Add detailed security feature documentation

Seems slightly overcomplicated...

.gitea/workflows/test.yml

@@ -360,6 +360,78 @@ jobs:
           echo ""
         done
 
+  test_dump_load:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: /root/vitastor/tests/test_dump_load.sh
+    - name: Print logs
+      if: always() && steps.test.outcome == 'failure'
+      run: |
+        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
+          echo "-------- $i --------"
+          cat $i
+          echo ""
+        done
+
+  test_dump_load_32k:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: TEST_NAME=32k OSD_ARGS="--data_csum_type crc32c --csum_block_size 32k" OFFSET_ARGS="$OSD_ARGS" /root/vitastor/tests/test_dump_load.sh
+    - name: Print logs
+      if: always() && steps.test.outcome == 'failure'
+      run: |
+        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
+          echo "-------- $i --------"
+          cat $i
+          echo ""
+        done
+
+  test_old_dump_load:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: OLD=1 /root/vitastor/tests/test_dump_load.sh
+    - name: Print logs
+      if: always() && steps.test.outcome == 'failure'
+      run: |
+        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
+          echo "-------- $i --------"
+          cat $i
+          echo ""
+        done
+
+  test_dump_load_old_32k:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: TEST_NAME=old_32k OLD=1 OSD_ARGS="--data_csum_type crc32c --csum_block_size 32k" OFFSET_ARGS="$OSD_ARGS" /root/vitastor/tests/test_dump_load.sh
+    - name: Print logs
+      if: always() && steps.test.outcome == 'failure'
+      run: |
+        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
+          echo "-------- $i --------"
+          cat $i
+          echo ""
+        done
+
   test_old_interrupted_rebalance:
     runs-on: ubuntu-latest
     needs: build
@@ -1584,6 +1656,24 @@ jobs:
           echo ""
         done
 
+  test_resize_last:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: /root/vitastor/tests/test_resize_last.sh
+    - name: Print logs
+      if: always() && steps.test.outcome == 'failure'
+      run: |
+        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
+          echo "-------- $i --------"
+          cat $i
+          echo ""
+        done
+
   test_resize_auto:
     runs-on: ubuntu-latest
     needs: build
@@ -1620,6 +1710,24 @@ jobs:
           echo ""
         done
 
+  test_old_resize_last:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: OLD=1 /root/vitastor/tests/test_resize_last.sh
+    - name: Print logs
+      if: always() && steps.test.outcome == 'failure'
+      run: |
+        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
+          echo "-------- $i --------"
+          cat $i
+          echo ""
+        done
+
   test_old_resize_auto:
     runs-on: ubuntu-latest
     needs: build

CMakeLists.txt

@@ -2,7 +2,7 @@ cmake_minimum_required(VERSION 2.8...3.30)
 
 project(vitastor)
 
-set(VITASTOR_VERSION "3.0.12")
+set(VITASTOR_VERSION "3.0.15")
 
 include(CTest)
 

csi/Makefile

@@ -1,4 +1,4 @@
-VITASTOR_VERSION ?= v3.0.12
+VITASTOR_VERSION ?= v3.0.15
 
 all: build push
 

csi/deploy/004-csi-nodeplugin.yaml

@@ -49,7 +49,7 @@ spec:
             capabilities:
               add: ["SYS_ADMIN"]
             allowPrivilegeEscalation: true
-          image: vitalif/vitastor-csi:v3.0.12
+          image: vitalif/vitastor-csi:v3.0.15
           args:
             - "--node=$(NODE_ID)"
             - "--endpoint=$(CSI_ENDPOINT)"

csi/deploy/007-csi-provisioner.yaml

@@ -121,7 +121,7 @@ spec:
             privileged: true
             capabilities:
               add: ["SYS_ADMIN"]
-          image: vitalif/vitastor-csi:v3.0.12
+          image: vitalif/vitastor-csi:v3.0.15
           args:
             - "--node=$(NODE_ID)"
             - "--endpoint=$(CSI_ENDPOINT)"

csi/src/config.go

@@ -5,7 +5,7 @@ package vitastor
 
 const (
     vitastorCSIDriverName    = "csi.vitastor.io"
-    vitastorCSIDriverVersion = "3.0.12"
+    vitastorCSIDriverVersion = "3.0.15"
 )
 
 // Config struct fills the parameters of request or user input

debian/changelog

@@ -1,4 +1,4 @@
-vitastor (3.0.12-1) unstable; urgency=medium
+vitastor (3.0.15-1) unstable; urgency=medium
 
   * Bugfixes
 

debian/rules

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

debian/vitastor-build.sh

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

docker/Makefile

@@ -1,4 +1,4 @@
-VITASTOR_VERSION ?= v3.0.12
+VITASTOR_VERSION ?= v3.0.15
 
 all: build push
 

docker/etc/vitastor/docker.conf

@@ -4,7 +4,7 @@
 #
 
 # Desired Vitastor version
-VITASTOR_VERSION=v3.0.12
+VITASTOR_VERSION=v3.0.15
 
 # Additional arguments for all containers
 # For example, you may want to specify a custom logging driver here

docs/config/monitor.en.md

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

docs/config/monitor.ru.md

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

docs/config/security.en.md

@@ -10,13 +10,36 @@ These parameters affect your Vitastor installation security and apply to OSDs, m
 
 Most of them can be set in /etc/vitastor/vitastor.conf and in etcd, but don't support online modification.
 
+All certificate and private key parameters mentioned may contain a path to a PEM file or just
+a PEM string with certificate or a private key. In the latter case, the string must begin with
+"-----BEGIN CERTIFICATE-----" or "-----BEGIN PRIVATE KEY-----".
+
+- [use_perms](#use_perms)
+- [cert](#cert)
+- [pkey](#pkey)
+- [etcd_ca](#etcd_ca)
+- [client_ca](#client_ca)
+- [osd_ca](#osd_ca)
+- [mon_ca](#mon_ca)
+- [antietcd_cert](#antietcd_cert)
+- [antietcd_key](#antietcd_key)
+- [etcd_proxy.urls](#etcd_proxyurls)
+- [etcd_proxy.cert](#etcd_proxycert)
+- [etcd_proxy.key](#etcd_proxykey)
+- [etcd_proxy.ca](#etcd_proxyca)
+- [osd_cert](#osd_cert)
+- [osd_pkey](#osd_pkey)
+- [api_cert](#api_cert)
+- [api_pkey](#api_pkey)
 - [etcd_client_cert](#etcd_client_cert)
 - [etcd_client_key](#etcd_client_key)
-- [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)
+- [proto_checksums](#proto_checksums)
+- [force_proto_checksums](#force_proto_checksums)
+- [max_cipher_pool_size](#max_cipher_pool_size)
 - [vault_url](#vault_url)
 - [vault_secret_api_path](#vault_secret_api_path)
 - [vault_client_cert](#vault_client_cert)
@@ -25,54 +48,195 @@ Most of them can be set in /etc/vitastor/vitastor.conf and in etcd, but don't su
 - [vault_timeout_ms](#vault_timeout_ms)
 - [vault_error_timeout_sec](#vault_error_timeout_sec)
 - [vault_refresh_leeway_sec](#vault_refresh_leeway_sec)
-- [max_cipher_pool_size](#max_cipher_pool_size)
 
-## etcd_client_cert
+## use_perms
+
+- Type: boolean
+- Default: false
+
+Enable client permissions in a Vitastor cluster, including Antietcd built into the Monitor.
+Requires configured encryption. Also note that separate Antietcd requires separate configuration
+to use permissions (see [security documentation](../intro/security.en.md) for details).
+
+## cert
 
 - Type: string
 
-Client 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-----".
+Client certificate of the current Vitastor user. Required for Vitastor protocol encryption.
+Must be signed with [client_ca](#client_ca). Also used as the client certificate for etcd/Antietcd
+connections by default.
 
-## etcd_client_key
+## pkey
 
 - Type: string
 
-Private key for etcd_client_cert (also a file or a PEM string).
+Private key of the current Vitastor user.
 
 ## 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.
+Trusted TLS CA to verify etcd server certificate. Or just the etcd server's
+certificate itself - it's fine to use it for etcd_ca.
+
+## client_ca
+
+- Type: string
+
+Trusted TLS CA to verify Vitastor client certificates.
+Mandatory for Vitastor protocol encryption.
+
+## osd_ca
+
+- Type: string
+
+Trusted TLS CA to verify Vitastor OSD certificates. Also mandatory for Vitastor protocol
+encryption. Must be different from client_ca. May be equal to osd_cert - different OSDs
+don't require separate certificates at the moment because their permissions don't differ.
+
+## mon_ca
+
+- Type: string
+
+Trusted TLS CA to verify Vitastor Monitor certificates. Used only for separate Antietcd,
+not required when a monitor built-in Antietcd is used. May be equal to mon_client_etcd_cert.
+
+## antietcd_cert
+
+- Type: string
+
+Server TLS certificate for Antietcd built into the Monitor.
+
+## antietcd_key
+
+- Type: string
+
+Private key for antietcd_cert.
+
+## etcd_proxy.urls
+
+- Type: string or array of strings
+
+etcd URLs for Antietcd etcd proxy mode.
+See [Mon as Etcd proxy](../intro/security.en.md#mon-as-etcd-proxy) for details.
+
+## etcd_proxy.cert
+
+- Type: string
+
+Client certificate for Antietcd connections to etcd in proxy mode.
+
+## etcd_proxy.key
+
+- Type: string
+
+Private key for etcd_proxy.cert.
+
+## etcd_proxy.ca
+
+- Type: string
+
+Trusted TLS CA to verify etcd server certificate when connecting to it from Antietcd.
+
+## osd_cert
+
+- Type: string
+
+Vitastor OSD server certificate. Required for Vitastor protocol encryption. May be equal
+to [osd_ca](#osd_ca) - all OSDs share the same permission set for now. Also used as the client
+certificate for connections from OSD to etcd/Antietcd by default.
+
+## osd_pkey
+
+- Type: string
+
+Private key for osd_cert.
+
+## api_cert
+
+- Type: string
+
+Server TLS certificate for [vitastor-cli serve](../usage/cli.en.md#serve) API server.
+
+## api_pkey
+
+- Type: string
+
+Private key for api_cert.
+
+## etcd_client_cert
+
+- Type: string
+
+Client TLS certificate to use for connections from Vitastor clients to etcd/Antietcd if you don't want
+to use the common client certificate [cert](#cert).
+
+## etcd_client_key
+
+- Type: string
+
+Private key for etcd_client_cert.
 
 ## osd_etcd_client_cert
 
 - Type: string
 
-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.
+Client TLS certificate to use for connections from Vitastor OSDs to etcd/Antietcd if you don't want
+to use the common OSD certificate [osd_cert](#osd_cert).
 
 ## osd_etcd_client_key
 
 - Type: string
 
-Same as [etcd_client_key](#etcd_client_key), but only for OSDs.
+Private key for osd_etcd_client_cert.
 
 ## mon_etcd_client_cert
 
 - Type: string
 
-Same as [etcd_client_cert](#etcd_client_cert), but only for Vitastor monitors.
+Client TLS certificate to use for connections from Vitastor Monitors to etcd/Antietcd - required
+if you don't use the built-in Antietcd. In case you use it Monitor has direct access to Antietcd data
+and doesn't require any connection.
 
 ## mon_etcd_client_key
 
 - Type: string
 
-Same as [etcd_client_key](#etcd_client_key), but only for Vitastor monitors.
+Private key for mon_etcd_client_cert.
+
+## proto_checksums
+
+- Type: string
+- Default: payload
+
+One of "full", "payload", "gcm", "none":
+- "full" means calculate and verify transport level checksums from the full message data
+  including the header - recommended for unencrypted setups.
+- "payload" enables checksums only for the actual read/write data, but skips them for message
+  headers - recommended for encrypted setups because headers are already protected by AES-GCM.
+- "gcm" disables checksums and enables AES-GCM encryption of the whole messages including headers
+  and data - AES-GCM already includes MAC which is actually a stronger checksum. This option is
+  slower and is only recommended for untrusted networks.
+- "none" disables transport level checksums at all.
+
+## force_proto_checksums
+
+- Type: string
+
+To allow older clients to connect to a Vitastor cluster with enabled checksums, Vitastor OSDs
+allow clients to downgrade their proto_checksums by default. force_proto_checksums sets the
+minimum security level allowed for connecting clients. When encryption is disabled, default
+force_proto_checksums is none and clients without checksums are allowed. With enabled
+encryption, force_proto_checksums becomes "payload" by default to block unauthenticated data
+on the transport level.
+
+## max_cipher_pool_size
+
+- Type: integer
+- Default: 256
+
+Maximum number of OpenSSL cipher contexts cached in OSD memory, counted separately
+for each cipher and for encryption/decryption. Probably doesn't require modification.
 
 ## vault_url
 
@@ -103,14 +267,14 @@ Vault v1 secret API mount path to use.
 
 - 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.
+Client TLS certificate to use for Vault connections if you don't want to use the common Vitastor
+client certificate [cert](#cert) which is also used for Vault connections by default.
 
 ## vault_client_key
 
 - Type: string
 
-Private key for vault_client_cert (also a file or a PEM string).
+Private key for the vault_client_cert certificate.
 
 ## vault_ca
 
@@ -140,11 +304,3 @@ Time (in seconds) to wait before retrying after receiving an error from Vault.
 
 Extra time (in seconds) before real Vault token lease_timeout to refresh it, just
 in case of system clock drift.
-
-## max_cipher_pool_size
-
-- Type: integer
-- Default: 256
-
-Maximum number of OpenSSL cipher contexts cached in OSD memory, counted separately
-for each cipher and for encryption/decryption. Probably doesn't require modification.

docs/config/security.ru.md

@@ -12,13 +12,36 @@ OSD, мониторами и клиентами.
 Большая их часть может задаваться в /etc/vitastor/vitastor.conf и в etcd, но не
 поддерживает онлайн-изменение.
 
+Все параметры сертификатов и закрытых ключей могут быть путём к файлу или просто
+строкой с сертификатом в формате PEM. В последнем случае строка должна начинаться с
+"-----BEGIN CERTIFICATE-----" или "-----BEGIN PRIVATE KEY-----".
+
+- [use_perms](#use_perms)
+- [cert](#cert)
+- [pkey](#pkey)
+- [etcd_ca](#etcd_ca)
+- [client_ca](#client_ca)
+- [osd_ca](#osd_ca)
+- [mon_ca](#mon_ca)
+- [antietcd_cert](#antietcd_cert)
+- [antietcd_key](#antietcd_key)
+- [etcd_proxy.urls](#etcd_proxyurls)
+- [etcd_proxy.cert](#etcd_proxycert)
+- [etcd_proxy.key](#etcd_proxykey)
+- [etcd_proxy.ca](#etcd_proxyca)
+- [osd_cert](#osd_cert)
+- [osd_pkey](#osd_pkey)
+- [api_cert](#api_cert)
+- [api_pkey](#api_pkey)
 - [etcd_client_cert](#etcd_client_cert)
 - [etcd_client_key](#etcd_client_key)
-- [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)
+- [proto_checksums](#proto_checksums)
+- [force_proto_checksums](#force_proto_checksums)
+- [max_cipher_pool_size](#max_cipher_pool_size)
 - [vault_url](#vault_url)
 - [vault_secret_api_path](#vault_secret_api_path)
 - [vault_client_cert](#vault_client_cert)
@@ -27,56 +50,199 @@ OSD, мониторами и клиентами.
 - [vault_timeout_ms](#vault_timeout_ms)
 - [vault_error_timeout_sec](#vault_error_timeout_sec)
 - [vault_refresh_leeway_sec](#vault_refresh_leeway_sec)
-- [max_cipher_pool_size](#max_cipher_pool_size)
 
-## etcd_client_cert
+## use_perms
+
+- Тип: булево (да/нет)
+- Значение по умолчанию: false
+
+Включает клиентские привилегии в кластере Vitastor, в том числе во встроенном в мониторе Antietcd.
+Требует настроенного шифрования протокола. Также обратите внимание, что отдельно установленный Antietcd
+требует отдельной настройки привилегий (подробности смотрите в [документации безопасности](../intro/security.ru.md)).
+
+## cert
 
 - Тип: строка
 
-Клиентский 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-----".
+Клиентский сертификат текущего пользователя Vitastor. Требуется для шифрования протокола Vitastor.
+Должен быть подписан [client_ca](#client_ca). Также по умолчанию используется как клиентский
+сертификат для подключения к etcd/Antietcd и Vault.
 
-## etcd_client_key
+## pkey
 
 - Тип: строка
 
-Закрытый ключ для сертификата etcd_client_cert (также путь к файлу или PEM строка).
+Закрытый ключ текущего пользователя Vitastor.
 
 ## 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.
+Либо же просто сам сертификат сервера etcd - его можно использовать как etcd_ca.
+
+## client_ca
+
+- Тип: строка
+
+Доверенный TLS-сертификат для проверки сертификатов клиентов Vitastor.
+Требуется для шифрования протокола Vitastor.
+
+## osd_ca
+
+- Тип: строка
+
+Доверенный TLS-сертификат для проверки сертификатов OSD Vitastor. Также обязателен
+для шифрования протокола Vitastor. Должен отличаться от client_ca. Может быть равен
+osd_cert - разные OSD не требуют разных сертификатов, потому что на данный момент
+привилегии разных OSD никак не отличаются.
+
+## mon_ca
+
+- Тип: строка
+
+Доверенный TLS-сертификат для проверки сертификатов мониторов Vitastor. Используется
+только отдельно установленным Antietcd, не требуется при использовании встроенного в монитор
+Antietcd. Может быть равен mon_client_etcd_cert.
+
+## antietcd_cert
+
+- Тип: строка
+
+Серверный TLS-сертификат для Antietcd, встроенного в монитор.
+
+## antietcd_key
+
+- Тип: строка
+
+Закрытый ключ для сертификата antietcd_cert.
+
+## etcd_proxy.urls
+
+- Тип: строка или массив строк
+
+Адреса etcd для режима Antietcd etcd-прокси.
+Смотрите подробности в разделе [Mon в роли Etcd proxy](../intro/security.ru.md#mon-в-роли-etcd-proxy).
+
+## etcd_proxy.cert
+
+- Тип: строка
+
+Клиентский сертификат для подключений от Antietcd к etcd в режиме прокси.
+
+## etcd_proxy.key
+
+- Тип: строка
+
+Закрытый ключ для сертификата etcd_proxy.cert.
+
+## etcd_proxy.ca
+
+- Тип: строка
+
+Доверенный TLS-сертификат для проверки сертификата сервера etcd при подключениях от Antietcd.
+
+## osd_cert
+
+- Тип: строка
+
+Сертификат сервера Vitastor OSD. Требуется для шифрования протокола Vitastor. Может быть равен
+[osd_ca](#osd_ca) - все OSD на данный момент имеют одинаковые привилегии. Также по умолчанию
+используется как клиентский сертификат для подключения от OSD к etcd/Antietcd.
+
+## osd_pkey
+
+- Тип: строка
+
+Закрытый ключ для сертификата osd_cert.
+
+## api_cert
+
+- Тип: строка
+
+Серверный TLS-сертификат для API-сервера [vitastor-cli serve](../usage/cli.ru.md#serve).
+
+## api_pkey
+
+- Тип: строка
+
+Закрытый ключ для сертификата api_cert.
+
+## etcd_client_cert
+
+- Тип: строка
+
+Клиентский TLS сертификат для подключений от клиентов Vitastor к etcd/Antietcd, если вы не хотите
+использовать общий клиентский сертификат [cert](#cert).
+
+## etcd_client_key
+
+- Тип: строка
+
+Закрытый ключ для сертификата etcd_client_cert.
 
 ## osd_etcd_client_cert
 
 - Тип: строка
 
-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.
+Клиентский TLS сертификат для подключений от Vitastor OSD к etcd/Antietcd, если вы не хотите
+использовать общий сертификат OSD [osd_cert](#osd_cert).
 
 ## osd_etcd_client_key
 
 - Тип: строка
 
-Аналогично [etcd_client_key](#etcd_client_key), но только для OSD.
+Закрытый ключ для сертификата osd_etcd_client_cert.
 
 ## mon_etcd_client_cert
 
 - Тип: строка
 
-Аналогично [etcd_client_cert](#etcd_client_cert), но только для мониторов Vitastor.
+Клиентский TLS сертификат для подключений от мониторов Vitastor к etcd/Antietcd - требуется, если
+вы не используете встроенный в монитор Antietcd. Если вы используете его, то монитор и так имеет
+прямой доступ к данным Antietcd и не требует никаких соединений.
 
 ## mon_etcd_client_key
 
 - Тип: строка
 
-Аналогично [etcd_client_key](#etcd_client_key), но только для мониторов Vitastor.
+Закрытый ключ для сертификата mon_etcd_client_cert.
+
+## proto_checksums
+
+- Тип: строка
+- Значение по умолчанию: payload
+
+Одно из значений "full", "payload", "gcm" и "none":
+- "full" означает расчёт и проверку контрольных сумм на транспортном уровне от полных сообщений,
+  включая их заголовки и данные - рекомендуется для кластеров без шифрования.
+- "payload" включает контрольные суммы только для данных сообщений, но пропускает заголовки -
+  такая настройка рекомендуется для кластеров с включённым шифрованием, потому что в них заголовки
+  и так защищены шифрованием AES-GCM.
+- "gcm" отключает контрольные суммы и включает шифрование полных сообщений включая заголовки и
+  данные - AES-GCM уже включает в себя MAC, который по сути является криптостойкой контрольной
+  суммой. Такая настройка медленнее и рекомендуется только для недоверенных сетей.
+- "none" полностью отключает контрольные суммы на транспортном уровне.
+
+## force_proto_checksums
+
+- Тип: строка
+
+Чтобы старые клиенты Vitastor могли подключаться к кластеру с включёнными контрольными
+суммами, Vitastor OSD по умолчанию разрешают клиентам отключать контрольные суммы
+данных (proto_checksums). Настройка force_proto_checksums задаёт минимальный уровень
+безопасности, разрешённый для подключающихся клиентов. Когда шифрование отключено,
+force_proto_checksums по умолчанию равно none и подключения клиентов без контрольных
+сумм разрешаются. При включённом шифровании значение по умолчанию force_proto_checksums
+становится "payload", чтобы блокировать подключения с неаутентифицированными данными.
+
+## max_cipher_pool_size
+
+- Тип: целое число
+- Значение по умолчанию: 256
+
+Максимальное количество кэшируемых в памяти OSD контекстов шифра OpenSSL, учитываемое
+отдельно для каждого шифра и для шифрования и расшифровки. Вряд ли требует изменения.
 
 ## vault_url
 
@@ -106,14 +272,14 @@ OSD, клиенты и мониторы должны иметь разные п
 
 - 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.
+Клиентский TLS сертификат для подключений к Vault на тот случай, если вы не хотите использовать
+общий сертификат клиента Vitastor [cert](#cert), используемый для подключений к Vault по умолчанию.
 
 ## vault_client_key
 
 - Тип: строка
 
-Закрытый ключ для сертификата vault_client_cert (также путь к файлу или PEM строка).
+Закрытый ключ для сертификата vault_client_cert.
 
 ## vault_ca
 
@@ -144,11 +310,3 @@ OSD, клиенты и мониторы должны иметь разные п
 
 Зазор времени (в секундах), чтобы обновлять токены Vault чуть раньше их реального
 lease_timeout, на случай "ухода" системных часов.
-
-## max_cipher_pool_size
-
-- Тип: целое число
-- Значение по умолчанию: 256
-
-Максимальное количество кэшируемых в памяти OSD контекстов шифра OpenSSL, учитываемое
-отдельно для каждого шифра и для шифрования и расшифровки. Вряд ли требует изменения.

docs/config/src/make.js

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

docs/config/src/monitor.yml

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

docs/config/src/security.en.md

@@ -3,3 +3,7 @@
 These parameters affect your Vitastor installation security and apply to OSDs, monitors and clients.
 
 Most of them can be set in /etc/vitastor/vitastor.conf and in etcd, but don't support online modification.
+
+All certificate and private key parameters mentioned may contain a path to a PEM file or just
+a PEM string with certificate or a private key. In the latter case, the string must begin with
+"-----BEGIN CERTIFICATE-----" or "-----BEGIN PRIVATE KEY-----".

docs/config/src/security.ru.md

@@ -5,3 +5,7 @@ OSD, мониторами и клиентами.
 
 Большая их часть может задаваться в /etc/vitastor/vitastor.conf и в etcd, но не
 поддерживает онлайн-изменение.
+
+Все параметры сертификатов и закрытых ключей могут быть путём к файлу или просто
+строкой с сертификатом в формате PEM. В последнем случае строка должна начинаться с
+"-----BEGIN CERTIFICATE-----" или "-----BEGIN PRIVATE KEY-----".

docs/config/src/security.yml

@@ -1,49 +1,203 @@
-- name: etcd_client_cert
+- name: use_perms
+  type: bool
+  default: false
+  info: |
+    Enable client permissions in a Vitastor cluster, including Antietcd built into the Monitor.
+    Requires configured encryption. Also note that separate Antietcd requires separate configuration
+    to use permissions (see [security documentation](../intro/security.en.md) for details).
+  info_ru: |
+    Включает клиентские привилегии в кластере Vitastor, в том числе во встроенном в мониторе Antietcd.
+    Требует настроенного шифрования протокола. Также обратите внимание, что отдельно установленный Antietcd
+    требует отдельной настройки привилегий (подробности смотрите в [документации безопасности](../intro/security.ru.md)).
+- name: cert
   type: string
   info: |
-    Client 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-----".
+    Client certificate of the current Vitastor user. Required for Vitastor protocol encryption.
+    Must be signed with [client_ca](#client_ca). Also used as the client certificate for etcd/Antietcd
+    connections by default.
   info_ru: |
-    Клиентский 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
+    Клиентский сертификат текущего пользователя Vitastor. Требуется для шифрования протокола Vitastor.
+    Должен быть подписан [client_ca](#client_ca). Также по умолчанию используется как клиентский
+    сертификат для подключения к etcd/Antietcd и Vault.
+- name: pkey
   type: string
-  info: Private key 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).
+  info: Private key of the current Vitastor user.
+  info_ru: Закрытый ключ текущего пользователя Vitastor.
 - name: etcd_ca
   type: string
   info: |
-    Trusted TLS CA to verify etcd server certificate. May be path to a file,
-    directory or just a PEM string with certificate.
+    Trusted TLS CA to verify etcd server certificate. Or just the etcd server's
+    certificate itself - it's fine to use it for etcd_ca.
   info_ru: |
     Доверенный корневой TLS-сертификат для проверки сертификата сервера etcd.
-    Может быть путём к файлу, директории или просто строкой с сертификатом в
-    формате PEM.
+    Либо же просто сам сертификат сервера etcd - его можно использовать как etcd_ca.
+- name: client_ca
+  type: string
+  info: |
+    Trusted TLS CA to verify Vitastor client certificates.
+    Mandatory for Vitastor protocol encryption.
+  info_ru: |
+    Доверенный TLS-сертификат для проверки сертификатов клиентов Vitastor.
+    Требуется для шифрования протокола Vitastor.
+- name: osd_ca
+  type: string
+  info: |
+    Trusted TLS CA to verify Vitastor OSD certificates. Also mandatory for Vitastor protocol
+    encryption. Must be different from client_ca. May be equal to osd_cert - different OSDs
+    don't require separate certificates at the moment because their permissions don't differ.
+  info_ru: |
+    Доверенный TLS-сертификат для проверки сертификатов OSD Vitastor. Также обязателен
+    для шифрования протокола Vitastor. Должен отличаться от client_ca. Может быть равен
+    osd_cert - разные OSD не требуют разных сертификатов, потому что на данный момент
+    привилегии разных OSD никак не отличаются.
+- name: mon_ca
+  type: string
+  info: |
+    Trusted TLS CA to verify Vitastor Monitor certificates. Used only for separate Antietcd,
+    not required when a monitor built-in Antietcd is used. May be equal to mon_client_etcd_cert.
+  info_ru: |
+    Доверенный TLS-сертификат для проверки сертификатов мониторов Vitastor. Используется
+    только отдельно установленным Antietcd, не требуется при использовании встроенного в монитор
+    Antietcd. Может быть равен mon_client_etcd_cert.
+- name: antietcd_cert
+  type: string
+  info: Server TLS certificate for Antietcd built into the Monitor.
+  info_ru: Серверный TLS-сертификат для Antietcd, встроенного в монитор.
+- name: antietcd_key
+  type: string
+  info: Private key for antietcd_cert.
+  info_ru: Закрытый ключ для сертификата antietcd_cert.
+- name: etcd_proxy.urls
+  type: string or array of strings
+  type_ru: строка или массив строк
+  info: |
+    etcd URLs for Antietcd etcd proxy mode.
+    See [Mon as Etcd proxy](../intro/security.en.md#mon-as-etcd-proxy) for details.
+  info_ru: |
+    Адреса etcd для режима Antietcd etcd-прокси.
+    Смотрите подробности в разделе [Mon в роли Etcd proxy](../intro/security.ru.md#mon-в-роли-etcd-proxy).
+- name: etcd_proxy.cert
+  type: string
+  info: Client certificate for Antietcd connections to etcd in proxy mode.
+  info_ru: Клиентский сертификат для подключений от Antietcd к etcd в режиме прокси.
+- name: etcd_proxy.key
+  type: string
+  info: Private key for etcd_proxy.cert.
+  info_ru: Закрытый ключ для сертификата etcd_proxy.cert.
+- name: etcd_proxy.ca
+  type: string
+  info: Trusted TLS CA to verify etcd server certificate when connecting to it from Antietcd.
+  info_ru: Доверенный TLS-сертификат для проверки сертификата сервера etcd при подключениях от Antietcd.
+- name: osd_cert
+  type: string
+  info: |
+    Vitastor OSD server certificate. Required for Vitastor protocol encryption. May be equal
+    to [osd_ca](#osd_ca) - all OSDs share the same permission set for now. Also used as the client
+    certificate for connections from OSD to etcd/Antietcd by default.
+  info_ru: |
+    Сертификат сервера Vitastor OSD. Требуется для шифрования протокола Vitastor. Может быть равен
+    [osd_ca](#osd_ca) - все OSD на данный момент имеют одинаковые привилегии. Также по умолчанию
+    используется как клиентский сертификат для подключения от OSD к etcd/Antietcd.
+- name: osd_pkey
+  type: string
+  info: Private key for osd_cert.
+  info_ru: Закрытый ключ для сертификата osd_cert.
+- name: api_cert
+  type: string
+  info: Server TLS certificate for [vitastor-cli serve](../usage/cli.en.md#serve) API server.
+  info_ru: Серверный TLS-сертификат для API-сервера [vitastor-cli serve](../usage/cli.ru.md#serve).
+- name: api_pkey
+  type: string
+  info: Private key for api_cert.
+  info_ru: Закрытый ключ для сертификата api_cert.
+- name: etcd_client_cert
+  type: string
+  info: |
+    Client TLS certificate to use for connections from Vitastor clients to etcd/Antietcd if you don't want
+    to use the common client certificate [cert](#cert).
+  info_ru: |
+    Клиентский TLS сертификат для подключений от клиентов Vitastor к etcd/Antietcd, если вы не хотите
+    использовать общий клиентский сертификат [cert](#cert).
+- name: etcd_client_key
+  type: string
+  info: Private key for etcd_client_cert.
+  info_ru: Закрытый ключ для сертификата etcd_client_cert.
 - name: osd_etcd_client_cert
   type: string
   info: |
-    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.
+    Client TLS certificate to use for connections from Vitastor OSDs to etcd/Antietcd if you don't want
+    to use the common OSD certificate [osd_cert](#osd_cert).
   info_ru: |
-    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.
+    Клиентский TLS сертификат для подключений от Vitastor OSD к etcd/Antietcd, если вы не хотите
+    использовать общий сертификат OSD [osd_cert](#osd_cert).
 - name: osd_etcd_client_key
   type: string
-  info: 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.
+  info: Private key for osd_etcd_client_cert.
+  info_ru: Закрытый ключ для сертификата osd_etcd_client_cert.
 - name: mon_etcd_client_cert
   type: string
-  info: 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.
+  info: |
+    Client TLS certificate to use for connections from Vitastor Monitors to etcd/Antietcd - required
+    if you don't use the built-in Antietcd. In case you use it Monitor has direct access to Antietcd data
+    and doesn't require any connection.
+  info_ru: |
+    Клиентский TLS сертификат для подключений от мониторов Vitastor к etcd/Antietcd - требуется, если
+    вы не используете встроенный в монитор Antietcd. Если вы используете его, то монитор и так имеет
+    прямой доступ к данным Antietcd и не требует никаких соединений.
 - name: mon_etcd_client_key
   type: string
-  info: 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.
+  info: Private key for mon_etcd_client_cert.
+  info_ru: Закрытый ключ для сертификата mon_etcd_client_cert.
+- name: proto_checksums
+  type: string
+  default: payload
+  info: |
+    One of "full", "payload", "gcm", "none":
+    - "full" means calculate and verify transport level checksums from the full message data
+      including the header - recommended for unencrypted setups.
+    - "payload" enables checksums only for the actual read/write data, but skips them for message
+      headers - recommended for encrypted setups because headers are already protected by AES-GCM.
+    - "gcm" disables checksums and enables AES-GCM encryption of the whole messages including headers
+      and data - AES-GCM already includes MAC which is actually a stronger checksum. This option is
+      slower and is only recommended for untrusted networks.
+    - "none" disables transport level checksums at all.
+  info_ru: |
+    Одно из значений "full", "payload", "gcm" и "none":
+    - "full" означает расчёт и проверку контрольных сумм на транспортном уровне от полных сообщений,
+      включая их заголовки и данные - рекомендуется для кластеров без шифрования.
+    - "payload" включает контрольные суммы только для данных сообщений, но пропускает заголовки -
+      такая настройка рекомендуется для кластеров с включённым шифрованием, потому что в них заголовки
+      и так защищены шифрованием AES-GCM.
+    - "gcm" отключает контрольные суммы и включает шифрование полных сообщений включая заголовки и
+      данные - AES-GCM уже включает в себя MAC, который по сути является криптостойкой контрольной
+      суммой. Такая настройка медленнее и рекомендуется только для недоверенных сетей.
+    - "none" полностью отключает контрольные суммы на транспортном уровне.
+- name: force_proto_checksums
+  type: string
+  info: |
+    To allow older clients to connect to a Vitastor cluster with enabled checksums, Vitastor OSDs
+    allow clients to downgrade their proto_checksums by default. force_proto_checksums sets the
+    minimum security level allowed for connecting clients. When encryption is disabled, default
+    force_proto_checksums is none and clients without checksums are allowed. With enabled
+    encryption, force_proto_checksums becomes "payload" by default to block unauthenticated data
+    on the transport level.
+  info_ru: |
+    Чтобы старые клиенты Vitastor могли подключаться к кластеру с включёнными контрольными
+    суммами, Vitastor OSD по умолчанию разрешают клиентам отключать контрольные суммы
+    данных (proto_checksums). Настройка force_proto_checksums задаёт минимальный уровень
+    безопасности, разрешённый для подключающихся клиентов. Когда шифрование отключено,
+    force_proto_checksums по умолчанию равно none и подключения клиентов без контрольных
+    сумм разрешаются. При включённом шифровании значение по умолчанию force_proto_checksums
+    становится "payload", чтобы блокировать подключения с неаутентифицированными данными.
+- name: max_cipher_pool_size
+  type: int
+  default: 256
+  info: |
+    Maximum number of OpenSSL cipher contexts cached in OSD memory, counted separately
+    for each cipher and for encryption/decryption. Probably doesn't require modification.
+  info_ru: |
+    Максимальное количество кэшируемых в памяти OSD контекстов шифра OpenSSL, учитываемое
+    отдельно для каждого шифра и для шифрования и расшифровки. Вряд ли требует изменения.
 - name: vault_url
   type: string
   info: |
@@ -81,15 +235,15 @@
 - 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.
+    Client TLS certificate to use for Vault connections if you don't want to use the common Vitastor
+    client certificate [cert](#cert) which is also used for Vault connections by default.
   info_ru: |
-    Клиентский TLS сертификат для подключений к Vault. Как и [etcd_client_cert](#etcd_client_cert),
-    может быть путём к файлу или просто PEM-строкой с сертификатом.
+    Клиентский TLS сертификат для подключений к Vault на тот случай, если вы не хотите использовать
+    общий сертификат клиента Vitastor [cert](#cert), используемый для подключений к Vault по умолчанию.
 - name: vault_client_key
   type: string
-  info: Private key for 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).
+  info: Private key for the vault_client_cert certificate.
+  info_ru: Закрытый ключ для сертификата vault_client_cert.
 - name: vault_ca
   type: string
   info: |
@@ -120,12 +274,3 @@
   info_ru: |
     Зазор времени (в секундах), чтобы обновлять токены Vault чуть раньше их реального
     lease_timeout, на случай "ухода" системных часов.
-- name: max_cipher_pool_size
-  type: int
-  default: 256
-  info: |
-    Maximum number of OpenSSL cipher contexts cached in OSD memory, counted separately
-    for each cipher and for encryption/decryption. Probably doesn't require modification.
-  info_ru: |
-    Максимальное количество кэшируемых в памяти OSD контекстов шифра OpenSSL, учитываемое
-    отдельно для каждого шифра и для шифрования и расшифровки. Вряд ли требует изменения.

docs/installation/docker.en.md

@@ -26,9 +26,9 @@ at Vitastor Kubernetes operator: https://github.com/Antilles7227/vitastor-operat
 The instruction is very simple.
 
 1. Download a Docker image of the desired version: \
-   `docker pull vitalif/vitastor:v3.0.12`
+   `docker pull vitalif/vitastor:v3.0.15`
 2. Install scripts to the host system: \
-   `docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.0.12 install.sh`
+   `docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.0.15 install.sh`
 3. Reload udev rules: \
    `udevadm control --reload-rules`
 4. Enable the vitastor-host service: \

docs/installation/docker.ru.md

@@ -25,9 +25,9 @@ Vitastor можно установить в Docker/Podman. При этом etcd,
 Инструкция по установке максимально простая.
 
