WIP Batch encrypted records to reduce EncryptInit/EncryptFinal costs

Seems slightly overcomplicated...

.gitea/workflows/test.yml

@@ -63,7 +63,7 @@ jobs:
     container: ${{env.TEST_IMAGE}}:${{github.sha}}
     steps:
     # leak sanitizer sometimes crashes
-    - run: cd /root/vitastor/build && ASAN_OPTIONS=detect_leaks=0 make -j16 test
+    - run: cd /root/vitastor/build && ASAN_OPTIONS=detect_leaks=0 make -j16 build_tests test
 
   npm_lint:
     runs-on: ubuntu-latest
@@ -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
@@ -648,6 +702,24 @@ 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_old_snapshot_chain:
     runs-on: ubuntu-latest
     needs: build
@@ -1224,6 +1296,24 @@ 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
@@ -2052,3 +2142,39 @@ 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

@@ -1,20 +1,20 @@
-cmake_minimum_required(VERSION 2.8.12)
+cmake_minimum_required(VERSION 2.8...3.30)
 
 project(vitastor)
 
-set(VITASTOR_VERSION "3.0.5")
+set(VITASTOR_VERSION "3.0.9")
 
 include(CTest)
 
 add_custom_target(build_tests)
-add_custom_target(test
-	COMMAND
-	echo leak:tcmalloc > ${CMAKE_CURRENT_BINARY_DIR}/lsan-suppress.txt &&
-	env LSAN_OPTIONS=suppressions=${CMAKE_CURRENT_BINARY_DIR}/lsan-suppress.txt ${CMAKE_CTEST_COMMAND}
+set_property(TEST PROPERTY ENVIRONMENT LSAN_OPTIONS=suppressions=${CMAKE_CURRENT_BINARY_DIR}/lsan-suppress.txt)
+add_test(gen_lsan_suppress
+    ${CMAKE_COMMAND} -E echo leak:tcmalloc > "${CMAKE_CURRENT_BINARY_DIR}/lsan-suppress.txt"
 )
+set_tests_properties(gen_lsan_suppress PROPERTIES FIXTURES_SETUP f_lsan_suppress)
+set_property(TEST PROPERTY FIXTURES_REQUIRED f_lsan_suppress)
 # make -j16 -C ../../build test_heap && ../../build/src/test/test_heap
 # make -j16 -C ../../build test_heap && rm -f $(find ../../build -name '*.gcda') && ctest -V -T test -T coverage -R heap --test-dir ../../build && (cd ../../build; gcovr -f ../src --html --html-nested -o coverage/index.html; cd ../src/test)
 # make -j16 -C ../../build test_blockstore && rm -f $(find ../../build -name '*.gcda') && ctest -V -T test -T coverage -R blockstore --test-dir ../../build && (cd ../../build; gcovr -f ../src --html --html-nested -o coverage/index.html; cd ../src/test)
 # kcov --include-path=../../../src ../../kcov ./test_blockstore
-add_dependencies(test build_tests)
 add_subdirectory(src)

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 ?= v3.0.5
+VITASTOR_VERSION ?= v3.0.9
 
 all: build push
 

csi/deploy/004-csi-nodeplugin.yaml

@@ -49,7 +49,7 @@ spec:
             capabilities:
               add: ["SYS_ADMIN"]
             allowPrivilegeEscalation: true
-          image: vitalif/vitastor-csi:v3.0.5
+          image: vitalif/vitastor-csi:v3.0.9
           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:v3.0.5
+          image: vitalif/vitastor-csi:v3.0.9
           args:
             - "--node=$(NODE_ID)"
             - "--endpoint=$(CSI_ENDPOINT)"

csi/src/config.go

@@ -5,7 +5,7 @@ package vitastor
 
 const (
     vitastorCSIDriverName    = "csi.vitastor.io"
-    vitastorCSIDriverVersion = "3.0.5"
+    vitastorCSIDriverVersion = "3.0.9"
 )
 
 // Config struct fills the parameters of request or user input

debian/changelog

@@ -1,4 +1,4 @@
-vitastor (3.0.5-1) unstable; urgency=medium
+vitastor (3.0.9-1) 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/vitastor-build.sh

@@ -44,7 +44,7 @@ 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" && -e ../vitastor-bookworm/vitastor_$VER.orig.tar.xz ]]; then
+if [[ ( "$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)
     cp ../vitastor-bookworm/vitastor_$VER.orig.tar.xz .
 else

debian/vitastor-buildenv.Dockerfile

@@ -25,7 +25,7 @@ RUN set -e -x; \
     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 \
         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 ?= v3.0.5
+VITASTOR_VERSION ?= v3.0.9
 
 all: build push
 

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

@@ -7,7 +7,7 @@ PartOf=vitastor.target
 [Service]
 Restart=always
 EnvironmentFile=/etc/vitastor/docker.conf
-ExecStart=bash -c 'docker run --rm -i -v /etc/vitastor:/etc/vitastor -v /dev:/dev -v /run:/run \
+ExecStart=bash -c 'docker run --rm -i -v /etc/vitastor:/etc/vitastor -v /dev:/dev -v /run:/run -e SYSTEMD_IN_CHROOT=0 \
     --security-opt seccomp=unconfined --privileged --pid=host --log-driver none --network host --name vitastor vitastor:$VITASTOR_VERSION \
     sleep.sh'
 ExecStartPost=udevadm trigger

docker/etc/vitastor/docker.conf

@@ -4,7 +4,7 @@
 #
 
 # Desired Vitastor version
-VITASTOR_VERSION=v3.0.5
+VITASTOR_VERSION=v3.0.9
 
 # Additional arguments for all containers
 # For example, you may want to specify a custom logging driver here

docker/install.sh

@@ -2,8 +2,7 @@
 
 set -e
 
-cp -urv /etc/default /host-etc/
-cp -urv /etc/systemd /host-etc/
-cp -urv /etc/udev /host-etc/
+cp -urv /etc/systemd/system/vitastor* /host-etc/systemd/system/
+cp -urv /etc/udev/rules.d /host-etc/udev/
 cp -urnv /etc/vitastor /host-etc/
 cp -urnv /opt/scripts/* /host-bin/

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/osd.en.md

@@ -70,6 +70,7 @@ with an OSD restart or, for some of them, even without restarting by updating co
 - [use_atomic_flag](#use_atomic_flag)
 - [pg_reshard_chunk_size](#pg_reshard_chunk_size)
 - [pg_reshard_chunk_pause_ms](#pg_reshard_chunk_pause_ms)
+- [gc_on_start](#gc_on_start)
 
 ## bind_address
 
@@ -753,3 +754,9 @@ This option sets the maximum number of object is a chunk. Moving 100k objects us
 - Default: 100
 
 This option sets the interval between handling two PG count change chunks.
+
+## gc_on_start
+
+- Type: boolean
+
+Forcibly clean all garbage entries in the new store on every OSD restart.

docs/config/osd.ru.md

@@ -71,6 +71,7 @@
 - [use_atomic_flag](#use_atomic_flag)
 - [pg_reshard_chunk_size](#pg_reshard_chunk_size)
 - [pg_reshard_chunk_pause_ms](#pg_reshard_chunk_pause_ms)
+- [gc_on_start](#gc_on_start)
 
 ## bind_address
 
@@ -793,3 +794,9 @@ pg_minsize OSD во время переключений, что может по
 - Значение по умолчанию: 100
 
 Данная опция задаёт интервал между обработкой двух порций изменения числа PG пулов.
+
+## gc_on_start
+
+- Тип: булево (да/нет)
+
+Принудительно очищать все мусорные записи в новом хранилище при каждом запуске OSD.

docs/config/security.en.md

@@ -0,0 +1,150 @@
+[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.
+
+- [etcd_client_cert](#etcd_client_cert)
+- [etcd_client_key](#etcd_client_key)
+- [etcd_ca](#etcd_ca)
+- [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)
+- [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)
+- [max_aes_xts_pool_size](#max_aes_xts_pool_size)
+
+## etcd_client_cert
+
+- Type: string
+
+Client TLS certificate to use for Vitastor client (not OSD and not monitor)
+etcd https connections. May be path to a file or just a PEM string with certificate.
+In the latter case, string must begin with "-----BEGIN CERTIFICATE-----".
+
+## etcd_client_key
+
+- Type: string
+
+Private key for etcd_client_cert (also a file or a PEM string).
+
+## etcd_ca
+
+- Type: string
+
+Trusted TLS CA to verify etcd server certificate. May be path to a file,
+directory or just a PEM string with certificate.
+
+## osd_etcd_client_cert
+
+- Type: string
+
+Same as [etcd_client_cert](#etcd_client_cert), but only for OSDs.
+OSDs, clients and monitors should have different permissions, so they should
+use different certificates.
+
+## osd_etcd_client_key
+
+- Type: string
+
+Same as [etcd_client_key](#etcd_client_key), but only for OSDs.
+
+## mon_etcd_client_cert
+
+- Type: string
+
+Same as [etcd_client_cert](#etcd_client_cert), but only for Vitastor monitors.
+
+## mon_etcd_client_key
+
+- Type: string
+
+Same as [etcd_client_key](#etcd_client_key), but only for Vitastor monitors.
+
+## 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. Just like [etcd_client_cert](#etcd_client_cert),
+may be path to a file or just a certificate in PEM string.
+
+## vault_client_key
+
+- Type: string
+
+Private key for vault_client_cert (also a file or a PEM string).
+
+## 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.
+
+## max_aes_xts_pool_size
+
+- Type: integer
+- Default: 256
+
+Maximum number of OpenSSL encryption contexts cached in OSD memory. Probably
+doesn't require modification.

docs/config/security.ru.md

@@ -0,0 +1,154 @@
+[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.
+
+- [etcd_client_cert](#etcd_client_cert)
+- [etcd_client_key](#etcd_client_key)
+- [etcd_ca](#etcd_ca)
+- [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)
+- [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)
+- [max_aes_xts_pool_size](#max_aes_xts_pool_size)
+
+## etcd_client_cert
+
+- Тип: строка
+
+Клиентский TLS сертификат для https-подключений к etcd для клиентов Vitastor
+(не OSD и не мониторов). Может быть путём к файлу или просто строкой с
+сертификатом в формате PEM. В последнем случае строка должна начинаться с
+"-----BEGIN CERTIFICATE-----".
+
+## etcd_client_key
+
+- Тип: строка
+
+Закрытый ключ для сертификата etcd_client_cert (также путь к файлу или PEM строка).
+
+## etcd_ca
+
+- Тип: строка
+
+Доверенный корневой TLS-сертификат для проверки сертификата сервера etcd.
+Может быть путём к файлу, директории или просто строкой с сертификатом в
+формате PEM.
+
+## osd_etcd_client_cert
+
+- Тип: строка
+
+Аналогично [etcd_client_cert](#etcd_client_cert), но только для OSD.
+OSD, клиенты и мониторы должны иметь разные привилегии, поэтому они должны
+использовать разные сертификаты.
+
+## osd_etcd_client_key
+
+- Тип: строка
+
+Аналогично [etcd_client_key](#etcd_client_key), но только для OSD.
+
+## mon_etcd_client_cert
+
+- Тип: строка
+
+Аналогично [etcd_client_cert](#etcd_client_cert), но только для мониторов Vitastor.
+
+## mon_etcd_client_key
+
+- Тип: строка
+
+Аналогично [etcd_client_key](#etcd_client_key), но только для мониторов Vitastor.
+
+## 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. Как и [etcd_client_cert](#etcd_client_cert),
+может быть путём к файлу или просто PEM-строкой с сертификатом.
+
+## vault_client_key
+
+- Тип: строка
+
+Закрытый ключ для сертификата vault_client_cert (также путь к файлу или PEM строка).
+
+## 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.
+
+## max_aes_xts_pool_size
+
+- Тип: целое число
+- Значение по умолчанию: 256
+
+Максимальное количество кэшируемых в памяти OSD контекстов шифрования OpenSSL.
+Вряд ли требует изменения.

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/osd.yml

@@ -938,3 +938,7 @@
     This option sets the interval between handling two PG count change chunks.
   info_ru: |
     Данная опция задаёт интервал между обработкой двух порций изменения числа PG пулов.
+- name: gc_on_start
+  type: bool
+  info: Forcibly clean all garbage entries in the new store on every OSD restart.
+  info_ru: Принудительно очищать все мусорные записи в новом хранилище при каждом запуске OSD.

docs/config/src/package.json

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

docs/config/src/security.en.md

@@ -0,0 +1,5 @@
+# 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.

docs/config/src/security.ru.md

@@ -0,0 +1,7 @@
+# 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.

docs/config/src/security.yml

@@ -0,0 +1,131 @@
+- name: etcd_client_cert
+  type: string
+  info: |
+    Client TLS certificate to use for Vitastor client (not OSD and not monitor)
+    etcd https connections. May be path to a file or just a PEM string with certificate.
+    In the latter case, string must begin with "-----BEGIN CERTIFICATE-----".
+  info_ru: |
+    Клиентский TLS сертификат для https-подключений к etcd для клиентов Vitastor
+    (не OSD и не мониторов). Может быть путём к файлу или просто строкой с
+    сертификатом в формате PEM. В последнем случае строка должна начинаться с
+    "-----BEGIN CERTIFICATE-----".
+- name: etcd_client_key
+  type: string
+  info: Private key for etcd_client_cert (also a file or a PEM string).
+  info_ru: Закрытый ключ для сертификата etcd_client_cert (также путь к файлу или PEM строка).
+- name: etcd_ca
+  type: string
+  info: |
+    Trusted TLS CA to verify etcd 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 etcd.
+    Может быть путём к файлу, директории или просто строкой с сертификатом в
+    формате PEM.
+- name: osd_etcd_client_cert
+  type: string
+  info: |
+    Same as [etcd_client_cert](#etcd_client_cert), but only for OSDs.
+    OSDs, clients and monitors should have different permissions, so they should
+    use different certificates.
+  info_ru: |
+    Аналогично [etcd_client_cert](#etcd_client_cert), но только для OSD.
+    OSD, клиенты и мониторы должны иметь разные привилегии, поэтому они должны
+    использовать разные сертификаты.
+- name: osd_etcd_client_key
+  type: string
+  info: Same as [etcd_client_key](#etcd_client_key), but only for OSDs.
+  info_ru: Аналогично [etcd_client_key](#etcd_client_key), но только для OSD.
+- name: mon_etcd_client_cert
+  type: string
+  info: Same as [etcd_client_cert](#etcd_client_cert), but only for Vitastor monitors.
+  info_ru: Аналогично [etcd_client_cert](#etcd_client_cert), но только для мониторов Vitastor.
+- name: mon_etcd_client_key
+  type: string
+  info: Same as [etcd_client_key](#etcd_client_key), but only for Vitastor monitors.
+  info_ru: Аналогично [etcd_client_key](#etcd_client_key), но только для мониторов Vitastor.
+- 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. Just like [etcd_client_cert](#etcd_client_cert),
+    may be path to a file or just a certificate in PEM string.
+  info_ru: |
+    Клиентский TLS сертификат для подключений к Vault. Как и [etcd_client_cert](#etcd_client_cert),
+    может быть путём к файлу или просто PEM-строкой с сертификатом.
+- name: vault_client_key
+  type: string
+  info: Private key for vault_client_cert (also a file or a PEM string).
+  info_ru: Закрытый ключ для сертификата vault_client_cert (также путь к файлу или PEM строка).
+- 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.
+- name: max_aes_xts_pool_size
+  type: int
+  default: 256
+  info: |
+    Maximum number of OpenSSL encryption contexts cached in OSD memory. Probably
+    doesn't require modification.
+  info_ru: |
+    Максимальное количество кэшируемых в памяти OSD контекстов шифрования OpenSSL.
+    Вряд ли требует изменения.

docs/installation/docker.en.md

@@ -26,13 +26,37 @@ 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:v3.0.5`
+   `docker pull vitalif/vitastor:v3.0.9`
 2. Install scripts to the host system: \
-   `docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.0.5 install.sh`
+   `docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.0.9 install.sh`
 3. Reload udev rules: \
    `udevadm control --reload-rules`
+4. Enable the vitastor-host service: \
+   `systemctl enable --now vitastor-host`
 
-And you can return to [Quick Start](../intro/quickstart.en.md).
+After these steps, you can return to [Quick Start](../intro/quickstart.en.md).
+
+## Podman
+
+If you use Podman, run the following commands as root before installing Vitastor containers:
+
+```
+ln -s podman /usr/bin/docker
+
+mkdir -p /etc/systemd/system/systemd-udevd.service.d
+
+cat >/etc/systemd/system/systemd-udevd.service.d/override.conf <<EOF
+[Service]
+CapabilityBoundingSet=~
+SystemCallFilter=@mount capset
+EOF
+
+systemctl daemon-reload
+
+systemctl restart systemd-udevd
+```
+
+Without it, udev fails to do calls into a Podman container and Vitastor disk detection doesn't work.
 
 ## Upgrading Containers
 

docs/installation/docker.ru.md

@@ -25,14 +25,39 @@ Vitastor можно установить в Docker/Podman. При этом etcd,
 Инструкция по установке максимально простая.
 
 1. Скачайте Docker-образ желаемой версии: \
-   `docker pull vitalif/vitastor:v3.0.5`
+   `docker pull vitalif/vitastor:v3.0.9`
 2. Установите скрипты в хост-систему командой: \
-   `docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.0.5 install.sh`
+   `docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.0.9 install.sh`
 3. Перезагрузите правила udev: \
    `udevadm control --reload-rules`
+4. Включите сервис vitastor-host: \
+   `systemctl enable --now vitastor-host`
 
 После этого вы можете возвращаться к разделу [Быстрый старт](../intro/quickstart.ru.md).
 
+## Podman
+
+Если вы используете Podman, перед установкой контейнеров Vitastor выполните следующие
+команды от имени суперпользователя:
+
+```
+ln -s podman /usr/bin/docker
+
+mkdir -p /etc/systemd/system/systemd-udevd.service.d
+
+cat >/etc/systemd/system/systemd-udevd.service.d/override.conf <<EOF
+[Service]
+CapabilityBoundingSet=~
+SystemCallFilter=@mount capset
+EOF
+
+systemctl daemon-reload
+
+systemctl restart systemd-udevd
+```
+
+Без этих настроек udev не может делать вызовы внутрь Podman-контейнеров и определение дисков Vitastor не работает.
+
 ## Обновление контейнеров
 
 Сначала обязательно проверьте раздел [Обновление Vitastor](../usage/admin.ru.md#обновление-vitastor),

docs/installation/packages.en.md

@@ -17,6 +17,7 @@
   - Debian 10 (Buster): `deb https://vitastor.io/debian buster main`
   - Ubuntu 22.04 (Jammy): `deb https://vitastor.io/debian jammy main`
   - Ubuntu 24.04 (Noble): `deb https://vitastor.io/debian noble main`
+  - Ubuntu 26.04 (Resolute): `deb https://vitastor.io/debian resolute main`
   - Add `-oldstable` to bookworm/bullseye/buster in this line to install the last
     stable version from 0.9.x branch instead of 1.x
 - To always prefer vitastor-patched QEMU and Libvirt versions, add the following to `/etc/apt/preferences`:

docs/installation/packages.ru.md

@@ -17,6 +17,7 @@
   - Debian 10 (Buster): `deb https://vitastor.io/debian buster main`
   - Ubuntu 22.04 (Jammy): `deb https://vitastor.io/debian jammy main`
   - Ubuntu 24.04 (Noble): `deb https://vitastor.io/debian noble main`
+  - Ubuntu 26.04 (Resolute): `deb https://vitastor.io/debian resolute main`
   - Добавьте `-oldstable` к слову bookworm/bullseye/buster в этой строке, чтобы
     установить последнюю стабильную версию из ветки 0.9.x вместо 1.x
 - Чтобы всегда предпочитались версии пакетов QEMU и Libvirt с патчами Vitastor, добавьте в `/etc/apt/preferences`:

docs/installation/source.en.md

@@ -15,8 +15,8 @@
 - 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
-- ISA-L, libibverbs and librdmacm headers and libraries (optional)
+- jerasure, c-ares headers and libraries
+- ISA-L, libibverbs, librdmacm, libnl3 headers and libraries (optional)
 - tcmalloc (google-perftools-dev)
 
 ## Basic instructions

docs/installation/source.ru.md

@@ -15,8 +15,8 @@
 - 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
-- Опционально - заголовки и библиотеки ISA-L, libibverbs, librdmacm
+- Заголовки и библиотеки jerasure, c-ares
+- Опционально - заголовки и библиотеки ISA-L, libibverbs, librdmacm, libnl3
 - tcmalloc (google-perftools-dev)
 
 ## 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/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

@@ -18,7 +18,7 @@ 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 +26,8 @@ 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 = cluster.map(s => s.split(':', 3)).filter(ip => is_local[ip[1].substr(2)] && (!cfg_port || ip[2] == cfg_port));
             if (selected.length > 1)
             {
                 console.error('More than 1 etcd_address matches local IPs, please specify port');
@@ -35,16 +36,30 @@ 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][1].substr(2),
+                    port: selected[0][2],
+                    cert: config.antietcd_cert,
+                    key: config.antietcd_key,
+                    ca: config.etcd_ca,
+                    data: config.antietcd_data_file || ((config.antietcd_data_dir || '/var/lib/vitastor') + '/mon_'+selected[0][2]+'.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: selected[0][1].substr(2)+':'+selected[0][2], // node_id = ip:port
+                    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,
                 };
+                if (config.use_auth)
+                {
+                    antietcd_config.client_cert_auth = true;
+                    antietcd_config.auth_filter = require('./vitastor_auth_filter.js');
+                    antietcd_config.peer_ca = config.antietcd_server_ca;
+                    if (!config.antietcd_server_ca || config.antietcd_server_ca == config.etcd_ca)
+                    {
+                        console.error('Secure setup requires separate antietcd_server_ca (for signing antietcd server certificates) and etcd_ca (for signing client certificates)');
+                        process.exit(1);
+                    }
+                }
                 for (const key in config)
                 {
                     if (key.substr(0, 9) === 'antietcd_')
@@ -169,7 +184,7 @@ 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, { username: 'root' });
                 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/http_server.js

@@ -16,7 +16,7 @@ async function create_http_server(cfg, handler)
         };
         if (cfg.mon_https_ca)
         {
-            tls.mon_https_ca = await fsp.readFile(cfg.mon_https_ca);
+            tls.ca = await fsp.readFile(cfg.mon_https_ca);
         }
         if (cfg.mon_https_client_auth)
         {

mon/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vitastor-mon",
-  "version": "3.0.5",
+  "version": "3.0.9",
   "description": "Vitastor SDS monitor service",
   "main": "mon-main.js",
   "scripts": {

mon/stats.js

@@ -37,6 +37,7 @@ function derive_osd_stats(st, prev, prev_diff)
         const n = c.count - BigInt(pr && pr.count||0);
         diff.recovery_stats[op] = { ...c, bps: n > 0 ? b*1000n/timediff : 0n, iops: n > 0 ? n*1000n/timediff : 0n };
     }
+    diff.inode_stats = {};
     for (const pool_id in st.inode_stats||{})
     {
         diff.inode_stats[pool_id] = {};

mon/vitastor_auth_filter.js

@@ -0,0 +1,471 @@
+// 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: osd, mon, admin, client.
+//      osd, mon types should be used by OSDs/monitors.
+//      admin should be used for administrative access from vitastor-cli.
+//      client should be used for regular clients.
+//    - 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
+
+const static_perms = {
+    invalid: {
+        keys: {},
+        prefixes: {},
+    },
+    osd: {
+        keys: { '/pg/config': false },
+        prefixes: { '/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/stats/': false, '/pg/state/': false, '/index/maxid/': false },
+    },
+};
+
+const api_perms = {
+    osd: { lease_grant: true, lease_revoke: true, lease_keepalive: true },
+    mon: { lease_grant: true, lease_revoke: true, lease_keepalive: true },
+    admin: { maintenance_status: true },
+    client: {},
+};
+
+class VitastorAuthFilter
+{
+    constructor(antietcd)
+    {
+        this.cfg = antietcd.cfg;
+        this.antietcd = antietcd;
+        this.prefix = this.cfg.vitastor_prefix || '/vitastor';
+        this.prefix_parts = this.prefix.split('/');
+    }
+
+    _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;
+            }
+        }
+        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(username)
+    {
+        if (!username)
+        {
+            return null;
+        }
+        let userInfo = this._get([ ...this.prefix_parts, 'config', 'user', username ], true);
+        if (!userInfo)
+        {
+            userInfo = { type: 'client' };
+        }
+        userInfo.perms = static_perms[userInfo.type] || static_perms['invalid'];
+        userInfo.name = 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(username, api/*, data*/)
+    {
+        if (username === 'root')
+        {
+            return true;
+        }
+        const userInfo = this._get([ ...this.prefix_parts, 'config', 'user', username ], true);
+        return userInfo && api_perms[userInfo.type] && api_perms[userInfo.type][api];
+    }
+
+    filter_txn(username, txn)
+    {
+        if (username === 'root')
+        {
+            return true;
+        }
+        const userInfo = this._get_user(username);
+        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(username, txn, res)
+    {
+        if (!res.responses || username === 'root')
+        {
+            return;
+        }
+        const userInfo = this._get_user(username);
+        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(username, msg)
+    {
+        if (!msg.result || !msg.result.events || username === 'root')
+        {
+            return;
+        }
+        const userInfo = this._get_user(username);
+        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": "3.0.5",
+  "version": "3.0.9",
   "description": "Low-level native bindings to Vitastor client library",
   "main": "index.js",
   "keywords": [

patches/cinder-vitastor.py

@@ -50,7 +50,7 @@ from cinder.volume import configuration
 from cinder.volume import driver
 from cinder.volume import volume_utils
 
-VITASTOR_VERSION = '3.0.5'
+VITASTOR_VERSION = '3.0.9'
 
 LOG = logging.getLogger(__name__)
 
@@ -275,7 +275,7 @@ class VitastorDriver(driver.CloneableImageVD,
             LOG.exception('error getting vitastor pool stats: '+str(e))
 
         self._stats = stats
-        
+
     def get_volume_stats(self, refresh=False):
         """Get volume stats.
         If 'refresh' is True, run update the stats first.
@@ -291,6 +291,14 @@ class VitastorDriver(driver.CloneableImageVD,
         else:
             return (1 + resp['kvs'][0]['value'], resp['kvs'][0]['mod_revision'])
 
+    def _cli(self, descr, *args):
+        args = [ 'vitastor-cli', *args, *(self._vitastor_args()) ]
+        try:
+            self._execute(*args)
+        except processutils.ProcessExecutionError as exc:
+            LOG.error("Failed to "+descr+": "+exc)
+            raise exception.VolumeBackendAPIException(data = exc.stderr)
+
     def create_volume(self, volume):
         """Creates a logical volume."""
 
