HTTP "hello, world!"

Seems slightly overcomplicated...

.gitea/workflows/test.yml

@@ -234,6 +234,60 @@ jobs:
           echo ""
         done
 
+  test_etcd_fail_https:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 10
+      run: ETCD_SCHEME=https /root/vitastor/tests/test_etcd_fail.sh
+    - name: Print logs
+      if: always() && steps.test.outcome == 'failure'
+      run: |
+        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
+          echo "-------- $i --------"
+          cat $i
+          echo ""
+        done
+
+  test_etcd_fail_https_antietcd:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 10
+      run: ETCD_SCHEME=https ANTIETCD=1 /root/vitastor/tests/test_etcd_fail.sh
+    - name: Print logs
+      if: always() && steps.test.outcome == 'failure'
+      run: |
+        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
+          echo "-------- $i --------"
+          cat $i
+          echo ""
+        done
+
+  test_snapshot_https:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: ETCD_SCHEME=https /root/vitastor/tests/test_snapshot.sh
+    - name: Print logs
+      if: always() && steps.test.outcome == 'failure'
+      run: |
+        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
+          echo "-------- $i --------"
+          cat $i
+          echo ""
+        done
+
   test_interrupted_rebalance:
     runs-on: ubuntu-latest
     needs: build
@@ -648,6 +702,24 @@ jobs:
           echo ""
         done
 
+  test_snapshot_chain_encrypted:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: ENCRYPTED=1 /root/vitastor/tests/test_snapshot_chain.sh
+    - name: Print logs
+      if: always() && steps.test.outcome == 'failure'
+      run: |
+        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
+          echo "-------- $i --------"
+          cat $i
+          echo ""
+        done
+
   test_old_snapshot_chain:
     runs-on: ubuntu-latest
     needs: build
@@ -1224,6 +1296,24 @@ jobs:
           echo ""
         done
 
+  test_checksum_xxhash:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: TEST_NAME=xxhash OSD_ARGS="--data_csum_type xxh3_32" /root/vitastor/tests/test_checksum.sh
+    - name: Print logs
+      if: always() && steps.test.outcome == 'failure'
+      run: |
+        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
+          echo "-------- $i --------"
+          cat $i
+          echo ""
+        done
+
   test_old_checksum:
     runs-on: ubuntu-latest
     needs: build
@@ -2052,3 +2142,39 @@ jobs:
           echo ""
         done
 
+  test_write_encrypted:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: /root/vitastor/tests/test_write_encrypted.sh
+    - name: Print logs
+      if: always() && steps.test.outcome == 'failure'
+      run: |
+        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
+          echo "-------- $i --------"
+          cat $i
+          echo ""
+        done
+
+  test_write_encrypted_ec:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: SCHEME=ec /root/vitastor/tests/test_write_encrypted.sh
+    - name: Print logs
+      if: always() && steps.test.outcome == 'failure'
+      run: |
+        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
+          echo "-------- $i --------"
+          cat $i
+          echo ""
+        done
+

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

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

.gitignore

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

README-ru.md

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

README.md

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

debian/control

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

debian/vitastor-buildenv.Dockerfile

@@ -25,7 +25,7 @@ RUN set -e -x; \
     echo 'APT::Install-Suggests false;' >> /etc/apt/apt.conf
 
 RUN apt-get update && \
-    apt-get -y install fio libgoogle-perftools-dev devscripts libjerasure-dev cmake \
+    apt-get -y install fio libgoogle-perftools-dev devscripts libjerasure-dev cmake libc-ares-dev \
         libibverbs-dev librdmacm-dev libisal-dev libnl-3-dev libnl-genl-3-dev curl nodejs npm node-nan node-bindings && \
     apt-get -y build-dep fio && \
     apt-get --download-only source fio

