name, $vmid) = $class->parse_volname($volname);
+    $name = $prefix.$name;
     $name .= '@'.$snapname if $snapname;
-
     my $mapped = run_cli($scfg, [ 'ls' ], binary => '/usr/bin/vitastor-nbd');
-    my ($kerneldev) = grep { $mapped->{$_}->{image} eq $prefix.$name } keys %$mapped;
-    if ($kerneldev && -b $kerneldev)
+
+    my @kerneldevs = grep {
+        $mapped->{$_} && $mapped->{$_}->{image} && $mapped->{$_}->{image} eq $name
+    } keys %$mapped;
+
+    for my $kerneldev (@kerneldevs)
     {
-        run_cli($scfg, [ 'unmap', $kerneldev ], binary => '/usr/bin/vitastor-nbd', json => 0);
+        next if !$kerneldev || !-b $kerneldev;
+        eval { run_cli($scfg, [ 'unmap', $kerneldev ], binary => '/usr/bin/vitastor-nbd', json => 0); };
+        warn "Failed to unmap Vitastor image $name from $kerneldev: $@" if $@;
     }
 
     return 1;
@@ -405,7 +434,13 @@ sub activate_volume
 sub deactivate_volume
 {
     my ($class, $storeid, $scfg, $volname, $snapname, $cache) = @_;
-    $class->unmap_volume($storeid, $scfg, $volname, $snapname) if $scfg->{vitastor_nbd};
+
+    # Even with vitastor_nbd=0, Proxmox may call map_volume() for special
+    # volumes like tpmstate0 because swtpm needs a local file/block path.
+    # Therefore, always try to unmap an existing NBD mapping here.
+    # unmap_volume() is a no-op if the volume is not currently mapped.
+    $class->unmap_volume($storeid, $scfg, $volname, $snapname);
+
     return 1;
 }
 

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 = '3.0.13'
+VITASTOR_VERSION = '3.0.15'
 
 LOG = logging.getLogger(__name__)
 

rpm/vitastor-el10.Dockerfile

@@ -11,7 +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 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
+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

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        3.0.13
+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-3.0.13.el10.tar.gz
+Source0:        vitastor-3.0.15.el10.tar.gz
 
 BuildRequires:  gperftools-devel
 BuildRequires:  gcc-c++
@@ -16,6 +16,7 @@ BuildRequires:  gf-complete-devel
 BuildRequires:  rdma-core-devel
 BuildRequires:  cmake
 BuildRequires:  libnl3-devel
+BuildRequires:  c-ares-devel
 Requires:       vitastor-osd = %{version}-%{release}
 Requires:       vitastor-mon = %{version}-%{release}
 Requires:       vitastor-client = %{version}-%{release}

rpm/vitastor-el7.Dockerfile

@@ -15,7 +15,7 @@ RUN yum -y --enablerepo=extras install centos-release-scl epel-release yum-utils
 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-1.0-1.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 gf-complete-devel rdma-core-devel libnl3-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

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        3.0.13
+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-3.0.13.el7.tar.gz
+Source0:        vitastor-3.0.15.el7.tar.gz
 
 BuildRequires:  gperftools-devel
 BuildRequires:  devtoolset-9-gcc-c++
@@ -17,6 +17,7 @@ BuildRequires:  gf-complete-devel
 BuildRequires:  rdma-core-devel
 BuildRequires:  cmake3
 BuildRequires:  libnl3-devel
+BuildRequires:  c-ares-devel
 Requires:       vitastor-osd = %{version}-%{release}
 Requires:       vitastor-mon = %{version}-%{release}
 Requires:       vitastor-client = %{version}-%{release}

rpm/vitastor-el8.Dockerfile

@@ -13,7 +13,7 @@ RUN dnf -y install centos-release-advanced-virtualization epel-release dnf-plugi
 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-1.0-1.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
+    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

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        3.0.13
+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-3.0.13.el8.tar.gz
+Source0:        vitastor-3.0.15.el8.tar.gz
 
 BuildRequires:  gperftools-devel
 BuildRequires:  gcc-toolset-9-gcc-c++
@@ -16,6 +16,7 @@ BuildRequires:  gf-complete-devel
 BuildRequires:  rdma-core-devel
 BuildRequires:  cmake
 BuildRequires:  libnl3-devel
+BuildRequires:  c-ares-devel
 Requires:       vitastor-osd = %{version}-%{release}
 Requires:       vitastor-mon = %{version}-%{release}
 Requires:       vitastor-client = %{version}-%{release}

rpm/vitastor-el9.Dockerfile

@@ -10,7 +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 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
+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

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        3.0.13
+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-3.0.13.el9.tar.gz
+Source0:        vitastor-3.0.15.el9.tar.gz
 
 BuildRequires:  gperftools-devel
 BuildRequires:  gcc-c++
@@ -16,6 +16,7 @@ BuildRequires:  gf-complete-devel
 BuildRequires:  rdma-core-devel
 BuildRequires:  cmake
 BuildRequires:  libnl3-devel
+BuildRequires:  c-ares-devel
 Requires:       vitastor-osd = %{version}-%{release}
 Requires:       vitastor-mon = %{version}-%{release}
 Requires:       vitastor-client = %{version}-%{release}

src/CMakeLists.txt

@@ -20,7 +20,7 @@ if("${CMAKE_INSTALL_PREFIX}" MATCHES "^/usr/local/?$")
 endif()
 set(ENABLE_COVERAGE false CACHE BOOL "Enable code coverage")
 
-add_definitions(-DVITASTOR_VERSION="3.0.13")
+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})
@@ -69,11 +69,21 @@ pkg_check_modules(ISAL libisal)
 if (ISAL_LIBRARIES)
 	add_definitions(-DWITH_ISAL)
 endif (ISAL_LIBRARIES)
+pkg_check_modules(ISAL_CRYPTO libisal_crypto)
+if (ISAL_CRYPTO_LIBRARIES)
+	add_definitions(-DWITH_ISAL_CRYPTO)
+endif (ISAL_CRYPTO_LIBRARIES)
 pkg_check_modules(RDMACM librdmacm)
 if (RDMACM_LIBRARIES)
 	add_definitions(-DWITH_RDMACM)
 endif (RDMACM_LIBRARIES)
 
+find_package(OpenSSL REQUIRED)
+add_definitions(-DWITH_OPENSSL)
+
+pkg_check_modules(CARES REQUIRED libcares)
+include_directories(${CARES_INCLUDE_DIRS})
+
 if (${WITH_SYSTEM_LIBURING})
 	pkg_check_modules(LIBURING REQUIRED liburing>=2.10)
 	include_directories(${LIBURING_INCLUDE_DIRS})

src/blockstore/blockstore_disk.cpp

@@ -83,13 +83,17 @@ void blockstore_disk_t::parse_config(std::map<std::string, std::string> & config
     {
         data_csum_type = BLOCKSTORE_CSUM_CRC32C;
     }
+    else if (config["data_csum_type"] == "xxh3_32")
+    {
+        data_csum_type = BLOCKSTORE_CSUM_XXH3_32;
+    }
     else if (config["data_csum_type"] == "" || config["data_csum_type"] == "none")
     {
         data_csum_type = BLOCKSTORE_CSUM_NONE;
     }
     else
     {
-        throw std::runtime_error("data_csum_type="+config["data_csum_type"]+" is unsupported, only \"crc32c\" and \"none\" are supported");
+        throw std::runtime_error("data_csum_type="+config["data_csum_type"]+" is unsupported, only \"crc32c\", \"xxh3_32\" and \"none\" are supported");
     }
     csum_block_size = parse_size(config["csum_block_size"]);
     discard_on_start = config.find("discard_on_start") != config.end() &&
@@ -541,17 +545,21 @@ int blockstore_disk_t::trim_data(std::function<bool(uint64_t)> is_used)
                     if (range[0] % discard_granularity)
                         range[0] = range[0] + discard_granularity - (range[0] % discard_granularity);
                     if (range[0] >= range[1])
-                        continue;
-                    range[1] -= range[0];
+                        range[1] = 0;
+                    else
+                        range[1] -= range[0];
                 }
-                r = ioctl(data_fd, BLKDISCARD, &range);
-                if (r != 0)
+                if (range[1] > 0)
                 {
-                    fprintf(stderr, "Failed to execute BLKDISCARD %ju+%ju on %s: %s (code %d)\n",
-                        range[0], range[1], data_device.c_str(), strerror(-r), r);
-                    return -errno;
+                    r = ioctl(data_fd, BLKDISCARD, &range);
+                    if (r != 0)
+                    {
+                        fprintf(stderr, "Failed to execute BLKDISCARD %ju+%ju on %s: %s (code %d)\n",
+                            range[0], range[1], data_device.c_str(), strerror(-r), r);
+                        return -errno;
+                    }
+                    discarded += range[1];
                 }
-                discarded += range[1];
             }
             j = i+1;
         }

src/blockstore/blockstore_disk.h

@@ -16,6 +16,7 @@
 #define BLOCKSTORE_CSUM_NONE 0
 // Lower byte of checksum type is its length
 #define BLOCKSTORE_CSUM_CRC32C 0x104
+#define BLOCKSTORE_CSUM_XXH3_32 0x204
 
 #define MOCK_DATA_FD 1000
 #define MOCK_META_FD 1001

src/blockstore/blockstore_heap.cpp

@@ -12,6 +12,7 @@
 #include "blockstore_heap.h"
 #include "../util/allocator.h"
 #include "../util/crc32c.h"
+#include "../util/xxh_x86dispatch.h"
 #include "../util/malloc_or_die.h"
 
 #define BS_HEAP_FREE_MVCC 1
@@ -22,7 +23,7 @@
 
 #define HEAP_INFLIGHT_DONE 1
 #define HEAP_INFLIGHT_COMPACTABLE 2
-#define HEAP_INFLIGHT_COMPACTED 4
+#define HEAP_INFLIGHT_OVERWRITE 4
 #define HEAP_INFLIGHT_GC 8
 #define HEAP_INFLIGHT_EXPLICIT 16
 
@@ -65,19 +66,19 @@ uint32_t blockstore_heap_t::get_simple_entry_size()
 uint32_t blockstore_heap_t::get_big_entry_size()
 {
     return sizeof(heap_big_write_t) + dsk->clean_entry_bitmap_size*2 +
-        (!dsk->data_csum_type ? 0 : dsk->data_block_size/dsk->csum_block_size * (dsk->data_csum_type & 0xFF));
+        (!dsk->csum_block_size ? 0 : dsk->data_block_size/dsk->csum_block_size * (dsk->data_csum_type & 0xFF));
 }
 
 uint32_t blockstore_heap_t::get_big_intent_entry_size()
 {
     return sizeof(heap_big_intent_t) + dsk->clean_entry_bitmap_size*2 +
-        (!dsk->data_csum_type ? 4 : dsk->data_block_size/dsk->csum_block_size * (dsk->data_csum_type & 0xFF));
+        (!dsk->csum_block_size ? 4 : dsk->data_block_size/dsk->csum_block_size * (dsk->data_csum_type & 0xFF));
 }
 
 uint32_t blockstore_heap_t::get_small_entry_size(uint32_t offset, uint32_t len)
 {
     return sizeof(heap_small_write_t) + dsk->clean_entry_bitmap_size +
-        (!dsk->data_csum_type ? 4 : (dsk->data_csum_type & 0xFF) *
+        (!dsk->csum_block_size ? 4 : (dsk->data_csum_type & 0xFF) *
             ((offset+len+dsk->csum_block_size-1)/dsk->csum_block_size - offset/dsk->csum_block_size));
 }
 
@@ -92,7 +93,7 @@ uint32_t blockstore_heap_t::get_csum_size(heap_entry_t *wr)
 
 uint32_t blockstore_heap_t::get_csum_size(uint32_t entry_type, uint32_t offset, uint32_t len)
 {
-    if (!dsk->data_csum_type)
+    if (!dsk->csum_block_size)
     {
         return 0;
     }
@@ -122,6 +123,8 @@ uint32_t heap_entry_t::get_size(blockstore_heap_t *heap)
     }
     if (type() == BS_HEAP_SMALL_WRITE || type() == BS_HEAP_INTENT_WRITE)
     {
+        if (size < sizeof(heap_small_write_t))
+            return heap->get_small_entry_size(0, 0);
         return heap->get_small_entry_size(small().offset, small().len);
     }
     return heap->get_simple_entry_size();
@@ -215,15 +218,24 @@ void heap_entry_t::set_big_location(blockstore_heap_t *heap, uint64_t location)
     big().block_num = location / heap->dsk->data_block_size;
 }
 
-uint32_t heap_entry_t::calc_crc32c()
+uint32_t heap_entry_t::calc_checksum(blockstore_disk_t *dsk)
 {
-    auto old_crc32c = crc32c;
-    crc32c = 0;
-    uint32_t res = ::crc32c(0, (uint8_t*)this, size);
-    crc32c = old_crc32c;
+    auto old_checksum = checksum;
+    checksum = 0;
+    uint32_t res = 0;
+    if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32)
+        res = (uint32_t)XXH3_64bits(this, size);
+    else
+        res = ::crc32c(0, (uint8_t*)this, size);
+    checksum = old_checksum;
     return res;
 }
 
+uint32_t heap_entry_t::calc_checksum(blockstore_heap_t *heap)
+{
+    return calc_checksum(heap->dsk);
+}
+
 uint64_t blockstore_heap_t::get_pg_id(inode_t inode, uint64_t stripe)
 {
     uint64_t pg_num = 0;
@@ -364,14 +376,11 @@ corrupted_object:
                     return EDOM;
                 }
             }
-            if (((wr->entry_type & BS_HEAP_TYPE) == BS_HEAP_SMALL_WRITE ||
-                (wr->entry_type & BS_HEAP_TYPE) == BS_HEAP_INTENT_WRITE) &&
-                wr->size < sizeof(heap_small_write_t))
+            if (wr->size != wr->get_size(this))
             {
-                // Small writes require accessing offset & len to calculate correct length,
-                // so require at least sizeof(heap_small_write_t) for them
-                fprintf(stderr, "Error: entry %jx:%jx v%ju has invalid size in metadata block %u at %u (%u < min %zu bytes)\n",
-                    wr->inode, wr->stripe, wr->version, block_num, block_offset, wr->size, sizeof(heap_small_write_t));
+                // Check entry size
+                fprintf(stderr, "Error: entry %jx:%jx v%ju has invalid size in metadata block %u at %u (%u != %u bytes)\n",
+                    wr->inode, wr->stripe, wr->version, block_num, block_offset, wr->size, wr->get_size(this));
                 goto corrupted_object;
             }
             if (wr->entry_type == BS_HEAP_COMMIT && !wr->version)
@@ -381,12 +390,12 @@ corrupted_object:
                 goto corrupted_object;
             }
             // Verify crc
-            uint32_t expected_crc32c = wr->calc_crc32c();
-            if (wr->crc32c != expected_crc32c)
+            uint32_t expected_checksum = wr->calc_checksum(this);
+            if (wr->checksum != expected_checksum)
             {
-                fprintf(stderr, "Error: entry %jx:%jx v%ju l%ju in metadata block %u at %u is corrupt (crc32c mismatch: expected %08x, got %08x). ",
+                fprintf(stderr, "Error: entry %jx:%jx v%ju l%ju in metadata block %u at %u is corrupt (checksum mismatch: expected %08x, got %08x). ",
                     wr->inode, wr->stripe, wr->version, wr->lsn,
-                    block_num, block_offset, expected_crc32c, wr->crc32c);
+                    block_num, block_offset, expected_checksum, wr->checksum);
                 goto corrupted_object;
             }
             // Verify offset & len
@@ -683,10 +692,14 @@ int blockstore_heap_t::mark_used_blocks()
                         }
                         use_data(wr->inode, wr->big_location(this));
                     }
-                    if (wr->is_compactable() && !added)
+                    if (wr->is_compactable())
                     {
-                        compact_queue.push_back((object_id){ .inode = wr->inode, .stripe = wr->stripe });
-                        added = true;
+                        to_compact_count++;
+                        if (!added)
+                        {
+                            compact_queue.push_back((object_id){ .inode = wr->inode, .stripe = wr->stripe });
+                            added = true;
+                        }
                     }
                     if (wr->is_overwrite())
                     {
@@ -696,6 +709,11 @@ int blockstore_heap_t::mark_used_blocks()
             });
         }
     }
+    for (auto li: init_erase_items)
+    {
+        unlink_list_item(li);
+    }
+    init_erase_items.clear();
     if (dsk->gc_on_start)
     {
         recheck_full_gc();
@@ -731,7 +749,6 @@ void blockstore_heap_t::init_erase_bad_entry(heap_list_item_t *li)
         inf.garbage_space -= (li->entry.is_garbage() ? li->entry.size : 0);
     });
     recheck_modified_blocks.insert(li->block_num);
-    unlink_list_item(li);
 }
 
 bool blockstore_heap_t::init_erase_double_claim(heap_list_item_t *prev_li, heap_list_item_t *cur_li)
@@ -786,6 +803,8 @@ bool blockstore_heap_t::init_erase_double_claim(heap_list_item_t *prev_li, heap_
                 overwritten = erase_li->entry.is_overwrite();
             }
             init_erase_bad_entry(erase_li);
+            // Can't erase (mutate map) while iterating, so postpone it
+            init_erase_items.push_back(erase_li);
             erase_li = prev_erase_li;
         }
     }
@@ -799,6 +818,8 @@ bool blockstore_heap_t::init_erase_double_claim(heap_list_item_t *prev_li, heap_
             auto next_erase_li = erase_li->next;
             init_free_bad_entry(&erase_li->entry);
             init_erase_bad_entry(erase_li);
+            // Can't erase (mutate map) while iterating, so postpone it
+            init_erase_items.push_back(erase_li);
             erase_li = next_erase_li;
         }
         erase_li = cur_li;
@@ -807,6 +828,8 @@ bool blockstore_heap_t::init_erase_double_claim(heap_list_item_t *prev_li, heap_
         {
             auto prev_erase_li = erase_li->prev;
             init_erase_bad_entry(erase_li);
+            // Can't erase (mutate map) while iterating, so postpone it
+            init_erase_items.push_back(erase_li);
             erase_li = prev_erase_li;
         }
     }
@@ -877,6 +900,7 @@ void blockstore_heap_t::recheck_drop_entries(heap_entry_t *obj, heap_entry_t *ba
         auto prev = li->prev;
         assert(li->entry.type() == bad_wr->type());
         init_erase_bad_entry(li);
+        unlink_list_item(li);
         li = prev;
     }
 }
@@ -1069,7 +1093,11 @@ bool blockstore_heap_t::calc_checksums(heap_entry_t *wr, uint8_t *data, bool set
             len = wr->big_intent().len;
         else
             assert(0);
-        uint32_t real_csum = crc32c(0, data, len);
+        uint32_t real_csum = 0;
+        if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32)
+            real_csum = (uint32_t)XXH3_64bits(data, len);
+        else
+            real_csum = crc32c(0, data, len);
         if (set)
         {
             *wr_csum = real_csum;
@@ -1119,11 +1147,26 @@ static uint32_t crc32c_iter(uint32_t prev_crc, const std::function<uint8_t*(uint
     return prev_crc;
 }
 
+static void xxh3_iter(XXH3_state_t* xxh3_state, const std::function<uint8_t*(uint32_t start, uint32_t & len)> & next, uint32_t pos, uint32_t size)
+{
+    uint32_t cur_len = 0;
+    while (size > 0)
+    {
+        uint8_t *data = next(pos, cur_len);
+        assert(data);
+        cur_len = (cur_len < size ? cur_len : size);
+        XXH3_64bits_update(xxh3_state, data, cur_len);
+        pos += cur_len;
+        size -= cur_len;
+    }
+}
+
 bool blockstore_heap_t::calc_block_checksums(uint32_t *block_csums, uint8_t *bitmap,
     uint32_t start, uint32_t end, std::function<uint8_t*(uint32_t start, uint32_t & len)> next,
     bool set, std::function<void(uint32_t, uint32_t, uint32_t)> bad_block_cb)
 {
     bool res = true;
+    XXH3_state_t* xxh3_state = NULL;
     uint32_t pos = start;
     uint32_t block_end = (start/dsk->csum_block_size + 1)*dsk->csum_block_size;
     uint32_t block_crc = 0;
@@ -1139,43 +1182,90 @@ bool blockstore_heap_t::calc_block_checksums(uint32_t *block_csums, uint8_t *bit
                 while (pos < end && pos < block_end && !(bitmap[pos/dsk->bitmap_granularity/8] & (1 << ((pos/dsk->bitmap_granularity) % 8))))
                     pos += dsk->bitmap_granularity;
                 // zero padding at the beginning or at the end of the block is not counted
-                if (pos > prev && prev > 0 && pos < block_end)
-                    block_crc = crc32c_pad(block_crc, NULL, 0, pos-prev, 0);
+                if (pos > prev && prev > blk_start && pos < block_end)
+                {
+                    if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32)
+                    {
+                        if (!xxh3_state)
+                        {
+                            xxh3_state = XXH3_createState();
+                            XXH3_64bits_reset(xxh3_state);
+                        }
+                        uint32_t zeropad = pos-prev;
+                        while (zeropad > 0)
+                        {
+                            uint32_t zerolen = zeropad > 4096 ? 4096 : zeropad;
+                            XXH3_64bits_update(xxh3_state, zero_page, zerolen);
+                            zeropad -= zerolen;
+                        }
+                    }
+                    else
+                        block_crc = crc32c_pad(block_crc, NULL, 0, pos-prev, 0);
+                }
                 prev = pos;
                 while (pos < end && pos < block_end && (bitmap[pos/dsk->bitmap_granularity/8] & (1 << ((pos/dsk->bitmap_granularity) % 8))))
                     pos += dsk->bitmap_granularity;
                 if (pos > prev)
                 {
                     isset = true;
-                    block_crc = crc32c_iter(block_crc, next, prev, pos-prev);
+                    if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32)
+                    {
+                        if (!xxh3_state)
+                        {
+                            xxh3_state = XXH3_createState();
+                            XXH3_64bits_reset(xxh3_state);
+                        }
+                        xxh3_iter(xxh3_state, next, prev, pos-prev);
+                    }
+                    else
+                        block_crc = crc32c_iter(block_crc, next, prev, pos-prev);
                 }
                 prev = pos;
             }
         }
         else
         {
-            block_crc = crc32c_iter(block_crc, next, pos, (end > block_end ? block_end : end)-pos);
+            if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32)
+            {
+                if (!xxh3_state)
+                {
+                    xxh3_state = XXH3_createState();
+                    XXH3_64bits_reset(xxh3_state);
+                }
+                xxh3_iter(xxh3_state, next, pos, (end > block_end ? block_end : end)-pos);
+            }
+            else
+                block_crc = crc32c_iter(block_crc, next, pos, (end > block_end ? block_end : end)-pos);
             pos = (end > block_end ? block_end : end);
             isset = true;
         }
+        if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32 && xxh3_state)
+        {
+            block_crc = (uint32_t)XXH3_64bits_digest(xxh3_state);
+            XXH3_64bits_reset(xxh3_state);
+        }
         if (set)
         {
             *block_csums = block_crc;
         }
         else if (isset && block_crc != *block_csums)
         {
+            res = false;
             if (bad_block_cb)
-            {
                 bad_block_cb(blk_start, *block_csums, block_crc);
-                res = false;
-            }
             else
-                return false;
+                break;
         }
         block_end += dsk->csum_block_size;
         block_crc = 0;
         block_csums++;
     }
+    if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32 && xxh3_state)
+    {
+        block_crc = (uint32_t)XXH3_64bits_digest(xxh3_state);
+        XXH3_freeState(xxh3_state);
+        xxh3_state = NULL;
+    }
     return res;
 }
 
@@ -1537,12 +1627,12 @@ int blockstore_heap_t::add_entry(uint32_t wr_size, uint32_t *modified_block,
     // Remember the object as dirty and remove older entries when this block is written and fsynced
     push_inflight_lsn(next_lsn, new_wr,
         (explicit_complete ? HEAP_INFLIGHT_EXPLICIT : 0) |
-        (new_wr->is_overwrite() ? HEAP_INFLIGHT_COMPACTED : 0) |
+        (new_wr->is_overwrite() ? HEAP_INFLIGHT_OVERWRITE : 0) |
         (new_wr->is_compactable() ? HEAP_INFLIGHT_COMPACTABLE : 0));
     insert_list_items(&li, 1, false);
     li->block_num = block_num;
     new_wr->size = wr_size;
-    new_wr->crc32c = new_wr->calc_crc32c();
+    new_wr->checksum = new_wr->calc_checksum(this);
     return 0;
 }
 
@@ -1571,10 +1661,22 @@ int blockstore_heap_t::add_small_write(object_id oid, heap_entry_t **obj_ptr, ui
         wr->small().location = location;
         if (bitmap)
             memcpy(wr->get_ext_bitmap(this), bitmap, dsk->clean_entry_bitmap_size);
-        else if (obj)
-            memcpy(wr->get_ext_bitmap(this), obj->get_ext_bitmap(this), dsk->clean_entry_bitmap_size);
         else
-            memset(wr->get_ext_bitmap(this), 0, dsk->clean_entry_bitmap_size);
+        {
+            bool found = false;
+            iterate_with_stable(obj, UINT64_MAX, [&](heap_entry_t *old_wr, bool stable)
+            {
+                if (old_wr->get_ext_bitmap(this))
+                {
+                    found = true;
+                    memcpy(wr->get_ext_bitmap(this), old_wr->get_ext_bitmap(this), dsk->clean_entry_bitmap_size);
+                    return false;
+                }
+                return true;
+            });
+            if (!found)
+                memset(wr->get_ext_bitmap(this), 0, dsk->clean_entry_bitmap_size);
+        }
         calc_checksums(wr, (uint8_t*)data, true);
         *obj_ptr = wr;
     });
@@ -1602,7 +1704,7 @@ int blockstore_heap_t::add_big_write(object_id oid, heap_entry_t *old_head, bool
             memset(wr->get_ext_bitmap(this), 0, dsk->clean_entry_bitmap_size);
         memset(wr->get_int_bitmap(this), 0, dsk->clean_entry_bitmap_size);
         bitmap_set(wr->get_int_bitmap(this), offset, len, dsk->bitmap_granularity);
-        if (dsk->data_csum_type)
+        if (dsk->csum_block_size)
         {
             memset(wr->get_checksums(this), 0, get_csum_size(wr));
             calc_checksums(wr, (uint8_t*)data, true, offset, len);
@@ -1631,7 +1733,7 @@ int blockstore_heap_t::add_redirect_intent(object_id oid, heap_entry_t **obj_ptr
             memset(wr->get_ext_bitmap(this), 0, dsk->clean_entry_bitmap_size);
         memset(wr->get_int_bitmap(this), 0, dsk->clean_entry_bitmap_size);
         bitmap_set(wr->get_int_bitmap(this), offset, len, dsk->bitmap_granularity);
-        if (dsk->data_csum_type)
+        if (dsk->csum_block_size)
             memset(wr->get_checksums(this), 0, get_csum_size(wr));
         calc_checksums(wr, (uint8_t*)data, true);
         *obj_ptr = wr;
@@ -1669,7 +1771,7 @@ int blockstore_heap_t::add_big_intent(object_id oid, heap_entry_t **obj_ptr, uin
             memcpy(wr->get_ext_bitmap(this), obj->get_ext_bitmap(this), dsk->clean_entry_bitmap_size);
         memcpy(wr->get_int_bitmap(this), obj->get_int_bitmap(this), dsk->clean_entry_bitmap_size);
         bitmap_set(wr->get_int_bitmap(this), offset, len, dsk->bitmap_granularity);
-        if (dsk->data_csum_type)
+        if (dsk->csum_block_size)
         {
             if (checksums)
                 memcpy(wr->get_checksums(this), checksums, get_csum_size(wr));
@@ -1716,7 +1818,7 @@ int blockstore_heap_t::add_compact(heap_entry_t *obj, uint64_t compact_version,
         new_wr->set_big_location(this, compact_location);
         memcpy(new_wr->get_int_bitmap(this), new_int_bitmap, dsk->clean_entry_bitmap_size);
         memcpy(new_wr->get_ext_bitmap(this), new_ext_bitmap, dsk->clean_entry_bitmap_size);
-        if (dsk->data_csum_type && new_csums)
+        if (dsk->csum_block_size && new_csums)
             memcpy(new_wr->get_checksums(this), new_csums, dsk->data_block_size/dsk->csum_block_size*(dsk->data_csum_type & 0xFF));
     });
 }
@@ -1735,7 +1837,7 @@ int blockstore_heap_t::punch_holes(heap_entry_t *wr, uint8_t *new_bitmap, uint8_
     *modified_block = block_num;
     memcpy(wr->get_int_bitmap(this), new_bitmap, dsk->clean_entry_bitmap_size);
     memcpy(wr->get_checksums(this), new_csums, dsk->data_block_size/dsk->csum_block_size*(dsk->data_csum_type & 0xFF));
-    wr->crc32c = wr->calc_crc32c();
+    wr->checksum = wr->calc_checksum(dsk);
     return 0;
 }
 