@@ -302,7 +310,7 @@ class VitastorDriver(driver.CloneableImageVD,
 
         LOG.debug("creating volume '%s'", vol_name)
 
-        self._create_image(vol_name, { 'size': size })
+        self._cli('create volume', 'create', vol_name, '--size', size)
 
         if volume.encryption_key_id:
             self._create_encrypted_volume(volume, volume.obj_context)
@@ -346,7 +354,7 @@ class VitastorDriver(driver.CloneableImageVD,
         snap_name = utils.convert_str(snapshot.name)
         if snap_name.find('@') >= 0 or snap_name.find('/') >= 0:
             raise exception.VolumeBackendAPIException(data = '@ and / are forbidden in volume and snapshot names')
-        self._create_snapshot(vol_name, vol_name+'@'+snap_name)
+        self._cli('create snapshot', 'snap-create', vol_name+'@'+snap_name)
 
     def snapshot_revert_use_temp_snapshot(self):
         """Disable the use of a temporary snapshot on revert."""
@@ -359,21 +367,8 @@ class VitastorDriver(driver.CloneableImageVD,
         snap_name = utils.convert_str(snapshot.name)
 
         # Delete the image and recreate it from the snapshot
-        args = [ 'vitastor-cli', 'rm', vol_name, *(self._vitastor_args()) ]
-        try:
-            self._execute(*args)
-        except processutils.ProcessExecutionError as exc:
-            LOG.error("Failed to delete image "+vol_name+": "+exc)
-            raise exception.VolumeBackendAPIException(data = exc.stderr)
-        args = [
-            'vitastor-cli', 'create', '--parent', vol_name+'@'+snap_name,
-            vol_name, *(self._vitastor_args())
-        ]
-        try:
-            self._execute(*args)
-        except processutils.ProcessExecutionError as exc:
-            LOG.error("Failed to recreate image "+vol_name+" from "+vol_name+"@"+snap_name+": "+exc)
-            raise exception.VolumeBackendAPIException(data = exc.stderr)
+        self._cli('delete image', 'rm', vol_name)
+        self._cli('recreate image', 'create', '--parent', vol_name+'@'+snap_name, vol_name)
 
     def delete_snapshot(self, snapshot):
         """Deletes a snapshot."""
@@ -381,15 +376,7 @@ class VitastorDriver(driver.CloneableImageVD,
         vol_name = utils.convert_str(snapshot.volume_name)
         snap_name = utils.convert_str(snapshot.name)
 
-        args = [
-            'vitastor-cli', 'rm', vol_name+'@'+snap_name,
-            *(self._vitastor_args())
-        ]
-        try:
-            self._execute(*args)
-        except processutils.ProcessExecutionError as exc:
-            LOG.error("Failed to remove snapshot "+vol_name+'@'+snap_name+": "+exc)
-            raise exception.VolumeBackendAPIException(data = exc.stderr)
+        self._cli('remove snapshot', 'rm', vol_name+'@'+snap_name)
 
     def _child_count(self, parents):
         children = 0
@@ -427,13 +414,7 @@ class VitastorDriver(driver.CloneableImageVD,
         if src_vref.admin_metadata.get('readonly') == 'True':
             # source volume is a volume-image cache entry or other readonly volume
             # clone without intermediate snapshot
-            src = self._get_image(src_name)
-            LOG.debug("creating image '%s' from '%s'", dest_name, src_name)
-            new_cfg = self._create_image(dest_name, {
-                'size': size,
-                'parent_id': src['idx']['id'],
-                'parent_pool_id': src['idx']['pool_id'],
-            })
+            self._cli('create clone', 'create', '--parent', src_name, '--size', size, dest_name)
             return {}
 
         clone_snap = "%s@%s.clone_snap" % (src_name, dest_name)
@@ -446,15 +427,12 @@ class VitastorDriver(driver.CloneableImageVD,
             clone_snap = dest_name
             make_img = False
 
-        LOG.debug("creating layer '%s' under '%s'", clone_snap, src_name)
-        new_cfg = self._create_snapshot(src_name, clone_snap, True)
+        LOG.debug("creating snapshot '%s'", clone_snap)
+        self._cli('create base snapshot', 'snap-create', '--allow-existing', '1', clone_snap)
+
         if make_img:
             # Then create a clone from it
-            new_cfg = self._create_image(dest_name, {
-                'size': size,
-                'parent_id': new_cfg['parent_id'],
-                'parent_pool_id': new_cfg['parent_pool_id'],
-            })
+            self._cli('create clone', 'create', '--parent', clone_snap, '--size', size, dest_name)
 
         return {}
 
@@ -464,7 +442,8 @@ class VitastorDriver(driver.CloneableImageVD,
         vol_name = utils.convert_str(volume.name)
         snap_name = utils.convert_str(snapshot.name)
 
-        snap = self._get_image('volume-'+snapshot.volume_id+'@'+snap_name)
+        src_snap = 'volume-'+snapshot.volume_id+'@'+snap_name
+        snap = self._get_image(src_snap)
         if not snap:
             raise exception.SnapshotNotFound(snapshot_id = snap_name)
         snap_inode_id = int(resp['responses'][0]['kvs'][0]['value']['id'])
@@ -473,12 +452,8 @@ class VitastorDriver(driver.CloneableImageVD,
         size = snap['cfg']['size']
         if int(volume.size):
             size = int(volume.size) * units.Gi
-        new_cfg = self._create_image(vol_name, {
-            'size': size,
-            'parent_id': snap['idx']['id'],
-            'parent_pool_id': snap['idx']['pool_id'],
-        })
 
+        self._cli('create clone', 'create', vol_name, '--size', size, '--parent', src_snap)
         return {}
 
     def _vitastor_args(self):
@@ -505,49 +480,7 @@ class VitastorDriver(driver.CloneableImageVD,
         """Deletes a logical volume."""
 
         vol_name = utils.convert_str(volume.name)
-
-        # Find the volume and all its snapshots
-        range_end = b'index/image/' + vol_name.encode('utf-8')
-        range_end = range_end[0 : len(range_end)-1] + six.int2byte(range_end[len(range_end)-1] + 1)
-        resp = self._etcd_txn({ 'success': [
-            { 'request_range': { 'key': 'index/image/'+vol_name, 'range_end': range_end } },
-        ] })
-        if len(resp['responses'][0]['kvs']) == 0:
-            # already deleted
-            LOG.info("volume %s no longer exists in backend", vol_name)
-            return
-        layers = resp['responses'][0]['kvs']
-        layer_ids = {}
-        for kv in layers:
-            inode_id = int(kv['value']['id'])
-            pool_id = int(kv['value']['pool_id'])
-            inode_pool_id = (pool_id << 48) | (inode_id & 0xffffffffffff)
-            layer_ids[inode_pool_id] = True
-
-        # Check if the volume has clones and raise 'busy' if so
-        children = self._child_count(layer_ids)
-        if children > 0:
-            raise exception.VolumeIsBusy(volume_name = vol_name)
-
-        # Clear data
-        for kv in layers:
-            args = [
-                'vitastor-cli', 'rm-data', '--pool', str(kv['value']['pool_id']),
-                '--inode', str(kv['value']['id']), '--progress', '0',
-                *(self._vitastor_args())
-            ]
-            try:
-                self._execute(*args)
-            except processutils.ProcessExecutionError as exc:
-                LOG.error("Failed to remove layer "+kv['key']+": "+exc)
-                raise exception.VolumeBackendAPIException(data = exc.stderr)
-
-        # Delete all layers from etcd
-        requests = []
-        for kv in layers:
-            requests.append({ 'request_delete_range': { 'key': kv['key'] } })
-            requests.append({ 'request_delete_range': { 'key': 'config/inode/'+str(kv['value']['pool_id'])+'/'+str(kv['value']['id']) } })
-        self._etcd_txn({ 'success': requests })
+        self._cli('delete volume', 'rm', '--matching', vol_name, vol_name+'@*', '--progress', '0')
 
     def retype(self, context, volume, new_type, diff, host):
         """Change extra type specifications for a volume."""
@@ -567,98 +500,6 @@ class VitastorDriver(driver.CloneableImageVD,
         """Removes an export for a logical volume."""
         pass
 
-    def _create_image(self, vol_name, cfg):
-        pool_s = str(self.cfg['pool_id'])
-        image_id = 0
-        while image_id == 0:
-            # check if the image already exists and find a free ID
-            resp = self._etcd_txn({ 'success': [
-                { 'request_range': { 'key': 'index/image/'+vol_name } },
-                { 'request_range': { 'key': 'index/maxid/'+pool_s } },
-            ] })
-            if len(resp['responses'][0]['kvs']) > 0:
-                # already exists
-                raise exception.VolumeBackendAPIException(data = 'Volume '+vol_name+' already exists')
-            image_id, id_mod = self._next_id(resp['responses'][1])
-            # try to create the image
-            resp = self._etcd_txn({ 'compare': [
-                { 'target': 'MOD', 'mod_revision': id_mod, 'key': 'index/maxid/'+pool_s },
-                { 'target': 'VERSION', 'version': 0, 'key': 'index/image/'+vol_name },
-                { 'target': 'VERSION', 'version': 0, 'key': 'config/inode/'+pool_s+'/'+str(image_id) },
-            ], 'success': [
-                { 'request_put': { 'key': 'index/maxid/'+pool_s, 'value': image_id } },
-                { 'request_put': { 'key': 'index/image/'+vol_name, 'value': json.dumps({
-                    'id': image_id, 'pool_id': self.cfg['pool_id']
-                }) } },
-                { 'request_put': { 'key': 'config/inode/'+pool_s+'/'+str(image_id), 'value': json.dumps({
-                    **cfg, 'name': vol_name,
-                }) } },
-            ] })
-            if not resp.get('succeeded'):
-                # repeat
-                image_id = 0
-
-    def _create_snapshot(self, vol_name, snap_vol_name, allow_existing = False):
-        while True:
-            # check if the image already exists and snapshot doesn't
-            resp = self._etcd_txn({ 'success': [
-                { 'request_range': { 'key': 'index/image/'+vol_name } },
-                { 'request_range': { 'key': 'index/image/'+snap_vol_name } },
-            ] })
-            if len(resp['responses'][0]['kvs']) == 0:
-                raise exception.VolumeBackendAPIException(data = 'Volume '+vol_name+' does not exist')
-            if len(resp['responses'][1]['kvs']) > 0:
-                if allow_existing:
-                    snap_idx = resp['responses'][1]['kvs'][0]['value']
-                    resp = self._etcd_txn({ 'success': [
-                        { 'request_range': { 'key': 'config/inode/'+str(snap_idx['pool_id'])+'/'+str(snap_idx['id']) } },
-                    ] })
-                    if len(resp['responses'][0]['kvs']) == 0:
-                        raise exception.VolumeBackendAPIException(data =
-                            'Volume '+snap_vol_name+' is already indexed, but does not exist'
-                        )
-                    return resp['responses'][0]['kvs'][0]['value']
-                raise exception.VolumeBackendAPIException(
-                    data = 'Volume '+snap_vol_name+' already exists'
-                )
-            vol_idx = resp['responses'][0]['kvs'][0]['value']
-            vol_idx_mod = resp['responses'][0]['kvs'][0]['mod_revision']
-            # get image inode config and find a new ID
-            resp = self._etcd_txn({ 'success': [
-                { 'request_range': { 'key': 'config/inode/'+str(vol_idx['pool_id'])+'/'+str(vol_idx['id']) } },
-                { 'request_range': { 'key': 'index/maxid/'+str(self.cfg['pool_id']) } },
-            ] })
-            if len(resp['responses'][0]['kvs']) == 0:
-                raise exception.VolumeBackendAPIException(data = 'Volume '+vol_name+' does not exist')
-            vol_cfg = resp['responses'][0]['kvs'][0]['value']
-            vol_mod = resp['responses'][0]['kvs'][0]['mod_revision']
-            new_id, id_mod = self._next_id(resp['responses'][1])
-            # try to redirect image to the new inode
-            new_cfg = {
-                **vol_cfg, 'name': vol_name, 'parent_id': vol_idx['id'], 'parent_pool_id': vol_idx['pool_id']
-            }
-            resp = self._etcd_txn({ 'compare': [
-                { 'target': 'MOD', 'mod_revision': vol_idx_mod, 'key': 'index/image/'+vol_name },
-                { 'target': 'MOD', 'mod_revision': vol_mod, 'key': 'config/inode/'+str(vol_idx['pool_id'])+'/'+str(vol_idx['id']) },
-                { 'target': 'MOD', 'mod_revision': id_mod, 'key': 'index/maxid/'+str(self.cfg['pool_id']) },
-                { 'target': 'VERSION', 'version': 0, 'key': 'index/image/'+snap_vol_name },
-                { 'target': 'VERSION', 'version': 0, 'key': 'config/inode/'+str(self.cfg['pool_id'])+'/'+str(new_id) },
-            ], 'success': [
-                { 'request_put': { 'key': 'index/maxid/'+str(self.cfg['pool_id']), 'value': new_id } },
-                { 'request_put': { 'key': 'index/image/'+vol_name, 'value': json.dumps({
-                    'id': new_id, 'pool_id': self.cfg['pool_id']
-                }) } },
-                { 'request_put': { 'key': 'config/inode/'+str(self.cfg['pool_id'])+'/'+str(new_id), 'value': json.dumps(new_cfg) } },
-                { 'request_put': { 'key': 'index/image/'+snap_vol_name, 'value': json.dumps({
-                    'id': vol_idx['id'], 'pool_id': vol_idx['pool_id']
-                }) } },
-                { 'request_put': { 'key': 'config/inode/'+str(vol_idx['pool_id'])+'/'+str(vol_idx['id']), 'value': json.dumps({
-                    **vol_cfg, 'name': snap_vol_name, 'readonly': True
-                }) } }
-            ] })
-            if resp.get('succeeded'):
-                return new_cfg
-
     def initialize_connection(self, volume, connector):
         data = {
             'driver_volume_type': 'vitastor',
@@ -697,13 +538,9 @@ class VitastorDriver(driver.CloneableImageVD,
                     size = int(volume.size) * units.Gi
                     dest_name = utils.convert_str(volume.name)
                     # Find or create the base snapshot
-                    snap_cfg = self._create_snapshot(base_vol.name, base_vol.name+'@.clone_snap', True)
+                    self._cli('create base snapshot', 'create', '--allow-existing', '1', base_vol.name+'@.clone_snap')
                     # Then create a clone from it
-                    new_cfg = self._create_image(dest_name, {
-                        'size': size,
-                        'parent_id': snap_cfg['parent_id'],
-                        'parent_pool_id': snap_cfg['parent_pool_id'],
-                    })
+                    self._cli('create clone', 'create', dest_name, '--size', size, '--parent', base_vol.name+'@.clone_snap')
                     return ({}, True)
         return ({}, False)
 
@@ -770,26 +607,8 @@ class VitastorDriver(driver.CloneableImageVD,
     def extend_volume(self, volume, new_size):
         """Extend an existing volume."""
         vol_name = utils.convert_str(volume.name)
-        while True:
-            vol = self._get_image(vol_name)
-            if not vol:
-                raise exception.VolumeBackendAPIException(data = 'Volume '+vol_name+' does not exist')
-            # change size
-            size = int(new_size) * units.Gi
-            if size == vol['cfg']['size']:
-                break
-            resp = self._etcd_txn({ 'compare': [ {
-                'target': 'MOD',
-                'mod_revision': vol['cfg_mod'],
-                'key': 'config/inode/'+str(vol['idx']['pool_id'])+'/'+str(vol['idx']['id']),
-            } ], 'success': [
-                { 'request_put': {
-                    'key': 'config/inode/'+str(vol['idx']['pool_id'])+'/'+str(vol['idx']['id']),
-                    'value': json.dumps({ **vol['cfg'], 'size': size }),
-                } },
-            ] })
-            if resp.get('succeeded'):
-                break
+        size = int(new_size) * units.Gi
+        self._cli('extend volume', 'modify', vol_name, '--resize', new_size)
         LOG.debug(
             "Extend volume from %(old_size)s GB to %(new_size)s GB.",
             {'old_size': volume.size, 'new_size': new_size}
@@ -862,28 +681,7 @@ class VitastorDriver(driver.CloneableImageVD,
         """
         from_name = self._get_existing_name(existing_ref)
         to_name = utils.convert_str(volume.name)
-        self._rename(from_name, to_name)
-
-    def _rename(self, from_name, to_name):
-        while True:
-            vol = self._get_image(from_name)
-            if not vol:
-                raise exception.VolumeBackendAPIException(data = 'Volume '+from_name+' does not exist')
-            to = self._get_image(to_name)
-            if to:
-                raise exception.VolumeBackendAPIException(data = 'Volume '+to_name+' already exists')
-            resp = self._etcd_txn({ 'compare': [
-                { 'target': 'MOD', 'mod_revision': vol['idx_mod'], 'key': 'index/image/'+vol['cfg']['name'] },
-                { 'target': 'MOD', 'mod_revision': vol['cfg_mod'], 'key': 'config/inode/'+str(vol['idx']['pool_id'])+'/'+str(vol['idx']['id']) },
-                { 'target': 'VERSION', 'version': 0, 'key': 'index/image/'+to_name },
-            ], 'success': [
-                { 'request_delete_range': { 'key': 'index/image/'+vol['cfg']['name'] } },
-                { 'request_put': { 'key': 'index/image/'+to_name, 'value': json.dumps(vol['idx']) } },
-                { 'request_put': { 'key': 'config/inode/'+str(vol['idx']['pool_id'])+'/'+str(vol['idx']['id']),
-                    'value': json.dumps({ **vol['cfg'], 'name': to_name }) } },
-            ] })
-            if resp.get('succeeded'):
-                break
+        self._cli('rename', 'modify', from_name, '--rename', to_name)
 
     def unmanage(self, volume):
         pass
@@ -956,7 +754,7 @@ class VitastorDriver(driver.CloneableImageVD,
         snap_name = self._get_existing_name(existing_ref)
         from_name = vol_name+'@'+snap_name
         to_name = vol_name+'@'+utils.convert_str(snapshot.name)
-        self._rename(from_name, to_name)
+        self._cli('rename', 'modify', from_name, '--rename', to_name)
 
     def unmanage_snapshot(self, snapshot):
         """Removes the specified snapshot from Cinder management."""