 1. Скачайте Docker-образ желаемой версии: \
-   `docker pull vitalif/vitastor:v3.0.12`
+   `docker pull vitalif/vitastor:v3.0.15`
 2. Установите скрипты в хост-систему командой: \
-   `docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.0.12 install.sh`
+   `docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.0.15 install.sh`
 3. Перезагрузите правила udev: \
    `udevadm control --reload-rules`
 4. Включите сервис vitastor-host: \

docs/intro/quickstart.en.md

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

docs/intro/quickstart.ru.md

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

docs/intro/security.en.md

@@ -0,0 +1,657 @@
+[Documentation](../../README.md#documentation) → Introduction → Security in Vitastor
+
+-----
+
+[Читать на русском](security.ru.md)
+
+# Security in Vitastor
+
+- [Overview](#overview)
+- [Quick setup](#quick-setup)
+- Principles of operation
+  - [etcd transport encryption (TLS)](#etcd-transport-encryption-tls)
+  - [OSD transport encryption (AES-GCM)](#osd-transport-encryption-aes-gcm)
+  - [End-to-end image data encryption (AES-XTS)](#end-to-end-image-data-encryption-aes-xts)
+  - [Certificate-based authentication](#certificate-based-authentication)
+  - [Users and access rights](#users-and-access-rights)
+  - [etcd privileges](#etcd-privileges)
+- Manual setup
+  - [Configuring OSD transport encryption](#configuring-osd-transport-encryption)
+  - etcd/Antietcd setup options
+    - [Mon with embedded Antietcd](#mon-with-embedded-antietcd)
+    - [Mon as an Etcd proxy](#mon-as-an-etcd-proxy)
+    - [Mon with a separate Antietcd Proxy](#mon-with-a-separate-antietcd-proxy)
+    - [Standalone Antietcd without etcd](#standalone-antietcd-without-etcd)
+  - [Vault/OpenBao setup](#vaultopenbao-setup)
+    - [Vault setup example](#vault-setup-example)
+- Lists of allowed operations
+  - [etcd data access rights](#etcd-data-access-rights)
+  - [OSD data access rights](#osd-data-access-rights)
+  - [API access rights](#api-access-rights)
+- [Encryption performance](#encryption-performance)
+
+## Overview
+
+Starting from version 3.1.0, Vitastor provides full data protection:
+control plane protection (etcd), data plane protection (OSDs), and end-to-end data encryption.
+
+- Control plane protection:
+  - etcd transport encryption (TLS)
+  - Authentication via client TLS (X.509) certificates
+  - Access control of clients to etcd data
+- Data plane protection:
+  - Full AES-GCM encryption of OSD transport (similar to TLS, but faster)
+  - Alternatively, AES-GCM encryption of just operation headers with data checksums using a secret "salt"
+  - Authentication via client TLS (X.509) certificates
+  - Access control of clients on the OSD side
+- End-to-end encryption:
+  - Data is encrypted using AES-XTS on the client side, the Vitastor cluster has no access to plaintext data
+  - AES-XTS keys can be stored in etcd or in an external Vault/OpenBao
+
+All features are optional and disabled in the simplest configuration. By default, only
+transport-level data checksums ([proto_checksums](../config/security.en.md#proto_checksums)=payload)
+are enabled for clients that support them (>= 3.1.0). For older clients, connections
+without data checksums are allowed by default ([force_proto_checksums](../config/security.en.md#force_proto_checksums) is empty).
+
+For a quick setup, jump to the [Quick setup](#quick-setup) section.
+
+Descriptions of all security-related parameters can be found [here](../config/security.en.md).
+
+## Quick setup
+
+For a quick setup, use the `/usr/lib/vitastor/mon/make-etcd` script:
+
+1. Log in to the node where the first monitor and etcd will be located.
+2. Create `/etc/vitastor/vitastor.conf` with minimal parameters: etcd_address,
+   osd_network and, if you want to enable privileges, use_perms (note `https://`
+   in etcd addresses):
+   ```
+   {
+       "etcd_address": ["https://10.0.0.10:2379","https://10.0.0.11:2379","https://10.0.0.12:2379"],
+       "osd_network": "10.0.0.0/24",
+       "use_perms": true
+   }
+   ```
+3. Run `/usr/lib/vitastor/mon/make-etcd` without parameters or with the `--antietcd-only`
+   parameter if you want to initialize the cluster with Antietcd only, without etcd.
+4. The script will generate all necessary certificates and offer to copy them to the other
+   monitor nodes (agree!).
+5. Log in to all other monitor nodes and repeat the `/usr/lib/vitastor/mon/make-etcd` call there.
+6. If you also have nodes with OSDs only (without monitors), run the following command to
+   copy only the required configuration to these nodes:
+   ```
+   /usr/lib/vitastor/mon/make-etcd --copy-to-osd osdnode1,osdnode2,...
+   ```
+
+After that, you can proceed with OSD initialization.
+
+If you want to understand the setup in more detail, read the [Principles of operation](#principles-of-operation)
+and [Manual setup](#manual-setup) sections below.
+
+## Principles of operation
+
+### etcd transport encryption (TLS)
+
+Possible setups:
+- Without encryption (http)
+- With encryption (https)
+- With encryption and client certificate authentication. Either the same certificate
+  used for authentication on the OSD side (`cert`+`pkey` / `osd_cert`+`osd_pkey`)
+  is used, or a separately specified certificate (`etcd_client_cert`+`etcd_client_key`).
+
+### OSD transport encryption (AES-GCM)
+
+Possible setups:
+- Unencrypted transport without checksums: `proto_checksums=none`.
+- Unencrypted transport with data checksums: `proto_checksums=payload` (may be omitted,
+  this is the default value). It's allowed to disable checksums on the client side, or
+  use an older client that does not support checksums. If you want to block connections
+  from clients without checksums, use the option `force_proto_checksums=payload`.
+- Header-only encryption with data checksums: activated when the options
+  `cert`, `pkey`, `osd_ca` are set on the client side and `osd_cert`, `osd_pkey`, `osd_ca`, `client_ca`
+  on the OSD side, with `proto_checksums=payload`. In this mode, disabling checksums on the client
+  side is forbidden by default, i.e. `force_proto_checksums=payload` is used.
+- Full transport encryption of all traffic: same as the previous option, but with `proto_checksums=gcm`.
+  In this case, clients are by default allowed to downgrade to checksums only, but this
+  can also be forbidden via `force_proto_checksums=gcm`. This is the slowest setup and
+  it's only recommended for insecure (public) networks. In particular, full traffic
+  encryption together with end-to-end AES-XTS image encryption encrypts data twice.
+
+Encryption uses the AES-256-GCM algorithm and a custom simplified key exchange protocol,
+fully analogous to TLS 1.3 ECDHE.
+
+### End-to-end image data encryption (AES-XTS)
+
+The Vitastor client supports encrypting each image's data with its own key. In this case,
+data is encrypted by the client before sending it to OSDs and OSDs can't see it in plain.
+The encryption key can be changed when cloning/creating image snapshots. For example,
+you can make a base VM image (say, Debian Linux) unencrypted, but have encrypted client VM
+images inheriting from it.
+
+Image encryption keys can be stored in etcd or in an external Vault. In the latter case,
+etcd only stores key IDs and Vitastor cluster can't decrypt the data at all. To use
+Vault, create an image with the `--enc_key vault:ID` option, specify vault_url and vault_ca
+options in the configuration, create accounts for all clients in Vault, and grant them access
+to the required v1 secrets.
+
+Once again, if AES-XTS is used together with full traffic encryption (`proto_checksums=gcm`),
+image data is encrypted twice — first with AES-XTS, and then with AES-GCM. Use it only if
+you are completely paranoid :-).
+
+### Certificate-based authentication
+
+When encryption is enabled, Vitastor clients, OSDs, and monitors authenticate via certificates
+for both etcd (Antietcd) and OSD connections.
+
+Separate certificates must be used for OSDs and monitors — either self-signed, or signed
+by separate CAs (`osd_ca` and `mon_ca`). All OSDs can use the same certificate, and all
+monitors can also use the same certificate, since the privileges of different OSDs or
+different monitors do not differ (theoretically, one could differentiate OSD certificates
+by pool, but there has been no need for this so far).
+
+Also, a monitor certificate may not be needed at all if Antietcd is embedded into the monitor
+itself. In this case, the monitor already has access to all etcd data directly in memory.
+
+### Users and access rights
+
+When transport encryption is disabled, Vitastor operates without access control, i.e.,
+any cluster client has full access to both the management layer and the data layer. This
+option is suitable for dedicated trusted storage networks.
+
+When OSD transport encryption is enabled (at least for headers), you can enable access
+rights by turning on the `use_perms=true` option. When this option is enabled, each user
+can perform only the operations that they are permitted, and even OSDs and monitors are
+also forbidden from performing "unnecessary" operations.
+
+Each user (or administrator) must have their own certificate signed by a common root
+certificate for clients (`client_ca`), with a Common Name equal to the user name.
+Privilege settings are stored in etcd. OSDs and monitors don't need user accounts;
+they authenticate via separate certificates.
+
+User privileges are stored in etcd data under the keys `/vitastor/config/user/<name>`.
+The following is defined per user in this key:
+- Type:
+  - Client (`type=client` or omitted) — can only read and modify explicitly permitted images.
+  - Administrator (`type=admin`) — can read and modify all images, and also administer the
+    cluster: view overall statistics and status, create and delete OSDs, etc.
+- List of group names the user is a member of.
+
+Images have the following properties:
+- Owner (owner) — the user name that is allowed to both read and modify the image
+- Owner group (owner_group) — the owner group name
+- Reader group (reader_group) — the name of the group of users allowed to read the image
+
+And there is also a property on the pool:
+- Creator group (creator_group) — the name of the group of users allowed to create images in the pool
+
+For the list of allowed operations on image data on the OSD side, see the
+[OSD data access rights](#osd-data-access-rights) section.
+
+### etcd privileges
+
+etcd privileges are implemented through Antietcd in all modes of operation.
+
+Built-in etcd privileges are not supported due to numerous inconveniences:
+- Certificate-based authentication does not work at all in etcd's REST interface,
+- Privileges are stored separately from k/v data and cannot participate in transactions,
+- Only the administrator (root) can change privileges,
+- There is no support for filtering range read responses by privileges.
+
+If etcd is used, Antietcd acts as a filtering proxy and can be embedded in the Vitastor
+monitor or run separately. In this case, etcd must allow incoming connections only from
+Antietcd, and all other components must connect to Antietcd.
+
+If Antietcd runs as a part of the Vitastor monitor, it is sufficient to enable the option
+`use_perms=true` and set the required certificates. If Antietcd is run separately, privileges
+have to be enabled separately using Antietcd options. For more details on the setup, see
+the [etcd/Antietcd setup options](#etcdantietcd-setup-options) section.
+
+For the list of allowed operations with etcd data, see the
+[etcd data access rights](#etcd-data-access-rights) section.
+
+## Manual setup
+
+### Configuring OSD transport encryption
+
+You need 2 certificates: one for OSDs and one for signing all client certificates.
+For OSDs, you can use a self-signed certificate (osd_ca.crt) or a separate certificate (osd.crt)
+signed by a trusted osd_ca.crt certificate. For clients, you must use separate certificates
+signed by a common trusted (client_ca.crt).
+
+Add to the Vitastor configuration on OSD servers:
+- use_perms: true
+- osd_ca: osd_ca.crt
+- client_ca: client_ca.crt
+- osd_cert: osd_ca.crt
+- osd_pkey: osd_ca.key
+
+On the client side:
+- use_perms: true
+- cert: client.crt
+- pkey: client.key
+
+### etcd/Antietcd setup options
+
+The following configuration options are available:
+
+#### Mon with embedded Antietcd
+
+The simplest option. You need 1 certificate for Antietcd (antietcd.crt), plus root
+certificates for OSDs and clients.
+
+Vitastor settings (`/etc/vitastor/vitastor.conf`):
+- etcd_address: [ "http://mon1:2379", ... ] (addresses of your monitors with port 2379)
+- use_perms: true
+- use_antietcd: true
+- antietcd_cert: antietcd.crt
+- antietcd_key: antietcd.key
+- etcd_ca: antietcd.crt
+- osd_ca: osd_ca.crt
+- client_ca: client_ca.crt
+
+#### Mon as an Etcd proxy
+
+If you want to enable privileges, but stay on etcd, you can use etcd proxy mode.
+
+You will need 2 separate certificates: one for etcd (etcd.crt) and one for antietcd (antietcd.crt).
+The etcd client port must be different from the standard 2379 — for example, you can pick 2381.
+OSD and client certificates are also needed.
+
+Vitastor settings:
+- etcd_address: [ "http://mon1:2379", ... ] (addresses of your monitors with port 2379)
+- use_perms: true
+- use_antietcd: true
+- etcd_proxy:
+  ```
+  {
+    "urls": [ "http://mon1:2381", ... ], // addresses of your etcd with port 2381
+    "cert": "antietcd.crt",
+    "key": "antietcd.key",
+    "ca": "etcd.crt"
+  }
+  ```
+- antietcd_cert: antietcd.crt
+- antietcd_key: antietcd.key
+- etcd_ca: antietcd.crt
+- osd_ca: osd_ca.crt
+- client_ca: client_ca.crt
+
+etcd command-line options:
+```
+--advertise-client-urls=https://<ADDRESS>:2381 --listen-client-urls=https://<ADDRESS>:2381 \
+--client-cert-auth --cert-file=etcd.crt --key-file=etcd.key --trusted-ca-file=antietcd.crt \
+--peer-client-cert-auth --peer-cert-file=etcd.crt --peer-key-file=etcd.key --peer-trusted-ca-file=etcd.crt
+```
+
+#### Mon with a separate Antietcd Proxy
+
+If in addition to the previous option you want to offload Antietcd from the Vitastor monitor's
+tasks, you can run it separately.
+
+Similar to the previous option, 2 certificates are needed: one for etcd and one for antietcd,
+plus separate certificates for clients, OSDs, and monitors will be needed.
+
+Vitastor settings:
+- etcd_address: [ "http://mon1:2379", ... ] (addresses of your monitors with port 2379)
+- use_perms: true
+- use_antietcd: false
+- etcd_ca: antietcd.crt
+- osd_ca: osd_ca.crt
+- client_ca: client_ca.crt
+- mon_etcd_client_cert: mon_ca.crt
+- mon_etcd_client_key: mon_ca.key
+
+Antietcd command-line options:
+```
+--port 2379 \
+--client_cert_auth 1 --auth_filter vitastor_auth_filter.js --etcd_proxy url1,url2,... \
+--cert antietcd.crt --key antietcd.key --ca client_ca.crt --osd_ca osd_ca.crt --mon_ca mon_ca.crt \
+--etcd_cert antietcd.crt --etcd_key antietcd.key --etcd_ca etcd.crt
+```
+
+etcd command-line options (same as in the previous option):
+```
+--advertise-client-urls=https://<ADDRESS>:2381 --listen-client-urls=https://<ADDRESS>:2381 \
+--client-cert-auth --cert-file=etcd.crt --key-file=etcd.key --trusted-ca-file=antietcd.crt \
+--peer-client-cert-auth --peer-cert-file=etcd.crt --peer-key-file=etcd.key --peer-trusted-ca-file=etcd.crt
+```
+
+#### Standalone Antietcd without etcd
+
+Same as the previous option, but etcd and its certificate are not needed:
+
+Vitastor settings (same as in the previous option):
+- etcd_address: [ "http://mon1:2379", ... ] (addresses of your monitors with port 2379)
+- use_perms: true
+- use_antietcd: false
+- etcd_ca: antietcd.crt
+- osd_ca: osd_ca.crt
+- client_ca: client_ca.crt
+- mon_etcd_client_cert: mon_ca.crt
+- mon_etcd_client_key: mon_ca.key
+
+Antietcd command-line options:
+```
+--port 2379 \
+--client_cert_auth 1 --auth_filter vitastor_auth_filter.js \
+--persist_filter vitastor_persist_filter.js \
+--cert antietcd.crt --key antietcd.key --ca client_ca.crt --osd_ca osd_ca.crt --mon_ca mon_ca.crt
+```
+
+### Vault/OpenBao setup
+
+To use Vault, each client that needs to get image keys from Vault needs a Vault account.
+Vitastor only supports client certificate-based authentication, so all client certificates
+(`cert`+`pkey`) must be registered in Vault, and they must be granted access to the
+corresponding secrets (v1 secrets API is supported).
+
+The required format of a Vault secret is a single `key` field as a hexadecimal string.
+The AES-256-XTS algorithm is used, so the key length is 64 bytes, i.e., the string must
+consist of 128 hexadecimal digits.
+
+To connect to Vault, set the following settings in Vitastor.conf:
+- `vault_url` — Vault address (e.g., `https://vault:8200`)
+- `vault_ca` — Vault's own certificate
+
+After that, if you create an image (`vitastor-cli create`) with the option `--enc_key vault:<ID>`,
+Vitastor clients will first contact Vault to obtain a token at `/v1/auth/cert/login`,
+and then request the actual secret from Vault at `/v1/secret/<ID>`.
+
+#### Vault setup example
+
+Step-by-step instructions for setting up a test Vault using OpenBao as an example:
+
+1. If TLS is not yet configured, generate a self-signed TLS certificate for Vault:
+   ```
+   openssl req -days 3650 -x509 -addext basicConstraints=critical,CA:TRUE,pathlen:1 --addext subjectAltName=DNS:vault \
+       -new -newkey rsa:4096 -nodes -keyout /etc/openbao/vault.key -out /etc/openbao/vault.crt
+   ```
+   Configure it in `/etc/openbao/openbao.hcl`:
+   ```
+   listener "tcp" {
+       address       = "0.0.0.0:8200"
+       tls_cert_file = "/etc/openbao/vault.crt"
+       tls_key_file  = "/etc/openbao/vault.key"
+   }
+   ```
+   And restart OpenBao (`systemctl restart openbao`).
+2. Copy Vault's TLS certificate for Vitastor:
+   ```
+   cp /etc/openbao/vault.crt /etc/vitastor/vault.crt
+   ```
+   Transfer it to all client nodes and specify it in `/etc/vitastor/vitastor.conf`:
+   ```
+   {
+       ...
+       "vault_url": "http://vault:8200",
+       "vault_ca": "/etc/vitastor/vault.crt"
+   }
+   ```
+3. Check Vault status:
+   ```
+   bao status -ca-cert /etc/openbao/vault.crt -address=https://vault:8200
+   ```
+4. Initialize Vault in test mode from 1 node (with 1 key share):
+   ```
+   bao operator init -n 1 -t 1 -ca-cert /etc/openbao/vault.crt -address=https://vault:8200
+   ```
+5. Unseal Vault:
+   ```
+   bao operator unseal -ca-cert /etc/openbao/vault.crt -address=https://vault:8200
+   ```
+6. Enable certificate-based authentication:
+   ```
+   bao auth enable -ca-cert /etc/openbao/vault.crt -address=https://vault:8200 cert
+   ```
+7. Enable v1 secrets:
+   ```
+   bao secrets enable -ca-cert /etc/openbao/vault.crt -address=https://vault:8200 -path=secret kv-v1
+   ```
+8. Create a test secret:
+   ```
+   bao kv put -ca-cert /etc/openbao/vault.crt -address=https://vault:8200 secret/vitastor/testimg3 key=$(openssl rand -hex 64)
+   ```
+9. Generate a signed certificate for a Vitastor user (on a machine where you have `client_ca.crt` and `client_ca.key`):
+   ```
+   openssl req -subj '/CN=testimg3' -nodes -new -keyout testimg3.key -out testimg3.csr
+   openssl x509 -req -days 3650 -CA client_ca.crt -CAkey client_ca.key -CAcreateserial -in testimg3.csr -out testimg3.crt
+   rm testimg3.csr
+   ```
+10. Create a user in Vault and grant it access to the secret:
+    ```
+    cat >testimg3.policy <<EOF
+    path "/secret/vitastor/testimg3" {
+        capabilities = ["read"]
+    }
+    EOF
+
+    bao policy write -ca-cert /etc/openbao/vault.crt -address=https://vault:8200 testimg3 testimg3.policy
+
+    bao write -ca-cert /etc/openbao/vault.crt -address=https://vault:8200 auth/cert/certs/testimg3 \
+        certificate=@testimg3.crt display_name=testimg3 token_ttl=24h token_policies=testimg3
+    ```
+11. Test access to the secret:
+    ```
+    curl --cacert /etc/vitastor/vault.crt --cert testimg3.crt --key testimg3.key \
+        --json '{}' https://vault:8200/v1/auth/cert/login
+    ```
+    A token will be printed, substitute it into the following request:
+    ```
+    curl --cacert /etc/vitastor/vault.crt --cert testimg3.crt --key testimg3.key \
+        -H 'X-Vault-Token: <RECEIVED TOKEN>' https://vault:8200/v1/secret/vitastor/testimg3
+    ```
+12. Create an image in Vitastor with the given secret (as an administrator or someone who
+    has the right to create images in your pool):
+    ```
+    vitastor-cli create -s 100G --enc_key vault:vitastor/testimg3 --owner testimg3 testimg3
+    ```
+13. Test access to the image as user testimg3:
+    ```
+    vitastor-cli --cert testimg3.crt --pkey testimg3.key dd if=/dev/urandom oimg=testimg3 bs=1M count=100
+    ```
+
+## Lists of allowed operations
+
+### etcd data access rights
+
+Below, all key names are given without the common prefix `/vitastor`.
+
+Allowed operations with keys in Antietcd for clients (`type=client`):
+- Read-only:
+  - Always allowed:
+    - `/config/global`
+    - `/config/node_placement`
+    - `/config/pools`
+    - `/pg/config`
+    - `/osd/state/*`
+    - `/pg/state/*`
+    - `/index/maxid/*`
+  - For images [readable by the user](#users-and-access-rights):
+    - `/config/inode/*`
+    - `/index/image/*`
+    - `/inode/stats/*`
+- Read and write:
+  - For pools in which the user can create images:
+    - `/index/maxid/*`
+  - For images owned by the user:
+    - `/config/inode/*`
+    - `/index/image/*`
+
+Allowed operations with keys in Antietcd for administrators (`type=admin`):
+- Read:
+  - `/stats`
+  - `/mon/*`
+  - `/pg/*`
+  - `/pgstats/*`
+  - `/inode/stats/*`
+  - `/pool/stats/*`
+- Read and write:
+  - `/config/*`
+  - `/osd/*`
+  - `/index/*`
+  - `/pg/history/*`
+
+Allowed operations with keys in etcd for OSDs:
+- Read:
+  - `/pg/config`
+  - `/config/*`
+- Read and write:
+  - `/osd/*`
+  - `/pg/state/*`
+  - `/pg/history/*`
+  - `/pgstats/*`
+
+Allowed operations with keys in etcd for monitors:
+- Read:
+  - `/config/*`
+  - `/osd/*`
+  - `/pgstats/*`
+- Read and write:
+  - `/pg/config`
+  - `/stats`
+  - `/history/last_clean_pgs`
+  - `/mon/*`
+  - `/pg/history/*`
+  - `/inode/stats/*`
+  - `/pool/stats/*`
+
+### OSD data access rights
+
+When the `use_perms` option and encryption are enabled, OSDs authenticate clients via
+certificates and allow each client only what is allowed by the access control model.
+
+Client operations:
+- READ — allowed for images the user has read access to.
+- WRITE, DELETE, SCRUB — allowed for images the user has write access to.
+- SYNC — the operation is not tied to an image and is always allowed.
+- DESCRIBE — the operation is allowed only for administrators (used by the commands
+  `vitastor-cli describe` and `fix`).
+- PING — the operation is always allowed.
+- SHOW_CONFIG — the operation is always allowed, however, if the client presents
+  itself as an OSD in it, then it is verified that it uses a certificate signed by `osd_ca`.
+- SEC_LIST (listing) — allowed for other OSDs and administrators with any parameters,
+  and for regular clients only allowed for requests limited to an image the user has
+  read access to.
+
+Cluster operations — allowed only for other OSDs:
+- SEC_READ
+- SEC_WRITE
+- SEC_WRITE_STABLE
+- SEC_SYNC
+- SEC_STABILIZE
+- SEC_ROLLBACK
+- SEC_DELETE
+- SEC_READ_BMP
+- SEC_LOCK
+
+### API access rights
+
+[vitastor-cli serve](../usage/cli.en.md#serve) also supports client authentication
+via certificates. Only certificates signed by `client_ca` are accepted. A separate
+certificate `server_cert` with the key `server_pkey` is used as the server certificate.
+
+For `vitastor-cli serve` to work correctly, it itself must use a certificate
+(`cert`+`pkey`) of a user with administrator rights (`type=admin`) to access Vitastor.
+
+Regular clients, when accessing the API, are only allowed API operations on images
+available to them either for reading (for reads) or for writing (for modification).
+All other API calls are allowed only for administrators.
+
+List of allowed API operations:
+
+Clients (users with `type=client`) are allowed the following operations:
+- image/list — for images the user can read.
+- image/create — for pools in which the user is allowed to create images, or for
+  creating snapshots of images owned by the user.
+- image/delete, image/flatten, image/modify — for images owned by the user.
+
+All other operations are allowed only for administrators (`type=admin`).
+
+## Encryption performance
+
+You may wonder — how fast is all this wonderful encryption?
+
+The answer is — it depends heavily on the CPU. On modern processors (with AVX512 with VAES
+support) it is very fast — AES encryption speed can reach 10-20 GB/s and above. This
+primarily concerns the CPU of client machines, because end-to-end encryption is performed
+entirely on the client, and client uses its signle thread for transport encryption too,
+while there are many OSDs on the server side, and it is easier to add resources there.
+
+On older processors, the speed is noticeably worse — for example, on a Xeon E5 v4 it is
+only 3 GB/s.
+
+You can evaluate the performance of your processors using the `vitastor-cli cpubench` command.
+
+Example output (💪 AMD EPYC 9575F):
+
+```
+$ vitastor-cli cpubench
+Vitastor transport encryption benchmark (AES-256-GCM, AES-256-XTS and xxhash3)
+
+Warmup...
+
+No transport encryption, data checksums enabled, e2e unencrypted image
+xxhash3 1 M block... 209000 iterations in 2001 ms = 104447.78 MB/s
+xxhash3 4 K block... 37000000 iterations in 2022 ms = 71479.35 MB/s
+
+Header encryption with payload checksums, e2e unencrypted image
+AES-256-GCM encrypt header + xxhash3 1 M block... 210000 iterations in 2015 ms = 104218.36 MB/s
+AES-256-GCM encrypt header + xxhash3 4 K block... 26000000 iterations in 2073 ms = 48993.01 MB/s
+
+Full transport encryption, e2e unencrypted image
+AES-256-GCM encrypt header and 1 M block... 54000 iterations in 2000 ms = 27000.00 MB/s
+AES-256-GCM encrypt header and 4 K block... 11700000 iterations in 2014 ms = 22692.71 MB/s
+
+No transport encryption, no checksums, e2e encrypted image
+AES-256-XTS encrypt 1 M block... 50000 iterations in 2039 ms = 24521.82 MB/s
+AES-256-XTS encrypt 4 K block... 12600000 iterations in 2009 ms = 24499.13 MB/s
+
+No transport encryption, e2e encrypted image, data checksums enabled
+AES-256-XTS encrypt + xxhash3 1 M block... 40000 iterations in 2013 ms = 19870.84 MB/s
+AES-256-XTS encrypt + xxhash3 4 K block... 10200000 iterations in 2011 ms = 19812.90 MB/s
+
+Header encryption with payload checksums, e2e encrypted image
+AES-256-GCM encrypt header + AES-256-XTS encrypt + xxhash3 1 M block... 40000 iterations in 2014 ms = 19860.97 MB/s
+AES-256-GCM encrypt header + AES-256-XTS encrypt + xxhash3 4 K block... 8700000 iterations in 2011 ms = 16899.24 MB/s
+
+Full transport encryption, e2e encrypted image
+AES-256-XTS + AES-256-GCM encrypt 1 M block... 26000 iterations in 2062 ms = 12609.12 MB/s
+AES-256-XTS + AES-256-GCM encrypt 4 K block... 6300000 iterations in 2006 ms = 12267.88 MB/s
+```
+
+And here is Xeon E5-2680v4:
+
+```
+$ vitastor-cli cpubench
+Vitastor transport encryption benchmark (AES-256-GCM, AES-256-XTS and xxhash3)
+
+Warmup...
+
+No transport encryption, data checksums enabled, e2e unencrypted image
+xxhash3 1 M block... 62000 iterations in 2021 ms = 30677.88 MB/s
+xxhash3 4 K block... 12400000 iterations in 2006 ms = 24146.31 MB/s
+
+Header encryption with payload checksums, e2e unencrypted image
+AES-256-GCM encrypt header + xxhash3 1 M block... 62000 iterations in 2027 ms = 30587.07 MB/s
+AES-256-GCM encrypt header + xxhash3 4 K block... 6800000 iterations in 2011 ms = 13208.60 MB/s
+
+Full transport encryption, e2e unencrypted image
+AES-256-GCM encrypt header and 1 M block... 7000 iterations in 2317 ms = 3021.15 MB/s
+AES-256-GCM encrypt header and 4 K block... 1500000 iterations in 2102 ms = 2787.52 MB/s
+
+No transport encryption, no checksums, e2e encrypted image
+AES-256-XTS encrypt 1 M block... 7000 iterations in 2317 ms = 3021.15 MB/s
+AES-256-XTS encrypt 4 K block... 1600000 iterations in 2088 ms = 2993.30 MB/s
+
+No transport encryption, e2e encrypted image, data checksums enabled
+AES-256-XTS encrypt + xxhash3 1 M block... 6000 iterations in 2188 ms = 2742.23 MB/s
+AES-256-XTS encrypt + xxhash3 4 K block... 1400000 iterations in 2053 ms = 2663.78 MB/s
+
+Header encryption with payload checksums, e2e encrypted image
+AES-256-GCM encrypt header + AES-256-XTS encrypt + xxhash3 1 M block... 6000 iterations in 2190 ms = 2739.73 MB/s
+AES-256-GCM encrypt header + AES-256-XTS encrypt + xxhash3 4 K block... 1300000 iterations in 2101 ms = 2417.00 MB/s
+
+Full transport encryption, e2e encrypted image
+AES-256-XTS + AES-256-GCM encrypt 1 M block... 4000 iterations in 2666 ms = 1500.38 MB/s
+AES-256-XTS + AES-256-GCM encrypt 4 K block... 800000 iterations in 2113 ms = 1478.94 MB/s
+```

docs/intro/security.ru.md

@@ -1,56 +1,105 @@
-[Документация](../../README-ru.md#документация) → Безопасность
+[Документация](../../README-ru.md#документация) → Введение → Безопасность в Vitastor
 
 -----
 
 [Read in English](security.en.md)
 
-# Оглавление
-
-⚠️ Предупреждение: детальное описание настроек безопасности достаточно длинное.
-
-Если не боитесь - читайте [Подробное описание](#подробное-описание).
-
-Если хотите просто быстро настроить Vitastor с шифрованием - читайте начало статьи.
+# Безопасность в Vitastor
 
+- [Обзор](#обзор)
 - [Быстрая настройка](#быстрая-настройка)
-- 
+- Принципы работы
+  - [Шифрование соединений с etcd (TLS)](#шифрование-соединений-с-etcd-tls)
+  - [Шифрование соединений с OSD (AES-GCM)](#шифрование-соединений-с-osd-aes-gcm)
+  - [Сквозное шифрование данных образов (AES-XTS)](#сквозное-шифрование-данных-образов-aes-xts)
+  - [Аутентификация по сертификатам](#аутентификация-по-сертификатам)
+  - [Пользователи и права доступа](#пользователи-и-права-доступа)
+  - [Привилегии etcd](#привилегии-etcd)
+- Ручная настройка
+  - [Настройка шифрования соединений OSD](#настройка-шифрования-соединений-osd)
+  - Варианты настройки etcd/Antietcd
+    - [Mon со встроенным Antietcd](#mon-со-встроенным-antietcd)
+    - [Mon в роли Etcd proxy](#mon-в-роли-etcd-proxy)
+    - [Mon с отдельным Antietcd Proxy](#mon-с-отдельным-antietcd-proxy)
+    - [Отдельный Antietcd без etcd](#отдельный-antietcd-без-etcd)
+  - [Настройка Vault/OpenBao](#настройка-vaultopenbao)
+    - [Пример настройки Vault](#пример-настройки-vault)
+- Списки разрешённых операций
+  - [Права доступа к данным etcd](#права-доступа-к-данным-etcd)
+  - [Права доступа к данным OSD](#права-доступа-к-данным-osd)
+  - [Права доступа к API](#права-доступа-к-api)
+- [Производительность шифрования](#производительность-шифрования)
 
-# Быстрая настройка
+## Обзор
 
+Начиная с версии 3.1.0, Vitastor предоставляет полную защиту данных: защиту слоя
+управления (etcd), защиту слоя данных (OSD) и сквозное шифрование данных.
 
+- Защита слоя управления:
+  - Шифрование соединений с etcd (TLS)
+  - Аутентификация по клиентским TLS (X.509) сертификатам
+  - Разграничение прав доступа клиентов к данным etcd
+- Защита слоя данных:
+  - Либо полное AES-GCM шифрование соединений с OSD (аналогично TLS, но быстрее)
+  - Либо шифрование AES-GCM только заголовков команд с контрольными суммами данных с секретной "солью"
+  - Аутентификация по клиентским TLS (X.509) сертификатам
+  - Разграничение прав доступа клиентов на стороне OSD
+- Сквозное шифрование:
+  - Данные шифруются AES-XTS на стороне клиента, кластер Vitastor не имеет доступа к открытым данным
+  - Ключи AES-XTS могут храниться в etcd или во внешнем Vault/OpenBao
 
-# Пользовательские сценарии
+Все функции опциональны и в простейшем варианте настройки выключены. По умолчанию включены
+только контрольные суммы данных на транспортном уровне ([proto_checksums](../config/security.ru.md#proto_checksums)=payload) для
+поддерживающих их клиентов (>= 3.1.0). Для более старых клиентов по умолчанию разрешены
+соединения без контрольных сумм данных ([force_proto_checksums](../config/security.ru.md#force_proto_checksums) пусто).
 
-Зачем всё это нужно вам?
+Для быстрой настройки перейдите к разделу [Быстрая настройка](#быстрая-настройка).
 
+Описания всех параметров, связанных с безопасностью, читайте [здесь](../config/security.ru.md).
 
+## Быстрая настройка
 
-# Подробное описание
+Для быстрой настройки используйте скрипт `/usr/lib/vitastor/mon/make-etcd`:
 
-Начиная с версии 3.1.0, в Vitastor есть следующие функции:
-1. Шифрование соединений с etcd (TLS)
-2. Шифрование соединений с OSD (AES-GCM) - по выбору либо только заголовков, либо и заголовков, и данных
-3. Сквозное шифрование данных образов (AES-XTS)
-4. Хранения ключей шифрования AES-XTS во внешнем Vault
-5. Контрольных сумм данных на транспортном уровне с секретной "солью"
-6. Аутентификация с помощью TLS (X.509) сертификатов и закрытых ключей
-7. Разграничение прав доступа клиентов к данным etcd
-8. Разграничение прав доступа клиентов к данным самих образов (на стороне OSD)
+1. Зайдите на узел, на котором будет располагаться первый монитор и etcd.
+2. Создайте там минимальный `/etc/vitastor/vitastor.conf` с параметрами etcd_address,
+   osd_network и, если хотите включить привилегии - use_perms (обратите внимание на `https://`
+   в адресах etcd):
+   ```
+   {
+       "etcd_address": ["https://10.0.0.10:2379","https://10.0.0.11:2379","https://10.0.0.12:2379"],
+       "osd_network": "10.0.0.0/24",
+       "use_perms": true
+   }
+   ```
+3. Запустите `/usr/lib/vitastor/mon/make-etcd` без параметров или с параметром `--antietcd-only`,
+   если хотите инициализировать кластер только с Antietcd без etcd.
+4. Скрипт сгенерирует все необходимые сертификаты и предложит скопировать их на остальные узлы
+   мониторов (соглашайтесь!).
+5. Зайдите на все остальные узлы мониторов и повторите там вызов `/usr/lib/vitastor/mon/make-etcd`.
+6. Если у вас будут узлы только с OSD без мониторов, выполните следующую команду, чтобы скопировать
+   только нужную конфигурацию на эти узлы:
+   ```
+   /usr/lib/vitastor/mon/make-etcd --copy-to-osd osdnode1,osdnode2,...
+   ```
 
-По умолчанию шифрование, аутентификация и авторизация отключены, но, начиная с 3.1.0,
-используются контрольные суммы данных на транспортном уровне (`proto_checksums=payload`).
+После этого можете переходить к инициализации OSD.
 
-## Шифрование соединений с etcd (TLS)
+Если хотите разобраться в настройке подробнее, читайте далее разделы [Принципы работы](#принципы-работы)
+и [Ручная настройка](#ручная-настройка).
+
+## Принципы работы
+
+### Шифрование соединений с etcd (TLS)
 
 Варианты настройки:
 - Без шифрования (http)
 - С шифрованием (https)
-- С клиентским сертификатом, но при выключенной авторизации (`use_auth=false`) - используется
-  отдельный сертификат и ключ: `etcd_client_cert`, `etcd_client_key`
-- С клиентским сертификатом, при включённой аутентификации на уровне OSD - используется общий
-  сертификат и ключ: для OSD - `osd_cert` и `osd_pkey`, для клиентов - `cert` и `pkey`
+- С шифрованием и аутентификацией по клиентским сертификатам. Используется либо тот
+  же сертификат, что используется для аутентификации на стороне OSD (`cert`+`pkey` / `osd_cert`+`osd_pkey`),
+  либо отдельно указанный сертификат (`etcd_client_cert`+`etcd_client_key`)
 
-## Шифрование соединений с OSD (AES-GCM)
+### Шифрование соединений с OSD (AES-GCM)
 
 Варианты настройки:
 - Без шифрования и без контрольных сумм: `proto_checksums=none`.
@@ -65,46 +114,31 @@
   отключение контрольных сумм на уровне клиента, то есть используется `force_proto_checksums=payload`.
 - С полным шифрованием всего трафика: аналогично прошлому варианту, но с `proto_checksums=gcm`.
   Клиенту при этом по умолчанию разрешается понизить уровень защиты до контрольных сумм, но
-  это тоже можно запретить через `force_proto_checksums=gcm`. Данный вариант не является рекомендуемым,
-  так как добавлен в первую очередь для возможной поддержки небезопасных (публичных) сетей и
-  больше всего снижает производительность. В частности, если одновременно использовать полное
-  шифрование трафика и сквозное шифрование образов AES-XTS, то данные будут шифроваться дважды.
+  это тоже можно запретить через `force_proto_checksums=gcm`. Данный вариант самый медленный и
+  рекомендуется только для небезопасных (публичных) сетей. В том числе потому, что при использовании
+  и полного шифрования трафика, и сквозного шифрования образов AES-XTS, данные шифруются дважды.
 
 Для шифрования используется алгоритм AES-256-GCM и собственный упрощённый протокол согласования
 ключей, полностью аналогичный TLS 1.3 ECDHE.
 
-## Сквозное шифрование данных образов (AES-XTS)
+### Сквозное шифрование данных образов (AES-XTS)
 
 Клиент Vitastor поддерживает шифрование данных каждого образа своим ключом. В этом случае на OSD
-уходят уже зашифрованные данные и сами OSD не видят настоящее содержимое образов. Разные ключи
-в том числе могут иметь разные снимки или клоны одного и того же образа. Например, можно сделать
-базовый образ ВМ (условный Debian Linux) нешифрованным, но наследовать от него шифрованные образы
-клиентских ВМ.
+уходят уже зашифрованные данные и сами OSD не видят исходные данные клиента. При этом ключ можно
+менять при клонировании/создании снимков образов. Например, можно сделать базовый образ ВМ
+(условный Debian Linux) нешифрованным, но наследовать от него шифрованные образы клиентских ВМ.
 
 Ключи шифрования образов могут храниться либо в etcd, либо во внешнем Vault. Во втором случае
 в etcd хранятся только ID ключей, а Vitastor вообще не имеет доступа к данным образов. Для
-использования Vault нужно создать образ с опцией `--enc_key vault:ID`, а в конфигурации указать
-опции:
-- vault_url
-- vault_ca
-- vault_client_cert
-- vault_client_key
+использования Vault нужно создать образ с опцией `--enc_key vault:ID`, в конфигурации указать
+опции vault_url, и vault_ca, создать всем клиентам учётные записи в Vault и дать им доступ
+к требуемым секретам v1.
 
 Ещё раз повторимся, что если AES-XTS используется с полным шифрованием трафика (`proto_checksums=gcm`),
 то данные образов шифруются дважды - сначала AES-XTS, а потом AES-GCM. Можете использовать,
 только если вы совсем параноик :-).
 