@@ -2274,7 +2376,7 @@ void blockstore_heap_t::use_data(inode_t inode, uint64_t location)
 {
     auto sh_it = pool_shard_settings.find(INODE_POOL(inode));
     if (sh_it != pool_shard_settings.end() && sh_it->second.no_inode_stats)
-        inode = (INODE_POOL(inode) << POOL_ID_BITS);
+        inode = INODE_WITH_POOL(INODE_POOL(inode), 0);
     assert(!data_alloc->get(location / dsk->data_block_size));
     data_alloc->set(location / dsk->data_block_size, true);
     inode_space_stats[inode] += dsk->data_block_size;
@@ -2285,7 +2387,7 @@ void blockstore_heap_t::free_data(inode_t inode, uint64_t location)
 {
     auto sh_it = pool_shard_settings.find(INODE_POOL(inode));
     if (sh_it != pool_shard_settings.end() && sh_it->second.no_inode_stats)
-        inode = (INODE_POOL(inode) << POOL_ID_BITS);
+        inode = INODE_WITH_POOL(INODE_POOL(inode), 0);
     assert(data_alloc->get(location / dsk->data_block_size));
     data_alloc->set(location / dsk->data_block_size, false);
     auto sp_it = inode_space_stats.find(inode);
@@ -2322,7 +2424,8 @@ void blockstore_heap_t::use_buffer_area(inode_t inode, uint64_t location, uint64
         return;
     }
     assert(!(size % dsk->bitmap_granularity));
-    buffer_alloc->use(location / dsk->bitmap_granularity, size / dsk->bitmap_granularity);
+    bool ok = buffer_alloc->use(location / dsk->bitmap_granularity, size / dsk->bitmap_granularity);
+    assert(ok);
     buffer_area_used_space += size;
 }
 
@@ -2364,7 +2467,7 @@ void blockstore_heap_t::get_meta_block(uint32_t block_num, uint8_t *buffer)
     }
 }
 
-void blockstore_heap_t::fill_block_empty_space(uint8_t *buffer, uint32_t pos)
+void blockstore_heap_t::fill_block_empty_space(uint8_t *buffer, uint64_t pos)
 {
     if (pos > dsk->meta_block_size)
     {
@@ -2452,7 +2555,7 @@ uint64_t blockstore_heap_t::get_garbage_memory()
 void blockstore_heap_t::push_inflight_lsn(uint64_t lsn, heap_entry_t *wr, uint64_t flags)
 {
     uint64_t next_inf = first_inflight_lsn + inflight_lsn.size();
-    if (flags & (HEAP_INFLIGHT_COMPACTABLE|HEAP_INFLIGHT_COMPACTED))
+    if (flags & (HEAP_INFLIGHT_COMPACTABLE|HEAP_INFLIGHT_OVERWRITE))
     {
         to_compact_count++;
     }
@@ -2519,7 +2622,7 @@ void blockstore_heap_t::mark_lsn_fsynced(uint64_t lsn)
 void blockstore_heap_t::apply_inflight(heap_inflight_lsn_t & inflight)
 {
     auto wr = inflight.wr;
-    if (inflight.flags & HEAP_INFLIGHT_COMPACTED)
+    if (inflight.flags & HEAP_INFLIGHT_OVERWRITE)
     {
         // Mark previous entries as garbage, sequentially
         mark_garbage_up_to(wr);

src/blockstore/blockstore_heap.h

@@ -43,7 +43,7 @@ struct __attribute__((__packed__)) heap_entry_t
 {
     uint16_t size;
     uint16_t entry_type;
-    uint32_t crc32c;
+    uint32_t checksum;
     uint64_t lsn;
     uint64_t inode;
     uint64_t stripe;
@@ -69,7 +69,8 @@ struct __attribute__((__packed__)) heap_entry_t
     uint32_t *get_checksum(blockstore_heap_t *heap);
     uint64_t big_location(blockstore_heap_t *heap);
     void set_big_location(blockstore_heap_t *heap, uint64_t location);
-    uint32_t calc_crc32c();
+    uint32_t calc_checksum(blockstore_heap_t *heap);
+    uint32_t calc_checksum(blockstore_disk_t *dsk);
 };
 
 struct __attribute__((__packed__)) heap_small_write_t
@@ -80,7 +81,7 @@ struct __attribute__((__packed__)) heap_small_write_t
     uint32_t offset;
     uint32_t len;
 
-    // Also includes 1 bitmap and 1 crc32c after the bitmap if checksums are disabled
+    // Also includes 1 bitmap and 1 checksum after the bitmap if block checksums are disabled
 };
 
 struct __attribute__((__packed__)) heap_big_write_t
@@ -98,7 +99,7 @@ struct __attribute__((__packed__)) heap_big_intent_t
     uint32_t offset;
     uint32_t len;
 
-    // Also includes 2 bitmaps and 1 crc32c if checksums are disabled
+    // Also includes 2 bitmaps and 1 checksums if block checksums are disabled
 };
 
 struct __attribute__((__packed__)) heap_list_item_t
@@ -219,6 +220,7 @@ class blockstore_heap_t
     bool marked_used_blocks = false;
     bool recheck_queue_filled = false;
     std::vector<heap_list_item_t*> postponed_items;
+    std::vector<heap_list_item_t*> init_erase_items;
     std::set<uint32_t> recheck_modified_blocks;
     std::deque<heap_entry_t*> recheck_queue;
     std::map<heap_entry_t*, heap_recheck_state_t> recheck_states;
@@ -361,7 +363,7 @@ public:
 
     // get metadata block data buffer and used space
     void get_meta_block(uint32_t block_num, uint8_t *buffer);
-    void fill_block_empty_space(uint8_t *buffer, uint32_t pos);
+    void fill_block_empty_space(uint8_t *buffer, uint64_t pos);
     uint32_t get_meta_block_used_space(uint32_t block_num);
 
     // get space usage statistics

src/blockstore/blockstore_impl.h

@@ -117,9 +117,12 @@ public:
 
     journal_flusher_t *flusher;
     int write_iodepth = 0;
-    int inflight_big = 0;
     int intent_write_counter = 0;
-    bool fsyncing_data = false;
+    uint64_t data_fsync_next = 0;
+    uint64_t data_fsync_cur = 0;
+    uint64_t data_fsync_sent = 0;
+    uint64_t data_fsync_done = 0;
+    std::deque<bool> data_fsyncs;
 
     bool live = false, queue_stall = false;
     ring_loop_i *ringloop = NULL;

src/blockstore/blockstore_init.cpp

@@ -10,7 +10,6 @@
 #define INIT_META_EMPTY 0
 #define INIT_META_READING 1
 #define INIT_META_READ_DONE 2
-#define INIT_META_WRITING 3
 
 #define GET_SQE() \
     sqe = bs->get_sqe();\
@@ -23,14 +22,15 @@ blockstore_init_meta::blockstore_init_meta(blockstore_impl_t *bs)
     this->bs = bs;
 }
 
-void blockstore_init_meta::handle_event(ring_data_t *data, int buf_num)
+void blockstore_init_meta::handle_event(ring_data_t *data, int buf_num, const char *op)
 {
-    if (data->res < 0)
+    if (data->res != data->iov.iov_len)
     {
-        throw std::runtime_error(
-            std::string("read metadata failed at offset ") + std::to_string(buf_num >= 0 ? bufs[buf_num].offset : last_read_offset) +
-            std::string(": ") + strerror(-data->res)
-        );
+        throw std::runtime_error(strprintf(
+            "%s failed at offset %ju: got %s (code %d), but expected %zu",
+            op, (buf_num >= 0 ? bufs[buf_num].offset : last_read_offset), strerror(-data->res),
+            data->res, data->iov.iov_len
+        ));
     }
     if (buf_num >= 0)
     {
@@ -60,7 +60,7 @@ int blockstore_init_meta::loop()
     GET_SQE();
     last_read_offset = 0;
     data->iov = { bs->meta_superblock, (size_t)bs->dsk.meta_block_size };
-    data->callback = [this](ring_data_t *data) { handle_event(data, -1); };
+    data->callback = [this](ring_data_t *data) { handle_event(data, -1, "read metadata header"); };
     io_uring_prep_readv(sqe, bs->dsk.meta_fd, &data->iov, 1, bs->dsk.meta_offset);
     bs->ringloop->submit();
     submitted++;
@@ -72,25 +72,19 @@ resume_1:
     }
     if (is_zero((uint64_t*)bs->meta_superblock, bs->dsk.meta_block_size))
     {
-        {
-            blockstore_meta_header_v3_t *hdr = (blockstore_meta_header_v3_t *)bs->meta_superblock;
-            hdr->zero = 0;
-            hdr->magic = BLOCKSTORE_META_MAGIC_V1;
-            hdr->version = bs->dsk.meta_format;
-            hdr->meta_block_size = bs->dsk.meta_block_size;
-            hdr->data_block_size = bs->dsk.data_block_size;
-            hdr->bitmap_granularity = bs->dsk.bitmap_granularity;
-            if (bs->dsk.meta_format >= BLOCKSTORE_META_FORMAT_V2)
-            {
-                hdr->data_csum_type = bs->dsk.data_csum_type;
-                hdr->csum_block_size = bs->dsk.csum_block_size;
-            }
-            if (bs->dsk.meta_format >= BLOCKSTORE_META_FORMAT_HEAP)
-            {
-                hdr->meta_area_size = bs->dsk.meta_area_size;
-            }
-            hdr->set_crc32c();
-        }
+        assert(bs->dsk.meta_format == BLOCKSTORE_META_FORMAT_HEAP);
+        blockstore_meta_header_v3_t *hdr = (blockstore_meta_header_v3_t *)bs->meta_superblock;
+        hdr->zero = 0;
+        hdr->magic = BLOCKSTORE_META_MAGIC_V1;
+        hdr->version = bs->dsk.meta_format;
+        hdr->meta_block_size = bs->dsk.meta_block_size;
+        hdr->data_block_size = bs->dsk.data_block_size;
+        hdr->bitmap_granularity = bs->dsk.bitmap_granularity;
+        hdr->completed_lsn = 0;
+        hdr->data_csum_type = bs->dsk.data_csum_type;
+        hdr->csum_block_size = bs->dsk.csum_block_size;
+        hdr->meta_area_size = bs->dsk.meta_area_size;
+        hdr->set_crc32c();
         if (bs->readonly)
         {
             printf("Skipping metadata initialization because blockstore is readonly\n");
@@ -98,21 +92,8 @@ resume_1:
         else
         {
             printf("Initializing metadata area\n");
-            GET_SQE();
-            last_read_offset = 0;
-            data->iov = (struct iovec){ bs->meta_superblock, (size_t)bs->dsk.meta_block_size };
-            data->callback = [this](ring_data_t *data) { handle_event(data, -1); };
-            io_uring_prep_writev(sqe, bs->dsk.meta_fd, &data->iov, 1, bs->dsk.meta_offset);
-            bs->ringloop->submit();
-            submitted++;
-        resume_2:
-            if (submitted > 0)
-            {
-                wait_state = 2;
-                return 1;
-            }
-            zero_on_init = true;
         }
+        zero_on_init = true;
     }
     else
     {
@@ -163,7 +144,7 @@ resume_1:
         hdr->header_csum = csum;
     }
     bs->heap->start_load(((blockstore_meta_header_v3_t *)bs->meta_superblock)->completed_lsn);
-    if (bs->dsk.inmemory_journal)
+    if (bs->dsk.inmemory_journal && !zero_on_init)
     {
         // Read buffer area
         printf("Reading buffered data\n");
@@ -175,7 +156,7 @@ resume_1:
                 bs->buffer_area + md_offset,
                 (size_t)(bs->dsk.journal_len - md_offset < bs->metadata_buf_size ? bs->dsk.journal_len - md_offset : bs->metadata_buf_size),
             };
-            data->callback = [this](ring_data_t *data) { handle_event(data, -1); };
+            data->callback = [this](ring_data_t *data) { handle_event(data, -1, "read buffer area"); };
             io_uring_prep_readv(sqe, bs->dsk.journal_fd, &data->iov, 1, bs->dsk.journal_offset + md_offset);
             md_offset += data->iov.iov_len;
             submitted++;
@@ -194,7 +175,7 @@ resume_3:
     next_offset = md_offset;
     // Read the rest of the metadata
 resume_4:
-    if (next_offset < bs->dsk.meta_area_size && submitted == 0)
+    if (next_offset < bs->dsk.meta_area_size && submitted == 0 && (!zero_on_init || !bs->readonly))
     {
         // Submit one read
         for (int i = 0; i < 2; i++)
@@ -211,12 +192,15 @@ resume_4:
                 GET_SQE();
                 assert(bufs[i].size <= 0x7fffffff);
                 data->iov = { bufs[i].buf, (size_t)bufs[i].size };
-                data->callback = [this, i](ring_data_t *data) { handle_event(data, i); };
                 if (!zero_on_init)
+                {
+                    data->callback = [this, i](ring_data_t *data) { handle_event(data, i, "read metadata"); };
                     io_uring_prep_readv(sqe, bs->dsk.meta_fd, &data->iov, 1, bs->dsk.meta_offset + bufs[i].offset);
+                }
                 else
                 {
                     // Fill metadata with empty block pattern
+                    data->callback = [this, i](ring_data_t *data) { handle_event(data, i, "clear metadata"); };
                     memset(bufs[i].buf, 0, bufs[i].size);
                     for (uint64_t o = 0; o < bufs[i].size; o += bs->dsk.meta_block_size)
                         bs->heap->fill_block_empty_space(bufs[i].buf + o, 0);
@@ -232,11 +216,14 @@ resume_4:
         if (bufs[i].state == INIT_META_READ_DONE)
         {
             // Handle result
-            uint64_t loaded = 0;
-            int r = bs->heap->load_blocks(bufs[i].offset-bs->dsk.meta_block_size, bufs[i].size, bufs[i].buf, bs->skip_corrupted_meta_entries, loaded);
-            if (r != 0)
-                exit(1);
-            entries_loaded += loaded;
+            if (!zero_on_init)
+            {
+                uint64_t loaded = 0;
+                int r = bs->heap->load_blocks(bufs[i].offset-bs->dsk.meta_block_size, bufs[i].size, bufs[i].buf, bs->skip_corrupted_meta_entries, loaded);
+                if (r != 0)
+                    exit(1);
+                entries_loaded += loaded;
+            }
             bufs[i].state = 0;
             bs->ringloop->wakeup();
         }
@@ -246,7 +233,7 @@ resume_4:
         wait_state = 4;
         return 1;
     }
-    // metadata read finished
+    // metadata read/clear finished
     bs->heap->finish_load();
     printf("Metadata entries loaded: %ju, rechecking unfinished writes and garbage entries\n", entries_loaded);
     // asynchronous recheck
@@ -329,13 +316,14 @@ resume_9:
     }
     free(metadata_buffer);
     metadata_buffer = NULL;
+do_fsync:
     if (!bs->dsk.disable_meta_fsync && !bs->readonly)
     {
         GET_SQE();
         io_uring_prep_fsync(sqe, bs->dsk.meta_fd, IORING_FSYNC_DATASYNC);
         last_read_offset = 0;
         data->iov = { 0 };
-        data->callback = [this](ring_data_t *data) { handle_event(data, -1); };
+        data->callback = [this](ring_data_t *data) { handle_event(data, -1, "fsync metadata"); };
         submitted++;
         bs->ringloop->submit();
     resume_5:
@@ -345,6 +333,27 @@ resume_9:
             return 1;
         }
     }
+    if (zero_on_init && !header_written && !bs->readonly)
+    {
+        GET_SQE();
+        header_written = true;
+        last_read_offset = 0;
+        data->iov = (struct iovec){ bs->meta_superblock, (size_t)bs->dsk.meta_block_size };
+        data->callback = [this](ring_data_t *data) { handle_event(data, -1, "write metadata header"); };
+        io_uring_prep_writev(sqe, bs->dsk.meta_fd, &data->iov, 1, bs->dsk.meta_offset);
+        bs->ringloop->submit();
+        submitted++;
+    resume_2:
+        if (submitted > 0)
+        {
+            wait_state = 2;
+            return 1;
+        }
+        if (!bs->dsk.disable_meta_fsync)
+        {
+            goto do_fsync;
+        }
+    }
     printf("Loading finished. Data used: %ju / %ju bytes (%s / %s)\n",
         bs->heap->get_data_used_space(), bs->dsk.block_count * bs->dsk.data_block_size,
         format_size(bs->heap->get_data_used_space()).c_str(),

src/blockstore/blockstore_init.h

@@ -17,6 +17,7 @@ class blockstore_init_meta
     int wait_state = 0;
     int wait_count = 0;
     bool zero_on_init = false;
+    bool header_written = false;
     void *metadata_buffer = NULL;
     blockstore_init_meta_buf bufs[2] = {};
     int submitted = 0;
@@ -29,7 +30,7 @@ class blockstore_init_meta
     std::vector<uint32_t> recheck_mod;
     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);
+    void handle_event(ring_data_t *data, int buf_num, const char *op);
 public:
     blockstore_init_meta(blockstore_impl_t *bs);
     int loop();

src/blockstore/blockstore_mock.cpp

@@ -0,0 +1,113 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+#include "blockstore_mock.h"
+
+blockstore_mock_t::blockstore_mock_t(const blockstore_config_t & config)
+{
+}
+
+void blockstore_mock_t::parse_config(blockstore_config_t & config)
+{
+}
+
+void* blockstore_mock_t::reshard_start(pool_id_t pool, uint32_t pg_count, uint32_t pg_stripe_size, uint64_t chunk_limit)
+{
+    return NULL;
+}
+
+bool blockstore_mock_t::reshard_continue(void *reshard_state, uint64_t chunk_limit)
+{
+    return true;
+}
+
+void blockstore_mock_t::loop()
+{
+}
+
+bool blockstore_mock_t::is_started()
+{
+    return true;
+}
+
+bool blockstore_mock_t::is_stalled()
+{
+    return false;
+}
+
+bool blockstore_mock_t::is_safe_to_stop()
+{
+    return true;
+}
+
+void blockstore_mock_t::enqueue_op(blockstore_op_t *op)
+{
+}
+
+int blockstore_mock_t::read_bitmap(object_id oid, uint64_t target_version, void *bitmap, uint64_t *result_version)
+{
+    return -EIO;
+}
+
+const std::map<uint64_t, uint64_t> & blockstore_mock_t::get_inode_space_stats()
+{
+    return inode_space;
+}
+
+void blockstore_mock_t::set_no_inode_stats(const std::vector<uint64_t> & pool_ids)
+{
+}
+
+void blockstore_mock_t::dump_diagnostics()
+{
+}
+
+std::string blockstore_mock_t::get_op_diag(blockstore_op_t *op)
+{
+    return "";
+}
+
+uint32_t blockstore_mock_t::get_block_size()
+{
+    return block_size;
+}
+
+uint64_t blockstore_mock_t::get_block_count()
+{
+    return block_count;
+}
+
+uint64_t blockstore_mock_t::get_free_block_count()
+{
+    return block_count;
+}
+
+uint64_t blockstore_mock_t::get_journal_size()
+{
+    return 32*1024*1024;
+}
+
+uint32_t blockstore_mock_t::get_bitmap_granularity()
+{
+    return bitmap_granularity;
+}
+
+uint64_t blockstore_mock_t::get_live_entries()
+{
+    return 0;
+}
+
+uint64_t blockstore_mock_t::get_live_memory()
+{
+    return 0;
+}
+
+uint64_t blockstore_mock_t::get_garbage_entries()
+{
+    return 0;
+}
+
+uint64_t blockstore_mock_t::get_garbage_memory()
+{
+    return 0;
+}

src/blockstore/blockstore_mock.h

@@ -0,0 +1,39 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+#pragma once
+
+#include "blockstore.h"
+
+class blockstore_mock_t: public blockstore_i
+{
+public:
+    uint32_t block_size = 128*1024;
+    uint32_t bitmap_granularity = 4096;
+    uint64_t block_count = 100*1024*8;
+    std::map<uint64_t, uint64_t> inode_space;
+
+    blockstore_mock_t(const blockstore_config_t & config);
+    void parse_config(blockstore_config_t & config) override;
+    void* reshard_start(pool_id_t pool, uint32_t pg_count, uint32_t pg_stripe_size, uint64_t chunk_limit) override;
+    bool reshard_continue(void *reshard_state, uint64_t chunk_limit) override;
+    void loop() override;
+    bool is_started() override;
+    bool is_stalled() override;
+    bool is_safe_to_stop() override;
+    void enqueue_op(blockstore_op_t *op) override;
+    int read_bitmap(object_id oid, uint64_t target_version, void *bitmap, uint64_t *result_version = NULL) override;
+    const std::map<uint64_t, uint64_t> & get_inode_space_stats() override;
+    void set_no_inode_stats(const std::vector<uint64_t> & pool_ids) override;
+    void dump_diagnostics() override;
+    std::string get_op_diag(blockstore_op_t *op) override;
+    uint32_t get_block_size() override;
+    uint64_t get_block_count() override;
+    uint64_t get_free_block_count() override;
+    uint64_t get_journal_size() override;
+    uint32_t get_bitmap_granularity() override;
+    uint64_t get_live_entries() override;
+    uint64_t get_live_memory() override;
+    uint64_t get_garbage_entries() override;
+    uint64_t get_garbage_memory() override;
+};

src/blockstore/blockstore_stable.cpp

@@ -28,6 +28,7 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
                 FINISH_OP(op);
                 return 2;
             }
+            priv->modified_block2 = UINT32_MAX;
             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);
@@ -52,11 +53,6 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
                     FINISH_OP(op);
                     return 2;
                 }
-                if (priv->modified_block2 != UINT32_MAX)
-                {
-                    priv->stab_pos--;
-                    goto resume_1;
-                }
                 priv->wait_for = WAIT_COMPACTION;
                 priv->wait_detail = heap->get_compacted_count();
                 flusher->request_trim();

src/blockstore/blockstore_sync.cpp

@@ -29,9 +29,12 @@ bool blockstore_impl_t::has_unsynced()
 
 bool blockstore_impl_t::submit_fsyncs(int & wait_count)
 {
-    int n = (unsynced_meta_write_count > 0 && !dsk.disable_meta_fsync) +
-        (unsynced_buffer_write_count > 0 && !dsk.disable_journal_fsync && dsk.journal_fd != dsk.meta_fd) +
-        (unsynced_data_write_count > 0 && !dsk.disable_data_fsync && dsk.data_fd != dsk.meta_fd && dsk.data_fd != dsk.journal_fd);
+    int n = (unsynced_meta_write_count > 0 && !dsk.disable_meta_fsync ? 1 : 0) +
+        (unsynced_buffer_write_count > 0 && !dsk.disable_journal_fsync &&
+            (!unsynced_meta_write_count || dsk.journal_fd != dsk.meta_fd) ? 1 : 0) +
+        (unsynced_data_write_count > 0 && !dsk.disable_data_fsync &&
+            (!unsynced_meta_write_count || dsk.data_fd != dsk.meta_fd) &&
+            (!unsynced_buffer_write_count || dsk.data_fd != dsk.journal_fd) ? 1 : 0);
     if (ringloop->space_left() < n)
     {
         return false;
@@ -60,7 +63,8 @@ bool blockstore_impl_t::submit_fsyncs(int & wait_count)
         data->callback = cb;
         wait_count++;
     }
-    if (unsynced_buffer_write_count > 0 && !dsk.disable_journal_fsync && dsk.meta_fd != dsk.journal_fd)
+    if (unsynced_buffer_write_count > 0 && !dsk.disable_journal_fsync &&
+        (!unsynced_meta_write_count || dsk.journal_fd != dsk.meta_fd))
     {
         // fsync buffer
         io_uring_sqe *sqe = get_sqe();
@@ -71,7 +75,9 @@ bool blockstore_impl_t::submit_fsyncs(int & wait_count)
         data->callback = cb;
         wait_count++;
     }
-    if (unsynced_data_write_count > 0 && !dsk.disable_data_fsync && dsk.data_fd != dsk.meta_fd && dsk.data_fd != dsk.journal_fd)
+    if (unsynced_data_write_count > 0 && !dsk.disable_data_fsync &&
+        (!unsynced_meta_write_count || dsk.data_fd != dsk.meta_fd) &&
+        (!unsynced_buffer_write_count || dsk.data_fd != dsk.journal_fd))
     {
         // fsync data
         io_uring_sqe *sqe = get_sqe();
@@ -109,6 +115,7 @@ int blockstore_impl_t::do_sync(blockstore_op_t *op, int base_state)
     PRIV(op)->lsn = heap->get_completed_lsn();
     if (!submit_fsyncs(PRIV(op)->pending_ops))
     {
+        PRIV(op)->lsn = 0;
         PRIV(op)->wait_detail = 1;
         PRIV(op)->wait_for = WAIT_SQE;
         return 0;