patches/libvirt-12.0-vitastor.diff

@@ -0,0 +1,637 @@
+diff --git a/include/libvirt/libvirt-storage.h b/include/libvirt/libvirt-storage.h
+index aaad4a3da1..5f5daa8341 100644
+--- a/include/libvirt/libvirt-storage.h
++++ b/include/libvirt/libvirt-storage.h
+@@ -326,6 +326,7 @@ typedef enum {
+     VIR_CONNECT_LIST_STORAGE_POOLS_ZFS           = 1 << 17, /* (Since: 1.2.8) */
+     VIR_CONNECT_LIST_STORAGE_POOLS_VSTORAGE      = 1 << 18, /* (Since: 3.1.0) */
+     VIR_CONNECT_LIST_STORAGE_POOLS_ISCSI_DIRECT  = 1 << 19, /* (Since: 5.6.0) */
++    VIR_CONNECT_LIST_STORAGE_POOLS_VITASTOR      = 1 << 20, /* (Since: 5.0.0) */
+ } virConnectListAllStoragePoolsFlags;
+ 
+ int                     virConnectListAllStoragePools(virConnectPtr conn,
+diff --git a/src/conf/domain_conf.c b/src/conf/domain_conf.c
+index 9ca5c2450c..cc52f00c0c 100644
+--- a/src/conf/domain_conf.c
++++ b/src/conf/domain_conf.c
+@@ -7453,7 +7453,8 @@ virDomainDiskSourceNetworkParse(xmlNodePtr node,
+     src->configFile = virXPathString("string(./config/@file)", ctxt);
+ 
+     if (src->protocol == VIR_STORAGE_NET_PROTOCOL_HTTP ||
+-        src->protocol == VIR_STORAGE_NET_PROTOCOL_HTTPS)
++        src->protocol == VIR_STORAGE_NET_PROTOCOL_HTTPS ||
++        src->protocol == VIR_STORAGE_NET_PROTOCOL_VITASTOR)
+         src->query = virXMLPropString(node, "query");
+ 
+     if (virDomainStorageNetworkParseHosts(node, ctxt, &src->hosts, &src->nhosts) < 0)
+@@ -32187,6 +32188,7 @@ virDomainStorageSourceTranslateSourcePool(virStorageSource *src,
+ 
+     case VIR_STORAGE_POOL_MPATH:
+     case VIR_STORAGE_POOL_RBD:
++    case VIR_STORAGE_POOL_VITASTOR:
+     case VIR_STORAGE_POOL_SHEEPDOG:
+     case VIR_STORAGE_POOL_GLUSTER:
+     case VIR_STORAGE_POOL_LAST:
+diff --git a/src/conf/domain_validate.c b/src/conf/domain_validate.c
+index 7346a61731..83e94d762e 100644
+--- a/src/conf/domain_validate.c
++++ b/src/conf/domain_validate.c
+@@ -520,6 +520,7 @@ virDomainDiskDefValidateSourceChainOne(const virStorageSource *src)
+         case VIR_STORAGE_NET_PROTOCOL_RBD:
+             break;
+ 
++        case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
+         case VIR_STORAGE_NET_PROTOCOL_NBD:
+         case VIR_STORAGE_NET_PROTOCOL_SHEEPDOG:
+         case VIR_STORAGE_NET_PROTOCOL_GLUSTER:
+@@ -592,7 +593,7 @@ virDomainDiskDefValidateSourceChainOne(const virStorageSource *src)
+         }
+     }
+ 
+-    /* internal snapshots and config files are currently supported only with rbd: */
++    /* internal snapshots are currently supported only with rbd: */
+     if (virStorageSourceGetActualType(src) != VIR_STORAGE_TYPE_NETWORK &&
+         src->protocol != VIR_STORAGE_NET_PROTOCOL_RBD) {
+         if (src->snapshot) {
+@@ -600,10 +601,14 @@ virDomainDiskDefValidateSourceChainOne(const virStorageSource *src)
+                            _("<snapshot> element is currently supported only with 'rbd' disks"));
+             return -1;
+         }
+-
++    }
++    /* config files are currently supported only with rbd and vitastor: */
++    if (virStorageSourceGetActualType(src) != VIR_STORAGE_TYPE_NETWORK &&
++        src->protocol != VIR_STORAGE_NET_PROTOCOL_RBD &&
++        src->protocol != VIR_STORAGE_NET_PROTOCOL_VITASTOR) {
+         if (src->configFile) {
+             virReportError(VIR_ERR_XML_ERROR, "%s",
+-                           _("<config> element is currently supported only with 'rbd' disks"));
++                           _("<config> element is currently supported only with 'rbd' and 'vitastor' disks"));
+             return -1;
+         }
+     }
+diff --git a/src/conf/schemas/domaincommon.rng b/src/conf/schemas/domaincommon.rng
+index 114dd3f96f..c71f9a3277 100644
+--- a/src/conf/schemas/domaincommon.rng
++++ b/src/conf/schemas/domaincommon.rng
+@@ -2093,6 +2093,35 @@
+     </element>
+   </define>
+ 
++  <define name="diskSourceNetworkProtocolVitastor">
++    <element name="source">
++      <interleave>
++        <attribute name="protocol">
++          <value>vitastor</value>
++        </attribute>
++        <ref name="diskSourceCommon"/>
++        <optional>
++          <attribute name="name"/>
++        </optional>
++        <optional>
++          <attribute name="query"/>
++        </optional>
++        <zeroOrMore>
++          <ref name="diskSourceNetworkHost"/>
++        </zeroOrMore>
++        <optional>
++          <element name="config">
++            <attribute name="file">
++              <ref name="absFilePath"/>
++            </attribute>
++            <empty/>
++          </element>
++        </optional>
++        <empty/>
++      </interleave>
++    </element>
++  </define>
++
+   <define name="diskSourceNetworkProtocolISCSI">
+     <element name="source">
+       <attribute name="protocol">
+@@ -2443,6 +2472,7 @@
+       <ref name="diskSourceNetworkProtocolSimple"/>
+       <ref name="diskSourceNetworkProtocolVxHS"/>
+       <ref name="diskSourceNetworkProtocolNFS"/>
++      <ref name="diskSourceNetworkProtocolVitastor"/>
+     </choice>
+   </define>
+ 
+diff --git a/src/conf/storage_conf.c b/src/conf/storage_conf.c
+index 1dc9365bf2..a8a736be81 100644
+--- a/src/conf/storage_conf.c
++++ b/src/conf/storage_conf.c
+@@ -56,7 +56,7 @@ VIR_ENUM_IMPL(virStoragePool,
+               "logical", "disk", "iscsi",
+               "iscsi-direct", "scsi", "mpath",
+               "rbd", "sheepdog", "gluster",
+-              "zfs", "vstorage",
++              "zfs", "vstorage", "vitastor",
+ );
+ 
+ VIR_ENUM_IMPL(virStoragePoolFormatFileSystem,
+@@ -242,6 +242,18 @@ static virStoragePoolTypeInfo poolTypeInfo[] = {
+           .formatToString = virStorageFileFormatTypeToString,
+       }
+     },
++    {.poolType = VIR_STORAGE_POOL_VITASTOR,
++     .poolOptions = {
++         .flags = (VIR_STORAGE_POOL_SOURCE_HOST |
++                   VIR_STORAGE_POOL_SOURCE_NETWORK |
++                   VIR_STORAGE_POOL_SOURCE_NAME),
++      },
++      .volOptions = {
++          .defaultFormat = VIR_STORAGE_FILE_RAW,
++          .formatFromString = virStorageVolumeFormatFromString,
++          .formatToString = virStorageFileFormatTypeToString,
++      }
++    },
+     {.poolType = VIR_STORAGE_POOL_SHEEPDOG,
+      .poolOptions = {
+          .flags = (VIR_STORAGE_POOL_SOURCE_HOST |
+@@ -538,6 +550,11 @@ virStoragePoolDefParseSource(xmlXPathContextPtr ctxt,
+                        _("element 'name' is mandatory for RBD pool"));
+         return -1;
+     }
++    if (pool_type == VIR_STORAGE_POOL_VITASTOR && source->name == NULL) {
++        virReportError(VIR_ERR_XML_ERROR, "%s",
++                       _("element 'name' is mandatory for Vitastor pool"));
++        return -1;
++    }
+ 
+     if (options->formatFromString) {
+         g_autofree char *format = NULL;
+@@ -1127,6 +1144,7 @@ virStoragePoolDefFormatBuf(virBuffer *buf,
+     /* RBD, Sheepdog, Gluster and Iscsi-direct devices are not local block devs nor
+      * files, so they don't have a target */
+     if (def->type != VIR_STORAGE_POOL_RBD &&
++        def->type != VIR_STORAGE_POOL_VITASTOR &&
+         def->type != VIR_STORAGE_POOL_SHEEPDOG &&
+         def->type != VIR_STORAGE_POOL_GLUSTER &&
+         def->type != VIR_STORAGE_POOL_ISCSI_DIRECT) {
+diff --git a/src/conf/storage_conf.h b/src/conf/storage_conf.h
+index fc67957cfe..720c07ef74 100644
+--- a/src/conf/storage_conf.h
++++ b/src/conf/storage_conf.h
+@@ -103,6 +103,7 @@ typedef enum {
+     VIR_STORAGE_POOL_GLUSTER,  /* Gluster device */
+     VIR_STORAGE_POOL_ZFS,      /* ZFS */
+     VIR_STORAGE_POOL_VSTORAGE, /* Virtuozzo Storage */
++    VIR_STORAGE_POOL_VITASTOR, /* Vitastor */
+ 
+     VIR_STORAGE_POOL_LAST,
+ } virStoragePoolType;
+@@ -454,6 +455,7 @@ VIR_ENUM_DECL(virStoragePartedFs);
+                  VIR_CONNECT_LIST_STORAGE_POOLS_SCSI     | \
+                  VIR_CONNECT_LIST_STORAGE_POOLS_MPATH    | \
+                  VIR_CONNECT_LIST_STORAGE_POOLS_RBD      | \
++                 VIR_CONNECT_LIST_STORAGE_POOLS_VITASTOR | \
+                  VIR_CONNECT_LIST_STORAGE_POOLS_SHEEPDOG | \
+                  VIR_CONNECT_LIST_STORAGE_POOLS_GLUSTER  | \
+                  VIR_CONNECT_LIST_STORAGE_POOLS_ZFS      | \
+diff --git a/src/conf/storage_source_conf.c b/src/conf/storage_source_conf.c
+index d7b9bdfecb..38aefd0dd4 100644
+--- a/src/conf/storage_source_conf.c
++++ b/src/conf/storage_source_conf.c
+@@ -90,6 +90,7 @@ VIR_ENUM_IMPL(virStorageNetProtocol,
+               "ssh",
+               "vxhs",
+               "nfs",
++              "vitastor",
+ );
+ 
+ 
+@@ -1317,6 +1318,7 @@ virStorageSourceNetworkDefaultPort(virStorageNetProtocol protocol)
+         case VIR_STORAGE_NET_PROTOCOL_GLUSTER:
+             return 24007;
+ 
++        case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
+         case VIR_STORAGE_NET_PROTOCOL_RBD:
+             /* we don't provide a default for RBD */
+             return 0;
+diff --git a/src/conf/storage_source_conf.h b/src/conf/storage_source_conf.h
+index 22c35d420d..f1e32ea83d 100644
+--- a/src/conf/storage_source_conf.h
++++ b/src/conf/storage_source_conf.h
+@@ -131,6 +131,7 @@ typedef enum {
+     VIR_STORAGE_NET_PROTOCOL_SSH,
+     VIR_STORAGE_NET_PROTOCOL_VXHS,
+     VIR_STORAGE_NET_PROTOCOL_NFS,
++    VIR_STORAGE_NET_PROTOCOL_VITASTOR,
+ 
+     VIR_STORAGE_NET_PROTOCOL_LAST
+ } virStorageNetProtocol;
+diff --git a/src/conf/virstorageobj.c b/src/conf/virstorageobj.c
+index 59fa5da372..4739167f5f 100644
+--- a/src/conf/virstorageobj.c
++++ b/src/conf/virstorageobj.c
+@@ -1438,6 +1438,7 @@ virStoragePoolObjSourceFindDuplicateCb(const void *payload,
+             return 1;
+         break;
+ 
++    case VIR_STORAGE_POOL_VITASTOR:
+     case VIR_STORAGE_POOL_ISCSI_DIRECT:
+     case VIR_STORAGE_POOL_RBD:
+     case VIR_STORAGE_POOL_LAST:
+@@ -1921,6 +1922,8 @@ virStoragePoolObjMatch(virStoragePoolObj *obj,
+                (obj->def->type == VIR_STORAGE_POOL_MPATH))   ||
+               (MATCH(VIR_CONNECT_LIST_STORAGE_POOLS_RBD) &&
+                (obj->def->type == VIR_STORAGE_POOL_RBD))     ||
++              (MATCH(VIR_CONNECT_LIST_STORAGE_POOLS_VITASTOR) &&
++               (obj->def->type == VIR_STORAGE_POOL_VITASTOR)) ||
+               (MATCH(VIR_CONNECT_LIST_STORAGE_POOLS_SHEEPDOG) &&
+                (obj->def->type == VIR_STORAGE_POOL_SHEEPDOG)) ||
+               (MATCH(VIR_CONNECT_LIST_STORAGE_POOLS_GLUSTER) &&
+diff --git a/src/libvirt-storage.c b/src/libvirt-storage.c
+index db7660aac4..561df34709 100644
+--- a/src/libvirt-storage.c
++++ b/src/libvirt-storage.c
+@@ -94,6 +94,7 @@ virStoragePoolGetConnect(virStoragePoolPtr pool)
+  * VIR_CONNECT_LIST_STORAGE_POOLS_SCSI
+  * VIR_CONNECT_LIST_STORAGE_POOLS_MPATH
+  * VIR_CONNECT_LIST_STORAGE_POOLS_RBD
++ * VIR_CONNECT_LIST_STORAGE_POOLS_VITASTOR
+  * VIR_CONNECT_LIST_STORAGE_POOLS_SHEEPDOG
+  * VIR_CONNECT_LIST_STORAGE_POOLS_GLUSTER
+  * VIR_CONNECT_LIST_STORAGE_POOLS_ZFS
+diff --git a/src/libxl/libxl_conf.c b/src/libxl/libxl_conf.c
+index 2b988157fa..9d0eb47b25 100644
+--- a/src/libxl/libxl_conf.c
++++ b/src/libxl/libxl_conf.c
+@@ -1069,6 +1069,7 @@ libxlMakeNetworkDiskSrcStr(virStorageSource *src,
+     case VIR_STORAGE_NET_PROTOCOL_SSH:
+     case VIR_STORAGE_NET_PROTOCOL_VXHS:
+     case VIR_STORAGE_NET_PROTOCOL_NFS:
++    case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
+     case VIR_STORAGE_NET_PROTOCOL_LAST:
+     case VIR_STORAGE_NET_PROTOCOL_NONE:
+         virReportError(VIR_ERR_NO_SUPPORT,
+diff --git a/src/libxl/xen_xl.c b/src/libxl/xen_xl.c
+index e72e7d7f44..8482c21805 100644
+--- a/src/libxl/xen_xl.c
++++ b/src/libxl/xen_xl.c
+@@ -1461,6 +1461,7 @@ xenFormatXLDiskSrcNet(virStorageSource *src)
+     case VIR_STORAGE_NET_PROTOCOL_SSH:
+     case VIR_STORAGE_NET_PROTOCOL_VXHS:
+     case VIR_STORAGE_NET_PROTOCOL_NFS:
++    case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
+     case VIR_STORAGE_NET_PROTOCOL_LAST:
+     case VIR_STORAGE_NET_PROTOCOL_NONE:
+         virReportError(VIR_ERR_NO_SUPPORT,
+diff --git a/src/qemu/qemu_block.c b/src/qemu/qemu_block.c
+index 9b43279797..459d8e8a65 100644
+--- a/src/qemu/qemu_block.c
++++ b/src/qemu/qemu_block.c
+@@ -743,6 +743,38 @@ qemuBlockStorageSourceGetRBDProps(virStorageSource *src,
+ }
+ 
+ 
++static virJSONValue *
++qemuBlockStorageSourceGetVitastorProps(virStorageSource *src)
++{
++    virJSONValue *ret = NULL;
++    virStorageNetHostDef *host;
++    size_t i;
++    g_auto(virBuffer) buf = VIR_BUFFER_INITIALIZER;
++    g_autofree char *etcd = NULL;
++
++    for (i = 0; i < src->nhosts; i++) {
++        host = src->hosts + i;
++        if ((virStorageNetHostTransport)host->transport != VIR_STORAGE_NET_HOST_TRANS_TCP) {
++            return NULL;
++        }
++        virBufferAsprintf(&buf, i > 0 ? ",%s:%u" : "%s:%u", host->name, host->port);
++    }
++    if (src->nhosts > 0) {
++        etcd = virBufferContentAndReset(&buf);
++    }
++
++    if (virJSONValueObjectAdd(&ret,
++                              "S:etcd-host", etcd,
++                              "S:etcd-prefix", src->query,
++                              "S:config-path", src->configFile,
++                              "s:image", src->path,
++                              NULL) < 0)
++        return NULL;
++
++    return ret;
++}
++
++
+ static virJSONValue *
+ qemuBlockStorageSourceGetSshProps(virStorageSource *src)
+ {
+@@ -1094,6 +1126,12 @@ qemuBlockStorageSourceGetBackendProps(virStorageSource *src,
+                 return NULL;
+             break;
+ 
++        case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
++            driver = "vitastor";
++            if (!(fileprops = qemuBlockStorageSourceGetVitastorProps(src)))
++                return NULL;
++            break;
++
+         case VIR_STORAGE_NET_PROTOCOL_SSH:
+             driver = "ssh";
+             if (!(fileprops = qemuBlockStorageSourceGetSshProps(src)))
+@@ -1997,6 +2035,7 @@ qemuBlockGetBackingStoreString(virStorageSource *src,
+ 
+             case VIR_STORAGE_NET_PROTOCOL_SHEEPDOG:
+             case VIR_STORAGE_NET_PROTOCOL_RBD:
++            case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
+             case VIR_STORAGE_NET_PROTOCOL_VXHS:
+             case VIR_STORAGE_NET_PROTOCOL_NFS:
+             case VIR_STORAGE_NET_PROTOCOL_SSH:
+@@ -2377,6 +2416,12 @@ qemuBlockStorageSourceCreateGetStorageProps(virStorageSource *src,
+                 return -1;
+             break;
+ 
++        case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
++            driver = "vitastor";
++            if (!(location = qemuBlockStorageSourceGetVitastorProps(src)))
++                return -1;
++            break;
++
+         case VIR_STORAGE_NET_PROTOCOL_SSH:
+             if (srcPriv->nbdkitProcess) {
+                 /* disk creation not yet supported with nbdkit, and even if it
+diff --git a/src/qemu/qemu_domain.c b/src/qemu/qemu_domain.c
+index ac56fc7cb4..9e407b4aab 100644
+--- a/src/qemu/qemu_domain.c
++++ b/src/qemu/qemu_domain.c
+@@ -4677,7 +4677,8 @@ qemuDomainValidateStorageSource(virStorageSource *src,
+     if (src->query &&
+         (actualType != VIR_STORAGE_TYPE_NETWORK ||
+          (src->protocol != VIR_STORAGE_NET_PROTOCOL_HTTPS &&
+-          src->protocol != VIR_STORAGE_NET_PROTOCOL_HTTP))) {
++          src->protocol != VIR_STORAGE_NET_PROTOCOL_HTTP &&
++          src->protocol != VIR_STORAGE_NET_PROTOCOL_VITASTOR))) {
+         virReportError(VIR_ERR_CONFIG_UNSUPPORTED, "%s",
+                        _("query is supported only with HTTP(S) protocols"));
+         return -1;
+@@ -9103,6 +9104,7 @@ qemuDomainPrepareStorageSourceTLS(virStorageSource *src,
+         break;
+ 
+     case VIR_STORAGE_NET_PROTOCOL_RBD:
++    case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
+     case VIR_STORAGE_NET_PROTOCOL_SHEEPDOG:
+     case VIR_STORAGE_NET_PROTOCOL_GLUSTER:
+     case VIR_STORAGE_NET_PROTOCOL_ISCSI:
+diff --git a/src/qemu/qemu_snapshot.c b/src/qemu/qemu_snapshot.c
+index e738afffc3..37d64f469b 100644
+--- a/src/qemu/qemu_snapshot.c
++++ b/src/qemu/qemu_snapshot.c
+@@ -665,6 +665,7 @@ qemuSnapshotPrepareDiskExternalInactive(virDomainSnapshotDiskDef *snapdisk,
+         case VIR_STORAGE_NET_PROTOCOL_NONE:
+         case VIR_STORAGE_NET_PROTOCOL_NBD:
+         case VIR_STORAGE_NET_PROTOCOL_RBD:
++        case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
+         case VIR_STORAGE_NET_PROTOCOL_SHEEPDOG:
+         case VIR_STORAGE_NET_PROTOCOL_GLUSTER:
+         case VIR_STORAGE_NET_PROTOCOL_ISCSI:
+@@ -893,6 +894,7 @@ qemuSnapshotPrepareDiskInternal(virDomainDiskDef *disk,
+         case VIR_STORAGE_NET_PROTOCOL_NONE:
+         case VIR_STORAGE_NET_PROTOCOL_NBD:
+         case VIR_STORAGE_NET_PROTOCOL_RBD:
++        case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
+         case VIR_STORAGE_NET_PROTOCOL_SHEEPDOG:
+         case VIR_STORAGE_NET_PROTOCOL_GLUSTER:
+         case VIR_STORAGE_NET_PROTOCOL_ISCSI:
+diff --git a/src/storage/storage_driver.c b/src/storage/storage_driver.c
+index e19e032427..59f91f4710 100644
+--- a/src/storage/storage_driver.c
++++ b/src/storage/storage_driver.c
+@@ -1626,6 +1626,7 @@ storageVolLookupByPathCallback(virStoragePoolObj *obj,
+ 
+         case VIR_STORAGE_POOL_GLUSTER:
+         case VIR_STORAGE_POOL_RBD:
++        case VIR_STORAGE_POOL_VITASTOR:
+         case VIR_STORAGE_POOL_SHEEPDOG:
+         case VIR_STORAGE_POOL_ZFS:
+         case VIR_STORAGE_POOL_LAST:
+diff --git a/src/storage_file/storage_source_backingstore.c b/src/storage_file/storage_source_backingstore.c
+index 821378883c..2211f6891b 100644
+--- a/src/storage_file/storage_source_backingstore.c
++++ b/src/storage_file/storage_source_backingstore.c
+@@ -264,6 +264,75 @@ virStorageSourceParseRBDColonString(const char *rbdstr,
+ }
+ 
+ 
++static int
++virStorageSourceParseVitastorColonString(const char *colonstr,
++                                         virStorageSource *src)
++{
++    char *p, *e, *next;
++    g_autofree char *options = NULL;
++
++    /* optionally skip the "vitastor:" prefix if provided */
++    if (STRPREFIX(colonstr, "vitastor:"))
++        colonstr += strlen("vitastor:");
++
++    options = g_strdup(colonstr);
++
++    p = options;
++    while (*p) {
++        /* find : delimiter or end of string */
++        for (e = p; *e && *e != ':'; ++e) {
++            if (*e == '\\') {
++                e++;
++                if (*e == '\0')
++                    break;
++            }
++        }
++        if (*e == '\0') {
++            next = e;    /* last kv pair */
++        } else {
++            next = e + 1;
++            *e = '\0';
++        }
++
++        if (STRPREFIX(p, "image=")) {
++            src->path = g_strdup(p + strlen("image="));
++        } else if (STRPREFIX(p, "etcd-prefix=")) {
++            src->query = g_strdup(p + strlen("etcd-prefix="));
++        } else if (STRPREFIX(p, "config-path=")) {
++            src->configFile = g_strdup(p + strlen("config-path="));
++        } else if (STRPREFIX(p, "etcd-host=")) {
++            char *h, *sep;
++
++            h = p + strlen("etcd-host=");
++            while (h < e) {
++                for (sep = h; sep < e; ++sep) {
++                    if (*sep == '\\' && (sep[1] == ',' ||
++                                         sep[1] == ';' ||
++                                         sep[1] == ' ')) {
++                        *sep = '\0';
++                        sep += 2;
++                        break;
++                    }
++                }
++
++                if (virStorageSourceRBDAddHost(src, h) < 0)
++                    return -1;
++
++                h = sep;
++            }
++        }
++
++        p = next;
++    }
++
++    if (!src->path) {
++        return -1;
++    }
++
++    return 0;
++}
++
++
+ static int
+ virStorageSourceParseNBDColonString(const char *nbdstr,
+                                     virStorageSource *src)
+@@ -379,6 +448,11 @@ virStorageSourceParseBackingColon(virStorageSource *src,
+             return -1;
+         break;
+ 
++    case VIR_STORAGE_NET_PROTOCOL_VITASTOR:
++        if (virStorageSourceParseVitastorColonString(path, src) < 0)
++            return -1;
++        break;
++
+     case VIR_STORAGE_NET_PROTOCOL_SHEEPDOG:
+     case VIR_STORAGE_NET_PROTOCOL_LAST:
+     case VIR_STORAGE_NET_PROTOCOL_NONE:
+@@ -953,6 +1027,54 @@ virStorageSourceParseBackingJSONRBD(virStorageSource *src,
+     return 0;
+ }
+ 
++static int
++virStorageSourceParseBackingJSONVitastor(virStorageSource *src,
++                                         virJSONValue *json,
++                                         const char *jsonstr G_GNUC_UNUSED,
++                                         int opaque G_GNUC_UNUSED)
++{
++    const char *filename;
++    const char *image = virJSONValueObjectGetString(json, "image");
++    const char *conf = virJSONValueObjectGetString(json, "config-path");
++    const char *etcd_prefix = virJSONValueObjectGetString(json, "etcd-prefix");
++    virJSONValue *servers = virJSONValueObjectGetArray(json, "server");
++    size_t nservers;
++    size_t i;
++
++    src->type = VIR_STORAGE_TYPE_NETWORK;
++    src->protocol = VIR_STORAGE_NET_PROTOCOL_VITASTOR;
++
++    /* legacy syntax passed via 'filename' option */
++    if ((filename = virJSONValueObjectGetString(json, "filename")))
++        return virStorageSourceParseVitastorColonString(filename, src);
++
++    if (!image) {
++        virReportError(VIR_ERR_INVALID_ARG, "%s",
++                       _("missing image name in Vitastor backing volume "
++                         "JSON specification"));
++        return -1;
++    }
++
++    src->path = g_strdup(image);
++    src->configFile = g_strdup(conf);
++    src->query = g_strdup(etcd_prefix);
++
++    if (servers) {
++        nservers = virJSONValueArraySize(servers);
++
++        src->hosts = g_new0(virStorageNetHostDef, nservers);
++        src->nhosts = nservers;
++
++        for (i = 0; i < nservers; i++) {
++            if (virStorageSourceParseBackingJSONInetSocketAddress(src->hosts + i,
++                                                                  virJSONValueArrayGet(servers, i)) < 0)
++                return -1;
++        }
++    }
++
++    return 0;
++}
++
+ static int
+ virStorageSourceParseBackingJSONRaw(virStorageSource *src,
+                                     virJSONValue *json,
+@@ -1130,6 +1252,7 @@ static const struct virStorageSourceJSONDriverParser jsonParsers[] = {
+     {"sheepdog", false, virStorageSourceParseBackingJSONSheepdog, 0},
+     {"ssh", false, virStorageSourceParseBackingJSONSSH, 0},
+     {"rbd", false, virStorageSourceParseBackingJSONRBD, 0},
++    {"vitastor", false, virStorageSourceParseBackingJSONVitastor, 0},
+     {"raw", true, virStorageSourceParseBackingJSONRaw, 0},
+     {"nfs", false, virStorageSourceParseBackingJSONNFS, 0},
+     {"vxhs", false, virStorageSourceParseBackingJSONVxHS, 0},
+diff --git a/src/test/test_driver.c b/src/test/test_driver.c
+index 1165689de7..bba846351c 100644
+--- a/src/test/test_driver.c
++++ b/src/test/test_driver.c
+@@ -7345,6 +7345,7 @@ testStorageVolumeTypeForPool(int pooltype)
+     case VIR_STORAGE_POOL_ISCSI_DIRECT:
+     case VIR_STORAGE_POOL_GLUSTER:
+     case VIR_STORAGE_POOL_RBD:
++    case VIR_STORAGE_POOL_VITASTOR:
+         return VIR_STORAGE_VOL_NETWORK;
+     case VIR_STORAGE_POOL_LOGICAL:
+     case VIR_STORAGE_POOL_DISK:
+diff --git a/tests/storagepoolcapsschemadata/poolcaps-fs.xml b/tests/storagepoolcapsschemadata/poolcaps-fs.xml
+index eee75af746..8bd0a57bdd 100644
+--- a/tests/storagepoolcapsschemadata/poolcaps-fs.xml
++++ b/tests/storagepoolcapsschemadata/poolcaps-fs.xml
+@@ -204,4 +204,11 @@
+       </enum>
+     </volOptions>
+   </pool>
++  <pool type='vitastor' supported='no'>
++    <volOptions>
++      <defaultFormat type='raw'/>
++      <enum name='targetFormatType'>
++      </enum>
++    </volOptions>
++  </pool>
+ </storagepoolCapabilities>
+diff --git a/tests/storagepoolcapsschemadata/poolcaps-full.xml b/tests/storagepoolcapsschemadata/poolcaps-full.xml
+index 805950a937..852df0de16 100644
+--- a/tests/storagepoolcapsschemadata/poolcaps-full.xml
++++ b/tests/storagepoolcapsschemadata/poolcaps-full.xml
+@@ -204,4 +204,11 @@
+       </enum>
+     </volOptions>
+   </pool>
++  <pool type='vitastor' supported='yes'>
++    <volOptions>
++      <defaultFormat type='raw'/>
++      <enum name='targetFormatType'>
++      </enum>
++    </volOptions>
++  </pool>
+ </storagepoolCapabilities>
+diff --git a/tests/storagepoolxml2argvtest.c b/tests/storagepoolxml2argvtest.c
+index d5c2531ab8..b19308ac38 100644
+--- a/tests/storagepoolxml2argvtest.c
++++ b/tests/storagepoolxml2argvtest.c
+@@ -57,6 +57,7 @@ testCompareXMLToArgvFiles(bool shouldFail,
+     case VIR_STORAGE_POOL_GLUSTER:
+     case VIR_STORAGE_POOL_ZFS:
+     case VIR_STORAGE_POOL_VSTORAGE:
++    case VIR_STORAGE_POOL_VITASTOR:
+     case VIR_STORAGE_POOL_LAST:
+     default:
+         VIR_TEST_DEBUG("pool type '%s' has no xml2argv test", defTypeStr);
+diff --git a/tools/virsh-pool.c b/tools/virsh-pool.c
+index 2010ef1356..072e2ff9e8 100644
+--- a/tools/virsh-pool.c
++++ b/tools/virsh-pool.c
+@@ -1187,6 +1187,9 @@ cmdPoolList(vshControl *ctl, const vshCmd *cmd G_GNUC_UNUSED)
+             case VIR_STORAGE_POOL_VSTORAGE:
+                 flags |= VIR_CONNECT_LIST_STORAGE_POOLS_VSTORAGE;
+                 break;
++            case VIR_STORAGE_POOL_VITASTOR:
++                flags |= VIR_CONNECT_LIST_STORAGE_POOLS_VITASTOR;
++                break;
+             case VIR_STORAGE_POOL_LAST:
+                 break;
+             }

patches/qemu-10.1-vitastor.patch

@@ -1,29 +1,172 @@
-diff --git a/src/client/qemu_driver.c b/src/client/qemu_driver.c
-index d8356dab..5f4cd50d 100644
---- a/src/client/qemu_driver.c
-+++ b/src/client/qemu_driver.c
-@@ -974,14 +974,21 @@ static void vitastor_co_read_bitmap_cb(void *opaque, long retval, uint8_t *bitma
- #endif
- }
+diff --git a/block/meson.build b/block/meson.build
+index 34b1b2a306..24ca0f1e52 100644
+--- a/block/meson.build
++++ b/block/meson.build
+@@ -114,6 +114,7 @@ foreach m : [
+   [libnfs, 'nfs', files('nfs.c')],
+   [libssh, 'ssh', files('ssh.c')],
+   [rbd, 'rbd', files('rbd.c')],
++  [vitastor, 'vitastor', files('vitastor.c')],
+ ]
+   if m[0].found()
+     module_ss = ss.source_set()
+diff --git a/meson.build b/meson.build
+index 50c774a195..e5c7a3a4b1 100644
+--- a/meson.build
++++ b/meson.build
+@@ -1652,6 +1652,26 @@ if not get_option('rbd').auto() or have_block
+   endif
+ endif
  
--static int coroutine_fn vitastor_co_block_status(
--    BlockDriverState *bs, bool want_zero, int64_t offset, int64_t bytes,
--    int64_t *pnum, int64_t *map, BlockDriverState **file)
-+static int coroutine_fn vitastor_co_block_status(BlockDriverState *bs,
-+#if QEMU_VERSION_MAJOR > 10 || QEMU_VERSION_MAJOR == 10 && QEMU_VERSION_MINOR >= 1
-+    unsigned int mode,
-+#else
-+    bool want_zero,
-+#endif
-+    int64_t offset, int64_t bytes, int64_t *pnum, int64_t *map, BlockDriverState **file)
- {
-     // Allocated => return BDRV_BLOCK_DATA|BDRV_BLOCK_OFFSET_VALID
-     // Not allocated => return 0
-     // Error => return -errno
-     // Set pnum to length of the extent, `*map` = `offset`, `*file` = `bs`
-+#if QEMU_VERSION_MAJOR > 10 || QEMU_VERSION_MAJOR == 10 && QEMU_VERSION_MINOR >= 1
-+    int want_zero = (mode == BDRV_WANT_PRECISE);
-+#endif
-     VitastorRPC task;
-     VitastorClient *client = bs->opaque;
-     uint64_t inode = client->watch ? vitastor_c_inode_get_num(client->watch) : client->inode;
++vitastor = not_found
++if not get_option('vitastor').auto() or have_block
++  libvitastor_client = cc.find_library('vitastor_client', has_headers: ['vitastor_c.h'],
++    required: get_option('vitastor'))
++  if libvitastor_client.found()
++    if cc.links('''
++      #include <vitastor_c.h>
++      int main(void) {
++        vitastor_c_create_qemu(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0);
++        return 0;
++      }''', dependencies: libvitastor_client)
++      vitastor = declare_dependency(dependencies: libvitastor_client)
++    elif get_option('vitastor').enabled()
++      error('could not link libvitastor_client')
++    else
++      warning('could not link libvitastor_client, disabling')
++    endif
++  endif
++endif
++
+ glusterfs = not_found
+ glusterfs_ftruncate_has_stat = false
+ glusterfs_iocb_has_stat = false
+@@ -2547,6 +2567,7 @@ endif
+ config_host_data.set('CONFIG_OPENGL', opengl.found())
+ config_host_data.set('CONFIG_PLUGIN', get_option('plugins'))
+ config_host_data.set('CONFIG_RBD', rbd.found())
++config_host_data.set('CONFIG_VITASTOR', vitastor.found())
+ config_host_data.set('CONFIG_RDMA', rdma.found())
+ config_host_data.set('CONFIG_RELOCATABLE', get_option('relocatable'))
+ config_host_data.set('CONFIG_SAFESTACK', get_option('safe_stack'))
+@@ -4972,6 +4993,7 @@ summary_info += {'fdt support':       fdt_opt == 'internal' ? 'internal' : fdt}
+ summary_info += {'libcap-ng support': libcap_ng}
+ summary_info += {'bpf support':       libbpf}
+ summary_info += {'rbd support':       rbd}
++summary_info += {'vitastor support':  vitastor}
+ summary_info += {'smartcard support': cacard}
+ summary_info += {'U2F support':       u2f}
+ summary_info += {'libusb':            libusb}
+diff --git a/meson_options.txt b/meson_options.txt
+index fff1521e58..f0844c0e00 100644
+--- a/meson_options.txt
++++ b/meson_options.txt
+@@ -202,6 +202,8 @@ option('pvg', type: 'feature', value: 'auto',
+        description: 'macOS paravirtualized graphics support')
+ option('rbd', type : 'feature', value : 'auto',
+        description: 'Ceph block device driver')
++option('vitastor', type : 'feature', value : 'auto',
++       description: 'Vitastor block device driver')
+ option('opengl', type : 'feature', value : 'auto',
+        description: 'OpenGL support')
+ option('rdma', type : 'feature', value : 'auto',
+diff --git a/qapi/block-core.json b/qapi/block-core.json
+index dc6eb4ae23..d043f4340e 100644
+--- a/qapi/block-core.json
++++ b/qapi/block-core.json
+@@ -3280,7 +3280,7 @@
+             'parallels', 'preallocate', 'qcow', 'qcow2', 'qed', 'quorum',
+             'raw', 'rbd',
+             { 'name': 'replication', 'if': 'CONFIG_REPLICATION' },
+-            'ssh', 'throttle', 'vdi', 'vhdx',
++            'ssh', 'throttle', 'vdi', 'vhdx', 'vitastor',
+             { 'name': 'virtio-blk-vfio-pci', 'if': 'CONFIG_BLKIO' },
+             { 'name': 'virtio-blk-vhost-user', 'if': 'CONFIG_BLKIO' },
+             { 'name': 'virtio-blk-vhost-vdpa', 'if': 'CONFIG_BLKIO' },
+@@ -4363,6 +4363,28 @@
+             '*key-secret': 'str',
+             '*server': ['InetSocketAddressBase'] } }
+ 
++##
++# @BlockdevOptionsVitastor:
++#
++# Driver specific block device options for vitastor
++#
++# @image:       Image name
++# @inode:       Inode number
++# @pool:        Pool ID
++# @size:        Desired image size in bytes
++# @config-path: Path to Vitastor configuration
++# @etcd-host:   etcd connection address(es)
++# @etcd-prefix: etcd key/value prefix
++##
++{ 'struct': 'BlockdevOptionsVitastor',
++  'data': { '*inode': 'uint64',
++            '*pool': 'uint64',
++            '*size': 'uint64',
++            '*image': 'str',
++            '*config-path': 'str',
++            '*etcd-host': 'str',
++            '*etcd-prefix': 'str' } }
++
+ ##
+ # @ReplicationMode:
+ #
+@@ -4831,6 +4853,7 @@
+       'throttle':   'BlockdevOptionsThrottle',
+       'vdi':        'BlockdevOptionsGenericFormat',
+       'vhdx':       'BlockdevOptionsGenericFormat',
++      'vitastor':   'BlockdevOptionsVitastor',
+       'virtio-blk-vfio-pci':
+                     { 'type': 'BlockdevOptionsVirtioBlkVfioPci',
+                       'if': 'CONFIG_BLKIO' },
+@@ -5304,6 +5327,20 @@
+             '*cluster-size' :   'size',
+             '*encrypt' :        'RbdEncryptionCreateOptions' } }
+ 
++##
++# @BlockdevCreateOptionsVitastor:
++#
++# Driver specific image creation options for Vitastor.
++#
++# @location: Where to store the new image file.  This location cannot
++#     point to a snapshot.
++#
++# @size: Size of the virtual disk in bytes
++##
++{ 'struct': 'BlockdevCreateOptionsVitastor',
++  'data': { 'location':         'BlockdevOptionsVitastor',
++            'size':             'size' } }
++
+ ##
+ # @BlockdevVmdkSubformat:
+ #
+@@ -5526,6 +5563,7 @@
+       'ssh':            'BlockdevCreateOptionsSsh',
+       'vdi':            'BlockdevCreateOptionsVdi',
+       'vhdx':           'BlockdevCreateOptionsVhdx',
++      'vitastor':       'BlockdevCreateOptionsVitastor',
+       'vmdk':           'BlockdevCreateOptionsVmdk',
+       'vpc':            'BlockdevCreateOptionsVpc'
+   } }
+diff --git a/scripts/meson-buildoptions.sh b/scripts/meson-buildoptions.sh
+index 0ebe6bc52a..2c37ad3892 100644
+--- a/scripts/meson-buildoptions.sh
++++ b/scripts/meson-buildoptions.sh
+@@ -175,6 +175,7 @@ meson_options_help() {
+   printf "%s\n" '  qga-vss         build QGA VSS support (broken with MinGW)'
+   printf "%s\n" '  qpl             Query Processing Library support'
+   printf "%s\n" '  rbd             Ceph block device driver'
++  printf "%s\n" '  vitastor        Vitastor block device driver'
+   printf "%s\n" '  rdma            Enable RDMA-based migration'
+   printf "%s\n" '  replication     replication support'
+   printf "%s\n" '  rust            Rust support'
+@@ -459,6 +460,8 @@ _meson_option_parse() {
+     --disable-qpl) printf "%s" -Dqpl=disabled ;;
+     --enable-rbd) printf "%s" -Drbd=enabled ;;
+     --disable-rbd) printf "%s" -Drbd=disabled ;;
++    --enable-vitastor) printf "%s" -Dvitastor=enabled ;;
++    --disable-vitastor) printf "%s" -Dvitastor=disabled ;;
+     --enable-rdma) printf "%s" -Drdma=enabled ;;
+     --disable-rdma) printf "%s" -Drdma=disabled ;;
+     --enable-relocatable) printf "%s" -Drelocatable=true ;;