-## Производительность шифрования
-
-У вас может возникнуть вопрос - а как быстро всё это прекрасное шифрование работает?
-
-Ответ - скорость сильно зависит от процессора. Складывается она из нескольких вещей:
-
-- 
-
-TODO: vitastor-cli bench.
-
-## Аутентификация по сертификатам
+### Аутентификация по сертификатам
 
 При включённом шифровании клиенты, OSD и мониторы Vitastor аутентифицируются по сертификатам
 как при соединениях с etcd (Antietcd), так и с OSD.
@@ -118,14 +152,24 @@ TODO: vitastor-cli bench.
 Также сертификат монитора может быть вообще не нужен, если Antietcd встраивается в сам монитор.
 В этом случае монитор и так имеет доступ ко всем данным etcd прямо в памяти.
 
-Каждый клиент должен иметь свой сертификат, подписанный общим корневым сертификатом
-для клиентов (`client_ca`). Common Name сертификата должно равняться имени пользователя.
+### Пользователи и права доступа
 
-## Модель прав доступа
+При отключённом шифровании трафика Vitastor работает без разграничения прав доступа, то есть,
+любой клиент кластера имеет полный доступ как к слою управлению, так и к слою данных. Такой
+вариант подходит для выделенных доверенных сетей хранения.
+
+При включённом шифровании трафика OSD (хотя бы заголовков) есть возможность задействовать
+права доступа, включив опцию `use_perms=true`. При включённой опции каждый пользователь может
+выполнять только те операции, которые ему разрешены, и даже OSD и мониторам также запрещены
+"лишние" операции.
+
+Каждый пользователь (или администратор) должен иметь свой сертификат, подписанный общим
+корневым сертификатом для клиентов (`client_ca`), с Common Name, равным имени пользователя.
+Настройки привилегий же хранятся в etcd. Для OSD и мониторов учётные записи не нужны,
+они аутентифицируются по отдельным сертификатам.
 
 Привилегии пользователей хранятся в данных etcd в ключах `/vitastor/config/user/<имя>`.
-
-У пользователя есть 2 свойства:
+В этом ключе для каждого пользователя задаётся:
 - Тип:
   - Клиент (`type=client` или не указано) - может читать и модифицировать только явным образом
     разрешённые образы.
@@ -133,67 +177,285 @@ TODO: vitastor-cli bench.
     кластер: смотреть общую статистику и состояние, создавать и удалять OSD и так далее.
 - Список имён групп, членом которых пользователь является.
 
-У образов есть 3 свойства:
+У образов есть следующие свойства:
 - Владелец (owner) - имя пользователя, которому разрешено и читать, и менять образ
 - Группа владельцев (owner_group) - имя группы владельцев
-- Группа читатетей (reader_group) - имя группы пользователей, которым разрешено читать образ
+- Группа читателей (reader_group) - имя группы пользователей, которым разрешено читать образ
 
-У пулов есть 1 свойство:
+И также есть свойство у пула:
 - Группа создателей (creator_group) - имя группы пользователей, которым разрешено создавать образы в пуле
 
-## Права доступа к данным etcd
+Перечень разрешённых операций с данными образов на стороне OSD смотрите в разделе
+[Права доступа к данным OSD](#права-доступа-к-данным-osd).
 
-Привилегии реализуются через Antietcd во всех режимах работы. Если используется etcd, то
-Antietcd выступает в роли фильтрующего прокси, при этом он может быть встроен в монитор
-Vitastor или запущен отдельно. В этом случае etcd должен разрешать входящие подключения
-только от Antietcd, а все остальные компоненты должны соединяться с Antietcd.
+### Привилегии etcd
 
-Если же используется Antietcd, то привилегии реализуются в нём самом.
+Привилегии etcd реализуются через Antietcd во всех режимах работы.
 
-Если используется встроенный в монитор Antietcd, то привилегии включаются либо параметром
-`use_auth: true`, либо, если этот параметр не указан - включается автоматически, если задан
-любой из параметров `client_ca`, `osd_ca`, `mon_ca`. При этом монитор требует указания
-параметров `client_ca` и `osd_ca`, а если не используется режим проксирования в etcd -
-также `antietcd_server_ca`, чтобы Antietcd мог отличать кластерные соединения от клиентских.
+Встроенные привилегии etcd не поддерживаются по причине их многочисленных неудобств:
+- Аутентификация по сертификатам вообще не работает в REST интерфейсе etcd,
+- Привилегии хранятся отдельно от k/v данных и не могут участвовать в транзакциях,
+- Менять привилегии может только администратор (root),
+- Нет поддержки фильтрации диапазонных ответов чтения по привилегиям.
 
-Если используется отдельно стоящий Antietcd, привилегии нужно включать явным образом.
+Если используется etcd, то Antietcd выступает в роли фильтрующего прокси, при этом он
+может быть встроен в монитор Vitastor или запущен отдельно. В этом случае etcd должен
+разрешать входящие подключения только от Antietcd, а все остальные компоненты должны
+соединяться с Antietcd.
 
-Встроенные привилегии etcd не поддерживаются по причине их многочисленных недоработок:
-- Аутентификация по сертификатам не работает в REST интерфейсе etcd,
-- Привилегии хранятся отдельно от k/v и не могут участвовать в транзакциях,
-- Менять привилегии может только администратор (root)
-- Нет поддержки фильтрации ответов чтения по привилегиям.
+Если Antietcd запускается в составе монитора Vitastor, то достаточно включить опцию
+`use_perms=true` и задать нужные сертификаты. Если Antietcd запускается отдельно, то
+привилегии нужно включать отдельно опциями Antietcd. Подробнее о настройке смотрите
+раздел [Варианты настройки etcd/Antietcd](#варианты-настройки-etcdantietcd).
 
-Подробный список привилегий на ключи в etcd [смотрите ниже](#привилегии-etcd).
+Перечень разрешённых операций с данными etcd смотрите в разделе
+[Права доступа к данным etcd](#права-доступа-к-данным-etcd).
 
-## Права доступа к данным OSD
+## Ручная настройка
 
-Регулируется опцией `use_auth`, либо, если она не указана, включается автоматически,
-если используется шифрование, то есть, если заданы опции `osd_ca` и `client_ca`.
+### Настройка шифрования соединений OSD
 
-OSD аутентифицирует клиентов по сертификатам и разрешает каждому клиенту только
-то, что ему разрешено согласно модели прав доступа.
+Вам нужно 2 сертификата: один для OSD и один для подписи сертификатов всех клиентов.
+Для OSD можно использовать самоподписанный сертификат (osd_ca.crt) или отдельный сертификат (osd.crt),
+подписанный доверенным сертификатом osd_ca.crt. Для клиентов нужно использовать отдельные
+сертификаты, подписанные общим доверенным (client_ca.crt).
 
-Подробный список разрешаемых OSD операций [смотрите ниже](#привилегии-osd).
+В конфигурацию Vitastor на серверах OSD нужно добавить:
+- use_perms: true
+- osd_ca: osd_ca.crt
+- client_ca: client_ca.crt
+- osd_cert: osd_ca.crt
+- osd_pkey: osd_ca.key
 
-## Права доступа к API
+На стороне клиентов:
+- use_perms: true
+- cert: client.crt
+- pkey: client.key
 
-[vitastor-cli serve](../usage/cli.ru.md#serve) также поддерживает клиентскую
-аутентификацию по сертификатам. Принимаются только сертификаты, подписанные
-`client_ca`. В качестве серверного сертификата используется отдельный сертификат
-`server_cert` с ключом `server_key`.
+### Варианты настройки etcd/Antietcd
 
-При этом для корректной работы `vitastor-cli serve` он сам должен использовать
-для доступа в Vitastor сертификат (`cert`+`pkey`) пользователя с правами
-администратора (`type=admin`).
+Доступны следующие варианты настройки:
 
-Обычным клиентам при доступе к API разрешаются только API-операции с образами,
-доступными им либо на чтение (для чтения), либо на запись (для модификации).
-Все остальные API-вызовы разрешаются только для администраторов.
+#### Mon со встроенным Antietcd
 
-Подробный список разрешаемых API операций [смотрите ниже](#привилегии-api).
+Самый простой вариант. Вам нужен 1 сертификат для Antietcd (antietcd.crt), плюс
+корневые сертификаты для OSD и клиентов.
 
-## Привилегии etcd
+Настройки Vitastor (`/etc/vitastor/vitastor.conf`):
+- etcd_address: [ "http://mon1:2379", ... ] (адреса ваших мониторов с портом 2379)
+- use_perms: true
+- use_antietcd: true
+- antietcd_cert: antietcd.crt
+- antietcd_key: antietcd.key
+- etcd_ca: antietcd.crt
+- osd_ca: osd_ca.crt
+- client_ca: client_ca.crt
+
+#### Mon в роли Etcd proxy
+
+Если вы хотите включить привилегии, но остаться на etcd, можно задействовать режим etcd proxy.
+
+Вам понадобится 2 отдельных сертификата: один для etcd (etcd.crt) и один для antietcd (antietcd.crt).
+Клиентский порт etcd должен отличаться от стандартного 2379, например, можно выбрать 2381.
+Также нужны сертификаты OSD и клиентов.
+
+Настройки Vitastor:
+- etcd_address: [ "http://mon1:2379", ... ] (адреса ваших мониторов с портом 2379)
+- use_perms: true
+- use_antietcd: true
+- etcd_proxy:
+  ```
+  {
+    "urls": [ "http://mon1:2381", ... ], // адреса ваших etcd с портом 2381
+    "cert": "antietcd.crt",
+    "key": "antietcd.key",
+    "ca": "etcd.crt"
+  }
+  ```
+- antietcd_cert: antietcd.crt
+- antietcd_key: antietcd.key
+- etcd_ca: antietcd.crt
+- osd_ca: osd_ca.crt
+- client_ca: client_ca.crt
+
+Опции командной строки etcd:
+```
+--advertise-client-urls=https://<АДРЕС>:2381 --listen-client-urls=https://<АДРЕС>:2381 \
+--client-cert-auth --cert-file=etcd.crt --key-file=etcd.key --trusted-ca-file=antietcd.crt \
+--peer-client-cert-auth --peer-cert-file=etcd.crt --peer-key-file=etcd.key --peer-trusted-ca-file=etcd.crt
+```
+
+#### Mon с отдельным Antietcd Proxy
+
+Если в дополнение к предыдущему варианту вы хотите разгрузить Antietcd от задач монитора Vitastor,
+можно запустить его отдельно.
+
+Аналогично предыдущему варианту нужно 2 сертификата: один для etcd и один для antietcd, плюс понадобятся
+отдельные сертификаты для клиентов, OSD и монитора.
+
+Настройки Vitastor:
+- etcd_address: [ "http://mon1:2379", ... ] (адреса ваших мониторов с портом 2379)
+- use_perms: true
+- use_antietcd: false
+- etcd_ca: antietcd.crt
+- osd_ca: osd_ca.crt
+- client_ca: client_ca.crt
+- mon_etcd_client_cert: mon_ca.crt
+- mon_etcd_client_key: mon_ca.key
+
+Опции командной строки Antietcd:
+```
+--port 2379 \
+--client_cert_auth 1 --auth_filter vitastor_auth_filter.js --etcd_proxy url1,url2,... \
+--cert antietcd.crt --key antietcd.key --ca client_ca.crt --osd_ca osd_ca.crt --mon_ca mon_ca.crt \
+--etcd_cert antietcd.crt --etcd_key antietcd.key --etcd_ca etcd.crt
+```
+
+Опции командной строки etcd (не отличаются от предыдущего варианта):
+```
+--advertise-client-urls=https://<АДРЕС>:2381 --listen-client-urls=https://<АДРЕС>:2381 \
+--client-cert-auth --cert-file=etcd.crt --key-file=etcd.key --trusted-ca-file=antietcd.crt \
+--peer-client-cert-auth --peer-cert-file=etcd.crt --peer-key-file=etcd.key --peer-trusted-ca-file=etcd.crt
+```
+
+#### Отдельный Antietcd без etcd
+
+Аналогично предыдущему варианту, но etcd и его сертификат не нужны:
+
+Настройки Vitastor (не отличаются от предыдущего варианта):
+- etcd_address: [ "http://mon1:2379", ... ] (адреса ваших мониторов с портом 2379)
+- use_perms: true
+- use_antietcd: false
+- etcd_ca: antietcd.crt
+- osd_ca: osd_ca.crt
+- client_ca: client_ca.crt
+- mon_etcd_client_cert: mon_ca.crt
+- mon_etcd_client_key: mon_ca.key
+
+Опции командной строки Antietcd:
+```
+--port 2379 \
+--client_cert_auth 1 --auth_filter vitastor_auth_filter.js \
+--persist_filter vitastor_persist_filter.js \
+--cert antietcd.crt --key antietcd.key --ca client_ca.crt --osd_ca osd_ca.crt --mon_ca mon_ca.crt
+```
+
+### Настройка Vault/OpenBao
+
+Для использования Vault каждому клиенту, который будет получать из Vault ключи
+образов, нужна учётная запись в Vault. Vitastor поддерживает только аутентификацию
+по клиентским сертификатам, так что все сертификаты клиентов (`cert`+`pkey`) должны
+быть зарегистрированы в Vault и им должен быть дан доступ к соответствующим секретам
+(поддерживается API секретов v1).
+
+Требуемый формат секрета Vault - одно поле `key` в формате шестнадцатеричной строки.
+Используется алгоритм AES-256-XTS, так что длина ключа - 64 байта, то есть строка
+должна состоять из 128 шестнадцатеричных цифр.
+
+Для подключения Vault включите следующие настройки в Vitastor.conf:
+- `vault_url` - адрес Vault (например, `https://vault:8200`)
+- `vault_ca` - сертификат самого Vault
+
+После этого, если создать образ (`vitastor-cli create`) с опцией `--enc_key vault:<ID>`,
+то для получения ключа клиенты Vitastor сначала обратятся к Vault для получения токена
+по адресу `/v1/auth/cert/login`, а потом запросят из Vault сам секрет по адресу `/v1/secret/<ID>`.
+
+#### Пример настройки Vault
+
+Пошаговая инструкция для настройки тестового Vault на примере OpenBao:
+
+1. Если ещё не настроен TLS, генерируем самоподписанный TLS сертификат для Vault:
+   ```
+   openssl req -days 3650 -x509 -addext basicConstraints=critical,CA:TRUE,pathlen:1 --addext subjectAltName=DNS:vault \
+       -new -newkey rsa:4096 -nodes -keyout /etc/openbao/vault.key -out /etc/openbao/vault.crt
+   ```
+   Настраиваем его в `/etc/openbao/openbao.hcl`:
+   ```
+   listener "tcp" {
+       address       = "0.0.0.0:8200"
+       tls_cert_file = "/etc/openbao/vault.crt"
+       tls_key_file  = "/etc/openbao/vault.key"
+   }
+   ```
+   И перезапускаем OpenBao (`systemctl restart openbao`).
+2. Копируем TLS сертификат Vault для Vitastor:
+   ```
+   cp /etc/openbao/vault.crt /etc/vitastor/vault.crt
+   ```
+   Переносим его на все клиентские ноды и прописываем в `/etc/vitastor/vitastor.conf`:
+   ```
+   {
+       ...
+       "vault_url": "http://vault:8200",
+       "vault_ca": "/etc/vitastor/vault.crt"
+   }
+   ```
+3. Проверяем статус Vault:
+   ```
+   bao status -ca-cert /etc/openbao/vault.crt -address=https://vault:8200
+   ```
+4. Инициализируем Vault в тестовом режиме из 1 ноды (с 1 частью ключа):
+   ```
+   bao operator init -n 1 -t 1 -ca-cert /etc/openbao/vault.crt -address=https://vault:8200
+   ```
+5. Разблокируем Vault:
+   ```
+   bao operator unseal -ca-cert /etc/openbao/vault.crt -address=https://vault:8200
+   ```
+6. Включаем аутентификацию по сертификатам:
+   ```
+   bao auth enable -ca-cert /etc/openbao/vault.crt -address=https://vault:8200 cert
+   ```
+7. Включаем секреты v1:
+   ```
+   bao secrets enable -ca-cert /etc/openbao/vault.crt -address=https://vault:8200 -path=secret kv-v1
+   ```
+8. Создаём тестовый секрет:
+   ```
+   bao kv put -ca-cert /etc/openbao/vault.crt -address=https://vault:8200 secret/vitastor/testimg3 key=$(openssl rand -hex 64)
+   ```
+9. Генерируем подписанный сертификат для пользователя Vitastor (там, где у вас есть `client_ca.crt` и `client_ca.key`):
+   ```
+   openssl req -subj '/CN=testimg3' -nodes -new -keyout testimg3.key -out testimg3.csr
+   openssl x509 -req -days 3650 -CA client_ca.crt -CAkey client_ca.key -CAcreateserial -in testimg3.csr -out testimg3.crt
+   rm testimg3.csr
+   ```
+10. Создаём пользователя в Vault и даём ему доступ к секрету:
+    ```
+    cat >testimg3.policy <<EOF
+    path "/secret/vitastor/testimg3" {
+        capabilities = ["read"]
+    }
+    EOF
+
+    bao policy write -ca-cert /etc/openbao/vault.crt -address=https://vault:8200 testimg3 testimg3.policy
+
+    bao write -ca-cert /etc/openbao/vault.crt -address=https://vault:8200 auth/cert/certs/testimg3 \
+        certificate=@testimg3.crt display_name=testimg3 token_ttl=24h token_policies=testimg3
+    ```
+11. Тестируем доступ к секрету:
+    ```
+    curl --cacert /etc/vitastor/vault.crt --cert testimg3.crt --key testimg3.key \
+        --json '{}' https://vault:8200/v1/auth/cert/login
+    ```
+    Будет выведен токен, подставляем его в следующий запрос:
+    ```
+    curl --cacert /etc/vitastor/vault.crt --cert testimg3.crt --key testimg3.key \
+        -H 'X-Vault-Token: <ПОЛУЧЕННЫЙ ТОКЕН>' https://vault:8200/v1/secret/vitastor/testimg3
+    ```
+12. Создаём образ в Vitastor с заданным секретом (от имени администратора или того, кто имеет
+    право создавать образы в вашем пуле):
+    ```
+    vitastor-cli create -s 100G --enc_key vault:vitastor/testimg3 --owner testimg3 testimg3
+    ```
+13. Тестируем доступ к образу от имени пользователя testimg3:
+    ```
+    vitastor-cli --cert testimg3.crt --pkey testimg3.key dd if=/dev/urandom oimg=testimg3 bs=1M count=100
+    ```
+
+## Списки разрешённых операций
+
+### Права доступа к данным etcd
 
 Ниже все названия ключей приведены без общего префикса `/vitastor`.
 
@@ -207,7 +469,7 @@ OSD аутентифицирует клиентов по сертификата
     - `/osd/state/*`
     - `/pg/state/*`
     - `/index/maxid/*`
-  - Для образов, которые [может читать пользователь](#модель-прав-доступа):
+  - Для образов, которые [может читать пользователь](#пользователи-и-права-доступа):
     - `/config/inode/*`
     - `/index/image/*`
     - `/inode/stats/*`
@@ -256,7 +518,10 @@ OSD аутентифицирует клиентов по сертификата
   - `/inode/stats/*`
   - `/pool/stats/*`
 
-## Привилегии OSD
+### Права доступа к данным OSD
+
+При включённой опции `use_perms` и шифровании OSD аутентифицирует клиентов по сертификатам
+и разрешает каждому клиенту только то, что ему разрешено согласно модели прав доступа.
 
 Клиентские операции:
 - READ - разрешено для образов, доступных пользователю на чтение.
@@ -282,7 +547,22 @@ OSD аутентифицирует клиентов по сертификата
 - SEC_READ_BMP
 - SEC_LOCK
 
-## Привилегии API
+### Права доступа к API
+
+[vitastor-cli serve](../usage/cli.ru.md#serve) также поддерживает клиентскую
+аутентификацию по сертификатам. Принимаются только сертификаты, подписанные
+`client_ca`. В качестве серверного сертификата используется отдельный сертификат
+`server_cert` с ключом `server_pkey`.
+
+При этом для корректной работы `vitastor-cli serve` он сам должен использовать
+для доступа в Vitastor сертификат (`cert`+`pkey`) пользователя с правами
+администратора (`type=admin`).
+
+Обычным клиентам при доступе к API разрешаются только API-операции с образами,
+доступными им либо на чтение (для чтения), либо на запись (для модификации).
+Все остальные API-вызовы разрешаются только для администраторов.
+
+Список разрешённых операций API:
 
 Клиентам (пользователям с `type=client`) разрешаются операции:
 - image/list - для образов, которые пользователь может читать.
@@ -292,148 +572,91 @@ OSD аутентифицирует клиентов по сертификата
 
 Все остальные операции разрешаются только администраторам (`type=admin`).
 
+## Производительность шифрования
 
+У вас может возникнуть вопрос - а как быстро всё это прекрасное шифрование работает?
 
+Ответ - сильно зависит от процессора. На современных процессорах (при наличии AVX512 с VAES)
+очень быстро - скорость шифрования AES может составлять 10-20 Гбайт/с и выше. В первую очередь
+подразумевается CPU клиентских машин, потому что сквозное шифрование выполняется целиком на
+клиенте, а транспортное хоть также и затрагивает OSD, но у клиента поток один, а OSD на стороне
+сервера много и добавить там ресурсов легче.
 
+На более старых процессорах скорость заметно хуже, например, на Xeon E5 v4 она составляет
+буквально 3 Гбайт/с.
 
+Вы можете оценить производительность своих процессоров с помощью команды `vitastor-cli cpubench`.
 
+Пример вывода (💪 AMD EPYC 9575F):
 
+```
+$ vitastor-cli cpubench
+Vitastor transport encryption benchmark (AES-256-GCM, AES-256-XTS and xxhash3)
 
+Warmup...
 
+No transport encryption, data checksums enabled, e2e unencrypted image
+xxhash3 1 M block... 209000 iterations in 2001 ms = 104447.78 MB/s
+xxhash3 4 K block... 37000000 iterations in 2022 ms = 71479.35 MB/s
 
+Header encryption with payload checksums, e2e unencrypted image
+AES-256-GCM encrypt header + xxhash3 1 M block... 210000 iterations in 2015 ms = 104218.36 MB/s
+AES-256-GCM encrypt header + xxhash3 4 K block... 26000000 iterations in 2073 ms = 48993.01 MB/s
 
+Full transport encryption, e2e unencrypted image
+AES-256-GCM encrypt header and 1 M block... 54000 iterations in 2000 ms = 27000.00 MB/s
+AES-256-GCM encrypt header and 4 K block... 11700000 iterations in 2014 ms = 22692.71 MB/s
 
+No transport encryption, no checksums, e2e encrypted image
+AES-256-XTS encrypt 1 M block... 50000 iterations in 2039 ms = 24521.82 MB/s
+AES-256-XTS encrypt 4 K block... 12600000 iterations in 2009 ms = 24499.13 MB/s
 
+No transport encryption, e2e encrypted image, data checksums enabled
+AES-256-XTS encrypt + xxhash3 1 M block... 40000 iterations in 2013 ms = 19870.84 MB/s
+AES-256-XTS encrypt + xxhash3 4 K block... 10200000 iterations in 2011 ms = 19812.90 MB/s
 
+Header encryption with payload checksums, e2e encrypted image
+AES-256-GCM encrypt header + AES-256-XTS encrypt + xxhash3 1 M block... 40000 iterations in 2014 ms = 19860.97 MB/s
+AES-256-GCM encrypt header + AES-256-XTS encrypt + xxhash3 4 K block... 8700000 iterations in 2011 ms = 16899.24 MB/s
 
+Full transport encryption, e2e encrypted image
+AES-256-XTS + AES-256-GCM encrypt 1 M block... 26000 iterations in 2062 ms = 12609.12 MB/s
+AES-256-XTS + AES-256-GCM encrypt 4 K block... 6300000 iterations in 2006 ms = 12267.88 MB/s
+```
 
+А вот Xeon E5-2680v4:
 
+```
+$ vitastor-cli cpubench
+Vitastor transport encryption benchmark (AES-256-GCM, AES-256-XTS and xxhash3)
 
-Таким образом, доступны следующие варианты настройки:
+Warmup...
 
-### Mon в роли Etcd proxy
+No transport encryption, data checksums enabled, e2e unencrypted image
+xxhash3 1 M block... 62000 iterations in 2021 ms = 30677.88 MB/s
+xxhash3 4 K block... 12400000 iterations in 2006 ms = 24146.31 MB/s
 
-Mon
-- use_antietcd: true
-- etcd_proxy = {
-    urls: [],
-    cert = <antietcd.pem>,
-    key,
-    ca = <etcd.pem>,
-}
-- antietcd_cert = antietcd.pem
-- antietcd_key
+Header encryption with payload checksums, e2e unencrypted image
+AES-256-GCM encrypt header + xxhash3 1 M block... 62000 iterations in 2027 ms = 30587.07 MB/s
+AES-256-GCM encrypt header + xxhash3 4 K block... 6800000 iterations in 2011 ms = 13208.60 MB/s
 
-etcd
---client-cert-auth --cert-file=etcd.pem --key-file=etcd.key --trusted-ca-file=antietcd.pem \
---peer-client-cert-auth --peer-cert-file=etcd.pem --peer-key-file=etcd.key --peer-trusted-ca-file=etcd.pem
+Full transport encryption, e2e unencrypted image
+AES-256-GCM encrypt header and 1 M block... 7000 iterations in 2317 ms = 3021.15 MB/s
+AES-256-GCM encrypt header and 4 K block... 1500000 iterations in 2102 ms = 2787.52 MB/s
 
-### Mon с отдельным Antietcd Proxy
+No transport encryption, no checksums, e2e encrypted image
+AES-256-XTS encrypt 1 M block... 7000 iterations in 2317 ms = 3021.15 MB/s
+AES-256-XTS encrypt 4 K block... 1600000 iterations in 2088 ms = 2993.30 MB/s
 
-Mon
-- use_antietcd: false
-- etcd_ca = antietcd.pem
+No transport encryption, e2e encrypted image, data checksums enabled
+AES-256-XTS encrypt + xxhash3 1 M block... 6000 iterations in 2188 ms = 2742.23 MB/s
+AES-256-XTS encrypt + xxhash3 4 K block... 1400000 iterations in 2053 ms = 2663.78 MB/s
 
-Antietcd
---client_cert_auth 1 --auth_filter vitastor_auth_filter.js --etcd_proxy url1,url2,... \
---cert antietcd.pem --key antietcd.key --ca client_ca.pem --osd_ca osd_ca.pem \
---etcd_cert antietcd.pem --etcd_key antietcd.key --etcd_ca etcd.pem
+Header encryption with payload checksums, e2e encrypted image
+AES-256-GCM encrypt header + AES-256-XTS encrypt + xxhash3 1 M block... 6000 iterations in 2190 ms = 2739.73 MB/s
+AES-256-GCM encrypt header + AES-256-XTS encrypt + xxhash3 4 K block... 1300000 iterations in 2101 ms = 2417.00 MB/s
 
-etcd
---client-cert-auth --cert-file=etcd.pem --key-file=etcd.key --trusted-ca-file=antietcd.pem \
---peer-client-cert-auth --peer-cert-file=etcd.pem --peer-key-file=etcd.key --peer-trusted-ca-file=etcd.pem
-
-### Mon со встроенным Antietcd
-
-Mon
-- use_antietcd: true
-- use_auth: true
-- antietcd_cert = antietcd.pem
-- antietcd_key
-
-### Отдельный Antietcd
-
-Mon
-- use_antietcd: false
-- etcd_ca = antietcd.pem
-
-Antietcd
---client_cert_auth 1 --auth_filter vitastor_auth_filter.js
-
-## Варианты настройки
-
-### Настройка по умолчанию
-
-Используются только контрольные суммы данных на транспортном уровне. Соединения с etcd не шифруются.
-Аутентификация и авторизация не используется, любой клиент имеет доступ ко всем данным кластера.
-
-Аналог настройки:
-- proto_checksums: payload
-
-### Полная защита
-
-Везде
-- osd_ca
-- client_ca
-- etcd_ca = antietcd.pem
-
-OSD
-- osd_cert
-- osd_pkey
-
-Клиент
-- cert
-- pkey
-
-### Только защита etcd
-
-- etcd_ca
-- etcd_cert
-- etcd_key
-
-### antietcd и только защита antietcd
-
-- etcd_ca
-- etcd_cert
-- etcd_key
-- use_antietcd: true
-- antietcd_cert = etcd_ca
-- antietcd_key
-- antietcd_ca = etcd_cert
-
-### Полное шифрование протокола, включая данные
-
-Внимание: если включить этот вариант защиты и при этом 
-
-### Только контрольные суммы на транспортном уровне, без шифрования
-
-## Настройка Vault/OpenBao
-
-openssl req -days 3650 -x509 -addext basicConstraints=critical,CA:TRUE,pathlen:1 --addext subjectAltName=DNS:vault \
-    -new -newkey rsa:4096 -nodes -keyout vault.key -out vault.crt
-
-bao status -ca-cert /etc/openbao/tls/vault.crt -address=https://vault:8200
-
-bao operator init -n 1 -t 1 -ca-cert /etc/openbao/tls/vault.crt -address=https://vault:8200
-
-bao operator unseal -ca-cert /etc/openbao/tls/vault.crt -address=https://vault:8200
-
-bao auth enable -ca-cert /etc/openbao/tls/vault.crt -address=https://vault:8200 cert
-
-bao secrets enable -ca-cert /etc/openbao/tls/vault.crt -address=https://vault:8200 -path=secret kv-v1
-
-bao kv put -ca-cert /etc/openbao/tls/vault.crt -address=https://vault:8200 secret/vitastor/testimg3 key=$(openssl rand -hex 64)
-
-cat >testimg3.policy <<EOF
-path "/secret/vitastor/testimg3" {
-  capabilities = ["read"]
-}
-EOF
-
-bao policy write -ca-cert /etc/openbao/tls/vault.crt -address=https://vault:8200 testimg3 testimg3.policy
-
-bao write -ca-cert /etc/openbao/tls/vault.crt -address=https://vault:8200 auth/cert/certs/testimg3 certificate=@testimg3.crt display_name=testimg3 token_ttl=24h token_policies=testimg3
-
-curl --cacert /etc/vitastor/vault.crt --cert testimg3.crt --key testimg3.key --json '{}' https://vault:8200/v1/auth/cert/login
-
-curl --cacert /etc/vitastor/vault.crt --cert testimg3.crt --key testimg3.key -H 'X-Vault-Token: s.Qkrm78BeK7Rqdz5MA3eJZNbu' https://vault:8200/v1/secret/vitastor/testimg3
+Full transport encryption, e2e encrypted image
+AES-256-XTS + AES-256-GCM encrypt 1 M block... 4000 iterations in 2666 ms = 1500.38 MB/s
+AES-256-XTS + AES-256-GCM encrypt 4 K block... 800000 iterations in 2113 ms = 1478.94 MB/s
+```

mon/antietcd_adapter.js

@@ -28,32 +28,38 @@ class AntiEtcdAdapter
             is_local['::'] = true;
             is_local[''] = true;
             // split :, 3 -> <schema>:<//ip>:<port>
-            const cluster_local = cluster.map(s =>
+            const selected = [];
+            for (let i = 0; i < cluster.length; i++)
             {
-                const m = /^https?:\/\/(?:\[(.*)\]|([^\[\:]+))(?::(\d+))?$/.exec(s);
-                return [ m[2] || m[1], m[3] || 2379 ];
-            });
-            const selected = cluster_local.filter(ip => is_local[ip[0]] && (!cfg_port || ip[1] == cfg_port));
+                const m = /^(https?:\/\/)?(?:\[(.*)\]|([^\[\:]+))(?::(\d+))?$/.exec(cluster[i]);
+                if (!m)
+                    continue;
+                const ip = m[3] || m[2];
+                const port = m[4] || 2379;
+                if (is_local[ip] && (!cfg_port || port == cfg_port))
+                    selected.push({ idx: i, ip, port });
+            }
             if (selected.length > 1)
             {
-                console.error('More than 1 etcd_address matches local IPs: '+(selected.join(', '))+', please specify port');
+                console.error('More than 1 etcd_address matches local IPs, please specify port');
                 process.exit(1);
             }
             else if (selected.length == 1)
             {
                 const antietcd_config = {
-                    ip: selected[0][0],
-                    port: selected[0][1],
+                    ip: selected[0].ip,
+                    port: selected[0].port,
                     cert: config.antietcd_cert,
                     key: config.antietcd_key,
-                    ca: config.antietcd_ca,
-                    data: config.antietcd_data_file || ((config.antietcd_data_dir || '/var/lib/vitastor') + '/mon_'+selected[0][1]+'.json.gz'),
+                    ca: config.client_ca,
+                    data: config.antietcd_data_file || ((config.antietcd_data_dir || '/var/lib/vitastor') + '/mon_'+selected[0].port+'.json.gz'),
                     persist_filter: vitastor_persist_filter({ vitastor_prefix: config.etcd_prefix || '/vitastor' }),
-                    node_id: selected[0][0]+':'+selected[0][1], // node_id = ip:port
+                    node_id: cluster[selected[0].idx].replace(/^(https?:\/\/)/, ''), // same as in <cluster> below
                     cluster: (cluster.length == 1 ? null : cluster.reduce((a, c) => { a[c.replace(/^(https?:\/\/)/, '')] = c; return a; }, {})),
                     cluster_key: (config.etcd_prefix || '/vitastor'),
                     stale_read: 1,
                     log_level: 1,
+                    logs: { cluster: true },
                 };
                 if (config.etcd_proxy)
                 {
@@ -72,23 +78,13 @@ class AntiEtcdAdapter
                     delete antietcd_config.cluster;
                     delete antietcd_config.cluster_key;
                 }
-                const use_auth = config.use_auth || config.use_auth == null && config.client_ca;
-                if (use_auth)
+                if (config.use_perms)
                 {
                     antietcd_config.client_cert_auth = true;
                     antietcd_config.auth_filter = vitastor_auth_filter;
                     antietcd_config.ca = config.client_ca;
                     antietcd_config.osd_ca = config.osd_ca;
                     antietcd_config.mon_ca = config.mon_ca;
-                    if (!config.etcd_proxy)
-                    {
-                        antietcd_config.peer_ca = config.antietcd_server_ca;
-                        if (!config.antietcd_server_ca || config.antietcd_server_ca == config.client_ca)
-                        {
-                            console.error('Secure setup requires separate antietcd_server_ca (for signing antietcd server certificates) and client_ca (for signing client certificates)');
-                            process.exit(1);
-                        }
-                    }
                 }
                 for (const key in config)
                 {

mon/lp_optimizer/lp_optimizer.js

@@ -112,9 +112,10 @@ function make_cyclic(pgs, parity_space)
 {
     if (parity_space > 1)
     {
-        for (const pg in pgs)
+        for (const id in pgs)
         {
-            for (let i = 1; i < pg.size; i++)
+            const pg = pgs[id];
+            for (let i = 1; i < pg.length; i++)
             {
                 const cyclic = [ ...pg.slice(i), ...pg.slice(0, i) ];
                 pgs['pg_'+cyclic.join('_')] = cyclic;

mon/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vitastor-mon",
-  "version": "3.0.12",
+  "version": "3.0.15",
   "description": "Vitastor SDS monitor service",
   "main": "mon-main.js",
   "scripts": {
@@ -9,7 +9,7 @@
   "author": "Vitaliy Filippov",
   "license": "UNLICENSED",
   "dependencies": {
-    "antietcd": "^1.2.4",
+    "antietcd": "^1.3.1",
     "sprintf-js": "^1.1.2",
     "ws": "^7.2.5"
   },

mon/scripts/make-etcd

@@ -6,38 +6,39 @@
 const child_process = require('child_process');
 const fs = require('fs');
 const os = require('os');
+const path = require('path');
 const readline = require('readline');
 
 run().catch(e => { console.error(e); process.exit(1); });
 
 const help_text = `Initialize a Vitastor cluster (etcd, vitastor.conf and TLS certificates)
-(c) Vitaliy Filippov, 2019+ (MIT)
+(c) Vitaliy Filippov, 2026+ (MIT)
 
 USAGE:
-1) Create a minimal vitastor.conf with etcd_address, osd_network and (optionally) use_auth.
-   Example: {"etcd_address":["http://10.0.0.10:2379","http://10.0.0.11:2379","http://10.0.0.12:2379"],"use_auth":false,"osd_network":"10.0.0.0/24"}
-   Or: {"etcd_address":["https://10.0.0.10:2379","https://10.0.0.11:2379","https://10.0.0.12:2379"],"use_auth":true,"osd_network":"10.0.0.0/24"}
-2) Run: ${process.argv[1]} [./vitastor.conf]
+1) Create a minimal vitastor.conf with etcd_address, osd_network and (optionally) use_perms.
+   Non-encrypted: {"etcd_address":["http://10.0.0.10:2379","http://10.0.0.11:2379","http://10.0.0.12:2379"],"osd_network":"10.0.0.0/24"}
+   Encrypted: {"etcd_address":["https://10.0.0.10:2379","https://10.0.0.11:2379","https://10.0.0.12:2379"],"use_perms":true,"osd_network":"10.0.0.0/24"}
+   (Note https:// etcd URLs!)
+2) Run: ${process.argv[1]} [./vitastor.conf] [--antietcd-only]
    You can run it on etcd/monitor nodes or on an external node.
-   It configures etcd, generates TLS certificates (on the first or external node), copies them
-   to other etcd/monitor nodes, and updates vitastor.conf with TLS options.
+   It configures etcd, generates TLS certificates (on the first or external node), copies
+   them to other etcd/monitor nodes, and updates vitastor.conf with TLS options.
 3) If you have OSD-only nodes, run:
    ${process.argv[1]} --copy-to-osd-node NODE_NAME ./vitastor.conf
    It copies vitastor.conf and required TLS certificates to that node.
 