src/blockstore/blockstore_write.cpp

@@ -180,13 +180,16 @@ 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);
+        if (!dsk.disable_data_fsync)
+        {
+            // use PRIV->lsn for fsync_data_id
+            PRIV(op)->lsn = ++data_fsync_next;
+            data_fsyncs.push_back(false);
+        }
         PRIV(op)->pending_ops++;
         write_iodepth++;
         if (PRIV(op)->write_type == BS_HEAP_BIG_WRITE)
-        {
             PRIV(op)->op_state = 1;
-            inflight_big++;
-        }
         else
             PRIV(op)->op_state = 3;
     }
@@ -297,8 +300,6 @@ again:
         goto resume_10;
     else if (op_state == 11)
         goto resume_11;
-    else if (op_state == 12)
-        goto resume_12;
     else
     {
         // In progress
@@ -317,38 +318,44 @@ again:
 resume_2:
     // We must fsync all big writes to avoid complex write workflows
     // It's OK for all HDDs and for server SSDs, but slightly worse for desktop SSDs
-    inflight_big--;
     if (!dsk.disable_data_fsync)
     {
-        // fsync data in a batch
-resume_11:
-        if (inflight_big > 0)
+        // Mark our data write as completed and advance data_fsync_cur
+        data_fsyncs[PRIV(op)->lsn - data_fsync_cur - 1] = true;
+        while (data_fsyncs.size() > 0 && data_fsyncs.front())
+        {
+            data_fsyncs.pop_front();
+            data_fsync_cur++;
+        }
+        PRIV(op)->op_state = 11;
+        // Then wait for all other data writes currently in progress to do less fsync calls
+        // I.e. to fsync data in batches
+        PRIV(op)->lsn = data_fsync_cur + data_fsyncs.size();
+resume_11:
+        if (data_fsync_cur < PRIV(op)->lsn)
         {
-            PRIV(op)->op_state = 11;
             return 1;
         }
-        if (fsyncing_data)
+        if (PRIV(op)->lsn > data_fsync_sent)
         {
-resume_12:
-            if (fsyncing_data)
+            BS_SUBMIT_GET_SQE(sqe, data);
+            io_uring_prep_fsync(sqe, dsk.data_fd, IORING_FSYNC_DATASYNC);
+            data->iov = { 0 };
+            data->callback = [this, op, fs = data_fsync_cur](ring_data_t *data)
             {
-                PRIV(op)->op_state = 12;
-                return 1;
-            }
-            goto resume_4;
+                if (fs > data_fsync_done)
+                {
+                    data_fsync_done = fs;
+                    ringloop->wakeup();
+                }
+            };
+            data_fsync_sent = data_fsync_cur;
         }
-        fsyncing_data = true;
-        BS_SUBMIT_GET_SQE(sqe, data);
-        io_uring_prep_fsync(sqe, dsk.data_fd, IORING_FSYNC_DATASYNC);
-        data->iov = { 0 };
-        data->callback = [this, op](ring_data_t *data)
+        if (PRIV(op)->lsn > data_fsync_done)
         {
-            fsyncing_data = false;
-            handle_write_event(data, op);
-        };
-        PRIV(op)->pending_ops++;
-        PRIV(op)->op_state = 3;
-        return 1;
+            return 1;
+        }
+        PRIV(op)->lsn = 0;
     }
 resume_4:
     {

src/blockstore/multilist.cpp

@@ -171,7 +171,7 @@ void multilist_alloc_t::print()
     printf("\n");
 }
 
-void multilist_alloc_t::use(uint32_t pos, uint32_t size)
+bool multilist_alloc_t::use(uint32_t pos, uint32_t size)
 {
     assert(pos+size <= count && size > 0);
     if (sizes[pos] <= 0)
@@ -182,7 +182,8 @@ void multilist_alloc_t::use(uint32_t pos, uint32_t size)
         else
             while (start > 0 && !sizes[start])
                 start--;
-        assert(sizes[start] >= size);
+        if (sizes[start] < size+(pos-start))
+            return false;
         use_full(start);
         uint32_t full = sizes[start];
         sizes[pos-1] = -pos+start;
@@ -199,7 +200,8 @@ void multilist_alloc_t::use(uint32_t pos, uint32_t size)
     }
     else
     {
-        assert(sizes[pos] >= size);
+        if (sizes[pos] < size)
+            return false;
         use_full(pos);
         if (sizes[pos] > size)
         {
@@ -214,12 +216,13 @@ void multilist_alloc_t::use(uint32_t pos, uint32_t size)
 #ifdef MULTILIST_TRACE
     print();
 #endif
+    return true;
 }
 
 void multilist_alloc_t::use_full(uint32_t pos)
 {
     uint32_t prevsize = sizes[pos];
-    assert(prevsize);
+    assert(prevsize > 0);
     assert(nexts[pos]);
     uint32_t pi = (prevsize < maxn ? prevsize : maxn)-1;
     if (heads[pi] == pos+1)

src/blockstore/multilist.h

@@ -17,7 +17,7 @@ struct multilist_alloc_t
     bool is_free(uint32_t pos);
     uint32_t find(uint32_t size);
     void use_full(uint32_t pos);
-    void use(uint32_t pos, uint32_t size);
+    bool use(uint32_t pos, uint32_t size);
     void do_free(uint32_t pos);
     void free(uint32_t pos);
     void verify();

src/blockstore/v1/flush.cpp

@@ -71,6 +71,11 @@ bool journal_flusher_t::is_active()
     return active_flushers > 0 || dequeuing;
 }
 
+size_t journal_flusher_t::get_queue_size()
+{
+    return flush_queue.size();
+}
+
 void journal_flusher_t::loop()
 {
     target_flusher_count = bs->write_iodepth*2;
@@ -384,6 +389,7 @@ stop_flusher:
         wait_state = 0;
         return true;
     }
+    copy_count = 0;
     try_trim = true;
     cur.oid = flusher->flush_queue.front();
     cur.version = flusher->flush_versions[cur.oid];
@@ -511,6 +517,31 @@ resume_2:
             {
                 uo_it->second.was_changed = true;
             }
+            if (!bs->journal.inmemory)
+            {
+                // Verify journaled data checksums (but not COALESCED)
+                for (it = v.begin(); it != v.end(); it++)
+                {
+                    if (it->copy_flags == COPY_BUF_JOURNAL)
+                    {
+                        iovec iov = { .iov_base = it->buf, .iov_len = it->len };
+                        bs->verify_journal_checksums(
+                            it->csum_buf, it->offset, &iov, 1,
+                            [&](uint32_t bad_block, uint32_t calc_csum, uint32_t stored_csum)
+                            {
+                                printf(
+                                    "Checksum mismatch in object %jx:%jx v%ju in journal at 0x%jx, checksum block #%u: got %08x, expected %08x\n",
+                                    cur.oid.inode, cur.oid.stripe, cur.version, it->disk_offset,
+                                    bad_block / bs->dsk.csum_block_size, calc_csum, stored_csum
+                                );
+                                bad_block += it->offset;
+                                assert(!(bad_block % bs->dsk.csum_block_size) && bad_block < bs->dsk.data_block_size);
+                                mangle_csum_blocks.insert(bad_block);
+                            }
+                        );
+                    }
+                }
+            }
         }
         // Submit data writes
         for (it = v.begin(); it != v.end(); it++)
@@ -634,6 +665,7 @@ resume_2:
         }
         // All done
         flusher->active_flushers--;
+        copy_count = 0; // used by is_mutated()...
         wait_state = 0;
         goto resume_0;
     }
@@ -815,35 +847,21 @@ bool journal_flusher_co::clear_incomplete_csum_block_bits(int wait_base)
                 bs->verify_padded_checksums(new_clean_bitmap, new_clean_bitmap + 2*bs->dsk.clean_entry_bitmap_size,
                     v[i].offset, &iov, 1, [&](uint32_t bad_block, uint32_t calc_csum, uint32_t stored_csum)
                 {
-                    printf("Checksum mismatch in object %jx:%jx v%ju in data area at offset 0x%jx+0x%x: got %08x, expected %08x\n",
+                    printf("Checksum mismatch in object %jx:%jx v%ju in data area at offset 0x%jx+0x%x during flush: got %08x, expected %08x\n",
                         cur.oid.inode, cur.oid.stripe, old_clean_ver, old_clean_loc, bad_block, calc_csum, stored_csum);
-                    for (uint32_t j = 0; j < bs->dsk.csum_block_size; j += bs->dsk.bitmap_granularity)
-                    {
-                        // Simplest method of mangling: flip one byte in every sector
-                        ((uint8_t*)v[i].buf)[j+bad_block-v[i].offset] ^= 0xff;
-                    }
+                    assert(!(bad_block % bs->dsk.csum_block_size) && bad_block < bs->dsk.data_block_size);
+                    mangle_csum_blocks.insert(bad_block);
                 });
             }
             else
             {
                 bs->verify_journal_checksums(v[i].csum_buf, v[i].offset, &iov, 1, [&](uint32_t bad_block, uint32_t calc_csum, uint32_t stored_csum)
                 {
-                    printf("Checksum mismatch in object %jx:%jx v%ju in journal at offset 0x%jx+0x%x (block offset 0x%jx): got %08x, expected %08x\n",
+                    printf("Checksum mismatch in object %jx:%jx v%ju in journal at offset 0x%jx+0x%x (block offset 0x%jx) during flush: got %08x, expected %08x\n",
                         cur.oid.inode, cur.oid.stripe, old_clean_ver,
                         v[i].disk_offset, bad_block, v[i].offset, calc_csum, stored_csum);
-                    bad_block += (v[i].offset/bs->dsk.csum_block_size) * bs->dsk.csum_block_size;
-                    uint32_t bad_block_end = bad_block + bs->dsk.csum_block_size + (v[i].offset/bs->dsk.csum_block_size) * bs->dsk.csum_block_size;
-                    if (bad_block < v[i].offset)
-                        bad_block = v[i].offset;
-                    if (bad_block_end > v[i].offset+v[i].len)
-                        bad_block_end = v[i].offset+v[i].len;
-                    bad_block -= v[i].offset;
-                    bad_block_end -= v[i].offset;
-                    for (uint32_t j = bad_block; j < bad_block_end; j += bs->dsk.bitmap_granularity)
-                    {
-                        // Simplest method of mangling: flip one byte in every sector
-                        ((uint8_t*)v[i].buf)[j] ^= 0xff;
-                    }
+                    assert(!(bad_block % bs->dsk.csum_block_size) && bad_block < bs->dsk.data_block_size);
+                    mangle_csum_blocks.insert(bad_block);
                 });
             }
         }
@@ -952,6 +970,11 @@ void journal_flusher_co::calc_block_checksums(uint32_t *new_data_csums, bool ski
     }
     // `v` should contain aligned items, possibly split into pieces
     assert(!block_done);
+    for (uint32_t mangle_block: mangle_csum_blocks)
+    {
+        // Flip 1 bit
+        new_data_csums[mangle_block / bs->dsk.csum_block_size] ^= 1;
+    }
 }
 
 void journal_flusher_co::scan_dirty()
@@ -1088,7 +1111,8 @@ void journal_flusher_co::scan_dirty()
             last--;
         read_to_fill_incomplete = bs->fill_partial_checksum_blocks(
             v, fulfilled, bmp_ptr, NULL, false, NULL, v[0].offset/bs->dsk.csum_block_size * bs->dsk.csum_block_size,
-            ((v[last].offset+v[last].len-1) / bs->dsk.csum_block_size + 1) * bs->dsk.csum_block_size
+            ((v[last].offset+v[last].len-1) / bs->dsk.csum_block_size + 1) * bs->dsk.csum_block_size,
+            0, bs->dsk.data_block_size
         );
     }
     else if (fill_incomplete && clean_init_bitmap)
@@ -1118,6 +1142,7 @@ bool journal_flusher_co::read_dirty(int wait_base)
     if (wait_state == wait_base)        goto resume_0;
     else if (wait_state == wait_base+1) goto resume_1;
     wait_count = wait_journal_count = 0;
+    mangle_csum_blocks.clear();
     if (bs->journal.inmemory && !read_to_fill_incomplete)
     {
         // Happy path: nothing to read :)

src/blockstore/v1/flush.h

@@ -66,6 +66,7 @@ class journal_flusher_co
     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;
 
@@ -123,6 +124,7 @@ public:
     void loop();
     bool is_trim_wanted() { return trim_wanted; }
     bool is_active();
+    size_t get_queue_size();
     void mark_trim_possible();
     void request_trim();
     void release_trim();

src/blockstore/v1/impl.cpp

@@ -6,11 +6,12 @@
 
 namespace v1 {
 
-blockstore_impl_t::blockstore_impl_t(blockstore_config_t & config, ring_loop_i *ringloop, timerfd_manager_t *tfd)
+blockstore_impl_t::blockstore_impl_t(blockstore_config_t & config, ring_loop_i *ringloop, timerfd_manager_t *tfd, bool mock_mode)
 {
     assert(sizeof(blockstore_op_private_t) <= BS_OP_PRIVATE_DATA_SIZE);
     this->tfd = tfd;
     this->ringloop = ringloop;
+    dsk.mock_mode = mock_mode;
     ring_consumer.loop = [this]() { loop(); };
     ringloop->register_consumer(&ring_consumer);
     initialized = 0;
@@ -35,6 +36,11 @@ blockstore_impl_t::blockstore_impl_t(blockstore_config_t & config, ring_loop_i *
 
 blockstore_impl_t::~blockstore_impl_t()
 {
+    for (auto& obj: dirty_db)
+    {
+        if (obj.second.dyn_data)
+            free(obj.second.dyn_data);
+    }
     delete data_alloc;
     delete flusher;
     if (zero_object)

src/blockstore/v1/impl.h

@@ -30,6 +30,8 @@
 
 //#define BLOCKSTORE_DEBUG
 
+struct bs_test_t;
+
 namespace v1 {
 
 #include "journal.h"
@@ -96,7 +98,7 @@ struct blockstore_op_private_t
     int op_state;
 
     // Read
-    uint64_t clean_block_used;
+    uint64_t clean_loc_used;
     std::vector<copy_buffer_t> read_vec;
 
     // Sync, write
@@ -122,6 +124,7 @@ typedef uint64_t pool_pg_id_t;
 
 class blockstore_impl_t: public blockstore_i
 {
+    friend struct ::bs_test_t;
     blockstore_disk_t dsk;
 
     /******* OPTIONS *******/
@@ -220,6 +223,7 @@ class blockstore_impl_t: public blockstore_i
 
     // Read
     int dequeue_read(blockstore_op_t *read_op);
+    void release_clean(blockstore_op_t *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,
@@ -230,7 +234,8 @@ class blockstore_impl_t: public blockstore_i
         uint8_t *clean_entry_bitmap, int *dyn_data,
         uint32_t item_start, uint32_t item_end, uint64_t clean_loc, uint64_t clean_ver);
     int fill_partial_checksum_blocks(std::vector<copy_buffer_t> & rv, uint64_t & fulfilled,
-        uint8_t *clean_entry_bitmap, int *dyn_data, bool from_journal, uint8_t *read_buf, uint64_t read_offset, uint64_t read_end);
+        uint8_t *clean_entry_bitmap, int *dyn_data, bool from_journal, uint8_t *read_buf,
+        uint32_t read_offset, uint32_t read_end, uint32_t item_start, uint32_t item_end);
     int pad_journal_read(std::vector<copy_buffer_t> & rv, copy_buffer_t & cp,
         uint64_t dirty_offset, uint64_t dirty_end, uint64_t dirty_loc, uint8_t *csum_ptr, int *dyn_data,
         uint64_t offset, uint64_t submit_len, uint64_t & blk_begin, uint64_t & blk_end, uint8_t* & blk_buf);
@@ -281,7 +286,7 @@ class blockstore_impl_t: public blockstore_i
 
 public:
 
-    blockstore_impl_t(blockstore_config_t & config, ring_loop_i *ringloop, timerfd_manager_t *tfd);
+    blockstore_impl_t(blockstore_config_t & config, ring_loop_i *ringloop, timerfd_manager_t *tfd, bool mock_mode = false);
     ~blockstore_impl_t();
 
     void parse_config(blockstore_config_t & config);

src/blockstore/v1/init.cpp

@@ -1,6 +1,7 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 (see README.md for details)
 
+#include "str_util.h"
 #include "impl.h"
 #include "internal.h"
 
@@ -30,14 +31,15 @@ blockstore_init_meta::blockstore_init_meta(blockstore_impl_t *bs)
     this->bs = bs;
 }
 
-void blockstore_init_meta::handle_event(ring_data_t *data, int buf_num)
+void blockstore_init_meta::handle_event(ring_data_t *data, int buf_num, const char *op)
 {
-    if (data->res < 0)
+    if (data->res != data->iov.iov_len)
     {
-        throw std::runtime_error(
-            std::string("read metadata failed at offset ") + std::to_string(buf_num >= 0 ? bufs[buf_num].offset : last_read_offset) +
-            std::string(": ") + strerror(-data->res)
-        );
+        throw std::runtime_error(strprintf(
+            "%s failed at offset %ju: got %s (code %d), but expected %zu",
+            op, (buf_num >= 0 ? bufs[buf_num].offset : last_read_offset), strerror(-data->res),
+            data->res, data->iov.iov_len
+        ));
     }
     if (buf_num >= 0)
     {
@@ -65,10 +67,11 @@ int blockstore_init_meta::loop()
     if (!metadata_buffer)
         throw std::runtime_error("Failed to allocate metadata read buffer");
     // Read superblock
+    hdr = (blockstore_meta_header_v2_t *)memalign_or_die(MEM_ALIGNMENT, bs->dsk.meta_block_size);
     GET_SQE();
     last_read_offset = 0;
-    data->iov = { metadata_buffer, (size_t)bs->dsk.meta_block_size };
-    data->callback = [this](ring_data_t *data) { handle_event(data, -1); };
+    data->iov = { hdr, (size_t)bs->dsk.meta_block_size };
+    data->callback = [this](ring_data_t *data) { handle_event(data, -1, "read metadata header"); };
     io_uring_prep_readv(sqe, bs->dsk.meta_fd, &data->iov, 1, bs->dsk.meta_offset);
     bs->ringloop->submit();
     submitted++;
@@ -78,24 +81,8 @@ resume_1:
         wait_state = 1;
         return 1;
     }
-    if (iszero((uint64_t*)metadata_buffer, bs->dsk.meta_block_size / sizeof(uint64_t)))
+    if (iszero((uint64_t*)hdr, bs->dsk.meta_block_size / sizeof(uint64_t)))
     {
-        {
-            blockstore_meta_header_v2_t *hdr = (blockstore_meta_header_v2_t *)metadata_buffer;
-            hdr->zero = 0;
-            hdr->magic = BLOCKSTORE_META_MAGIC_V1;
-            hdr->version = bs->dsk.meta_format;
-            hdr->meta_block_size = bs->dsk.meta_block_size;
-            hdr->data_block_size = bs->dsk.data_block_size;
-            hdr->bitmap_granularity = bs->dsk.bitmap_granularity;
-            if (bs->dsk.meta_format >= BLOCKSTORE_META_FORMAT_V2)
-            {
-                hdr->data_csum_type = bs->dsk.data_csum_type;
-                hdr->csum_block_size = bs->dsk.csum_block_size;
-                hdr->header_csum = 0;
-                hdr->header_csum = crc32c(0, hdr, sizeof(*hdr));
-            }
-        }
         if (bs->readonly)
         {
             printf("Skipping metadata initialization because blockstore is readonly\n");
@@ -103,25 +90,11 @@ resume_1:
         else
         {
             printf("Initializing metadata area\n");
-            GET_SQE();
-            last_read_offset = 0;
-            data->iov = (struct iovec){ metadata_buffer, (size_t)bs->dsk.meta_block_size };
-            data->callback = [this](ring_data_t *data) { handle_event(data, -1); };
-            io_uring_prep_writev(sqe, bs->dsk.meta_fd, &data->iov, 1, bs->dsk.meta_offset);
-            bs->ringloop->submit();
-            submitted++;
-        resume_3:
-            if (submitted > 0)
-            {
-                wait_state = 3;
-                return 1;
-            }
-            zero_on_init = true;
         }
+        zero_on_init = true;
     }
     else
     {
-        blockstore_meta_header_v2_t *hdr = (blockstore_meta_header_v2_t *)metadata_buffer;
         if (hdr->zero != 0 || hdr->magic != BLOCKSTORE_META_MAGIC_V1 || hdr->version < BLOCKSTORE_META_FORMAT_V1)
         {
             printf(
@@ -223,12 +196,15 @@ resume_2:
                 GET_SQE();
                 assert(bufs[i].size <= 0x7fffffff);
                 data->iov = { bufs[i].buf, (size_t)bufs[i].size };
-                data->callback = [this, i](ring_data_t *data) { handle_event(data, i); };
                 if (!zero_on_init)
+                {
+                    data->callback = [this, i](ring_data_t *data) { handle_event(data, i, "read metadata"); };
                     io_uring_prep_readv(sqe, bs->dsk.meta_fd, &data->iov, 1, bs->dsk.meta_offset + bufs[i].offset);
+                }
                 else
                 {
                     // Fill metadata with zeroes
+                    data->callback = [this, i](ring_data_t *data) { handle_event(data, i, "clear metadata"); };
                     memset(data->iov.iov_base, 0, data->iov.iov_len);
                     io_uring_prep_writev(sqe, bs->dsk.meta_fd, &data->iov, 1, bs->dsk.meta_offset + bufs[i].offset);
                 }
@@ -256,7 +232,7 @@ resume_2:
                 GET_SQE();
                 assert(bufs[i].size <= 0x7fffffff);
                 data->iov = { bufs[i].buf, (size_t)bufs[i].size };
-                data->callback = [this, i](ring_data_t *data) { handle_event(data, i); };
+                data->callback = [this, i](ring_data_t *data) { handle_event(data, i, "write metadata"); };
                 io_uring_prep_writev(sqe, bs->dsk.meta_fd, &data->iov, 1, bs->dsk.meta_offset + bufs[i].offset);
                 bs->ringloop->submit();
                 bufs[i].state = INIT_META_WRITING;
@@ -285,7 +261,7 @@ resume_2:
             GET_SQE();
             last_read_offset = (1+next_offset)*bs->dsk.meta_block_size;
             data->iov = { metadata_buffer, (size_t)bs->dsk.meta_block_size };
-            data->callback = [this](ring_data_t *data) { handle_event(data, -1); };
+            data->callback = [this](ring_data_t *data) { handle_event(data, -1, "read metadata"); };
             io_uring_prep_readv(sqe, bs->dsk.meta_fd, &data->iov, 1, bs->dsk.meta_offset + (1+next_offset)*bs->dsk.meta_block_size);
             bs->ringloop->submit();
             submitted++;
@@ -302,7 +278,7 @@ resume_5:
             }
             GET_SQE();
             data->iov = { metadata_buffer, (size_t)bs->dsk.meta_block_size };
-            data->callback = [this](ring_data_t *data) { handle_event(data, -1); };
+            data->callback = [this](ring_data_t *data) { handle_event(data, -1, "write metadata"); };
             io_uring_prep_writev(sqe, bs->dsk.meta_fd, &data->iov, 1, bs->dsk.meta_offset + (1+next_offset)*bs->dsk.meta_block_size);
             bs->ringloop->submit();
             submitted++;
@@ -317,27 +293,64 @@ resume_6:
     }
     // metadata read finished
     printf("Metadata entries loaded: %ju, free blocks: %ju / %ju\n", entries_loaded, bs->data_alloc->get_free_count(), bs->dsk.block_count);
+    if (zero_on_init && !bs->readonly)
+    {
+do_fsync:
+        if (!bs->disable_meta_fsync)
+        {
+            GET_SQE();
+            io_uring_prep_fsync(sqe, bs->dsk.meta_fd, IORING_FSYNC_DATASYNC);
+            last_read_offset = 0;
+            data->iov = { 0 };
+            data->callback = [this](ring_data_t *data) { handle_event(data, -1, "fsync metadata"); };
+            submitted++;
+            bs->ringloop->submit();
+resume_4:
+            if (submitted > 0)
+            {
+                wait_state = 4;
+                return 1;
+            }
+        }
+        if (!header_written)
+        {
+            GET_SQE();
+            hdr->zero = 0;
+            hdr->magic = BLOCKSTORE_META_MAGIC_V1;
+            hdr->version = bs->dsk.meta_format;
+            hdr->meta_block_size = bs->dsk.meta_block_size;
+            hdr->data_block_size = bs->dsk.data_block_size;
+            hdr->bitmap_granularity = bs->dsk.bitmap_granularity;
+            if (bs->dsk.meta_format >= BLOCKSTORE_META_FORMAT_V2)
+            {
+                hdr->data_csum_type = bs->dsk.data_csum_type;
+                hdr->csum_block_size = bs->dsk.csum_block_size;
+                hdr->header_csum = 0;
+                hdr->header_csum = crc32c(0, hdr, sizeof(*hdr));
+            }
+            header_written = true;
+            last_read_offset = 0;
+            data->iov = (struct iovec){ hdr, (size_t)bs->dsk.meta_block_size };
+            data->callback = [this](ring_data_t *data) { handle_event(data, -1, "write metadata header"); };
+            io_uring_prep_writev(sqe, bs->dsk.meta_fd, &data->iov, 1, bs->dsk.meta_offset);
+            bs->ringloop->submit();
+            submitted++;
+resume_3:
+            if (submitted > 0)
+            {
+                wait_state = 3;
+                return 1;
+            }
+            goto do_fsync;
+        }
+    }
     if (!bs->inmemory_meta)
     {
         free(metadata_buffer);
         metadata_buffer = NULL;
     }
-    if (zero_on_init && !bs->disable_meta_fsync)
-    {
-        GET_SQE();
-        io_uring_prep_fsync(sqe, bs->dsk.meta_fd, IORING_FSYNC_DATASYNC);
-        last_read_offset = 0;
-        data->iov = { 0 };
-        data->callback = [this](ring_data_t *data) { handle_event(data, -1); };
-        submitted++;
-        bs->ringloop->submit();
-    resume_4:
-        if (submitted > 0)
-        {
-            wait_state = 4;
-            return 1;
-        }
-    }
+    free(hdr);
+    hdr = NULL;
     return 0;
 }
 
@@ -345,6 +358,8 @@ bool blockstore_init_meta::handle_meta_block(uint8_t *buf, uint64_t entries_per_
 {
     bool updated = false;
     uint64_t max_i = entries_per_block;
+    if (done_cnt > bs->dsk.block_count)
+        return false;
     if (max_i > bs->dsk.block_count-done_cnt)
         max_i = bs->dsk.block_count-done_cnt;
     for (uint64_t i = 0; i < max_i; i++)
@@ -455,21 +470,21 @@ blockstore_init_journal::blockstore_init_journal(blockstore_impl_t *bs)
     };
 }
 