patches/qemu-10.2-vitastor.patch

@@ -0,0 +1,172 @@
+diff --git a/block/meson.build b/block/meson.build
+index 34b1b2a306..24ca0f1e52 100644
+--- a/block/meson.build
++++ b/block/meson.build
+@@ -114,6 +114,7 @@ foreach m : [
+   [libnfs, 'nfs', files('nfs.c')],
+   [libssh, 'ssh', files('ssh.c')],
+   [rbd, 'rbd', files('rbd.c')],
++  [vitastor, 'vitastor', files('vitastor.c')],
+ ]
+   if m[0].found()
+     module_ss = ss.source_set()
+diff --git a/meson.build b/meson.build
+index d9293294d8..776a5becc6 100644
+--- a/meson.build
++++ b/meson.build
+@@ -1665,6 +1665,26 @@ if not get_option('rbd').auto() or have_block
+   endif
+ endif
+ 
++vitastor = not_found
++if not get_option('vitastor').auto() or have_block
++  libvitastor_client = cc.find_library('vitastor_client', has_headers: ['vitastor_c.h'],
++    required: get_option('vitastor'))
++  if libvitastor_client.found()
++    if cc.links('''
++      #include <vitastor_c.h>
++      int main(void) {
++        vitastor_c_create_qemu(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0);
++        return 0;
++      }''', dependencies: libvitastor_client)
++      vitastor = declare_dependency(dependencies: libvitastor_client)
++    elif get_option('vitastor').enabled()
++      error('could not link libvitastor_client')
++    else
++      warning('could not link libvitastor_client, disabling')
++    endif
++  endif
++endif
++
+ glusterfs = not_found
+ glusterfs_ftruncate_has_stat = false
+ glusterfs_iocb_has_stat = false
+@@ -2509,6 +2529,7 @@ endif
+ config_host_data.set('CONFIG_OPENGL', opengl.found())
+ config_host_data.set('CONFIG_PLUGIN', get_option('plugins'))
+ config_host_data.set('CONFIG_RBD', rbd.found())
++config_host_data.set('CONFIG_VITASTOR', vitastor.found())
+ config_host_data.set('CONFIG_RDMA', rdma.found())
+ config_host_data.set('CONFIG_RELOCATABLE', get_option('relocatable'))
+ config_host_data.set('CONFIG_SAFESTACK', get_option('safe_stack'))
+@@ -4948,6 +4969,7 @@ summary_info += {'fdt support':       fdt_opt == 'internal' ? 'internal' : fdt}
+ summary_info += {'libcap-ng support': libcap_ng}
+ summary_info += {'bpf support':       libbpf}
+ summary_info += {'rbd support':       rbd}
++summary_info += {'vitastor support':  vitastor}
+ summary_info += {'smartcard support': cacard}
+ summary_info += {'U2F support':       u2f}
+ summary_info += {'libusb':            libusb}
+diff --git a/meson_options.txt b/meson_options.txt
+index 2836156257..148086cc6f 100644
+--- a/meson_options.txt
++++ b/meson_options.txt
+@@ -206,6 +206,8 @@ option('pvg', type: 'feature', value: 'auto',
+        description: 'macOS paravirtualized graphics support')
+ option('rbd', type : 'feature', value : 'auto',
+        description: 'Ceph block device driver')
++option('vitastor', type : 'feature', value : 'auto',
++       description: 'Vitastor block device driver')
+ option('opengl', type : 'feature', value : 'auto',
+        description: 'OpenGL support')
+ option('rdma', type : 'feature', value : 'auto',
+diff --git a/qapi/block-core.json b/qapi/block-core.json
+index b82af74256..f25a6f5ce8 100644
+--- a/qapi/block-core.json
++++ b/qapi/block-core.json
+@@ -3351,7 +3351,7 @@
+             'parallels', 'preallocate', 'qcow', 'qcow2', 'qed', 'quorum',
+             'raw', 'rbd',
+             { 'name': 'replication', 'if': 'CONFIG_REPLICATION' },
+-            'ssh', 'throttle', 'vdi', 'vhdx',
++            'ssh', 'throttle', 'vdi', 'vhdx', 'vitastor',
+             { 'name': 'virtio-blk-vfio-pci', 'if': 'CONFIG_BLKIO' },
+             { 'name': 'virtio-blk-vhost-user', 'if': 'CONFIG_BLKIO' },
+             { 'name': 'virtio-blk-vhost-vdpa', 'if': 'CONFIG_BLKIO' },
+@@ -4434,6 +4434,28 @@
+             '*key-secret': 'str',
+             '*server': ['InetSocketAddressBase'] } }
+ 
++##
++# @BlockdevOptionsVitastor:
++#
++# Driver specific block device options for vitastor
++#
++# @image:       Image name
++# @inode:       Inode number
++# @pool:        Pool ID
++# @size:        Desired image size in bytes
++# @config-path: Path to Vitastor configuration
++# @etcd-host:   etcd connection address(es)
++# @etcd-prefix: etcd key/value prefix
++##
++{ 'struct': 'BlockdevOptionsVitastor',
++  'data': { '*inode': 'uint64',
++            '*pool': 'uint64',
++            '*size': 'uint64',
++            '*image': 'str',
++            '*config-path': 'str',
++            '*etcd-host': 'str',
++            '*etcd-prefix': 'str' } }
++
+ ##
+ # @ReplicationMode:
+ #
+@@ -4902,6 +4924,7 @@
+       'throttle':   'BlockdevOptionsThrottle',
+       'vdi':        'BlockdevOptionsGenericFormat',
+       'vhdx':       'BlockdevOptionsGenericFormat',
++      'vitastor':   'BlockdevOptionsVitastor',
+       'virtio-blk-vfio-pci':
+                     { 'type': 'BlockdevOptionsVirtioBlkVfioPci',
+                       'if': 'CONFIG_BLKIO' },
+@@ -5376,6 +5399,20 @@
+             '*cluster-size' :   'size',
+             '*encrypt' :        'RbdEncryptionCreateOptions' } }
+ 
++##
++# @BlockdevCreateOptionsVitastor:
++#
++# Driver specific image creation options for Vitastor.
++#
++# @location: Where to store the new image file.  This location cannot
++#     point to a snapshot.
++#
++# @size: Size of the virtual disk in bytes
++##
++{ 'struct': 'BlockdevCreateOptionsVitastor',
++  'data': { 'location':         'BlockdevOptionsVitastor',
++            'size':             'size' } }
++
+ ##
+ # @BlockdevVmdkSubformat:
+ #
+@@ -5598,6 +5635,7 @@
+       'ssh':            'BlockdevCreateOptionsSsh',
+       'vdi':            'BlockdevCreateOptionsVdi',
+       'vhdx':           'BlockdevCreateOptionsVhdx',
++      'vitastor':       'BlockdevCreateOptionsVitastor',
+       'vmdk':           'BlockdevCreateOptionsVmdk',
+       'vpc':            'BlockdevCreateOptionsVpc'
+   } }
+diff --git a/scripts/meson-buildoptions.sh b/scripts/meson-buildoptions.sh
+index 3d0d132344..65ee8c855e 100644
+--- a/scripts/meson-buildoptions.sh
++++ b/scripts/meson-buildoptions.sh
+@@ -177,6 +177,7 @@ meson_options_help() {
+   printf "%s\n" '  qga-vss         build QGA VSS support (broken with MinGW)'
+   printf "%s\n" '  qpl             Query Processing Library support'
+   printf "%s\n" '  rbd             Ceph block device driver'
++  printf "%s\n" '  vitastor        Vitastor block device driver'
+   printf "%s\n" '  rdma            Enable RDMA-based migration'
+   printf "%s\n" '  replication     replication support'
+   printf "%s\n" '  rust            Rust support'
+@@ -464,6 +465,8 @@ _meson_option_parse() {
+     --disable-qpl) printf "%s" -Dqpl=disabled ;;
+     --enable-rbd) printf "%s" -Drbd=enabled ;;
+     --disable-rbd) printf "%s" -Drbd=disabled ;;
++    --enable-vitastor) printf "%s" -Dvitastor=enabled ;;
++    --disable-vitastor) printf "%s" -Dvitastor=disabled ;;
+     --enable-rdma) printf "%s" -Drdma=enabled ;;
+     --disable-rdma) printf "%s" -Drdma=disabled ;;
+     --enable-relocatable) printf "%s" -Drelocatable=true ;;

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:        3.0.5
+Version:        3.0.9
 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-3.0.5.el10.tar.gz
+Source0:        vitastor-3.0.9.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:        3.0.5
+Version:        3.0.9
 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-3.0.5.el7.tar.gz
+Source0:        vitastor-3.0.9.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:        3.0.5
+Version:        3.0.9
 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-3.0.5.el8.tar.gz
+Source0:        vitastor-3.0.9.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:        3.0.5
+Version:        3.0.9
 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-3.0.5.el9.tar.gz
+Source0:        vitastor-3.0.9.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

@@ -1,9 +1,8 @@
-cmake_minimum_required(VERSION 2.8.12)
+cmake_minimum_required(VERSION 2.8...3.30)
 
 project(vitastor)
 
 include(GNUInstallDirs)
-include(CTest)
 include(CheckIncludeFile)
 
 find_package(PkgConfig)
@@ -21,7 +20,7 @@ if("${CMAKE_INSTALL_PREFIX}" MATCHES "^/usr/local/?$")
 endif()
 set(ENABLE_COVERAGE false CACHE BOOL "Enable code coverage")
 
-add_definitions(-DVITASTOR_VERSION="3.0.5")
+add_definitions(-DVITASTOR_VERSION="3.0.9")
 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})
@@ -75,6 +74,14 @@ if (RDMACM_LIBRARIES)
 	add_definitions(-DWITH_RDMACM)
 endif (RDMACM_LIBRARIES)
 
+find_package(OpenSSL REQUIRED)
+if (OPENSSL_FOUND)
+	add_definitions(-DWITH_OPENSSL)
+endif (OPENSSL_FOUND)
+
+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/CMakeLists.txt

@@ -1,4 +1,4 @@
-cmake_minimum_required(VERSION 2.8.12)
+cmake_minimum_required(VERSION 2.8...3.30)
 
 project(vitastor)
 

src/blockstore/blockstore.h

@@ -228,4 +228,9 @@ public:
     virtual uint64_t get_journal_size() = 0;
 
     virtual uint32_t get_bitmap_granularity() = 0;
+
+    virtual uint64_t get_live_entries() = 0;
+    virtual uint64_t get_live_memory() = 0;
+    virtual uint64_t get_garbage_entries() = 0;
+    virtual uint64_t get_garbage_memory() = 0;
 };

src/blockstore/blockstore_disk.cpp

@@ -83,17 +83,23 @@ 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() &&
         (config["discard_on_start"] == "true" || config["discard_on_start"] == "1" || config["discard_on_start"] == "yes");
+    gc_on_start = config.find("gc_on_start") == config.end() ||
+        (config["gc_on_start"] == "true" || config["gc_on_start"] == "1" || config["gc_on_start"] == "yes");
     min_discard_size = parse_size(config["min_discard_size"]);
     if (!min_discard_size)
         min_discard_size = 1024*1024;

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
@@ -57,6 +58,8 @@ struct blockstore_disk_t
     bool inmemory_journal = true;
     // Data discard granularity and minimum size (for the sake of performance)
     bool discard_on_start = false;
+    // GC on start (new store)
+    bool gc_on_start = true;
     uint64_t min_discard_size = 1024*1024;
     uint64_t discard_granularity = 0;
 

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
@@ -29,6 +30,8 @@
 #define IMAP_MALLOC_LOW_BITS ((size_t)0x0F)
 #define IMAP_MAX_LOW 16
 
+#define list_item_overhead(a) (((a) + sizeof(heap_list_item_t) - sizeof(heap_entry_t) + sizeof(void*) + 15) & ~15)
+
 void inode_map_put(void* & inode_idx, heap_list_item_t* li);
 void inode_map_get(void *inode_idx, heap_inode_map_t::iterator & li_it, heap_list_item_t* & li, uint64_t stripe);
 void inode_map_free(void* inode_idx);
@@ -62,19 +65,19 @@ 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));
 }
 
@@ -89,7 +92,7 @@ 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;
     }
@@ -212,15 +215,24 @@ 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;
@@ -244,7 +256,7 @@ blockstore_heap_t::blockstore_heap_t(blockstore_disk_t *dsk, uint8_t *buffer_are
     assert(dsk->meta_block_size < 32768);
     assert(dsk->meta_area_size > 0);
     assert(dsk->journal_len > 0);
-    meta_alloc = new multilist_index_t(meta_block_count, META_ALLOC_LEVELS+1, 0);
+    meta_alloc = new multilist_index_t(meta_block_count, META_ALLOC_LEVELS+1, 2);
     block_info.resize(meta_block_count);
     assert(dsk->block_count <= 0xFFFF0000);
     data_alloc = new allocator_t(dsk->block_count);
@@ -332,19 +344,12 @@ corrupted_block:
                     block_num, block_offset, wr->size, sizeof(heap_entry_t));
                 goto corrupted_block;
             }
-            if (wr->is_garbage())
-            {
-                // Garbage collection is only performed when writing new entries into the block
-                // because it needs a fake LSN and modified blocks require consecutive modified LSNs
-                // That's why garbage entries may persist on disk
-                if (log_level > 5)
-                {
-                    fprintf(stderr, "Notice: skipping garbage entry %jx:%jx v%ju l%ju in metadata block %u at %u\n",
-                        wr->inode, wr->stripe, wr->version, wr->lsn, block_num, block_offset);
-                }
-                block_offset += wr->size;
-                continue;
-            }
+            // Garbage collection is only performed when writing new entries into the block
+            // because it needs a fake LSN and modified blocks require consecutive modified LSNs
+            // At the same time, further modifications _after_ putting new entries into the block,
+            // but _before_ writing it, may mark some entries in it as garbage. That's why garbage
+            // entries may still be present on disk.
+            wr->entry_type &= ~BS_HEAP_GARBAGE;
             if ((wr->entry_type & BS_HEAP_TYPE) < BS_HEAP_BIG_WRITE ||
                 (wr->entry_type & BS_HEAP_TYPE) > BS_HEAP_ROLLBACK ||
                 (wr->entry_type & ~(BS_HEAP_TYPE|BS_HEAP_STABLE)) ||
@@ -385,12 +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
@@ -426,7 +431,10 @@ int blockstore_heap_t::load_blocks(uint64_t disk_offset, uint64_t size, uint8_t
     entries_loaded = 0;
     return read_blocks(disk_offset, size, buf, allow_corrupted, [&](uint32_t block_num, heap_entry_t *wr_orig)
     {
-        heap_list_item_t *li = (heap_list_item_t*)malloc_or_die(wr_orig->size + sizeof(heap_list_item_t) - sizeof(heap_entry_t));
+        auto alloc_size = wr_orig->size + sizeof(heap_list_item_t) - sizeof(heap_entry_t);
+        heap_list_item_t *li = (heap_list_item_t*)malloc_or_die(alloc_size);
+        live_entries++;
+        live_memory += list_item_overhead(wr_orig->size);
         li->block_num = block_num;
         li->prev = li->next = NULL;
         memcpy(&li->entry, wr_orig, wr_orig->size);
@@ -611,10 +619,11 @@ int blockstore_heap_t::mark_used_blocks()
                 if (wr->entry_type == (BS_HEAP_DELETE|BS_HEAP_STABLE) && !li->prev)
                 {
                     wr->set_garbage();
+                    garbage_entries++;
+                    garbage_memory += list_item_overhead(wr->size);
                     modify_alloc(li->block_num, [&](heap_block_info_t & inf)
                     {
-                        inf.used_space -= wr->size;
-                        inf.has_garbage = true;
+                        inf.garbage_space += wr->size;
                     });
                     li = NULL;
                 }
@@ -624,10 +633,11 @@ int blockstore_heap_t::mark_used_blocks()
                     if (overwritten)
                     {
                         wr->set_garbage();
+                        garbage_entries++;
+                        garbage_memory += list_item_overhead(wr->size);
                         modify_alloc(li->block_num, [&](heap_block_info_t & inf)
                         {
-                            inf.used_space -= wr->size;
-                            inf.has_garbage = true;
+                            inf.garbage_space += wr->size;
                         });
                         continue;
                     }
@@ -666,15 +676,66 @@ int blockstore_heap_t::mark_used_blocks()
             });
         }
     }
+    if (dsk->gc_on_start)
+    {
+        recheck_full_gc();
+    }
     return res;
 }
 
+void blockstore_heap_t::recheck_full_gc()
+{
+    uint32_t block_num = 0;
+    for (auto & inf: block_info)
+    {
+        // Instantly collect all garbage on restart
+        if (inf.garbage_space > 0)
+        {
+            if (log_level > 5)
+            {
+                fprintf(stderr, "Clearing %u out of %u garbage bytes in block %u\n", inf.garbage_space, inf.used_space, block_num);
+            }
+            uint32_t collected_garbage = 0;
+            size_t i = 0, j = 0;
+            for (; i < inf.entries.size(); i++)
+            {
+                if (inf.entries[i]->entry.is_garbage())
+                {
+                    collected_garbage += inf.entries[i]->entry.size;
+                    remove_list_item(inf.entries[i]);
+                }
+                else
+                {
+                    if (j != i)
+                        inf.entries[j] = inf.entries[i];
+                    j++;
+                }
+            }
+            inf.entries.resize(j);
+            modify_alloc(block_num, [&](heap_block_info_t & inf)
+            {
+                inf.used_space -= collected_garbage;
+                inf.garbage_space -= collected_garbage;
+            });
+            recheck_modified_blocks.insert(block_num);
+        }
+        block_num++;
+    }
+}
+
 void blockstore_heap_t::recheck_buffer(heap_entry_t *cwr, uint8_t *buf)
 {
     auto free_entry = [&](heap_list_item_t *li)
     {
         uint32_t block_num = li->block_num;
         auto wr_size = li->entry.size;
+        if (li->entry.is_garbage())
+        {
+            garbage_entries--;
+            garbage_memory -= list_item_overhead(wr_size);
+        }
+        live_entries--;
+        live_memory -= list_item_overhead(wr_size);
         free(li);
         modify_alloc(block_num, [&](heap_block_info_t & inf)
         {
@@ -712,7 +773,12 @@ void blockstore_heap_t::recheck_buffer(heap_entry_t *cwr, uint8_t *buf)
             assert(li->entry.entry_type == cwr->entry_type);
             auto prev = li->prev;
             li->next = li->prev = NULL;
-            li->entry.set_garbage();
+            if (!li->entry.is_garbage())
+            {
+                garbage_entries++;
+                garbage_memory += list_item_overhead(li->entry.size);
+                li->entry.set_garbage();
+            }
             li = prev;
             rolled_back++;
         }
@@ -878,7 +944,11 @@ 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;
@@ -928,11 +998,26 @@ 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;
@@ -949,42 +1034,89 @@ bool blockstore_heap_t::calc_block_checksums(uint32_t *block_csums, uint8_t *bit
                     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 (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;
 }
 
@@ -1162,7 +1294,7 @@ heap_entry_t *blockstore_heap_t::read_entry(object_id oid)
 
 void blockstore_heap_t::gc_block(heap_block_info_t & inf)
 {
-    if (inf.has_garbage)
+    if (inf.garbage_space > 0)
     {
         size_t i = 0, j = 0;
         for (; i < inf.entries.size(); i++)
@@ -1185,7 +1317,8 @@ void blockstore_heap_t::gc_block(heap_block_info_t & inf)
             }
         }
         inf.entries.resize(j);
-        inf.has_garbage = false;
+        inf.used_space -= inf.garbage_space;
+        inf.garbage_space = 0;
     }
 }
 
@@ -1195,11 +1328,12 @@ int blockstore_heap_t::allocate_entry(uint32_t entry_size, uint32_t *block_num,
     {
         // First try to write into the same block as the previous time
         auto & inf = block_info.at(last_allocated_block);
-        auto free_space = dsk->meta_block_size - inf.used_space;
-        if (inf.is_writing || free_space < entry_size ||
+        if (inf.is_writing ||
+            inf.used_space - inf.garbage_space + entry_size > dsk->meta_block_size ||
             // Do not allow to make the last non-nearfull block nearfull
             !allow_last_free && meta_nearfull_blocks >= meta_block_count-1 &&
-            free_space >= max_entry_size && free_space < max_entry_size+entry_size)
+            inf.used_space - inf.garbage_space <= dsk->meta_block_size-max_entry_size &&
+            inf.used_space - inf.garbage_space + entry_size > dsk->meta_block_size-max_entry_size)
         {
             last_allocated_block = UINT32_MAX;
         }
@@ -1216,8 +1350,8 @@ int blockstore_heap_t::allocate_entry(uint32_t entry_size, uint32_t *block_num,
         {
             // Do not allow to make the last non-nearfull block nearfull
             auto & inf = block_info.at(last_allocated_block);
-            auto free_space = dsk->meta_block_size - inf.used_space;
-            if (free_space >= max_entry_size && free_space < max_entry_size+entry_size)
+            if (inf.used_space - inf.garbage_space <= dsk->meta_block_size-max_entry_size &&
+                inf.used_space - inf.garbage_space + entry_size > dsk->meta_block_size-max_entry_size)
             {
                 last_allocated_block = UINT32_MAX;
             }
@@ -1228,8 +1362,7 @@ int blockstore_heap_t::allocate_entry(uint32_t entry_size, uint32_t *block_num,
             for (uint32_t b = meta_alloc->find(META_ALLOC_LEVELS-1); b != UINT32_MAX; b = meta_alloc->next(b))
             {
                 auto & inf = block_info.at(b);
-                auto free_space = dsk->meta_block_size - inf.used_space;
-                if (free_space >= entry_size)
+                if (inf.used_space - inf.garbage_space + entry_size <= dsk->meta_block_size)
                 {
                     last_allocated_block = b;
                     break;
@@ -1246,20 +1379,27 @@ int blockstore_heap_t::allocate_entry(uint32_t entry_size, uint32_t *block_num,
     {
         // Do not allow to make the last non-nearfull block nearfull
         auto & inf = block_info.at(last_allocated_block);
-        if (dsk->meta_block_size-inf.used_space >= max_entry_size &&
-            dsk->meta_block_size-inf.used_space+entry_size < max_entry_size)
+        if (inf.used_space - inf.garbage_space <= dsk->meta_block_size-max_entry_size &&
+            inf.used_space - inf.garbage_space + entry_size > dsk->meta_block_size-max_entry_size)
         {
             last_allocated_block = UINT32_MAX;
             return ENOSPC;
         }
     }
     // Write into the same block
-    auto & inf = block_info.at(last_allocated_block);
-    gc_block(inf);
     *block_num = last_allocated_block;
     modify_alloc(last_allocated_block, [&](heap_block_info_t & inf)
     {
+        // Write just 1 entry to the block to collect garbage
+        if (inf.garbage_space > (inf.used_space-inf.garbage_space)/2)
+            last_allocated_block = UINT32_MAX;
+        gc_block(inf);
         inf.used_space += entry_size;
+        assert(inf.used_space - inf.garbage_space <= dsk->meta_block_size);
+        assert(!inf.mod_lsn_to || inf.mod_lsn_to == next_lsn);
+        ++next_lsn;
+        inf.mod_lsn = inf.mod_lsn ? inf.mod_lsn : next_lsn;
+        inf.mod_lsn_to = next_lsn;
     });
     return 0;
 }
@@ -1318,16 +1458,15 @@ int blockstore_heap_t::add_entry(uint32_t wr_size, uint32_t *modified_block,
         *modified_block = block_num;
     }
     auto li = (heap_list_item_t*)malloc_or_die(wr_size + sizeof(heap_list_item_t) - sizeof(heap_entry_t));
+    live_entries++;
+    live_memory += list_item_overhead(wr_size);
     auto new_wr = &li->entry;
     auto & inf = block_info.at(block_num);
     if (!inf.entries.size())
         inf.entries.reserve(dsk->meta_block_size / max_entry_size);
     inf.entries.push_back(li);
-    assert(!inf.mod_lsn_to || inf.mod_lsn_to == next_lsn);
-    new_wr->lsn = ++next_lsn;
+    new_wr->lsn = next_lsn;
     fill_entry(new_wr);
-    inf.mod_lsn = inf.mod_lsn ? inf.mod_lsn : next_lsn;
-    inf.mod_lsn_to = next_lsn;
     // 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) |
@@ -1336,7 +1475,7 @@ int blockstore_heap_t::add_entry(uint32_t wr_size, uint32_t *modified_block,
     insert_list_item(li);
     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;
 }
 
@@ -1396,7 +1535,7 @@ 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);
@@ -1425,7 +1564,7 @@ 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;
@@ -1463,7 +1602,7 @@ 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));
@@ -1510,7 +1649,7 @@ 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));
     });
 }
@@ -1666,28 +1805,38 @@ int blockstore_heap_t::add_delete(heap_entry_t *obj, uint32_t *modified_block)
 
 uint32_t blockstore_heap_t::meta_alloc_pos(const heap_block_info_t & inf)
 {
-    if (inf.is_writing || inf.used_space > dsk->meta_block_size-sizeof(heap_entry_t))
+    auto real_used = (inf.used_space-inf.garbage_space);
+    if (inf.is_writing || inf.mod_lsn || real_used > dsk->meta_block_size-sizeof(heap_entry_t))
     {
         // 100% full - no entry can be written into this block at all
         return META_ALLOC_LEVELS;
     }
-    if (inf.used_space > dsk->meta_block_size-max_entry_size)
+    if (real_used > dsk->meta_block_size-max_entry_size)
     {
         // nearfull - big_entries won't fit into this block so it can't be used for compaction
         return META_ALLOC_LEVELS-1;
     }
-    // normal block
-    return inf.used_space / ((dsk->meta_block_size-max_entry_size+META_ALLOC_LEVELS-2) / (META_ALLOC_LEVELS-1));
+    // First we want to write to blocks with most garbage:
+    // >= 2*used, >= used/2
+    // (i.e. 66% garbage, 33% garbage)
+    // Then to mostly free blocks:
+    // >= 75% free, >= 50% free, >= 25% free
+    if (inf.garbage_space > real_used*2)
+        return 0;
+    if (inf.garbage_space > real_used/2)
+        return 1;
+    // META_ALLOC_LEVELS-3 levels left
+    return 2 + real_used / ((dsk->meta_block_size-max_entry_size+META_ALLOC_LEVELS-4) / (META_ALLOC_LEVELS-3));
 }
 
 void blockstore_heap_t::modify_alloc(uint32_t block_num, std::function<void(heap_block_info_t &)> change_cb)
 {
     auto & inf = block_info.at(block_num);
     uint32_t old_pos = meta_alloc_pos(inf);
-    uint32_t old_used = inf.used_space;
+    uint32_t old_used = inf.used_space-inf.garbage_space;
     change_cb(inf);
     uint32_t new_pos = meta_alloc_pos(inf);
-    uint32_t new_used = inf.used_space;
+    uint32_t new_used = inf.used_space-inf.garbage_space;
     meta_alloc->change(block_num, old_pos, new_pos);
     meta_used_space -= old_used;
     meta_used_space += new_used;
@@ -1699,11 +1848,19 @@ void blockstore_heap_t::modify_alloc(uint32_t block_num, std::function<void(heap
 
 void blockstore_heap_t::start_block_write(uint32_t block_num)
 {
-    modify_alloc(block_num, [&](heap_block_info_t & inf)
+    auto & inf = block_info.at(block_num);
+    assert(!inf.is_writing);
+    if (!inf.mod_lsn)
+    {
+        modify_alloc(block_num, [&](heap_block_info_t & inf)
+        {
+            inf.is_writing = true;
+        });
+    }
+    else
     {
-        assert(!inf.is_writing);
         inf.is_writing = true;
-    });
+    }
 }
 
 void blockstore_heap_t::complete_block_write(uint32_t block_num)
@@ -1768,6 +1925,8 @@ void blockstore_heap_t::mark_garbage_up_to(heap_entry_t *wr)
 void blockstore_heap_t::mark_garbage(uint32_t block_num, heap_entry_t *prev_wr, uint32_t used_big)
 {
     prev_wr->set_garbage();
+    garbage_entries++;
+    garbage_memory += list_item_overhead(prev_wr->size);
     // And this is the moment when we can free the data reference
     if (prev_wr->type() == BS_HEAP_SMALL_WRITE && prev_wr->small().len > 0)
     {
@@ -1783,8 +1942,7 @@ void blockstore_heap_t::mark_garbage(uint32_t block_num, heap_entry_t *prev_wr,
     }
     modify_alloc(block_num, [&](heap_block_info_t & inf)
     {
-        inf.used_space -= prev_wr->size;
-        inf.has_garbage = true;
+        inf.garbage_space += prev_wr->size;
     });
 }
 
@@ -2154,7 +2312,7 @@ void blockstore_heap_t::fill_block_empty_space(uint8_t *buffer, uint32_t pos)
 uint32_t blockstore_heap_t::get_meta_block_used_space(uint32_t block_num)
 {
     auto & inf = block_info.at(block_num);
-    return inf.used_space;
+    return inf.used_space - inf.garbage_space;
 }
 
 uint64_t blockstore_heap_t::get_data_used_space()
@@ -2197,6 +2355,26 @@ uint64_t blockstore_heap_t::get_compacted_count()
     return compacted_count;
 }
 