 OPTIONS:
+--antietcd-only
+    disable etcd (proxy or direct mode), use only antietcd
 --gen-certs
     force certificate generation even if it's not the first node
 --no-certs
     disable certificate generation
---no-copy
-    do not copy initial certificates to other nodes
+--copy yes|no|ask
+    copy vitastor.conf and TLS certificates for monitor&etcd to monitor nodes using scp
+    (default is ask)
 --copy-to-osd-node NODE[,NODE2,...]
-    copy vitastor.conf and TLS certificates required for OSDs to NODES using scp
---copy-to-mon-node NODE[,NODE2,...]
-    copy vitastor.conf and TLS certificates required for monitor and etcd to NODES using scp
---copy-to-client-node NODE[,NODE2,...]
-    copy vitastor.conf and TLS certificates required for clients to NODES using scp
+    copy vitastor.conf and TLS certificates for OSDs to NODES using scp
 `;
 
 async function run()
@@ -45,21 +46,35 @@ async function run()
     let config_path = '/etc/vitastor/vitastor.conf';
     let config_dir = '/etc/vitastor/';
     let gen_certs = 'auto';
-    let copy_initial = true;
-    let copy_to_osd = 
+    let antietcd_only = false;
+    let copy = 'ask';
+    let copy_to_osd = null;
     for (let i = 2; i < process.argv.length; i++)
     {
+        const arg = process.argv[i];
         if (arg == '-h' || arg == '--help')
         {
             console.log(help_text);
             process.exit(0);
         }
-        else if (arg == '--only-certs')
+        else if (arg == '--gen-certs')
+        {
+            gen_certs = true;
+        }
+        else if (arg == '--no-certs')
+        {
+            gen_certs = false;
+        }
+        else if (arg == '--antietcd-only')
+        {
+            antietcd_only = true;
+        }
+        else if (arg == '--copy-to-osd-node' && i < process.argv.length-1)
         {
             i++;
-            gen_certs = 
+            copy_to_osd = process.argv[i].split(/,/);
         }
-        else if (arg == '--copy')
+        else if (arg == '--copy' && i < process.argv.length-1)
         {
             i++;
             copy = process.argv[i];
@@ -77,6 +92,7 @@ async function run()
         else
         {
             config_path = arg;
+            config_dir = path.dirname(arg);
         }
     }
     if (!fs.existsSync(config_path))
@@ -100,18 +116,22 @@ async function run()
             port: s[3],
         }));
     const tls = etcds.filter(e => e.scheme === 'https').length > 0;
-    const use_auth = tls && config.use_auth;
+    const use_perms = tls && config.use_perms;
     const num = select_local_etcd(etcds);
+    if (copy_to_osd)
+    {
+        copy_to_osd_nodes(copy_to_osd, config_dir, use_perms, antietcd_only);
+        process.exit(0);
+    }
     if (tls)
     {
-        if (gen_certs === 'yes')
+        const etcd_ca = config_dir+'/'+path.basename(config.etcd_ca);
+        if (gen_certs === true)
         {
-            gen_certs = true;
             console.log('Certificate generation is requested explicitly, generating');
         }
-        else if (gen_certs === 'no')
+        else if (gen_certs === false)
         {
-            gen_certs = false;
             console.log('Certificate generation is disabled explicitly, skipping');
         }
         else if (num < 0)
@@ -119,10 +139,10 @@ async function run()
             gen_certs = true;
             console.log('No matching IPs in etcd_address from '+config_path+', only generating certificates');
         }
-        else if (fs.existsSync("/etc/vitastor/etcd.crt"))
+        else if (config.etcd_ca && fs.existsSync(etcd_ca))
         {
             gen_certs = false;
-            console.log('/etc/vitastor/etcd.crt already exists, assuming certificates are already generated');
+            console.log(etcd_ca+' already exists, assuming certificates are already generated');
         }
         else if (num === 0)
         {
@@ -131,24 +151,25 @@ async function run()
         }
         else
         {
-            console.log('This is monitor node '+(num+1)+', /etc/vitastor/etcd.crt does not exist, please copy certificates to this node');
+            console.log('This is monitor node '+(num+1)+', '+etcd_ca+' does not exist, please copy certificates to this node');
             process.exit(1);
         }
+        await write_auth_config(config, config_path, etcds, use_perms, antietcd_only);
         if (gen_certs)
         {
             if (copy === 'ask')
                 copy = await ask_copy('Copy certificates and vitastor.conf to other nodes after generation?');
             copy = (copy === 'y' || copy === 'yes');
-            await make_certs(dir, copy);
+            await make_certs(config_dir, copy, etcds, use_perms, antietcd_only);
         }
-        await write_auth_config(config, config_path);
     }
     if (num < 0)
     {
         console.log('No matching IPs in etcd_address from '+config_path);
         process.exit(tls && gen_certs ? 0 : 1);
     }
-    await configure_etcd();
+    await configure_etcd(etcds, num, tls, use_perms);
+    await enable_mon();
     process.exit(0);
 }
 
@@ -160,83 +181,98 @@ async function ask_copy(question)
         prompt: '> ',
     });
     let copy;
-    while (true)
+    while (copy != 'y' && copy != 'n' && copy != 'yes' && copy != 'no')
     {
-        copy = await new Promise(ok => rl.question(question, ok));
-        if (copy != 'y' && copy != 'n' && copy != 'yes' && copy != 'no')
+        if (copy)
             console.log('Please type "yes" or "no"');
-        else
-            break;
+        copy = await new Promise(ok => rl.question(question, ok));
     }
     return copy;
 }
 
-async function make_certs(dir, copy)
+async function copy_to_osd_nodes(to, dir, use_perms, antietcd_only)
+{
+    const osd_to_copy = [ 'vitastor.conf' ];
+    if (!antietcd_only && !use_perms)
+        osd_to_copy.push('etcd_ca.crt');
+    else
+        osd_to_copy.push('antietcd_ca.crt');
+    if (use_perms)
+        osd_to_copy.push('osd.crt', 'osd.key', 'client_ca.crt');
+    console.warn('Copying configuration to OSD nodes '+to.join(', '));
+    for (const node of to)
+        await system("scp "+dir+osd_to_copy.join(" "+dir)+" root@"+node+":/etc/vitastor/");
+}
+
+async function make_certs(dir, copy, etcds, use_perms, antietcd_only)
 {
     console.log(`-----
 Generating certificates in ${dir}
 -----
 `);
-    await make_ca("/O=Vitastor etcd CA", dir+"etcd_ca");
-    await make_signed("/CN=Vitastor etcd", dir+"etcd", dir+"etcd_ca", etcds.map(e => "IP:"+e.ip).join(','));
-    if (use_auth)
+    const to_copy = [ 'vitastor.conf' ];
+    const osd_to_copy = [ 'vitastor.conf' ];
+    if (!antietcd_only)
+    {
+        await make_ca("/O=Vitastor etcd CA", dir+"etcd_ca");
+        await make_signed("/CN=Vitastor etcd", dir+"etcd", dir+"etcd_ca", etcds.map(e => "IP:"+e.ip).join(','));
+        to_copy.push('etcd_ca.crt', 'etcd.crt', 'etcd.key');
+        if (!use_perms)
+            osd_to_copy.push('etcd_ca.crt');
+    }
+    if (use_perms || antietcd_only)
     {
         await make_ca("/O=Vitastor Antietcd CA", dir+"antietcd_ca");
         await make_signed("/CN=Vitastor Antietcd", dir+"antietcd", dir+"antietcd_ca", etcds.map(e => "IP:"+e.ip).join(','));
+        to_copy.push('antietcd_ca.crt', 'antietcd.crt', 'antietcd.key');
+        osd_to_copy.push('antietcd_ca.crt');
+    }
+    if (use_perms)
+    {
         await make_ca("/CN=Vitastor OSD", dir+"osd");
         await make_ca("/O=Vitastor Client CA", dir+"client_ca");
         await make_signed("/CN=admin", dir+"admin", dir+"client_ca");
+        to_copy.push('osd.crt', 'osd.key', 'client_ca.crt');
+        osd_to_copy.push('osd.crt', 'osd.key', 'client_ca.crt');
     }
-    if (use_auth)
-    {
-        console.log(`-----
+    console.log(`-----
 Certificates generated, commands to copy them:
 - Monitor+OSD node:
-  cd ${dir} && scp antietcd_ca.crt antietcd.crt antietcd.key osd.crt osd.key client_ca.crt etcd_ca.crt etcd.crt etcd.key root@NODE:/etc/vitastor/
+  cd ${dir} && scp ${to_copy.join(' ')} root@NODE:/etc/vitastor/
 - Monitor node:
-  cd ${dir} && scp antietcd_ca.crt antietcd.crt antietcd.key osd.crt client_ca.crt etcd_ca.crt etcd.crt etcd.key root@NODE:/etc/vitastor/
+  cd ${dir} && scp ${to_copy.filter(f => f != 'osd.key').join(' ')} root@NODE:/etc/vitastor/
 - OSD node:
-  cd ${dir} && scp antietcd_ca.crt osd.crt osd.key client_ca.crt root@NODE:/etc/vitastor/
+  cd ${dir} && scp ${osd_to_copy.join(' ')} root@NODE:/etc/vitastor/
 -----
 `);
-    }
-    else
-    {
-        console.log(`-----
-Certificates generated, commands to copy them:
-- Monitor node:
-  cd ${dir} && scp etcd_ca.crt etcd.crt etcd.key root@NODE:/etc/vitastor/
------
-`);
-    }
     if (copy)
     {
-        const to_copy = use_auth
-            ? [ "antietcd_ca.crt", "antietcd.crt", "antietcd.key", "osd.crt", "osd.key", "client_ca.crt", "etcd_ca.crt", "etcd.crt", "etcd.key" ]
-            : [ "etcd_ca.crt", "etcd.crt", "etcd.key" ];
         for (const node of etcds)
         {
-            await system("scp "+dir+to_copy.join(" "+dir)+" root@"+node.ip+"/etc/vitastor/");
+            await system("scp "+dir+to_copy.join(" "+dir)+" root@"+node.ip+":/etc/vitastor/");
         }
     }
     else
     {
-        console.warn('Certificates generated in /etc/vitastor, please copy them to other nodes');
+        console.warn('Certificates generated in '+dir+', please copy them to other nodes');
     }
 }
 
-async function write_auth_config(config, config_path)
+async function write_auth_config(config, config_path, etcds, use_perms, antietcd_only)
 {
     const auth = {};
-    if (use_auth)
+    if (use_perms)
     {
         auth["use_antietcd"] = true;
-        auth["etcd_proxy"] = {
-            urls: etcds.map(e => e.ip+':2381'),
-            cert: "/etc/vitastor/antietcd.crt",
-            key: "/etc/vitastor/antietcd.key",
-            ca: "/etc/vitastor/etcd_ca.crt",
-        };
+        if (!antietcd_only)
+        {
+            auth["etcd_proxy"] = {
+                urls: etcds.map(e => e.ip+':2381'),
+                cert: "/etc/vitastor/antietcd.crt",
+                key: "/etc/vitastor/antietcd.key",
+                ca: "/etc/vitastor/etcd_ca.crt",
+            };
+        }
         auth["antietcd_cert"] = "/etc/vitastor/antietcd.crt";
         auth["antietcd_key"] = "/etc/vitastor/antietcd.key";
         auth["etcd_ca"] = "/etc/vitastor/antietcd_ca.crt";
@@ -249,7 +285,17 @@ async function write_auth_config(config, config_path)
     }
     else
     {
-        auth["etcd_ca"] = "/etc/vitastor/etcd.crt";
+        if (antietcd_only)
+        {
+            auth["use_antietcd"] = true;
+            auth["antietcd_cert"] = "/etc/vitastor/antietcd.crt";
+            auth["antietcd_key"] = "/etc/vitastor/antietcd.key";
+            auth["etcd_ca"] = "/etc/vitastor/antietcd_ca.crt";
+        }
+        else
+        {
+            auth["etcd_ca"] = "/etc/vitastor/etcd.crt";
+        }
     }
     for (const k in auth)
     {
@@ -271,7 +317,7 @@ Updating ${config_path}
     fs.writeFileSync(config_path, JSON.stringify(config, 0, 4));
 }
 
-async configure_etcd()
+async function configure_etcd(etcds, num, tls, use_perms)
 {
     const in_docker = fs.existsSync("/etc/vitastor/etcd.conf") &&
         fs.existsSync("/etc/vitastor/docker.conf");
@@ -288,8 +334,8 @@ async configure_etcd()
     const etcd_url = etcds[num].scheme + '://' + etcds[num].addr;
     const options = {
         name: 'etcd'+etcds[num].ip.replace(/[^0-9a-z_]/ig, '_'),
-        advertise_client_urls: etcd_url+':'+(use_auth ? 2381 : 2379),
-        listen_client_urls: etcd_url+':'+(use_auth ? 2381 : 2379),
+        advertise_client_urls: etcd_url+':'+(use_perms ? 2381 : 2379),
+        listen_client_urls: etcd_url+':'+(use_perms ? 2381 : 2379),
         initial_advertise_peer_urls: etcd_url+':2380',
         listen_peer_urls: etcd_url+':2380',
         initial_cluster_token: 'vitastor-etcd-1',
@@ -305,14 +351,14 @@ async configure_etcd()
     {
         options['cert_file'] = '/etc/vitastor/etcd.crt';
         options['key_file'] = '/etc/vitastor/etcd.key';
-        if (use_auth)
+        if (use_perms)
         {
             options['client_cert_auth'] = '1';
             options['trusted_ca_file'] = '/etc/vitastor/antietcd.crt';
         }
         options['peer_cert_file'] = '/etc/vitastor/etcd.crt';
         options['peer_key_file'] = '/etc/vitastor/etcd.key';
-        if (use_auth)
+        if (use_perms)
         {
             options['peer_client_cert_auth'] = '1';
             options['peer_trusted_ca_file'] = '/etc/vitastor/etcd.crt';
@@ -333,8 +379,7 @@ async configure_etcd()
     }
     await system(`mkdir -p /var/lib/etcd/vitastor`);
     fs.writeFileSync(
-        "/etc/systemd/system/vitastor-etcd.service",
-`[Unit]
+        "/etc/systemd/system/vitastor-etcd.service", `[Unit]
 Description=etcd for vitastor
 After=network-online.target local-fs.target time-sync.target
 Wants=network-online.target local-fs.target time-sync.target
@@ -363,6 +408,11 @@ WantedBy=multi-user.target
     await system(`systemctl enable --now vitastor-etcd`);
 }
 
+async function enable_mon()
+{
+    await system(`systemctl enable --now vitastor-mon`);
+}
+
 function replace_env(text, key, value)
 {
     let found = false;

mon/vitastor_auth_filter.js

@@ -68,10 +68,10 @@ class VitastorAuthFilter
 
     async init()
     {
-        if (!this.cfg.cert || !this.cfg.key || !this.cfg.osd_ca || !this.cfg.etcd_proxy && !this.cfg.peer_ca || !this.cfg.client_cert_auth)
+        if (!this.cfg.cert || !this.cfg.key || !this.cfg.ca || !this.cfg.osd_ca || !this.cfg.client_cert_auth)
         {
             throw new Error('Authenticated Vitastor setups require enabled client_cert_auth, cert, key'+
-                ' and separate ca (client CA), osd_ca'+(this.cfg.etcd_proxy ? '' : ', peer_ca')+' and optionally mon_ca');
+                ' and separate ca (client CA), osd_ca and optionally mon_ca');
         }
         this.osd_ca = await this.antietcd.readPEM(this.cfg.osd_ca);
         this.osd_ca_obj = new X509Certificate(this.osd_ca);

node-binding/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vitastor",
-  "version": "3.0.12",
+  "version": "3.0.15",
   "description": "Low-level native bindings to Vitastor client library",
   "main": "index.js",
   "keywords": [

patches/VitastorPlugin.pm

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

patches/cinder-vitastor.py

@@ -50,7 +50,7 @@ from cinder.volume import configuration
 from cinder.volume import driver
 from cinder.volume import volume_utils
 
-VITASTOR_VERSION = '3.0.12'
+VITASTOR_VERSION = '3.0.15'
 
 LOG = logging.getLogger(__name__)
 

rpm/vitastor-el10.spec

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        3.0.12
+Version:        3.0.15
 Release:        1%{?dist}
 Summary:        Vitastor, a fast software-defined clustered block storage
 
 License:        Vitastor Network Public License 1.1
 URL:            https://vitastor.io/
-Source0:        vitastor-3.0.12.el10.tar.gz
+Source0:        vitastor-3.0.15.el10.tar.gz
 
 BuildRequires:  gperftools-devel
 BuildRequires:  gcc-c++

rpm/vitastor-el7.spec

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        3.0.12
+Version:        3.0.15
 Release:        1%{?dist}
 Summary:        Vitastor, a fast software-defined clustered block storage
 
 License:        Vitastor Network Public License 1.1
 URL:            https://vitastor.io/
-Source0:        vitastor-3.0.12.el7.tar.gz
+Source0:        vitastor-3.0.15.el7.tar.gz
 
 BuildRequires:  gperftools-devel
 BuildRequires:  devtoolset-9-gcc-c++

rpm/vitastor-el8.spec

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        3.0.12
+Version:        3.0.15
 Release:        1%{?dist}
 Summary:        Vitastor, a fast software-defined clustered block storage
 
 License:        Vitastor Network Public License 1.1
 URL:            https://vitastor.io/
-Source0:        vitastor-3.0.12.el8.tar.gz
+Source0:        vitastor-3.0.15.el8.tar.gz
 
 BuildRequires:  gperftools-devel
 BuildRequires:  gcc-toolset-9-gcc-c++

rpm/vitastor-el9.spec

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        3.0.12
+Version:        3.0.15
 Release:        1%{?dist}
 Summary:        Vitastor, a fast software-defined clustered block storage
 
 License:        Vitastor Network Public License 1.1
 URL:            https://vitastor.io/
-Source0:        vitastor-3.0.12.el9.tar.gz
+Source0:        vitastor-3.0.15.el9.tar.gz
 
 BuildRequires:  gperftools-devel
 BuildRequires:  gcc-c++

src/CMakeLists.txt

@@ -20,7 +20,7 @@ if("${CMAKE_INSTALL_PREFIX}" MATCHES "^/usr/local/?$")
 endif()
 set(ENABLE_COVERAGE false CACHE BOOL "Enable code coverage")
 
-add_definitions(-DVITASTOR_VERSION="3.0.12")
+add_definitions(-DVITASTOR_VERSION="3.0.15")
 add_definitions(-D_GNU_SOURCE -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -Wall -Wno-sign-compare -Wno-comment -Wno-parentheses -Wno-pointer-arith -fdiagnostics-color=always -fno-omit-frame-pointer -fvisibility=hidden -I ${CMAKE_SOURCE_DIR}/src)
 add_link_options(-fno-omit-frame-pointer)
 if (${WITH_ASAN})

src/blockstore/blockstore_disk.cpp

@@ -521,7 +521,7 @@ void blockstore_disk_t::close_all()
 
 // Sadly DISCARD only works through ioctl(), but it seems to always block the device queue,
 // so it's not a big deal that we can only run it synchronously.
-int blockstore_disk_t::trim_data(std::function<bool(uint64_t)> is_free)
+int blockstore_disk_t::trim_data(std::function<bool(uint64_t)> is_used)
 {
     if (mock_mode)
     {
@@ -532,7 +532,7 @@ int blockstore_disk_t::trim_data(std::function<bool(uint64_t)> is_free)
     uint64_t discarded = 0;
     for (; i <= block_count; i++)
     {
-        if (i >= block_count || is_free(i))
+        if (i >= block_count || is_used(i))
         {
             if (i > j && (i-j)*data_block_size >= min_discard_size)
             {
@@ -545,17 +545,21 @@ int blockstore_disk_t::trim_data(std::function<bool(uint64_t)> is_free)
                     if (range[0] % discard_granularity)
                         range[0] = range[0] + discard_granularity - (range[0] % discard_granularity);
                     if (range[0] >= range[1])
-                        continue;
-                    range[1] -= range[0];
+                        range[1] = 0;
+                    else
+                        range[1] -= range[0];
                 }
-                r = ioctl(data_fd, BLKDISCARD, &range);
-                if (r != 0)
+                if (range[1] > 0)
                 {
-                    fprintf(stderr, "Failed to execute BLKDISCARD %ju+%ju on %s: %s (code %d)\n",
-                        range[0], range[1], data_device.c_str(), strerror(-r), r);
-                    return -errno;
+                    r = ioctl(data_fd, BLKDISCARD, &range);
+                    if (r != 0)
+                    {
+                        fprintf(stderr, "Failed to execute BLKDISCARD %ju+%ju on %s: %s (code %d)\n",
+                            range[0], range[1], data_device.c_str(), strerror(-r), r);
+                        return -errno;
+                    }
+                    discarded += range[1];
                 }
-                discarded += range[1];
             }
             j = i+1;
         }

src/blockstore/blockstore_disk.h

@@ -84,7 +84,7 @@ struct blockstore_disk_t
     void calc_lengths(bool skip_meta_check = false);
     void check_lengths();
     void close_all();
-    int trim_data(std::function<bool(uint64_t)> is_free);
+    int trim_data(std::function<bool(uint64_t)> is_used);
 
     inline uint64_t dirty_dyn_size(uint64_t offset, uint64_t len)
     {

src/blockstore/blockstore_flush.cpp

@@ -289,30 +289,41 @@ resume_1:
     {
         init_fsync_data();
     }
-    if (bs->log_level > 10)
+    if (compact_info.do_delete)
     {
-        printf("Compacting %jx:%jx v%ju..v%ju / l%ju..l%ju (%d writes)\n", cur_oid.inode, cur_oid.stripe,
-            compact_info.clean_wr->version, compact_info.compact_version,
-            compact_info.clean_wr->lsn, compact_info.compact_lsn, copy_count);
-    }
-    mem_or(new_bmp, compact_info.clean_wr->get_int_bitmap(bs->heap), bs->dsk.clean_entry_bitmap_size);
-    if (!bitmap_copied)
-    {
-        memcpy(new_ext_bmp, compact_info.clean_wr->get_ext_bitmap(bs->heap), bs->dsk.clean_entry_bitmap_size);
-        bitmap_copied = true;
-    }
-    if (bs->dsk.csum_block_size && bs->dsk.csum_block_size <= bs->dsk.bitmap_granularity)
-    {
-        memcpy(new_csums, compact_info.clean_wr->get_checksums(bs->heap), bs->dsk.data_block_size/bs->dsk.csum_block_size * (bs->dsk.data_csum_type & 0xFF));
-        for (size_t i = csum_copy.size(); i > 0; i--)
+        if (bs->log_level > 10)
         {
-            auto wr = csum_copy[i-1];
-            memcpy(new_csums + wr->small().offset/bs->dsk.csum_block_size*(bs->dsk.data_csum_type & 0xFF),
-                wr->get_checksums(bs->heap), wr->small().len/bs->dsk.csum_block_size*(bs->dsk.data_csum_type & 0xFF));
+            printf("Compacting %jx:%jx up to l%ju (delete)\n", cur_oid.inode, cur_oid.stripe, compact_info.compact_lsn);
         }
-        csum_copy.clear();
+        clean_loc = UINT64_MAX;
+    }
+    else
+    {
+        if (bs->log_level > 10)
+        {
+            printf("Compacting %jx:%jx v%ju..v%ju / l%ju..l%ju (%d writes)\n", cur_oid.inode, cur_oid.stripe,
+                compact_info.clean_wr->version, compact_info.compact_version,
+                compact_info.clean_wr->lsn, compact_info.compact_lsn, copy_count);
+        }
+        mem_or(new_bmp, compact_info.clean_wr->get_int_bitmap(bs->heap), bs->dsk.clean_entry_bitmap_size);
+        if (!bitmap_copied)
+        {
+            memcpy(new_ext_bmp, compact_info.clean_wr->get_ext_bitmap(bs->heap), bs->dsk.clean_entry_bitmap_size);
+            bitmap_copied = true;
+        }
+        if (bs->dsk.csum_block_size && bs->dsk.csum_block_size <= bs->dsk.bitmap_granularity)
+        {
+            memcpy(new_csums, compact_info.clean_wr->get_checksums(bs->heap), bs->dsk.data_block_size/bs->dsk.csum_block_size * (bs->dsk.data_csum_type & 0xFF));
+            for (size_t i = csum_copy.size(); i > 0; i--)
+            {
+                auto wr = csum_copy[i-1];
+                memcpy(new_csums + wr->small().offset/bs->dsk.csum_block_size*(bs->dsk.data_csum_type & 0xFF),
+                    wr->get_checksums(bs->heap), wr->small().len/bs->dsk.csum_block_size*(bs->dsk.data_csum_type & 0xFF));
+            }
+            csum_copy.clear();
+        }
+        clean_loc = compact_info.clean_wr->big_location(bs->heap);
     }
-    clean_loc = compact_info.clean_wr->big_location(bs->heap);
     overwrite_start = overwrite_end = 0;
     if (read_vec.size() > 0)
     {
@@ -625,7 +636,7 @@ int journal_flusher_co::check_and_punch_checksums()
 
 bool journal_flusher_co::calc_block_checksums()
 {
-    if (bs->dsk.csum_block_size <= bs->dsk.bitmap_granularity)
+    if (bs->dsk.csum_block_size <= bs->dsk.bitmap_granularity || compact_info.do_delete)
     {
         return true;
     }

src/blockstore/blockstore_heap.cpp

@@ -23,7 +23,7 @@
 
 #define HEAP_INFLIGHT_DONE 1
 #define HEAP_INFLIGHT_COMPACTABLE 2
-#define HEAP_INFLIGHT_COMPACTED 4
+#define HEAP_INFLIGHT_OVERWRITE 4
 #define HEAP_INFLIGHT_GC 8
 #define HEAP_INFLIGHT_EXPLICIT 16
 
@@ -123,6 +123,8 @@ uint32_t heap_entry_t::get_size(blockstore_heap_t *heap)
     }
     if (type() == BS_HEAP_SMALL_WRITE || type() == BS_HEAP_INTENT_WRITE)
     {
+        if (size < sizeof(heap_small_write_t))
+            return heap->get_small_entry_size(0, 0);
         return heap->get_small_entry_size(small().offset, small().len);
     }
     return heap->get_simple_entry_size();
@@ -374,14 +376,11 @@ corrupted_object:
                     return EDOM;
                 }
             }
-            if (((wr->entry_type & BS_HEAP_TYPE) == BS_HEAP_SMALL_WRITE ||
-                (wr->entry_type & BS_HEAP_TYPE) == BS_HEAP_INTENT_WRITE) &&
-                wr->size < sizeof(heap_small_write_t))
+            if (wr->size != wr->get_size(this))
             {
-                // Small writes require accessing offset & len to calculate correct length,
-                // so require at least sizeof(heap_small_write_t) for them
-                fprintf(stderr, "Error: entry %jx:%jx v%ju has invalid size in metadata block %u at %u (%u < min %zu bytes)\n",
-                    wr->inode, wr->stripe, wr->version, block_num, block_offset, wr->size, sizeof(heap_small_write_t));
+                // Check entry size
+                fprintf(stderr, "Error: entry %jx:%jx v%ju has invalid size in metadata block %u at %u (%u != %u bytes)\n",
+                    wr->inode, wr->stripe, wr->version, block_num, block_offset, wr->size, wr->get_size(this));
                 goto corrupted_object;
             }
             if (wr->entry_type == BS_HEAP_COMMIT && !wr->version)
@@ -570,7 +569,7 @@ void blockstore_heap_t::finish_load()
         size_t s = 0, e, n = postponed_items.size();
         for (e = 1; e <= n; e++)
         {
-            if (e >= n || postponed_items[e]->entry.inode != postponed_items[s]->entry.inode &&
+            if (e >= n || postponed_items[e]->entry.inode != postponed_items[s]->entry.inode ||
                 postponed_items[e]->entry.stripe != postponed_items[s]->entry.stripe)
             {
                 insert_list_items(postponed_items.data()+s, e-s, false);
@@ -693,10 +692,14 @@ int blockstore_heap_t::mark_used_blocks()
                         }
                         use_data(wr->inode, wr->big_location(this));
                     }
-                    if (wr->is_compactable() && !added)
+                    if (wr->is_compactable())
                     {
-                        compact_queue.push_back((object_id){ .inode = wr->inode, .stripe = wr->stripe });
-                        added = true;
+                        to_compact_count++;
+                        if (!added)
+                        {
+                            compact_queue.push_back((object_id){ .inode = wr->inode, .stripe = wr->stripe });
+                            added = true;
+                        }
                     }
                     if (wr->is_overwrite())
                     {
@@ -706,6 +709,11 @@ int blockstore_heap_t::mark_used_blocks()
             });
         }
     }
+    for (auto li: init_erase_items)
+    {
+        unlink_list_item(li);
+    }
+    init_erase_items.clear();
     if (dsk->gc_on_start)
     {
         recheck_full_gc();
@@ -741,7 +749,6 @@ void blockstore_heap_t::init_erase_bad_entry(heap_list_item_t *li)
         inf.garbage_space -= (li->entry.is_garbage() ? li->entry.size : 0);
     });
     recheck_modified_blocks.insert(li->block_num);
-    unlink_list_item(li);
 }
 
 bool blockstore_heap_t::init_erase_double_claim(heap_list_item_t *prev_li, heap_list_item_t *cur_li)
@@ -796,6 +803,8 @@ bool blockstore_heap_t::init_erase_double_claim(heap_list_item_t *prev_li, heap_
                 overwritten = erase_li->entry.is_overwrite();
             }
             init_erase_bad_entry(erase_li);
+            // Can't erase (mutate map) while iterating, so postpone it
+            init_erase_items.push_back(erase_li);
             erase_li = prev_erase_li;
         }
     }
@@ -809,6 +818,8 @@ bool blockstore_heap_t::init_erase_double_claim(heap_list_item_t *prev_li, heap_
             auto next_erase_li = erase_li->next;
             init_free_bad_entry(&erase_li->entry);
             init_erase_bad_entry(erase_li);
+            // Can't erase (mutate map) while iterating, so postpone it
+            init_erase_items.push_back(erase_li);
             erase_li = next_erase_li;
         }
         erase_li = cur_li;
@@ -817,6 +828,8 @@ bool blockstore_heap_t::init_erase_double_claim(heap_list_item_t *prev_li, heap_
         {
             auto prev_erase_li = erase_li->prev;
             init_erase_bad_entry(erase_li);
+            // Can't erase (mutate map) while iterating, so postpone it
+            init_erase_items.push_back(erase_li);
             erase_li = prev_erase_li;
         }
     }
@@ -887,6 +900,7 @@ void blockstore_heap_t::recheck_drop_entries(heap_entry_t *obj, heap_entry_t *ba
         auto prev = li->prev;
         assert(li->entry.type() == bad_wr->type());
         init_erase_bad_entry(li);
+        unlink_list_item(li);
         li = prev;
     }
 }
@@ -1168,7 +1182,7 @@ bool blockstore_heap_t::calc_block_checksums(uint32_t *block_csums, uint8_t *bit
                 while (pos < end && pos < block_end && !(bitmap[pos/dsk->bitmap_granularity/8] & (1 << ((pos/dsk->bitmap_granularity) % 8))))
                     pos += dsk->bitmap_granularity;
                 // zero padding at the beginning or at the end of the block is not counted
-                if (pos > prev && prev > 0 && pos < block_end)
+                if (pos > prev && prev > blk_start && pos < block_end)
                 {
                     if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32)
                     {
@@ -1613,7 +1627,7 @@ int blockstore_heap_t::add_entry(uint32_t wr_size, uint32_t *modified_block,
     // Remember the object as dirty and remove older entries when this block is written and fsynced
     push_inflight_lsn(next_lsn, new_wr,
         (explicit_complete ? HEAP_INFLIGHT_EXPLICIT : 0) |
-        (new_wr->is_overwrite() ? HEAP_INFLIGHT_COMPACTED : 0) |
+        (new_wr->is_overwrite() ? HEAP_INFLIGHT_OVERWRITE : 0) |
         (new_wr->is_compactable() ? HEAP_INFLIGHT_COMPACTABLE : 0));
     insert_list_items(&li, 1, false);
     li->block_num = block_num;
@@ -1647,10 +1661,22 @@ int blockstore_heap_t::add_small_write(object_id oid, heap_entry_t **obj_ptr, ui
         wr->small().location = location;
         if (bitmap)
             memcpy(wr->get_ext_bitmap(this), bitmap, dsk->clean_entry_bitmap_size);
-        else if (obj)
-            memcpy(wr->get_ext_bitmap(this), obj->get_ext_bitmap(this), dsk->clean_entry_bitmap_size);
         else
-            memset(wr->get_ext_bitmap(this), 0, dsk->clean_entry_bitmap_size);
+        {
+            bool found = false;
+            iterate_with_stable(obj, UINT64_MAX, [&](heap_entry_t *old_wr, bool stable)
+            {
+                if (old_wr->get_ext_bitmap(this))
+                {
+                    found = true;
+                    memcpy(wr->get_ext_bitmap(this), old_wr->get_ext_bitmap(this), dsk->clean_entry_bitmap_size);
+                    return false;
+                }
+                return true;
+            });
+            if (!found)
+                memset(wr->get_ext_bitmap(this), 0, dsk->clean_entry_bitmap_size);
+        }
         calc_checksums(wr, (uint8_t*)data, true);
         *obj_ptr = wr;
     });
@@ -2114,7 +2140,10 @@ void blockstore_heap_t::iterate_with_stable(heap_entry_t *obj, uint64_t max_lsn,
     {
         if (old_wr->type() == BS_HEAP_ROLLBACK)
         {
-            rollback_version = old_wr->version;
+            if (rollback_version > old_wr->version)
+            {
+                rollback_version = old_wr->version;
+            }
         }
         else if (old_wr->type() == BS_HEAP_COMMIT)
         {
@@ -2166,7 +2195,10 @@ heap_compact_t blockstore_heap_t::iterate_compaction(heap_entry_t *obj, uint64_t
                 res.compact_lsn = wr->lsn;
                 res.compact_version = wr->version;
             }
-            rollback_version = wr->version;
+            if (rollback_version > wr->version)
+            {
+                rollback_version = wr->version;
+            }
             continue;
         }
         if (wr->type() == BS_HEAP_COMMIT && wr->lsn <= fsynced_lsn)
@@ -2344,7 +2376,7 @@ void blockstore_heap_t::use_data(inode_t inode, uint64_t location)
 {
     auto sh_it = pool_shard_settings.find(INODE_POOL(inode));
     if (sh_it != pool_shard_settings.end() && sh_it->second.no_inode_stats)
-        inode = (INODE_POOL(inode) << POOL_ID_BITS);
+        inode = INODE_WITH_POOL(INODE_POOL(inode), 0);
     assert(!data_alloc->get(location / dsk->data_block_size));
     data_alloc->set(location / dsk->data_block_size, true);
     inode_space_stats[inode] += dsk->data_block_size;
@@ -2355,7 +2387,7 @@ void blockstore_heap_t::free_data(inode_t inode, uint64_t location)
 {
     auto sh_it = pool_shard_settings.find(INODE_POOL(inode));
     if (sh_it != pool_shard_settings.end() && sh_it->second.no_inode_stats)
-        inode = (INODE_POOL(inode) << POOL_ID_BITS);
+        inode = INODE_WITH_POOL(INODE_POOL(inode), 0);
     assert(data_alloc->get(location / dsk->data_block_size));
     data_alloc->set(location / dsk->data_block_size, false);
     auto sp_it = inode_space_stats.find(inode);
@@ -2392,7 +2424,8 @@ void blockstore_heap_t::use_buffer_area(inode_t inode, uint64_t location, uint64
         return;
     }
     assert(!(size % dsk->bitmap_granularity));
-    buffer_alloc->use(location / dsk->bitmap_granularity, size / dsk->bitmap_granularity);
+    bool ok = buffer_alloc->use(location / dsk->bitmap_granularity, size / dsk->bitmap_granularity);
+    assert(ok);
     buffer_area_used_space += size;
 }
 
@@ -2434,7 +2467,7 @@ void blockstore_heap_t::get_meta_block(uint32_t block_num, uint8_t *buffer)
     }
 }
 
-void blockstore_heap_t::fill_block_empty_space(uint8_t *buffer, uint32_t pos)
+void blockstore_heap_t::fill_block_empty_space(uint8_t *buffer, uint64_t pos)
 {
     if (pos > dsk->meta_block_size)
     {
@@ -2522,7 +2555,7 @@ uint64_t blockstore_heap_t::get_garbage_memory()
 void blockstore_heap_t::push_inflight_lsn(uint64_t lsn, heap_entry_t *wr, uint64_t flags)
 {
     uint64_t next_inf = first_inflight_lsn + inflight_lsn.size();
-    if (flags & (HEAP_INFLIGHT_COMPACTABLE|HEAP_INFLIGHT_COMPACTED))
+    if (flags & (HEAP_INFLIGHT_COMPACTABLE|HEAP_INFLIGHT_OVERWRITE))
     {
         to_compact_count++;
     }
@@ -2589,7 +2622,7 @@ void blockstore_heap_t::mark_lsn_fsynced(uint64_t lsn)
 void blockstore_heap_t::apply_inflight(heap_inflight_lsn_t & inflight)
 {
     auto wr = inflight.wr;
-    if (inflight.flags & HEAP_INFLIGHT_COMPACTED)
+    if (inflight.flags & HEAP_INFLIGHT_OVERWRITE)
     {
         // Mark previous entries as garbage, sequentially
         mark_garbage_up_to(wr);

src/blockstore/blockstore_heap.h

@@ -220,6 +220,7 @@ class blockstore_heap_t
     bool marked_used_blocks = false;
     bool recheck_queue_filled = false;
     std::vector<heap_list_item_t*> postponed_items;
+    std::vector<heap_list_item_t*> init_erase_items;
     std::set<uint32_t> recheck_modified_blocks;
     std::deque<heap_entry_t*> recheck_queue;
     std::map<heap_entry_t*, heap_recheck_state_t> recheck_states;
@@ -362,7 +363,7 @@ public:
 
     // get metadata block data buffer and used space
     void get_meta_block(uint32_t block_num, uint8_t *buffer);
-    void fill_block_empty_space(uint8_t *buffer, uint32_t pos);
+    void fill_block_empty_space(uint8_t *buffer, uint64_t pos);
     uint32_t get_meta_block_used_space(uint32_t block_num);
 
     // get space usage statistics

src/blockstore/blockstore_impl.h

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

src/blockstore/blockstore_init.cpp

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

src/blockstore/blockstore_init.h

@@ -17,6 +17,7 @@ class blockstore_init_meta
     int wait_state = 0;
     int wait_count = 0;
     bool zero_on_init = false;
+    bool header_written = false;
     void *metadata_buffer = NULL;
     blockstore_init_meta_buf bufs[2] = {};
     int submitted = 0;
@@ -29,7 +30,7 @@ class blockstore_init_meta
     std::vector<uint32_t> recheck_mod;
     int i = 0, j = 0;
     bool handle_meta_block(uint8_t *buf, uint64_t count, uint64_t done_cnt);
-    void handle_event(ring_data_t *data, int buf_num);
+    void handle_event(ring_data_t *data, int buf_num, const char *op);
 public:
     blockstore_init_meta(blockstore_impl_t *bs);
     int loop();

src/blockstore/blockstore_mock.cpp

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

src/blockstore/blockstore_mock.h

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

src/blockstore/blockstore_stable.cpp

@@ -28,6 +28,7 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
                 FINISH_OP(op);
                 return 2;
             }
+            priv->modified_block2 = UINT32_MAX;
             int res = op->opcode == BS_OP_STABLE
                 ? heap->add_commit(obj, v[priv->stab_pos].version, &priv->modified_block2)
                 : heap->add_rollback(obj, v[priv->stab_pos].version, &priv->modified_block2);
@@ -52,11 +53,6 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
                     FINISH_OP(op);
                     return 2;
                 }
-                if (priv->modified_block2 != UINT32_MAX)
-                {
-                    priv->stab_pos--;
-                    goto resume_1;
-                }
                 priv->wait_for = WAIT_COMPACTION;
                 priv->wait_detail = heap->get_compacted_count();
                 flusher->request_trim();

src/blockstore/blockstore_sync.cpp

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

src/blockstore/blockstore_write.cpp

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

src/blockstore/multilist.cpp

@@ -12,7 +12,7 @@ multilist_alloc_t::multilist_alloc_t(uint32_t count, uint32_t maxn):
     count(count), maxn(maxn)
 {
     // not-so-memory-efficient: 16 MB memory per 1 GB buffer space, but buffer spaces are small, so OK
-    assert(count > 1 && count < 0x80000000);
+    assert(count > 1 && count < 0x80000000 && count >= maxn);
     sizes.resize(count);
     nexts.resize(count); // nexts[i] = 0 -> area is used; nexts[i] = 1 -> no next; nexts[i] >= 2 -> next item
     prevs.resize(count);
@@ -171,7 +171,7 @@ void multilist_alloc_t::print()
     printf("\n");
 }
 