-void blockstore_init_journal::handle_event(ring_data_t *data1)
+void blockstore_init_journal::handle_event(ring_data_t *data)
 {
-    if (data1->res <= 0)
+    if (data->res != data->iov.iov_len)
     {
-        throw std::runtime_error(
-            std::string("read journal failed at offset ") + std::to_string(journal_pos) +
-            std::string(": ") + strerror(-data1->res)
-        );
+        throw std::runtime_error(strprintf(
+            "read journal failed at offset %ju: got %s (code %d), but expected %zu",
+            journal_pos, strerror(-data->res), data->res, data->iov.iov_len
+        ));
     }
     done.push_back({
         .buf = submitted_buf,
         .pos = journal_pos,
-        .len = (uint64_t)data1->res,
+        .len = (uint64_t)data->res,
     });
-    journal_pos += data1->res;
+    journal_pos += data->res;
     if (journal_pos >= bs->journal.len)
     {
         // Continue from the beginning

src/blockstore/v1/init.h

@@ -16,7 +16,9 @@ class blockstore_init_meta
     blockstore_impl_t *bs;
     int wait_state = 0;
     bool zero_on_init = false;
+    bool header_written = false;
     void *metadata_buffer = NULL;
+    blockstore_meta_header_v2_t *hdr = NULL;
     blockstore_init_meta_buf bufs[2] = {};
     int submitted = 0;
     struct io_uring_sqe *sqe;
@@ -29,7 +31,7 @@ class blockstore_init_meta
     int i = 0, j = 0;
     std::vector<uint64_t> entries_to_zero;
     bool handle_meta_block(uint8_t *buf, uint64_t count, uint64_t done_cnt);
-    void handle_event(ring_data_t *data, int buf_num);
+    void handle_event(ring_data_t *data, int buf_num, const char *op);
 public:
     blockstore_init_meta(blockstore_impl_t *bs);
     int loop();

src/blockstore/v1/read.cpp

@@ -101,8 +101,8 @@ int blockstore_impl_t::fulfill_read(blockstore_op_t *read_op,
                     .copy_flags = COPY_BUF_JOURNAL|COPY_BUF_CSUM_FILL,
                     .offset = blk_begin,
                     .len = blk_end-blk_begin,
-                    .csum_buf = (csum + (blk_begin/dsk.csum_block_size -
-                        item_start/dsk.csum_block_size) * (dsk.data_csum_type & 0xFF)),
+                    .csum_buf = (!csum ? NULL : (csum + (blk_begin/dsk.csum_block_size -
+                        item_start/dsk.csum_block_size) * (dsk.data_csum_type & 0xFF))),
                     .dyn_data = dyn_data,
                 });
                 if (dyn_data)
@@ -134,7 +134,7 @@ int blockstore_impl_t::fulfill_read(blockstore_op_t *read_op,
                 // If we don't track it then we may IN THEORY read another object's data:
                 // submit read -> remove the object -> flush remove -> overwrite with another object -> finish read
                 // Very improbable, but possible
-                PRIV(read_op)->clean_block_used = 1;
+                PRIV(read_op)->clean_loc_used = UINT64_MAX;
             }
             rv.insert(rv.begin() + pos, el);
             fulfilled += el.len;
@@ -167,7 +167,8 @@ uint8_t* blockstore_impl_t::get_clean_entry_bitmap(uint64_t block_loc, int offse
 }
 
 int blockstore_impl_t::fill_partial_checksum_blocks(std::vector<copy_buffer_t> & rv, uint64_t & fulfilled,
-    uint8_t *clean_entry_bitmap, int *dyn_data, bool from_journal, uint8_t *read_buf, uint64_t read_offset, uint64_t read_end)
+    uint8_t *clean_entry_bitmap, int *dyn_data, bool from_journal, uint8_t *read_buf,
+    uint32_t read_offset, uint32_t read_end, uint32_t item_start, uint32_t item_end)
 {
     if (read_end == read_offset)
         return 0;
@@ -175,10 +176,38 @@ int blockstore_impl_t::fill_partial_checksum_blocks(std::vector<copy_buffer_t> &
     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;
     uint32_t end_block = 0;
+    auto zero_range = [&](int pos, bool alloc, uint32_t cur_start, uint32_t cur_end)
+    {
+        if (alloc)
+            return 0;
+        copy_buffer_t el = {
+            .copy_flags = COPY_BUF_ZERO,
+            .offset = cur_start,
+            .len = cur_end-cur_start,
+        };
+        rv.insert(rv.begin() + pos, el);
+        if (read_buf)
+            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, zero_range);
+        read_offset = item_start;
+    }
+    if (read_end > item_end)
+    {
+        // Zero-fill the end
+        find_holes(rv, item_end, read_end, zero_range);
+        read_end = item_end;
+    }
     while (start_block <= last_block)
     {
-        if (read_range_fulfilled(rv, fulfilled, read_buf, clean_entry_bitmap,
+        if (read_range_fulfilled(rv, fulfilled, read_buf, from_journal ? NULL : clean_entry_bitmap,
             start_block*dsk.csum_block_size < read_offset ? read_offset : start_block*dsk.csum_block_size,
             (start_block+1)*dsk.csum_block_size > read_end ? read_end : (start_block+1)*dsk.csum_block_size))
         {
@@ -190,7 +219,7 @@ int blockstore_impl_t::fill_partial_checksum_blocks(std::vector<copy_buffer_t> &
             // Find a sequence of checksum blocks required to be read
             end_block = start_block;
             while ((end_block+1)*dsk.csum_block_size < read_end &&
-                !read_range_fulfilled(rv, fulfilled, read_buf, clean_entry_bitmap,
+                !read_range_fulfilled(rv, fulfilled, read_buf, from_journal ? NULL : clean_entry_bitmap,
                     (end_block+1)*dsk.csum_block_size < read_offset ? read_offset : (end_block+1)*dsk.csum_block_size,
                     (end_block+2)*dsk.csum_block_size > read_end ? read_end : (end_block+2)*dsk.csum_block_size))
             {
@@ -202,8 +231,10 @@ int blockstore_impl_t::fill_partial_checksum_blocks(std::vector<copy_buffer_t> &
                 .copy_flags = COPY_BUF_CSUM_FILL | (from_journal ? COPY_BUF_JOURNALED_BIG : 0),
                 .offset = start_block*dsk.csum_block_size,
                 .len = (end_block-start_block)*dsk.csum_block_size,
-                // save clean_entry_bitmap if we're reading clean data from the journal
-                .csum_buf = from_journal ? clean_entry_bitmap : NULL,
+                // 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)
+                    : NULL,
                 .dyn_data = dyn_data,
             });
             if (dyn_data)
@@ -226,6 +257,11 @@ bool blockstore_impl_t::read_range_fulfilled(std::vector<copy_buffer_t> & rv, ui
     {
         if (alloc)
             return 0;
+        if (!clean_entry_bitmap)
+        {
+            all_done = false;
+            return 0;
+        }
         int diff = 0;
         uint32_t bmp_start = cur_start/dsk.bitmap_granularity;
         uint32_t bmp_end = cur_end/dsk.bitmap_granularity;
@@ -323,7 +359,7 @@ bool blockstore_impl_t::read_checksum_block(blockstore_op_t *op, int rv_pos, uin
             {
                 iov[n_iov++] = (struct iovec){ (uint8_t*)op->buf+cur_start-op->offset, lim_end-cur_start };
                 rv.insert(rv.begin() + pos, (copy_buffer_t){
-                    .copy_flags = COPY_BUF_DATA,
+                    .copy_flags = COPY_BUF_DATA|COPY_BUF_COALESCED,
                     .offset = cur_start,
                     .len = lim_end-cur_start,
                 });
@@ -361,10 +397,10 @@ bool blockstore_impl_t::read_checksum_block(blockstore_op_t *op, int rv_pos, uin
         PRIV(op)->pending_ops++;
         io_uring_prep_readv(sqe, submit_fd, iov + n_pos, n_cur, submit_offset + clean_loc + item_start + d_pos);
         data->callback = [this, op](ring_data_t *data) { handle_read_event(data, op); };
-        if (n_pos > 0 || n_pos + IOV_MAX < n_iov)
+        if (n_pos > 0 || n_iov > IOV_MAX)
         {
             uint32_t d_len = 0;
-            for (int i = 0; i < IOV_MAX; i++)
+            for (int i = 0; i < n_cur; i++)
                 d_len += iov[n_pos+i].iov_len;
             data->iov.iov_len = d_len;
             d_pos += d_len;
@@ -376,7 +412,7 @@ bool blockstore_impl_t::read_checksum_block(blockstore_op_t *op, int rv_pos, uin
     {
         // Reads running parallel to flushes of the same clean block may read
         // a mixture of old and new data. So we don't verify checksums for such blocks.
-        PRIV(op)->clean_block_used = 1;
+        PRIV(op)->clean_loc_used = UINT64_MAX;
     }
     return true;
 }
@@ -402,7 +438,7 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
     }
     uint64_t fulfilled = 0;
     PRIV(read_op)->pending_ops = 0;
-    PRIV(read_op)->clean_block_used = 0;
+    PRIV(read_op)->clean_loc_used = 0;
     auto & rv = PRIV(read_op)->read_vec;
     uint64_t result_version = 0;
     if (dirty_found)
@@ -515,26 +551,50 @@ int blockstore_impl_t::dequeue_read(blockstore_op_t *read_op)
     return 2;
 undo_read:
     // need to wait. undo added requests, don't dequeue op
-    if (dsk.csum_block_size > dsk.bitmap_granularity)
+    release_clean(read_op);
+    for (auto & vec: rv)
     {
-        for (auto & vec: rv)
+        if ((vec.copy_flags & COPY_BUF_CSUM_FILL) && vec.buf)
         {
-            if ((vec.copy_flags & COPY_BUF_CSUM_FILL) && vec.buf)
-            {
-                free(vec.buf);
-                vec.buf = NULL;
-            }
-            if (vec.dyn_data && --(*vec.dyn_data) == 0) // refcount
-            {
-                free(vec.dyn_data);
-                vec.dyn_data = NULL;
-            }
+            free(vec.buf);
+            vec.buf = NULL;
+        }
+        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)
+{
+    if (PRIV(op)->clean_loc_used == UINT64_MAX)
+    {
+        PRIV(op)->clean_loc_used = 0;
+    }
+    if (PRIV(op)->clean_loc_used)
+    {
+        // Release clean data block
+        auto uo_it = used_clean_objects.find(PRIV(op)->clean_loc_used - 1);
+        if (uo_it != used_clean_objects.end())
+        {
+            uo_it->second.refs--;
+            if (uo_it->second.refs <= 0)
+            {
+                if (uo_it->second.was_freed)
+                {
+                    data_alloc->set((PRIV(op)->clean_loc_used - 1) / dsk.data_block_size, false);
+                }
+                used_clean_objects.erase(uo_it);
+            }
+        }
+        PRIV(op)->clean_loc_used = 0;
+    }
+}
+
 int blockstore_impl_t::pad_journal_read(std::vector<copy_buffer_t> & rv, copy_buffer_t & cp,
     // FIXME Passing dirty_entry& would be nicer
     uint64_t dirty_offset, uint64_t dirty_end, uint64_t dirty_loc, uint8_t *csum_ptr, int *dyn_data,
@@ -598,11 +658,15 @@ bool blockstore_impl_t::fulfill_clean_read(blockstore_op_t *read_op, uint64_t &
     {
         auto & rv = PRIV(read_op)->read_vec;
         int req = fill_partial_checksum_blocks(rv, fulfilled, clean_entry_bitmap, dyn_data, from_journal,
-            (uint8_t*)read_op->buf, read_op->offset, read_op->offset+read_op->len);
+            (uint8_t*)read_op->buf, read_op->offset, read_op->offset+read_op->len, item_start, item_end);
         if (!inmemory_meta && !from_journal && req > 0)
         {
             // Read checksums from disk
             uint8_t *csum_buf = read_clean_meta_block(read_op, clean_loc, rv.size()-req);
+            if (!csum_buf)
+            {
+                return false;
+            }
             for (int i = req; i > 0; i--)
             {
                 rv[rv.size()-i].csum_buf = csum_buf;
@@ -615,7 +679,7 @@ bool blockstore_impl_t::fulfill_clean_read(blockstore_op_t *read_op, uint64_t &
                 return false;
             }
         }
-        PRIV(read_op)->clean_block_used = req > 0;
+        PRIV(read_op)->clean_loc_used = req > 0 ? UINT64_MAX : 0;
     }
     else if (from_journal)
     {
@@ -665,6 +729,10 @@ bool blockstore_impl_t::fulfill_clean_read(blockstore_op_t *read_op, uint64_t &
                 {
                     // Read checksums from disk
                     csum_buf = read_clean_meta_block(read_op, clean_loc, PRIV(read_op)->read_vec.size());
+                    if (!csum_buf)
+                    {
+                        return false;
+                    }
                     csum_done = true;
                 }
                 uint8_t *csum = !dsk.csum_block_size ? 0 : (csum_buf + 2*dsk.clean_entry_bitmap_size + bmp_start*(dsk.data_csum_type & 0xFF));
@@ -679,13 +747,13 @@ bool blockstore_impl_t::fulfill_clean_read(blockstore_op_t *read_op, uint64_t &
         }
     }
     // Increment reference counter if clean data is being read from the disk
-    if (PRIV(read_op)->clean_block_used)
+    if (PRIV(read_op)->clean_loc_used == UINT64_MAX)
     {
         auto & uo = used_clean_objects[clean_loc];
         uo.refs++;
         if (dsk.csum_block_size && flusher->is_mutated(clean_loc))
             uo.was_changed = true;
-        PRIV(read_op)->clean_block_used = clean_loc;
+        PRIV(read_op)->clean_loc_used = clean_loc + 1;
     }
     return true;
 }
@@ -725,12 +793,18 @@ bool blockstore_impl_t::verify_padded_checksums(uint8_t *clean_entry_bitmap, uin
         while (pos < iov[i].iov_len)
         {
             uint32_t start = pos;
-            uint8_t bit = (clean_entry_bitmap[bmp_pos >> 3] >> (bmp_pos & 0x7)) & 1;
-            while (pos < iov[i].iov_len && ((clean_entry_bitmap[bmp_pos >> 3] >> (bmp_pos & 0x7)) & 1) == bit)
+            uint8_t bit = 1;
+            if (clean_entry_bitmap)
             {
-                pos += dsk.bitmap_granularity;
-                bmp_pos++;
+                bit = (clean_entry_bitmap[bmp_pos >> 3] >> (bmp_pos & 0x7)) & 1;
+                while (pos < iov[i].iov_len && ((clean_entry_bitmap[bmp_pos >> 3] >> (bmp_pos & 0x7)) & 1) == bit)
+                {
+                    pos += dsk.bitmap_granularity;
+                    bmp_pos++;
+                }
             }
+            else
+                pos = iov[i].iov_len;
             uint32_t len = pos-start;
             auto buf = (uint8_t*)iov[i].iov_base+start;
             while (block_done+len >= dsk.csum_block_size)
@@ -807,7 +881,7 @@ bool blockstore_impl_t::verify_clean_padded_checksums(blockstore_op_t *op, uint6
 {
     uint32_t offset = clean_loc % dsk.data_block_size;
     if (from_journal)
-        return verify_padded_checksums(dyn_data, dyn_data + dsk.clean_entry_bitmap_size, offset, iov, n_iov, bad_block_cb);
+        return verify_padded_checksums(NULL, dyn_data, offset, iov, n_iov, bad_block_cb);
     clean_loc = (clean_loc / dsk.data_block_size) * dsk.data_block_size;
     if (!dyn_data)
     {
@@ -835,7 +909,7 @@ void blockstore_impl_t::handle_read_event(ring_data_t *data, blockstore_op_t *op
             void *meta_block = NULL;
             if (dsk.csum_block_size > dsk.bitmap_granularity)
             {
-                for (int i = rv.size()-1; i >= 0 && (rv[i].copy_flags & COPY_BUF_CSUM_FILL); i--)
+                for (int i = 0; i < rv.size(); i++)
                 {
                     if (rv[i].copy_flags & COPY_BUF_META_BLOCK)
                     {
@@ -845,8 +919,41 @@ void blockstore_impl_t::handle_read_event(ring_data_t *data, blockstore_op_t *op
                         rv[i].buf = NULL;
                         continue;
                     }
-                    struct iovec *iov = (struct iovec*)((uint8_t*)rv[i].buf + (rv[i].len & 0xFFFFFFFF));
-                    int n_iov = rv[i].len >> 32;
+                    if (rv[i].copy_flags & COPY_BUF_ZERO)
+                    {
+                        // Zero read
+                        continue;
+                    }
+                    if (rv[i].copy_flags & COPY_BUF_COALESCED)
+                    {
+                        // Sub-block shared with another read. Skip
+                        continue;
+                    }
+                    if ((rv[i].copy_flags & COPY_BUF_JOURNAL) && journal.inmemory)
+                    {
+                        // Do not check journal checksums in-memory
+                        continue;
+                    }
+                    iovec single_iov = {};
+                    iovec *iov = NULL;
+                    int n_iov = 0;
+                    if (rv[i].copy_flags & COPY_BUF_CSUM_FILL)
+                    {
+                        // Padded, buffer list passed using a 'creepy way'
+                        iov = (struct iovec*)((uint8_t*)rv[i].buf + (rv[i].len & 0xFFFFFFFF));
+                        n_iov = rv[i].len >> 32;
+                    }
+                    else
+                    {
+                        // Not padded, buffer is fully within the input buffer
+                        assert(op->buf);
+                        assert(rv[i].csum_buf);
+                        iov = &single_iov;
+                        n_iov = 1;
+                        assert(rv[i].offset >= op->offset);
+                        assert(rv[i].offset + rv[i].len <= op->offset + op->len);
+                        single_iov = { .iov_base = op->buf + rv[i].offset - op->offset, .iov_len = rv[i].len };
+                    }
                     bool ok = true;
                     if (rv[i].copy_flags & COPY_BUF_JOURNAL)
                     {
@@ -944,23 +1051,7 @@ void blockstore_impl_t::handle_read_event(ring_data_t *data, blockstore_op_t *op
                 meta_block = NULL;
             }
         }
-        if (PRIV(op)->clean_block_used)
-        {
-            // Release clean data block
-            auto uo_it = used_clean_objects.find(PRIV(op)->clean_block_used);
-            if (uo_it != used_clean_objects.end())
-            {
-                uo_it->second.refs--;
-                if (uo_it->second.refs <= 0)
-                {
-                    if (uo_it->second.was_freed)
-                    {
-                        data_alloc->set(PRIV(op)->clean_block_used, false);
-                    }
-                    used_clean_objects.erase(uo_it);
-                }
-            }
-        }
+        release_clean(op);
         if (!journal.inmemory)
         {
             // Release journal sector usage

src/blockstore/v1/stable.cpp

@@ -491,7 +491,7 @@ void blockstore_impl_t::mark_stable(obj_ver_id v, bool forget_dirty)
                     if (!exists)
                     {
                         uint64_t space_id = dirty_it->first.oid.inode;
-                        if (no_inode_stats[dirty_it->first.oid.inode >> (64-POOL_ID_BITS)])
+                        if (no_inode_stats.find(dirty_it->first.oid.inode >> (64-POOL_ID_BITS)) != no_inode_stats.end())
                             space_id = space_id & ~(((uint64_t)1 << (64-POOL_ID_BITS)) - 1);
                         inode_space_stats[space_id] += dsk.data_block_size;
                         used_blocks++;
@@ -501,7 +501,7 @@ void blockstore_impl_t::mark_stable(obj_ver_id v, bool forget_dirty)
                 else if (IS_DELETE(dirty_it->second.state))
                 {
                     uint64_t space_id = dirty_it->first.oid.inode;
-                    if (no_inode_stats[dirty_it->first.oid.inode >> (64-POOL_ID_BITS)])
+                    if (no_inode_stats.find(dirty_it->first.oid.inode >> (64-POOL_ID_BITS)) != no_inode_stats.end())
                         space_id = space_id & ~(((uint64_t)1 << (64-POOL_ID_BITS)) - 1);
                     auto & sp = inode_space_stats[space_id];
                     if (sp > dsk.data_block_size)

src/client/CMakeLists.txt

@@ -3,6 +3,27 @@ cmake_minimum_required(VERSION 2.8...3.30)
 project(vitastor)
 
 # libvitastor_common.a
+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})
+
+# libvitastor_net.a
 set(MSGR_RDMA "")
 if (IBVERBS_LIBRARIES)
 	set(MSGR_RDMA "msgr_rdma.cpp")
@@ -11,28 +32,48 @@ set(MSGR_RDMACM "")
 if (RDMACM_LIBRARIES)
 	set(MSGR_RDMACM "msgr_rdmacm.cpp")
 endif (RDMACM_LIBRARIES)
-add_library(vitastor_common STATIC
-	../util/epoll_manager.cpp etcd_state_client.cpp messenger.cpp msgr_iothread.cpp ../util/addr_util.cpp
-	msgr_stop.cpp msgr_op.cpp msgr_send.cpp msgr_receive.cpp ../util/ringloop.cpp ../../json11/json11.cpp
-	http_client.cpp osd_ops.cpp pg_states.cpp ../util/timerfd_manager.cpp ../util/str_util.cpp ../util/json_util.cpp ${MSGR_RDMA} ${MSGR_RDMACM}
+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}
 )
-target_link_libraries(vitastor_common pthread)
-target_compile_options(vitastor_common PUBLIC -fPIC)
+target_link_libraries(vitastor_net pthread vitastor_common ${CARES_LIBRARIES})
+target_compile_options(vitastor_net PUBLIC -fPIC)
 
-# libvitastor_client.so
-add_library(vitastor_client SHARED
+# libvitastor_client_int.a
+add_library(vitastor_client_int STATIC
 	cluster_client.cpp
+	cluster_client_real.cpp
 	cluster_client_list.cpp
 	cluster_client_wb.cpp
-	vitastor_c.cpp
+	cluster_client_icache.cpp
 )
-set_target_properties(vitastor_client PROPERTIES PUBLIC_HEADER "client/vitastor_c.h")
-target_link_libraries(vitastor_client
-	vitastor_common
+target_link_libraries(vitastor_client_int
+	vitastor_net
 	vitastor_cli
 	${LIBURING_LIBRARIES}
 	${IBVERBS_LIBRARIES}
 	${RDMACM_LIBRARIES}
+	${OPENSSL_LIBRARIES}
+	${ISAL_CRYPTO_LIBRARIES}
+)
+target_compile_options(vitastor_client_int PUBLIC -fPIC)
+
+# libvitastor_client.so
+add_library(vitastor_client SHARED
+	vitastor_c.cpp
+)
+set_target_properties(vitastor_client PROPERTIES PUBLIC_HEADER "client/vitastor_c.h")
+target_link_libraries(vitastor_client
+	vitastor_client_int
 )
 set_target_properties(vitastor_client PROPERTIES VERSION ${VITASTOR_VERSION} SOVERSION 0)
 configure_file(vitastor.pc.in vitastor.pc @ONLY)
@@ -95,11 +136,15 @@ endif (${WITH_QEMU})
 add_executable(test_cluster_client
 	EXCLUDE_FROM_ALL
 	../test/test_cluster_client.cpp
-	pg_states.cpp osd_ops.cpp cluster_client.cpp cluster_client_list.cpp cluster_client_wb.cpp msgr_op.cpp ../test/mock/messenger.cpp msgr_stop.cpp
-	etcd_state_client.cpp ../util/timerfd_manager.cpp ../util/addr_util.cpp ../util/str_util.cpp ../util/json_util.cpp ../../json11/json11.cpp
+	cluster_client.cpp
+	cluster_client_list.cpp
+	cluster_client_wb.cpp
+	cluster_client_icache.cpp
+	../test/mock/messenger.cpp
+	../test/mock/vault.cpp
+	etcd_state_client_mock.cpp
 )
-target_link_libraries(test_cluster_client ${LIBURING_LIBRARIES})
-target_compile_definitions(test_cluster_client PUBLIC -D__MOCK__)
+target_link_libraries(test_cluster_client vitastor_common ${LIBURING_LIBRARIES} ${OPENSSL_LIBRARIES} ${ISAL_CRYPTO_LIBRARIES})
 target_include_directories(test_cluster_client BEFORE PUBLIC ${CMAKE_SOURCE_DIR}/src/test/mock)
 add_dependencies(build_tests test_cluster_client)
 add_test(NAME test_cluster_client COMMAND test_cluster_client)

src/client/cluster_client.cpp

@@ -11,7 +11,7 @@
 #define TRY_SEND_CONNECTING 1
 #define TRY_SEND_OK 2
 
-cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json config)
+cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json config, std::unique_ptr<etcd_state_client_t> st_cli_ptr)
 {
     wb = new writeback_cache_t();
 
@@ -51,32 +51,31 @@ cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd
         msgr.stop_client(op->client_id);
         delete op;
     };
-    msgr.parse_config(config);
+    msgr.parse_config(config, true);
 
-    st_cli.tfd = tfd;
-    st_cli.on_load_config_hook = [this](json11::Json::object & cfg) { on_load_config_hook(cfg); };
-    st_cli.on_change_osd_state_hook = [this](uint64_t peer_osd) { on_change_osd_state_hook(peer_osd); };
-    st_cli.on_change_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); };
-    st_cli.on_change_node_placement_hook = [this]() { on_change_node_placement_hook(); };
-    st_cli.on_load_pgs_hook = [this](bool success) { on_load_pgs_hook(success); };
-    st_cli.on_reload_hook = [this]() { st_cli.load_global_config(); };
+    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](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); };
+    st_cli->on_change_node_placement_hook = [this]() { on_change_node_placement_hook(); };
+    st_cli->on_load_pgs_hook = [this](bool success) { on_load_pgs_hook(success); };
+    st_cli->on_reload_hook = [this]() { this->st_cli->load_global_config(); };
+    st_cli->on_inode_change_hook = [this](uint64_t inode, bool removed) { on_change_inode_hook(inode, removed); };
 