+uint64_t blockstore_heap_t::get_live_entries()
+{
+    return live_entries-garbage_entries;
+}
+
+uint64_t blockstore_heap_t::get_live_memory()
+{
+    return live_memory-garbage_memory;
+}
+
+uint64_t blockstore_heap_t::get_garbage_entries()
+{
+    return garbage_entries;
+}
+
+uint64_t blockstore_heap_t::get_garbage_memory()
+{
+    return 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();
@@ -2283,37 +2461,50 @@ void blockstore_heap_t::apply_inflight(heap_inflight_lsn_t & inflight)
     {
         // Remove entry
         auto li = list_item(wr);
-        auto prev = li->prev;
-        auto next = li->next;
-        if (prev)
-        {
-            prev->next = next;
-        }
-        if (!next)
-        {
-            // The last freed entry must be a deletion
-            assert(!prev);
-            assert(wr->entry_type == BS_HEAP_DELETE|BS_HEAP_STABLE);
-            auto & pg_idx = block_index[get_pg_id(wr->inode, wr->stripe)];
-            auto & inode_idx = pg_idx[wr->inode];
-            heap_inode_map_t::iterator li_it;
-            heap_list_item_t *old_li = NULL;
-            inode_map_get(inode_idx, li_it, old_li, wr->stripe);
-            inode_map_erase(pg_idx, inode_idx, li_it, old_li);
-        }
-        else
-        {
-            next->prev = prev;
-            if (!prev && next->entry.entry_type == (BS_HEAP_DELETE|BS_HEAP_STABLE))
-            {
-                // free BS_HEAP_DELETEs when all previous entries are also freed
-                mark_garbage(next->block_num, &next->entry, UINT32_MAX);
-            }
-        }
-        free(li);
+        remove_list_item(li);
     }
 }
 
+void blockstore_heap_t::remove_list_item(heap_list_item_t *li)
+{
+    auto prev = li->prev;
+    auto next = li->next;
+    if (prev)
+    {
+        prev->next = next;
+    }
+    if (!next)
+    {
+        // The last freed entry must be a deletion
+        assert(!prev);
+        auto wr = &li->entry;
+        assert(wr->entry_type == BS_HEAP_DELETE|BS_HEAP_STABLE);
+        auto & pg_idx = block_index[get_pg_id(wr->inode, wr->stripe)];
+        auto & inode_idx = pg_idx[wr->inode];
+        heap_inode_map_t::iterator li_it;
+        heap_list_item_t *old_li = NULL;
+        inode_map_get(inode_idx, li_it, old_li, wr->stripe);
+        inode_map_erase(pg_idx, inode_idx, li_it, old_li);
+    }
+    else
+    {
+        next->prev = prev;
+        if (!prev && next->entry.entry_type == (BS_HEAP_DELETE|BS_HEAP_STABLE))
+        {
+            // free BS_HEAP_DELETEs when all previous entries are also freed
+            mark_garbage(next->block_num, &next->entry, UINT32_MAX);
+        }
+    }
+    if (li->entry.is_garbage())
+    {
+        garbage_entries--;
+        garbage_memory -= list_item_overhead(li->entry.size);
+    }
+    live_entries--;
+    live_memory -= list_item_overhead(li->entry.size);
+    free(li);
+}
+
 bool blockstore_heap_t::is_lsn_completed(uint64_t lsn)
 {
     if (lsn <= completed_lsn)

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
@@ -117,10 +118,13 @@ struct heap_object_mvcc_t
 
 struct heap_block_info_t
 {
-    uint32_t used_space = 0;
+    struct __attribute__((__packed__))
+    {
+        uint32_t used_space = 0;
+        uint32_t garbage_space = 0;
+    };
     uint64_t mod_lsn = 0, mod_lsn_to = 0; // only 1 block write of LSN sequence is allowed at a moment
-    bool is_writing: 1;
-    bool has_garbage: 1;
+    bool is_writing = false;
     std::vector<heap_list_item_t*> entries;
 };
 
@@ -184,6 +188,11 @@ class blockstore_heap_t
     uint64_t buffer_area_used_space = 0;
     uint64_t data_used_space = 0;
 
+    uint64_t live_entries = 0;
+    uint64_t live_memory = 0;
+    uint64_t garbage_entries = 0;
+    uint64_t garbage_memory = 0;
+
     uint64_t next_lsn = 0;
     uint32_t last_allocated_block = UINT32_MAX;
     heap_mvcc_map_t object_mvcc;
@@ -212,6 +221,7 @@ class blockstore_heap_t
     bool validate_object(heap_entry_t *obj);
     void fill_recheck_queue();
     int mark_used_blocks();
+    void recheck_full_gc();
     void recheck_buffer(heap_entry_t *cwr, uint8_t *buf);
     void defragment_block(uint32_t block_num);
     void reshard_add(heap_reshard_state_t *st, heap_list_item_t *li);
@@ -219,6 +229,7 @@ class blockstore_heap_t
     void gc_block(heap_block_info_t & inf);
     int allocate_entry(uint32_t entry_size, uint32_t *block_num, bool allow_last_free);
     void insert_list_item(heap_list_item_t *li);
+    void remove_list_item(heap_list_item_t *li);
     int add_entry(uint32_t wr_size, uint32_t *modified_block, bool allow_last_free,
         bool explicit_complete, std::function<void(heap_entry_t *wr)> fill_entry);
     int add_simple(heap_entry_t *obj, uint64_t version, uint32_t *modified_block, uint32_t entry_type);
@@ -345,6 +356,10 @@ public:
     uint32_t get_compact_queue_size();
     uint32_t get_to_compact_count();
     uint64_t get_compacted_count();
+    uint64_t get_live_entries();
+    uint64_t get_live_memory();
+    uint64_t get_garbage_entries();
+    uint64_t get_garbage_memory();
 
     uint64_t entry_pos(uint32_t block_num, uint32_t offset);
     heap_entry_t *entry_from_pos(uint64_t entry_pos, bool allow_unallocated = false);

src/blockstore/blockstore_impl.h

@@ -229,4 +229,9 @@ public:
     uint64_t get_free_block_count();
     inline uint32_t get_bitmap_granularity() { return dsk.bitmap_granularity; }
     inline uint64_t get_journal_size() { return dsk.journal_len; }
+
+    inline uint64_t get_live_entries() { return heap->get_live_entries(); }
+    inline uint64_t get_live_memory() { return heap->get_live_memory(); }
+    inline uint64_t get_garbage_entries() { return heap->get_garbage_entries(); }
+    inline uint64_t get_garbage_memory() { return heap->get_garbage_memory(); }
 };

src/blockstore/blockstore_init.cpp

@@ -240,7 +240,7 @@ resume_4:
     }
     // metadata read finished
     bs->heap->finish_load();
-    printf("Metadata entries loaded: %ju, used blocks: %ju / %ju\n", entries_loaded, bs->heap->get_data_used_space() / bs->dsk.data_block_size, bs->dsk.block_count);
+    printf("Metadata entries loaded: %ju, rechecking unfinished writes and garbage entries\n", entries_loaded);
     if (zero_on_init && !bs->dsk.disable_meta_fsync)
     {
         GET_SQE();
@@ -293,6 +293,11 @@ resume_7:
     if (bs->readonly)
     {
         recheck_mod.clear();
+        printf("Actual metadata entries: %ju\n", bs->heap->get_live_entries());
+    }
+    else
+    {
+        printf("Actual metadata entries: %ju, clearing garbage in %zu metadata blocks\n", bs->heap->get_live_entries(), recheck_mod.size());
     }
     for (i = 0; i < recheck_mod.size(); i++)
     {
@@ -306,7 +311,7 @@ resume_8:
         uint32_t block_num = recheck_mod[i];
         uint64_t block_offset = bs->dsk.meta_offset + (uint64_t)(block_num+1) * bs->dsk.meta_block_size;
         data = ((ring_data_t*)sqe->user_data);
-        uint8_t *buf = (uint8_t*)malloc_or_die(bs->dsk.meta_block_size);
+        uint8_t *buf = (uint8_t*)memalign_or_die(MEM_ALIGNMENT, bs->dsk.meta_block_size);
         bs->heap->get_meta_block(block_num, buf);
         data->iov = { buf, bs->dsk.meta_block_size };
         data->callback = [this, buf, block_offset](ring_data_t *data)
@@ -332,5 +337,9 @@ resume_9:
     }
     free(metadata_buffer);
     metadata_buffer = NULL;
+    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(),
+        format_size(bs->dsk.block_count * bs->dsk.data_block_size).c_str());
     return 0;
 }

src/blockstore/blockstore_read.cpp

@@ -462,6 +462,10 @@ int blockstore_impl_t::read_bitmap(object_id oid, uint64_t target_version, void
         {
             if (target_version >= wr->version)
             {
+                if (wr->type() == BS_HEAP_DELETE)
+                {
+                    return false;
+                }
                 found = true;
                 if (result_version)
                 {

src/blockstore/v1/impl.cpp

@@ -855,4 +855,29 @@ std::string blockstore_impl_t::get_op_diag(blockstore_op_t *op)
     return std::string(buf);
 }
 
+uint64_t blockstore_impl_t::get_live_entries()
+{
+    return used_blocks;
+}
+
+uint64_t blockstore_impl_t::get_live_memory()
+{
+    uint64_t used = 0;
+    for (auto & kv: clean_db_shards)
+    {
+        used += kv.second.size() * sizeof(blockstore_clean_db_t::value_type);
+    }
+    return used;
+}
+
+uint64_t blockstore_impl_t::get_garbage_entries()
+{
+    return dirty_db.size();
+}
+
+uint64_t blockstore_impl_t::get_garbage_memory()
+{
+    return (sizeof(obj_ver_id) + sizeof(dirty_entry) + 32) * dirty_db.size();
+}
+
 } // namespace v1

src/blockstore/v1/impl.h

@@ -332,6 +332,10 @@ public:
     inline uint64_t get_free_block_count() { return dsk.block_count - used_blocks; }
     inline uint32_t get_bitmap_granularity() { return dsk.disk_alignment; }
     inline uint64_t get_journal_size() { return dsk.journal_len; }
+    uint64_t get_live_entries();
+    uint64_t get_live_memory();
+    uint64_t get_garbage_entries();
+    uint64_t get_garbage_memory();
 };
 
 } // namespace v1

src/blockstore/v1/read.cpp

@@ -620,8 +620,7 @@ bool blockstore_impl_t::fulfill_clean_read(blockstore_op_t *read_op, uint64_t &
     else if (from_journal)
     {
         // Don't scan bitmap - journal writes don't have holes (internal bitmap)!
-        uint8_t *csum = !dsk.csum_block_size ? 0 : (clean_entry_bitmap + dsk.clean_entry_bitmap_size +
-            item_start/dsk.csum_block_size*(dsk.data_csum_type & 0xFF));
+        uint8_t *csum = !dsk.csum_block_size ? 0 : (clean_entry_bitmap + dsk.clean_entry_bitmap_size);
         if (!fulfill_read(read_op, fulfilled, item_start, item_end,
             (BS_ST_BIG_WRITE | BS_ST_STABLE), 0, clean_loc + item_start, 0, csum, dyn_data))
         {

src/client/CMakeLists.txt

@@ -1,4 +1,4 @@
-cmake_minimum_required(VERSION 2.8.12)
+cmake_minimum_required(VERSION 2.8...3.30)
 
 project(vitastor)
 
@@ -12,11 +12,11 @@ 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 ../util/addr_util.cpp
-	msgr_stop.cpp msgr_op.cpp msgr_send.cpp msgr_receive.cpp ../util/ringloop.cpp ../../json11/json11.cpp
+	../util/epoll_manager.cpp etcd_state_client.cpp messenger.cpp ../util/addr_util.cpp ../util/xxh_x86dispatch.c
+	msgr_encrypt.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}
 )
-target_link_libraries(vitastor_common pthread)
+target_link_libraries(vitastor_common pthread ${OPENSSL_LIBRARIES} ${CARES_LIBRARIES})
 target_compile_options(vitastor_common PUBLIC -fPIC)
 
 # libvitastor_client.so
@@ -24,6 +24,7 @@ add_library(vitastor_client SHARED
 	cluster_client.cpp
 	cluster_client_list.cpp
 	cluster_client_wb.cpp
+	cluster_client_icache.cpp
 	vitastor_c.cpp
 )
 set_target_properties(vitastor_client PROPERTIES PUBLIC_HEADER "client/vitastor_c.h")
@@ -33,6 +34,7 @@ target_link_libraries(vitastor_client
 	${LIBURING_LIBRARIES}
 	${IBVERBS_LIBRARIES}
 	${RDMACM_LIBRARIES}
+	${OPENSSL_LIBRARIES}
 )
 set_target_properties(vitastor_client PROPERTIES VERSION ${VITASTOR_VERSION} SOVERSION 0)
 configure_file(vitastor.pc.in vitastor.pc @ONLY)
@@ -98,9 +100,10 @@ endif (${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
+	pg_states.cpp osd_ops.cpp cluster_client.cpp cluster_client_list.cpp cluster_client_wb.cpp cluster_client_icache.cpp msgr_op.cpp ../test/mock/messenger.cpp msgr_stop.cpp msgr_encrypt.cpp
+	etcd_state_client.cpp ../util/timerfd_manager.cpp ../util/addr_util.cpp ../util/str_util.cpp ../util/json_util.cpp ../util/xxh_x86dispatch.c ../../json11/json11.cpp
 )
+target_link_libraries(test_cluster_client ${OPENSSL_LIBRARIES})
 target_compile_definitions(test_cluster_client PUBLIC -D__MOCK__)
 target_include_directories(test_cluster_client BEFORE PUBLIC ${CMAKE_SOURCE_DIR}/src/test/mock)
 add_dependencies(build_tests test_cluster_client)

src/client/cluster_client.cpp

@@ -27,7 +27,7 @@ cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd
     msgr.ringloop = ringloop;
     msgr.repeer_pgs = [this](osd_num_t peer_osd)
     {
-        if (msgr.osd_peer_fds.find(peer_osd) != msgr.osd_peer_fds.end())
+        if (msgr.osd_peers.find(peer_osd) != msgr.osd_peers.end())
         {
             // peer_osd just connected
             continue_ops();
@@ -47,8 +47,8 @@ cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd
     msgr.exec_op = [this](osd_op_t *op)
     {
         // Garbage in
-        fprintf(stderr, "Incoming garbage from peer %d\n", op->peer_fd);
-        msgr.stop_client(op->peer_fd);
+        fprintf(stderr, "Can't handle incoming operation from client %lu\n", op->client_id);
+        msgr.stop_client(op->client_id);
         delete op;
     };
     msgr.parse_config(config);
@@ -62,6 +62,7 @@ cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd
     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.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;
@@ -70,13 +71,11 @@ cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd
         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);
 }
 
 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;
 }
@@ -156,7 +154,7 @@ void cluster_client_t::continue_raw_ops(osd_num_t peer_osd)
     {
         auto op = it->second;
         op->op_type = OSD_OP_OUT;
-        op->peer_fd = msgr.osd_peer_fds.at(peer_osd);
+        op->client_id = msgr.osd_peers.at(peer_osd)->client_id;
         msgr.outbox_push(op);
         raw_ops.erase(it++);
     }
@@ -481,6 +479,8 @@ void cluster_client_t::on_load_config_hook(json11::Json::object & etcd_global_co
         self_tree_metrics.clear();
         client_hostname = new_hostname;
     }
+    // vault
+    vault_parse_config();
     msgr.parse_config(config);
     st_cli.parse_config(config);
     st_cli.load_pgs();
@@ -590,7 +590,7 @@ void cluster_client_t::on_change_pool_config_hook()
         {
             if (log_level > 2 && pg_counts[pool_item.first])
             {
-                printf("Pool %u (%s) PG count changed from %lu to %lu\n", pool_item.first, pool_item.second.name.c_str(),
+                fprintf(stderr, "Pool %u (%s) PG count changed from %lu to %lu\n", pool_item.first, pool_item.second.name.c_str(),
                     pg_counts[pool_item.first], pool_item.second.real_pg_count);
             }
             // At this point, all pool operations should have been suspended
@@ -607,6 +607,9 @@ 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();
 }
 
@@ -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)
@@ -871,13 +878,13 @@ void cluster_client_t::execute_cas(cluster_op_t *op)
         if (op->retval != expected && op->retval >= 0)
             op->retval = -EIO;
         op->retval = op->retval == -EPIPE ? -EINTR : op->retval;
-        auto peer_it = msgr.osd_peer_fds.find(op->parts[0].osd_num);
+        auto peer_it = msgr.osd_peers.find(op->parts[0].osd_num);
         if (op->retval != 0 || (op->flags & OP_IMMEDIATE_COMMIT))
         {
             auto cb = std::move(op->callback);
             cb(op);
         }
-        else if (peer_it == msgr.osd_peer_fds.end())
+        else if (peer_it == msgr.osd_peers.end())
         {
             // Care must be taken to make sure that the client doesn't reconnect to the OSD
             // before executing the previously completed operation callback (!)
@@ -888,10 +895,10 @@ void cluster_client_t::execute_cas(cluster_op_t *op)
         else
         {
             // CAS writes have a built-in sync
-            auto peer_fd = peer_it->second;
+            osd_client_t *cl = peer_it->second;
             *part = (osd_op_t){
                 .op_type = OSD_OP_OUT,
-                .peer_fd = peer_fd,
+                .client_id = cl->client_id,
                 .req = {
                     .hdr = {
                         .magic = SECONDARY_OSD_OP_MAGIC,
@@ -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);
@@ -972,43 +1009,49 @@ 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;
 }
 
 void cluster_client_t::execute_raw(osd_num_t osd_num, osd_op_t *op)
 {
-    auto fd_it = msgr.osd_peer_fds.find(osd_num);
-    if (fd_it != msgr.osd_peer_fds.end())
+    auto peer_it = msgr.osd_peers.find(osd_num);
+    if (peer_it != msgr.osd_peers.end())
     {
         op->op_type = OSD_OP_OUT;
-        op->peer_fd = fd_it->second;
+        op->client_id = peer_it->second->client_id;
         msgr.outbox_push(op);
     }
     else
@@ -1121,31 +1164,33 @@ resume_2:
         // because if some operations were invalid for the new PG count we'd get errors
         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;
@@ -1196,7 +1241,7 @@ 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)
@@ -1204,7 +1249,7 @@ 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);
             }
@@ -1214,7 +1259,7 @@ 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);
             }
@@ -1223,16 +1268,10 @@ 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);
     }
 }
 
@@ -1252,7 +1291,11 @@ 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);
@@ -1294,10 +1337,10 @@ 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;
@@ -1308,11 +1351,11 @@ 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++;
@@ -1321,7 +1364,7 @@ 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;
@@ -1401,15 +1444,18 @@ int cluster_client_t::try_send(cluster_op_t *op, int i, std::function<void(osd_o
                 primary_osd = nearest_osd;
         }
         part->osd_num = primary_osd;
-        auto peer_it = msgr.osd_peer_fds.find(primary_osd);
-        if (peer_it != msgr.osd_peer_fds.end())
+        auto peer_it = msgr.osd_peers.find(primary_osd);
+        if (peer_it != msgr.osd_peers.end())
         {
-            int peer_fd = peer_it->second;
+            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)
             {
@@ -1419,7 +1465,7 @@ int cluster_client_t::try_send(cluster_op_t *op, int i, std::function<void(osd_o
             }
             part->op = (osd_op_t){
                 .op_type = OSD_OP_OUT,
-                .peer_fd = peer_fd,
+                .client_id = cl->client_id,
                 .req = { .rw = {
                     .header = {
                         .magic = SECONDARY_OSD_OP_MAGIC,
@@ -1428,6 +1474,7 @@ 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,
                 } },
@@ -1435,6 +1482,7 @@ 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);
@@ -1468,8 +1516,8 @@ int cluster_client_t::continue_sync(cluster_op_t *op)
     for (auto do_it = dirty_osds.begin(); do_it != dirty_osds.end(); )
     {
         osd_num_t sync_osd = *do_it;
-        auto peer_it = msgr.osd_peer_fds.find(sync_osd);
-        if (peer_it == msgr.osd_peer_fds.end())
+        auto peer_it = msgr.osd_peers.find(sync_osd);
+        if (peer_it == msgr.osd_peers.end())
             dirty_osds.erase(do_it++);
         else
             do_it++;
@@ -1522,12 +1570,12 @@ resume_1:
 
 void cluster_client_t::send_sync(cluster_op_t *op, cluster_op_part_t *part)
 {
-    auto peer_fd = msgr.osd_peer_fds.at(part->osd_num);
+    osd_client_t *cl = msgr.osd_peers.at(part->osd_num);
     part->flags |= PART_SENT;
     op->inflight_count++;
     part->op = (osd_op_t){
         .op_type = OSD_OP_OUT,
-        .peer_fd = peer_fd,
+        .client_id = cl->client_id,
         .req = {
             .hdr = {
                 .magic = SECONDARY_OSD_OP_MAGIC,
@@ -1567,10 +1615,10 @@ void cluster_client_t::handle_op_part(cluster_op_part_t *part)
             // Error priority: EIO > ENOSPC > ETIMEDOUT > EPIPE
             op->retval = part->op.reply.hdr.retval;
         }
-        int stop_fd = -1;
+        uint64_t stop_client_id = 0;
         if (op->retval != -EINTR && op->retval != -EIO && op->retval != -ENOSPC)
         {
-            stop_fd = part->op.peer_fd;
+            stop_client_id = part->op.client_id;
             if (op->retval != -EPIPE || log_level > 0)
             {
                 fprintf(
@@ -1597,9 +1645,9 @@ void cluster_client_t::handle_op_part(cluster_op_part_t *part)
             op->retry_after = op->retval != -EPIPE ? client_eio_retry_interval : client_retry_interval;
         }
         reset_retry_timer(op->retry_after);
-        if (stop_fd >= 0)
+        if (stop_client_id)
         {
-            msgr.stop_client(stop_fd);
+            msgr.stop_client(stop_client_id);
         }
         op->inflight_count--;
         if (op->inflight_count == 0 && !op->retry_after)
@@ -1634,7 +1682,7 @@ void cluster_client_t::handle_op_part(cluster_op_part_t *part)
             if (op->opcode == OSD_OP_READ || op->opcode == OSD_OP_READ_BITMAP || op->opcode == OSD_OP_READ_CHAIN_BITMAP)
             {
                 for (auto & part: op->parts)
-                    if (part.flags == (PART_SENT|PART_VALID|PART_DONE))
+                    if ((part.flags & (PART_SENT|PART_VALID|PART_DONE)) == (PART_SENT|PART_VALID|PART_DONE))
                         copy_part_bitmap(op, &part);
             }
             if (op->opcode == OSD_OP_SYNC)

src/client/cluster_client.h

@@ -5,6 +5,7 @@
 
 #include "messenger.h"
 #include "etcd_state_client.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;
 };
@@ -80,6 +82,25 @@ 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
 {
@@ -89,8 +110,8 @@ public:
     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;
@@ -102,12 +123,23 @@ public:
     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;
@@ -115,15 +147,28 @@ public:
     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;
@@ -163,6 +208,13 @@ protected:
 #endif
     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);
+
 protected:
     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);
@@ -173,6 +225,7 @@ 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);
@@ -189,6 +242,7 @@ 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 +1,367 @@
+// 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 "http_client.h"
+#include "str_util.h"
+
+#define VAULT_KEY_NOT_LOADED 0
+#define VAULT_KEY_LOADING 1
+#define VAULT_KEY_LOADED 2
+#define VAULT_KEY_ERROR 3
+
+inode_cache_t::~inode_cache_t()
+{
+    if (key_data)
+    {
+        free(key_data);
+        key_data = NULL;
+        op_enc = NULL;
+    }
+}
+
+void cluster_client_t::vault_destroy()
+{
+    if (vault_http_ctx)
+    {
+#ifndef __MOCK__
+        http_destroy(vault_http_cli);
+        http_context_destroy(vault_http_ctx);
+        vault_http_cli = NULL;
+        vault_http_ctx = NULL;
+#endif
+    }
+}
+
+void cluster_client_t::vault_parse_config()
+{
+    vault_url = config["vault_url"].string_value();
+    vault_client_cert = config["vault_client_cert"].string_value();
+    vault_client_key = config["vault_client_key"].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;
+}
+
+#ifndef __MOCK__
+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;
+}
+#endif
+
+void cluster_client_t::vault_load_keys()
+{
+    if (vault_loading || !vault_key_load_queue.size())
+    {
+        return;
+    }
+#ifdef __MOCK__
+    vault_loading = true;
+#else
+    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);
+        }
+    );
+#endif
+}
+
+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

src/client/cluster_client_list.cpp

@@ -295,7 +295,7 @@ int cluster_client_t::start_pg_listing(inode_list_pg_t *pg)
     bool conn = true;
     for (osd_num_t peer_osd: all_peers)
     {
-        if (msgr.osd_peer_fds.find(peer_osd) == msgr.osd_peer_fds.end())
+        if (msgr.osd_peers.find(peer_osd) == msgr.osd_peers.end())
         {
             // Initiate connection
             if (st_cli.peer_states[peer_osd].is_null())
@@ -340,7 +340,7 @@ void cluster_client_t::send_list(inode_list_osd_t *cur_list)
     osd_op_t *op = new osd_op_t();
     op->op_type = OSD_OP_OUT;
     // Already checked that it exists above, but anyway
-    op->peer_fd = msgr.osd_peer_fds.at(cur_list->osd_num);
+    op->client_id = msgr.osd_peers.at(cur_list->osd_num)->client_id;
     op->req = (osd_any_op_t){
         .sec_list = {
             .header = {

src/client/etcd_state_client.cpp

@@ -1,7 +1,10 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
+#include <assert.h>
+#include "malloc_or_die.h"
 #include "osd_ops.h"
+#include "msgr_op.h"
 #include "pg_states.h"
 #include "etcd_state_client.h"
 #ifndef __MOCK__
@@ -22,14 +25,19 @@ etcd_state_client_t::~etcd_state_client_t()
     stop_ws_keepalive();
     if (etcd_watch_ws)
     {
-        http_close(etcd_watch_ws);
+        http_destroy(etcd_watch_ws);
         etcd_watch_ws = NULL;
     }
     if (keepalive_client)
     {
-        http_close(keepalive_client);
+        http_destroy(keepalive_client);
         keepalive_client = NULL;
     }
+    if (http_ctx)
+    {
+        http_context_destroy(http_ctx);
+        http_ctx = NULL;
+    }
 #endif
     if (load_pgs_timer_id >= 0)
     {
@@ -72,55 +80,51 @@ 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,
+http_context_t *etcd_state_client_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_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)
 {
-    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);
-    }
-    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)
+    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_close(http_cli);
-    };
-    http_request(http_cli, etcd_address, req, { .timeout = timeout }, cb);
+        http_destroy(http_cli);
+    });
 }
 
-void etcd_state_client_t::etcd_call(std::string api, json11::Json payload, int timeout,
+void etcd_state_client_t::etcd_call(const std::string & api, json11::Json payload, int timeout,
     int retries, int interval, std::function<void(std::string, json11::Json)> callback)
 {
-    if (!etcd_addresses.size() && !etcd_local.size())
+    pick_next_etcd([=]()
     {
-        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_call_selected(api, payload, timeout, retries, interval, callback);
+    });
+}
+
+void etcd_state_client_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 "+etcd_api_path+api+" HTTP/1.1\r\n"
-        "Host: "+etcd_address+"\r\n"
+    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"
@@ -128,15 +132,15 @@ void etcd_state_client_t::etcd_call(std::string api, json11::Json payload, int t
         "\r\n"+req;
     retries--;
     auto cb = [this, api, payload, timeout, retries, interval, callback,
-        cur_addr = selected_etcd_address](const http_response_t *response)
+        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_address)
-                selected_etcd_address = "";
+            if (cur_addr == selected_etcd_url.addr)
+                selected_etcd_url = (http_url_t){};
             if (retries > 0)
             {
                 if (this->log_level > 0)
@@ -164,54 +168,58 @@ void etcd_state_client_t::etcd_call(std::string api, json11::Json payload, int t
             callback(err, data);
     };
     if (!keepalive_client)
-    {
-        keepalive_client = http_init(tfd);
-    }
-    http_request(keepalive_client, etcd_address, req, { .timeout = timeout, .keepalive = true }, cb);
+        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_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 +227,9 @@ 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();
@@ -239,7 +250,19 @@ void etcd_state_client_t::parse_config(const json11::Json & config)
             add_etcd_url(ea.string_value());
         }
     }
+    if (this->osd_num)
+    {
+        this->etcd_client_cert = config["osd_etcd_client_cert"].string_value();
+        this->etcd_client_key = config["osd_etcd_client_key"].string_value();
+    }
+    else
+    {
+        this->etcd_client_cert = config["etcd_client_cert"].string_value();
+        this->etcd_client_key = config["etcd_client_key"].string_value();
+    }
+    this->etcd_ca = config["etcd_ca"].string_value();
     this->etcd_prefix = config["etcd_prefix"].string_value();
+    this->use_auth = config["use_auth"].bool_value();
     if (this->etcd_prefix == "")
     {
         this->etcd_prefix = "/vitastor";
@@ -291,66 +314,130 @@ void etcd_state_client_t::parse_config(const json11::Json & config)
     }
 }
 