-void multilist_alloc_t::use(uint32_t pos, uint32_t size)
+bool multilist_alloc_t::use(uint32_t pos, uint32_t size)
 {
     assert(pos+size <= count && size > 0);
     if (sizes[pos] <= 0)
@@ -182,7 +182,8 @@ void multilist_alloc_t::use(uint32_t pos, uint32_t size)
         else
             while (start > 0 && !sizes[start])
                 start--;
-        assert(sizes[start] >= size);
+        if (sizes[start] < size+(pos-start))
+            return false;
         use_full(start);
         uint32_t full = sizes[start];
         sizes[pos-1] = -pos+start;
@@ -199,7 +200,8 @@ void multilist_alloc_t::use(uint32_t pos, uint32_t size)
     }
     else
     {
-        assert(sizes[pos] >= size);
+        if (sizes[pos] < size)
+            return false;
         use_full(pos);
         if (sizes[pos] > size)
         {
@@ -214,12 +216,13 @@ void multilist_alloc_t::use(uint32_t pos, uint32_t size)
 #ifdef MULTILIST_TRACE
     print();
 #endif
+    return true;
 }
 
 void multilist_alloc_t::use_full(uint32_t pos)
 {
     uint32_t prevsize = sizes[pos];
-    assert(prevsize);
+    assert(prevsize > 0);
     assert(nexts[pos]);
     uint32_t pi = (prevsize < maxn ? prevsize : maxn)-1;
     if (heads[pi] == pos+1)

src/blockstore/multilist.h

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

src/blockstore/v1/flush.cpp

@@ -71,6 +71,11 @@ bool journal_flusher_t::is_active()
     return active_flushers > 0 || dequeuing;
 }
 
+size_t journal_flusher_t::get_queue_size()
+{
+    return flush_queue.size();
+}
+
 void journal_flusher_t::loop()
 {
     target_flusher_count = bs->write_iodepth*2;
@@ -384,6 +389,7 @@ stop_flusher:
         wait_state = 0;
         return true;
     }
+    copy_count = 0;
     try_trim = true;
     cur.oid = flusher->flush_queue.front();
     cur.version = flusher->flush_versions[cur.oid];
@@ -511,6 +517,31 @@ resume_2:
             {
                 uo_it->second.was_changed = true;
             }
+            if (!bs->journal.inmemory)
+            {
+                // Verify journaled data checksums (but not COALESCED)
+                for (it = v.begin(); it != v.end(); it++)
+                {
+                    if (it->copy_flags == COPY_BUF_JOURNAL)
+                    {
+                        iovec iov = { .iov_base = it->buf, .iov_len = it->len };
+                        bs->verify_journal_checksums(
+                            it->csum_buf, it->offset, &iov, 1,
+                            [&](uint32_t bad_block, uint32_t calc_csum, uint32_t stored_csum)
+                            {
+                                printf(
+                                    "Checksum mismatch in object %jx:%jx v%ju in journal at 0x%jx, checksum block #%u: got %08x, expected %08x\n",
+                                    cur.oid.inode, cur.oid.stripe, cur.version, it->disk_offset,
+                                    bad_block / bs->dsk.csum_block_size, calc_csum, stored_csum
+                                );
+                                bad_block += it->offset;
+                                assert(!(bad_block % bs->dsk.csum_block_size) && bad_block < bs->dsk.data_block_size);
+                                mangle_csum_blocks.insert(bad_block);
+                            }
+                        );
+                    }
+                }
+            }
         }
         // Submit data writes
         for (it = v.begin(); it != v.end(); it++)
@@ -634,6 +665,7 @@ resume_2:
         }
         // All done
         flusher->active_flushers--;
+        copy_count = 0; // used by is_mutated()...
         wait_state = 0;
         goto resume_0;
     }
@@ -815,35 +847,21 @@ bool journal_flusher_co::clear_incomplete_csum_block_bits(int wait_base)
                 bs->verify_padded_checksums(new_clean_bitmap, new_clean_bitmap + 2*bs->dsk.clean_entry_bitmap_size,
                     v[i].offset, &iov, 1, [&](uint32_t bad_block, uint32_t calc_csum, uint32_t stored_csum)
                 {
-                    printf("Checksum mismatch in object %jx:%jx v%ju in data area at offset 0x%jx+0x%x: got %08x, expected %08x\n",
+                    printf("Checksum mismatch in object %jx:%jx v%ju in data area at offset 0x%jx+0x%x during flush: got %08x, expected %08x\n",
                         cur.oid.inode, cur.oid.stripe, old_clean_ver, old_clean_loc, bad_block, calc_csum, stored_csum);
-                    for (uint32_t j = 0; j < bs->dsk.csum_block_size; j += bs->dsk.bitmap_granularity)
-                    {
-                        // Simplest method of mangling: flip one byte in every sector
-                        ((uint8_t*)v[i].buf)[j+bad_block-v[i].offset] ^= 0xff;
-                    }
+                    assert(!(bad_block % bs->dsk.csum_block_size) && bad_block < bs->dsk.data_block_size);
+                    mangle_csum_blocks.insert(bad_block);
                 });
             }
             else
             {
                 bs->verify_journal_checksums(v[i].csum_buf, v[i].offset, &iov, 1, [&](uint32_t bad_block, uint32_t calc_csum, uint32_t stored_csum)
                 {
-                    printf("Checksum mismatch in object %jx:%jx v%ju in journal at offset 0x%jx+0x%x (block offset 0x%jx): got %08x, expected %08x\n",
+                    printf("Checksum mismatch in object %jx:%jx v%ju in journal at offset 0x%jx+0x%x (block offset 0x%jx) during flush: got %08x, expected %08x\n",
                         cur.oid.inode, cur.oid.stripe, old_clean_ver,
                         v[i].disk_offset, bad_block, v[i].offset, calc_csum, stored_csum);
-                    bad_block += (v[i].offset/bs->dsk.csum_block_size) * bs->dsk.csum_block_size;
-                    uint32_t bad_block_end = bad_block + bs->dsk.csum_block_size + (v[i].offset/bs->dsk.csum_block_size) * bs->dsk.csum_block_size;
-                    if (bad_block < v[i].offset)
-                        bad_block = v[i].offset;
-                    if (bad_block_end > v[i].offset+v[i].len)
-                        bad_block_end = v[i].offset+v[i].len;
-                    bad_block -= v[i].offset;
-                    bad_block_end -= v[i].offset;
-                    for (uint32_t j = bad_block; j < bad_block_end; j += bs->dsk.bitmap_granularity)
-                    {
-                        // Simplest method of mangling: flip one byte in every sector
-                        ((uint8_t*)v[i].buf)[j] ^= 0xff;
-                    }
+                    assert(!(bad_block % bs->dsk.csum_block_size) && bad_block < bs->dsk.data_block_size);
+                    mangle_csum_blocks.insert(bad_block);
                 });
             }
         }
@@ -952,6 +970,11 @@ void journal_flusher_co::calc_block_checksums(uint32_t *new_data_csums, bool ski
     }
     // `v` should contain aligned items, possibly split into pieces
     assert(!block_done);
+    for (uint32_t mangle_block: mangle_csum_blocks)
+    {
+        // Flip 1 bit
+        new_data_csums[mangle_block / bs->dsk.csum_block_size] ^= 1;
+    }
 }
 
 void journal_flusher_co::scan_dirty()
@@ -1088,7 +1111,8 @@ void journal_flusher_co::scan_dirty()
             last--;
         read_to_fill_incomplete = bs->fill_partial_checksum_blocks(
             v, fulfilled, bmp_ptr, NULL, false, NULL, v[0].offset/bs->dsk.csum_block_size * bs->dsk.csum_block_size,
-            ((v[last].offset+v[last].len-1) / bs->dsk.csum_block_size + 1) * bs->dsk.csum_block_size
+            ((v[last].offset+v[last].len-1) / bs->dsk.csum_block_size + 1) * bs->dsk.csum_block_size,
+            0, bs->dsk.data_block_size
         );
     }
     else if (fill_incomplete && clean_init_bitmap)
@@ -1118,6 +1142,7 @@ bool journal_flusher_co::read_dirty(int wait_base)
     if (wait_state == wait_base)        goto resume_0;
     else if (wait_state == wait_base+1) goto resume_1;
     wait_count = wait_journal_count = 0;
+    mangle_csum_blocks.clear();
     if (bs->journal.inmemory && !read_to_fill_incomplete)
     {
         // Happy path: nothing to read :)
@@ -1349,7 +1374,7 @@ bool journal_flusher_co::fsync_batch(bool fsync_meta, int wait_base)
         cur_sync->ready_count++;
         flusher->syncing_flushers++;
     resume_1:
-        if (!cur_sync->state)
+        if (cur_sync->state == 0)
         {
             if (flusher->syncing_flushers >= flusher->active_flushers || !flusher->flush_queue.size())
             {
@@ -1377,6 +1402,12 @@ bool journal_flusher_co::fsync_batch(bool fsync_meta, int wait_base)
                 return false;
             }
         }
+        else if (cur_sync->state == 1)
+        {
+            // Wait for fsync completion
+            wait_state = wait_base+1;
+            return false;
+        }
         flusher->syncing_flushers--;
         cur_sync->ready_count--;
         if (cur_sync->ready_count == 0)

src/blockstore/v1/flush.h

@@ -66,6 +66,7 @@ class journal_flusher_co
     uint64_t clean_bitmap_offset, clean_bitmap_len;
     uint8_t *clean_init_dyn_ptr;
     uint8_t *new_clean_bitmap;
+    std::unordered_set<uint32_t> mangle_csum_blocks;
 
     uint64_t new_trim_pos;
 
@@ -123,6 +124,7 @@ public:
     void loop();
     bool is_trim_wanted() { return trim_wanted; }
     bool is_active();
+    size_t get_queue_size();
     void mark_trim_possible();
     void request_trim();
     void release_trim();

src/blockstore/v1/impl.cpp

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

src/blockstore/v1/impl.h

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

src/blockstore/v1/init.cpp

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

src/blockstore/v1/init.h

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

src/blockstore/v1/read.cpp

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

src/blockstore/v1/stable.cpp

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

src/client/CMakeLists.txt

@@ -3,6 +3,27 @@ cmake_minimum_required(VERSION 2.8...3.30)
 project(vitastor)
 
 # libvitastor_common.a
+add_library(vitastor_common STATIC
+	etcd_state_client.cpp
+	msgr_stop.cpp
+	msgr_op.cpp
+	../../json11/json11.cpp
+	osd_ops.cpp
+	pg_states.cpp
+	msgr_encrypt.cpp
+	msgr_handshake.cpp
+	../util/allocator.cpp
+	../util/addr_util.cpp
+	../util/timerfd_manager.cpp
+	../util/str_util.cpp
+	../util/json_util.cpp
+	../util/xxh_x86dispatch.c
+	 ../util/openssl_util.cpp
+)
+target_compile_options(vitastor_common PUBLIC -fPIC)
+target_link_libraries(vitastor_common ${OPENSSL_LIBRARIES} ${ISAL_CRYPTO_LIBRARIES})
+
+# libvitastor_net.a
 set(MSGR_RDMA "")
 if (IBVERBS_LIBRARIES)
 	set(MSGR_RDMA "msgr_rdma.cpp")
@@ -11,25 +32,32 @@ set(MSGR_RDMACM "")
 if (RDMACM_LIBRARIES)
 	set(MSGR_RDMACM "msgr_rdmacm.cpp")
 endif (RDMACM_LIBRARIES)
-add_library(vitastor_common STATIC
-	../util/epoll_manager.cpp etcd_state_client.cpp messenger.cpp msgr_iothread.cpp ../util/addr_util.cpp ../util/xxh_x86dispatch.c ../util/openssl_util.cpp
-	msgr_encrypt.cpp msgr_handshake.cpp msgr_stop.cpp msgr_op.cpp msgr_send.cpp msgr_receive.cpp ../util/ringloop.cpp ../../json11/json11.cpp
-	http_client.cpp osd_ops.cpp pg_states.cpp ../util/timerfd_manager.cpp ../util/str_util.cpp ../util/json_util.cpp ${MSGR_RDMA} ${MSGR_RDMACM}
+add_library(vitastor_net STATIC
+	../util/epoll_manager.cpp
+	etcd_state_client_http.cpp
+	messenger.cpp
+	msgr_iothread.cpp
+	msgr_send.cpp
+	msgr_receive.cpp
+	msgr_encrypt.cpp
+	../util/ringloop.cpp
+	http_client.cpp
+	${MSGR_RDMA}
+	${MSGR_RDMACM}
 )
-target_link_libraries(vitastor_common pthread ${OPENSSL_LIBRARIES} ${CARES_LIBRARIES} ${ISAL_CRYPTO_LIBRARIES})
-target_compile_options(vitastor_common PUBLIC -fPIC)
+target_link_libraries(vitastor_net pthread vitastor_common ${CARES_LIBRARIES})
+target_compile_options(vitastor_net PUBLIC -fPIC)
 
-# libvitastor_client.so
-add_library(vitastor_client SHARED
+# libvitastor_client_int.a
+add_library(vitastor_client_int STATIC
 	cluster_client.cpp
+	cluster_client_real.cpp
 	cluster_client_list.cpp
 	cluster_client_wb.cpp
 	cluster_client_icache.cpp
-	vitastor_c.cpp
 )
-set_target_properties(vitastor_client PROPERTIES PUBLIC_HEADER "client/vitastor_c.h")
-target_link_libraries(vitastor_client
-	vitastor_common
+target_link_libraries(vitastor_client_int
+	vitastor_net
 	vitastor_cli
 	${LIBURING_LIBRARIES}
 	${IBVERBS_LIBRARIES}
@@ -37,6 +65,16 @@ target_link_libraries(vitastor_client
 	${OPENSSL_LIBRARIES}
 	${ISAL_CRYPTO_LIBRARIES}
 )
+target_compile_options(vitastor_client_int PUBLIC -fPIC)
+
+# libvitastor_client.so
+add_library(vitastor_client SHARED
+	vitastor_c.cpp
+)
+set_target_properties(vitastor_client PROPERTIES PUBLIC_HEADER "client/vitastor_c.h")
+target_link_libraries(vitastor_client
+	vitastor_client_int
+)
 set_target_properties(vitastor_client PROPERTIES VERSION ${VITASTOR_VERSION} SOVERSION 0)
 configure_file(vitastor.pc.in vitastor.pc @ONLY)
 
@@ -98,11 +136,15 @@ endif (${WITH_QEMU})
 add_executable(test_cluster_client
 	EXCLUDE_FROM_ALL
 	../test/test_cluster_client.cpp
-	pg_states.cpp osd_ops.cpp cluster_client.cpp cluster_client_list.cpp cluster_client_wb.cpp 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 ../util/openssl_util.cpp ../../json11/json11.cpp
+	cluster_client.cpp
+	cluster_client_list.cpp
+	cluster_client_wb.cpp
+	cluster_client_icache.cpp
+	../test/mock/messenger.cpp
+	../test/mock/vault.cpp
+	etcd_state_client_mock.cpp
 )
-target_link_libraries(test_cluster_client ${LIBURING_LIBRARIES} ${OPENSSL_LIBRARIES} ${ISAL_CRYPTO_LIBRARIES})
-target_compile_definitions(test_cluster_client PUBLIC -D__MOCK__)
+target_link_libraries(test_cluster_client vitastor_common ${LIBURING_LIBRARIES} ${OPENSSL_LIBRARIES} ${ISAL_CRYPTO_LIBRARIES})
 target_include_directories(test_cluster_client BEFORE PUBLIC ${CMAKE_SOURCE_DIR}/src/test/mock)
 add_dependencies(build_tests test_cluster_client)
 add_test(NAME test_cluster_client COMMAND test_cluster_client)

src/client/cluster_client.cpp

@@ -11,7 +11,7 @@
 #define TRY_SEND_CONNECTING 1
 #define TRY_SEND_OK 2
 
-cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json config)
+cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json config, std::unique_ptr<etcd_state_client_t> st_cli_ptr)
 {
     wb = new writeback_cache_t();
 
@@ -53,24 +53,24 @@ cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd
     };
     msgr.parse_config(config, true);
 
-    st_cli.tfd = tfd;
-    st_cli.on_load_config_hook = [this](json11::Json::object & cfg) { on_load_config_hook(cfg); };
-    st_cli.on_change_osd_state_hook = [this](uint64_t peer_osd) { on_change_osd_state_hook(peer_osd); };
-    st_cli.on_change_pool_config_hook = [this]() { on_change_pool_config_hook(); };
-    st_cli.on_change_pg_config_hook = [this]() { on_change_pool_config_hook(); };
-    st_cli.on_change_pg_state_hook = [this](pool_id_t pool_id, pg_num_t pg_num, osd_num_t prev_primary) { on_change_pg_state_hook(pool_id, pg_num, prev_primary); };
-    st_cli.on_change_node_placement_hook = [this]() { on_change_node_placement_hook(); };
-    st_cli.on_load_pgs_hook = [this](bool success) { on_load_pgs_hook(success); };
-    st_cli.on_reload_hook = [this]() { st_cli.load_global_config(); };
-    st_cli.on_inode_change_hook = [this](uint64_t inode, bool removed) { on_change_inode_hook(inode, removed); };
+    st_cli = std::move(st_cli_ptr);
+    st_cli->on_load_config_hook = [this](json11::Json::object & cfg) { on_load_config_hook(cfg); };
+    st_cli->on_change_osd_state_hook = [this](uint64_t peer_osd) { on_change_osd_state_hook(peer_osd); };
+    st_cli->on_change_pool_config_hook = [this]() { on_change_pool_config_hook(); };
+    st_cli->on_change_pg_config_hook = [this]() { on_change_pool_config_hook(); };
+    st_cli->on_change_pg_state_hook = [this](pool_id_t pool_id, pg_num_t pg_num, osd_num_t prev_primary) { on_change_pg_state_hook(pool_id, pg_num, prev_primary); };
+    st_cli->on_change_node_placement_hook = [this]() { on_change_node_placement_hook(); };
+    st_cli->on_load_pgs_hook = [this](bool success) { on_load_pgs_hook(success); };
+    st_cli->on_reload_hook = [this]() { this->st_cli->load_global_config(); };
+    st_cli->on_inode_change_hook = [this](uint64_t inode, bool removed) { on_change_inode_hook(inode, removed); };
 
-    st_cli.parse_config(config);
-    st_cli.infinite_start = false;
+    st_cli->parse_config(config);
+    st_cli->infinite_start = false;
     if (!config["client_infinite_start"].is_null())
     {
-        st_cli.infinite_start = config["client_infinite_start"].bool_value();
+        st_cli->infinite_start = config["client_infinite_start"].bool_value();
     }
-    st_cli.load_global_config();
+    st_cli->load_global_config();
 }
 
 cluster_client_t::~cluster_client_t()
@@ -467,7 +467,7 @@ void cluster_client_t::on_load_config_hook(json11::Json::object & etcd_global_co
         auto etcd_report_interval = config["etcd_report_interval"].uint64_value();
         if (!etcd_report_interval)
             etcd_report_interval = 5;
-        client_wait_up_timeout = 1+etcd_report_interval+(st_cli.max_etcd_attempts*(2*st_cli.etcd_quick_timeout)+999)/1000;
+        client_wait_up_timeout = 1+etcd_report_interval+(st_cli->max_etcd_attempts*(2*st_cli->etcd_quick_timeout)+999)/1000;
     }
     // log_level
     log_level = config["log_level"].uint64_value();
@@ -482,8 +482,8 @@ void cluster_client_t::on_load_config_hook(json11::Json::object & etcd_global_co
     // vault
     vault_parse_config();
     msgr.parse_config(config, false);
-    st_cli.parse_config(config);
-    st_cli.load_pgs();
+    st_cli->parse_config(config);
+    st_cli->load_pgs();
 }
 
 osd_num_t cluster_client_t::select_random_osd(const std::vector<osd_num_t> & osds)
@@ -492,7 +492,7 @@ osd_num_t cluster_client_t::select_random_osd(const std::vector<osd_num_t> & osd
     int alive_count = 0;
     for (auto & osd_num: osds)
     {
-        if (!st_cli.peer_states[osd_num].is_null())
+        if (!st_cli->peer_states[osd_num].is_null())
             alive_set[alive_count++] = osd_num;
     }
     if (!alive_count)
@@ -509,7 +509,7 @@ osd_num_t cluster_client_t::select_nearest_osd(const std::vector<osd_num_t> & os
         while (self_tree_metrics.find(cur_id) == self_tree_metrics.end())
         {
             self_tree_metrics[cur_id] = metric++;
-            json11::Json cur_placement = st_cli.node_placement[cur_id];
+            json11::Json cur_placement = st_cli->node_placement[cur_id];
             cur_id = cur_placement["parent"].string_value();
         }
         if (cur_id != "")
@@ -529,7 +529,7 @@ osd_num_t cluster_client_t::select_nearest_osd(const std::vector<osd_num_t> & os
         }
         else
         {
-            auto & peer_state = st_cli.peer_states[osd_num];
+            auto & peer_state = st_cli->peer_states[osd_num];
             if (!peer_state.is_null())
             {
                 metric = self_tree_metrics[""];
@@ -539,7 +539,7 @@ osd_num_t cluster_client_t::select_nearest_osd(const std::vector<osd_num_t> & os
                 while (seen.find(cur_id) == seen.end())
                 {
                     seen.insert(cur_id);
-                    json11::Json cur_placement = st_cli.node_placement[cur_id];
+                    json11::Json cur_placement = st_cli->node_placement[cur_id];
                     std::string cur_parent = cur_placement["parent"].string_value();
                     cur_id = (!first || cur_parent != "" ? cur_parent : peer_state["host"].string_value());
                     first = false;
@@ -564,7 +564,7 @@ osd_num_t cluster_client_t::select_nearest_osd(const std::vector<osd_num_t> & os
 
 void cluster_client_t::on_load_pgs_hook(bool success)
 {
-    for (auto & pool_item: st_cli.pool_config)
+    for (auto & pool_item: st_cli->pool_config)
     {
         pg_counts[pool_item.first] = pool_item.second.real_pg_count;
     }
@@ -584,7 +584,7 @@ void cluster_client_t::on_load_pgs_hook(bool success)
 
 void cluster_client_t::on_change_pool_config_hook()
 {
-    for (auto & pool_item: st_cli.pool_config)
+    for (auto & pool_item: st_cli->pool_config)
     {
         if (pg_counts[pool_item.first] != pool_item.second.real_pg_count)
         {
@@ -615,7 +615,7 @@ void cluster_client_t::on_change_pool_config_hook()
 
 void cluster_client_t::on_change_pg_state_hook(pool_id_t pool_id, pg_num_t pg_num, osd_num_t prev_primary)
 {
-    auto & pg_cfg = st_cli.pool_config[pool_id].pg_config[pg_num];
+    auto & pg_cfg = st_cli->pool_config[pool_id].pg_config[pg_num];
     if (pg_cfg.cur_primary != prev_primary)
     {
         // Repeat this PG operations because an OSD which stopped being primary may not fsync operations
@@ -633,8 +633,8 @@ bool cluster_client_t::get_immediate_commit(uint64_t inode)
     pool_id_t pool_id = INODE_POOL(inode);
     if (!pool_id)
         return true;
-    auto pool_it = st_cli.pool_config.find(pool_id);
-    if (pool_it == st_cli.pool_config.end())
+    auto pool_it = st_cli->pool_config.find(pool_id);
+    if (pool_it == st_cli->pool_config.end())
         return true;
     return pool_it->second.immediate_commit == IMMEDIATE_ALL;
 }
@@ -644,7 +644,7 @@ void cluster_client_t::on_change_osd_state_hook(uint64_t peer_osd)
     osd_tree_metrics.erase(peer_osd);
     if (msgr.wanted_peers.find(peer_osd) != msgr.wanted_peers.end())
     {
-        msgr.connect_peer(peer_osd, st_cli.peer_states[peer_osd]);
+        msgr.connect_peer(peer_osd, st_cli->peer_states[peer_osd]);
         continue_lists();
     }
 }
@@ -943,8 +943,8 @@ bool cluster_client_t::check_rw(cluster_op_t *op)
         cb(op);
         return false;
     }
-    auto pool_it = st_cli.pool_config.find(pool_id);
-    if (pool_it == st_cli.pool_config.end() || pool_it->second.real_pg_count == 0)
+    auto pool_it = st_cli->pool_config.find(pool_id);
+    if (pool_it == st_cli->pool_config.end() || pool_it->second.real_pg_count == 0)
     {
         // Pools are loaded, but this one is unknown
         op->retval = -EINVAL;
@@ -1057,7 +1057,7 @@ void cluster_client_t::execute_raw(osd_num_t osd_num, osd_op_t *op)
     else
     {
         if (msgr.wanted_peers.find(osd_num) == msgr.wanted_peers.end())
-            msgr.connect_peer(osd_num, st_cli.peer_states[osd_num]);
+            msgr.connect_peer(osd_num, st_cli->peer_states[osd_num]);
         raw_ops.emplace(osd_num, op);
     }
 }
@@ -1206,7 +1206,7 @@ resume_2:
         op->retval = op->len;
         if (op->opcode == OSD_OP_READ_BITMAP || op->opcode == OSD_OP_READ_CHAIN_BITMAP)
         {
-            auto & pool_cfg = st_cli.pool_config.at(INODE_POOL(op->inode));
+            auto & pool_cfg = st_cli->pool_config.at(INODE_POOL(op->inode));
             op->retval = op->len / pool_cfg.bitmap_granularity;
         }
         if (op->flush_id)
@@ -1286,7 +1286,7 @@ void cluster_client_t::slice_rw(cluster_op_t *op)
 {
     // Slice the request into individual object stripe requests
     // Primary OSDs still operate individual stripes, but their size is multiplied by PG minsize in case of EC
-    auto & pool_cfg = st_cli.pool_config.at(INODE_POOL(op->cur_inode));
+    auto & pool_cfg = st_cli->pool_config.at(INODE_POOL(op->cur_inode));
     uint32_t pg_data_size = (pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks);
     uint64_t pg_block_size = pool_cfg.data_block_size * pg_data_size;
     uint64_t first_stripe = (op->offset / pg_block_size) * pg_block_size;
@@ -1389,7 +1389,7 @@ bool cluster_client_t::affects_pg(uint64_t inode, uint64_t offset, uint64_t len,
     {
         return false;
     }
-    auto & pool_cfg = st_cli.pool_config.at(INODE_POOL(inode));
+    auto & pool_cfg = st_cli->pool_config.at(INODE_POOL(inode));
     uint32_t pg_data_size = (pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks);
     uint64_t pg_block_size = pool_cfg.data_block_size * pg_data_size;
     uint64_t first_stripe = (offset / pg_block_size) * pg_block_size;
@@ -1408,7 +1408,7 @@ bool cluster_client_t::affects_pg(uint64_t inode, uint64_t offset, uint64_t len,
 
 bool cluster_client_t::affects_osd(uint64_t inode, uint64_t offset, uint64_t len, osd_num_t osd)
 {
-    auto & pool_cfg = st_cli.pool_config.at(INODE_POOL(inode));
+    auto & pool_cfg = st_cli->pool_config.at(INODE_POOL(inode));
     uint32_t pg_data_size = (pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks);
     uint64_t pg_block_size = pool_cfg.data_block_size * pg_data_size;
     uint64_t first_stripe = (offset / pg_block_size) * pg_block_size;
@@ -1432,7 +1432,7 @@ int cluster_client_t::try_send(cluster_op_t *op, int i, std::function<void(osd_o
         init_msgr();
     }
     auto part = &op->parts[i];
-    auto & pool_cfg = st_cli.pool_config.at(INODE_POOL(op->cur_inode));
+    auto & pool_cfg = st_cli->pool_config.at(INODE_POOL(op->cur_inode));
     auto pg_it = pool_cfg.pg_config.find(part->pg_num);
     if (pg_it != pool_cfg.pg_config.end() &&
         !pg_it->second.pause && pg_it->second.cur_primary &&
@@ -1466,8 +1466,8 @@ int cluster_client_t::try_send(cluster_op_t *op, int i, std::function<void(osd_o
             uint64_t meta_rev = 0;
             if (op->opcode != OSD_OP_READ_BITMAP && op->opcode != OSD_OP_DELETE && !op->deoptimise_snapshot)
             {
-                auto ino_it = st_cli.inode_config.find(op->cur_inode);
-                if (ino_it != st_cli.inode_config.end())
+                auto ino_it = st_cli->inode_config.find(op->cur_inode);
+                if (ino_it != st_cli->inode_config.end())
                     meta_rev = ino_it->second.mod_revision;
             }
             part->op = (osd_op_t){
@@ -1501,7 +1501,7 @@ int cluster_client_t::try_send(cluster_op_t *op, int i, std::function<void(osd_o
         }
         else if (msgr.wanted_peers.find(primary_osd) == msgr.wanted_peers.end())
         {
-            msgr.connect_peer(primary_osd, st_cli.peer_states[primary_osd]);
+            msgr.connect_peer(primary_osd, st_cli->peer_states[primary_osd]);
             return TRY_SEND_CONNECTING;
         }
     }
@@ -1694,7 +1694,7 @@ void cluster_client_t::handle_op_part(cluster_op_part_t *part)
 void cluster_client_t::copy_part_bitmap(cluster_op_t *op, cluster_op_part_t *part)
 {
     // Copy (OR) bitmap
-    auto & pool_cfg = st_cli.pool_config.at(INODE_POOL(op->cur_inode));
+    auto & pool_cfg = st_cli->pool_config.at(INODE_POOL(op->cur_inode));
     uint32_t pg_block_size = pool_cfg.data_block_size * (
         pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks
     );

src/client/cluster_client.h

@@ -4,7 +4,7 @@
 #pragma once
 
 #include "messenger.h"
-#include "etcd_state_client.h"
+#include "etcd_state_client_http.h"
 #include "../util/robin_hood.h"
 
 #define DEFAULT_CLIENT_MAX_DIRTY_BYTES 32*1024*1024
@@ -176,7 +176,7 @@ class __attribute__((visibility("default"))) cluster_client_t
     bool msgr_initialized = false;
 
 public:
-    etcd_state_client_t st_cli;
+    std::unique_ptr<etcd_state_client_t> st_cli;
 
     osd_messenger_t msgr;
     void init_msgr();
@@ -184,7 +184,8 @@ public:
     json11::Json::object cli_config, file_config, etcd_global_config;
     json11::Json::object config;
 
-    cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json config);
+    static cluster_client_t* create(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json config);
+    cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd, json11::Json config, std::unique_ptr<etcd_state_client_t> st_cli);
     ~cluster_client_t();
     void execute(cluster_op_t *op);
     void execute_raw(osd_num_t osd_num, osd_op_t *op);

src/client/cluster_client_icache.cpp

@@ -4,14 +4,8 @@
 #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)
@@ -22,24 +16,15 @@ inode_cache_t::~inode_cache_t()
     }
 }
 
-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();
+    if (vault_client_cert.empty())
+        vault_client_cert = config["cert"].string_value();
     vault_client_key = config["vault_client_key"].string_value();
+    if (vault_client_key.empty())
+        vault_client_key = config["pkey"].string_value();
     vault_ca = config["vault_ca"].string_value();
     vault_secret_api_path = "/v1/secret/";
     if (config["vault_secret_api_path"].is_string())
@@ -91,14 +76,14 @@ std::shared_ptr<inode_cache_t> cluster_client_t::inode_cache_get(inode_t ino)
         return icache_it->second;
     }
     // Fill inode cache
-    auto ino_it = st_cli.inode_config.find(ino);
-    if (ino_it == st_cli.inode_config.end())
+    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())
+    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;
@@ -125,11 +110,11 @@ std::shared_ptr<inode_cache_t> cluster_client_t::inode_cache_get(inode_t ino)
                 break;
             }
             seen.insert(parent_id);
-            ino_it = st_cli.inode_config.find(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())
+                if (ino_it == st_cli->inode_config.end())
                     chain_cfg.push_back(NULL);
                 else
                 {
@@ -140,7 +125,7 @@ std::shared_ptr<inode_cache_t> cluster_client_t::inode_cache_get(inode_t ino)
             }
             else if (!icache->other_pool_parent_id)
                 icache->other_pool_parent_id = parent_id;
-            if (ino_it == st_cli.inode_config.end())
+            if (ino_it == st_cli->inode_config.end())
                 break;
             parent_id = ino_it->second.parent_id;
         }
@@ -223,113 +208,6 @@ std::shared_ptr<inode_cache_t> cluster_client_t::inode_cache_get(inode_t 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;

src/client/cluster_client_impl.h

@@ -17,6 +17,11 @@
 #define OP_FLUSH_BUFFER 0x02
 #define OP_IMMEDIATE_COMMIT 0x04
 
+#define VAULT_KEY_NOT_LOADED 0
+#define VAULT_KEY_LOADING 1
+#define VAULT_KEY_LOADED 2
+#define VAULT_KEY_ERROR 3
+
 struct cluster_buffer_t
 {
     uint8_t *buf;

src/client/cluster_client_list.cpp

@@ -63,14 +63,14 @@ void cluster_client_t::list_inode(inode_t inode, uint64_t min_offset, uint64_t m
 {
     init_msgr();
     pool_id_t pool_id = INODE_POOL(inode);
-    if (!pool_id || st_cli.pool_config.find(pool_id) == st_cli.pool_config.end())
+    if (!pool_id || st_cli->pool_config.find(pool_id) == st_cli->pool_config.end())
     {
         if (log_level > 0)
             fprintf(stderr, "Pool %u does not exist\n", pool_id);
         pg_callback(-EINVAL, 0, 0, std::set<object_id>());
         return;
     }
-    auto pg_stripe_size = st_cli.pool_config.at(pool_id).pg_stripe_size;
+    auto pg_stripe_size = st_cli->pool_config.at(pool_id).pg_stripe_size;
     if (min_offset)
         min_offset = (min_offset/pg_stripe_size) * pg_stripe_size;
     inode_list_t *lst = new inode_list_t();
@@ -110,13 +110,13 @@ bool cluster_client_t::continue_listing(inode_list_t *lst)
 
 bool cluster_client_t::restart_listing(inode_list_t* lst)
 {
-    auto pool_it = st_cli.pool_config.find(lst->pool_id);
+    auto pool_it = st_cli->pool_config.find(lst->pool_id);
     // We want listing to be consistent. To achieve it we should:
     // 1) retry listing of each PG if its state changes
     // 2) abort listing if PG count changes during listing
     // 3) ideally, only talk to the primary OSD - this will be done separately
     // So first we add all PGs without checking their state
-    if (pool_it == st_cli.pool_config.end() ||
+    if (pool_it == st_cli->pool_config.end() ||
         lst->real_pg_count != pool_it->second.real_pg_count)
     {
         for (auto pg: lst->pgs)
@@ -136,7 +136,7 @@ bool cluster_client_t::restart_listing(inode_list_t* lst)
             fprintf(stderr, "PG count in pool %u changed during listing\n", lst->pool_id);
         }
         lst->pgs.clear();
-        if (pool_it == st_cli.pool_config.end())
+        if (pool_it == st_cli->pool_config.end())
         {
             // Unknown pool
             lst->callback(-EINVAL, 0, 0, std::set<object_id>());
@@ -248,7 +248,7 @@ void cluster_client_t::set_list_retry_timeout(int ms, timespec new_time)
 
 int cluster_client_t::start_pg_listing(inode_list_pg_t *pg)
 {
-    auto & pool_cfg = st_cli.pool_config.at(pg->lst->pool_id);
+    auto & pool_cfg = st_cli->pool_config.at(pg->lst->pool_id);
     auto pg_it = pool_cfg.pg_config.find(pg->pg_num);
     assert(pg->lst->real_pg_count == pool_cfg.real_pg_count);
     if (pg_it == pool_cfg.pg_config.end() ||
@@ -277,7 +277,7 @@ int cluster_client_t::start_pg_listing(inode_list_pg_t *pg)
         for (auto peer_it = all_peers.begin(); peer_it != all_peers.end(); )
         {
             if (*peer_it != pg_it->second.cur_primary &&
-                st_cli.peer_states[*peer_it].is_null())
+                st_cli->peer_states[*peer_it].is_null())
             {
                 pg->inactive_osds.push_back(*peer_it);
                 all_peers.erase(peer_it++);
@@ -298,11 +298,11 @@ int cluster_client_t::start_pg_listing(inode_list_pg_t *pg)
         if (msgr.osd_peers.find(peer_osd) == msgr.osd_peers.end())
         {
             // Initiate connection
-            if (st_cli.peer_states[peer_osd].is_null())
+            if (st_cli->peer_states[peer_osd].is_null())
             {
                 return LIST_PG_WAIT_ACTIVE;
             }
-            msgr.connect_peer(peer_osd, st_cli.peer_states[peer_osd]);
+            msgr.connect_peer(peer_osd, st_cli->peer_states[peer_osd]);
             conn = false;
         }
     }
@@ -336,7 +336,7 @@ void cluster_client_t::send_list(inode_list_osd_t *cur_list)
     if (!cur_list->pg->inflight_ops)
         cur_list->pg->lst->inflight_pgs++;
     cur_list->pg->inflight_ops++;
-    auto & pool_cfg = st_cli.pool_config[cur_list->pg->lst->pool_id];
+    auto & pool_cfg = st_cli->pool_config[cur_list->pg->lst->pool_id];
     osd_op_t *op = new osd_op_t();
     op->op_type = OSD_OP_OUT;
     // Already checked that it exists above, but anyway

src/client/cluster_client_real.cpp

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

src/client/cluster_client_wb.cpp

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

src/client/etcd_state_client.cpp

@@ -1,16 +1,12 @@
 // 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__
 #include "addr_util.h"
-#include "http_client.h"
-#endif
 #include "str_util.h"
 #include "json_util.h"
 
@@ -21,33 +17,8 @@ etcd_state_client_t::~etcd_state_client_t()
         delete watch;
     }
     watches.clear();
-    etcd_watches_initialised = -1;
-#ifndef __MOCK__
-    stop_ws_keepalive();
-    if (etcd_watch_ws)
-    {
-        http_destroy(etcd_watch_ws);
-        etcd_watch_ws = NULL;
-    }
-    if (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)
-    {
-        tfd->clear_timer(load_pgs_timer_id);
-        load_pgs_timer_id = -1;
-    }
 }
 
-#ifndef __MOCK__
 etcd_kv_t etcd_state_client_t::parse_etcd_kv(const json11::Json & kv_json)
 {
     etcd_kv_t kv;
@@ -121,99 +92,6 @@ bool etcd_state_client_t::check_image_perm(const std::shared_ptr<user_info_t> &
     return write ? (perm_item.perm == user_perm_t::OWNER) : (perm_item.perm != user_perm_t::DENY);
 }
 