-    st_cli.parse_config(config);
-    st_cli.infinite_start = false;
+    st_cli->parse_config(config);
+    st_cli->infinite_start = false;
     if (!config["client_infinite_start"].is_null())
     {
-        st_cli.infinite_start = config["client_infinite_start"].bool_value();
+        st_cli->infinite_start = config["client_infinite_start"].bool_value();
     }
-    st_cli.load_global_config();
-
-    scrap_buffer_size = SCRAP_BUFFER_SIZE;
-    scrap_buffer = malloc_or_die(scrap_buffer_size);
+    st_cli->load_global_config();
 }
 
 cluster_client_t::~cluster_client_t()
 {
+    vault_destroy();
     if (retry_timeout_id >= 0)
     {
         tfd->clear_timer(retry_timeout_id);
@@ -94,7 +93,6 @@ cluster_client_t::~cluster_client_t()
     {
         ringloop->unregister_consumer(&consumer);
     }
-    free(scrap_buffer);
     delete wb;
     wb = NULL;
 }
@@ -469,7 +467,7 @@ void cluster_client_t::on_load_config_hook(json11::Json::object & etcd_global_co
         auto etcd_report_interval = config["etcd_report_interval"].uint64_value();
         if (!etcd_report_interval)
             etcd_report_interval = 5;
-        client_wait_up_timeout = 1+etcd_report_interval+(st_cli.max_etcd_attempts*(2*st_cli.etcd_quick_timeout)+999)/1000;
+        client_wait_up_timeout = 1+etcd_report_interval+(st_cli->max_etcd_attempts*(2*st_cli->etcd_quick_timeout)+999)/1000;
     }
     // log_level
     log_level = config["log_level"].uint64_value();
@@ -481,9 +479,11 @@ void cluster_client_t::on_load_config_hook(json11::Json::object & etcd_global_co
         self_tree_metrics.clear();
         client_hostname = new_hostname;
     }
-    msgr.parse_config(config);
-    st_cli.parse_config(config);
-    st_cli.load_pgs();
+    // vault
+    vault_parse_config();
+    msgr.parse_config(config, false);
+    st_cli->parse_config(config);
+    st_cli->load_pgs();
 }
 
 osd_num_t cluster_client_t::select_random_osd(const std::vector<osd_num_t> & osds)
@@ -492,7 +492,7 @@ osd_num_t cluster_client_t::select_random_osd(const std::vector<osd_num_t> & osd
     int alive_count = 0;
     for (auto & osd_num: osds)
     {
-        if (!st_cli.peer_states[osd_num].is_null())
+        if (!st_cli->peer_states[osd_num].is_null())
             alive_set[alive_count++] = osd_num;
     }
     if (!alive_count)
@@ -509,7 +509,7 @@ osd_num_t cluster_client_t::select_nearest_osd(const std::vector<osd_num_t> & os
         while (self_tree_metrics.find(cur_id) == self_tree_metrics.end())
         {
             self_tree_metrics[cur_id] = metric++;
-            json11::Json cur_placement = st_cli.node_placement[cur_id];
+            json11::Json cur_placement = st_cli->node_placement[cur_id];
             cur_id = cur_placement["parent"].string_value();
         }
         if (cur_id != "")
@@ -529,7 +529,7 @@ osd_num_t cluster_client_t::select_nearest_osd(const std::vector<osd_num_t> & os
         }
         else
         {
-            auto & peer_state = st_cli.peer_states[osd_num];
+            auto & peer_state = st_cli->peer_states[osd_num];
             if (!peer_state.is_null())
             {
                 metric = self_tree_metrics[""];
@@ -539,7 +539,7 @@ osd_num_t cluster_client_t::select_nearest_osd(const std::vector<osd_num_t> & os
                 while (seen.find(cur_id) == seen.end())
                 {
                     seen.insert(cur_id);
-                    json11::Json cur_placement = st_cli.node_placement[cur_id];
+                    json11::Json cur_placement = st_cli->node_placement[cur_id];
                     std::string cur_parent = cur_placement["parent"].string_value();
                     cur_id = (!first || cur_parent != "" ? cur_parent : peer_state["host"].string_value());
                     first = false;
@@ -564,7 +564,7 @@ osd_num_t cluster_client_t::select_nearest_osd(const std::vector<osd_num_t> & os
 
 void cluster_client_t::on_load_pgs_hook(bool success)
 {
-    for (auto & pool_item: st_cli.pool_config)
+    for (auto & pool_item: st_cli->pool_config)
     {
         pg_counts[pool_item.first] = pool_item.second.real_pg_count;
     }
@@ -584,7 +584,7 @@ void cluster_client_t::on_load_pgs_hook(bool success)
 
 void cluster_client_t::on_change_pool_config_hook()
 {
-    for (auto & pool_item: st_cli.pool_config)
+    for (auto & pool_item: st_cli->pool_config)
     {
         if (pg_counts[pool_item.first] != pool_item.second.real_pg_count)
         {
@@ -607,12 +607,15 @@ void cluster_client_t::on_change_pool_config_hook()
             pg_counts[pool_item.first] = pool_item.second.real_pg_count;
         }
     }
+    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];
+    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
@@ -630,8 +633,8 @@ bool cluster_client_t::get_immediate_commit(uint64_t inode)
     pool_id_t pool_id = INODE_POOL(inode);
     if (!pool_id)
         return true;
-    auto pool_it = st_cli.pool_config.find(pool_id);
-    if (pool_it == st_cli.pool_config.end())
+    auto pool_it = st_cli->pool_config.find(pool_id);
+    if (pool_it == st_cli->pool_config.end())
         return true;
     return pool_it->second.immediate_commit == IMMEDIATE_ALL;
 }
@@ -641,7 +644,7 @@ 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())
     {
-        msgr.connect_peer(peer_osd, st_cli.peer_states[peer_osd]);
+        msgr.connect_peer(peer_osd, st_cli->peer_states[peer_osd]);
         continue_lists();
     }
 }
@@ -673,6 +676,10 @@ bool cluster_client_t::flush()
 {
     if (!ringloop)
     {
+        if (vault_loading)
+        {
+            return false;
+        }
         if (wb->writeback_queue.size())
         {
             wb->start_writebacks(this, 0);
@@ -695,7 +702,7 @@ bool cluster_client_t::flush()
         sync_done = true;
     };
     execute(sync);
-    while (!sync_done)
+    while (!sync_done || vault_loading)
     {
         ringloop->loop();
         if (!sync_done)
@@ -936,8 +943,8 @@ bool cluster_client_t::check_rw(cluster_op_t *op)
         cb(op);
         return false;
     }
-    auto pool_it = st_cli.pool_config.find(pool_id);
-    if (pool_it == st_cli.pool_config.end() || pool_it->second.real_pg_count == 0)
+    auto pool_it = st_cli->pool_config.find(pool_id);
+    if (pool_it == st_cli->pool_config.end() || pool_it->second.real_pg_count == 0)
     {
         // Pools are loaded, but this one is unknown
         op->retval = -EINVAL;
@@ -958,10 +965,40 @@ bool cluster_client_t::check_rw(cluster_op_t *op)
     {
         op->flags |= OP_IMMEDIATE_COMMIT;
     }
+    bool searched = false;
+    std::shared_ptr<inode_cache_t> icache;
+    if (op->opcode == OSD_OP_READ || op->opcode == OSD_OP_WRITE)
+    {
+        if (!searched)
+        {
+            icache = inode_cache_get(op->inode);
+            searched = true;
+        }
+        if (icache && icache->has_parent_loop && op->opcode == OSD_OP_READ)
+        {
+            op->retval = -EINVAL;
+            auto cb = std::move(op->callback);
+            cb(op);
+            return false;
+        }
+        if (icache && icache->op_enc)
+        {
+            // Use shared_ptr aliasing to attach op_enc to the inode cache entry
+            op->enc = std::shared_ptr<osd_op_enc_t>(icache, icache->op_enc);
+        }
+        else
+            op->enc.reset();
+    }
+    else
+        op->enc.reset();
     if ((op->opcode == OSD_OP_WRITE || op->opcode == OSD_OP_DELETE) && !(op->flags & OSD_OP_IGNORE_READONLY))
     {
-        auto ino_it = st_cli.inode_config.find(op->inode);
-        if (ino_it != st_cli.inode_config.end() && ino_it->second.readonly)
+        if (!searched)
+        {
+            icache = inode_cache_get(op->inode);
+            searched = true;
+        }
+        if (icache && icache->readonly)
         {
             op->retval = -EROFS;
             auto cb = std::move(op->callback);
@@ -972,33 +1009,39 @@ bool cluster_client_t::check_rw(cluster_op_t *op)
     op->deoptimise_snapshot = false;
     if (enable_writeback && (op->opcode == OSD_OP_READ || op->opcode == OSD_OP_READ_BITMAP || op->opcode == OSD_OP_READ_CHAIN_BITMAP))
     {
-        auto ino_it = st_cli.inode_config.find(op->inode);
-        if (ino_it != st_cli.inode_config.end())
+        if (!searched)
         {
-            int chain_size = 0;
-            while (ino_it != st_cli.inode_config.end() && ino_it->second.parent_id)
+            icache = inode_cache_get(op->inode);
+            searched = true;
+        }
+        if (icache)
+        {
+            for (auto & parent: icache->chain)
             {
-                // Check for loops - FIXME check it in etcd_state_client
-                if (ino_it->second.parent_id == op->inode ||
-                    chain_size > st_cli.inode_config.size())
-                {
-                    op->retval = -EINVAL;
-                    auto cb = std::move(op->callback);
-                    cb(op);
-                    return false;
-                }
-                if (INODE_POOL(ino_it->second.parent_id) == INODE_POOL(ino_it->first) &&
-                    wb->has_inode(ino_it->second.parent_id))
+                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;
                     break;
                 }
-                chain_size++;
-                ino_it = st_cli.inode_config.find(ino_it->second.parent_id);
             }
         }
     }
+    if (icache && icache->err_code)
+    {
+        if (icache->err_code == EPERM)
+        {
+            op->retval = -EPERM;
+            auto cb = std::move(op->callback);
+            cb(op);
+            return false;
+        }
+        else if (icache->err_code == EAGAIN)
+        {
+            key_wait_ops.push_back(op);
+            return false;
+        }
+    }
     return true;
 }
 
@@ -1014,7 +1057,7 @@ void cluster_client_t::execute_raw(osd_num_t osd_num, osd_op_t *op)
     else
     {
         if (msgr.wanted_peers.find(osd_num) == msgr.wanted_peers.end())
-            msgr.connect_peer(osd_num, st_cli.peer_states[osd_num]);
+            msgr.connect_peer(osd_num, st_cli->peer_states[osd_num]);
         raw_ops.emplace(osd_num, op);
     }
 }
@@ -1128,31 +1171,33 @@ resume_2:
         }
         if (op->opcode == OSD_OP_READ || op->opcode == OSD_OP_READ_CHAIN_BITMAP)
         {
-            // Check parent inode
-            auto ino_it = st_cli.inode_config.find(op->cur_inode);
-            // Skip parents from the same pool
-            int skipped = 0;
-            while (!op->deoptimise_snapshot &&
-                ino_it != st_cli.inode_config.end() && ino_it->second.parent_id &&
-                INODE_POOL(ino_it->second.parent_id) == INODE_POOL(op->cur_inode))
+            uint64_t next_inode = 0;
+            auto icache = inode_cache_get(op->cur_inode);
+            if (icache)
             {
-                // Check for loops - FIXME check it in etcd_state_client
-                if (ino_it->second.parent_id == op->inode ||
-                    skipped > st_cli.inode_config.size())
+                if (icache->has_parent_loop)
                 {
                     op->retval = -EINVAL;
                     erase_op(op);
                     return 1;
                 }
-                skipped++;
-                ino_it = st_cli.inode_config.find(ino_it->second.parent_id);
+                if (op->deoptimise_snapshot)
+                {
+                    if (icache->chain.size() > 1)
+                        next_inode = icache->chain[1];
+                }
+                else
+                {
+                    if (icache->other_pool_parent_id)
+                        next_inode = icache->other_pool_parent_id;
+                }
             }
-            if (ino_it != st_cli.inode_config.end() &&
-                ino_it->second.parent_id &&
-                ino_it->second.parent_id != op->inode)
+            if (next_inode)
             {
                 // Continue reading from the parent inode
-                op->cur_inode = ino_it->second.parent_id;
+                icache = inode_cache_get(next_inode);
+                op->cur_inode = next_inode;
+                op->enc = (icache && icache->op_enc ? std::shared_ptr<osd_op_enc_t>(icache, icache->op_enc) : nullptr);
                 op->parts.clear();
                 op->done_count = 0;
                 goto resume_0;
@@ -1161,7 +1206,7 @@ resume_2:
         op->retval = op->len;
         if (op->opcode == OSD_OP_READ_BITMAP || op->opcode == OSD_OP_READ_CHAIN_BITMAP)
         {
-            auto & pool_cfg = st_cli.pool_config.at(INODE_POOL(op->inode));
+            auto & pool_cfg = st_cli->pool_config.at(INODE_POOL(op->inode));
             op->retval = op->len / pool_cfg.bitmap_granularity;
         }
         if (op->flush_id)
@@ -1203,7 +1248,7 @@ resume_2:
     return 0;
 }
 
-static void add_iov(int size, bool skip, cluster_op_t *op, int &iov_idx, size_t &iov_pos, osd_op_buf_list_t &iov, void *scrap, int scrap_len)
+static void add_iov(int size, int skip, cluster_op_t *op, int &iov_idx, size_t &iov_pos, osd_op_buf_list_t &iov)
 {
     int left = size;
     while (left > 0 && iov_idx < op->iov.count)
@@ -1211,7 +1256,7 @@ static void add_iov(int size, bool skip, cluster_op_t *op, int &iov_idx, size_t
         int cur_left = op->iov.buf[iov_idx].iov_len - iov_pos;
         if (cur_left < left)
         {
-            if (!skip)
+            if (skip == 0)
             {
                 iov.push_back((uint8_t*)op->iov.buf[iov_idx].iov_base + iov_pos, cur_left);
             }
@@ -1221,7 +1266,7 @@ static void add_iov(int size, bool skip, cluster_op_t *op, int &iov_idx, size_t
         }
         else
         {
-            if (!skip)
+            if (skip == 0)
             {
                 iov.push_back((uint8_t*)op->iov.buf[iov_idx].iov_base + iov_pos, left);
             }
@@ -1230,16 +1275,10 @@ static void add_iov(int size, bool skip, cluster_op_t *op, int &iov_idx, size_t
         }
     }
     assert(left == 0);
-    if (skip && scrap_len > 0)
+    if (skip == 1)
     {
-        // All skipped ranges are read into the same useless buffer
-        left = size;
-        while (left > 0)
-        {
-            int cur_left = scrap_len < left ? scrap_len : left;
-            iov.push_back(scrap, cur_left);
-            left -= cur_left;
-        }
+        // data read into a NULL buffer will be discarded by messenger
+        iov.push_back(NULL, size);
     }
 }
 
@@ -1247,7 +1286,7 @@ void cluster_client_t::slice_rw(cluster_op_t *op)
 {
     // Slice the request into individual object stripe requests
     // Primary OSDs still operate individual stripes, but their size is multiplied by PG minsize in case of EC
-    auto & pool_cfg = st_cli.pool_config.at(INODE_POOL(op->cur_inode));
+    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;
@@ -1259,7 +1298,11 @@ void cluster_client_t::slice_rw(cluster_op_t *op)
         // Allocate memory for the bitmap
         unsigned object_bitmap_size = ((op->len / pool_cfg.bitmap_granularity + 7) / 8);
         object_bitmap_size = (object_bitmap_size < 8 ? 8 : object_bitmap_size);
-        unsigned bitmap_mem = object_bitmap_size + (pool_cfg.data_block_size / pool_cfg.bitmap_granularity / 8 * pg_data_size) * op->parts.size();
+        unsigned bitmap_mem = object_bitmap_size +
+            op->parts.size() * pg_data_size *
+            (pool_cfg.data_block_size / pool_cfg.bitmap_granularity / 8
+            // read chain_info - max 4 bytes per block
+            + (op->enc ? osd_op_rw_t::chain_info_bytes(op->enc->chain_size)*op->len/pool_cfg.bitmap_granularity : 0));
         if (!op->bitmap_buf || op->bitmap_buf_size < bitmap_mem)
         {
             op->bitmap_buf = realloc_or_die(op->bitmap_buf, bitmap_mem);
@@ -1301,10 +1344,10 @@ void cluster_client_t::slice_rw(cluster_op_t *op)
                         {
                             begin = cur;
                             // Just advance iov_idx & iov_pos
-                            add_iov(cur-prev, true, op, iov_idx, iov_pos, op->parts[i].iov, NULL, 0);
+                            add_iov(cur-prev, 2, op, iov_idx, iov_pos, op->parts[i].iov);
                         }
                         else
-                            add_iov(cur-prev, skip_prev, op, iov_idx, iov_pos, op->parts[i].iov, scrap_buffer, scrap_buffer_size);
+                            add_iov(cur-prev, skip_prev ? 1 : 0, op, iov_idx, iov_pos, op->parts[i].iov);
                     }
                     skip_prev = skip;
                     prev = cur;
@@ -1315,11 +1358,11 @@ void cluster_client_t::slice_rw(cluster_op_t *op)
             if (skip_prev)
             {
                 // Just advance iov_idx & iov_pos
-                add_iov(end-prev, true, op, iov_idx, iov_pos, op->parts[i].iov, NULL, 0);
+                add_iov(end-prev, 2, op, iov_idx, iov_pos, op->parts[i].iov);
                 end = prev;
             }
             else
-                add_iov(cur-prev, skip_prev, op, iov_idx, iov_pos, op->parts[i].iov, scrap_buffer, scrap_buffer_size);
+                add_iov(cur-prev, skip_prev ? 1 : 0, op, iov_idx, iov_pos, op->parts[i].iov);
             if (end == begin)
             {
                 op->done_count++;
@@ -1328,7 +1371,7 @@ void cluster_client_t::slice_rw(cluster_op_t *op)
         }
         else if (op->opcode != OSD_OP_READ_BITMAP && op->opcode != OSD_OP_READ_CHAIN_BITMAP && op->opcode != OSD_OP_DELETE)
         {
-            add_iov(end-begin, false, op, iov_idx, iov_pos, op->parts[i].iov, NULL, 0);
+            add_iov(end-begin, 0, op, iov_idx, iov_pos, op->parts[i].iov);
         }
         op->parts[i].parent = op;
         op->parts[i].offset = begin;
@@ -1346,7 +1389,7 @@ bool cluster_client_t::affects_pg(uint64_t inode, uint64_t offset, uint64_t len,
     {
         return false;
     }
-    auto & pool_cfg = st_cli.pool_config.at(INODE_POOL(inode));
+    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;
@@ -1365,7 +1408,7 @@ bool cluster_client_t::affects_pg(uint64_t inode, uint64_t offset, uint64_t len,
 
 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));
+    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;
@@ -1389,7 +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_cfg = st_cli.pool_config.at(INODE_POOL(op->cur_inode));
+    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 &&
@@ -1414,14 +1457,17 @@ int cluster_client_t::try_send(cluster_op_t *op, int i, std::function<void(osd_o
             osd_client_t *cl = peer_it->second;
             part->flags |= PART_SENT|PART_VALID;
             op->inflight_count++;
-            uint64_t pg_bitmap_size = (pool_cfg.data_block_size / pool_cfg.bitmap_granularity / 8) * (
-                pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks
-            );
+            uint32_t pg_data_size = (pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks);
+            uint64_t pg_bitmap_size = pg_data_size * (pool_cfg.data_block_size / pool_cfg.bitmap_granularity / 8
+                // read chain_info - max 4 bytes per block
+                + (op->opcode == OSD_OP_READ && op->enc
+                    ? osd_op_rw_t::chain_info_bytes(op->enc->chain_size)*pool_cfg.data_block_size/pool_cfg.bitmap_granularity
+                    : 0));
             uint64_t meta_rev = 0;
             if (op->opcode != OSD_OP_READ_BITMAP && op->opcode != OSD_OP_DELETE && !op->deoptimise_snapshot)
             {
-                auto ino_it = st_cli.inode_config.find(op->cur_inode);
-                if (ino_it != st_cli.inode_config.end())
+                auto ino_it = st_cli->inode_config.find(op->cur_inode);
+                if (ino_it != st_cli->inode_config.end())
                     meta_rev = ino_it->second.mod_revision;
             }
             part->op = (osd_op_t){
@@ -1435,6 +1481,7 @@ int cluster_client_t::try_send(cluster_op_t *op, int i, std::function<void(osd_o
                     .inode = op->cur_inode,
                     .offset = part->offset,
                     .len = part->len,
+                    .flags = op->opcode == OSD_OP_READ && op->enc && !op->deoptimise_snapshot ? OSD_OP_RETURN_CHAIN : 0,
                     .meta_revision = meta_rev,
                     .version = op->opcode == OSD_OP_WRITE || op->opcode == OSD_OP_DELETE ? op->version : 0,
                 } },
@@ -1442,6 +1489,7 @@ int cluster_client_t::try_send(cluster_op_t *op, int i, std::function<void(osd_o
                     ? (uint8_t*)op->part_bitmaps + pg_bitmap_size*i : NULL),
                 .bitmap_len = (unsigned)(op->opcode == OSD_OP_READ || op->opcode == OSD_OP_READ_BITMAP || op->opcode == OSD_OP_READ_CHAIN_BITMAP
                     ? pg_bitmap_size : 0),
+                .enc = op->enc,
                 .callback = cb ? cb : [this, part](osd_op_t *op_part)
                 {
                     handle_op_part(part);
@@ -1453,7 +1501,7 @@ int cluster_client_t::try_send(cluster_op_t *op, int i, std::function<void(osd_o
         }
         else if (msgr.wanted_peers.find(primary_osd) == msgr.wanted_peers.end())
         {
-            msgr.connect_peer(primary_osd, st_cli.peer_states[primary_osd]);
+            msgr.connect_peer(primary_osd, st_cli->peer_states[primary_osd]);
             return TRY_SEND_CONNECTING;
         }
     }
@@ -1559,6 +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;
@@ -1567,15 +1618,10 @@ void cluster_client_t::handle_op_part(cluster_op_part_t *part)
     {
         // Operation failed, retry
         part->flags |= PART_ERROR;
-        if (!op->retval || op->retval == -EPIPE ||
-            part->op.reply.hdr.retval == -ENOSPC && op->retval == -ETIMEDOUT ||
-            part->op.reply.hdr.retval == -EIO)
-        {
-            // Error priority: EIO > ENOSPC > ETIMEDOUT > EPIPE
+        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)
+        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)
@@ -1648,7 +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_cfg = st_cli.pool_config.at(INODE_POOL(op->cur_inode));
+    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

@@ -4,7 +4,8 @@
 #pragma once
 
 #include "messenger.h"
-#include "etcd_state_client.h"
+#include "etcd_state_client_http.h"
+#include "../util/robin_hood.h"
 
 #define DEFAULT_CLIENT_MAX_DIRTY_BYTES 32*1024*1024
 #define DEFAULT_CLIENT_MAX_DIRTY_OPS 1024
@@ -71,6 +72,7 @@ protected:
     cluster_op_t *prev = NULL, *next = NULL;
     int prev_wait = 0;
     uint64_t flush_id = 0;
+    std::shared_ptr<osd_op_enc_t> enc;
     friend class cluster_client_t;
     friend class writeback_cache_t;
 };
@@ -80,14 +82,33 @@ struct inode_list_osd_t;
 struct inode_list_pg_t;
 class writeback_cache_t;
 
+struct inode_cache_t
+{
+    std::vector<inode_t> chain; // only parents from the same pool
+    uint8_t *key_data = NULL;
+    osd_op_enc_t *op_enc = NULL;
+    bool readonly = false;
+    bool has_parent_loop = false;
+    inode_t other_pool_parent_id = 0;
+    int err_code = 0;
+
+    ~inode_cache_t();
+};
+
+struct vault_load_key_t
+{
+    int key_state = 0;
+    std::string key;
+};
+
 // FIXME: Split into public and private interfaces
 class __attribute__((visibility("default"))) cluster_client_t
 {
     timerfd_manager_t *tfd = NULL;
     ring_loop_t *ringloop = NULL;
 
-    std::map<pool_id_t, uint64_t> pg_counts;
-    std::map<pool_pg_num_t, osd_num_t> pg_primary;
+    // config:
+
     // client_max_dirty_* is actually "max unsynced", for the case when immediate_commit is off
     uint64_t client_max_dirty_bytes = 0;
     uint64_t client_max_dirty_ops = 0;
@@ -99,12 +120,23 @@ class __attribute__((visibility("default"))) cluster_client_t
     uint64_t client_max_writeback_iodepth = 0;
     std::string conf_hostname;
 
+    std::string vault_url;
+    std::string vault_client_cert;
+    std::string vault_client_key;
+    std::string vault_ca;
+    std::string vault_secret_api_path;
+    uint64_t vault_timeout_ms = 0;
+    uint64_t vault_error_timeout_sec = 0;
+    uint64_t vault_refresh_leeway_sec = 0;
+
     int log_level = 0;
     int client_retry_interval = 50; // ms
     int client_eio_retry_interval = 1000; // ms
     bool client_retry_enospc = true;
     int client_wait_up_timeout = 16; // sec (for listings)
 
+    // state:
+
     std::string client_hostname;
     std::map<std::string, int> self_tree_metrics;
     std::map<osd_num_t, int> osd_tree_metrics;
@@ -112,15 +144,28 @@ class __attribute__((visibility("default"))) cluster_client_t
     int retry_timeout_id = -1;
     int retry_timeout_duration = 0;
     std::vector<cluster_op_t*> offline_ops;
+    std::vector<cluster_op_t*> key_wait_ops;
     cluster_op_t *op_queue_head = NULL, *op_queue_tail = NULL;
     writeback_cache_t *wb = NULL;
     std::set<osd_num_t> dirty_osds;
     uint64_t dirty_bytes = 0, dirty_ops = 0;
 
-    void *scrap_buffer = NULL;
-    unsigned scrap_buffer_size = 0;
+    // inodes require some extra state for read/write, it's stored here.
+    // moreover, robin_hood access is slightly faster than std::map :)
+    robin_hood::unordered_flat_map<inode_t, std::shared_ptr<inode_cache_t>> inode_cache;
+    std::set<std::pair<inode_t, inode_t>> inode_cache_children;
+
+    http_context_t *vault_http_ctx = NULL;
+    http_co_t *vault_http_cli = NULL;
+    bool vault_loading = false;
+    std::string vault_token;
+    bool vault_auth_error = false;
+    timespec vault_token_expire = {};
+    std::vector<std::string> vault_key_load_queue;
+    std::map<std::string, vault_load_key_t> vault_keys;
 
     bool pgs_loaded = false;
+    std::map<pool_id_t, uint64_t> pg_counts;
     ring_consumer_t consumer;
     std::vector<std::function<void(void)>> on_ready_hooks;
     int list_retry_timeout_id = -1;
@@ -131,7 +176,7 @@ class __attribute__((visibility("default"))) cluster_client_t
     bool msgr_initialized = false;
 
 public:
-    etcd_state_client_t st_cli;
+    std::unique_ptr<etcd_state_client_t> st_cli;
 
     osd_messenger_t msgr;
     void init_msgr();
@@ -139,7 +184,8 @@ public:
     json11::Json::object cli_config, file_config, etcd_global_config;
     json11::Json::object config;
 
-    cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json config);
+    static cluster_client_t* create(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json config);
+    cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json config, std::unique_ptr<etcd_state_client_t> st_cli);
     ~cluster_client_t();
     void execute(cluster_op_t *op);
     void execute_raw(osd_num_t osd_num, osd_op_t *op);
@@ -155,6 +201,13 @@ public:
 protected:
     void continue_ops(int time_passed = 0);
 
+    std::shared_ptr<inode_cache_t> inode_cache_get(inode_t ino);
+    void vault_parse_config();
+    bool vault_check_token();
+    void vault_load_keys();
+    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);
 