-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()
+void etcd_state_client_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);
     }
-    pick_next_etcd();
-    std::string etcd_address = selected_etcd_address;
-    std::string etcd_api_path;
-    int pos = etcd_address.find('/');
-    if (pos >= 0)
+    if (selected_etcd_url.addr != "")
     {
-        etcd_api_path = etcd_address.substr(pos);
-        etcd_address = etcd_address.substr(0, pos);
+        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_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_t::start_etcd_watcher()
+{
+    pick_next_etcd([this]()
+    {
+        start_etcd_watcher_selected();
+    });
+}
+
+void etcd_state_client_t::start_etcd_watcher_selected()
+{
+    const auto & url = selected_etcd_url;
     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(),
+        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);
     }
-    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 (!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())
         {
@@ -393,7 +480,6 @@ void etcd_state_client_t::start_etcd_watcher()
                                 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();
                             }
@@ -438,7 +524,7 @@ void etcd_state_client_t::start_etcd_watcher()
                         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();
+                    etcd_urls_to_try.clear();
                 }
                 // First gather all changes into a hash to remove multiple overwrites
                 std::map<std::string, etcd_kv_t> changes;
@@ -468,13 +554,8 @@ void etcd_state_client_t::start_etcd_watcher()
         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 (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>
@@ -549,12 +630,7 @@ void etcd_state_client_t::start_ws_keepalive()
             {
                 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;
+                    fprintf(stderr, "Websocket ping failed, disconnecting from etcd %s\n", selected_etcd_url.addr.c_str());
                 }
                 start_etcd_watcher();
             }
@@ -942,6 +1018,8 @@ 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")
@@ -1265,33 +1343,7 @@ void etcd_state_client_t::parse_state(const etcd_kv_t & kv)
             }
             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));
             }
         }
     }
@@ -1302,6 +1354,14 @@ 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 (use_auth && key.substr(0, etcd_prefix.length()+13) == etcd_prefix+"/config/user/")
+    {
+        // <etcd_prefix>/config/user/<username>
+        if (!value.is_object())
+            user_info.erase(key.substr(etcd_prefix.length()+13));
+        else
+            user_info[key.substr(etcd_prefix.length()+13)] = value;
+    }
 }
 
 uint32_t etcd_state_client_t::parse_immediate_commit(const std::string & immediate_commit_str, uint32_t default_value)
@@ -1380,6 +1440,10 @@ 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;
@@ -1388,6 +1452,18 @@ 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;
@@ -1395,6 +1471,53 @@ 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

@@ -4,6 +4,7 @@
 #pragma once
 
 #include <set>
+#include <memory>
 
 #include "json11/json11.hpp"
 #include "object_id.h"
@@ -19,6 +20,8 @@
 #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 +69,7 @@ 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 +87,9 @@ 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 +102,41 @@ struct inode_watch_t
     inode_config_t cfg = {};
 };
 
+struct http_url_t
+{
+    bool ssl;
+    std::string addr;
+    std::string hostname;
+    std::string path;
+};
+
 struct http_co_t;
+struct http_context_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;
+    size_t local_to_try = 0;
+    std::vector<http_url_t> etcd_urls_to_try;
+    http_url_t selected_etcd_url;
+    size_t resolve_count = 0;
     std::vector<inode_watch_t*> watches;
+    std::vector<std::function<void()>> on_resolve_queue;
     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 pick_next_etcd(std::function<void()> cb);
+    void pick_next_etcd_on_resolve();
+    void etcd_call_selected(const std::string & api, json11::Json payload, int timeout, int retries, int interval, std::function<void(std::string, json11::Json)> callback);
+    void start_etcd_watcher_selected();
 public:
     int etcd_keepalive_timeout = 30;
     int etcd_ws_keepalive_interval = 5;
@@ -120,14 +145,20 @@ public:
     int etcd_slow_timeout = 5000;
     int etcd_min_reload_interval = 1000;
     bool infinite_start = true;
+    bool use_auth = false;
     uint64_t global_block_size = DEFAULT_BLOCK_SIZE;
     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_context_t *http_ctx = NULL;
     http_co_t *etcd_watch_ws = NULL, *keepalive_client = NULL;
     int etcd_watches_initialised = 0;
     uint64_t etcd_watch_revision_config = 0;
@@ -140,6 +171,7 @@ public:
     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;
+    std::map<std::string, json11::Json> user_info;
     json11::Json node_placement;
 
     std::function<void(std::map<std::string, etcd_kv_t> &)> on_change_hook;
@@ -158,10 +190,12 @@ 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);
+    http_context_t *get_http_ctx();
+    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);
+    void etcd_call(const std::string & api, json11::Json payload, int timeout, int retries, int interval, std::function<void(std::string, json11::Json)> callback);
     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();

src/client/http_client.h

@@ -8,6 +8,10 @@
 #include <functional>
 #include "json11/json11.hpp"
 
+#ifdef WITH_OPENSSL
+#include <openssl/types.h>
+#endif
+
 #define WS_CONTINUATION 0
 #define WS_TEXT 1
 #define WS_BINARY 2
@@ -17,14 +21,19 @@
 
 class timerfd_manager_t;
 
+#pragma GCC visibility push(default)
+
 struct http_options_t
 {
     int timeout;
     bool want_streaming;
     bool keepalive;
+    bool ssl;
 };
 
-struct http_response_t
+struct http_context_t;
+
+struct http_message_t
 {
     std::string error;
 
@@ -41,10 +50,34 @@ struct http_response_t
 // Opened websocket or keepalive HTTP connection
 struct http_co_t;
 
-http_co_t* http_init(timerfd_manager_t *tfd);
-http_co_t* open_websocket(timerfd_manager_t *tfd, const std::string & host, const std::string & path,
-    int timeout, std::function<void(const http_response_t *msg)> on_message);
+http_context_t* http_context_init(timerfd_manager_t *tfd, const std::string & ssl_cert, const std::string & ssl_key,
+    const std::string & ssl_ca, bool verify_peer, std::string & error);
+std::string http_context_get_ssl_cn(http_context_t *ctx);
+void http_resolve(http_context_t *ctx, bool ssl, std::string host,
+    std::function<void(const std::string & error, const std::vector<std::string> & addrs)> cb);
+void http_context_destroy(http_context_t *ctx);
+http_co_t* http_init(http_context_t *ctx = NULL);
+void open_websocket(http_co_t *handler, const std::string & addr, const std::string & hostname, const std::string & path,
+    const http_options_t & options, std::function<void(http_message_t *msg)> on_message);
 void http_request(http_co_t *handler, const std::string & host, const std::string & request,
-    const http_options_t & options, std::function<void(const http_response_t *response)> response_callback);
+    const http_options_t & options, std::function<void(http_message_t *response)> response_callback);
+void http_get(http_co_t *handler, const std::string & url, const std::string & headers,
+    const http_options_t & options, std::function<void(http_message_t *response)> response_callback);
+void http_json_post(http_co_t *handler, const std::string & url, json11::Json body, const std::string & headers,
+    const http_options_t & options, std::function<void(http_message_t *response)> response_callback);
 void http_post_message(http_co_t *handler, uint8_t type, const std::string & msg);
+void http_serve(http_co_t *handler, int peer_fd, const http_options_t & options,
+    std::function<void(http_message_t *msg)> request_callback);
+void http_reply(http_co_t *handler, const std::string & reply);
 void http_close(http_co_t *co);
+void http_destroy(http_co_t *co);
+
+#pragma GCC visibility pop
+
+#ifdef WITH_OPENSSL
+bool openssl_ctx_add_ca(SSL_CTX *ssl_ctx, const std::string & file_or_pem);
+bool openssl_ctx_use_ca(SSL_CTX *ssl_ctx, const std::string & file_or_pem);
+std::string openssl_get_cn(X509 *x509);
+bool openssl_ctx_use_cert(SSL_CTX *ssl_ctx, const std::string & file_or_pem, std::string & common_name);
+bool openssl_ctx_use_key(SSL_CTX *ssl_ctx, const std::string & file_or_pem);
+#endif

src/client/messenger.cpp

@@ -10,10 +10,18 @@
 #include <stdexcept>
 
 #include "addr_util.h"
+#include "str_util.h"
 #include "messenger.h"
 #ifdef WITH_RDMA
 #include "msgr_rdma.h"
 #endif
+#include "http_client.h"
+#ifdef WITH_OPENSSL
+#include <openssl/bio.h>
+#include <openssl/err.h>
+#include <openssl/pem.h>
+#include <openssl/ssl.h>
+#endif
 
 #include <sys/poll.h>
 
@@ -117,6 +125,50 @@ void msgr_iothread_t::run()
 
 void osd_messenger_t::init()
 {
+    if (!tls_cert.empty() || !tls_key.empty() || !osd_tls_ca.empty() || !client_tls_ca.empty())
+    {
+        // Initialize TLS context
+        // FIXME: require OpenSSL
+#ifndef WITH_OPENSSL
+        fprintf(stderr, "Vitastor is built without OpenSSL support\n");
+        exit(1);
+#else
+        if (tls_cert.empty() || tls_key.empty() || osd_tls_ca.empty() || osd_num && client_tls_ca.empty())
+        {
+            if (osd_num)
+                fprintf(stderr, "Vitastor OSD TLS requires osd_tls_cert, osd_tls_key, osd_tls_ca, client_tls_ca\n");
+            else
+                fprintf(stderr, "Vitastor client TLS requires tls_cert, tls_key and osd_tls_ca\n");
+            exit(1);
+        }
+        else
+        {
+            ssl_ctx = SSL_CTX_new(TLS_method());
+            if (!ssl_ctx)
+            {
+init_err:
+                fprintf(stderr, "OpenSSL initialization failed: %s\n", ERR_error_string(ERR_get_error(), NULL));
+                exit(1);
+            }
+            SSL_CTX_set_verify(ssl_ctx, SSL_VERIFY_PEER, NULL);
+            bool ok = SSL_CTX_set_min_proto_version(ssl_ctx, TLS1_3_VERSION);
+            ok = ok && openssl_ctx_add_ca(ssl_ctx, osd_tls_ca);
+            if (osd_num)
+            {
+                // OSD uses 2 separate root certificates to distinguish between clients and peer OSDs
+                ok = ok && openssl_ctx_add_ca(ssl_ctx, client_tls_ca);
+            }
+            ok = ok && openssl_ctx_use_cert(ssl_ctx, tls_cert, tls_cn);
+            ok = ok && openssl_ctx_use_key(ssl_ctx, tls_key);
+            if (!ok)
+            {
+                SSL_CTX_free(ssl_ctx);
+                ssl_ctx = NULL;
+                goto init_err;
+            }
+        }
+#endif
+    }
 #ifdef WITH_RDMACM
     if (use_rdmacm)
     {
@@ -182,12 +234,13 @@ void osd_messenger_t::init()
     }
     keepalive_timer_id = tfd->set_timer(1000, true, [this](int)
     {
+        std::vector<uint64_t> clients_to_stop;
+        std::vector<osd_op_t*> ops_to_send;
         auto cl_it = clients.begin();
         while (cl_it != clients.end())
         {
             auto cl = cl_it->second;
             cl_it++;
-            auto peer_fd = cl->peer_fd;
             if (!cl->osd_num && !cl->in_osd_num || cl->peer_state != PEER_CONNECTED && cl->peer_state != PEER_RDMA)
             {
                 // Do not run keepalive on regular clients
@@ -199,10 +252,9 @@ void osd_messenger_t::init()
                 if (!cl->ping_time_remaining)
                 {
                     // Ping timed out, stop the client
-                    fprintf(stderr, "Ping timed out for OSD %ju (client %d), disconnecting peer\n", cl->in_osd_num ? cl->in_osd_num : cl->osd_num, cl->peer_fd);
-                    stop_client(peer_fd, true);
-                    // Restart iterator because it may be invalidated
-                    cl_it = clients.upper_bound(peer_fd);
+                    fprintf(stderr, "Ping timed out for OSD %ju (client %ju), disconnecting peer\n",
+                        cl->in_osd_num ? cl->in_osd_num : cl->osd_num, cl->client_id);
+                    clients_to_stop.push_back(cl->client_id);
                 }
             }
             else if (cl->idle_time_remaining > 0)
@@ -213,7 +265,7 @@ void osd_messenger_t::init()
                     // Connection is idle for <osd_idle_time>, send ping
                     osd_op_t *op = new osd_op_t();
                     op->op_type = OSD_OP_OUT;
-                    op->peer_fd = cl->peer_fd;
+                    op->client_id = cl->client_id;
                     op->req = (osd_any_op_t){
                         .hdr = {
                             .magic = SECONDARY_OSD_OP_MAGIC,
@@ -222,28 +274,26 @@ void osd_messenger_t::init()
                     };
                     op->callback = [this, cl](osd_op_t *op)
                     {
-                        auto cl_it = clients.find(op->peer_fd);
-                        if (cl_it == clients.end() || cl_it->second != cl)
+                        auto cl_it = clients.find(op->client_id);
+                        if (cl_it == clients.end())
                         {
                             // client is already dropped
                             delete op;
                             return;
                         }
-                        int fail_fd = (op->reply.hdr.retval != 0 ? op->peer_fd : -1);
+                        uint64_t fail_client_id = (op->reply.hdr.retval != 0 ? op->client_id : 0);
                         auto fail_osd_num = cl->in_osd_num ? cl->in_osd_num : cl->osd_num;
                         cl->ping_time_remaining = 0;
                         delete op;
-                        if (fail_fd >= 0)
+                        if (fail_client_id)
                         {
-                            fprintf(stderr, "Ping failed for OSD %ju (client %d), disconnecting peer\n", fail_osd_num, fail_fd);
-                            stop_client(fail_fd, true);
+                            fprintf(stderr, "Ping failed for OSD %ju (client %ju), disconnecting peer\n", fail_osd_num, fail_client_id);
+                            stop_client(fail_client_id);
                         }
                     };
                     cl->ping_time_remaining = osd_ping_timeout;
                     cl->idle_time_remaining = osd_idle_timeout;
-                    outbox_push(op);
-                    // Restart iterator because it may be invalidated
-                    cl_it = clients.upper_bound(peer_fd);
+                    ops_to_send.push_back(op);
                 }
             }
             else
@@ -251,6 +301,14 @@ void osd_messenger_t::init()
                 cl->idle_time_remaining = osd_idle_timeout;
             }
         }
+        for (uint64_t client_id: clients_to_stop)
+        {
+            stop_client(client_id);
+        }
+        for (osd_op_t *op: ops_to_send)
+        {
+            outbox_push(op);
+        }
     });
 }
 
@@ -263,7 +321,7 @@ osd_messenger_t::~osd_messenger_t()
     }
     while (clients.size() > 0)
     {
-        stop_client(clients.begin()->first, true, true);
+        stop_client(clients.begin()->first, true);
     }
     if (iothreads.size())
     {
@@ -288,6 +346,21 @@ osd_messenger_t::~osd_messenger_t()
         rdma_destroy_event_channel(rdmacm_evch);
         rdmacm_evch = NULL;
     }
+#endif
+    for (auto encrypt_ctx: encrypt_ctx_pool)
+    {
+        destroy_aes_xts_encrypt(encrypt_ctx);
+    }
+    for (auto decrypt_ctx: decrypt_ctx_pool)
+    {
+        destroy_aes_xts_decrypt(decrypt_ctx);
+    }
+#ifdef WITH_OPENSSL
+    if (ssl_ctx)
+    {
+        SSL_CTX_free(ssl_ctx);
+        ssl_ctx = NULL;
+    }
 #endif
 }
 
@@ -323,6 +396,33 @@ void osd_messenger_t::parse_config(const json11::Json & config)
     if (!this->rdma_max_msg || this->rdma_max_msg > 128*1024*1024)
         this->rdma_max_msg = 129*1024;
 #endif
+    this->max_aes_xts_pool_size = config["max_aes_xts_pool_size"].uint64_value();
+    if (!this->max_aes_xts_pool_size)
+        this->max_aes_xts_pool_size = 256;
+    if (config["proto_checksums"].is_null())
+        this->use_proto_checksums = MSGR_CSUM_PAYLOAD;
+    else if (config["proto_checksums"].is_bool())
+        this->use_proto_checksums = config["proto_checksums"].bool_value() ? MSGR_CSUM_FULL : 0;
+    else if (config["proto_checksums"].string_value() != "")
+        this->use_proto_checksums = config["proto_checksums"].string_value() == "full" ? MSGR_CSUM_FULL : MSGR_CSUM_PAYLOAD;
+    else
+        this->use_proto_checksums = 0;
+    if (!osd_num)
+    {
+        tls_cert = config["tls_cert"].string_value();
+        tls_key = config["tls_key"].string_value();
+        osd_tls_ca = config["osd_tls_ca"].string_value();
+    }
+    else
+    {
+        tls_cert = config["osd_tls_cert"].string_value();
+        tls_key = config["osd_tls_key"].string_value();
+        osd_tls_ca = config["osd_tls_ca"].string_value();
+        client_tls_ca = config["client_tls_ca"].string_value();
+    }
+    test_osd_aes_key.resize(32);
+    if (fromhexstr(config["test_osd_aes_key"].string_value(), 32, (uint8_t*)test_osd_aes_key.data()) != 32)
+        test_osd_aes_key.clear();
     if (!osd_num)
         this->iothread_count = (uint32_t)config["client_iothread_count"].uint64_value();
     else
@@ -331,7 +431,7 @@ void osd_messenger_t::parse_config(const json11::Json & config)
     if (!this->receive_buffer_size || this->receive_buffer_size > 1024*1024*1024)
         this->receive_buffer_size = 65536;
     this->use_sync_send_recv = config["use_sync_send_recv"].bool_value() ||
-        config["use_sync_send_recv"].uint64_value();
+        config["use_sync_send_recv"].uint64_value() || !ringloop;
     this->min_zerocopy_send_size = config["min_zerocopy_send_size"].is_null()
         ? DEFAULT_MIN_ZEROCOPY_SEND_SIZE
         : (int)config["min_zerocopy_send_size"].int64_value();
@@ -440,7 +540,7 @@ void osd_messenger_t::try_connect_peer(uint64_t peer_osd)
     {
         return;
     }
-    if (osd_peer_fds.find(peer_osd) != osd_peer_fds.end())
+    if (osd_peers.find(peer_osd) != osd_peers.end())
     {
         wanted_peers.erase(peer_osd);
         return;
@@ -467,20 +567,20 @@ void osd_messenger_t::try_connect_peer_tcp(osd_num_t peer_osd, const char *peer_
 #ifdef WITH_RDMACM
     if (disable_tcp)
     {
-        on_connect_peer(peer_osd, -EINVAL);
+        on_connect_peer(peer_osd, -EINVAL, 0);
         return;
     }
 #endif
     struct sockaddr_storage addr;
     if (!string_to_addr(peer_host, 0, peer_port, &addr))
     {
-        on_connect_peer(peer_osd, -EINVAL);
+        on_connect_peer(peer_osd, -EINVAL, 0);
         return;
     }
     int peer_fd = socket(addr.ss_family, SOCK_STREAM, 0);
     if (peer_fd < 0)
     {
-        on_connect_peer(peer_osd, -errno);
+        on_connect_peer(peer_osd, -errno, 0);
         return;
     }
     fcntl(peer_fd, F_SETFL, fcntl(peer_fd, F_GETFL, 0) | O_NONBLOCK);
@@ -488,21 +588,25 @@ void osd_messenger_t::try_connect_peer_tcp(osd_num_t peer_osd, const char *peer_
     if (r < 0 && errno != EINPROGRESS)
     {
         close(peer_fd);
-        on_connect_peer(peer_osd, -errno);
+        on_connect_peer(peer_osd, -errno, 0);
         return;
     }
-    clients[peer_fd] = new osd_client_t();
+    const uint64_t client_id = next_client_id++;
+    osd_client_t *cl = new osd_client_t();
     if (log_level > 0)
     {
-        fprintf(stderr, "Connecting to OSD %ju at %s:%d (client %d)\n", peer_osd, peer_host, peer_port, peer_fd);
+        fprintf(stderr, "Connecting to OSD %ju at %s:%d (client %ju, FD %d)\n", peer_osd, peer_host, peer_port, client_id, peer_fd);
     }
-    clients[peer_fd]->peer_addr = addr;
-    clients[peer_fd]->peer_port = peer_port;
-    clients[peer_fd]->peer_fd = peer_fd;
-    clients[peer_fd]->peer_state = PEER_CONNECTING;
-    clients[peer_fd]->connect_timeout_id = -1;
-    clients[peer_fd]->osd_num = peer_osd;
-    clients[peer_fd]->in_buf = malloc_or_die(receive_buffer_size);
+    cl->client_id = client_id;
+    cl->peer_addr = addr;
+    cl->peer_port = peer_port;
+    cl->peer_fd = peer_fd;
+    cl->peer_state = PEER_CONNECTING;
+    cl->connect_timeout_id = -1;
+    cl->osd_num = peer_osd;
+    cl->in_buf = (uint8_t*)malloc_or_die(receive_buffer_size);
+    clients[client_id] = cl;
+    clients_by_fd[peer_fd] = cl;
     tfd->set_fd_handler(peer_fd, true, [this](int peer_fd, int epoll_events)
     {
         // Either OUT (connected) or HUP
@@ -510,11 +614,11 @@ void osd_messenger_t::try_connect_peer_tcp(osd_num_t peer_osd, const char *peer_
     });
     if (peer_connect_timeout > 0)
     {
-        clients[peer_fd]->connect_timeout_id = tfd->set_timer(1000*peer_connect_timeout, false, [this, peer_fd](int timer_id)
+        cl->connect_timeout_id = tfd->set_timer(1000*peer_connect_timeout, false, [this, client_id](int timer_id)
         {
-            osd_num_t peer_osd = clients.at(peer_fd)->osd_num;
-            stop_client(peer_fd, true);
-            on_connect_peer(peer_osd, -EPIPE);
+            osd_num_t peer_osd = clients.at(client_id)->osd_num;
+            stop_client(client_id);
+            on_connect_peer(peer_osd, -EPIPE, 0);
             return;
         });
     }
@@ -522,7 +626,7 @@ void osd_messenger_t::try_connect_peer_tcp(osd_num_t peer_osd, const char *peer_
 
 void osd_messenger_t::handle_connect_epoll(int peer_fd)
 {
-    auto cl = clients[peer_fd];
+    auto cl = clients_by_fd.at(peer_fd);
     if (cl->connect_timeout_id >= 0)
     {
         tfd->clear_timer(cl->connect_timeout_id);
@@ -537,8 +641,8 @@ void osd_messenger_t::handle_connect_epoll(int peer_fd)
     }
     if (result != 0)
     {
-        stop_client(peer_fd, true);
-        on_connect_peer(peer_osd, -result);
+        stop_client(cl->client_id);
+        on_connect_peer(peer_osd, -result, 0);
         return;
     }
     int one = 1;
@@ -549,29 +653,30 @@ void osd_messenger_t::handle_connect_epoll(int peer_fd)
         handle_peer_epoll(peer_fd, epoll_events);
     });
     // Check OSD number
+    ssl_init(cl, false);
     check_peer_config(cl);
 }
 
 void osd_messenger_t::handle_peer_epoll(int peer_fd, int epoll_events)
 {
     // Mark client as ready (i.e. some data is available)
+    auto cl = clients_by_fd.at(peer_fd);
     if (epoll_events & EPOLLRDHUP)
     {
         // Stop client
         if (log_level > 0)
         {
-            fprintf(stderr, "[OSD %ju] client %d disconnected\n", this->osd_num, peer_fd);
+            fprintf(stderr, "[OSD %ju] client %ju disconnected\n", this->osd_num, cl->client_id);
         }
-        stop_client(peer_fd, true);
+        stop_client(cl->client_id);
     }
     else if (epoll_events & EPOLLIN)
     {
         // Mark client as ready (i.e. some data is available)
-        auto cl = clients[peer_fd];
         cl->read_ready++;
         if (cl->read_ready == 1)
         {
-            read_ready_clients.push_back(cl->peer_fd);
+            read_ready_clients.push_back(cl->client_id);
             if (ringloop)
                 ringloop->wakeup();
             else
@@ -580,13 +685,13 @@ void osd_messenger_t::handle_peer_epoll(int peer_fd, int epoll_events)
     }
 }
 
-void osd_messenger_t::on_connect_peer(osd_num_t peer_osd, int peer_fd)
+void osd_messenger_t::on_connect_peer(osd_num_t peer_osd, int errcode, uint64_t client_id)
 {
     auto & wp = wanted_peers.at(peer_osd);
     wp.connecting = false;
-    if (peer_fd < 0)
+    if (errcode < 0)
     {
-        fprintf(stderr, "Failed to connect to peer OSD %ju address %s port %d: %s\n", peer_osd, wp.cur_addr.c_str(), wp.cur_port, strerror(-peer_fd));
+        fprintf(stderr, "Failed to connect to peer OSD %ju address %s port %d: %s\n", peer_osd, wp.cur_addr.c_str(), wp.cur_port, strerror(-errcode));
         if (wp.address_changed)
         {
             wp.address_changed = false;
@@ -613,7 +718,7 @@ void osd_messenger_t::on_connect_peer(osd_num_t peer_osd, int peer_fd)
     }
     if (log_level > 0)
     {
-        fprintf(stderr, "[OSD %ju] Connected with peer OSD %ju (client %d)\n", osd_num, peer_osd, peer_fd);
+        fprintf(stderr, "[OSD %ju] Connected with peer OSD %ju (client %ju)\n", osd_num, peer_osd, client_id);
     }
     wanted_peers.erase(peer_osd);
     repeer_pgs(peer_osd);
@@ -623,7 +728,7 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
 {
     osd_op_t *op = new osd_op_t();
     op->op_type = OSD_OP_OUT;
-    op->peer_fd = cl->peer_fd;
+    op->client_id = cl->client_id;
     op->req = (osd_any_op_t){
         .show_conf = {
             .header = {
@@ -638,7 +743,12 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
         // Inform that we're OSD <osd_num>
         payload["osd_num"] = osd_num;
     }
-    payload["features"] = json11::Json::object{ { "check_sequencing", true } };
+    auto features = json11::Json::object{ { "check_sequencing", true } };
+    if (use_proto_checksums)
+    {
+        features["proto_checksums"] = use_proto_checksums;
+    }
+    payload["features"] = features;
 #ifdef WITH_RDMA
     if (!use_rdmacm && rdma_contexts.size())
     {
@@ -647,7 +757,7 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
         if (!selected_ctx)
         {
             if (log_level > 0)
-                fprintf(stderr, "No RDMA context for OSD %ju connection (peer %d), using only TCP\n", cl->osd_num, cl->peer_fd);
+                fprintf(stderr, "No RDMA context for OSD %ju connection (client %ju), using only TCP\n", cl->osd_num, cl->client_id);
         }
         else
         {
@@ -708,11 +818,19 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
         if (err)
         {
             osd_num_t peer_osd = cl->osd_num;
-            stop_client(op->peer_fd);
-            on_connect_peer(peer_osd, -EINVAL);
+            stop_client(op->client_id);
+            on_connect_peer(peer_osd, -EINVAL, 0);
             delete op;
             return;
         }
+        if (use_proto_checksums)
+        {
+            auto peer_csums = config["features"]["proto_checksums"].uint64_value();
+            if (peer_csums == MSGR_CSUM_FULL && use_proto_checksums == MSGR_CSUM_FULL)
+                cl->proto_csum_status = MSGR_CSUM_FULL;
+            else if (peer_csums && use_proto_checksums)
+                cl->proto_csum_status = MSGR_CSUM_PAYLOAD;
+        }
 #ifdef WITH_RDMA
         if (!use_rdmacm && cl->rdma_conn && config["rdma_address"].is_string())
         {
@@ -739,21 +857,13 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
                     fprintf(stderr, "Connected to OSD %ju using RDMA\n", cl->osd_num);
                 }
                 cl->peer_state = PEER_RDMA;
-                tfd->set_fd_handler(cl->peer_fd, false, [this](int peer_fd, int epoll_events)
-                {
-                    // Do not miss the disconnection!
-                    if (epoll_events & EPOLLRDHUP)
-                    {
-                        handle_peer_epoll(peer_fd, epoll_events);
-                    }
-                });
                 // Add the initial receive request
                 init_recv_rdma(cl);
             }
         }
 #endif
-        osd_peer_fds[cl->osd_num] = cl->peer_fd;
-        on_connect_peer(cl->osd_num, cl->peer_fd);
+        osd_peers[cl->osd_num] = cl;
+        on_connect_peer(cl->osd_num, 0, cl->client_id);
         delete op;
     };
     outbox_push(op);
@@ -768,20 +878,24 @@ void osd_messenger_t::accept_connections(int listen_fd)
     while ((peer_fd = accept(listen_fd, (sockaddr*)&addr, &peer_addr_size)) >= 0)
     {
         assert(peer_fd != 0);
-        fprintf(stderr, "[OSD %ju] new client %d: connection from %s\n", this->osd_num, peer_fd,
+        const uint64_t client_id = next_client_id++;
+        fprintf(stderr, "[OSD %ju] new client %ju (FD %d): connection from %s\n", this->osd_num, client_id, peer_fd,
             addr_to_string(addr).c_str());
         fcntl(peer_fd, F_SETFL, fcntl(peer_fd, F_GETFL, 0) | O_NONBLOCK);
         int one = 1;
         setsockopt(peer_fd, SOL_TCP, TCP_NODELAY, &one, sizeof(one));
         auto cl = new osd_client_t();
-        clients[peer_fd] = cl;
+        cl->client_id = client_id;
+        clients[cl->client_id] = cl;
+        clients_by_fd[peer_fd] = cl;
         cl->is_incoming = true;
         cl->peer_addr = addr;
         cl->peer_addr = addr;
         cl->peer_port = ntohs(((sockaddr_in*)&addr)->sin_port);
         cl->peer_fd = peer_fd;
         cl->peer_state = PEER_CONNECTED;
-        cl->in_buf = malloc_or_die(receive_buffer_size);
+        cl->in_buf = (uint8_t*)malloc_or_die(receive_buffer_size);
+        ssl_init(cl, true);
         // Add FD to epoll
         tfd->set_fd_handler(peer_fd, false, [this](int peer_fd, int epoll_events)
         {
@@ -796,6 +910,52 @@ void osd_messenger_t::accept_connections(int listen_fd)
     }
 }
 