-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)
-{
-    auto http_cli = http_init(get_http_ctx());
-    http_json_post(http_cli, etcd_url+api, payload, "", { .timeout = timeout }, [http_cli, callback](http_message_t *response)
-    {
-        std::string err;
-        json11::Json data;
-        response->parse_json_response(err, data);
-        callback(err, data);
-        http_destroy(http_cli);
-    });
-}
-
-void etcd_state_client_t::etcd_call(const std::string & api, json11::Json payload, int timeout,
-    int retries, int interval, std::function<void(std::string, json11::Json)> callback)
-{
-    pick_next_etcd([=]()
-    {
-        etcd_call_selected(api, payload, timeout, retries, interval, callback);
-    });
-}
-
-void etcd_state_client_t::etcd_call_selected(const std::string & api, json11::Json payload, int timeout,
-    int retries, int interval, std::function<void(std::string, json11::Json)> callback)
-{
-    const auto & url = selected_etcd_url;
-    std::string req = payload.dump();
-    req = "POST "+url.path+api+" HTTP/1.1\r\n"
-        "Host: "+url.hostname+"\r\n"
-        "Content-Type: application/json\r\n"
-        "Content-Length: "+std::to_string(req.size())+"\r\n"
-        "Connection: keep-alive\r\n"
-        "Keep-Alive: timeout="+std::to_string(etcd_keepalive_timeout)+"\r\n"
-        "\r\n"+req;
-    retries--;
-    auto cb = [this, api, payload, timeout, retries, interval, callback,
-        cur_addr = url.addr](http_message_t *response)
-    {
-        std::string err;
-        json11::Json data;
-        response->parse_json_response(err, data);
-        if (err != "")
-        {
-            if (cur_addr == selected_etcd_url.addr)
-                selected_etcd_url = (http_url_t){};
-            if (retries > 0)
-            {
-                if (this->log_level > 0)
-                {
-                    fprintf(
-                        stderr, "Warning: etcd request failed: %s, retrying %d more times\n",
-                        err.c_str(), retries
-                    );
-                }
-                if (interval > 0)
-                {
-                    // FIXME: Prevent destruction of etcd_state_client if timers or requests are active
-                    tfd->set_timer(interval, false, [this, api, payload, timeout, retries, interval, callback](int)
-                    {
-                        etcd_call(api, payload, timeout, retries, interval, callback);
-                    });
-                }
-                else
-                    etcd_call(api, payload, timeout, retries, interval, callback);
-            }
-            else
-                callback(err, data);
-        }
-        else
-            callback(err, data);
-    };
-    if (!keepalive_client)
-        keepalive_client = http_init(get_http_ctx());
-    http_request(keepalive_client, url.addr, req, { .timeout = timeout, .keepalive = true, .ssl = url.ssl }, cb);
-}
-
-
 void etcd_state_client_t::add_etcd_url(std::string etcd_address)
 {
     if (etcd_address.size() > 0)
@@ -324,7 +202,6 @@ void etcd_state_client_t::parse_config(const json11::Json & config)
         if (this->etcd_keepalive_timeout < 30)
             this->etcd_keepalive_timeout = 30;
     }
-    auto old_etcd_ws_keepalive_interval = this->etcd_ws_keepalive_interval;
     this->etcd_ws_keepalive_interval = config["etcd_ws_keepalive_interval"].uint64_value();
     if (this->etcd_ws_keepalive_interval <= 0)
     {
@@ -350,347 +227,9 @@ void etcd_state_client_t::parse_config(const json11::Json & config)
     {
         this->etcd_min_reload_interval = 50;
     }
-    if (this->etcd_ws_keepalive_interval != old_etcd_ws_keepalive_interval && ws_keepalive_timer >= 0)
-    {
-#ifndef __MOCK__
-        stop_ws_keepalive();
-        start_ws_keepalive();
-#endif
-    }
 }
 
-void etcd_state_client_t::pick_next_etcd(std::function<void()> cb)
-{
-    if (!etcd_addresses.size() && !etcd_local.size())
-    {
-        fprintf(stderr, "etcd_address is missing in Vitastor configuration\n");
-        exit(1);
-    }
-    if (selected_etcd_url.addr != "")
-    {
-        cb();
-        return;
-    }
-    if (etcd_urls_to_try.size() != 0)
-    {
-        selected_etcd_url = std::move(etcd_urls_to_try[0]);
-        etcd_urls_to_try.erase(etcd_urls_to_try.begin());
-        cb();
-        return;
-    }
-    on_resolve_queue.push_back(std::move(cb));
-    if (on_resolve_queue.size() > 1)
-    {
-        // Already resolving
-        return;
-    }
-    assert(!resolve_count);
-    local_to_try = 0;
-    for (auto & url: etcd_local_addr_urls)
-    {
-        // Prefer local IPs, if any
-        etcd_urls_to_try.push_back(url);
-        local_to_try++;
-    }
-    for (auto & url: etcd_nonlocal_addr_urls)
-    {
-        etcd_urls_to_try.push_back(url);
-    }
-    resolve_count++;
-    for (auto & url: etcd_name_urls)
-    {
-        resolve_count++;
-        http_resolve(get_http_ctx(), url.ssl, url.addr, [this, url](const std::string & error, const std::vector<std::string>& addresses)
-        {
-            if (error != "")
-                fprintf(stderr, "Error resolving %s: %s\n", url.addr.c_str(), error.c_str());
-            for (auto & addr: addresses)
-            {
-                auto url_copy = url;
-                url_copy.addr = addr;
-                if (local_ips.find(addr) != local_ips.end())
-                {
-                    etcd_urls_to_try.insert(etcd_urls_to_try.begin(), std::move(url_copy));
-                    local_to_try++;
-                }
-                else
-                    etcd_urls_to_try.push_back(std::move(url_copy));
-            }
-            resolve_count--;
-            if (!resolve_count)
-                pick_next_etcd_on_resolve();
-        });
-    }
-    resolve_count--;
-    if (!resolve_count)
-    {
-        pick_next_etcd_on_resolve();
-    }
-}
-
-void etcd_state_client_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 (this->log_level > 1)
-    {
-        fprintf(stderr, "Trying to connect to etcd websocket at %s%s%s (hostname %s), watch from revision %ju/%ju/%ju\n",
-            url.ssl ? "https://" : "http://", url.addr.c_str(), url.path.c_str(), url.hostname.c_str(),
-            etcd_watch_revision_config, etcd_watch_revision_osd, etcd_watch_revision_pg);
-    }
-    if (!etcd_watch_ws)
-        etcd_watch_ws = http_init(get_http_ctx());
-    else
-        http_close(etcd_watch_ws);
-    open_websocket(etcd_watch_ws, url.addr, url.hostname, url.path+"/watch", { .timeout = etcd_slow_timeout, .ssl = url.ssl },
-        [this, cur_addr = url.addr](http_message_t *msg)
-    {
-        if (msg->body.length())
-        {
-            ws_alive = 1;
-            std::string json_err;
-            json11::Json data = json11::Json::parse(msg->body, json_err);
-            if (json_err != "")
-            {
-                fprintf(stderr, "Bad JSON in etcd event: %s, ignoring event\n", json_err.c_str());
-            }
-            else
-            {
-                uint64_t watch_id = data["result"]["watch_id"].uint64_value();
-                if (data["result"]["created"].bool_value())
-                {
-                    if (watch_id == ETCD_CONFIG_WATCH_ID ||
-                        watch_id == ETCD_PG_STATE_WATCH_ID ||
-                        watch_id == ETCD_OSD_STATE_WATCH_ID)
-                    {
-                        etcd_watches_initialised++;
-                    }
-                    if (etcd_watches_initialised == ETCD_TOTAL_WATCHES && this->log_level > 0)
-                    {
-                        fprintf(stderr, "Successfully subscribed to etcd at %s, revision %ju/%ju/%ju\n", cur_addr.c_str(),
-                            etcd_watch_revision_config, etcd_watch_revision_osd, etcd_watch_revision_pg);
-                    }
-                }
-                if (data["result"]["canceled"].bool_value())
-                {
-                    // etcd watch canceled, maybe because the revision was compacted
-                    if (data["result"]["compact_revision"].uint64_value())
-                    {
-                        // we may miss events if we proceed
-                        // so we should restart from the beginning if we can
-                        if (on_reload_hook != NULL)
-                        {
-                            // check to not trigger on_reload_hook multiple times
-                            if (etcd_watch_ws != NULL)
-                            {
-                                fprintf(stderr, "Revisions before %ju were compacted by etcd, reloading state\n",
-                                    data["result"]["compact_revision"].uint64_value());
-                                http_close(etcd_watch_ws);
-                                etcd_watch_revision_config = etcd_watch_revision_osd = etcd_watch_revision_pg = 0;
-                                on_reload_hook();
-                            }
-                            return;
-                        }
-                        else
-                        {
-                            fprintf(stderr, "Revisions before %ju were compacted by etcd, exiting\n",
-                                data["result"]["compact_revision"].uint64_value());
-                            exit(1);
-                        }
-                    }
-                    else
-                    {
-                        fprintf(stderr, "Watch canceled by etcd, reason: %s, exiting\n", data["result"]["cancel_reason"].string_value().c_str());
-                        exit(1);
-                    }
-                }
-                // Save revision only if it's present in the message - because sometimes etcd sends something without a header, like:
-                // {"error": {"grpc_code": 14, "http_code": 503, "http_status": "Service Unavailable", "message": "error reading from server: EOF"}}
-                // Also don't save revision from the initial created: true messages because they always contain the latest revision
-                if (etcd_watches_initialised == ETCD_TOTAL_WATCHES &&
-                    !data["result"]["header"]["revision"].is_null() &&
-                    !data["result"]["created"].bool_value())
-                {
-                    // Restart watchers from the same revision number as in the last received message,
-                    // not from the next one to protect against revision being split into multiple messages,
-                    // even though etcd guarantees not to do that **within a single watcher** without fragment=true:
-                    // https://etcd.io/docs/v3.5/learning/api_guarantees/#watch-apis
-                    // Revision contents are ALWAYS split into separate messages for different watchers though!
-                    // So generally we have to resume each watcher from its own revision...
-                    // Progress messages may have watch_id=-1 if sent on behalf of multiple watchers though.
-                    // And antietcd has an advanced semantic which merges the same revision for all watchers
-                    // into one message and just omits watch_id.
-                    // So we also have to handle the case where watch_id is -1 or not present (0).
-                    auto watch_rev = data["result"]["header"]["revision"].uint64_value();
-                    if (!watch_id || watch_id == UINT64_MAX)
-                        etcd_watch_revision_config = etcd_watch_revision_osd = etcd_watch_revision_pg = watch_rev;
-                    else if (watch_id == ETCD_CONFIG_WATCH_ID)
-                        etcd_watch_revision_config = watch_rev;
-                    else if (watch_id == ETCD_PG_STATE_WATCH_ID)
-                        etcd_watch_revision_pg = watch_rev;
-                    else if (watch_id == ETCD_OSD_STATE_WATCH_ID)
-                        etcd_watch_revision_osd = watch_rev;
-                    etcd_urls_to_try.clear();
-                }
-                // First gather all changes into a hash to remove multiple overwrites
-                std::map<std::string, etcd_kv_t> changes;
-                for (auto & ev: data["result"]["events"].array_items())
-                {
-                    auto kv = parse_etcd_kv(ev["kv"]);
-                    if (kv.key != "")
-                    {
-                        changes[kv.key] = kv;
-                    }
-                }
-                for (auto & kv: changes)
-                {
-                    if (this->log_level > 3)
-                    {
-                        fprintf(stderr, "Incoming event: %s -> %s\n", kv.first.c_str(), kv.second.value.dump().c_str());
-                    }
-                    parse_state(kv.second);
-                }
-                // React to changes
-                if (on_change_hook != NULL)
-                {
-                    on_change_hook(changes);
-                }
-            }
-        }
-        if (msg->eof)
-        {
-            fprintf(stderr, "Disconnected from etcd %s\n", cur_addr.c_str());
-            if (cur_addr == selected_etcd_url.addr)
-                selected_etcd_url = (http_url_t){};
-            if (etcd_watches_initialised == 0)
-            {
-                // Connection not established, retry in <etcd_quick_timeout>
-                tfd->set_timer(etcd_quick_timeout, false, [this](int)
-                {
-                    start_etcd_watcher();
-                });
-            }
-            else if (etcd_watches_initialised > 0)
-            {
-                // Connection was live, retry immediately
-                etcd_watches_initialised = 0;
-                start_etcd_watcher();
-            }
-        }
-    });
-    http_post_message(etcd_watch_ws, WS_TEXT, json11::Json(json11::Json::object {
-        { "create_request", json11::Json::object {
-            { "key", base64_encode(etcd_prefix+"/config/") },
-            { "range_end", base64_encode(etcd_prefix+"/config0") },
-            { "start_revision", etcd_watch_revision_config },
-            { "watch_id", ETCD_CONFIG_WATCH_ID },
-            { "progress_notify", true },
-        } }
-    }).dump());
-    http_post_message(etcd_watch_ws, WS_TEXT, json11::Json(json11::Json::object {
-        { "create_request", json11::Json::object {
-            { "key", base64_encode(etcd_prefix+"/osd/state/") },
-            { "range_end", base64_encode(etcd_prefix+"/osd/state0") },
-            { "start_revision", etcd_watch_revision_osd },
-            { "watch_id", ETCD_OSD_STATE_WATCH_ID },
-            { "progress_notify", true },
-        } }
-    }).dump());
-    http_post_message(etcd_watch_ws, WS_TEXT, json11::Json(json11::Json::object {
-        { "create_request", json11::Json::object {
-            { "key", base64_encode(etcd_prefix+"/pg/") },
-            { "range_end", base64_encode(etcd_prefix+"/pg0") },
-            { "start_revision", etcd_watch_revision_pg },
-            { "watch_id", ETCD_PG_STATE_WATCH_ID },
-            { "progress_notify", true },
-        } }
-    }).dump());
-    // FIXME: Do not watch /pg/history/ at all in client code (not in OSD)
-    if (on_start_watcher_hook)
-    {
-        on_start_watcher_hook(etcd_watch_ws);
-    }
-    start_ws_keepalive();
-}
-
-void etcd_state_client_t::stop_ws_keepalive()
-{
-    if (ws_keepalive_timer >= 0)
-    {
-        tfd->clear_timer(ws_keepalive_timer);
-        ws_keepalive_timer = -1;
-    }
-}
-
-void etcd_state_client_t::start_ws_keepalive()
-{
-    if (ws_keepalive_timer < 0)
-    {
-        ws_keepalive_timer = tfd->set_timer(etcd_ws_keepalive_interval*1000, true, [this](int)
-        {
-            if (!etcd_watch_ws || etcd_watches_initialised < ETCD_TOTAL_WATCHES)
-            {
-                // Do nothing
-            }
-            else if (!ws_alive)
-            {
-                if (this->log_level > 0)
-                {
-                    fprintf(stderr, "Websocket ping failed, disconnecting from etcd %s\n", selected_etcd_url.addr.c_str());
-                }
-                start_etcd_watcher();
-            }
-            else
-            {
-                ws_alive = 0;
-                http_post_message(etcd_watch_ws, WS_TEXT, json11::Json(json11::Json::object {
-                    { "progress_request", json11::Json::object { } }
-                }).dump());
-            }
-        });
-    }
-}
-
-void etcd_state_client_t::load_global_config()
+void etcd_state_client_t::load_global_config(std::function<void(const std::string & error)> cb)
 {
     json11::Json::object req = { { "success", json11::Json::array {
         json11::Json::object {
@@ -704,22 +243,12 @@ void etcd_state_client_t::load_global_config()
             } }
         },
     } } };
-    etcd_txn(req, etcd_quick_timeout, max_etcd_attempts, 0, [this](std::string err, json11::Json data)
+    etcd_txn(req, etcd_quick_timeout, max_etcd_attempts, 0, [this, cb](std::string err, json11::Json data)
     {
         if (err != "")
         {
             fprintf(stderr, "Error reading configuration from etcd: %s\n", err.c_str());
-            if (infinite_start)
-            {
-                tfd->set_timer(etcd_slow_timeout, false, [this](int timer_id)
-                {
-                    load_global_config();
-                });
-            }
-            else
-            {
-                exit(1);
-            }
+            cb(err);
             return;
         }
         json11::Json config_kv = data["responses"][0]["response_range"]["kvs"][0];
@@ -750,28 +279,12 @@ void etcd_state_client_t::load_global_config()
             parse_state(kv);
         }
         on_load_config_hook(global_config);
+        cb("");
     });
 }
 
-void etcd_state_client_t::load_pgs()
+void etcd_state_client_t::load_pgs(std::function<void(const std::string &)> cb)
 {
-    timespec tv;
-    clock_gettime(CLOCK_REALTIME, &tv);
-    uint64_t ms_passed = (tv.tv_sec-etcd_last_reload.tv_sec)*1000 + (tv.tv_nsec-etcd_last_reload.tv_nsec)/1000000;
-    if (ms_passed < etcd_min_reload_interval)
-    {
-        if (load_pgs_timer_id < 0)
-        {
-            load_pgs_timer_id = tfd->set_timer(etcd_min_reload_interval+50-ms_passed, false, [this](int) { load_pgs(); });
-        }
-        return;
-    }
-    etcd_last_reload = tv;
-    if (load_pgs_timer_id >= 0)
-    {
-        tfd->clear_timer(load_pgs_timer_id);
-        load_pgs_timer_id = -1;
-    }
     json11::Json::array txn = {
         json11::Json::object {
             { "request_range", json11::Json::object {
@@ -809,16 +322,13 @@ void etcd_state_client_t::load_pgs()
     {
         req["compare"] = checks;
     }
-    etcd_txn_slow(req, [this](std::string err, json11::Json data)
+    etcd_txn_slow(req, [this, cb](std::string err, json11::Json data)
     {
         if (err != "")
         {
             // Retry indefinitely
             fprintf(stderr, "Error loading PGs from etcd: %s\n", err.c_str());
-            tfd->set_timer(etcd_slow_timeout, false, [this](int timer_id)
-            {
-                load_pgs();
-            });
+            cb(err);
             return;
         }
         if (!data["succeeded"].bool_value())
@@ -846,24 +356,9 @@ void etcd_state_client_t::load_pgs()
         }
         clean_nonexistent_pgs();
         on_load_pgs_hook(true);
-        start_etcd_watcher();
+        cb("");
     });
 }
-#else
-void etcd_state_client_t::parse_config(const json11::Json & config)
-{
-}
-
-void etcd_state_client_t::load_global_config()
-{
-    json11::Json::object global_config;
-    on_load_config_hook(global_config);
-}
-
-void etcd_state_client_t::load_pgs()
-{
-}
-#endif
 
 void etcd_state_client_t::reset_pg_exists()
 {
@@ -976,7 +471,8 @@ void etcd_state_client_t::parse_state(const etcd_kv_t & kv)
             if (pc.pg_size < 1 ||
                 pool_item.second["pg_size"].uint64_value() < 3 &&
                 (pc.scheme == POOL_SCHEME_XOR || pc.scheme == POOL_SCHEME_EC) ||
-                pool_item.second["pg_size"].uint64_value() > 256)
+                // limit is 64 because osd_peering_pg.cpp uses a 64-bit mask for has_roles
+                pool_item.second["pg_size"].uint64_value() > 64)
             {
                 fprintf(stderr, "Pool %u has invalid pg_size, skipping pool\n", pool_id);
                 continue;
@@ -1318,8 +814,14 @@ void etcd_state_client_t::parse_state(const etcd_kv_t & kv)
     else if (key.substr(0, etcd_prefix.length()+11) == etcd_prefix+"/osd/state/")
     {
         // <etcd_prefix>/osd/state/%d
-        osd_num_t peer_osd = std::stoull(key.substr(etcd_prefix.length()+11));
-        if (peer_osd > 0)
+        osd_num_t peer_osd = 0;
+        char null_byte = 0;
+        int scanned = sscanf(key.c_str() + etcd_prefix.length()+11, "%ju%c", &peer_osd, &null_byte);
+        if (scanned != 1 || !peer_osd)
+        {
+            fprintf(stderr, "Bad etcd key %s, ignoring\n", key.c_str());
+        }
+        else
         {
             if (value.is_object() && value["state"] == "up")
             {

src/client/etcd_state_client.h

@@ -136,7 +136,6 @@ struct user_info_t
 };
 
 struct http_co_t;
-struct http_context_t;
 
 struct __attribute__((visibility("default"))) etcd_state_client_t
 {
@@ -147,21 +146,13 @@ protected:
     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;
+    std::set<osd_num_t> seen_peers;
     bool new_pg_config = false;
-    int ws_keepalive_timer = -1;
-    int ws_alive = 0;
-    bool rand_initialized = false;
+
     void add_etcd_url(std::string);
-    void pick_next_etcd(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();
+    void reset_pg_exists();
+    void clean_nonexistent_pgs();
 public:
     int etcd_keepalive_timeout = 30;
     int etcd_ws_keepalive_interval = 5;
@@ -180,19 +171,13 @@ public:
     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;
     uint64_t etcd_watch_revision_osd = 0;
     uint64_t etcd_watch_revision_pg = 0;
-    timespec etcd_last_reload = {};
-    int load_pgs_timer_id = -1;
+
     std::map<pool_id_t, pool_config_t> pool_config;
     std::map<osd_num_t, json11::Json> peer_states;
-    std::set<osd_num_t> seen_peers;
     std::map<inode_t, inode_config_t> inode_config;
     std::map<std::string, inode_t> inode_by_name;
     robin_hood::unordered_flat_map<std::string, std::shared_ptr<user_info_t>> user_info;
@@ -220,25 +205,23 @@ public:
     std::vector<std::string> get_addresses();
     std::shared_ptr<user_info_t> get_user(const std::string & username);
     bool check_image_perm(const std::shared_ptr<user_info_t> & user_info, inode_t inode_num, bool write);
-    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);
+    virtual void etcd_call_oneshot(const std::string & etcd_address, const std::string & api, json11::Json payload, int timeout, std::function<void(std::string, json11::Json)> callback) = 0;
+    virtual void etcd_call(const std::string & api, json11::Json payload, int timeout, int retries, int interval, std::function<void(std::string, json11::Json)> callback) = 0;
     void etcd_txn(json11::Json txn, int timeout, int retries, int interval, std::function<void(std::string, json11::Json)> callback);
     void etcd_txn_slow(json11::Json txn, std::function<void(std::string, json11::Json)> callback);
-    void start_etcd_watcher();
-    void stop_ws_keepalive();
-    void start_ws_keepalive();
-    void load_global_config();
-    void load_pgs();
-    void reset_pg_exists();
-    void clean_nonexistent_pgs();
+    virtual void etcd_add_watch(json11::Json watch) = 0;
+    virtual std::string get_username() = 0;
+    void load_global_config(std::function<void(const std::string &)> cb);
+    virtual void load_global_config() = 0;
+    void load_pgs(std::function<void(const std::string &)> cb);
+    virtual void load_pgs() = 0;
     void parse_state(const etcd_kv_t & kv);
-    void parse_config(const json11::Json & config);
+    virtual void parse_config(const json11::Json & config);
     void insert_inode_config(const inode_config_t & cfg);
     inode_watch_t* watch_inode(std::string name);
     void close_watch(inode_watch_t* watch);
     int address_count();
-    ~etcd_state_client_t();
+    virtual ~etcd_state_client_t();
 
     static uint32_t parse_immediate_commit(const std::string & immediate_commit_str, uint32_t default_value);
     static uint32_t parse_scheme(const std::string & scheme_str);

src/client/etcd_state_client_http.cpp

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

src/client/etcd_state_client_http.h

@@ -0,0 +1,50 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#pragma once
+
+#include "etcd_state_client.h"
+
+struct http_context_t;
+
+struct __attribute__((visibility("default"))) etcd_state_client_http_t: public etcd_state_client_t
+{
+protected:
+    timerfd_manager_t *tfd = NULL;
+    int ws_keepalive_timer = -1;
+    int ws_alive = 0;
+    bool rand_initialized = false;
+    int etcd_watches_initialised = 0;
+    timespec etcd_last_reload = {};
+    int load_pgs_timer_id = -1;
+    http_co_t *keepalive_client = NULL;
+    http_co_t *etcd_watch_ws = NULL;
+    http_context_t *http_ctx = NULL;
+    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<std::function<void()>> on_resolve_queue;
+
+    void pick_next_etcd(std::function<void()> cb);
+    void pick_next_etcd_on_resolve();
+    void start_etcd_watcher();
+    void start_etcd_watcher_selected();
+    void stop_ws_keepalive();
+    void start_ws_keepalive();
+    http_context_t *get_http_ctx();
+public:
+    etcd_state_client_http_t(timerfd_manager_t *tfd);
+    void 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) override;
+    void etcd_call(const std::string & api, json11::Json payload, int timeout,
+        int retries, int interval, std::function<void(std::string, json11::Json)> callback) override;
+    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 etcd_add_watch(json11::Json watch) override;
+    std::string get_username() override;
+    void load_global_config() override;
+    void load_pgs() override;
+    void parse_config(const json11::Json & config) override;
+    ~etcd_state_client_http_t();
+};

src/client/etcd_state_client_mock.cpp

@@ -0,0 +1,210 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#include <assert.h>
+#include "etcd_state_client_mock.h"
+#include "str_util.h"
+
+etcd_state_client_mock_t::etcd_state_client_mock_t()
+{
+    timespec tv;
+    clock_gettime(CLOCK_REALTIME, &tv);
+    srand48(tv.tv_sec*1000000000 + tv.tv_nsec);
+}
+
+void etcd_state_client_mock_t::etcd_add_watch(json11::Json watch)
+{
+}
+
+std::string etcd_state_client_mock_t::get_username()
+{
+    return username;
+}
+
+void etcd_state_client_mock_t::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_state_client_mock_t::pause()
+{
+    paused = true;
+}
+
+void etcd_state_client_mock_t::resume()
+{
+    paused = false;
+    auto queue = std::move(this->queue);
+    for (auto& req: queue)
+    {
+        etcd_call(req.api, req.payload, req.timeout, req.retries, req.interval, req.callback);
+    }
+}
+
+void etcd_state_client_mock_t::set(const std::string& key, json11::Json data, uint64_t mod_revision, uint64_t lease_id)
+{
+    if (!mod_revision)
+        mod_revision = ++this->mod_revision;
+    this->data[key] = (etcd_mock_key_data_t){ .value = data.dump(), .mod_revision = mod_revision, .lease_id = lease_id };
+}
+
+void etcd_state_client_mock_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 (paused)
+    {
+        queue.push_back({ api, payload, timeout, retries, interval, callback });
+        return;
+    }
+    printf("+ etcd: %s\n", api.c_str());
+    if (api == "/kv/txn")
+    {
+        bool ok = true;
+        for (auto& check: payload["compare"].array_items())
+        {
+            auto key = base64_decode(check["key"].string_value());
+            etcd_mock_key_data_t *key_data = data.find(key) != data.end() ? &data.at(key) : NULL;
+            auto target = check["target"].string_value();
+            auto res = check["result"].string_value();
+            assert(res == "LESS" || res == "");
+            bool less = res == "LESS";
+            if (target == "MOD")
+            {
+                uint64_t rev = check["mod_revision"].uint64_value();
+                assert(!less || rev);
+                ok = ok && (less ? (!key_data || key_data->mod_revision < rev) : (key_data && key_data->mod_revision == rev));
+            }
+            else if (target == "CREATE")
+            {
+                uint64_t rev = check["create_revision"].uint64_value();
+                assert(rev == 0 && !less);
+                ok = ok && !key_data;
+            }
+            else if (target == "VERSION")
+            {
+                uint64_t rev = check["version"].uint64_value();
+                assert(rev == 0 && !less);
+                ok = ok && !key_data;
+            }
+            else if (target == "LEASE")
+            {
+                assert(!less);
+                uint64_t lease_id = check["lease"].uint64_value();
+                ok = ok && key_data && key_data->lease_id == lease_id;
+            }
+            else
+                assert(0);
+        }
+        std::map<std::string, etcd_kv_t> changes;
+        bool has_mod = false;
+        for (auto& op: payload[ok ? "success" : "failure"].array_items())
+        {
+            auto& obj = op.object_items();
+            has_mod = has_mod || obj.find("request_put") != obj.end() ||
+                obj.find("request_delete_range") != obj.end();
+        }
+        if (has_mod)
+        {
+            mod_revision++;
+        }
+        json11::Json::array responses;
+        for (auto& op_ptr: payload[ok ? "success" : "failure"].array_items())
+        {
+            auto& op = op_ptr.object_items();
+            if (op.find("request_range") != op.end())
+            {
+                json11::Json::array kvs;
+                auto req = op.at("request_range");
+                auto key = base64_decode(req["key"].string_value());
+                auto range_end = base64_decode(req["range_end"].string_value());
+                auto begin_it = range_end.empty() ? data.find(key) : data.lower_bound(key);
+                auto end_it = range_end.empty() ? (begin_it == data.end() ? begin_it : std::next(begin_it)) : data.lower_bound(range_end);
+                for (auto it = begin_it; it != end_it; it++)
+                {
+                    printf("\\- get: %s = %s, rev %ju\n", it->first.c_str(), it->second.value.c_str(), it->second.mod_revision);
+                    kvs.push_back(json11::Json::object {
+                        { "key", base64_encode(it->first) },
+                        { "value", base64_encode(it->second.value) },
+                        { "mod_revision", it->second.mod_revision },
+                    });
+                }
+                responses.push_back(json11::Json::object {
+                    { "response_range", json11::Json::object{ { "header", json11::Json::object{ { "revision", mod_revision } } }, { "kvs", kvs } } },
+                });
+            }
+            else if (op.find("request_put") != op.end())
+            {
+                auto req = op.at("request_put");
+                auto key = base64_decode(req["key"].string_value());
+                auto value = base64_decode(req["value"].string_value());
+                auto lease_id = req["lease"].uint64_value();
+                printf("\\- put: %s = %s, rev %ju, lease %ju\n", key.c_str(), value.c_str(), mod_revision, lease_id);
+                data[key] = {
+                    .value = value,
+                    .mod_revision = mod_revision,
+                    .lease_id = lease_id,
+                };
+                std::string err;
+                json11::Json json_value = json11::Json::parse(value, err);
+                if (err != "")
+                {
+                    fprintf(stderr, "Invalid JSON in etcd key %s during test: %s\n", key.c_str(), value.c_str());
+                    exit(1);
+                }
+                changes[key] = { .key = key, .value = json_value, .mod_revision = mod_revision };
+                responses.push_back(json11::Json::object {
+                    { "response_put", json11::Json::object{ { "header", json11::Json::object{ { "revision", mod_revision } } } } },
+                });
+            }
+            else if (op.find("request_delete_range") != op.end())
+            {
+                auto req = op.at("request_delete_range");
+                auto key = base64_decode(req["key"].string_value());
+                auto range_end = base64_decode(req["range_end"].string_value());
+                uint64_t n_del = 0;
+                for (auto it = data.lower_bound(key); it != data.end() && (range_end == "" || it->first < range_end); )
+                {
+                    auto & key = it->first;
+                    printf("\\- del: %s\n", key.c_str());
+                    changes[key] = { .key = key, .mod_revision = mod_revision };
+                    n_del++;
+                    data.erase(it++);
+                }
+                responses.push_back(json11::Json::object {
+                    { "response_delete_range", json11::Json::object{ { "header", json11::Json::object{ { "revision", mod_revision } } }, { "deleted", n_del } } },
+                });
+            }
+        }
+        callback("", json11::Json::object{
+            { "header", json11::Json::object{ { "revision", mod_revision } } },
+            { "succeeded", ok },
+            { "responses", responses }
+        });
+        // Push changes to watcher
+        if (changes.size())
+        {
+            for (auto & kv: changes)
+                parse_state(kv.second);
+            if (on_change_hook != NULL)
+                on_change_hook(changes);
+        }
+    }
+    else if (api == "/lease/grant")
+    {
+        uint64_t lease_id = (((uint64_t)lrand48()) << 32) | lrand48();
+        leases[lease_id] = payload["TTL"].uint64_value();
+        callback("", json11::Json::object{ { "ID", std::to_string(lease_id) } });
+    }
+    else
+        callback("Unsupported", json11::Json());
+}
+
+void etcd_state_client_mock_t::load_global_config()
+{
+    etcd_state_client_t::load_global_config([this](const std::string & err) {});
+}
+
+void etcd_state_client_mock_t::load_pgs()
+{
+    etcd_state_client_t::load_pgs([this](const std::string & err) {});
+}

src/client/etcd_state_client_mock.h

@@ -0,0 +1,44 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#pragma once
+
+#include "etcd_state_client.h"
+
+struct etcd_mock_key_data_t
+{
+    std::string value;
+    uint64_t mod_revision;
+    uint64_t lease_id;
+};
+
+struct etcd_mock_request_t
+{
+    std::string api;
+    json11::Json payload;
+    int timeout;
+    int retries;
+    int interval;
+    std::function<void(std::string, json11::Json)> callback;
+};
+
+struct etcd_state_client_mock_t: public etcd_state_client_t
+{
+    uint64_t mod_revision = 0;
+    bool paused = false;
+    std::vector<etcd_mock_request_t> queue;
+public:
+    std::map<uint64_t, uint64_t> leases;
+    std::map<std::string, etcd_mock_key_data_t> data;
+    std::string username;
+    etcd_state_client_mock_t();
+    void set(const std::string& key, json11::Json data, uint64_t mod_revision = 0, uint64_t lease_id = 0);
+    void pause();
+    void resume();
+    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) override;
+    void etcd_call(const std::string & api, json11::Json payload, int timeout, int retries, int interval, std::function<void(std::string, json11::Json)> callback) override;
+    void etcd_add_watch(json11::Json watch) override;
+    std::string get_username() override;
+    void load_global_config() override;
+    void load_pgs() override;
+};

src/client/messenger.h

@@ -318,7 +318,7 @@ public:
 
 #ifdef WITH_RDMA
     bool is_rdma_enabled();
-    bool connect_rdma(uint64_t client_id, std::string rdma_address, uint64_t client_max_msg);
+    json11::Json connect_rdma(uint64_t client_id, std::string rdma_address, uint64_t client_max_msg);
 #endif
 #ifdef WITH_RDMACM
     bool is_use_rdmacm();

src/client/msgr_encrypt.cpp

@@ -427,35 +427,37 @@ void osd_messenger_t::op_encrypted_copy_buf(osd_client_t *cl, uint8_t *enc_buf,
         assert(cl->write_op->enc->key_chain[0]);
         cl->xts_enc_ctx->start(cl, cl->write_op->enc->key_chain[0], cl->write_op->req.rw.offset, cl->write_op->enc->bitmap_granularity);
     }
+    size_t old_out = done_enc;
     while (done_plain < plain_len && done_enc < enc_len)
     {
         size_t done_in = 0;
         size_t done_out = 0;
         cl->xts_enc_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;
     }
+    if (cl->write_csum_state && done_enc > old_out)
+        XXH3_64bits_update(cl->write_csum_state, enc_buf+old_out, done_enc-old_out);
 }
 
 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);
+    size_t old_in = done_enc;
     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->xts_dec_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;
     }
+    if (cl->read_csum_state && done_enc > old_in)
+        XXH3_64bits_update(cl->read_csum_state, enc_buf+old_in, done_enc-old_in);
 }
 
 void osd_messenger_t::op_decrypt_start(osd_client_t* cl)

src/client/msgr_op.cpp

@@ -3,6 +3,7 @@
 
 #include <assert.h>
 
+#include "messenger.h"
 #include "msgr_op.h"
 
 osd_op_t::~osd_op_t()
@@ -38,3 +39,52 @@ bool osd_op_t::is_recovery_related()
         req.hdr.opcode == OSD_OP_SEC_SYNC &&
         (req.sec_sync.flags & OSD_OP_RECOVERY_RELATED);
 }
+
+void osd_messenger_t::measure_exec(osd_op_t *cur_op)
+{
+    // Measure execution latency
+    if (cur_op->req.hdr.opcode > OSD_OP_MAX)
+    {
+        return;
+    }
+    if (!cur_op->tv_end.tv_sec)
+    {
+        clock_gettime(CLOCK_REALTIME, &cur_op->tv_end);
+    }
+    uint64_t len = 0;
+    if (cur_op->req.hdr.opcode == OSD_OP_READ ||
+        cur_op->req.hdr.opcode == OSD_OP_WRITE ||
+        cur_op->req.hdr.opcode == OSD_OP_SCRUB)
+    {
+        // req.rw.len is internally set to the full object size for scrubs
+        len = cur_op->req.rw.len;
+    }
+    else if (cur_op->req.hdr.opcode == OSD_OP_SEC_READ ||
+        cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE ||
+        cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE)
+    {
+        len = cur_op->req.sec_rw.len;
+    }
+    inc_op_stats(stats, cur_op->req.hdr.opcode, cur_op->tv_begin, cur_op->tv_end, len);
+    if (cur_op->is_recovery_related())
+    {
+        inc_op_stats(recovery_stats, cur_op->req.hdr.opcode, cur_op->tv_begin, cur_op->tv_end, len);
+    }
+}
+
+void osd_messenger_t::inc_op_stats(osd_op_stats_t & stats, uint64_t opcode, timespec & tv_begin, timespec & tv_end, uint64_t len)
+{
+    uint64_t usecs = (
+        (tv_end.tv_sec - tv_begin.tv_sec)*1000000 +
+        (tv_end.tv_nsec - tv_begin.tv_nsec)/1000
+    );
+    stats.op_stat_count[opcode]++;
+    if (!stats.op_stat_count[opcode])
+    {
+        stats.op_stat_count[opcode] = 1;
+        stats.op_stat_sum[opcode] = 0;
+        stats.op_stat_bytes[opcode] = 0;
+    }
+    stats.op_stat_sum[opcode] += usecs;
+    stats.op_stat_bytes[opcode] += len;
+}

src/client/msgr_rdma.cpp

@@ -507,7 +507,7 @@ int msgr_rdma_connection_t::connect(msgr_rdma_address_t *dest)
     return 0;
 }
 
-bool osd_messenger_t::connect_rdma(uint64_t client_id, std::string rdma_address, uint64_t client_max_msg)
+json11::Json 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;
@@ -523,7 +523,7 @@ bool osd_messenger_t::connect_rdma(uint64_t client_id, std::string rdma_address,
         {
             if (log_level > 0)
                 fprintf(stderr, "No RDMA context for peer %ju, using only TCP\n", client_id);
-            return false;
+            return json11::Json();
         }
         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);
         if (rdma_conn)
@@ -542,11 +542,14 @@ bool osd_messenger_t::connect_rdma(uint64_t client_id, std::string rdma_address,
                 // Remember connection, but switch to RDMA only after sending the configuration response
                 cl->rdma_conn = rdma_conn;
                 cl->peer_state = PEER_RDMA_CONNECTING;
-                return true;
+                return json11::Json::object{
+                    {"rdma_address", rdma_conn->addr.to_string()},
+                    {"rdma_max_msg", rdma_conn->max_msg},
+                };
             }
         }
     }
-    return false;
+    return json11::Json();
 }
 
 static void try_send_rdma_wr(osd_client_t *cl, ibv_sge *sge, int op_sge)

src/client/msgr_send.cpp

@@ -486,55 +486,6 @@ void osd_messenger_t::outbox_push(osd_op_t *cur_op)
     }
 }
 