@@ -164,6 +217,7 @@ protected:
     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(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);
@@ -180,6 +234,7 @@ protected:
     void erase_op(cluster_op_t *op);
     void calc_wait(cluster_op_t *op);
     void inc_wait(uint64_t opcode, uint64_t flags, cluster_op_t *next, int inc);
+
     void continue_lists();
     bool continue_listing(inode_list_t *lst);
     bool restart_listing(inode_list_t* lst);

src/client/cluster_client_icache.cpp

@@ -0,0 +1,245 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#include <stdexcept>
+#include <assert.h>
+#include "cluster_client_impl.h"
+#include "str_util.h"
+
+inode_cache_t::~inode_cache_t()
+{
+    if (key_data)
+    {
+        free(key_data);
+        key_data = NULL;
+        op_enc = NULL;
+    }
+}
+
+void cluster_client_t::vault_parse_config()
+{
+    vault_url = config["vault_url"].string_value();
+    vault_client_cert = config["vault_client_cert"].string_value();
+    if (vault_client_cert.empty())
+        vault_client_cert = config["cert"].string_value();
+    vault_client_key = config["vault_client_key"].string_value();
+    if (vault_client_key.empty())
+        vault_client_key = config["pkey"].string_value();
+    vault_ca = config["vault_ca"].string_value();
+    vault_secret_api_path = "/v1/secret/";
+    if (config["vault_secret_api_path"].is_string())
+        vault_secret_api_path = config["vault_secret_api_path"].string_value();
+    vault_timeout_ms = config["vault_timeout_ms"].uint64_value();
+    if (!vault_timeout_ms)
+        vault_timeout_ms = 5000;
+    vault_error_timeout_sec = config["vault_error_timeout_sec"].uint64_value();
+    if (!vault_error_timeout_sec)
+        vault_error_timeout_sec = 60;
+    vault_refresh_leeway_sec = config["vault_refresh_leeway_sec"].uint64_value();
+    if (!vault_refresh_leeway_sec)
+        vault_refresh_leeway_sec = 60;
+}
+
+// FIXME: Rework client API by adding open/close and cache inode information in the "FD" (maybe)
+void cluster_client_t::on_change_inode_hook(uint64_t inode, bool removed)
+{
+    std::vector<inode_t> children = { inode };
+    for (size_t i = 0; i < children.size(); i++)
+    {
+        auto it = inode_cache_children.lower_bound(std::make_pair(children[i], (inode_t)0));
+        while (it != inode_cache_children.end() && it->first == children[i])
+        {
+            children.push_back(it->second);
+            it++;
+        }
+    }
+    for (auto & inode: children)
+    {
+        auto it = inode_cache.find(inode);
+        if (it != inode_cache.end())
+        {
+            auto icache = it->second;
+            for (auto & parent: icache->chain)
+            {
+                inode_cache_children.erase(std::make_pair(parent, inode));
+            }
+            inode_cache.erase(it);
+        }
+    }
+}
+
+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())
+    {
+        return icache_it->second;
+    }
+    // Fill inode cache
+    auto ino_it = st_cli->inode_config.find(ino);
+    if (ino_it == st_cli->inode_config.end())
+    {
+        inode_cache[ino] = NULL;
+        return NULL;
+    }
+    auto pool_it = st_cli->pool_config.find(INODE_POOL(ino));
+    if (pool_it == st_cli->pool_config.end())
+    {
+        inode_cache[ino] = NULL;
+        return NULL;
+    }
+    auto & inode_cfg = ino_it->second;
+    auto & pool_cfg = pool_it->second;
+    std::shared_ptr<inode_cache_t> icache = std::make_shared<inode_cache_t>();
+    icache->readonly = inode_cfg.readonly;
+    icache->chain.push_back(ino);
+    std::vector<inode_config_t*> chain_cfg;
+    // FIXME: Allow unencrypted read & write when all chain is encrypted with the same key
+    int enc_key_count = !inode_cfg.enc_key.empty() ? 1 : 0;
+    if (inode_cfg.parent_id)
+    {
+        // Check for loops and cache the chain
+        robin_hood::unordered_flat_set<inode_t> seen;
+        seen.insert(ino);
+        uint64_t parent_id = inode_cfg.parent_id;
+        while (parent_id)
+        {
+            if (seen.find(parent_id) != seen.end())
+            {
+                icache->has_parent_loop = true;
+                break;
+            }
+            seen.insert(parent_id);
+            ino_it = st_cli->inode_config.find(parent_id);
+            if (INODE_POOL(parent_id) == INODE_POOL(ino))
+            {
+                icache->chain.push_back(parent_id);
+                if (ino_it == st_cli->inode_config.end())
+                    chain_cfg.push_back(NULL);
+                else
+                {
+                    chain_cfg.push_back(&ino_it->second);
+                    if (!ino_it->second.enc_key.empty())
+                        enc_key_count++;
+                }
+            }
+            else if (!icache->other_pool_parent_id)
+                icache->other_pool_parent_id = parent_id;
+            if (ino_it == st_cli->inode_config.end())
+                break;
+            parent_id = ino_it->second.parent_id;
+        }
+    }
+    // Check external keys and wait for loading, if required
+    if (enc_key_count)
+    {
+        for (size_t i = 0; i <= chain_cfg.size(); i++)
+        {
+            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[inode_cfg.enc_key];
+                if (ik.key_state == VAULT_KEY_ERROR || vault_url.empty())
+                {
+                    icache->err_code = EPERM;
+                    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(inode_cfg.enc_key);
+                    vault_load_keys();
+                    icache->err_code = EAGAIN;
+                    enc_key_count = 0;
+                }
+                else if (ik.key_state == VAULT_KEY_LOADING)
+                {
+                    icache->err_code = EAGAIN;
+                    enc_key_count = 0;
+                }
+                else
+                {
+                    assert(ik.key_state == VAULT_KEY_LOADED);
+                }
+            }
+        }
+    }
+    // Generate encryption key chain, if applicable
+    if (enc_key_count)
+    {
+        uint8_t *key_data = (uint8_t*)malloc_or_die(
+            AES_256_XTS_KEY_SIZE * enc_key_count +
+            sizeof(uint8_t*) * icache->chain.size() +
+            sizeof(osd_op_enc_t)
+        );
+        uint8_t **keys = (uint8_t**)(key_data + AES_256_XTS_KEY_SIZE * enc_key_count);
+        osd_op_enc_t *enc = (osd_op_enc_t*)((uint8_t*)keys + sizeof(uint8_t*)*icache->chain.size());
+        size_t key_pos = 0;
+        for (size_t i = 0; i <= chain_cfg.size(); i++)
+        {
+            inode_config_t *cfg = !i ? &inode_cfg : chain_cfg[i-1];
+            if (cfg && !cfg->enc_key.empty())
+            {
+                const auto & key = cfg->enc_key.substr(0, strlen(VAULT_KEY_PREFIX)) == VAULT_KEY_PREFIX
+                    ? vault_keys.at(cfg->enc_key).key
+                    : cfg->enc_key;
+                assert(key_pos < AES_256_XTS_KEY_SIZE * enc_key_count);
+                assert(key.size() == 2*AES_256_XTS_KEY_SIZE);
+                keys[i] = key_data + key_pos;
+                fromhexstr(key, AES_256_XTS_KEY_SIZE, key_data + key_pos);
+                key_pos += AES_256_XTS_KEY_SIZE;
+            }
+            else
+                keys[i] = NULL;
+        }
+        enc->key_chain = keys;
+        enc->chain_size = icache->chain.size();
+        enc->read_chain_bitmap_pos = pool_cfg.data_block_size/pool_cfg.bitmap_granularity/8;
+        enc->bitmap_granularity = pool_cfg.bitmap_granularity;
+        icache->key_data = key_data;
+        icache->op_enc = enc;
+    }
+    inode_cache[ino] = icache;
+    for (auto & parent: icache->chain)
+    {
+        if (parent != ino)
+            inode_cache_children.insert(std::make_pair(parent, ino));
+    }
+    return icache;
+}
+
+void cluster_client_t::vault_parse_secret(const std::string & key_id, const std::string & err, json11::Json data)
+{
+    vault_loading = false;
+    auto & k = vault_keys[key_id];
+    if (err != "")
+    {
+        k.key_state = VAULT_KEY_ERROR;
+        fprintf(stderr, "Vault %s%s%s request failed: %s\n", vault_url.c_str(),
+            vault_secret_api_path.c_str(), key_id.c_str()+strlen(VAULT_KEY_PREFIX), err.c_str());
+    }
+    else
+    {
+        auto hexkey = data["data"]["key"].string_value();
+        if (hexkey.empty() || !ishexstr(hexkey) || hexkey.size() != 2*AES_256_XTS_KEY_SIZE)
+        {
+            k.key_state = VAULT_KEY_ERROR;
+            fprintf(stderr, "Vault /v1/secret/%s request failed: 'key' is empty or has invalid format\n", key_id.c_str());
+        }
+        else
+        {
+            k.key_state = VAULT_KEY_LOADED;
+            k.key = hexkey;
+        }
+    }
+    if (vault_key_load_queue.empty())
+    {
+        auto ops = std::move(key_wait_ops);
+        for (cluster_op_t *op: ops)
+            inode_cache.erase(op->inode);
+        for (cluster_op_t *op: ops)
+            execute_internal(op);
+    }
+    else
+        vault_load_keys();
+}

src/client/cluster_client_impl.h

@@ -5,7 +5,6 @@
 
 #include "cluster_client.h"
 
-#define SCRAP_BUFFER_SIZE 4*1024*1024
 #define PART_SENT 1
 #define PART_DONE 2
 #define PART_ERROR 4
@@ -18,6 +17,11 @@
 #define OP_FLUSH_BUFFER 0x02
 #define OP_IMMEDIATE_COMMIT 0x04
 
+#define VAULT_KEY_NOT_LOADED 0
+#define VAULT_KEY_LOADING 1
+#define VAULT_KEY_LOADED 2
+#define VAULT_KEY_ERROR 3
+
 struct cluster_buffer_t
 {
     uint8_t *buf;

src/client/cluster_client_list.cpp

@@ -63,14 +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);
-    if (!pool_id || st_cli.pool_config.find(pool_id) == 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 = st_cli.pool_config.at(pool_id).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();
@@ -110,13 +110,13 @@ bool cluster_client_t::continue_listing(inode_list_t *lst)
 
 bool cluster_client_t::restart_listing(inode_list_t* lst)
 {
-    auto pool_it = st_cli.pool_config.find(lst->pool_id);
+    auto pool_it = st_cli->pool_config.find(lst->pool_id);
     // We want listing to be consistent. To achieve it we should:
     // 1) retry listing of each PG if its state changes
     // 2) abort listing if PG count changes during listing
     // 3) ideally, only talk to the primary OSD - this will be done separately
     // So first we add all PGs without checking their state
-    if (pool_it == st_cli.pool_config.end() ||
+    if (pool_it == st_cli->pool_config.end() ||
         lst->real_pg_count != pool_it->second.real_pg_count)
     {
         for (auto pg: lst->pgs)
@@ -136,7 +136,7 @@ bool cluster_client_t::restart_listing(inode_list_t* lst)
             fprintf(stderr, "PG count in pool %u changed during listing\n", lst->pool_id);
         }
         lst->pgs.clear();
-        if (pool_it == st_cli.pool_config.end())
+        if (pool_it == st_cli->pool_config.end())
         {
             // Unknown pool
             lst->callback(-EINVAL, 0, 0, std::set<object_id>());
@@ -248,7 +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_cfg = st_cli.pool_config.at(pg->lst->pool_id);
+    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() ||
@@ -277,7 +277,7 @@ int cluster_client_t::start_pg_listing(inode_list_pg_t *pg)
         for (auto peer_it = all_peers.begin(); peer_it != all_peers.end(); )
         {
             if (*peer_it != pg_it->second.cur_primary &&
-                st_cli.peer_states[*peer_it].is_null())
+                st_cli->peer_states[*peer_it].is_null())
             {
                 pg->inactive_osds.push_back(*peer_it);
                 all_peers.erase(peer_it++);
@@ -298,11 +298,11 @@ int cluster_client_t::start_pg_listing(inode_list_pg_t *pg)
         if (msgr.osd_peers.find(peer_osd) == msgr.osd_peers.end())
         {
             // Initiate connection
-            if (st_cli.peer_states[peer_osd].is_null())
+            if (st_cli->peer_states[peer_osd].is_null())
             {
                 return LIST_PG_WAIT_ACTIVE;
             }
-            msgr.connect_peer(peer_osd, st_cli.peer_states[peer_osd]);
+            msgr.connect_peer(peer_osd, st_cli->peer_states[peer_osd]);
             conn = false;
         }
     }
@@ -336,7 +336,7 @@ void cluster_client_t::send_list(inode_list_osd_t *cur_list)
     if (!cur_list->pg->inflight_ops)
         cur_list->pg->lst->inflight_pgs++;
     cur_list->pg->inflight_ops++;
-    auto & pool_cfg = st_cli.pool_config[cur_list->pg->lst->pool_id];
+    auto & pool_cfg = st_cli->pool_config[cur_list->pg->lst->pool_id];
     osd_op_t *op = new osd_op_t();
     op->op_type = OSD_OP_OUT;
     // Already checked that it exists above, but anyway

src/client/cluster_client_real.cpp

@@ -0,0 +1,125 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#include "cluster_client.h"
+#include "cluster_client_impl.h"
+#include "etcd_state_client_http.h"
+#include "http_client.h"
+
+cluster_client_t* cluster_client_t::create(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json config)
+{
+    auto st_cli = new etcd_state_client_http_t(tfd);
+    return new cluster_client_t(ringloop, tfd, config, std::unique_ptr<etcd_state_client_t>(st_cli));
+}
+
+bool cluster_client_t::vault_check_token()
+{
+    timespec now;
+    clock_gettime(CLOCK_REALTIME, &now);
+    if (!vault_token_expire.tv_sec || vault_token_expire.tv_sec < now.tv_sec)
+    {
+        vault_loading = true;
+        http_json_post(
+            vault_http_cli, vault_url+"/v1/auth/cert/login", json11::Json::object{}, "",
+            (http_options_t){ .timeout = (int)vault_timeout_ms, .keepalive = true },
+            [this](http_message_t *response)
+            {
+                clock_gettime(CLOCK_REALTIME, &vault_token_expire);
+                vault_loading = false;
+                std::string err;
+                json11::Json data;
+                response->parse_json_response(err, data);
+                if (err != "")
+                {
+                    vault_token_expire.tv_sec += vault_error_timeout_sec;
+                    fprintf(stderr, "Vault request failed: %s\n", err.c_str());
+                }
+                else
+                {
+                    uint64_t ttl = data["auth"]["lease_duration"].uint64_value();
+                    vault_token = data["auth"]["client_token"].string_value();
+                    if (vault_token.empty() || !ttl)
+                    {
+                        vault_token_expire.tv_sec += vault_error_timeout_sec;
+                        fprintf(stderr, "No token or lease_duration in Vault response: %s\n", data.dump().c_str());
+                    }
+                    else
+                    {
+                        if (ttl < vault_refresh_leeway_sec)
+                            vault_token_expire.tv_sec += ttl/2;
+                        else
+                            vault_token_expire.tv_sec += ttl - vault_refresh_leeway_sec;
+                    }
+                }
+                vault_load_keys();
+            }
+        );
+        return false;
+    }
+    if (vault_token.empty())
+    {
+        // Auth error happened, mark all loads as failed
+        for (auto & key_id: vault_key_load_queue)
+        {
+            auto & k = vault_keys[key_id];
+            k.key_state = VAULT_KEY_ERROR;
+        }
+        vault_key_load_queue.clear();
+        auto ops = std::move(key_wait_ops);
+        for (cluster_op_t *op: ops)
+            inode_cache.erase(op->inode);
+        for (cluster_op_t *op: ops)
+            execute_internal(op);
+        return false;
+    }
+    return true;
+}
+
+void cluster_client_t::vault_destroy()
+{
+    if (vault_http_ctx)
+    {
+        http_destroy(vault_http_cli);
+        http_context_destroy(vault_http_ctx);
+        vault_http_cli = NULL;
+        vault_http_ctx = NULL;
+    }
+}
+
+void cluster_client_t::vault_load_keys()
+{
+    if (vault_loading || !vault_key_load_queue.size())
+    {
+        return;
+    }
+    if (!vault_http_ctx)
+    {
+        std::string error;
+        vault_http_ctx = http_context_init(tfd, vault_client_cert, vault_client_key, vault_ca, true, error);
+        if (!vault_http_ctx)
+        {
+            fprintf(stderr, "Failed to initialize HTTP context for Vault: %s\n", error.c_str());
+            exit(1);
+        }
+        vault_http_cli = http_init(vault_http_ctx);
+    }
+    if (!vault_check_token())
+    {
+        return;
+    }
+    std::string key_id = vault_key_load_queue[0];
+    vault_key_load_queue.erase(vault_key_load_queue.begin());
+    vault_loading = true;
+    http_get(
+        vault_http_cli, vault_url+vault_secret_api_path+key_id.substr(strlen(VAULT_KEY_PREFIX)), "X-Vault-Token: "+vault_token+"\r\n",
+        (http_options_t){ .timeout = (int)vault_timeout_ms, .keepalive = true },
+        [this, key_id](http_message_t *response)
+        {
+            vault_loading = false;
+            std::string err;
+            json11::Json data;
+            response->parse_json_response(err, data);
+            vault_parse_secret(key_id, err, data);
+        }
+    );
+}

src/client/cluster_client_wb.cpp

@@ -131,6 +131,7 @@ void writeback_cache_t::copy_write(cluster_op_t *op, int state, uint64_t new_flu
                         writeback_bytes -= op->len;
                     }
                     writeback_queue_size++;
+                    writeback_queue.push_back({ op->inode, new_end });
                 }
                 break;
             }
@@ -165,6 +166,7 @@ void writeback_cache_t::copy_write(cluster_op_t *op, int state, uint64_t new_flu
                 {
                     writeback_queue_size++;
                 }
+                writeback_queue.push_back({ op->inode, new_end });
             }
             auto new_dirty_it = dirty_buffers.emplace_hint(dirty_it, (object_id){
                 .inode = op->inode,

src/client/etcd_state_client.cpp

@@ -1,14 +1,14 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
+#include "malloc_or_die.h"
 #include "osd_ops.h"
+#include "msgr_op.h"
 #include "pg_states.h"
 #include "etcd_state_client.h"
-#ifndef __MOCK__
 #include "addr_util.h"
-#include "http_client.h"
-#endif
 #include "str_util.h"
+#include "json_util.h"
 
 etcd_state_client_t::~etcd_state_client_t()
 {
@@ -17,28 +17,8 @@ etcd_state_client_t::~etcd_state_client_t()
         delete watch;
     }
     watches.clear();
-    etcd_watches_initialised = -1;
-#ifndef __MOCK__
-    stop_ws_keepalive();
-    if (etcd_watch_ws)
-    {
-        http_close(etcd_watch_ws);
-        etcd_watch_ws = NULL;
-    }
-    if (keepalive_client)
-    {
-        http_close(keepalive_client);
-        keepalive_client = NULL;
-    }
-#endif
-    if (load_pgs_timer_id >= 0)
-    {
-        tfd->clear_timer(load_pgs_timer_id);
-        load_pgs_timer_id = -1;
-    }
 }
 
-#ifndef __MOCK__
 etcd_kv_t etcd_state_client_t::parse_etcd_kv(const json11::Json & kv_json)
 {
     etcd_kv_t kv;
@@ -72,146 +52,93 @@ std::vector<std::string> etcd_state_client_t::get_addresses()
     return addrs;
 }
 
-void etcd_state_client_t::etcd_call_oneshot(std::string etcd_address, std::string api, json11::Json payload,
-    int timeout, std::function<void(std::string, json11::Json)> callback)
+std::shared_ptr<user_info_t> etcd_state_client_t::get_user(const std::string & username)
 {
-    std::string etcd_api_path;
-    int pos = etcd_address.find('/');
-    if (pos >= 0)
+    auto user_it = user_info.find(username);
+    if (user_it != user_info.end())
     {
-        etcd_api_path = etcd_address.substr(pos);
-        etcd_address = etcd_address.substr(0, pos);
+        return user_it->second;
     }
-    std::string req = payload.dump();
-    req = "POST "+etcd_api_path+api+" HTTP/1.1\r\n"
-        "Host: "+etcd_address+"\r\n"
-        "Content-Type: application/json\r\n"
-        "Content-Length: "+std::to_string(req.size())+"\r\n"
-        "Connection: close\r\n"
-        "\r\n"+req;
-    auto http_cli = http_init(tfd);
-    auto cb = [http_cli, callback](const http_response_t *response)
-    {
-        std::string err;
-        json11::Json data;
-        response->parse_json_response(err, data);
-        callback(err, data);
-        http_close(http_cli);
-    };
-    http_request(http_cli, etcd_address, req, { .timeout = timeout }, cb);
+    auto inf = std::make_shared<user_info_t>();
+    inf->name = username;
+    return inf;
 }
 
-void etcd_state_client_t::etcd_call(std::string api, json11::Json payload, int timeout,
-    int retries, int interval, std::function<void(std::string, json11::Json)> callback)
+bool etcd_state_client_t::check_image_perm(const std::shared_ptr<user_info_t> & user_info, inode_t inode_num, bool write)
 {
-    if (!etcd_addresses.size() && !etcd_local.size())
+    if (user_info->type == user_type_t::ADMIN)
     {
-        fprintf(stderr, "etcd_address is missing in Vitastor configuration\n");
-        exit(1);
+        return true;
     }
-    pick_next_etcd();
-    std::string etcd_address = selected_etcd_address;
-    std::string etcd_api_path;
-    int pos = etcd_address.find('/');
-    if (pos >= 0)
+    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)
     {
-        etcd_api_path = etcd_address.substr(pos);
-        etcd_address = etcd_address.substr(0, pos);
+        return write ? (cache_it->second.perm == user_perm_t::OWNER) : (cache_it->second.perm != user_perm_t::DENY);
     }
-    std::string req = payload.dump();
-    req = "POST "+etcd_api_path+api+" HTTP/1.1\r\n"
-        "Host: "+etcd_address+"\r\n"
-        "Content-Type: application/json\r\n"
-        "Content-Length: "+std::to_string(req.size())+"\r\n"
-        "Connection: keep-alive\r\n"
-        "Keep-Alive: timeout="+std::to_string(etcd_keepalive_timeout)+"\r\n"
-        "\r\n"+req;
-    retries--;
-    auto cb = [this, api, payload, timeout, retries, interval, callback,
-        cur_addr = selected_etcd_address](const http_response_t *response)
+    auto inode_it = inode_config.find(inode_num);
+    if (inode_it == inode_config.end())
     {
-        std::string err;
-        json11::Json data;
-        response->parse_json_response(err, data);
-        if (err != "")
-        {
-            if (cur_addr == selected_etcd_address)
-                selected_etcd_address = "";
-            if (retries > 0)
-            {
-                if (this->log_level > 0)
-                {
-                    fprintf(
-                        stderr, "Warning: etcd request failed: %s, retrying %d more times\n",
-                        err.c_str(), retries
-                    );
-                }
-                if (interval > 0)
-                {
-                    // FIXME: Prevent destruction of etcd_state_client if timers or requests are active
-                    tfd->set_timer(interval, false, [this, api, payload, timeout, retries, interval, callback](int)
-                    {
-                        etcd_call(api, payload, timeout, retries, interval, callback);
-                    });
-                }
-                else
-                    etcd_call(api, payload, timeout, retries, interval, callback);
-            }
-            else
-                callback(err, data);
-        }
-        else
-            callback(err, data);
-    };
-    if (!keepalive_client)
-    {
-        keepalive_client = http_init(tfd);
+        return false;
     }
-    http_request(keepalive_client, etcd_address, req, { .timeout = timeout, .keepalive = true }, cb);
+    // FIXME Implement cache reset after reworking etcd interaction to not keep everything in memory
+    auto & perm_item = user_info->perm_cache[inode_num];
+    perm_item.mod_revision = user_perm_cache_revision;
+    perm_item.perm = (user_info->name == inode_it->second.owner || inode_it->second.owner_group != "" &&
+        user_info->groups.find(inode_it->second.owner_group) != user_info->groups.end()
+        ? user_perm_t::OWNER : (inode_it->second.reader_group != "" &&
+            user_info->groups.find(inode_it->second.reader_group) != user_info->groups.end()
+            ? user_perm_t::READER : user_perm_t::DENY));
+    return write ? (perm_item.perm == user_perm_t::OWNER) : (perm_item.perm != user_perm_t::DENY);
 }
 
-void etcd_state_client_t::add_etcd_url(std::string addr)
+void etcd_state_client_t::add_etcd_url(std::string etcd_address)
 {
-    if (addr.length() > 0)
+    if (etcd_address.size() > 0)
     {
-        if (strtolower(addr.substr(0, 7)) == "http://")
-            addr = addr.substr(7);
-        else if (strtolower(addr.substr(0, 8)) == "https://")
-        {
-            fprintf(stderr, "HTTPS is unsupported for etcd. Either use plain HTTP or setup a local proxy for etcd interaction\n");
-            exit(1);
-        }
         if (!local_ips.size())
-            local_ips = getifaddr_list(std::vector<addr_mask_t>(), true);
-        std::string check_addr;
-        int pos = addr.find('/');
-        int pos2 = addr.find(':');
-        if (pos2 >= 0)
-            check_addr = addr.substr(0, pos2);
-        else if (pos >= 0)
-            check_addr = addr.substr(0, pos);
-        else
-            check_addr = addr;
-        if (pos == std::string::npos)
-            addr += "/v3";
-        bool local = false;
-        int i;
-        for (i = 0; i < local_ips.size(); i++)
         {
-            if (local_ips[i] == check_addr)
-            {
-                local = true;
-                break;
-            }
+            // Fill local_ips
+            for (auto & ip: getifaddr_list(std::vector<addr_mask_t>(), true))
+                local_ips.insert(ip);
         }
-        auto & to = local ? this->etcd_local : this->etcd_addresses;
+        std::string etcd_api_path;
+        bool ssl = false;
+        if (etcd_address.substr(0, 8) == "https://")
+        {
+            ssl = true;
+            etcd_address = etcd_address.substr(8);
+        }
+        else if (etcd_address.substr(0, 7) == "http://")
+            etcd_address = etcd_address.substr(7);
+        auto pos = etcd_address.find('/');
+        if (pos != std::string::npos)
+        {
+            etcd_api_path = etcd_address.substr(pos);
+            etcd_address = etcd_address.substr(0, pos);
+        }
+        else
+            etcd_api_path = "/v3";
+        pos = etcd_address.find(':');
+        auto check_addr = (pos != std::string::npos ? etcd_address.substr(0, pos) : etcd_address);
+        bool is_local = local_ips.find(check_addr) != local_ips.end();
+        auto & to = (is_local ? etcd_local : etcd_addresses);
+        check_addr = (ssl ? "https://" : "http://") + etcd_address + etcd_api_path;
+        size_t i;
         for (i = 0; i < to.size(); i++)
         {
-            if (to[i] == addr)
+            if (to[i] == check_addr)
                 break;
         }
         if (i >= to.size())
-            to.push_back(addr);
+        {
+            to.push_back(check_addr);
+            // Check if it's a domain name
+            sockaddr_storage ss;
+            bool is_name = !is_local && !string_to_addr(etcd_address, true, 0, &ss);
+            auto & to_addr = (is_local ? etcd_local_addr_urls : (is_name ? etcd_name_urls : etcd_nonlocal_addr_urls));
+            to_addr.push_back((http_url_t){ .ssl = ssl, .addr = etcd_address, .hostname = etcd_address, .path = etcd_api_path });
+        }
     }
 }
 