+void osd_messenger_t::ssl_init(osd_client_t *cl, bool server_mode)
+{
+    if (!tls_cert.empty())
+    {
+        cl->write_to_ssl = BIO_new(BIO_s_mem());
+        cl->read_from_ssl = BIO_new(BIO_s_mem());
+        cl->ssl_cli = SSL_new(ssl_ctx);
+        if (!cl->ssl_cli)
+        {
+            fprintf(stderr, "OpenSSL initialization failed: %s\n", ERR_error_string(ERR_get_error(), NULL));
+            exit(1);
+        }
+        if (server_mode)
+        {
+            SSL_set_accept_state(cl->ssl_cli);
+        }
+        else
+        {
+            SSL_set_connect_state(cl->ssl_cli);
+        }
+        SSL_set_bio(cl->ssl_cli, cl->write_to_ssl, cl->read_from_ssl);
+        bool ok = ssl_do_handshake(cl);
+        assert(ok);
+    }
+    else if (!test_osd_aes_key.empty())
+    {
+        int r;
+        cl->enc_ctx = EVP_CIPHER_CTX_new();
+        assert(cl->enc_ctx);
+        r = EVP_EncryptInit_ex(cl->enc_ctx, EVP_aes_256_gcm(), NULL, NULL, NULL);
+        assert(r == 1);
+        r = EVP_CIPHER_CTX_set_padding(cl->enc_ctx, 0);
+        assert(r == 1);
+        r = EVP_CIPHER_CTX_ctrl(cl->enc_ctx, EVP_CTRL_GCM_SET_IVLEN, 12, NULL);
+        assert(r == 1);
+        cl->dec_ctx = EVP_CIPHER_CTX_new();
+        assert(cl->dec_ctx);
+        r = EVP_DecryptInit_ex(cl->dec_ctx, EVP_aes_256_gcm(), NULL, NULL, NULL);
+        assert(r == 1);
+        r = EVP_CIPHER_CTX_set_padding(cl->dec_ctx, 0);
+        assert(r == 1);
+        r = EVP_CIPHER_CTX_ctrl(cl->dec_ctx, EVP_CTRL_GCM_SET_IVLEN, 12, NULL);
+        assert(r == 1);
+    }
+}
+
 #ifdef WITH_RDMA
 msgr_rdma_context_t* osd_messenger_t::choose_rdma_context(osd_client_t *cl)
 {

src/client/messenger.h

@@ -12,6 +12,12 @@
 #include <deque>
 #include <vector>
 
+#ifdef WITH_OPENSSL
+#include <openssl/types.h>
+#endif
+
+#include "../util/xxh_x86dispatch.h"
+#include "../util/robin_hood.h"
 #include "malloc_or_die.h"
 #include "json11/json11.hpp"
 #include "msgr_op.h"
@@ -30,13 +36,14 @@
 #define PEER_RDMA 4
 #define PEER_STOPPED 5
 
+#define MSGR_CSUM_PAYLOAD 1
+#define MSGR_CSUM_FULL 2
+#define MSGR_CSUM_NEG 4
+
 #define VITASTOR_CONFIG_PATH "/etc/vitastor/vitastor.conf"
 
 #define DEFAULT_MIN_ZEROCOPY_SEND_SIZE 32*1024
 
-#define MSGR_SENDP_HDR 1
-#define MSGR_SENDP_FREE 2
-
 struct msgr_sendp_t
 {
     osd_op_t *op;
@@ -48,8 +55,22 @@ struct msgr_rdma_connection_t;
 struct msgr_rdma_context_t;
 #endif
 
+struct op_aes_xts_encrypt_t;
+struct op_aes_xts_decrypt_t;
+void destroy_aes_xts_encrypt(op_aes_xts_encrypt_t *encrypt_ctx);
+void destroy_aes_xts_decrypt(op_aes_xts_decrypt_t *decrypt_ctx);
+
+// Standard TLS record header. We are only interested in the record size
+struct __attribute__((__packed__)) msgr_tls_record_hdr_t
+{
+    uint8_t content_type;
+    uint16_t version;
+    uint16_t size;
+};
+
 struct osd_client_t
 {
+    uint64_t client_id = 0;
     int refs = 0;
 
     sockaddr_storage peer_addr = {};
@@ -63,40 +84,77 @@ struct osd_client_t
     osd_num_t in_osd_num = 0;
     bool is_incoming = false;
 
-    void *in_buf = NULL;
+    uint8_t *in_buf = NULL;
 
 #ifdef WITH_RDMA
     msgr_rdma_connection_t *rdma_conn = NULL;
 #endif
 
+#ifdef WITH_OPENSSL
+    SSL *ssl_cli = NULL;
+    BIO *write_to_ssl = NULL;
+    // FIXME: use custom bio to avoid 1 more memory copy?
+    BIO *read_from_ssl = NULL;
+    uint8_t *ssl_out_buf = NULL;
+    size_t ssl_out_buf_size = 0, ssl_out_buf_cap = 0;
+    bool ssl_handshake_done = false;
+    msgr_tls_record_hdr_t ssl_read_record;
+    size_t ssl_read_header_size = 0;
+    bool ssl_more_to_buffer = false;
+
+    EVP_CIPHER_CTX *enc_ctx = NULL;
+    uint8_t enc_tag[16];
+    size_t enc_tag_size = 0;
+    bool enc_batch = false;
+    EVP_CIPHER_CTX *dec_ctx = NULL;
+    uint8_t dec_tag[16];
+    size_t dec_tag_size = 0;
+    uint32_t dec_batch_size = 0;
+    size_t dec_batch_size_size = 0;
+    std::vector<osd_op_t*> unverified_ops;
+#endif
+
     // Read state
+    bool io_error = false;
     int read_ready = 0;
     osd_op_t *read_op = NULL;
+    size_t read_op_size = 0;
+    size_t read_op_pos = 0;
     iovec read_iov = { 0 };
     msghdr read_msg = { 0 };
-    int read_remaining = 0;
-    int read_state = 0;
-    osd_op_buf_list_t recv_list;
+    std::vector<iovec> recv_list;
+    std::vector<int> recv_flags;
     uint64_t read_op_id = 1;
     bool check_sequencing = false;
     bool enable_pg_locks = false;
+    op_aes_xts_decrypt_t *decrypt_ctx = NULL;
+    size_t read_op_inline_decrypt_pos = 0;
+    size_t read_op_inline_decrypt_in = 0;
+    int proto_csum_status = 0;
+    XXH3_state_t* read_csum_state = NULL;
 
     // Incoming operations
     std::vector<osd_op_t*> received_ops;
 
     // Outbound operations
-    std::map<uint64_t, osd_op_t*> sent_ops;
+    robin_hood::unordered_flat_map<uint64_t, osd_op_t*> sent_ops;
     uint64_t send_op_id = 0;
 
     // PGs dirtied by this client's primary-writes
     std::set<pool_pg_num_t> dirty_pgs;
 
     // Write state
+    std::deque<osd_op_t *> write_ops;
+    osd_op_t *write_op = NULL;
+    size_t write_op_pos = 0;
     msghdr write_msg = { 0 };
     int write_state = 0;
-    std::vector<iovec> send_list, next_send_list;
-    std::vector<msgr_sendp_t> outbox, next_outbox;
+    std::vector<iovec> send_list;
+    size_t send_list_size = 0;
+    std::deque<osd_op_t*> send_free_ops;
     std::vector<osd_op_t*> zc_free_list;
+    op_aes_xts_encrypt_t *encrypt_ctx = NULL;
+    XXH3_state_t* write_csum_state = NULL;
 
     ~osd_client_t();
     void cancel_ops();
@@ -174,9 +232,21 @@ struct osd_messenger_t;
 struct rdmacm_connecting_t;
 #endif
 
+class msgr_op_reader_t;
+class msgr_op_writer_t;
+
 struct __attribute__((visibility("default"))) osd_messenger_t
 {
 protected:
+    friend class copy_op_reader_t;
+    friend class ssl_op_reader_t;
+    friend class gcm_op_reader_t;
+    friend class get_op_reader_t;
+    friend class copy_op_writer_t;
+    friend class ssl_op_writer_t;
+    friend class gcm_op_writer_t;
+    friend class get_op_writer_t;
+
     int keepalive_timer_id = -1;
 
     uint32_t receive_buffer_size = 0;
@@ -188,6 +258,13 @@ protected:
     bool use_sync_send_recv = false;
     int min_zerocopy_send_size = DEFAULT_MIN_ZEROCOPY_SEND_SIZE;
     int iothread_count = 0;
+    int max_aes_xts_pool_size = 256;
+
+    std::string tls_cert;
+    std::string tls_key;
+    std::string osd_tls_ca;
+    std::string client_tls_ca;
+    std::string test_osd_aes_key; // FIXME Insecure, only for PoC tests
 
 #ifdef WITH_RDMA
     bool use_rdma = true;
@@ -201,38 +278,53 @@ protected:
     uint64_t rdma_max_sge = 0, rdma_max_send = 0, rdma_max_recv = 0;
     uint64_t rdma_max_msg = 0;
     rdma_event_channel *rdmacm_evch = NULL;
-    std::map<rdma_cm_id*, osd_client_t*> rdmacm_connections;
-    std::map<rdma_cm_id*, rdmacm_connecting_t*> rdmacm_connecting;
+    robin_hood::unordered_flat_map<rdma_cm_id*, osd_client_t*> rdmacm_connections;
+    robin_hood::unordered_flat_map<rdma_cm_id*, rdmacm_connecting_t*> rdmacm_connecting;
+#endif
+
+#ifdef WITH_OPENSSL
+    SSL_CTX *ssl_ctx = NULL;
+    std::string tls_cn;
+
+    void ssl_init(osd_client_t *cl, bool server_mode);
+    bool ssl_do_handshake(osd_client_t *cl);
 #endif
 
     std::vector<msgr_iothread_t*> iothreads;
-    std::vector<int> read_ready_clients;
-    std::vector<int> write_ready_clients;
+    std::vector<uint64_t> read_ready_clients;
+    std::vector<uint64_t> write_ready_clients;
     // We don't use ringloop->set_immediate here because we may have no ringloop in client :)
     std::deque<osd_op_t*> set_immediate_ops;
 
+    std::vector<op_aes_xts_encrypt_t*> encrypt_ctx_pool;
+    std::vector<op_aes_xts_decrypt_t*> decrypt_ctx_pool;
+
 public:
     timerfd_manager_t *tfd = NULL;
     ring_loop_t *ringloop = NULL;
     bool has_sendmsg_zc = false;
     // osd_num_t is only for logging and asserts
+    uint64_t next_client_id = 1;
     osd_num_t osd_num;
-    std::map<int, osd_client_t*> clients;
-    std::map<osd_num_t, osd_wanted_peer_t> wanted_peers;
-    std::map<uint64_t, int> osd_peer_fds;
+    robin_hood::unordered_flat_map<uint64_t, osd_client_t*> clients;
+    robin_hood::unordered_flat_map<uint64_t, osd_client_t*> osd_peers;
+    robin_hood::unordered_flat_map<int, osd_client_t*> clients_by_fd;
+    robin_hood::unordered_flat_map<osd_num_t, osd_wanted_peer_t> wanted_peers;
     std::vector<std::string> osd_networks;
     std::vector<addr_mask_t> osd_network_masks;
     std::vector<std::string> osd_cluster_networks;
     std::vector<addr_mask_t> osd_cluster_network_masks;
     std::vector<std::string> all_osd_networks;
     std::vector<addr_mask_t> all_osd_network_masks;
+    int use_proto_checksums = 0;
     // op statistics
     osd_op_stats_t stats, recovery_stats;
 
     void init();
     void parse_config(const json11::Json & config);
     void connect_peer(uint64_t osd_num, json11::Json peer_state);
-    void stop_client(int peer_fd, bool force = false, bool force_delete = false);
+    void stop_client(uint64_t client_id, bool force_delete = false);
+    void destroy_client(osd_client_t *cl);
     void outbox_push(osd_op_t *cur_op);
     std::function<void(osd_op_t*)> exec_op;
     std::function<void(osd_num_t)> repeer_pgs;
@@ -251,7 +343,7 @@ public:
 
 #ifdef WITH_RDMA
     bool is_rdma_enabled();
-    bool connect_rdma(int peer_fd, std::string rdma_address, uint64_t client_max_msg);
+    bool connect_rdma(uint64_t client_id, std::string rdma_address, uint64_t client_max_msg);
 #endif
 #ifdef WITH_RDMACM
     bool is_use_rdmacm();
@@ -267,22 +359,37 @@ protected:
     void try_connect_peer_tcp(osd_num_t peer_osd, const char *peer_host, int peer_port);
     void handle_peer_epoll(int peer_fd, int epoll_events);
     void handle_connect_epoll(int peer_fd);
-    void on_connect_peer(osd_num_t peer_osd, int peer_fd);
+    void on_connect_peer(osd_num_t peer_osd, int errcode, uint64_t client_id);
     void check_peer_config(osd_client_t *cl);
     void cancel_osd_ops(osd_client_t *cl);
     void cancel_op(osd_op_t *op);
 
     bool try_send(osd_client_t *cl);
     void handle_send(int result, bool prev, bool more, osd_client_t *cl);
+    bool op_write_to(osd_client_t *cl, msgr_op_writer_t & wr);
+    void next_write_op(osd_client_t *cl);
+    bool op_write_buf(osd_client_t *cl, uint8_t *src, size_t src_len, uint8_t *dst, size_t dst_len, bool skip_csum, size_t & from, size_t & done);
+    bool op_copy_data_to(osd_client_t *cl, uint8_t *dst, size_t dst_len, size_t & from, size_t & done);
+    size_t copy_ops_to(osd_client_t *cl, uint8_t *dst, size_t dst_len);
+    template<typename T> size_t copy_ops_to_with(osd_client_t *cl, uint8_t *dst, size_t dst_len);
 
-    bool handle_read(int result, osd_client_t *cl);
-    bool handle_read_buffer(osd_client_t *cl, void *curbuf, int remain);
-    bool handle_finished_read(osd_client_t *cl);
-    void handle_op_hdr(osd_client_t *cl);
-    bool handle_reply_hdr(osd_client_t *cl);
-    void handle_reply_ready(osd_op_t *op);
+    void handle_read(int result, osd_client_t *cl);
+    bool handle_read_buffer(osd_client_t *cl, uint8_t *curbuf, size_t bufsize);
+    template<typename T> bool handle_buffer_with(osd_client_t *cl, uint8_t *curbuf, size_t bufsize);
+    bool handle_hdr(osd_client_t *cl);
+    bool allocate_op_buffers(osd_client_t *cl);
+    bool allocate_reply_buffers(osd_client_t *cl, osd_op_t *op);
+    bool op_read_from(osd_client_t *cl, msgr_op_reader_t & rdr);
+    bool handle_finished_op(osd_client_t *cl);
+    void execute_verified_op(osd_client_t *cl, osd_op_t *op);
     void handle_immediate_ops();
-    void clear_immediate_ops(int peer_fd);
+
+    void op_encrypted_copy_buf(osd_client_t *cl, uint8_t *enc_buf, size_t enc_len, uint8_t *plain, size_t plain_len, size_t & done_plain, size_t & done_enc);
+    void op_encrypt_free(osd_client_t* cl);
+    void op_decrypted_copy_buf(osd_client_t *cl, uint8_t *enc_buf, size_t enc_len, uint8_t *plain, size_t plain_len, size_t & done_plain, size_t & done_enc);
+    void op_decrypt_start(osd_client_t* cl);
+    void op_decrypt_inline(osd_client_t* cl);
+    void op_decrypt_free(osd_client_t* cl);
 
 #ifdef WITH_RDMA
     void try_send_rdma(osd_client_t *cl);

src/client/msgr_encrypt.cpp

@@ -0,0 +1,467 @@
+// Copyright (c) Vitaliy Filippov, 2026+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#include <assert.h>
+
+#include "etcd_state_client.h"
+#include "messenger.h"
+#include "msgr_encrypt.h"
+
+op_aes_xts_encrypt_t::op_aes_xts_encrypt_t()
+{
+#ifdef WITH_OPENSSL
+    if (!(ctx = EVP_CIPHER_CTX_new()))
+    {
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+    EVP_CIPHER_CTX_set_padding(ctx, 0);
+    if (EVP_EncryptInit_ex(ctx, EVP_aes_256_xts(), NULL, NULL, NULL) != 1)
+    {
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+#else
+    fprintf(stderr, "Error: Vitastor is built without encryption support\n");
+    abort();
+#endif
+}
+
+op_aes_xts_encrypt_t::~op_aes_xts_encrypt_t()
+{
+    assert(!encrypted);
+#ifdef WITH_OPENSSL
+    EVP_CIPHER_CTX_free(ctx);
+#endif
+    if (tmp)
+        free(tmp);
+}
+
+void op_aes_xts_encrypt_t::start(uint8_t *key, uint64_t start_offset, size_t block_size)
+{
+    assert(!encrypted);
+    this->start_offset = start_offset;
+    this->key = key;
+    this->block_size = block_size;
+    this->offset = 0;
+    this->encrypted = false;
+    this->tmp_pos = 0;
+    if (tmp && tmp_size != block_size)
+    {
+        free(tmp);
+        tmp = NULL;
+        tmp_size = 0;
+    }
+#ifdef WITH_OPENSSL
+    if (EVP_EncryptInit_ex(ctx, NULL, NULL, key, NULL) != 1)
+    {
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+#endif
+}
+
+void op_aes_xts_encrypt_t::encrypt_block(uint8_t *in, uint8_t *out)
+{
+#ifdef WITH_OPENSSL
+    uint8_t iv[16] = { 0 };
+    *((uint64_t*)iv) = start_offset + offset - offset%block_size;
+    if (EVP_EncryptInit_ex(ctx, NULL, NULL, NULL, iv) != 1)
+    {
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+    int actual_out = 0;
+    if (EVP_EncryptUpdate(ctx, out, &actual_out, in, block_size) != 1)
+    {
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+    assert(actual_out == block_size);
+#endif
+}
+
+void op_aes_xts_encrypt_t::update(uint8_t *in, size_t max_in, uint8_t *out, size_t max_out, size_t & done_in, size_t & done_out)
+{
+    // Fucking AES-XTS implementations (all of them) don't have streaming support,
+    // crafting IV to resume encryption is slow, so we have to accumulate a full block
+    // and encrypt it at once :-(
+    // And then we have to support consuming it in parts because it's simpler for the
+    // higher layers.
+    if (encrypted)
+    {
+        // Copy accumulated and encrypted output
+        assert(tmp);
+        if (max_out > block_size - tmp_pos)
+            max_out = block_size - tmp_pos;
+        memcpy(out, tmp + tmp_pos, max_out);
+        done_out += max_out;
+        tmp_pos += max_out;
+        if (tmp_pos >= block_size)
+            encrypted = false;
+    }
+    else if (max_in < block_size - offset%block_size)
+    {
+        // Just accumulate input
+        if (!tmp)
+        {
+            tmp = (uint8_t*)malloc_or_die(block_size);
+            tmp_size = block_size;
+        }
+        memcpy(tmp + offset%block_size, in, max_in);
+        done_in += max_in;
+        offset += max_in;
+    }
+    else if (max_out < block_size)
+    {
+        // Accumulate and encrypt input in <tmp>, then copy part of it to <out>
+        if (!tmp)
+        {
+            tmp = (uint8_t*)malloc_or_die(block_size);
+            tmp_size = block_size;
+        }
+        max_in = block_size - offset%block_size;
+        memcpy(tmp + offset%block_size, in, max_in);
+        encrypt_block(tmp, tmp);
+        encrypted = true;
+        memcpy(out, tmp, max_out);
+        tmp_pos = max_out;
+        done_in += max_in;
+        offset += max_in;
+        done_out += max_out;
+    }
+    else if (!(offset%block_size))
+    {
+        // Full block - simplest case
+        encrypt_block(in, out);
+        done_in += block_size;
+        offset += block_size;
+        done_out += block_size;
+    }
+    else
+    {
+        // Accumulate input and encrypt directly to <output>
+        assert(tmp);
+        max_in = block_size - offset%block_size;
+        memcpy(tmp + offset%block_size, in, max_in);
+        encrypt_block(tmp, out);
+        done_in += max_in;
+        offset += max_in;
+        done_out += block_size;
+    }
+}
+
+void destroy_aes_xts_encrypt(op_aes_xts_encrypt_t *encrypt_ctx)
+{
+    delete encrypt_ctx;
+}
+
+op_aes_xts_decrypt_t::op_aes_xts_decrypt_t()
+{
+#ifdef WITH_OPENSSL
+    if (!(ctx = EVP_CIPHER_CTX_new()))
+    {
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+    EVP_CIPHER_CTX_set_padding(ctx, 0);
+    if (EVP_DecryptInit_ex(ctx, EVP_aes_256_xts(), NULL, NULL, NULL) != 1)
+    {
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+#else
+    fprintf(stderr, "Error: Vitastor is built without encryption support\n");
+    abort();
+#endif
+}
+
+op_aes_xts_decrypt_t::~op_aes_xts_decrypt_t()
+{
+    assert(!decrypted);
+#ifdef WITH_OPENSSL
+    EVP_CIPHER_CTX_free(ctx);
+#endif
+    if (tmp)
+        free(tmp);
+}
+
+void op_aes_xts_decrypt_t::start(uint8_t **key_chain, size_t chain_size, void *key_indexes, uint64_t start_offset, size_t block_size)
+{
+    assert(!decrypted);
+    this->start_offset = start_offset;
+    this->key_chain = chain_size > 1 ? key_chain : 0;
+    this->chain_size = chain_size > 1 ? chain_size : 0;
+    this->key_indexes = chain_size > 1 ? key_indexes : NULL;
+    this->key_index_bytes = osd_op_rw_t::chain_info_bytes(chain_size);
+    assert(chain_size <= 1 || key_indexes != NULL);
+    this->block_size = block_size;
+    this->offset = 0;
+    this->tmp_pos = 0;
+    if (tmp && tmp_size != block_size)
+    {
+        free(tmp);
+        tmp = NULL;
+        tmp_size = 0;
+    }
+#ifdef WITH_OPENSSL
+    if (chain_size == 1 && key_chain[0] && EVP_DecryptInit_ex(ctx, NULL, NULL, key_chain[0], NULL) != 1)
+    {
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+#endif
+}
+
+void op_aes_xts_decrypt_t::decrypt_block(uint8_t *in, uint8_t *out)
+{
+    uint8_t *key = NULL;
+    if (chain_size)
+    {
+        uint32_t key_index = key_index_bytes == 1
+            ? ((uint8_t*)key_indexes)[offset/block_size]
+            : (key_index_bytes == 2
+                ? ((uint16_t*)key_indexes)[offset/block_size]
+                : (key_index_bytes == 4
+                    ? ((uint32_t*)key_indexes)[offset/block_size]
+                    : UINT32_MAX));
+        assert(key_index < chain_size);
+        key = key_chain[key_index];
+        if (!key)
+        {
+            if (in != out)
+                memcpy(out, in, block_size);
+            return;
+        }
+    }
+#ifdef WITH_OPENSSL
+    uint8_t iv[16] = { 0 };
+    *((uint64_t*)iv) = start_offset + offset - offset%block_size;
+    if (EVP_DecryptInit_ex(ctx, NULL, NULL, key, iv) != 1)
+    {
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+    int actual_out = 0;
+    if (EVP_DecryptUpdate(ctx, out, &actual_out, in, block_size) != 1)
+    {
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+    assert(actual_out == block_size);
+#endif
+}
+
+// out may be NULL, in this case all input is still decrypted to calculate checksums,
+// but part of it is skipped and not copied to out
+void op_aes_xts_decrypt_t::update(uint8_t *in, size_t max_in, uint8_t *out, size_t max_out, size_t & done_in, size_t & done_out)
+{
+    // Fucking AES-XTS implementations (all of them) don't have streaming support,
+    // crafting IV to resume decryption is slow, so we have to accumulate a full block
+    // and decrypt it at once :-(
+    // And then we have to support consuming it in parts because clients sometimes need
+    // fragmented output.
+    if (decrypted)
+    {
+        // Copy accumulated and decrypted output
+        assert(tmp);
+        if (max_out > block_size - tmp_pos)
+            max_out = block_size - tmp_pos;
+        if (out)
+            memcpy(out, tmp + tmp_pos, max_out);
+        done_out += max_out;
+        tmp_pos += max_out;
+        if (tmp_pos >= block_size)
+            decrypted = false;
+    }
+    else if (max_in < block_size - offset%block_size)
+    {
+        // Just accumulate input
+        if (!tmp)
+        {
+            tmp = (uint8_t*)malloc_or_die(block_size);
+            tmp_size = block_size;
+        }
+        memcpy(tmp + offset%block_size, in, max_in);
+        done_in += max_in;
+        offset += max_in;
+    }
+    else if (max_out < block_size || !out)
+    {
+        // Accumulate and decrypt input in <tmp>, then copy part of it to <out>
+        if (!tmp)
+        {
+            tmp = (uint8_t*)malloc_or_die(block_size);
+            tmp_size = block_size;
+        }
+        max_in = block_size - offset%block_size;
+        memcpy(tmp + offset%block_size, in, max_in);
+        decrypt_block(tmp, tmp);
+        decrypted = true;
+        if (out)
+            memcpy(out, tmp, max_out);
+        tmp_pos = max_out;
+        done_in += max_in;
+        offset += max_in;
+        done_out += max_out;
+    }
+    else if (!(offset%block_size))
+    {
+        // Full block - simplest case
+        if (out)
+            decrypt_block(in, out);
+        done_in += block_size;
+        offset += block_size;
+        done_out += block_size;
+    }
+    else
+    {
+        // Accumulate input and decrypt directly to <output>
+        assert(tmp);
+        max_in = block_size - offset%block_size;
+        memcpy(tmp + offset%block_size, in, max_in);
+        assert(out);
+        decrypt_block(tmp, out);
+        done_in += max_in;
+        offset += max_in;
+        done_out += block_size;
+    }
+}
+
+void destroy_aes_xts_decrypt(op_aes_xts_decrypt_t *decrypt_ctx)
+{
+    delete decrypt_ctx;
+}
+
+void osd_messenger_t::op_encrypted_copy_buf(osd_client_t *cl, uint8_t *enc_buf, size_t enc_len, uint8_t *plain, size_t plain_len, size_t & done_plain, size_t & done_enc)
+{
+    if (!cl->encrypt_ctx)
+    {
+        if (encrypt_ctx_pool.size())
+        {
+            cl->encrypt_ctx = encrypt_ctx_pool.back();
+            encrypt_ctx_pool.pop_back();
+        }
+        else
+            cl->encrypt_ctx = new op_aes_xts_encrypt_t();
+        assert(cl->write_op->enc->key_chain[0]);
+        cl->encrypt_ctx->start(cl->write_op->enc->key_chain[0], cl->write_op->req.rw.offset, cl->write_op->enc->bitmap_granularity);
+    }
+    while (done_enc < enc_len && (done_plain < plain_len || cl->encrypt_ctx->has_buffered()))
+    {
+        size_t done_in = 0;
+        size_t done_out = 0;
+        cl->encrypt_ctx->update(plain+done_plain, plain_len-done_plain, enc_buf+done_enc, enc_len-done_enc, done_in, done_out);
+        if (cl->write_csum_state && done_out > 0)
+            XXH3_64bits_update(cl->write_csum_state, enc_buf+done_enc, done_out);
+        done_enc += done_out;
+        cl->write_op_pos += done_in;
+        done_plain += done_in;
+    }
+}
+
+void osd_messenger_t::op_decrypted_copy_buf(osd_client_t *cl, uint8_t *enc_buf, size_t enc_len, uint8_t *plain, size_t plain_len, size_t & done_plain, size_t & done_enc)
+{
+    op_decrypt_start(cl);
+    while (done_plain < plain_len && done_enc < enc_len)
+    {
+        size_t done_in = 0;
+        size_t done_out = 0;
+        // plain == NULL means skip output
+        cl->decrypt_ctx->update(enc_buf+done_enc, enc_len-done_enc, plain ? plain+done_plain : NULL, plain_len-done_plain, done_in, done_out);
+        if (cl->read_csum_state && done_in > 0)
+            XXH3_64bits_update(cl->read_csum_state, enc_buf+done_enc, done_in);
+        done_enc += done_in;
+        cl->read_op_pos += done_out;
+        cl->read_op_inline_decrypt_in += done_in;
+        done_plain += done_out;
+    }
+}
+
+void osd_messenger_t::op_decrypt_start(osd_client_t* cl)
+{
+    if (!cl->decrypt_ctx)
+    {
+        if (decrypt_ctx_pool.size())
+        {
+            cl->decrypt_ctx = decrypt_ctx_pool.back();
+            decrypt_ctx_pool.pop_back();
+        }
+        else
+            cl->decrypt_ctx = new op_aes_xts_decrypt_t();
+        auto & enc = cl->read_op->enc;
+        assert(cl->read_op->req.hdr.opcode == OSD_OP_READ);
+        cl->decrypt_ctx->start(enc->key_chain, enc->chain_size,
+            (cl->read_op->req.rw.flags & OSD_OP_RETURN_CHAIN) ? (uint8_t*)cl->read_op->bitmap + enc->read_chain_bitmap_pos : 0,
+            cl->read_op->req.rw.offset, enc->bitmap_granularity);
+    }
+}
+
+void osd_messenger_t::op_decrypt_inline(osd_client_t* cl)
+{
+    op_decrypt_start(cl);
+    osd_op_t *op = cl->read_op;
+    size_t from_in = cl->read_op_inline_decrypt_in;
+    int i = 0;
+    while (i < op->iov.count && from_in >= op->iov.buf[i].iov_len)
+    {
+        from_in -= op->iov.buf[i].iov_len;
+        i++;
+    }
+    size_t from_out = cl->read_op_inline_decrypt_pos - OSD_PACKET_SIZE - op->reply.rw.bitmap_len;
+    int j = 0;
+    while (j < op->iov.count && from_out >= op->iov.buf[j].iov_len)
+    {
+        from_out -= op->iov.buf[j].iov_len;
+        j++;
+    }
+    while (i < op->iov.count && j < op->iov.count)
+    {
+        uint8_t *in = (uint8_t*)op->iov.buf[i].iov_base + from_in;
+        size_t in_len = op->iov.buf[i].iov_len - from_in;
+        uint8_t *out = (uint8_t*)op->iov.buf[j].iov_base + from_out;
+        size_t out_len = op->iov.buf[j].iov_len - from_out;
+        size_t done_in = 0;
+        size_t done_out = 0;
+        cl->decrypt_ctx->update(in, in_len, out, out_len, done_in, done_out);
+        if (done_in >= in_len)
+        {
+            i++;
+            from_in = 0;
+        }
+        else
+            from_in += done_in;
+        if (done_out >= out_len)
+        {
+            j++;
+            from_out = 0;
+        }
+        else
+            from_out += done_out;
+    }
+    assert(j >= op->iov.count);
+}
+
+void osd_messenger_t::op_decrypt_free(osd_client_t* cl)
+{
+    if (cl->decrypt_ctx)
+    {
+        if (decrypt_ctx_pool.size() > max_aes_xts_pool_size)
+            delete cl->decrypt_ctx;
+        else
+            decrypt_ctx_pool.push_back(cl->decrypt_ctx);
+        cl->decrypt_ctx = NULL;
+    }
+}
+
+void osd_messenger_t::op_encrypt_free(osd_client_t* cl)
+{
+    if (cl->encrypt_ctx)
+    {
+        if (encrypt_ctx_pool.size() > max_aes_xts_pool_size)
+            delete cl->encrypt_ctx;
+        else
+            encrypt_ctx_pool.push_back(cl->encrypt_ctx);
+        cl->encrypt_ctx = NULL;
+    }
+}