-void osd_messenger_t::inc_op_stats(osd_op_stats_t & stats, uint64_t opcode, timespec & tv_begin, timespec & tv_end, uint64_t len)
-{
-    uint64_t usecs = (
-        (tv_end.tv_sec - tv_begin.tv_sec)*1000000 +
-        (tv_end.tv_nsec - tv_begin.tv_nsec)/1000
-    );
-    stats.op_stat_count[opcode]++;
-    if (!stats.op_stat_count[opcode])
-    {
-        stats.op_stat_count[opcode] = 1;
-        stats.op_stat_sum[opcode] = 0;
-        stats.op_stat_bytes[opcode] = 0;
-    }
-    stats.op_stat_sum[opcode] += usecs;
-    stats.op_stat_bytes[opcode] += len;
-}
-
-void osd_messenger_t::measure_exec(osd_op_t *cur_op)
-{
-    // Measure execution latency
-    if (cur_op->req.hdr.opcode > OSD_OP_MAX)
-    {
-        return;
-    }
-    if (!cur_op->tv_end.tv_sec)
-    {
-        clock_gettime(CLOCK_REALTIME, &cur_op->tv_end);
-    }
-    uint64_t len = 0;
-    if (cur_op->req.hdr.opcode == OSD_OP_READ ||
-        cur_op->req.hdr.opcode == OSD_OP_WRITE ||
-        cur_op->req.hdr.opcode == OSD_OP_SCRUB)
-    {
-        // req.rw.len is internally set to the full object size for scrubs
-        len = cur_op->req.rw.len;
-    }
-    else if (cur_op->req.hdr.opcode == OSD_OP_SEC_READ ||
-        cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE ||
-        cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE)
-    {
-        len = cur_op->req.sec_rw.len;
-    }
-    inc_op_stats(stats, cur_op->req.hdr.opcode, cur_op->tv_begin, cur_op->tv_end, len);
-    if (cur_op->is_recovery_related())
-    {
-        inc_op_stats(recovery_stats, cur_op->req.hdr.opcode, cur_op->tv_begin, cur_op->tv_end, len);
-    }
-}
-
 bool osd_messenger_t::try_send(osd_client_t *cl)
 {
     if (cl->peer_state == PEER_STOPPED || cl->peer_fd < 0)

src/client/nbd_proxy.cpp

@@ -301,10 +301,9 @@ const char *help_text =
     "     --nbd_disconnect_on_close 1\n"
     "       Disconnect the nbd device on close by last opener.\n"
 #endif
-#ifdef NBD_FLAG_READ_ONLY
+    "     --readonly\n"
     "     --nbd_ro 1\n"
     "       Set device into read only mode.\n"
-#endif
     "\n"
     "vitastor-nbd netlink-unmap /dev/nbdN\n"
     "  Unmap a device using netlink interface. Works with both netlink and ioctl mapped devices.\n"
@@ -348,6 +347,7 @@ protected:
     int read_ready = 0;
     msghdr read_msg = { 0 }, send_msg = { 0 };
     iovec read_iov = { 0 };
+    bool stop = false;
 
     std::string logfile = "/dev/null";
 
@@ -514,7 +514,7 @@ help:
         // Create client
         ringloop = new ring_loop_t(RINGLOOP_DEFAULT_SIZE);
         epmgr = new epoll_manager_t(ringloop);
-        cli = new cluster_client_t(ringloop, epmgr->tfd, cfg);
+        cli = cluster_client_t::create(ringloop, epmgr->tfd, cfg);
         if (!inode)
         {
             // Load image metadata
@@ -525,7 +525,7 @@ help:
                     break;
                 ringloop->wait();
             }
-            watch = cli->st_cli.watch_inode(image_name);
+            watch = cli->st_cli->watch_inode(image_name);
             device_size = watch->cfg.size;
             if (!watch->cfg.num || !device_size)
             {
@@ -581,10 +581,8 @@ help:
             }
             uint64_t flags = NBD_FLAG_SEND_FLUSH;
             uint64_t cflags = 0;
-#ifdef NBD_FLAG_READ_ONLY
-            if (!cfg["nbd_ro"].is_null())
+            if (!cfg["readonly"].is_null() || !cfg["nbd_ro"].is_null())
                 flags |= NBD_FLAG_READ_ONLY;
-#endif
 #ifdef NBD_CFLAG_DESTROY_ON_DISCONNECT
             if (!cfg["nbd_destroy_on_disconnect"].is_null())
                 cflags |= NBD_CFLAG_DESTROY_ON_DISCONNECT;
@@ -620,7 +618,10 @@ help:
             if (!cfg["dev_num"].is_null())
             {
                 int r;
-                if ((r = run_nbd(sockfd, cfg["dev_num"].int64_value(), device_size, NBD_FLAG_SEND_FLUSH, nbd_timeout, bg)) != 0)
+                uint64_t flags = NBD_FLAG_SEND_FLUSH;
+                if (!cfg["readonly"].is_null())
+                    flags |= NBD_FLAG_READ_ONLY;
+                if ((r = run_nbd(sockfd, cfg["dev_num"].int64_value(), device_size, flags, nbd_timeout, bg)) != 0)
                 {
                     fprintf(stderr, "run_nbd: %s\n", strerror(-r));
                     exit(1);
@@ -678,8 +679,7 @@ help:
         };
         ringloop->register_consumer(&consumer);
         // Add FD to epoll
-        bool stop = false;
-        epmgr->tfd->set_fd_handler(sockfd[0], false, [this, &stop](int peer_fd, int epoll_events)
+        epmgr->tfd->set_fd_handler(sockfd[0], false, [this](int peer_fd, int epoll_events)
         {
             if (epoll_events & EPOLLRDHUP)
             {
@@ -1118,7 +1118,8 @@ protected:
             {
                 // Disconnect
                 close(nbd_fd);
-                exit(0);
+                stop = true;
+                return;
             }
             if (be32toh(cur_req.magic) != NBD_REQUEST_MAGIC ||
                 req_type != NBD_CMD_READ && req_type != NBD_CMD_WRITE && req_type != NBD_CMD_FLUSH)

src/client/qemu_driver.c

@@ -705,6 +705,54 @@ static void vitastor_close(BlockDriverState *bs)
     client->last_bitmap = NULL;
 }
 
+// Unregister all event sources from the current AioContext. Called by the
+// block layer before bs is moved to a different AioContext (e.g. during live
+// migration, drain, dataplane switching). The block layer guarantees that no
+// requests are in flight at this point.
+static void vitastor_detach_aio_context(BlockDriverState *bs)
+{
+    VitastorClient *client = bs->opaque;
+    int i;
+#if defined VITASTOR_C_API_VERSION && VITASTOR_C_API_VERSION >= 2
+    if (client->uring_eventfd >= 0)
+    {
+        universal_aio_set_fd_handler(client->ctx, client->uring_eventfd, NULL, NULL, NULL);
+        // Wait until any scheduled B/H is processed before switching contexts:
+        // it would otherwise fire on the old context with stale state.
+        if (client->bh_uring_scheduled)
+        {
+            BDRV_POLL_WHILE(bs, client->bh_uring_scheduled);
+        }
+    }
+#endif
+    for (i = 0; i < client->fd_count; i++)
+    {
+        universal_aio_set_fd_handler(client->ctx, client->fds[i]->fd, NULL, NULL, NULL);
+    }
+}
+
+// (Re-)register all event sources on the new AioContext.
+static void vitastor_attach_aio_context(BlockDriverState *bs, AioContext *new_ctx)
+{
+    VitastorClient *client = bs->opaque;
+    int i;
+    client->ctx = new_ctx;
+#if defined VITASTOR_C_API_VERSION && VITASTOR_C_API_VERSION >= 2
+    if (client->uring_eventfd >= 0)
+    {
+        universal_aio_set_fd_handler(new_ctx, client->uring_eventfd, vitastor_uring_handler, NULL, client);
+    }
+#endif
+    for (i = 0; i < client->fd_count; i++)
+    {
+        VitastorFdData *fdd = client->fds[i];
+        universal_aio_set_fd_handler(new_ctx, fdd->fd,
+            fdd->fd_read ? vitastor_aio_fd_read : NULL,
+            fdd->fd_write ? vitastor_aio_fd_write : NULL,
+            fdd);
+    }
+}
+
 #if QEMU_VERSION_MAJOR >= 3 || QEMU_VERSION_MAJOR == 2 && QEMU_VERSION_MINOR >= 2
 static void vitastor_refresh_filename(BlockDriverState *bs)
 {
@@ -832,6 +880,25 @@ static int vitastor_refresh_limits(BlockDriverState *bs)
 //    return 0;
 //}
 
+// Move the running coroutine to the BlockDriverState's home AioContext.
+//
+// The block-coroutine-wrapper generator sets poll_state.ctx to
+// qemu_get_current_aio_context() in the sync wrappers (bdrv_flush(),
+// bdrv_pread() etc.). When bdrv_flush_all() runs under BQL from outside the
+// bs's iothread (e.g. on the migration thread inside do_vm_stop()), that is
+// the main AioContext, not the iothread that actually owns the bs. The
+// coroutine then runs on the wrong context while completions are delivered on
+// the iothread, and racing aio_co_schedule() vs. qemu_aio_coroutine_enter()
+// on the same coroutine triggers "Co-routine was already scheduled in
+// aio_co_schedule" and aborts the process (observed during live migration).
+//
+// aio_co_reschedule_self() is a no-op when we are already on the target ctx.
+#if QEMU_VERSION_MAJOR > 5 || QEMU_VERSION_MAJOR == 5 && QEMU_VERSION_MINOR >= 2
+#define vitastor_co_pin_to_bs_ctx(bs) aio_co_reschedule_self(bdrv_get_aio_context(bs))
+#else
+#define vitastor_co_pin_to_bs_ctx(bs) ((void)0)
+#endif
+
 static void vitastor_co_init_task(BlockDriverState *bs, VitastorRPC *task)
 {
     *task = (VitastorRPC) {
@@ -890,6 +957,7 @@ static int coroutine_fn vitastor_co_preadv(BlockDriverState *bs,
 {
     VitastorClient *client = bs->opaque;
     VitastorRPC task;
+    vitastor_co_pin_to_bs_ctx(bs);
     vitastor_co_init_task(bs, &task);
     task.iov = iov;
 
@@ -918,6 +986,7 @@ static int coroutine_fn vitastor_co_pwritev(BlockDriverState *bs,
 {
     VitastorClient *client = bs->opaque;
     VitastorRPC task;
+    vitastor_co_pin_to_bs_ctx(bs);
     vitastor_co_init_task(bs, &task);
     task.iov = iov;
 
@@ -991,6 +1060,7 @@ static int coroutine_fn vitastor_co_block_status(BlockDriverState *bs,
 #endif
     VitastorRPC task;
     VitastorClient *client = bs->opaque;
+    vitastor_co_pin_to_bs_ctx(bs);
     uint64_t inode = client->watch ? vitastor_c_inode_get_num(client->watch) : client->inode;
     uint8_t bit = 0;
     if (client->last_bitmap && client->last_bitmap_inode == inode &&
@@ -1103,6 +1173,7 @@ static int coroutine_fn vitastor_co_flush(BlockDriverState *bs)
 {
     VitastorClient *client = bs->opaque;
     VitastorRPC task;
+    vitastor_co_pin_to_bs_ctx(bs);
     vitastor_co_init_task(bs, &task);
 
     qemu_mutex_lock(&client->mutex);
@@ -1188,6 +1259,11 @@ static BlockDriver bdrv_vitastor = {
 #endif
     .bdrv_close                     = vitastor_close,
 
+    // Re-register fd handlers when the bs is moved to a different AioContext
+    // (live migration, drain, iothread reassignment).
+    .bdrv_detach_aio_context        = vitastor_detach_aio_context,
+    .bdrv_attach_aio_context        = vitastor_attach_aio_context,
+
     // Option list for the create operation
 #if QEMU_VERSION_MAJOR >= 3 || QEMU_VERSION_MAJOR == 2 && QEMU_VERSION_MINOR > 0
     .create_opts                    = &vitastor_create_opts,

src/client/ublk_server.cpp

@@ -62,6 +62,13 @@ const char *help_text =
     "All usual Vitastor config options like --config_path <path_to_config> may also be specified in CLI.\n"
 ;
 
+struct ublk_request
+{
+    uint64_t ublk_cmd;
+    int index;
+    int result;
+};
+
 class ublk_server
 {
 protected:
@@ -241,7 +248,7 @@ help:
 
         // Create client
         epmgr = new epoll_manager_t(ringloop);
-        cli = new cluster_client_t(ringloop, epmgr->tfd, cfg);
+        cli = cluster_client_t::create(ringloop, epmgr->tfd, cfg);
 
         // cli->config contains merged config
         if (!cfg["queue_depth"].is_null())
@@ -273,7 +280,7 @@ help:
         }
         if (!inode)
         {
-            watch = cli->st_cli.watch_inode(image_name);
+            watch = cli->st_cli->watch_inode(image_name);
             device_size = watch->cfg.size;
             if (!watch->cfg.num || !device_size)
             {
@@ -282,9 +289,9 @@ help:
                 exit(1);
             }
         }
-        const bool writeback = !cli->get_immediate_commit(inode);
-        auto pool_it = cli->st_cli.pool_config.find(INODE_POOL(inode ? inode : watch->cfg.num));
-        if (pool_it == cli->st_cli.pool_config.end())
+        const bool writeback = !cli->get_immediate_commit(inode ? inode : watch->cfg.num);
+        auto pool_it = cli->st_cli->pool_config.find(INODE_POOL(inode ? inode : watch->cfg.num));
+        if (pool_it == cli->st_cli->pool_config.end())
         {
             fprintf(stderr, "Pool %u does not exist\n", INODE_POOL(inode ? inode : watch->cfg.num));
             exit(1);
@@ -331,6 +338,11 @@ help:
             daemonize_fork(notifyfd);
             close(notifyfd[0]);
         }
+        consumer.loop = [this]()
+        {
+            submit_postponed();
+        };
+        ringloop->register_consumer(&consumer);
         start_device(recover);
         if (pidfile != "")
             write_pid();
@@ -350,6 +362,7 @@ help:
             ringloop->wait();
         }
         cli->flush();
+        ringloop->unregister_consumer(&consumer);
         delete cli;
         delete epmgr;
         cli = NULL;
@@ -553,6 +566,8 @@ protected:
     ublksrv_ctrl_dev_info ublk_dev = {};
     ublksrv_io_desc *ublk_queue = NULL;
     std::vector<uint8_t*> buffers;
+    ring_consumer_t consumer;
+    std::vector<ublk_request> postponed_requests;
 
     void open_control()
     {
@@ -734,9 +749,19 @@ protected:
         ctrl_fd = -1;
     }
 
-    void submit_request(uint64_t ublk_cmd, int i, int res)
+    bool submit_request(uint64_t ublk_cmd, int i, int res)
     {
         io_uring_sqe *sqe = ringloop->get_sqe();
+        if (!sqe)
+        {
+            // Handle full io_uring by postponing the request
+            postponed_requests.push_back((ublk_request){
+                .ublk_cmd = ublk_cmd,
+                .index = i,
+                .result = res,
+            });
+            return false;
+        }
         ring_data_t* data = ((ring_data_t*)sqe->user_data);
         sqe->fd = cdev_fd;
         sqe->opcode = IORING_OP_URING_CMD;
@@ -750,6 +775,22 @@ protected:
         cmd->addr = (uint64_t)buffers[i];
         cmd->result = res;
         data->callback = [this, i](ring_data_t *data) { exec_request(data->res, i); };
+        return true;
+    }
+
+    void submit_postponed()
+    {
+        int sent = 0;
+        while (postponed_requests.size())
+        {
+            ublk_request r = postponed_requests.back();
+            postponed_requests.pop_back();
+            if (!submit_request(r.ublk_cmd, r.index, r.result))
+                break;
+            sent++;
+        }
+        if (sent)
+            ringloop->submit();
     }
 
     void exec_request(int res, int i)
@@ -864,6 +905,11 @@ protected:
     int sync_ublk_cmd(uint32_t cmd_op, void *addr, uint32_t len, uint16_t dev_path_len = 0, uint64_t data0 = 0)
     {
         io_uring_sqe *sqe = ringloop->get_sqe();
+        if (!sqe)
+        {
+            fprintf(stderr, "Error: io_uring is full when trying to execute a control command\n");
+            exit(1);
+        }
         sqe->fd = ctrl_fd;
         sqe->opcode = IORING_OP_URING_CMD;
         sqe->ioprio = 0;

src/client/vitastor.pc.in

@@ -6,7 +6,7 @@ includedir=${prefix}/@CMAKE_INSTALL_INCLUDEDIR@
 
 Name: Vitastor
 Description: Vitastor client library
-Version: 3.0.12
+Version: 3.0.15
 Libs: -L${libdir} -lvitastor_client
 Cflags: -I${includedir}
 

src/client/vitastor_c.cpp

@@ -103,7 +103,7 @@ vitastor_c *vitastor_c_create_qemu(QEMUSetFDHandler *aio_set_fd_handler, void *a
         rdma_device, rdma_port_num, rdma_gid_index, rdma_mtu, log_level
     );
     auto self = vitastor_c_create_qemu_common(aio_set_fd_handler, aio_context);
-    self->cli = new cluster_client_t(NULL, self->tfd, cfg_json);
+    self->cli = cluster_client_t::create(NULL, self->tfd, cfg_json);
     return self;
 }
 
@@ -126,7 +126,7 @@ vitastor_c *vitastor_c_create_qemu_uring(QEMUSetFDHandler *aio_set_fd_handler, v
     );
     auto self = vitastor_c_create_qemu_common(aio_set_fd_handler, aio_context);
     self->ringloop = ringloop;
-    self->cli = new cluster_client_t(self->ringloop, self->tfd, cfg_json);
+    self->cli = cluster_client_t::create(self->ringloop, self->tfd, cfg_json);
     ringloop->loop();
     return self;
 }
@@ -150,7 +150,7 @@ vitastor_c *vitastor_c_create_uring(const char *config_path, const char *etcd_ho
     vitastor_c *self = new vitastor_c;
     self->ringloop = ringloop;
     self->epmgr = new epoll_manager_t(self->ringloop);
-    self->cli = new cluster_client_t(self->ringloop, self->epmgr->tfd, cfg_json);
+    self->cli = cluster_client_t::create(self->ringloop, self->epmgr->tfd, cfg_json);
     ringloop->loop();
     return self;
 }
@@ -191,7 +191,7 @@ vitastor_c *vitastor_c_create_uring_json(const char **options, int options_len)
     vitastor_c *self = new vitastor_c;
     self->ringloop = ringloop;
     self->epmgr = new epoll_manager_t(self->ringloop);
-    self->cli = new cluster_client_t(self->ringloop, self->epmgr->tfd, cfg_json);
+    self->cli = cluster_client_t::create(self->ringloop, self->epmgr->tfd, cfg_json);
     ringloop->loop();
     return self;
 }
@@ -206,7 +206,7 @@ vitastor_c *vitastor_c_create_epoll_json(const char **options, int options_len)
     json11::Json cfg_json(cfg);
     vitastor_c *self = new vitastor_c;
     self->epmgr = new epoll_manager_t(NULL);
-    self->cli = new cluster_client_t(NULL, self->epmgr->tfd, cfg_json);
+    self->cli = cluster_client_t::create(NULL, self->epmgr->tfd, cfg_json);
     return self;
 }
 
@@ -396,7 +396,7 @@ void vitastor_c_watch_inode(vitastor_c *client, char *image, VitastorIOHandler c
 {
     client->cli->on_ready([=]()
     {
-        auto watch = client->cli->st_cli.watch_inode(std::string(image));
+        auto watch = client->cli->st_cli->watch_inode(std::string(image));
         cb(opaque, (long)watch);
     });
     if (client->ringloop)
@@ -407,7 +407,7 @@ void vitastor_c_watch_inode(vitastor_c *client, char *image, VitastorIOHandler c
 
 void vitastor_c_close_watch(vitastor_c *client, void *handle)
 {
-    client->cli->st_cli.close_watch((inode_watch_t*)handle);
+    client->cli->st_cli->close_watch((inode_watch_t*)handle);
 }
 
 uint64_t vitastor_c_inode_get_size(void *handle)
@@ -424,8 +424,8 @@ uint64_t vitastor_c_inode_get_num(void *handle)
 
 uint32_t vitastor_c_inode_get_block_size(vitastor_c *client, uint64_t inode_num)
 {
-    auto pool_it = client->cli->st_cli.pool_config.find(INODE_POOL(inode_num));
-    if (pool_it == client->cli->st_cli.pool_config.end())
+    auto pool_it = client->cli->st_cli->pool_config.find(INODE_POOL(inode_num));
+    if (pool_it == client->cli->st_cli->pool_config.end())
         return 0;
     auto & pool_cfg = pool_it->second;
     uint32_t pg_data_size = (pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks);
@@ -434,8 +434,8 @@ uint32_t vitastor_c_inode_get_block_size(vitastor_c *client, uint64_t inode_num)
 
 uint32_t vitastor_c_inode_get_bitmap_granularity(vitastor_c *client, uint64_t inode_num)
 {
-    auto pool_it = client->cli->st_cli.pool_config.find(INODE_POOL(inode_num));
-    if (pool_it == client->cli->st_cli.pool_config.end())
+    auto pool_it = client->cli->st_cli->pool_config.find(INODE_POOL(inode_num));
+    if (pool_it == client->cli->st_cli->pool_config.end())
         return 0;
     // FIXME: READ_BITMAP may fails if parent bitmap granularity differs from inode bitmap granularity
     return pool_it->second.bitmap_granularity;
@@ -471,8 +471,8 @@ uint64_t vitastor_c_inode_get_mod_revision(void *handle)
 
 uint32_t vitastor_c_inode_get_immediate_commit(vitastor_c *client, uint64_t inode_num)
 {
-    auto pool_it = client->cli->st_cli.pool_config.find(INODE_POOL(inode_num));
-    if (pool_it == client->cli->st_cli.pool_config.end())
+    auto pool_it = client->cli->st_cli->pool_config.find(INODE_POOL(inode_num));
+    if (pool_it == client->cli->st_cli->pool_config.end())
         return 0;
     return pool_it->second.immediate_commit;
 }

src/cmd/CMakeLists.txt

@@ -15,6 +15,7 @@ add_custom_command(
 add_library(vitastor_cli STATIC
 	cli_common.cpp
 	cli_alloc_osd.cpp
+	cli_cpubench.cpp
 	cli_describe.cpp
 	cli_fix.cpp
 	cli_ls.cpp
@@ -50,5 +51,5 @@ add_executable(vitastor-cli
 	cli.cpp
 )
 target_link_libraries(vitastor-cli
-	vitastor_client
+	vitastor_client_int
 )

src/cmd/cli.cpp

@@ -260,12 +260,15 @@ static const char* help_text =
     "vitastor-cli rm-user|remove-user|delete-user <username>\n"
     "  Remove a user.\n"
     "\n"
+    "vitastor-cli cpubench [--json]\n"
+    "  Run CPU crypto performance tests: AES-256-GCM, AES-256-XTS and xxhash3.\n"
+    "\n"
     "vitastor-cli serve\n"
     "  Start HTTP server able to handle CLI commands over a REST API. Options:\n"
     "  --bind_address ADDR  Specify server IP address or addresses, separated by space. Default is 127.0.0.1.\n"
     "  --port 8080          Specify server port.\n"
-    "  --server_cert FILE   Path to server TLS certificate file (PEM format).\n"
-    "  --server_key FILE    Path to server TLS private key file.\n"
+    "  --api_cert FILE      Path to server TLS certificate file (PEM format).\n"
+    "  --api_pkey FILE      Path to server TLS private key file.\n"
     "  --client_ca FILE     Path to file with TLS CA certificates used to validate client connections.\n"
     "\n"
     "Use vitastor-cli --help <command> for command details or vitastor-cli --help --all for all details.\n"
@@ -629,6 +632,10 @@ std::function<bool(cli_result_t &)> cli_tool_t::start(json11::Json::object cfg,
         // Start HTTP server
         action_cb = start_serve(cfg);
     }
+    else if (cmd[0] == "cpubench")
+    {
+        action_cb = start_cpubench(cfg);
+    }
     else
     {
         result = { .err = EOPNOTSUPP, .text = "unknown command: "+cmd[0].string_value() };
@@ -648,7 +655,7 @@ static int run(cli_tool_t *p, json11::Json::object cfg)
         json11::Json cfg_j = cfg;
         p->ringloop = new ring_loop_t(RINGLOOP_DEFAULT_SIZE);
         p->epmgr = new epoll_manager_t(p->ringloop);
-        p->cli = new cluster_client_t(p->ringloop, p->epmgr->tfd, cfg_j);
+        p->cli = cluster_client_t::create(p->ringloop, p->epmgr->tfd, cfg_j);
         p->loop_and_wait(action_cb, [&](const cli_result_t & r)
         {
             result = r;

src/cmd/cli.h

@@ -67,6 +67,7 @@ public:
 
     std::function<bool(cli_result_t &)> start(json11::Json::object cfg, cli_result_t & result);
     std::function<bool(cli_result_t &)> start_alloc_osd(json11::Json);
+    std::function<bool(cli_result_t &)> start_cpubench(json11::Json);
     std::function<bool(cli_result_t &)> start_create(json11::Json);
     std::function<bool(cli_result_t &)> start_dd(json11::Json);
     std::function<bool(cli_result_t &)> start_describe(json11::Json);

src/cmd/cli_alloc_osd.cpp

@@ -35,7 +35,7 @@ struct alloc_osd_t
                         { "target", "VERSION" },
                         { "version", 0 },
                         { "key", base64_encode(
-                            parent->cli->st_cli.etcd_prefix+"/osd/stats/"+std::to_string(new_id)
+                            parent->cli->st_cli->etcd_prefix+"/osd/stats/"+std::to_string(new_id)
                         ) },
                     },
                 } },
@@ -43,7 +43,7 @@ struct alloc_osd_t
                     json11::Json::object {
                         { "request_put", json11::Json::object {
                             { "key", base64_encode(
-                                parent->cli->st_cli.etcd_prefix+"/osd/stats/"+std::to_string(new_id)
+                                parent->cli->st_cli->etcd_prefix+"/osd/stats/"+std::to_string(new_id)
                             ) },
                             { "value", base64_encode("{}") },
                         } },
@@ -52,8 +52,8 @@ struct alloc_osd_t
                 { "failure", json11::Json::array {
                     json11::Json::object {
                         { "request_range", json11::Json::object {
-                            { "key", base64_encode(parent->cli->st_cli.etcd_prefix+"/osd/stats/") },
-                            { "range_end", base64_encode(parent->cli->st_cli.etcd_prefix+"/osd/stats0") },
+                            { "key", base64_encode(parent->cli->st_cli->etcd_prefix+"/osd/stats/") },
+                            { "range_end", base64_encode(parent->cli->st_cli->etcd_prefix+"/osd/stats0") },
                             { "keys_only", true },
                         } },
                     },

src/cmd/cli_common.cpp

@@ -17,8 +17,8 @@ bool cli_tool_t::check_image_perm(const inode_config_t & cfg, bool write)
 
 json11::Json::object cli_tool_t::format_image(const inode_config_t & cfg)
 {
-    auto pool_it = cli->st_cli.pool_config.find(INODE_POOL(cfg.num));
-    bool good_pool = pool_it != cli->st_cli.pool_config.end();
+    auto pool_it = cli->st_cli->pool_config.find(INODE_POOL(cfg.num));
+    bool good_pool = pool_it != cli->st_cli->pool_config.end();
     auto img = json11::Json::object {
         { "name", cfg.name },
         { "size", cfg.size },
@@ -50,8 +50,8 @@ json11::Json::object cli_tool_t::format_image(const inode_config_t & cfg)
     }
     if (cfg.parent_id)
     {
-        auto parent_it = cli->st_cli.inode_config.find(cfg.parent_id);
-        if (parent_it != cli->st_cli.inode_config.end())
+        auto parent_it = cli->st_cli->inode_config.find(cfg.parent_id);
+        if (parent_it != cli->st_cli->inode_config.end())
         {
             img["parent_name"] = parent_it->second.name;
         }
@@ -64,8 +64,8 @@ json11::Json::object cli_tool_t::format_image(const inode_config_t & cfg)
 
 void cli_tool_t::change_parent(inode_t cur, inode_t new_parent, cli_result_t *result)
 {
-    auto cur_cfg_it = cli->st_cli.inode_config.find(cur);
-    if (cur_cfg_it == cli->st_cli.inode_config.end())
+    auto cur_cfg_it = cli->st_cli->inode_config.find(cur);
+    if (cur_cfg_it == cli->st_cli->inode_config.end())
     {
         char buf[128];
         snprintf(buf, 128, "Inode 0x%jx disappeared", cur);
@@ -74,13 +74,13 @@ void cli_tool_t::change_parent(inode_t cur, inode_t new_parent, cli_result_t *re
     }
     inode_config_t new_cfg = cur_cfg_it->second;
     std::string cur_name = new_cfg.name;
-    std::string cur_cfg_key = base64_encode(cli->st_cli.etcd_prefix+
+    std::string cur_cfg_key = base64_encode(cli->st_cli->etcd_prefix+
         "/config/inode/"+std::to_string(INODE_POOL(cur))+
         "/"+std::to_string(INODE_NO_POOL(cur)));
     new_cfg.parent_id = new_parent;
-    json11::Json::object cur_cfg_json = cli->st_cli.serialize_inode_cfg(&new_cfg);
+    json11::Json::object cur_cfg_json = cli->st_cli->serialize_inode_cfg(&new_cfg);
     waiting++;
-    cli->st_cli.etcd_txn_slow(json11::Json::object {
+    cli->st_cli->etcd_txn_slow(json11::Json::object {
         { "compare", json11::Json::array {
             json11::Json::object {
                 { "target", "MOD" },
@@ -109,8 +109,8 @@ void cli_tool_t::change_parent(inode_t cur, inode_t new_parent, cli_result_t *re
         }
         else if (new_parent)
         {
-            auto new_parent_it = cli->st_cli.inode_config.find(new_parent);
-            std::string new_parent_name = new_parent_it != cli->st_cli.inode_config.end()
+            auto new_parent_it = cli->st_cli->inode_config.find(new_parent);
+            std::string new_parent_name = new_parent_it != cli->st_cli->inode_config.end()
                 ? new_parent_it->second.name : "<unknown>";
             *result = (cli_result_t){
                 .text = "Parent of layer "+cur_name+" (inode "+std::to_string(INODE_NO_POOL(cur))+
@@ -133,7 +133,7 @@ void cli_tool_t::change_parent(inode_t cur, inode_t new_parent, cli_result_t *re
 void cli_tool_t::etcd_txn(json11::Json txn)
 {
     waiting++;
-    cli->st_cli.etcd_txn_slow(txn, [this](std::string err, json11::Json res)
+    cli->st_cli->etcd_txn_slow(txn, [this](std::string err, json11::Json res)
     {
         waiting--;
         if (err != "")
@@ -147,7 +147,7 @@ void cli_tool_t::etcd_txn(json11::Json txn)
 
 inode_config_t* cli_tool_t::get_inode_cfg(const std::string & name)
 {
-    for (auto & ic: cli->st_cli.inode_config)
+    for (auto & ic: cli->st_cli->inode_config)
     {
         if (ic.second.name == name)
         {
@@ -231,11 +231,11 @@ void cli_tool_t::iterate_kvs_1(json11::Json kvs, const std::string & prefix, std
     bool is_pool = prefix == "/pool/stats/";
     for (auto & kv_item: kvs.array_items())
     {
-        auto kv = cli->st_cli.parse_etcd_kv(kv_item);
+        auto kv = cli->st_cli->parse_etcd_kv(kv_item);
         uint64_t num = 0;
         char null_byte = 0;
         // OSD or pool number
-        int scanned = sscanf(kv.key.substr(cli->st_cli.etcd_prefix.size() + prefix.size()).c_str(), "%ju%c", &num, &null_byte);
+        int scanned = sscanf(kv.key.substr(cli->st_cli->etcd_prefix.size() + prefix.size()).c_str(), "%ju%c", &num, &null_byte);
         if (scanned != 1 || !num || is_pool && num >= POOL_ID_MAX)
         {
             fprintf(stderr, "Invalid key in etcd: %s\n", kv.key.c_str());
@@ -250,12 +250,12 @@ void cli_tool_t::iterate_kvs_2(json11::Json kvs, const std::string & prefix, std
     bool is_inode = prefix == "/config/inode/" || prefix == "/inode/stats/";
     for (auto & kv_item: kvs.array_items())
     {
-        auto kv = cli->st_cli.parse_etcd_kv(kv_item);
+        auto kv = cli->st_cli->parse_etcd_kv(kv_item);
         pool_id_t pool_id = 0;
         uint64_t num = 0;
         char null_byte = 0;
         // pool+pg or pool+inode
-        int scanned = sscanf(kv.key.substr(cli->st_cli.etcd_prefix.size() + prefix.size()).c_str(),
+        int scanned = sscanf(kv.key.substr(cli->st_cli->etcd_prefix.size() + prefix.size()).c_str(),
             "%u/%ju%c", &pool_id, &num, &null_byte);
         if (scanned != 2 || !pool_id || is_inode && INODE_POOL(num) || !is_inode && num >= UINT32_MAX)
         {