@@ -219,6 +146,9 @@ void etcd_state_client_t::parse_config(const json11::Json & config)
 {
     this->etcd_local.clear();
     this->etcd_addresses.clear();
+    this->etcd_local_addr_urls.clear();
+    this->etcd_nonlocal_addr_urls.clear();
+    this->etcd_name_urls.clear();
     if (config["etcd_address"].is_string())
     {
         std::string ea = config["etcd_address"].string_value();
@@ -239,6 +169,22 @@ void etcd_state_client_t::parse_config(const json11::Json & config)
             add_etcd_url(ea.string_value());
         }
     }
+    if (this->etcd_client_cert != "")
+    {
+        this->etcd_client_cert = config["etcd_client_cert"].string_value();
+        this->etcd_client_key = config["etcd_client_key"].string_value();
+    }
+    else if (this->osd_num)
+    {
+        this->etcd_client_cert = config["osd_cert"].string_value();
+        this->etcd_client_key = config["osd_pkey"].string_value();
+    }
+    else
+    {
+        this->etcd_client_cert = config["cert"].string_value();
+        this->etcd_client_key = config["pkey"].string_value();
+    }
+    this->etcd_ca = config["etcd_ca"].string_value();
     this->etcd_prefix = config["etcd_prefix"].string_value();
     if (this->etcd_prefix == "")
     {
@@ -256,7 +202,6 @@ void etcd_state_client_t::parse_config(const json11::Json & config)
         if (this->etcd_keepalive_timeout < 30)
             this->etcd_keepalive_timeout = 30;
     }
-    auto old_etcd_ws_keepalive_interval = this->etcd_ws_keepalive_interval;
     this->etcd_ws_keepalive_interval = config["etcd_ws_keepalive_interval"].uint64_value();
     if (this->etcd_ws_keepalive_interval <= 0)
     {
@@ -282,294 +227,9 @@ void etcd_state_client_t::parse_config(const json11::Json & config)
     {
         this->etcd_min_reload_interval = 50;
     }
-    if (this->etcd_ws_keepalive_interval != old_etcd_ws_keepalive_interval && ws_keepalive_timer >= 0)
-    {
-#ifndef __MOCK__
-        stop_ws_keepalive();
-        start_ws_keepalive();
-#endif
-    }
 }
 
-void etcd_state_client_t::pick_next_etcd()
-{
-    if (selected_etcd_address != "")
-        return;
-    if (addresses_to_try.size() == 0)
-    {
-        // Prefer local etcd, if any
-        for (int i = 0; i < etcd_local.size(); i++)
-            addresses_to_try.push_back(etcd_local[i]);
-        std::vector<int> ns;
-        for (int i = 0; i < etcd_addresses.size(); i++)
-            ns.push_back(i);
-        if (!rand_initialized)
-        {
-            timespec tv;
-            clock_gettime(CLOCK_REALTIME, &tv);
-            srand48(tv.tv_sec*1000000000 + tv.tv_nsec);
-            rand_initialized = true;
-        }
-        while (ns.size())
-        {
-            int i = lrand48() % ns.size();
-            addresses_to_try.push_back(etcd_addresses[ns[i]]);
-            ns.erase(ns.begin()+i, ns.begin()+i+1);
-        }
-    }
-    selected_etcd_address = addresses_to_try[0];
-    addresses_to_try.erase(addresses_to_try.begin(), addresses_to_try.begin()+1);
-}
-
-void etcd_state_client_t::start_etcd_watcher()
-{
-    if (!etcd_addresses.size() && !etcd_local.size())
-    {
-        fprintf(stderr, "etcd_address is missing in Vitastor configuration\n");
-        exit(1);
-    }
-    pick_next_etcd();
-    std::string etcd_address = selected_etcd_address;
-    std::string etcd_api_path;
-    int pos = etcd_address.find('/');
-    if (pos >= 0)
-    {
-        etcd_api_path = etcd_address.substr(pos);
-        etcd_address = etcd_address.substr(0, pos);
-    }
-    etcd_watches_initialised = 0;
-    ws_alive = 1;
-    if (etcd_watch_ws)
-    {
-        http_close(etcd_watch_ws);
-        etcd_watch_ws = NULL;
-    }
-    if (this->log_level > 1)
-    {
-        fprintf(stderr, "Trying to connect to etcd websocket at %s, watch from revision %ju/%ju/%ju\n", etcd_address.c_str(),
-            etcd_watch_revision_config, etcd_watch_revision_osd, etcd_watch_revision_pg);
-    }
-    etcd_watch_ws = open_websocket(tfd, etcd_address, etcd_api_path+"/watch", etcd_slow_timeout,
-        [this, cur_addr = selected_etcd_address](const http_response_t *msg)
-    {
-        if (msg->body.length())
-        {
-            ws_alive = 1;
-            std::string json_err;
-            json11::Json data = json11::Json::parse(msg->body, json_err);
-            if (json_err != "")
-            {
-                fprintf(stderr, "Bad JSON in etcd event: %s, ignoring event\n", json_err.c_str());
-            }
-            else
-            {
-                uint64_t watch_id = data["result"]["watch_id"].uint64_value();
-                if (data["result"]["created"].bool_value())
-                {
-                    if (watch_id == ETCD_CONFIG_WATCH_ID ||
-                        watch_id == ETCD_PG_STATE_WATCH_ID ||
-                        watch_id == ETCD_OSD_STATE_WATCH_ID)
-                    {
-                        etcd_watches_initialised++;
-                    }
-                    if (etcd_watches_initialised == ETCD_TOTAL_WATCHES && this->log_level > 0)
-                    {
-                        fprintf(stderr, "Successfully subscribed to etcd at %s, revision %ju/%ju/%ju\n", cur_addr.c_str(),
-                            etcd_watch_revision_config, etcd_watch_revision_osd, etcd_watch_revision_pg);
-                    }
-                }
-                if (data["result"]["canceled"].bool_value())
-                {
-                    // etcd watch canceled, maybe because the revision was compacted
-                    if (data["result"]["compact_revision"].uint64_value())
-                    {
-                        // we may miss events if we proceed
-                        // so we should restart from the beginning if we can
-                        if (on_reload_hook != NULL)
-                        {
-                            // check to not trigger on_reload_hook multiple times
-                            if (etcd_watch_ws != NULL)
-                            {
-                                fprintf(stderr, "Revisions before %ju were compacted by etcd, reloading state\n",
-                                    data["result"]["compact_revision"].uint64_value());
-                                http_close(etcd_watch_ws);
-                                etcd_watch_ws = NULL;
-                                etcd_watch_revision_config = etcd_watch_revision_osd = etcd_watch_revision_pg = 0;
-                                on_reload_hook();
-                            }
-                            return;
-                        }
-                        else
-                        {
-                            fprintf(stderr, "Revisions before %ju were compacted by etcd, exiting\n",
-                                data["result"]["compact_revision"].uint64_value());
-                            exit(1);
-                        }
-                    }
-                    else
-                    {
-                        fprintf(stderr, "Watch canceled by etcd, reason: %s, exiting\n", data["result"]["cancel_reason"].string_value().c_str());
-                        exit(1);
-                    }
-                }
-                // Save revision only if it's present in the message - because sometimes etcd sends something without a header, like:
-                // {"error": {"grpc_code": 14, "http_code": 503, "http_status": "Service Unavailable", "message": "error reading from server: EOF"}}
-                // Also don't save revision from the initial created: true messages because they always contain the latest revision
-                if (etcd_watches_initialised == ETCD_TOTAL_WATCHES &&
-                    !data["result"]["header"]["revision"].is_null() &&
-                    !data["result"]["created"].bool_value())
-                {
-                    // Restart watchers from the same revision number as in the last received message,
-                    // not from the next one to protect against revision being split into multiple messages,
-                    // even though etcd guarantees not to do that **within a single watcher** without fragment=true:
-                    // https://etcd.io/docs/v3.5/learning/api_guarantees/#watch-apis
-                    // Revision contents are ALWAYS split into separate messages for different watchers though!
-                    // So generally we have to resume each watcher from its own revision...
-                    // Progress messages may have watch_id=-1 if sent on behalf of multiple watchers though.
-                    // And antietcd has an advanced semantic which merges the same revision for all watchers
-                    // into one message and just omits watch_id.
-                    // So we also have to handle the case where watch_id is -1 or not present (0).
-                    auto watch_rev = data["result"]["header"]["revision"].uint64_value();
-                    if (!watch_id || watch_id == UINT64_MAX)
-                        etcd_watch_revision_config = etcd_watch_revision_osd = etcd_watch_revision_pg = watch_rev;
-                    else if (watch_id == ETCD_CONFIG_WATCH_ID)
-                        etcd_watch_revision_config = watch_rev;
-                    else if (watch_id == ETCD_PG_STATE_WATCH_ID)
-                        etcd_watch_revision_pg = watch_rev;
-                    else if (watch_id == ETCD_OSD_STATE_WATCH_ID)
-                        etcd_watch_revision_osd = watch_rev;
-                    addresses_to_try.clear();
-                }
-                // First gather all changes into a hash to remove multiple overwrites
-                std::map<std::string, etcd_kv_t> changes;
-                for (auto & ev: data["result"]["events"].array_items())
-                {
-                    auto kv = parse_etcd_kv(ev["kv"]);
-                    if (kv.key != "")
-                    {
-                        changes[kv.key] = kv;
-                    }
-                }
-                for (auto & kv: changes)
-                {
-                    if (this->log_level > 3)
-                    {
-                        fprintf(stderr, "Incoming event: %s -> %s\n", kv.first.c_str(), kv.second.value.dump().c_str());
-                    }
-                    parse_state(kv.second);
-                }
-                // React to changes
-                if (on_change_hook != NULL)
-                {
-                    on_change_hook(changes);
-                }
-            }
-        }
-        if (msg->eof)
-        {
-            fprintf(stderr, "Disconnected from etcd %s\n", cur_addr.c_str());
-            if (cur_addr == selected_etcd_address)
-                selected_etcd_address = "";
-            if (etcd_watch_ws)
-            {
-                http_close(etcd_watch_ws);
-                etcd_watch_ws = NULL;
-            }
-            if (etcd_watches_initialised == 0)
-            {
-                // Connection not established, retry in <etcd_quick_timeout>
-                tfd->set_timer(etcd_quick_timeout, false, [this](int)
-                {
-                    start_etcd_watcher();
-                });
-            }
-            else if (etcd_watches_initialised > 0)
-            {
-                // Connection was live, retry immediately
-                etcd_watches_initialised = 0;
-                start_etcd_watcher();
-            }
-        }
-    });
-    http_post_message(etcd_watch_ws, WS_TEXT, json11::Json(json11::Json::object {
-        { "create_request", json11::Json::object {
-            { "key", base64_encode(etcd_prefix+"/config/") },
-            { "range_end", base64_encode(etcd_prefix+"/config0") },
-            { "start_revision", etcd_watch_revision_config },
-            { "watch_id", ETCD_CONFIG_WATCH_ID },
-            { "progress_notify", true },
-        } }
-    }).dump());
-    http_post_message(etcd_watch_ws, WS_TEXT, json11::Json(json11::Json::object {
-        { "create_request", json11::Json::object {
-            { "key", base64_encode(etcd_prefix+"/osd/state/") },
-            { "range_end", base64_encode(etcd_prefix+"/osd/state0") },
-            { "start_revision", etcd_watch_revision_osd },
-            { "watch_id", ETCD_OSD_STATE_WATCH_ID },
-            { "progress_notify", true },
-        } }
-    }).dump());
-    http_post_message(etcd_watch_ws, WS_TEXT, json11::Json(json11::Json::object {
-        { "create_request", json11::Json::object {
-            { "key", base64_encode(etcd_prefix+"/pg/") },
-            { "range_end", base64_encode(etcd_prefix+"/pg0") },
-            { "start_revision", etcd_watch_revision_pg },
-            { "watch_id", ETCD_PG_STATE_WATCH_ID },
-            { "progress_notify", true },
-        } }
-    }).dump());
-    // FIXME: Do not watch /pg/history/ at all in client code (not in OSD)
-    if (on_start_watcher_hook)
-    {
-        on_start_watcher_hook(etcd_watch_ws);
-    }
-    start_ws_keepalive();
-}
-
-void etcd_state_client_t::stop_ws_keepalive()
-{
-    if (ws_keepalive_timer >= 0)
-    {
-        tfd->clear_timer(ws_keepalive_timer);
-        ws_keepalive_timer = -1;
-    }
-}
-
-void etcd_state_client_t::start_ws_keepalive()
-{
-    if (ws_keepalive_timer < 0)
-    {
-        ws_keepalive_timer = tfd->set_timer(etcd_ws_keepalive_interval*1000, true, [this](int)
-        {
-            if (!etcd_watch_ws || etcd_watches_initialised < ETCD_TOTAL_WATCHES)
-            {
-                // Do nothing
-            }
-            else if (!ws_alive)
-            {
-                if (this->log_level > 0)
-                {
-                    fprintf(stderr, "Websocket ping failed, disconnecting from etcd %s\n", selected_etcd_address.c_str());
-                }
-                if (etcd_watch_ws)
-                {
-                    http_close(etcd_watch_ws);
-                    etcd_watch_ws = NULL;
-                }
-                start_etcd_watcher();
-            }
-            else
-            {
-                ws_alive = 0;
-                http_post_message(etcd_watch_ws, WS_TEXT, json11::Json(json11::Json::object {
-                    { "progress_request", json11::Json::object { } }
-                }).dump());
-            }
-        });
-    }
-}
-
-void etcd_state_client_t::load_global_config()
+void etcd_state_client_t::load_global_config(std::function<void(const std::string & error)> cb)
 {
     json11::Json::object req = { { "success", json11::Json::array {
         json11::Json::object {
@@ -583,22 +243,12 @@ void etcd_state_client_t::load_global_config()
             } }
         },
     } } };
-    etcd_txn(req, etcd_quick_timeout, max_etcd_attempts, 0, [this](std::string err, json11::Json data)
+    etcd_txn(req, etcd_quick_timeout, max_etcd_attempts, 0, [this, cb](std::string err, json11::Json data)
     {
         if (err != "")
         {
             fprintf(stderr, "Error reading configuration from etcd: %s\n", err.c_str());
-            if (infinite_start)
-            {
-                tfd->set_timer(etcd_slow_timeout, false, [this](int timer_id)
-                {
-                    load_global_config();
-                });
-            }
-            else
-            {
-                exit(1);
-            }
+            cb(err);
             return;
         }
         json11::Json config_kv = data["responses"][0]["response_range"]["kvs"][0];
@@ -629,37 +279,17 @@ void etcd_state_client_t::load_global_config()
             parse_state(kv);
         }
         on_load_config_hook(global_config);
+        cb("");
     });
 }
 
-void etcd_state_client_t::load_pgs()
+void etcd_state_client_t::load_pgs(std::function<void(const std::string &)> cb)
 {
-    timespec tv;
-    clock_gettime(CLOCK_REALTIME, &tv);
-    uint64_t ms_passed = (tv.tv_sec-etcd_last_reload.tv_sec)*1000 + (tv.tv_nsec-etcd_last_reload.tv_nsec)/1000000;
-    if (ms_passed < etcd_min_reload_interval)
-    {
-        if (load_pgs_timer_id < 0)
-        {
-            load_pgs_timer_id = tfd->set_timer(etcd_min_reload_interval+50-ms_passed, false, [this](int) { load_pgs(); });
-        }
-        return;
-    }
-    etcd_last_reload = tv;
-    if (load_pgs_timer_id >= 0)
-    {
-        tfd->clear_timer(load_pgs_timer_id);
-        load_pgs_timer_id = -1;
-    }
     json11::Json::array txn = {
         json11::Json::object {
             { "request_range", json11::Json::object {
-                { "key", base64_encode(etcd_prefix+"/config/pools") },
-            } }
-        },
-        json11::Json::object {
-            { "request_range", json11::Json::object {
-                { "key", base64_encode(etcd_prefix+"/config/pgs") },
+                { "key", base64_encode(etcd_prefix+"/config/") },
+                { "range_end", base64_encode(etcd_prefix+"/config0") },
             } }
         },
         json11::Json::object {
@@ -667,12 +297,6 @@ void etcd_state_client_t::load_pgs()
                 { "key", base64_encode(etcd_prefix+"/pg/config") },
             } }
         },
-        json11::Json::object {
-            { "request_range", json11::Json::object {
-                { "key", base64_encode(etcd_prefix+"/config/inode/") },
-                { "range_end", base64_encode(etcd_prefix+"/config/inode0") },
-            } }
-        },
         json11::Json::object {
             { "request_range", json11::Json::object {
                 { "key", base64_encode(etcd_prefix+"/pg/history/") },
@@ -698,16 +322,13 @@ void etcd_state_client_t::load_pgs()
     {
         req["compare"] = checks;
     }
-    etcd_txn_slow(req, [this](std::string err, json11::Json data)
+    etcd_txn_slow(req, [this, cb](std::string err, json11::Json data)
     {
         if (err != "")
         {
             // Retry indefinitely
             fprintf(stderr, "Error loading PGs from etcd: %s\n", err.c_str());
-            tfd->set_timer(etcd_slow_timeout, false, [this](int timer_id)
-            {
-                load_pgs();
-            });
+            cb(err);
             return;
         }
         if (!data["succeeded"].bool_value())
@@ -735,24 +356,9 @@ void etcd_state_client_t::load_pgs()
         }
         clean_nonexistent_pgs();
         on_load_pgs_hook(true);
-        start_etcd_watcher();
+        cb("");
     });
 }
-#else
-void etcd_state_client_t::parse_config(const json11::Json & config)
-{
-}
-
-void etcd_state_client_t::load_global_config()
-{
-    json11::Json::object global_config;
-    on_load_config_hook(global_config);
-}
-
-void etcd_state_client_t::load_pgs()
-{
-}
-#endif
 
 void etcd_state_client_t::reset_pg_exists()
 {
@@ -865,7 +471,8 @@ void etcd_state_client_t::parse_state(const etcd_kv_t & kv)
             if (pc.pg_size < 1 ||
                 pool_item.second["pg_size"].uint64_value() < 3 &&
                 (pc.scheme == POOL_SCHEME_XOR || pc.scheme == POOL_SCHEME_EC) ||
-                pool_item.second["pg_size"].uint64_value() > 256)
+                // limit is 64 because osd_peering_pg.cpp uses a 64-bit mask for has_roles
+                pool_item.second["pg_size"].uint64_value() > 64)
             {
                 fprintf(stderr, "Pool %u has invalid pg_size, skipping pool\n", pool_id);
                 continue;
@@ -942,6 +549,8 @@ void etcd_state_client_t::parse_state(const etcd_kv_t & kv)
                 pc.used_for_app = "fs:"+pc.used_for_app;
             else
                 pc.used_for_app = pool_item.second["used_for_app"].as_string();
+            // Create group permission
+            pc.creator_group = pool_item.second["creator_group"].string_value();
             // Local Read Configuration
             std::string local_reads = pool_item.second["local_reads"].string_value();
             if (local_reads == "nearest")
@@ -1205,8 +814,14 @@ void etcd_state_client_t::parse_state(const etcd_kv_t & kv)
     else if (key.substr(0, etcd_prefix.length()+11) == etcd_prefix+"/osd/state/")
     {
         // <etcd_prefix>/osd/state/%d
-        osd_num_t peer_osd = std::stoull(key.substr(etcd_prefix.length()+11));
-        if (peer_osd > 0)
+        osd_num_t peer_osd = 0;
+        char null_byte = 0;
+        int scanned = sscanf(key.c_str() + etcd_prefix.length()+11, "%ju%c", &peer_osd, &null_byte);
+        if (scanned != 1 || !peer_osd)
+        {
+            fprintf(stderr, "Bad etcd key %s, ignoring\n", key.c_str());
+        }
+        else
         {
             if (value.is_object() && value["state"] == "up")
             {
@@ -1259,37 +874,15 @@ void etcd_state_client_t::parse_state(const etcd_kv_t & kv)
                 {
                     on_inode_change_hook(inode_num, true);
                 }
+                if (this->inode_config.find(inode_num) != this->inode_config.end())
+                {
+                    user_perm_cache_revision = kv.mod_revision;
+                }
                 this->inode_config.erase(inode_num);
             }
             else
             {
-                inode_t parent_inode_num = value["parent_id"].uint64_value();
-                if (parent_inode_num && !(parent_inode_num >> (64-POOL_ID_BITS)))
-                {
-                    uint64_t parent_pool_id = value["parent_pool"].uint64_value();
-                    if (!parent_pool_id)
-                        parent_inode_num |= pool_id << (64-POOL_ID_BITS);
-                    else if (parent_pool_id >= POOL_ID_MAX)
-                    {
-                        fprintf(
-                            stderr, "Inode %ju/%ju parent_pool value is invalid, ignoring parent setting\n",
-                            inode_num >> (64-POOL_ID_BITS), inode_num & (((uint64_t)1 << (64-POOL_ID_BITS)) - 1)
-                        );
-                        parent_inode_num = 0;
-                    }
-                    else
-                        parent_inode_num |= parent_pool_id << (64-POOL_ID_BITS);
-                }
-                insert_inode_config((inode_config_t){
-                    .num = inode_num,
-                    .name = value["name"].string_value(),
-                    .size = value["size"].uint64_value(),
-                    .parent_id = parent_inode_num,
-                    .readonly = value["readonly"].bool_value(),
-                    .deleted = value["deleted"].bool_value(),
-                    .meta = value["meta"],
-                    .mod_revision = kv.mod_revision,
-                });
+                insert_inode_config(deserialize_inode_cfg(inode_num, kv.value, kv.mod_revision));
             }
         }
     }
@@ -1300,6 +893,38 @@ void etcd_state_client_t::parse_state(const etcd_kv_t & kv)
         if (on_change_node_placement_hook)
             on_change_node_placement_hook();
     }
+    else if (key.substr(0, etcd_prefix.length()+13) == etcd_prefix+"/config/user/")
+    {
+        // <etcd_prefix>/config/user/<username>
+        auto name = key.substr(etcd_prefix.length()+13);
+        auto & inf = user_info[name];
+        if (!value.is_object())
+        {
+            if (inf)
+            {
+                inf->type = user_type_t::CLIENT;
+                inf->groups.clear();
+                inf->perm_cache.clear();
+            }
+            user_info.erase(name);
+        }
+        else
+        {
+            if (!inf)
+            {
+                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())
+            {
+                if (group.string_value() != "")
+                    inf->groups.insert(group.string_value());
+            }
+            inf->perm_cache.clear();
+        }
+    }
 }
 
 uint32_t etcd_state_client_t::parse_immediate_commit(const std::string & immediate_commit_str, uint32_t default_value)
@@ -1322,7 +947,12 @@ uint32_t etcd_state_client_t::parse_scheme(const std::string & scheme)
 
 void etcd_state_client_t::insert_inode_config(const inode_config_t & cfg)
 {
-    this->inode_config[cfg.num] = cfg;
+    auto & cfg_ref = this->inode_config[cfg.num];
+    if (cfg_ref.mod_revision != cfg.mod_revision)
+    {
+        user_perm_cache_revision = cfg.mod_revision;
+    }
+    cfg_ref = cfg;
     if (cfg.name != "")
     {
         this->inode_by_name[cfg.name] = cfg.num;
@@ -1378,6 +1008,10 @@ json11::Json::object etcd_state_client_t::serialize_inode_cfg(inode_config_t *cf
             new_cfg["parent_pool"] = (uint64_t)INODE_POOL(cfg->parent_id);
         new_cfg["parent_id"] = (uint64_t)INODE_NO_POOL(cfg->parent_id);
     }
+    if (!cfg->enc_key.empty())
+    {
+        new_cfg["enc_key"] = cfg->enc_key;
+    }
     if (cfg->readonly)
     {
         new_cfg["readonly"] = true;
@@ -1386,6 +1020,18 @@ json11::Json::object etcd_state_client_t::serialize_inode_cfg(inode_config_t *cf
     {
         new_cfg["deleted"] = true;
     }
+    if (!cfg->owner.empty())
+    {
+        new_cfg["owner"] = cfg->owner;
+    }
+    if (!cfg->owner_group.empty())
+    {
+        new_cfg["owner_group"] = cfg->owner_group;
+    }
+    if (!cfg->reader_group.empty())
+    {
+        new_cfg["reader_group"] = cfg->reader_group;
+    }
     if (cfg->meta.is_object())
     {
         new_cfg["meta"] = cfg->meta;
@@ -1393,6 +1039,53 @@ json11::Json::object etcd_state_client_t::serialize_inode_cfg(inode_config_t *cf
     return new_cfg;
 }
 
+inode_config_t etcd_state_client_t::deserialize_inode_cfg(uint64_t inode_num, json11::Json value, uint64_t mod_revision)
+{
+    inode_t parent_inode_num = value["parent_id"].uint64_value();
+    if (parent_inode_num && !INODE_POOL(parent_inode_num))
+    {
+        uint64_t parent_pool_id = value["parent_pool"].uint64_value();
+        if (!parent_pool_id)
+            parent_inode_num = INODE_WITH_POOL(INODE_POOL(inode_num), parent_inode_num);
+        else if (parent_pool_id >= POOL_ID_MAX)
+        {
+            fprintf(
+                stderr, "Inode %u/%ju parent_pool value is invalid, ignoring parent setting\n",
+                INODE_POOL(inode_num), INODE_NO_POOL(inode_num)
+            );
+            parent_inode_num = 0;
+        }
+        else
+            parent_inode_num |= parent_pool_id << (64-POOL_ID_BITS);
+    }
+    std::string enc_key;
+    if (!value["enc_key"].is_null())
+    {
+        enc_key = value["enc_key"].string_value();
+        if (enc_key.substr(0, strlen(VAULT_KEY_PREFIX)) != VAULT_KEY_PREFIX &&
+            (enc_key.size() != 2*AES_256_XTS_KEY_SIZE || !ishexstr(enc_key)))
+        {
+            enc_key = "";
+            fprintf(stderr, "Inode %u/%ju has invalid enc_key, should be %u bit hex string or Vault key reference\n",
+                INODE_POOL(inode_num), INODE_NO_POOL(inode_num), AES_256_XTS_KEY_SIZE);
+        }
+    }
+    return (inode_config_t){
+        .num = inode_num,
+        .name = value["name"].string_value(),
+        .size = value["size"].uint64_value(),
+        .parent_id = parent_inode_num,
+        .readonly = value["readonly"].bool_value(),
+        .deleted = value["deleted"].bool_value(),
+        .enc_key = std::move(enc_key),
+        .owner = value["owner"].string_value(),
+        .owner_group = value["owner_group"].string_value(),
+        .reader_group = value["reader_group"].string_value(),
+        .meta = value["meta"],
+        .mod_revision = mod_revision,
+    };
+}
+
 int etcd_state_client_t::address_count()
 {
     return etcd_addresses.size() + etcd_local.size();

src/client/etcd_state_client.h

@@ -4,10 +4,12 @@
 #pragma once
 
 #include <set>
+#include <memory>
 
 #include "json11/json11.hpp"
 #include "object_id.h"
 #include "timerfd_manager.h"
+#include "../util/robin_hood.h"
 