src/client/msgr_encrypt.h

@@ -0,0 +1,69 @@
+// Copyright (c) Vitaliy Filippov, 2026+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#include <stdint.h>
+
+#include "../util/xxh_x86dispatch.h"
+// WITH_OPENSSL is left to possibly support other crypto libraries
+#ifdef WITH_OPENSSL
+#include <openssl/conf.h>
+#include <openssl/evp.h>
+#include <openssl/err.h>
+#endif
+
+class op_aes_xts_encrypt_t
+{
+#ifdef WITH_OPENSSL
+    EVP_CIPHER_CTX *ctx = NULL;
+#endif
+    uint64_t start_offset = 0;
+    uint8_t *key = NULL;
+    size_t offset = 0;
+    size_t block_size = 0;
+    uint8_t *tmp = NULL;
+    size_t tmp_size = 0;
+    size_t tmp_pos = 0;
+    bool encrypted = false;
+
+    void encrypt_block(uint8_t *in, uint8_t *out);
+
+public:
+    op_aes_xts_encrypt_t();
+    ~op_aes_xts_encrypt_t();
+
+    inline bool has_buffered() { return encrypted; };
+    void start(uint8_t *key, uint64_t start_offset, size_t block_size);
+    void update(uint8_t *in, size_t max_in, uint8_t *out, size_t max_out, size_t & done_in, size_t & done_out);
+};
+
+void destroy_aes_xts_encrypt(op_aes_xts_encrypt_t *encrypt_ctx);
+
+class op_aes_xts_decrypt_t
+{
+#ifdef WITH_OPENSSL
+    EVP_CIPHER_CTX *ctx = NULL;
+#endif
+    uint64_t start_offset = 0;
+    uint8_t **key_chain = NULL;
+    size_t chain_size = 0;
+    void *key_indexes = NULL;
+    int key_index_bytes = 0;
+    size_t offset = 0;
+    size_t block_size = 0;
+    uint8_t *tmp = NULL;
+    size_t tmp_size = 0;
+    size_t tmp_pos = 0;
+    bool decrypted = false;
+
+    void decrypt_block(uint8_t *in, uint8_t *out);
+
+public:
+    op_aes_xts_decrypt_t();
+    ~op_aes_xts_decrypt_t();
+
+    inline bool has_buffered() { return decrypted; };
+    void start(uint8_t **key_chain, size_t chain_size, void *key_indexes, uint64_t start_offset, size_t block_size);
+    void update(uint8_t *in, size_t max_in, uint8_t *out, size_t max_out, size_t & done_in, size_t & done_out);
+};
+
+void destroy_aes_xts_decrypt(op_aes_xts_decrypt_t *decrypt_ctx);

src/client/msgr_op.cpp

@@ -8,7 +8,6 @@
 osd_op_t::~osd_op_t()
 {
     assert(!bs_op);
-    assert(!op_data);
     if (bitmap_buf)
     {
         free(bitmap_buf);
@@ -23,6 +22,14 @@ osd_op_t::~osd_op_t()
         // So we don't reuse it, but free it every time
         free(buf);
     }
+    if (enc_buf)
+    {
+        free(enc_buf);
+    }
+    if (op_data)
+    {
+        free(op_data);
+    }
 }
 
 bool osd_op_t::is_recovery_related()

src/client/msgr_op.h

@@ -3,6 +3,8 @@
 
 #pragma once
 
+#include <memory>
+
 #include <sys/uio.h>
 #include <stdint.h>
 #include <stdio.h>
@@ -16,6 +18,8 @@
 
 #define OSD_OP_INLINE_BUF_COUNT 16
 
+#define AES_256_XTS_KEY_SIZE 64
+
 // Kind of a vector with small-list-optimisation
 struct osd_op_buf_list_t
 {
@@ -152,11 +156,24 @@ struct blockstore_op_t;
 
 struct osd_primary_op_data_t;
 
+struct osd_op_enc_t
+{
+    // Keys may contain more information in the future, like encryption algorithm and key ID
+    // In this case, key_chain will become inode_key_t* with inode_key_t also being a structure
+    // Currently all keys are required to be 512 bit (64 byte) long, for AES-256-XTS
+    // Raw pointers are convenient for messenger code; external users may use shared_ptr aliasing
+    // to implement complex freeing of osd_op_enc_t along with their external inode cache info
+    uint8_t** key_chain = NULL;
+    size_t chain_size = 0;
+    uint32_t read_chain_bitmap_pos = 0;
+    uint32_t bitmap_granularity = 0;
+};
+
 struct __attribute__((visibility("default"))) osd_op_t
 {
     timespec tv_begin = { 0 }, tv_end = { 0 };
     uint64_t op_type = OSD_OP_IN;
-    int peer_fd;
+    uint64_t client_id = 0;
     osd_any_op_t req;
     osd_any_reply_t reply;
     blockstore_op_t *bs_op = NULL;
@@ -165,8 +182,11 @@ struct __attribute__((visibility("default"))) osd_op_t
     void *bitmap = NULL;
     unsigned bitmap_len = 0;
     unsigned bmp_data = 0;
-    void *bitmap_buf = NULL;
+    uint8_t *bitmap_buf = NULL;
     void *rmw_buf = NULL;
+    std::shared_ptr<osd_op_enc_t> enc;
+    uint8_t *enc_buf = NULL;
+    uint64_t csum = 0; // network layer checksum
     osd_primary_op_data_t* op_data = NULL;
     std::function<void(osd_op_t*)> callback;
 
@@ -176,4 +196,5 @@ struct __attribute__((visibility("default"))) osd_op_t
     void cancel();
 
     bool is_recovery_related();
+    uint64_t calc_data_checksum();
 };

src/client/msgr_rdma.cpp

@@ -187,6 +187,8 @@ std::vector<msgr_rdma_context_t*> msgr_rdma_context_t::create_all(const std::vec
     ibv_device **raw_dev_list = NULL;
     ibv_device **dev_list = NULL;
     ibv_device *single_list[2] = {};
+    int up_ports = 0;
+    int single_port_num = 0;
 
     raw_dev_list = dev_list = ibv_get_device_list(NULL);
     if (!dev_list || !*dev_list)
@@ -221,6 +223,7 @@ std::vector<msgr_rdma_context_t*> msgr_rdma_context_t::create_all(const std::vec
         dev_list = single_list;
     }
 
+retry:
     for (int i = 0; dev_list[i]; ++i)
     {
         auto dev = dev_list[i];
@@ -258,6 +261,9 @@ std::vector<msgr_rdma_context_t*> msgr_rdma_context_t::create_all(const std::vec
                 fprintf(stderr, "RDMA device %s port %d GID %d does not exist\n", ibv_get_device_name(dev), port_num, sel_gid_index);
                 continue;
             }
+            up_ports++;
+            single_port_num = port_num;
+            single_list[0] = dev;
             uint32_t port_mtu = sel_mtu ? sel_mtu : ibv_mtu_to_bytes(portinfo.active_mtu);
 #ifdef IBV_ADVISE_MR_ADVICE_PREFETCH_NO_FAULT
             if (sel_gid_index < 0)
@@ -298,6 +304,14 @@ cleanup_dev:
         ibv_close_device(context);
     }
 
+    if (!ret.size() && up_ports == 1 && dev_list != single_list)
+    {
+        // Auto-select the only available device/port if there is only one
+        dev_list = single_list;
+        sel_port_num = single_port_num;
+        goto retry;
+    }
+
 cleanup:
     if (raw_dev_list)
         ibv_free_device_list(raw_dev_list);
@@ -493,7 +507,7 @@ int msgr_rdma_connection_t::connect(msgr_rdma_address_t *dest)
     return 0;
 }
 
-bool osd_messenger_t::connect_rdma(int peer_fd, std::string rdma_address, uint64_t client_max_msg)
+bool osd_messenger_t::connect_rdma(uint64_t client_id, std::string rdma_address, uint64_t client_max_msg)
 {
     // Try to connect to the peer using RDMA
     msgr_rdma_address_t addr;
@@ -503,12 +517,12 @@ bool osd_messenger_t::connect_rdma(int peer_fd, std::string rdma_address, uint64
         {
             client_max_msg = rdma_max_msg;
         }
-        auto cl = clients.at(peer_fd);
+        auto cl = clients.at(client_id);
         msgr_rdma_context_t *selected_ctx = choose_rdma_context(cl);
         if (!selected_ctx)
         {
             if (log_level > 0)
-                fprintf(stderr, "No RDMA context for peer %d, using only TCP\n", cl->peer_fd);
+                fprintf(stderr, "No RDMA context for peer %ju, using only TCP\n", client_id);
             return false;
         }
         msgr_rdma_connection_t *rdma_conn = msgr_rdma_connection_t::create(selected_ctx, rdma_max_send, rdma_max_recv, rdma_max_sge, client_max_msg);
@@ -519,14 +533,13 @@ bool osd_messenger_t::connect_rdma(int peer_fd, std::string rdma_address, uint64
             {
                 delete rdma_conn;
                 fprintf(
-                    stderr, "Failed to connect RDMA queue pair to %s (client %d)\n",
-                    addr.to_string().c_str(), peer_fd
+                    stderr, "Failed to connect RDMA queue pair to %s (client %ju)\n",
+                    addr.to_string().c_str(), client_id
                 );
             }
             else
             {
                 // Remember connection, but switch to RDMA only after sending the configuration response
-                auto cl = clients.at(peer_fd);
                 cl->rdma_conn = rdma_conn;
                 cl->peer_state = PEER_RDMA_CONNECTING;
                 return true;
@@ -540,7 +553,7 @@ static void try_send_rdma_wr(osd_client_t *cl, ibv_sge *sge, int op_sge)
 {
     ibv_send_wr *bad_wr = NULL;
     ibv_send_wr wr = {
-        .wr_id = (uint64_t)(cl->peer_fd*2+1),
+        .wr_id = cl->client_id,
         .sg_list = sge,
         .num_sge = op_sge,
         .opcode = IBV_WR_SEND,
@@ -555,28 +568,6 @@ static void try_send_rdma_wr(osd_client_t *cl, ibv_sge *sge, int op_sge)
     cl->rdma_conn->cur_send++;
 }
 
-static int try_send_rdma_copy(osd_client_t *cl, uint8_t *dst, int dst_len)
-{
-    auto rc = cl->rdma_conn;
-    int total_dst_len = dst_len;
-    while (dst_len > 0 && rc->send_pos < cl->send_list.size())
-    {
-        iovec & iov = cl->send_list[rc->send_pos];
-        uint32_t len = (uint32_t)(iov.iov_len-rc->send_buf_pos < dst_len
-            ? iov.iov_len-rc->send_buf_pos : dst_len);
-        memcpy(dst, (uint8_t*)iov.iov_base+rc->send_buf_pos, len);
-        dst += len;
-        dst_len -= len;
-        rc->send_buf_pos += len;
-        if (rc->send_buf_pos >= iov.iov_len)
-        {
-            rc->send_pos++;
-            rc->send_buf_pos = 0;
-        }
-    }
-    return total_dst_len-dst_len;
-}
-
 void osd_messenger_t::try_send_rdma(osd_client_t *cl)
 {
     auto rc = cl->rdma_conn;
@@ -599,10 +590,44 @@ void osd_messenger_t::try_send_rdma(osd_client_t *cl)
     while (!rc->send_out_full && copied > 0 && rc->cur_send < rc->max_send)
     {
         dst = (uint8_t*)rc->send_out.buf + rc->send_out_pos;
-        dst_len = (rc->send_out_pos < rc->send_out_size ? rc->send_out_size-rc->send_out_pos : rc->send_done_pos-rc->send_out_pos);
+        if (rc->send_out_pos >= rc->send_done_pos)
+        {
+            dst_len = rc->send_out_size-rc->send_out_pos;
+            if (dst_len < 4096)
+            {
+                // free end of the buffer is too small, skip
+                rc->send_out_pos = 0;
+                if (rc->send_out_pos >= rc->send_done_pos)
+                    rc->send_out_full = true;
+                if (!rc->send_sizes.size())
+                {
+                    rc->send_done_pos += dst_len;
+                    rc->send_out_full = false;
+                    if (rc->send_done_pos == rc->send_out_size)
+                        rc->send_done_pos = 0;
+                }
+                else
+                    rc->send_sizes.back() += dst_len;
+                continue;
+            }
+        }
+        else
+        {
+            dst_len = rc->send_done_pos-rc->send_out_pos;
+            if (dst_len < 4096)
+            {
+                // too small buffer, stop
+                break;
+            }
+        }
         if (dst_len > rc->max_msg)
             dst_len = rc->max_msg;
-        copied = try_send_rdma_copy(cl, dst, dst_len);
+        copied = copy_ops_to(cl, dst, dst_len);
+        if (cl->io_error)
+        {
+            stop_client(cl->client_id);
+            return;
+        }
         if (copied > 0)
         {
             rc->send_out_pos += copied;
@@ -618,6 +643,7 @@ void osd_messenger_t::try_send_rdma(osd_client_t *cl)
             };
             try_send_rdma_wr(cl, &sge, 1);
             rc->send_sizes.push_back(copied);
+            cl->send_free_ops.push_back(NULL); // end marker
         }
     }
 }
@@ -631,7 +657,7 @@ static void try_recv_rdma_wr(osd_client_t *cl, void *buf)
     };
     ibv_recv_wr *bad_wr = NULL;
     ibv_recv_wr wr = {
-        .wr_id = (uint64_t)(cl->peer_fd*2),
+        .wr_id = cl->client_id,
         .sg_list = &sge,
         .num_sge = 1,
     };
@@ -688,37 +714,35 @@ void osd_messenger_t::handle_rdma_events(msgr_rdma_context_t *rdma_context)
         event_count = ibv_poll_cq(rdma_context->cq, RDMA_EVENTS_AT_ONCE, wc);
         for (int i = 0; i < event_count; i++)
         {
-            int client_id = wc[i].wr_id >> 1;
-            bool is_send = wc[i].wr_id & 1;
+            uint64_t client_id = wc[i].wr_id;
+            bool is_send = wc[i].opcode == IBV_WC_SEND;
             auto cl_it = clients.find(client_id);
             if (cl_it == clients.end())
             {
                 continue;
             }
             osd_client_t *cl = cl_it->second;
+            if (cl->peer_state == PEER_STOPPED)
+            {
+                continue;
+            }
             auto rc = cl->rdma_conn;
             if (wc[i].status != IBV_WC_SUCCESS)
             {
-                fprintf(stderr, "RDMA work request failed for client %d", client_id);
+                fprintf(stderr, "RDMA work request failed for client %ju", client_id);
                 if (cl->osd_num)
                 {
                     fprintf(stderr, " (OSD %ju)", cl->osd_num);
                 }
                 fprintf(stderr, " with status: %s, stopping client\n", ibv_wc_status_str(wc[i].status));
                 stop_client(client_id);
-                clear_immediate_ops(client_id);
                 continue;
             }
             if (!is_send)
             {
-                // Reset OSD ping state - client is obviously alive
-                cl->ping_time_remaining = 0;
-                cl->idle_time_remaining = osd_idle_timeout;
                 rc->cur_recv--;
                 if (!handle_read_buffer(cl, rc->recv_buffers[rc->next_recv_buf], wc[i].byte_len))
                 {
-                    // handle_read_buffer may stop the client
-                    clear_immediate_ops(client_id);
                     continue;
                 }
                 try_recv_rdma_wr(cl, rc->recv_buffers[rc->next_recv_buf]);
@@ -727,54 +751,27 @@ void osd_messenger_t::handle_rdma_events(msgr_rdma_context_t *rdma_context)
             else
             {
                 rc->cur_send--;
-                uint64_t sent_size = rc->send_sizes.at(0);
-                rc->send_sizes.erase(rc->send_sizes.begin(), rc->send_sizes.begin()+1);
+                // byte_len is not filled for send operations
+                uint64_t sent_size = rc->send_sizes.front();
+                rc->send_sizes.pop_front();
                 rc->send_done_pos += sent_size;
                 rc->send_out_full = false;
                 if (rc->send_done_pos == rc->send_out_size)
                     rc->send_done_pos = 0;
                 assert(rc->send_done_pos < rc->send_out_size);
-                int send_pos = 0, send_buf_pos = 0;
-                while (sent_size > 0)
+                while (cl->send_free_ops.front())
                 {
-                    if (sent_size >= cl->send_list.at(send_pos).iov_len)
-                    {
-                        sent_size -= cl->send_list[send_pos].iov_len;
-                        send_pos++;
-                    }
-                    else
-                    {
-                        send_buf_pos = sent_size;
-                        sent_size = 0;
-                    }
+                    delete cl->send_free_ops.front();
+                    cl->send_free_ops.pop_front();
                 }
-                assert(rc->send_pos >= send_pos);
-                if (rc->send_pos == send_pos)
+                cl->send_free_ops.pop_front();
+                if ((cl->proto_csum_status & MSGR_CSUM_NEG) && !cl->write_op && !cl->write_ops.size())
                 {
-                    rc->send_buf_pos -= send_buf_pos;
-                }
-                rc->send_pos -= send_pos;
-                for (int i = 0; i < send_pos; i++)
-                {
-                    if (cl->outbox[i].flags & MSGR_SENDP_FREE)
-                    {
-                        // Reply fully sent
-                        delete cl->outbox[i].op;
-                    }
-                }
-                if (send_pos > 0)
-                {
-                    cl->send_list.erase(cl->send_list.begin(), cl->send_list.begin()+send_pos);
-                    cl->outbox.erase(cl->outbox.begin(), cl->outbox.begin()+send_pos);
-                }
-                if (send_buf_pos > 0)
-                {
-                    cl->send_list[0].iov_base = (uint8_t*)cl->send_list[0].iov_base + send_buf_pos;
-                    cl->send_list[0].iov_len -= send_buf_pos;
+                    // Checksums negotiated, enable
+                    cl->proto_csum_status = cl->proto_csum_status & (~MSGR_CSUM_NEG);
                 }
                 try_send_rdma(cl);
             }
         }
     } while (event_count > 0);
-    handle_immediate_ops();
 }

src/client/msgr_rdma.h

@@ -8,8 +8,11 @@
 #include <infiniband/verbs.h>
 #include <string>
 #include <vector>
+#include <deque>
 #include "addr_util.h"
 
+struct osd_op_t;
+
 struct msgr_rdma_address_t
 {
     ibv_gid gid;
@@ -72,11 +75,11 @@ struct msgr_rdma_connection_t
     int cur_send = 0, cur_recv = 0;
     int send_pos = 0, send_buf_pos = 0;
     int next_recv_buf = 0;
-    std::vector<void*> recv_buffers;
+    std::vector<uint8_t*> recv_buffers;
     msgr_rdma_buf_t recv_buf;
-    std::vector<uint64_t> send_sizes;
+    std::deque<uint64_t> send_sizes;
     msgr_rdma_buf_t send_out;
-    int send_out_pos = 0, send_done_pos = 0, send_out_size = 0;
+    size_t send_out_pos = 0, send_done_pos = 0, send_out_size = 0;
     bool send_out_full = false;
 
     ~msgr_rdma_connection_t();

src/client/msgr_rdmacm.cpp

@@ -11,7 +11,7 @@
 struct rdmacm_connecting_t
 {
     rdma_cm_id *cmid = NULL;
-    int peer_fd = -1;
+    uint64_t client_id = 0;
     osd_num_t peer_osd = 0;
     std::string addr;
     sockaddr_storage parsed_addr = {};
@@ -117,9 +117,9 @@ void osd_messenger_t::handle_rdmacm_events()
             auto cli_it = rdmacm_connections.find(ev->id);
             if (cli_it != rdmacm_connections.end())
             {
-                fprintf(stderr, "Received %s event for peer %d, closing connection\n",
-                    event_type_name, cli_it->second->peer_fd);
-                stop_client(cli_it->second->peer_fd);
+                fprintf(stderr, "Received %s event for client %ju, closing connection\n",
+                    event_type_name, cli_it->second->client_id);
+                stop_client(cli_it->second->client_id);
             }
             else if (rdmacm_connecting.find(ev->id) != rdmacm_connecting.end())
             {
@@ -265,14 +265,6 @@ msgr_rdma_context_t* osd_messenger_t::rdmacm_create_qp(rdma_cm_id *cmid)
 
 void osd_messenger_t::rdmacm_accept(rdma_cm_event *ev)
 {
-    // Make a fake FD (FIXME: do not use FDs for identifying clients!)
-    int fake_fd = socket(AF_INET, SOCK_STREAM, 0);
-    if (fake_fd < 0)
-    {
-        fprintf(stderr, "Failed to allocate a fake socket for RDMA-CM client: %s (code %d)\n", strerror(errno), errno);
-        rdma_destroy_id(ev->id);
-        return;
-    }
     auto rdma_context = rdmacm_create_qp(ev->id);
     if (!rdma_context)
     {
@@ -297,12 +289,12 @@ void osd_messenger_t::rdmacm_accept(rdma_cm_event *ev)
     // Wait for RDMA_CM_ESTABLISHED, and enable the connection only after it
     auto conn = new rdmacm_connecting_t;
     conn->cmid = ev->id;
-    conn->peer_fd = fake_fd;
+    conn->client_id = next_client_id++;
     conn->parsed_addr = *(sockaddr_storage*)rdma_get_peer_addr(ev->id);
     conn->rdma_context = rdma_context;
     rdmacm_set_conn_timeout(conn);
     rdmacm_connecting[ev->id] = conn;
-    fprintf(stderr, "[OSD %ju] new client %d: connection from %s via RDMA-CM\n", this->osd_num, conn->peer_fd,
+    fprintf(stderr, "[OSD %ju] new client %ju: connection from %s via RDMA-CM\n", this->osd_num, conn->client_id,
         addr_to_string(conn->parsed_addr).c_str());
 }
 
@@ -332,8 +324,6 @@ void osd_messenger_t::rdmacm_on_connect_peer_error(rdma_cm_id *cmid, int res)
     auto peer_osd = conn->peer_osd;
     if (conn->timeout_id >= 0)
         tfd->clear_timer(conn->timeout_id);
-    if (conn->peer_fd >= 0)
-        close(conn->peer_fd);
     if (conn->rdma_context)
         conn->rdma_context->reserve_cqe(-rdma_max_send-rdma_max_recv);
     if (conn->cmid)
@@ -354,7 +344,7 @@ void osd_messenger_t::rdmacm_on_connect_peer_error(rdma_cm_id *cmid, int res)
         else
         {
             // TCP is disabled
-            on_connect_peer(peer_osd, res == 0 ? -EINVAL : (res > 0 ? -res : res));
+            on_connect_peer(peer_osd, res == 0 ? -EINVAL : (res > 0 ? -res : res), 0);
         }
     }
 }
@@ -365,7 +355,7 @@ void osd_messenger_t::rdmacm_try_connect_peer(uint64_t peer_osd, const std::stri
     if (!string_to_addr(addr, false, rdmacm_port, &sa))
     {
         fprintf(stderr, "Address %s is invalid\n", addr.c_str());
-        on_connect_peer(peer_osd, -EINVAL);
+        on_connect_peer(peer_osd, -EINVAL, 0);
         return;
     }
     rdma_cm_id *cmid = NULL;
@@ -376,17 +366,7 @@ void osd_messenger_t::rdmacm_try_connect_peer(uint64_t peer_osd, const std::stri
         if (!disable_tcp)
             try_connect_peer_tcp(peer_osd, addr.c_str(), fallback_tcp_port);
         else
-            on_connect_peer(peer_osd, res);
-        return;
-    }
-    // Make a fake FD (FIXME: do not use FDs for identifying clients!)
-    int fake_fd = socket(AF_INET, SOCK_STREAM, 0);
-    if (fake_fd < 0)
-    {
-        int res = -errno;
-        rdma_destroy_id(cmid);
-        // Can't create socket, pointless to try TCP
-        on_connect_peer(peer_osd, res);
+            on_connect_peer(peer_osd, res, 0);
         return;
     }
     if (log_level > 0)
@@ -394,7 +374,7 @@ void osd_messenger_t::rdmacm_try_connect_peer(uint64_t peer_osd, const std::stri
     auto conn = new rdmacm_connecting_t;
     rdmacm_connecting[cmid] = conn;
     conn->cmid = cmid;
-    conn->peer_fd = fake_fd;
+    conn->client_id = next_client_id++;
     conn->peer_osd = peer_osd;
     conn->addr = addr;
     conn->parsed_addr = sa;
@@ -511,13 +491,13 @@ void osd_messenger_t::rdmacm_established(rdma_cm_event *ev)
     auto cl = new osd_client_t();
     cl->peer_addr = conn->parsed_addr;
     cl->peer_port = conn->rdmacm_port;
-    cl->peer_fd = conn->peer_fd;
+    cl->client_id = conn->client_id;
     cl->peer_state = PEER_RDMA;
     cl->connect_timeout_id = -1;
     cl->osd_num = peer_osd;
-    cl->in_buf = malloc_or_die(receive_buffer_size);
+    cl->in_buf = (uint8_t*)malloc_or_die(receive_buffer_size);
     cl->rdma_conn = rc;
-    clients[conn->peer_fd] = cl;
+    clients[conn->client_id] = cl;
     if (conn->timeout_id >= 0)
     {
         tfd->clear_timer(conn->timeout_id);