src/cmd/cli_cpubench.cpp

@@ -0,0 +1,374 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+#include <openssl/rand.h>
+#include "cli.h"
+#include "messenger.h"
+#include "msgr_encrypt.h"
+#include "msgr_op.h"
+#include "str_util.h"
+#include "xxhash.h"
+#include "xxh_x86dispatch.h"
+
+// Prevent the compiler from proving that memory is unused.
+static inline void clobber_memory(const void* p, size_t n)
+{
+#if defined(__GNUC__) || defined(__clang__)
+    __asm__ __volatile__("" : : "r"(p), "m"(*(const char(*)[1])p) : "memory");
+#else
+    (void)p; (void)n;
+#endif
+}
+
+void init_gcm(osd_client_t *cl)
+{
+    cl->my_key.resize(AES_256_GCM_KEY_SIZE+AES_256_GCM_IV_SIZE);
+    RAND_bytes(cl->my_key.data(), AES_256_GCM_KEY_SIZE+AES_256_GCM_IV_SIZE);
+#ifdef WITH_ISAL_CRYPTO
+    cl->enc_ctx = (isal_gcm_context_data*)malloc_or_die(sizeof(isal_gcm_context_data));
+    isal_aes_gcm_pre_256(cl->my_key.data(), &cl->my_key_isal);
+#else
+    cl->enc_ctx = EVP_CIPHER_CTX_new();
+    assert(cl->enc_ctx);
+    int r = EVP_EncryptInit_ex(cl->enc_ctx, EVP_aes_256_gcm(), NULL, NULL, NULL);
+    if (r != 1)
+    {
+        fprintf(stderr, "EncryptInit error: ");
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+#endif
+}
+
+void init_gcm_round(osd_client_t *cl)
+{
+#ifdef WITH_ISAL_CRYPTO
+    int r = isal_aes_gcm_init_256(&cl->my_key_isal, cl->enc_ctx, cl->my_key.data() + AES_256_GCM_KEY_SIZE, NULL, 0);
+    if (r != 0)
+    {
+        fprintf(stderr, "isal_aes_gcm_init_256 error %d\n", r);
+        abort();
+    }
+#else
+    int r = EVP_EncryptInit_ex(cl->enc_ctx, NULL, NULL, (uint8_t*)cl->my_key.data(), cl->my_key.data() + AES_256_GCM_KEY_SIZE);
+    if (r != 1)
+    {
+        fprintf(stderr, "EncryptInit error: ");
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+#endif
+}
+
+void finalize_gcm_round(osd_client_t *cl)
+{
+    uint8_t tag[16];
+#ifdef WITH_ISAL_CRYPTO
+    int r = isal_aes_gcm_enc_256_finalize(&cl->my_key_isal, cl->enc_ctx, tag, 16);
+    assert(!r);
+#else
+    int actual_out = 0;
+    int r = EVP_EncryptFinal_ex(cl->enc_ctx, NULL, &actual_out);
+    if (r != 1)
+    {
+        fprintf(stderr, "EncryptFinal error: ");
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+    assert(actual_out == 0);
+    r = EVP_CIPHER_CTX_ctrl(cl->enc_ctx, EVP_CTRL_GCM_GET_TAG, 16, tag);
+    assert(r == 1);
+#endif
+    clobber_memory(&tag, sizeof(tag));
+}
+
+void encrypt_gcm(osd_client_t *cl, uint8_t *in_buf, uint8_t *out_buf, size_t bufsize)
+{
+#ifdef WITH_ISAL_CRYPTO
+    int r = isal_aes_gcm_enc_256_update(&cl->my_key_isal, cl->enc_ctx, out_buf, in_buf, bufsize);
+    assert(!r);
+#else
+    int actual_out;
+    if (EVP_EncryptUpdate(cl->enc_ctx, out_buf, &actual_out, in_buf, bufsize) != 1)
+    {
+        fprintf(stderr, "EncryptUpdate error: ");
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+    assert(actual_out == bufsize);
+#endif
+}
+
+void bench_aes_xts(uint64_t millis, size_t bufsize, int csum_status, bool quiet, int json)
+{
+    size_t check_interval = 100;
+    if (!quiet && !json)
+    {
+        printf("%s %s block... ",
+            csum_status == MSGR_CSUM_GCM ? "AES-256-XTS + AES-256-GCM encrypt" :
+                (csum_status == MSGR_CSUM_PAYLOAD ? "AES-256-GCM encrypt header + AES-256-XTS encrypt + xxhash3" :
+                (csum_status == MSGR_CSUM_FULL ? "AES-256-XTS encrypt + xxhash3" : "AES-256-XTS encrypt")),
+            format_size(bufsize).c_str());
+    }
+    uint8_t xts_key[AES_256_XTS_KEY_SIZE];
+    RAND_bytes(xts_key, AES_256_XTS_KEY_SIZE);
+    uint8_t *in_buf = (uint8_t*)malloc_or_die(bufsize);
+    uint8_t *out_buf = (uint8_t*)malloc_or_die(bufsize);
+    XXH3_state_t *hash_state = NULL;
+    osd_client_t *cl = new osd_client_t();
+    cl->proto_csum_status = csum_status;
+    if (csum_status == MSGR_CSUM_GCM || csum_status == MSGR_CSUM_PAYLOAD)
+    {
+        init_gcm(cl);
+    }
+    if (csum_status == MSGR_CSUM_PAYLOAD || csum_status == MSGR_CSUM_FULL)
+    {
+        hash_state = XXH3_createState();
+    }
+    op_aes_xts_encrypt_t enc;
+    timespec tv_begin, tv_end;
+    clock_gettime(CLOCK_REALTIME, &tv_begin);
+    uint64_t iters = 0;
+    while (true)
+    {
+        if (csum_status == MSGR_CSUM_GCM)
+        {
+            init_gcm_round(cl);
+        }
+        if (csum_status == MSGR_CSUM_PAYLOAD || csum_status == MSGR_CSUM_FULL)
+        {
+            XXH3_64bits_reset(hash_state);
+        }
+        if (csum_status == MSGR_CSUM_PAYLOAD)
+        {
+            init_gcm_round(cl);
+            encrypt_gcm(cl, in_buf, out_buf, OSD_PACKET_SIZE);
+        }
+        enc.start(cl, xts_key, 0, 4096);
+        size_t done_in = 0, done_out = 0;
+        while (done_in < bufsize || done_out < bufsize)
+        {
+            enc.update(in_buf, bufsize, out_buf, bufsize, done_in, done_out);
+        }
+        if (csum_status == MSGR_CSUM_PAYLOAD || csum_status == MSGR_CSUM_FULL)
+        {
+            XXH3_64bits_update(hash_state, in_buf, bufsize);
+        }
+        if (csum_status == MSGR_CSUM_GCM)
+        {
+            finalize_gcm_round(cl);
+        }
+        else if (csum_status)
+        {
+            uint64_t hash = XXH3_64bits_digest(hash_state);
+            clobber_memory(&hash, sizeof(hash));
+            if (csum_status == MSGR_CSUM_PAYLOAD)
+            {
+                encrypt_gcm(cl, (uint8_t*)&hash, out_buf+OSD_PACKET_SIZE, sizeof(hash));
+                finalize_gcm_round(cl);
+            }
+        }
+        if (!(iters % check_interval))
+        {
+            clock_gettime(CLOCK_REALTIME, &tv_end);
+            uint64_t passed = tv_end.tv_sec*1000 - tv_begin.tv_sec*1000 + tv_end.tv_nsec/1000000 - tv_begin.tv_nsec/1000000;
+            if (passed >= millis)
+                break;
+            else if (passed < 10)
+                check_interval *= 10;
+        }
+        iters++;
+    }
+    uint64_t result_ms = tv_end.tv_sec*1000 - tv_begin.tv_sec*1000 + tv_end.tv_nsec/1000000 - tv_begin.tv_nsec/1000000;
+    double result_mbps = 1000.0 * bufsize / 1048576 * iters / result_ms;
+    if (!quiet)
+    {
+        if (json)
+        {
+            printf(
+                "%s{ \"xxhash3\": %s, \"aes-256-xts\": true, \"aes-256-gcm\": %s, \"bufsize\": %zu, \"iters\": %ju, \"ms\": %ju, \"mbps\": %.2f }",
+                json == 1 ? "" : ",\n    ",
+                csum_status == MSGR_CSUM_PAYLOAD || csum_status == MSGR_CSUM_FULL ? "true" : "false",
+                csum_status == MSGR_CSUM_PAYLOAD ? "\"header\"" : (csum_status == MSGR_CSUM_GCM ? "\"full\"" : "\"none\""),
+                bufsize, iters, result_ms, result_mbps
+            );
+        }
+        else
+            printf("%ju iterations in %ju ms = %.2f MB/s\n", iters, result_ms, result_mbps);
+    }
+    if (csum_status == MSGR_CSUM_PAYLOAD || csum_status == MSGR_CSUM_FULL)
+    {
+        XXH3_freeState(hash_state);
+        hash_state = NULL;
+    }
+    delete cl;
+    free(out_buf);
+    free(in_buf);
+}
+
+void bench_aes_gcm(uint64_t millis, size_t bufsize, bool with_csum, int json)
+{
+    size_t check_interval = 100;
+    if (!json)
+    {
+        printf("%s %s block... ",
+            with_csum ? "AES-256-GCM encrypt header + xxhash3" : "AES-256-GCM encrypt header and",
+            format_size(bufsize).c_str());
+    }
+    uint8_t *in_buf = (uint8_t*)malloc_or_die(bufsize);
+    uint8_t *out_buf = (uint8_t*)malloc_or_die(bufsize);
+    XXH3_state_t *hash_state = NULL;
+    osd_client_t *cl = new osd_client_t();
+    init_gcm(cl);
+    if (with_csum)
+    {
+        hash_state = XXH3_createState();
+    }
+    timespec tv_begin, tv_end;
+    clock_gettime(CLOCK_REALTIME, &tv_begin);
+    uint64_t iters = 0;
+    while (true)
+    {
+        init_gcm_round(cl);
+        encrypt_gcm(cl, in_buf, out_buf, OSD_PACKET_SIZE);
+        if (with_csum)
+        {
+            XXH3_64bits_reset(hash_state);
+            XXH3_64bits_update(hash_state, in_buf, bufsize);
+            uint64_t hash = XXH3_64bits_digest(hash_state);
+            encrypt_gcm(cl, (uint8_t*)&hash, out_buf+OSD_PACKET_SIZE, sizeof(hash));
+        }
+        else
+        {
+            encrypt_gcm(cl, in_buf, out_buf, bufsize);
+        }
+        finalize_gcm_round(cl);
+        if (!(iters % check_interval))
+        {
+            clock_gettime(CLOCK_REALTIME, &tv_end);
+            uint64_t passed = tv_end.tv_sec*1000 - tv_begin.tv_sec*1000 + tv_end.tv_nsec/1000000 - tv_begin.tv_nsec/1000000;
+            if (passed >= millis)
+                break;
+            else if (passed < 10)
+                check_interval *= 10;
+        }
+        iters++;
+    }
+    uint64_t result_ms = tv_end.tv_sec*1000 - tv_begin.tv_sec*1000 + tv_end.tv_nsec/1000000 - tv_begin.tv_nsec/1000000;
+    double result_mbps = 1000.0 * bufsize / 1048576 * iters / result_ms;
+    if (json)
+    {
+        printf(
+            "%s{ \"xxhash3\": %s, \"aes-256-xts\": false, \"aes-256-gcm\": %s, \"bufsize\": %zu, \"iters\": %ju, \"ms\": %ju, \"mbps\": %.2f }",
+            json == 1 ? "" : ",\n    ",
+            with_csum ? "true" : "false",
+            with_csum ? "\"header\"" : "\"full\"",
+            bufsize, iters, result_ms, result_mbps
+        );
+    }
+    else
+    {
+        printf("%ju iterations in %ju ms = %.2f MB/s\n", iters, result_ms, result_mbps);
+    }
+    if (with_csum)
+    {
+        XXH3_freeState(hash_state);
+        hash_state = NULL;
+    }
+    delete cl;
+    free(out_buf);
+    free(in_buf);
+}
+
+void bench_xxh(uint64_t millis, size_t bufsize, int json)
+{
+    size_t check_interval = 100;
+    if (!json)
+    {
+        printf("xxhash3 %s block... ", format_size(bufsize).c_str());
+    }
+    uint8_t *in_buf = (uint8_t*)malloc_or_die(bufsize);
+    XXH3_state_t *hash_state = XXH3_createState();
+    timespec tv_begin, tv_end;
+    clock_gettime(CLOCK_REALTIME, &tv_begin);
+    uint64_t iters = 0;
+    while (true)
+    {
+        XXH3_64bits_reset(hash_state);
+        XXH3_64bits_update(hash_state, in_buf, bufsize);
+        uint64_t hash = XXH3_64bits_digest(hash_state);
+        clobber_memory(&hash, sizeof(hash));
+        if (!(iters % check_interval))
+        {
+            clock_gettime(CLOCK_REALTIME, &tv_end);
+            uint64_t passed = tv_end.tv_sec*1000 - tv_begin.tv_sec*1000 + tv_end.tv_nsec/1000000 - tv_begin.tv_nsec/1000000;
+            if (passed >= millis)
+                break;
+            else if (passed < 10)
+                check_interval *= 10;
+        }
+        iters++;
+    }
+    uint64_t result_ms = tv_end.tv_sec*1000 - tv_begin.tv_sec*1000 + tv_end.tv_nsec/1000000 - tv_begin.tv_nsec/1000000;
+    double result_mbps = 1000.0 * bufsize / 1048576 * iters / result_ms;
+    if (json)
+    {
+        printf(
+            "%s{ \"xxhash3\": true, \"aes-256-xts\": false, \"aes-256-gcm\": \"none\", \"bufsize\": %zu, \"iters\": %ju, \"ms\": %ju, \"mbps\": %.2f }",
+            json == 1 ? "" : ",\n    ",
+            bufsize, iters, result_ms, result_mbps
+        );
+    }
+    else
+    {
+        printf("%ju iterations in %ju ms = %.2f MB/s\n", iters, result_ms, result_mbps);
+    }
+    XXH3_freeState(hash_state);
+    hash_state = NULL;
+    free(in_buf);
+}
+
+// Run hardware performance tests (for now, only encryption-related)
+std::function<bool(cli_result_t &)> cli_tool_t::start_cpubench(json11::Json cfg)
+{
+    int json = !!json_output;
+    if (!json_output)
+    {
+        printf("Vitastor transport encryption benchmark (AES-256-GCM, AES-256-XTS and xxhash3)\n");
+        printf("\nWarmup...\n");
+    }
+    else
+        printf("[\n    ");
+    bench_aes_xts(1000, 1048576, 0, true, false);
+    if (!json_output)
+        printf("\nNo transport encryption, data checksums enabled, e2e unencrypted image\n");
+    bench_xxh(2000, 1048576, json ? json++ : 0);
+    bench_xxh(2000, 4096, json ? json++ : 0);
+    if (!json_output)
+        printf("\nHeader encryption with payload checksums, e2e unencrypted image\n");
+    bench_aes_gcm(2000, 1048576, true, json ? json++ : 0);
+    bench_aes_gcm(2000, 4096, true, json ? json++ : 0);
+    if (!json_output)
+        printf("\nFull transport encryption, e2e unencrypted image\n");
+    bench_aes_gcm(2000, 1048576, false, json ? json++ : 0);
+    bench_aes_gcm(2000, 4096, false, json ? json++ : 0);
+    if (!json_output)
+        printf("\nNo transport encryption, no checksums, e2e encrypted image\n");
+    bench_aes_xts(2000, 1048576, 0, false, json ? json++ : 0);
+    bench_aes_xts(2000, 4096, 0, false, json ? json++ : 0);
+    if (!json_output)
+        printf("\nNo transport encryption, e2e encrypted image, data checksums enabled\n");
+    bench_aes_xts(2000, 1048576, MSGR_CSUM_FULL, false, json ? json++ : 0);
+    bench_aes_xts(2000, 4096, MSGR_CSUM_FULL, false, json ? json++ : 0);
+    if (!json_output)
+        printf("\nHeader encryption with payload checksums, e2e encrypted image\n");
+    bench_aes_xts(2000, 1048576, MSGR_CSUM_PAYLOAD, false, json ? json++ : 0);
+    bench_aes_xts(2000, 4096, MSGR_CSUM_PAYLOAD, false, json ? json++ : 0);
+    if (!json_output)
+        printf("\nFull transport encryption, e2e encrypted image\n");
+    bench_aes_xts(2000, 1048576, MSGR_CSUM_GCM, false, json ? json++ : 0);
+    bench_aes_xts(2000, 4096, MSGR_CSUM_GCM, false, json ? json++ : 0);
+    if (json_output)
+        printf("\n]\n");
+    return NULL;
+}

src/cmd/cli_create.cpp

@@ -53,7 +53,7 @@ struct image_creator_t
 
     void loop()
     {
-        auto & pools = parent->cli->st_cli.pool_config;
+        auto & pools = parent->cli->st_cli->pool_config;
         if (state >= 1)
             goto resume_1;
         if (image_name == "")
@@ -125,8 +125,8 @@ struct image_creator_t
         {
             return true;
         }
-        auto pool_it = parent->cli->st_cli.pool_config.find(new_pool_id);
-        if (pool_it == parent->cli->st_cli.pool_config.end() ||
+        auto pool_it = parent->cli->st_cli->pool_config.find(new_pool_id);
+        if (pool_it == parent->cli->st_cli->pool_config.end() ||
             (pool_it->second.creator_group == "" || parent->user->groups.find(pool_it->second.creator_group) == parent->user->groups.end()))
         {
             result = (cli_result_t){ .err = EACCES, .text = "Pool image create permission denied" };
@@ -142,7 +142,7 @@ struct image_creator_t
             goto resume_2;
         else if (state == 3)
             goto resume_3;
-        for (auto & ic: parent->cli->st_cli.inode_config)
+        for (auto & ic: parent->cli->st_cli->inode_config)
         {
             if (ic.second.name == image_name)
             {
@@ -222,7 +222,7 @@ resume_3:
         } while (!parent->etcd_result["succeeded"].bool_value());
         // Save into inode_config for library users to be able to take it from there immediately
         new_cfg.mod_revision = parent->etcd_result["header"]["revision"].uint64_value();
-        parent->cli->st_cli.insert_inode_config(new_cfg);
+        parent->cli->st_cli->insert_inode_config(new_cfg);
         auto img = parent->format_image(new_cfg);
         result = (cli_result_t){
             .err = 0,
@@ -240,8 +240,8 @@ resume_3:
             goto resume_3;
         else if (state == 4)
             goto resume_4;
-        // FIXME: take all info from etcd requests, not mixed with st_cli.inode_config
-        for (auto & ic: parent->cli->st_cli.inode_config)
+        // FIXME: take all info from etcd requests, not mixed with st_cli->inode_config
+        for (auto & ic: parent->cli->st_cli->inode_config)
         {
             if (ic.second.name == image_name+"@"+new_snap)
             {
@@ -307,10 +307,10 @@ resume_4:
         } while (!parent->etcd_result["succeeded"].bool_value());
         // Save into inode_config for library users to be able to take it from there immediately
         new_cfg.mod_revision = parent->etcd_result["header"]["revision"].uint64_value();
-        parent->cli->st_cli.insert_inode_config(new_cfg);
+        parent->cli->st_cli->insert_inode_config(new_cfg);
         {
-            auto new_pool_it = parent->cli->st_cli.pool_config.find(new_pool_id);
-            new_pool_name = new_pool_it != parent->cli->st_cli.pool_config.end() ? new_pool_it->second.name : "";
+            auto new_pool_it = parent->cli->st_cli->pool_config.find(new_pool_id);
+            new_pool_name = new_pool_it != parent->cli->st_cli->pool_config.end() ? new_pool_it->second.name : "";
         }
         result = (cli_result_t){
             .err = 0,
@@ -337,7 +337,7 @@ resume_4:
         return json11::Json::object {
             { "request_range", json11::Json::object {
                 { "key", base64_encode(
-                    parent->cli->st_cli.etcd_prefix+"/index/maxid/"+std::to_string(new_pool_id)
+                    parent->cli->st_cli->etcd_prefix+"/index/maxid/"+std::to_string(new_pool_id)
                 ) },
             } },
         };
@@ -349,13 +349,13 @@ resume_4:
         max_id_mod_rev = 0;
         if (response["response_range"]["kvs"].array_items().size() > 0)
         {
-            auto kv = parent->cli->st_cli.parse_etcd_kv(response["response_range"]["kvs"][0]);
+            auto kv = parent->cli->st_cli->parse_etcd_kv(response["response_range"]["kvs"][0]);
             new_id = 1+INODE_NO_POOL(kv.value.uint64_value());
             max_id_mod_rev = kv.mod_revision;
         }
         // Also check existing inodes - for the case when some inodes are created without changing /index/maxid
-        auto ino_it = parent->cli->st_cli.inode_config.lower_bound(INODE_WITH_POOL(new_pool_id+1, 0));
-        if (ino_it != parent->cli->st_cli.inode_config.begin())
+        auto ino_it = parent->cli->st_cli->inode_config.lower_bound(INODE_WITH_POOL(new_pool_id+1, 0));
+        if (ino_it != parent->cli->st_cli->inode_config.begin())
         {
             ino_it--;
             if (INODE_POOL(ino_it->first) == new_pool_id && new_id < 1+INODE_NO_POOL(ino_it->first))
@@ -374,7 +374,7 @@ resume_4:
             json11::Json::object {
                 { "request_range", json11::Json::object {
                     { "key", base64_encode(
-                        parent->cli->st_cli.etcd_prefix+"/index/image/"+image_name
+                        parent->cli->st_cli->etcd_prefix+"/index/image/"+image_name
                     ) },
                 } },
             },
@@ -395,7 +395,7 @@ resume_2:
         idx_mod_rev = 0;
         if (parent->etcd_result["responses"][1]["response_range"]["kvs"].array_items().size() == 0)
         {
-            for (auto & ic: parent->cli->st_cli.inode_config)
+            for (auto & ic: parent->cli->st_cli->inode_config)
             {
                 if (ic.second.name == image_name)
                 {
@@ -411,7 +411,7 @@ resume_2:
         {
             // FIXME: Parse kvs in etcd_state_client automatically
             {
-                auto kv = parent->cli->st_cli.parse_etcd_kv(parent->etcd_result["responses"][1]["response_range"]["kvs"][0]);
+                auto kv = parent->cli->st_cli->parse_etcd_kv(parent->etcd_result["responses"][1]["response_range"]["kvs"][0]);
                 old_id = INODE_NO_POOL(kv.value["id"].uint64_value());
                 old_pool_id = (pool_id_t)kv.value["pool_id"].uint64_value();
                 idx_mod_rev = kv.mod_revision;
@@ -427,7 +427,7 @@ resume_2:
                     json11::Json::object {
                         { "request_range", json11::Json::object {
                             { "key", base64_encode(
-                                parent->cli->st_cli.etcd_prefix+"/config/inode/"+
+                                parent->cli->st_cli->etcd_prefix+"/config/inode/"+
                                 std::to_string(old_pool_id)+"/"+std::to_string(old_id)
                             ) },
                         } },
@@ -445,8 +445,8 @@ resume_3:
                 return;
             }
             {
-                auto kv = parent->cli->st_cli.parse_etcd_kv(parent->etcd_result["responses"][0]["response_range"]["kvs"][0]);
-                cur_cfg = parent->cli->st_cli.deserialize_inode_cfg(INODE_WITH_POOL(old_pool_id, old_id), kv.value, kv.mod_revision);
+                auto kv = parent->cli->st_cli->parse_etcd_kv(parent->etcd_result["responses"][0]["response_range"]["kvs"][0]);
+                cur_cfg = parent->cli->st_cli->deserialize_inode_cfg(INODE_WITH_POOL(old_pool_id, old_id), kv.value, kv.mod_revision);
                 size = cur_cfg.size;
             }
         }
@@ -474,7 +474,7 @@ resume_3:
         {
             new_cfg.enc_key = cur_cfg.enc_key;
         }
-        new_cfg.owner = http_context_get_ssl_cn(parent->cli->st_cli.get_http_ctx());
+        new_cfg.owner = parent->cli->st_cli->get_username();
         if (!new_owner.empty())
         {
             new_cfg.owner = new_owner;
@@ -492,7 +492,7 @@ resume_3:
                 { "target", "VERSION" },
                 { "version", 0 },
                 { "key", base64_encode(
-                    parent->cli->st_cli.etcd_prefix+"/config/inode/"+
+                    parent->cli->st_cli->etcd_prefix+"/config/inode/"+
                     std::to_string(new_pool_id)+"/"+std::to_string(new_id)
                 ) },
             },
@@ -500,31 +500,31 @@ resume_3:
                 { "target", "VERSION" },
                 { "version", 0 },
                 { "key", base64_encode(
-                    parent->cli->st_cli.etcd_prefix+"/index/image/"+image_name+
+                    parent->cli->st_cli->etcd_prefix+"/index/image/"+image_name+
                     (new_snap != "" ? "@"+new_snap : "")
                 ) },
             },
             json11::Json::object {
                 { "target", "MOD" },
                 { "mod_revision", max_id_mod_rev },
-                { "key", base64_encode(parent->cli->st_cli.etcd_prefix+"/index/maxid/"+std::to_string(new_pool_id)) },
+                { "key", base64_encode(parent->cli->st_cli->etcd_prefix+"/index/maxid/"+std::to_string(new_pool_id)) },
             },
         };
         json11::Json::array success = json11::Json::array {
             json11::Json::object {
                 { "request_put", json11::Json::object {
                     { "key", base64_encode(
-                        parent->cli->st_cli.etcd_prefix+"/config/inode/"+
+                        parent->cli->st_cli->etcd_prefix+"/config/inode/"+
                         std::to_string(new_pool_id)+"/"+std::to_string(new_id)
                     ) },
                     { "value", base64_encode(
-                        json11::Json(parent->cli->st_cli.serialize_inode_cfg(&new_cfg)).dump()
+                        json11::Json(parent->cli->st_cli->serialize_inode_cfg(&new_cfg)).dump()
                     ) },
                 } },
             },
             json11::Json::object {
                 { "request_put", json11::Json::object {
-                    { "key", base64_encode(parent->cli->st_cli.etcd_prefix+"/index/image/"+image_name) },
+                    { "key", base64_encode(parent->cli->st_cli->etcd_prefix+"/index/image/"+image_name) },
                     { "value", base64_encode(json11::Json(json11::Json::object{
                         { "id", new_id },
                         { "pool_id", (uint64_t)new_pool_id },
@@ -534,7 +534,7 @@ resume_3:
             json11::Json::object {
                 { "request_put", json11::Json::object {
                     { "key", base64_encode(
-                        parent->cli->st_cli.etcd_prefix+"/index/maxid/"+
+                        parent->cli->st_cli->etcd_prefix+"/index/maxid/"+
                         std::to_string(new_pool_id)
                     ) },
                     { "value", base64_encode(std::to_string(new_id)) }
@@ -545,7 +545,7 @@ resume_3:
             json11::Json::object {
                 { "request_range", json11::Json::object {
                     { "key", base64_encode(
-                        parent->cli->st_cli.etcd_prefix+"/index/image/"+
+                        parent->cli->st_cli->etcd_prefix+"/index/image/"+
                         image_name+(new_snap != "" ? "@"+new_snap : "")
                     ) },
                 } },
@@ -560,29 +560,29 @@ resume_3:
                 { "target", "MOD" },
                 { "mod_revision", cur_cfg.mod_revision },
                 { "key", base64_encode(
-                    parent->cli->st_cli.etcd_prefix+"/config/inode/"+
+                    parent->cli->st_cli->etcd_prefix+"/config/inode/"+
                     std::to_string(old_pool_id)+"/"+std::to_string(old_id)
                 ) },
             });
             checks.push_back(json11::Json::object {
                 { "target", "MOD" },
                 { "mod_revision", idx_mod_rev },
-                { "key", base64_encode(parent->cli->st_cli.etcd_prefix+"/index/image/"+image_name) }
+                { "key", base64_encode(parent->cli->st_cli->etcd_prefix+"/index/image/"+image_name) }
             });
             success.push_back(json11::Json::object {
                 { "request_put", json11::Json::object {
                     { "key", base64_encode(
-                        parent->cli->st_cli.etcd_prefix+"/config/inode/"+
+                        parent->cli->st_cli->etcd_prefix+"/config/inode/"+
                         std::to_string(old_pool_id)+"/"+std::to_string(old_id)
                     ) },
                     { "value", base64_encode(
-                        json11::Json(parent->cli->st_cli.serialize_inode_cfg(&snap_cfg)).dump()
+                        json11::Json(parent->cli->st_cli->serialize_inode_cfg(&snap_cfg)).dump()
                     ) },
                 } },
             });
             success.push_back(json11::Json::object {
                 { "request_put", json11::Json::object {
-                    { "key", base64_encode(parent->cli->st_cli.etcd_prefix+"/index/image/"+image_name+"@"+new_snap) },
+                    { "key", base64_encode(parent->cli->st_cli->etcd_prefix+"/index/image/"+image_name+"@"+new_snap) },
                     { "value", base64_encode(json11::Json(json11::Json::object{
                         { "id", old_id },
                         { "pool_id", (uint64_t)old_pool_id },

src/cmd/cli_dd.cpp

@@ -52,19 +52,19 @@ struct dd_in_info_t
         in_seekable = true;
         if (iimg != "")
         {
-            iwatch = parent->cli->st_cli.watch_inode(iimg);
+            iwatch = parent->cli->st_cli->watch_inode(iimg);
             if (!iwatch->cfg.num)
             {
                 result = (cli_result_t){ .err = ENOENT, .text = "Image "+iimg+" does not exist" };
-                parent->cli->st_cli.close_watch(iwatch);
+                parent->cli->st_cli->close_watch(iwatch);
                 iwatch = NULL;
                 return;
             }
-            auto pool_it = parent->cli->st_cli.pool_config.find(INODE_POOL(iwatch->cfg.num));
-            if (pool_it == parent->cli->st_cli.pool_config.end())
+            auto pool_it = parent->cli->st_cli->pool_config.find(INODE_POOL(iwatch->cfg.num));
+            if (pool_it == parent->cli->st_cli->pool_config.end())
             {
                 result = (cli_result_t){ .err = ENOENT, .text = "Pool of image "+iimg+" does not exist" };
-                parent->cli->st_cli.close_watch(iwatch);
+                parent->cli->st_cli->close_watch(iwatch);
                 iwatch = NULL;
                 return;
             }
@@ -131,7 +131,7 @@ struct dd_in_info_t
     {
         if (iimg != "")
         {
-            parent->cli->st_cli.close_watch(iwatch);
+            parent->cli->st_cli->close_watch(iwatch);
             iwatch = NULL;
         }
         else if (ifile != "")
@@ -163,11 +163,11 @@ struct dd_out_info_t
 
     pool_config_t *find_pool(cli_tool_t *parent, const std::string & name)
     {
-        if (name == "" && parent->cli->st_cli.pool_config.size() == 1)
+        if (name == "" && parent->cli->st_cli->pool_config.size() == 1)
         {
-            return &parent->cli->st_cli.pool_config.begin()->second;
+            return &parent->cli->st_cli->pool_config.begin()->second;
         }
-        for (auto & pp: parent->cli->st_cli.pool_config)
+        for (auto & pp: parent->cli->st_cli->pool_config)
         {
             if (pp.second.name == name)
             {
@@ -186,14 +186,14 @@ struct dd_out_info_t
         if (oimg != "")
         {
             out_seekable = true;
-            owatch = parent->cli->st_cli.watch_inode(oimg);
+            owatch = parent->cli->st_cli->watch_inode(oimg);
             if (owatch->cfg.num)
             {
-                auto pool_it = parent->cli->st_cli.pool_config.find(INODE_POOL(owatch->cfg.num));
-                if (pool_it == parent->cli->st_cli.pool_config.end())
+                auto pool_it = parent->cli->st_cli->pool_config.find(INODE_POOL(owatch->cfg.num));
+                if (pool_it == parent->cli->st_cli->pool_config.end())
                 {
                     result = (cli_result_t){ .err = ENOENT, .text = "Pool of image "+oimg+" does not exist" };
-                    parent->cli->st_cli.close_watch(owatch);
+                    parent->cli->st_cli->close_watch(owatch);
                     owatch = NULL;
                     return true;
                 }
@@ -209,7 +209,7 @@ struct dd_out_info_t
                 else
                 {
                     result = (cli_result_t){ .err = ENOENT, .text = "Pool to create output image "+oimg+" is not specified" };
-                    parent->cli->st_cli.close_watch(owatch);
+                    parent->cli->st_cli->close_watch(owatch);
                     owatch = NULL;
                     return true;
                 }
@@ -224,14 +224,14 @@ struct dd_out_info_t
                 if (!out_create)
                 {
                     result = (cli_result_t){ .err = ENOENT, .text = "Image "+oimg+" does not exist" };
-                    parent->cli->st_cli.close_watch(owatch);
+                    parent->cli->st_cli->close_watch(owatch);
                     owatch = NULL;
                     return true;
                 }
                 if (!out_size)
                 {
                     result = (cli_result_t){ .err = ENOENT, .text = "Input size is unknown, specify size to create output image "+oimg };
-                    parent->cli->st_cli.close_watch(owatch);
+                    parent->cli->st_cli->close_watch(owatch);
                     owatch = NULL;
                     return true;
                 }
@@ -247,7 +247,7 @@ struct dd_out_info_t
                 if (!out_size)
                 {
                     result = (cli_result_t){ .err = ENOENT, .text = "Input size is unknown, specify size to truncate output image" };
-                    parent->cli->st_cli.close_watch(owatch);
+                    parent->cli->st_cli->close_watch(owatch);
                     owatch = NULL;
                     return true;
                 }
@@ -275,7 +275,7 @@ resume_1:
             sub_cb = NULL;
             if (result.err)
             {
-                parent->cli->st_cli.close_watch(owatch);
+                parent->cli->st_cli->close_watch(owatch);
                 owatch = NULL;
                 return true;
             }
@@ -324,8 +324,14 @@ resume_2:
                 cluster_op_t *sync_op = new cluster_op_t;
                 sync_op->opcode = OSD_OP_SYNC;
                 parent->waiting++;
-                sync_op->callback = [parent](cluster_op_t *sync_op)
+                sync_op->callback = [this, parent](cluster_op_t *sync_op)
                 {
+                    if (sync_op->retval != 0 && !result.err)
+                    {
+                        // Just in case, actually OP_SYNC can't fail
+                        result.err = -sync_op->retval;
+                        result.text = "Failed to sync "+oimg+": "+std::string(strerror(result.err));
+                    }
                     parent->waiting--;
                     delete sync_op;
                     parent->ringloop->wakeup();
@@ -354,7 +360,7 @@ resume_2:
     {
         if (oimg != "")
         {
-            parent->cli->st_cli.close_watch(owatch);
+            parent->cli->st_cli->close_watch(owatch);
             owatch = NULL;
         }
         else

src/cmd/cli_describe.cpp

@@ -72,7 +72,7 @@ struct cli_describe_t
         only_pool = pool_id;
         if (!only_pool && pool_name != "")
         {
-            for (auto & pp: parent->cli->st_cli.pool_config)
+            for (auto & pp: parent->cli->st_cli->pool_config)
             {
                 if (pp.second.name == pool_name)
                 {
@@ -153,7 +153,7 @@ struct cli_describe_t
         {
             uint64_t min_pool = min_inode >> (64-POOL_ID_BITS);
             uint64_t max_pool = max_inode >> (64-POOL_ID_BITS);
-            for (auto & pp: parent->cli->st_cli.pool_config)
+            for (auto & pp: parent->cli->st_cli->pool_config)
             {
                 if (pp.first >= min_pool && (!max_pool || pp.first <= max_pool))
                 {

src/cmd/cli_fix.cpp

@@ -134,8 +134,8 @@ struct cli_fix_t
                 return;
             }
             auto & obj = objects[processed_count++];
-            auto pool_cfg_it = parent->cli->st_cli.pool_config.find(INODE_POOL(obj.inode));
-            if (pool_cfg_it == parent->cli->st_cli.pool_config.end())
+            auto pool_cfg_it = parent->cli->st_cli->pool_config.find(INODE_POOL(obj.inode));
+            if (pool_cfg_it == parent->cli->st_cli->pool_config.end())
             {
                 fprintf(stderr, "Object %jx:%jx is from unknown pool\n", obj.inode, obj.stripe);
                 continue;

src/cmd/cli_flatten.cpp

@@ -47,8 +47,8 @@ struct snap_flattener_t
         chain_list.push_back(cur->num);
         while (cur->parent_id != 0 && cur->parent_id != target_cfg->num)
         {
-            auto it = parent->cli->st_cli.inode_config.find(cur->parent_id);
-            if (it == parent->cli->st_cli.inode_config.end())
+            auto it = parent->cli->st_cli->inode_config.find(cur->parent_id);
+            if (it == parent->cli->st_cli->inode_config.end())
             {
                 result = (cli_result_t){
                     .err = ENOENT,
@@ -103,9 +103,9 @@ struct snap_flattener_t
             { "from", top_parent_name },
             { "to", target_name },
             { "target", target_name },
-            { "delete-source", false },
+            { "delete_source", false },
             { "cas", use_cas },
-            { "fsync-interval", fsync_interval },
+            { "fsync_interval", fsync_interval },
         });
         // Wait for it
 resume_1:

src/cmd/cli_ls.cpp

@@ -38,7 +38,7 @@ struct image_lister_t
     {
         if (list_pool_name != "")
         {
-            for (auto & ic: parent->cli->st_cli.pool_config)
+            for (auto & ic: parent->cli->st_cli->pool_config)
             {
                 if (ic.second.name == list_pool_name)
                 {
@@ -54,11 +54,11 @@ struct image_lister_t
             }
         }
         auto begin_it = list_pool_id
-            ? parent->cli->st_cli.inode_config.lower_bound(INODE_WITH_POOL(list_pool_id, 0))
-            : parent->cli->st_cli.inode_config.begin();
+            ? parent->cli->st_cli->inode_config.lower_bound(INODE_WITH_POOL(list_pool_id, 0))
+            : parent->cli->st_cli->inode_config.begin();
         auto end_it = list_pool_id
-            ? parent->cli->st_cli.inode_config.lower_bound(INODE_WITH_POOL(list_pool_id+1, 0))
-            : parent->cli->st_cli.inode_config.end();
+            ? parent->cli->st_cli->inode_config.lower_bound(INODE_WITH_POOL(list_pool_id+1, 0))
+            : parent->cli->st_cli->inode_config.end();
         for (auto it = begin_it; it != end_it; it++)
         {
             if (!parent->check_image_perm(it->second, false))
@@ -81,21 +81,21 @@ struct image_lister_t
                 json11::Json::object {
                     { "request_range", (list_pool_id
                         ? json11::Json::object {
-                            { "key", base64_encode(parent->cli->st_cli.etcd_prefix+"/pool/stats/"+std::to_string(list_pool_id)) },
+                            { "key", base64_encode(parent->cli->st_cli->etcd_prefix+"/pool/stats/"+std::to_string(list_pool_id)) },
                         }
                         : json11::Json::object {
-                            { "key", base64_encode(parent->cli->st_cli.etcd_prefix+"/pool/stats/") },
-                            { "range_end", base64_encode(parent->cli->st_cli.etcd_prefix+"/pool/stats0") },
+                            { "key", base64_encode(parent->cli->st_cli->etcd_prefix+"/pool/stats/") },
+                            { "range_end", base64_encode(parent->cli->st_cli->etcd_prefix+"/pool/stats0") },
                         }) },
                 },
                 json11::Json::object {
                     { "request_range", json11::Json::object {
                         { "key", base64_encode(
-                            parent->cli->st_cli.etcd_prefix+"/inode/stats"+
+                            parent->cli->st_cli->etcd_prefix+"/inode/stats"+
                             (list_pool_id ? "/"+std::to_string(list_pool_id) : "")+"/"
                         ) },
                         { "range_end", base64_encode(
-                            parent->cli->st_cli.etcd_prefix+"/inode/stats"+
+                            parent->cli->st_cli->etcd_prefix+"/inode/stats"+
                             (list_pool_id ? "/"+std::to_string(list_pool_id) : "")+"0"
                         ) },
                     } },
@@ -117,11 +117,11 @@ resume_1:
         std::map<pool_id_t, uint64_t> pool_pg_real_size;
         for (auto & kv_item: space_info["responses"][0]["response_range"]["kvs"].array_items())
         {
-            auto kv = parent->cli->st_cli.parse_etcd_kv(kv_item);
+            auto kv = parent->cli->st_cli->parse_etcd_kv(kv_item);
             // pool ID
             pool_id_t pool_id;
             char null_byte = 0;
-            int scanned = sscanf(kv.key.substr(parent->cli->st_cli.etcd_prefix.length()).c_str(), "/pool/stats/%u%c", &pool_id, &null_byte);
+            int scanned = sscanf(kv.key.substr(parent->cli->st_cli->etcd_prefix.length()).c_str(), "/pool/stats/%u%c", &pool_id, &null_byte);
             if (scanned != 1 || !pool_id || pool_id >= POOL_ID_MAX)
             {
                 fprintf(stderr, "Invalid key in etcd: %s\n", kv.key.c_str());
@@ -132,12 +132,12 @@ resume_1:
         }
         for (auto & kv_item: space_info["responses"][1]["response_range"]["kvs"].array_items())
         {
-            auto kv = parent->cli->st_cli.parse_etcd_kv(kv_item);
+            auto kv = parent->cli->st_cli->parse_etcd_kv(kv_item);
             // pool ID & inode number
             pool_id_t pool_id;
             inode_t only_inode_num;
             char null_byte = 0;
-            int scanned = sscanf(kv.key.substr(parent->cli->st_cli.etcd_prefix.length()).c_str(),
+            int scanned = sscanf(kv.key.substr(parent->cli->st_cli->etcd_prefix.length()).c_str(),
                 "/inode/stats/%u/%ju%c", &pool_id, &only_inode_num, &null_byte);
             if (scanned != 2 || !pool_id || pool_id >= POOL_ID_MAX || INODE_POOL(only_inode_num) != 0)
             {
@@ -152,8 +152,8 @@ resume_1:
                 continue;
             }
             // save stats
-            auto pool_it = parent->cli->st_cli.pool_config.find(pool_id);
-            if (pool_it != parent->cli->st_cli.pool_config.end())
+            auto pool_it = parent->cli->st_cli->pool_config.find(pool_id);
+            if (pool_it != parent->cli->st_cli->pool_config.end())
             {
                 auto & pool_cfg = pool_it->second;
                 used_size = used_size / (pool_pg_real_size[pool_id] ? pool_pg_real_size[pool_id] : 1)
@@ -166,7 +166,7 @@ resume_1:
                     { "size", 0 },
                     { "readonly", false },
                     { "pool_id", (uint64_t)INODE_POOL(inode_num) },
-                    { "pool_name", pool_it != parent->cli->st_cli.pool_config.end()
+                    { "pool_name", pool_it != parent->cli->st_cli->pool_config.end()
                         ? (pool_it->second.name == "" ? "<Unnamed>" : pool_it->second.name) : "?" },
                     { "inode_num", INODE_NO_POOL(inode_num) },
                     { "inode_id", inode_num },

src/cmd/cli_merge.cpp

@@ -110,8 +110,8 @@ struct snap_merger_t
             cur->parent_id != to_cfg->num &&
             cur->parent_id != 0)
         {
-            auto it = parent->cli->st_cli.inode_config.find(cur->parent_id);
-            if (it == parent->cli->st_cli.inode_config.end())
+            auto it = parent->cli->st_cli->inode_config.find(cur->parent_id);
+            if (it == parent->cli->st_cli->inode_config.end())
             {
                 result = (cli_result_t){
                     .err = ENOENT,
@@ -166,7 +166,7 @@ struct snap_merger_t
         //
         //    <from> - <layer 1> - <target> - <to>
         //          \- <layer 2> <---------X-------- NOT ALLOWED
-        for (auto & ic: parent->cli->st_cli.inode_config)
+        for (auto & ic: parent->cli->st_cli->inode_config)
         {
             auto it = sources.find(ic.second.num);
             if (it == sources.end() && ic.second.parent_id != 0)
@@ -182,7 +182,7 @@ struct snap_merger_t
                             .text = "Layers at or above "+(check_delete_source ? from_name : target_name)+
                                 ", but below "+to_name+" are not allowed to have other children, but "+
                                 ic.second.name+" is a child of "+
-                                parent->cli->st_cli.inode_config.at(ic.second.parent_id).name,
+                                parent->cli->st_cli->inode_config.at(ic.second.parent_id).name,
                         };
                         state = 100;
                         return;
@@ -213,7 +213,7 @@ struct snap_merger_t
 
     uint64_t get_block_size(inode_t inode, uint32_t *bitmap_granularity)
     {
-        auto & pool_cfg = parent->cli->st_cli.pool_config.at(INODE_POOL(inode));
+        auto & pool_cfg = parent->cli->st_cli->pool_config.at(INODE_POOL(inode));
         uint64_t pg_data_size = (pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks);
         if (bitmap_granularity)
             *bitmap_granularity = pool_cfg.bitmap_granularity;
@@ -251,6 +251,8 @@ struct snap_merger_t
             goto resume_100;
         // Get parents and so on
         start_merge();
+        if (state == 100)
+            return;
         // First list lower layers
         list_errcode.clear();
         list_layers(true);
@@ -418,7 +420,7 @@ struct snap_merger_t
                     }
                     if (!pgs_left)
                     {
-                        auto & name = parent->cli->st_cli.inode_config.at(src).name;
+                        auto & name = parent->cli->st_cli->inode_config.at(src).name;
                         if (list_errcode.find(src) != list_errcode.end())
                         {
                             fprintf(stderr, "Failed to get listing of layer %s (inode %ju in pool %u): %s (code %d)\n",
@@ -612,8 +614,10 @@ struct snap_merger_t
         subop->offset = offset;
         subop->len = 0;
         subop->flags = OSD_OP_IGNORE_READONLY | OSD_OP_WAIT_UP_TIMEOUT;
-        subop->callback = [](cluster_op_t *subop)
+        in_flight++;
+        subop->callback = [this](cluster_op_t *subop)
         {
+            in_flight--;
             if (subop->retval != 0)
             {
                 fprintf(stderr, "error deleting from layer 0x%jx at offset %jx: %s", subop->inode, subop->offset, strerror(-subop->retval));
@@ -640,8 +644,10 @@ struct snap_merger_t
                     uint64_t to = last_written_offset;
                     cluster_op_t *subop = new cluster_op_t;
                     subop->opcode = OSD_OP_SYNC;
+                    in_flight++;
                     subop->callback = [this, to](cluster_op_t *subop)
                     {
+                        in_flight--;
                         delete subop;
                         // We can now delete source data between <from> and <to>
                         // But to do this we have to keep all object lists in memory :-(