 #define ETCD_CONFIG_WATCH_ID 1
 #define ETCD_OSD_STATE_WATCH_ID 2
@@ -19,6 +21,8 @@
 #define MAX_DATA_BLOCK_SIZE 128*1024*1024
 #define DEFAULT_BITMAP_GRANULARITY 4096
 
+#define VAULT_KEY_PREFIX "vault:"
+
 #ifndef IMMEDIATE_NONE
 #define IMMEDIATE_NONE 0
 #define IMMEDIATE_SMALL 1
@@ -66,6 +70,7 @@ struct pool_config_t
     std::map<pg_num_t, pg_config_t> pg_config;
     uint64_t scrub_interval = 0;
     std::string used_for_app;
+    std::string creator_group;
     int backfillfull = 0;
     int local_reads = 0;
 
@@ -83,6 +88,9 @@ struct inode_config_t
     inode_t parent_id = 0;
     bool readonly = false;
     bool deleted = false;
+    std::string enc_key;
+    // Permissions
+    std::string owner, owner_group, reader_group;
     // Arbitrary metadata
     json11::Json meta;
     // Change revision of the metadata in etcd
@@ -95,23 +103,56 @@ struct inode_watch_t
     inode_config_t cfg = {};
 };
 
+struct http_url_t
+{
+    bool ssl;
+    std::string addr;
+    std::string hostname;
+    std::string path;
+};
+
+enum class user_type_t
+{
+    CLIENT = 0,
+    ADMIN = 1,
+};
+
+struct user_perm_t
+{
+    enum class perm_type_t: uint8_t;
+    constexpr static perm_type_t DENY = (perm_type_t)0;
+    constexpr static perm_type_t READER = (perm_type_t)1;
+    constexpr static perm_type_t OWNER = (perm_type_t)2;
+    uint64_t mod_revision = 0;
+    perm_type_t perm = DENY;
+};
+
+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;
+};
+
 struct http_co_t;
 
 struct __attribute__((visibility("default"))) etcd_state_client_t
 {
 protected:
-    std::vector<std::string> local_ips;
-    std::vector<std::string> etcd_addresses;
+    std::set<std::string> local_ips;
     std::vector<std::string> etcd_local;
-    std::string selected_etcd_address;
-    std::vector<std::string> addresses_to_try;
+    std::vector<std::string> etcd_addresses;
+    std::vector<http_url_t> etcd_local_addr_urls;
+    std::vector<http_url_t> etcd_nonlocal_addr_urls;
+    std::vector<http_url_t> etcd_name_urls;
     std::vector<inode_watch_t*> watches;
+    std::set<osd_num_t> seen_peers;
     bool new_pg_config = false;
-    int ws_keepalive_timer = -1;
-    int ws_alive = 0;
-    bool rand_initialized = false;
+
     void add_etcd_url(std::string);
-    void pick_next_etcd();
+    void reset_pg_exists();
+    void clean_nonexistent_pgs();
 public:
     int etcd_keepalive_timeout = 30;
     int etcd_ws_keepalive_interval = 5;
@@ -124,22 +165,23 @@ public:
     uint32_t global_bitmap_granularity = DEFAULT_BITMAP_GRANULARITY;
     uint32_t global_immediate_commit = IMMEDIATE_NONE;
 
+    uint64_t osd_num = 0;
     std::string etcd_prefix;
+    std::string etcd_client_cert;
+    std::string etcd_client_key;
+    std::string etcd_ca;
     int log_level = 0;
-    timerfd_manager_t *tfd = NULL;
 
-    http_co_t *etcd_watch_ws = NULL, *keepalive_client = NULL;
-    int etcd_watches_initialised = 0;
     uint64_t etcd_watch_revision_config = 0;
     uint64_t etcd_watch_revision_osd = 0;
     uint64_t etcd_watch_revision_pg = 0;
-    timespec etcd_last_reload = {};
-    int load_pgs_timer_id = -1;
+
     std::map<pool_id_t, pool_config_t> pool_config;
     std::map<osd_num_t, json11::Json> peer_states;
-    std::set<osd_num_t> seen_peers;
     std::map<inode_t, inode_config_t> inode_config;
     std::map<std::string, inode_t> inode_by_name;
+    robin_hood::unordered_flat_map<std::string, std::shared_ptr<user_info_t>> user_info;
+    uint64_t user_perm_cache_revision = 0;
     json11::Json node_placement;
 
     std::function<void(std::map<std::string, etcd_kv_t> &)> on_change_hook;
@@ -158,26 +200,28 @@ public:
     std::function<void(http_co_t *)> on_start_watcher_hook;
 
     json11::Json::object serialize_inode_cfg(inode_config_t *cfg);
+    inode_config_t deserialize_inode_cfg(uint64_t inode_num, json11::Json value, uint64_t mod_revision);
     etcd_kv_t parse_etcd_kv(const json11::Json & kv_json);
     std::vector<std::string> get_addresses();
-    void etcd_call_oneshot(std::string etcd_address, std::string api, json11::Json payload, int timeout, std::function<void(std::string, json11::Json)> callback);
-    void etcd_call(std::string api, json11::Json payload, int timeout, int retries, int interval, std::function<void(std::string, json11::Json)> callback);
+    std::shared_ptr<user_info_t> get_user(const std::string & username);
+    bool check_image_perm(const std::shared_ptr<user_info_t> & user_info, inode_t inode_num, bool write);
+    virtual void etcd_call_oneshot(const std::string & etcd_address, const std::string & api, json11::Json payload, int timeout, std::function<void(std::string, json11::Json)> callback) = 0;
+    virtual void etcd_call(const std::string & api, json11::Json payload, int timeout, int retries, int interval, std::function<void(std::string, json11::Json)> callback) = 0;
     void etcd_txn(json11::Json txn, int timeout, int retries, int interval, std::function<void(std::string, json11::Json)> callback);
     void etcd_txn_slow(json11::Json txn, std::function<void(std::string, json11::Json)> callback);
-    void start_etcd_watcher();
-    void stop_ws_keepalive();
-    void start_ws_keepalive();
-    void load_global_config();
-    void load_pgs();
-    void reset_pg_exists();
-    void clean_nonexistent_pgs();
+    virtual void etcd_add_watch(json11::Json watch) = 0;
+    virtual std::string get_username() = 0;
+    void load_global_config(std::function<void(const std::string &)> cb);
+    virtual void load_global_config() = 0;
+    void load_pgs(std::function<void(const std::string &)> cb);
+    virtual void load_pgs() = 0;
     void parse_state(const etcd_kv_t & kv);
-    void parse_config(const json11::Json & config);
+    virtual void parse_config(const json11::Json & config);
     void insert_inode_config(const inode_config_t & cfg);
     inode_watch_t* watch_inode(std::string name);
     void close_watch(inode_watch_t* watch);
     int address_count();
-    ~etcd_state_client_t();
+    virtual ~etcd_state_client_t();
 
     static uint32_t parse_immediate_commit(const std::string & immediate_commit_str, uint32_t default_value);
     static uint32_t parse_scheme(const std::string & scheme_str);

src/client/etcd_state_client_http.cpp

@@ -0,0 +1,545 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#include <assert.h>
+#include "etcd_state_client_http.h"
+#include "addr_util.h"
+#include "http_client.h"
+#include "str_util.h"
+
+etcd_state_client_http_t::etcd_state_client_http_t(timerfd_manager_t *tfd)
+{
+    this->tfd = tfd;
+}
+
+etcd_state_client_http_t::~etcd_state_client_http_t()
+{
+    stop_ws_keepalive();
+    if (etcd_watch_ws)
+    {
+        http_destroy(etcd_watch_ws);
+        etcd_watch_ws = NULL;
+    }
+    if (keepalive_client)
+    {
+        http_destroy(keepalive_client);
+        keepalive_client = NULL;
+    }
+    if (load_pgs_timer_id >= 0)
+    {
+        tfd->clear_timer(load_pgs_timer_id);
+        load_pgs_timer_id = -1;
+    }
+    if (http_ctx)
+    {
+        http_context_destroy(http_ctx);
+        http_ctx = NULL;
+    }
+    etcd_watches_initialised = -1;
+}
+
+void etcd_state_client_http_t::etcd_add_watch(json11::Json watch)
+{
+    if (etcd_watch_ws)
+    {
+        http_post_message(etcd_watch_ws, WS_TEXT, watch.dump());
+    }
+}
+
+std::string etcd_state_client_http_t::get_username()
+{
+    return http_context_get_ssl_cn(get_http_ctx());
+}
+
+http_context_t *etcd_state_client_http_t::get_http_ctx()
+{
+    if (!http_ctx)
+    {
+        std::string error;
+        http_ctx = http_context_init(tfd, etcd_client_cert, etcd_client_key, etcd_ca, true, error);
+        if (!http_ctx)
+        {
+            fprintf(stderr, "Failed to initialize HTTP context: %s\n", error.c_str());
+            exit(1);
+        }
+    }
+    return http_ctx;
+}
+
+void etcd_state_client_http_t::etcd_call_oneshot(const std::string & etcd_url, const std::string & api, json11::Json payload,
+    int timeout, std::function<void(std::string, json11::Json)> callback)
+{
+    auto http_cli = http_init(get_http_ctx());
+    http_json_post(http_cli, etcd_url+api, payload, "", { .timeout = timeout }, [http_cli, callback](http_message_t *response)
+    {
+        std::string err;
+        json11::Json data;
+        response->parse_json_response(err, data);
+        callback(err, data);
+        http_destroy(http_cli);
+    });
+}
+
+void etcd_state_client_http_t::etcd_call(const std::string & api, json11::Json payload, int timeout,
+    int retries, int interval, std::function<void(std::string, json11::Json)> callback)
+{
+    pick_next_etcd([=]()
+    {
+        etcd_call_selected(api, payload, timeout, retries, interval, callback);
+    });
+}
+
+void etcd_state_client_http_t::etcd_call_selected(const std::string & api, json11::Json payload, int timeout,
+    int retries, int interval, std::function<void(std::string, json11::Json)> callback)
+{
+    const auto & url = selected_etcd_url;
+    std::string req = payload.dump();
+    req = "POST "+url.path+api+" HTTP/1.1\r\n"
+        "Host: "+url.hostname+"\r\n"
+        "Content-Type: application/json\r\n"
+        "Content-Length: "+std::to_string(req.size())+"\r\n"
+        "Connection: keep-alive\r\n"
+        "Keep-Alive: timeout="+std::to_string(etcd_keepalive_timeout)+"\r\n"
+        "\r\n"+req;
+    retries--;
+    auto cb = [this, api, payload, timeout, retries, interval, callback,
+        cur_addr = url.addr](http_message_t *response)
+    {
+        std::string err;
+        json11::Json data;
+        response->parse_json_response(err, data);
+        if (err != "")
+        {
+            if (cur_addr == selected_etcd_url.addr)
+                selected_etcd_url = (http_url_t){};
+            if (retries > 0)
+            {
+                if (this->log_level > 0)
+                {
+                    fprintf(
+                        stderr, "Warning: etcd request failed: %s, retrying %d more times\n",
+                        err.c_str(), retries
+                    );
+                }
+                if (interval > 0)
+                {
+                    // FIXME: Prevent destruction of etcd_state_client if timers or requests are active
+                    tfd->set_timer(interval, false, [this, api, payload, timeout, retries, interval, callback](int)
+                    {
+                        etcd_call(api, payload, timeout, retries, interval, callback);
+                    });
+                }
+                else
+                    etcd_call(api, payload, timeout, retries, interval, callback);
+            }
+            else
+                callback(err, data);
+        }
+        else
+            callback(err, data);
+    };
+    if (!keepalive_client)
+        keepalive_client = http_init(get_http_ctx());
+    http_request(keepalive_client, url.addr, req, { .timeout = timeout, .keepalive = true, .ssl = url.ssl }, cb);
+}
+
+void etcd_state_client_http_t::parse_config(const json11::Json & config)
+{
+    auto old_etcd_ws_keepalive_interval = this->etcd_ws_keepalive_interval;
+    etcd_state_client_t::parse_config(config);
+    if (this->etcd_ws_keepalive_interval != old_etcd_ws_keepalive_interval && ws_keepalive_timer >= 0)
+    {
+        stop_ws_keepalive();
+        start_ws_keepalive();
+    }
+}
+
+void etcd_state_client_http_t::pick_next_etcd(std::function<void()> cb)
+{
+    if (!etcd_addresses.size() && !etcd_local.size())
+    {
+        fprintf(stderr, "etcd_address is missing in Vitastor configuration\n");
+        exit(1);
+    }
+    if (selected_etcd_url.addr != "")
+    {
+        cb();
+        return;
+    }
+    if (etcd_urls_to_try.size() != 0)
+    {
+        selected_etcd_url = std::move(etcd_urls_to_try[0]);
+        etcd_urls_to_try.erase(etcd_urls_to_try.begin());
+        cb();
+        return;
+    }
+    on_resolve_queue.push_back(std::move(cb));
+    if (on_resolve_queue.size() > 1)
+    {
+        // Already resolving
+        return;
+    }
+    assert(!resolve_count);
+    local_to_try = 0;
+    for (auto & url: etcd_local_addr_urls)
+    {
+        // Prefer local IPs, if any
+        etcd_urls_to_try.push_back(url);
+        local_to_try++;
+    }
+    for (auto & url: etcd_nonlocal_addr_urls)
+    {
+        etcd_urls_to_try.push_back(url);
+    }
+    resolve_count++;
+    for (auto & url: etcd_name_urls)
+    {
+        resolve_count++;
+        http_resolve(get_http_ctx(), url.ssl, url.addr, [this, url](const std::string & error, const std::vector<std::string>& addresses)
+        {
+            if (error != "")
+                fprintf(stderr, "Error resolving %s: %s\n", url.addr.c_str(), error.c_str());
+            for (auto & addr: addresses)
+            {
+                auto url_copy = url;
+                url_copy.addr = addr;
+                if (local_ips.find(addr) != local_ips.end())
+                {
+                    etcd_urls_to_try.insert(etcd_urls_to_try.begin(), std::move(url_copy));
+                    local_to_try++;
+                }
+                else
+                    etcd_urls_to_try.push_back(std::move(url_copy));
+            }
+            resolve_count--;
+            if (!resolve_count)
+                pick_next_etcd_on_resolve();
+        });
+    }
+    resolve_count--;
+    if (!resolve_count)
+    {
+        pick_next_etcd_on_resolve();
+    }
+}
+
+void etcd_state_client_http_t::pick_next_etcd_on_resolve()
+{
+    if (!etcd_urls_to_try.size())
+    {
+        fprintf(stderr, "None of etcd_address could be resolved\n");
+        exit(1);
+    }
+    if (!rand_initialized)
+    {
+        timespec tv;
+        clock_gettime(CLOCK_REALTIME, &tv);
+        srand48(tv.tv_sec*1000000000 + tv.tv_nsec);
+        rand_initialized = true;
+    }
+    // Shuffle addresses
+    for (size_t i = etcd_urls_to_try.size()-1; i > local_to_try; i--)
+    {
+        size_t j = local_to_try + lrand48() % (i - local_to_try);
+        if (j != i)
+            std::swap(etcd_urls_to_try[i], etcd_urls_to_try[j]);
+    }
+    selected_etcd_url = std::move(etcd_urls_to_try[0]);
+    etcd_urls_to_try.erase(etcd_urls_to_try.begin());
+    auto cbs = std::move(on_resolve_queue);
+    for (auto cb: cbs)
+    {
+        cb();
+    }
+}
+
+void etcd_state_client_http_t::start_etcd_watcher()
+{
+    pick_next_etcd([this]()
+    {
+        start_etcd_watcher_selected();
+    });
+}
+
+void etcd_state_client_http_t::start_etcd_watcher_selected()
+{
+    const auto & url = selected_etcd_url;
+    etcd_watches_initialised = 0;
+    ws_alive = 1;
+    if (this->log_level > 1)
+    {
+        fprintf(stderr, "Trying to connect to etcd websocket at %s%s%s (hostname %s), watch from revision %ju/%ju/%ju\n",
+            url.ssl ? "https://" : "http://", url.addr.c_str(), url.path.c_str(), url.hostname.c_str(),
+            etcd_watch_revision_config, etcd_watch_revision_osd, etcd_watch_revision_pg);
+    }
+    if (!etcd_watch_ws)
+        etcd_watch_ws = http_init(get_http_ctx());
+    else
+        http_close(etcd_watch_ws);
+    open_websocket(etcd_watch_ws, url.addr, url.hostname, url.path+"/watch", { .timeout = etcd_slow_timeout, .ssl = url.ssl },
+        [this, cur_addr = url.addr](http_message_t *msg)
+    {
+        if (msg->body.length())
+        {
+            ws_alive = 1;
+            std::string json_err;
+            json11::Json data = json11::Json::parse(msg->body, json_err);
+            if (json_err != "")
+            {
+                fprintf(stderr, "Bad JSON in etcd event: %s, ignoring event\n", json_err.c_str());
+            }
+            else
+            {
+                uint64_t watch_id = data["result"]["watch_id"].uint64_value();
+                if (data["result"]["created"].bool_value())
+                {
+                    if (watch_id == ETCD_CONFIG_WATCH_ID ||
+                        watch_id == ETCD_PG_STATE_WATCH_ID ||
+                        watch_id == ETCD_OSD_STATE_WATCH_ID)
+                    {
+                        etcd_watches_initialised++;
+                    }
+                    if (etcd_watches_initialised == ETCD_TOTAL_WATCHES && this->log_level > 0)
+                    {
+                        fprintf(stderr, "Successfully subscribed to etcd at %s, revision %ju/%ju/%ju\n", cur_addr.c_str(),
+                            etcd_watch_revision_config, etcd_watch_revision_osd, etcd_watch_revision_pg);
+                    }
+                }
+                if (data["result"]["canceled"].bool_value())
+                {
+                    // etcd watch canceled, maybe because the revision was compacted
+                    if (data["result"]["compact_revision"].uint64_value())
+                    {
+                        // we may miss events if we proceed
+                        // so we should restart from the beginning if we can
+                        if (on_reload_hook != NULL)
+                        {
+                            // check to not trigger on_reload_hook multiple times
+                            if (etcd_watch_ws != NULL)
+                            {
+                                fprintf(stderr, "Revisions before %ju were compacted by etcd, reloading state\n",
+                                    data["result"]["compact_revision"].uint64_value());
+                                http_close(etcd_watch_ws);
+                                etcd_watch_revision_config = etcd_watch_revision_osd = etcd_watch_revision_pg = 0;
+                                on_reload_hook();
+                            }
+                            return;
+                        }
+                        else
+                        {
+                            fprintf(stderr, "Revisions before %ju were compacted by etcd, exiting\n",
+                                data["result"]["compact_revision"].uint64_value());
+                            exit(1);
+                        }
+                    }
+                    else
+                    {
+                        fprintf(stderr, "Watch canceled by etcd, reason: %s, exiting\n", data["result"]["cancel_reason"].string_value().c_str());
+                        exit(1);
+                    }
+                }
+                // Save revision only if it's present in the message - because sometimes etcd sends something without a header, like:
+                // {"error": {"grpc_code": 14, "http_code": 503, "http_status": "Service Unavailable", "message": "error reading from server: EOF"}}
+                // Also don't save revision from the initial created: true messages because they always contain the latest revision
+                if (etcd_watches_initialised == ETCD_TOTAL_WATCHES &&
+                    !data["result"]["header"]["revision"].is_null() &&
+                    !data["result"]["created"].bool_value())
+                {
+                    // Restart watchers from the same revision number as in the last received message,
+                    // not from the next one to protect against revision being split into multiple messages,
+                    // even though etcd guarantees not to do that **within a single watcher** without fragment=true:
+                    // https://etcd.io/docs/v3.5/learning/api_guarantees/#watch-apis
+                    // Revision contents are ALWAYS split into separate messages for different watchers though!
+                    // So generally we have to resume each watcher from its own revision...
+                    // Progress messages may have watch_id=-1 if sent on behalf of multiple watchers though.
+                    // And antietcd has an advanced semantic which merges the same revision for all watchers
+                    // into one message and just omits watch_id.
+                    // So we also have to handle the case where watch_id is -1 or not present (0).
+                    auto watch_rev = data["result"]["header"]["revision"].uint64_value();
+                    if (!watch_id || watch_id == UINT64_MAX)
+                        etcd_watch_revision_config = etcd_watch_revision_osd = etcd_watch_revision_pg = watch_rev;
+                    else if (watch_id == ETCD_CONFIG_WATCH_ID)
+                        etcd_watch_revision_config = watch_rev;
+                    else if (watch_id == ETCD_PG_STATE_WATCH_ID)
+                        etcd_watch_revision_pg = watch_rev;
+                    else if (watch_id == ETCD_OSD_STATE_WATCH_ID)
+                        etcd_watch_revision_osd = watch_rev;
+                    etcd_urls_to_try.clear();
+                }
+                // First gather all changes into a hash to remove multiple overwrites
+                std::map<std::string, etcd_kv_t> changes;
+                for (auto & ev: data["result"]["events"].array_items())
+                {
+                    auto kv = parse_etcd_kv(ev["kv"]);
+                    if (kv.key != "")
+                    {
+                        changes[kv.key] = kv;
+                    }
+                }
+                for (auto & kv: changes)
+                {
+                    if (this->log_level > 3)
+                    {
+                        fprintf(stderr, "Incoming event: %s -> %s\n", kv.first.c_str(), kv.second.value.dump().c_str());
+                    }
+                    parse_state(kv.second);
+                }
+                // React to changes
+                if (on_change_hook != NULL)
+                {
+                    on_change_hook(changes);
+                }
+            }
+        }
+        if (msg->eof)
+        {
+            fprintf(stderr, "Disconnected from etcd %s\n", cur_addr.c_str());
+            if (cur_addr == selected_etcd_url.addr)
+                selected_etcd_url = (http_url_t){};
+            if (etcd_watches_initialised == 0)
+            {
+                // Connection not established, retry in <etcd_quick_timeout>
+                tfd->set_timer(etcd_quick_timeout, false, [this](int)
+                {
+                    start_etcd_watcher();
+                });
+            }
+            else if (etcd_watches_initialised > 0)
+            {
+                // Connection was live, retry immediately
+                etcd_watches_initialised = 0;
+                start_etcd_watcher();
+            }
+        }
+    });
+    http_post_message(etcd_watch_ws, WS_TEXT, json11::Json(json11::Json::object {
+        { "create_request", json11::Json::object {
+            { "key", base64_encode(etcd_prefix+"/config/") },
+            { "range_end", base64_encode(etcd_prefix+"/config0") },
+            { "start_revision", etcd_watch_revision_config },
+            { "watch_id", ETCD_CONFIG_WATCH_ID },
+            { "progress_notify", true },
+        } }
+    }).dump());
+    http_post_message(etcd_watch_ws, WS_TEXT, json11::Json(json11::Json::object {
+        { "create_request", json11::Json::object {
+            { "key", base64_encode(etcd_prefix+"/osd/state/") },
+            { "range_end", base64_encode(etcd_prefix+"/osd/state0") },
+            { "start_revision", etcd_watch_revision_osd },
+            { "watch_id", ETCD_OSD_STATE_WATCH_ID },
+            { "progress_notify", true },
+        } }
+    }).dump());
+    http_post_message(etcd_watch_ws, WS_TEXT, json11::Json(json11::Json::object {
+        { "create_request", json11::Json::object {
+            { "key", base64_encode(etcd_prefix+"/pg/") },
+            { "range_end", base64_encode(etcd_prefix+"/pg0") },
+            { "start_revision", etcd_watch_revision_pg },
+            { "watch_id", ETCD_PG_STATE_WATCH_ID },
+            { "progress_notify", true },
+        } }
+    }).dump());
+    // FIXME: Do not watch /pg/history/ at all in client code (not in OSD)
+    if (on_start_watcher_hook)
+    {
+        on_start_watcher_hook(etcd_watch_ws);
+    }
+    start_ws_keepalive();
+}
+
+void etcd_state_client_http_t::stop_ws_keepalive()
+{
+    if (ws_keepalive_timer >= 0)
+    {
+        tfd->clear_timer(ws_keepalive_timer);
+        ws_keepalive_timer = -1;
+    }
+}
+
+void etcd_state_client_http_t::start_ws_keepalive()
+{
+    if (ws_keepalive_timer < 0)
+    {
+        ws_keepalive_timer = tfd->set_timer(etcd_ws_keepalive_interval*1000, true, [this](int)
+        {
+            if (!etcd_watch_ws || etcd_watches_initialised < ETCD_TOTAL_WATCHES)
+            {
+                // Do nothing
+            }
+            else if (!ws_alive)
+            {
+                if (this->log_level > 0)
+                {
+                    fprintf(stderr, "Websocket ping failed, disconnecting from etcd %s\n", selected_etcd_url.addr.c_str());
+                }
+                start_etcd_watcher();
+            }
+            else
+            {
+                ws_alive = 0;
+                http_post_message(etcd_watch_ws, WS_TEXT, json11::Json(json11::Json::object {
+                    { "progress_request", json11::Json::object { } }
+                }).dump());
+            }
+        });
+    }
+}
+
+void etcd_state_client_http_t::load_global_config()
+{
+    etcd_state_client_t::load_global_config([this](const std::string & err)
+    {
+        if (err != "")
+        {
+            fprintf(stderr, "Error reading configuration from etcd: %s\n", err.c_str());
+            if (infinite_start)
+            {
+                tfd->set_timer(etcd_slow_timeout, false, [this](int timer_id)
+                {
+                    load_global_config();
+                });
+            }
+            else
+            {
+                exit(1);
+            }
+        }
+    });
+}
+
+void etcd_state_client_http_t::load_pgs()
+{
+    timespec tv;
+    clock_gettime(CLOCK_REALTIME, &tv);
+    uint64_t ms_passed = (tv.tv_sec-etcd_last_reload.tv_sec)*1000 + (tv.tv_nsec-etcd_last_reload.tv_nsec)/1000000;
+    if (ms_passed < etcd_min_reload_interval)
+    {
+        if (load_pgs_timer_id < 0)
+        {
+            load_pgs_timer_id = tfd->set_timer(etcd_min_reload_interval+50-ms_passed, false, [this](int) { load_pgs(); });
+        }
+        return;
+    }
+    etcd_last_reload = tv;
+    if (load_pgs_timer_id >= 0)
+    {
+        tfd->clear_timer(load_pgs_timer_id);
+        load_pgs_timer_id = -1;
+    }
+    etcd_state_client_t::load_pgs([this](const std::string & err)
+    {
+        if (err != "")
+        {
+            // Retry indefinitely
+            fprintf(stderr, "Error loading PGs from etcd: %s\n", err.c_str());
+            tfd->set_timer(etcd_slow_timeout, false, [this](int timer_id)
+            {
+                load_pgs();
+            });
+        }
+        else
+        {
+            start_etcd_watcher();
+        }
+    });
+}