docs/config.en.md

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

docs/config.ru.md

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

docs/config/layout-osd.en.md

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

docs/config/layout-osd.ru.md

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

docs/config/security.en.md

@@ -0,0 +1,150 @@
+[Documentation](../../README.md#documentation) → [Configuration](../config.en.md) → Security Parameters
+
+-----
+
+[Читать на русском](security.ru.md)
+
+# Security Parameters
+
+These parameters affect your Vitastor installation security and apply to OSDs, monitors and clients.
+
+Most of them can be set in /etc/vitastor/vitastor.conf and in etcd, but don't support online modification.
+
+- [etcd_client_cert](#etcd_client_cert)
+- [etcd_client_key](#etcd_client_key)
+- [etcd_ca](#etcd_ca)
+- [osd_etcd_client_cert](#osd_etcd_client_cert)
+- [osd_etcd_client_key](#osd_etcd_client_key)
+- [mon_etcd_client_cert](#mon_etcd_client_cert)
+- [mon_etcd_client_key](#mon_etcd_client_key)
+- [vault_url](#vault_url)
+- [vault_secret_api_path](#vault_secret_api_path)
+- [vault_client_cert](#vault_client_cert)
+- [vault_client_key](#vault_client_key)
+- [vault_ca](#vault_ca)
+- [vault_timeout_ms](#vault_timeout_ms)
+- [vault_error_timeout_sec](#vault_error_timeout_sec)
+- [vault_refresh_leeway_sec](#vault_refresh_leeway_sec)
+- [max_aes_xts_pool_size](#max_aes_xts_pool_size)
+
+## etcd_client_cert
+
+- Type: string
+
+Client TLS certificate to use for Vitastor client (not OSD and not monitor)
+etcd https connections. May be path to a file or just a PEM string with certificate.
+In the latter case, string must begin with "-----BEGIN CERTIFICATE-----".
+
+## etcd_client_key
+
+- Type: string
+
+Private key for etcd_client_cert (also a file or a PEM string).
+
+## etcd_ca
+
+- Type: string
+
+Trusted TLS CA to verify etcd server certificate. May be path to a file,
+directory or just a PEM string with certificate.
+
+## osd_etcd_client_cert
+
+- Type: string
+
+Same as [etcd_client_cert](#etcd_client_cert), but only for OSDs.
+OSDs, clients and monitors should have different permissions, so they should
+use different certificates.
+
+## osd_etcd_client_key
+
+- Type: string
+
+Same as [etcd_client_key](#etcd_client_key), but only for OSDs.
+
+## mon_etcd_client_cert
+
+- Type: string
+
+Same as [etcd_client_cert](#etcd_client_cert), but only for Vitastor monitors.
+
+## mon_etcd_client_key
+
+- Type: string
+
+Same as [etcd_client_key](#etcd_client_key), but only for Vitastor monitors.
+
+## vault_url
+
+- Type: string
+
+Vault base URL.
+
+Vitastor clients support AES-256-XTS image data encryption with different per-image keys.
+Encryption is performed by the client, OSDs don't have access to decrypted data.
+
+Encryption keys may be stored in etcd or, for the increased security level, in an external
+[HashiCorp Vault](https://developer.hashicorp.com/vault/) or [OpenBao](https://openbao.org/)
+instance.
+
+Vitastor clients use [v1 k/v secrets engine](https://openbao.org/api-docs/secret/kv/kv-v1/)
+and [TLS authentication engine](https://openbao.org/api-docs/auth/cert/) in Vault.
+
+In that case, only key IDs are stored in etcd.
+
+## vault_secret_api_path
+
+- Type: string
+- Default: /v1/secret/
+
+Vault v1 secret API mount path to use.
+
+## vault_client_cert
+
+- Type: string
+
+Client TLS certificate to use for Vault connections. Just like [etcd_client_cert](#etcd_client_cert),
+may be path to a file or just a certificate in PEM string.
+
+## vault_client_key
+
+- Type: string
+
+Private key for vault_client_cert (also a file or a PEM string).
+
+## vault_ca
+
+- Type: string
+
+Trusted TLS CA to verify Vault server certificate. May be path to a file,
+directory or just a PEM string with certificate.
+
+## vault_timeout_ms
+
+- Type: integer
+- Default: 5000
+
+Timeout for Vault requests in milliseconds.
+
+## vault_error_timeout_sec
+
+- Type: integer
+- Default: 60
+
+Time (in seconds) to wait before retrying after receiving an error from Vault.
+
+## vault_refresh_leeway_sec
+
+- Type: integer
+- Default: 60
+
+Extra time (in seconds) before real Vault token lease_timeout to refresh it, just
+in case of system clock drift.
+
+## max_aes_xts_pool_size
+
+- Type: integer
+- Default: 256
+
+Maximum number of OpenSSL encryption contexts cached in OSD memory. Probably
+doesn't require modification.

docs/config/security.ru.md

@@ -0,0 +1,154 @@
+[Документация](../../README-ru.md#документация) → [Конфигурация](../config.ru.md) → Параметры безопасности
+
+-----
+
+[Read in English](security.en.md)
+
+# Параметры безопасности
+
+Данные параметры затрагивают безопасность инсталляций Vitastor и используются
+OSD, мониторами и клиентами.
+
+Большая их часть может задаваться в /etc/vitastor/vitastor.conf и в etcd, но не
+поддерживает онлайн-изменение.
+
+- [etcd_client_cert](#etcd_client_cert)
+- [etcd_client_key](#etcd_client_key)
+- [etcd_ca](#etcd_ca)
+- [osd_etcd_client_cert](#osd_etcd_client_cert)
+- [osd_etcd_client_key](#osd_etcd_client_key)
+- [mon_etcd_client_cert](#mon_etcd_client_cert)
+- [mon_etcd_client_key](#mon_etcd_client_key)
+- [vault_url](#vault_url)
+- [vault_secret_api_path](#vault_secret_api_path)
+- [vault_client_cert](#vault_client_cert)
+- [vault_client_key](#vault_client_key)
+- [vault_ca](#vault_ca)
+- [vault_timeout_ms](#vault_timeout_ms)
+- [vault_error_timeout_sec](#vault_error_timeout_sec)
+- [vault_refresh_leeway_sec](#vault_refresh_leeway_sec)
+- [max_aes_xts_pool_size](#max_aes_xts_pool_size)
+
+## etcd_client_cert
+
+- Тип: строка
+
+Клиентский TLS сертификат для https-подключений к etcd для клиентов Vitastor
+(не OSD и не мониторов). Может быть путём к файлу или просто строкой с
+сертификатом в формате PEM. В последнем случае строка должна начинаться с
+"-----BEGIN CERTIFICATE-----".
+
+## etcd_client_key
+
+- Тип: строка
+
+Закрытый ключ для сертификата etcd_client_cert (также путь к файлу или PEM строка).
+
+## etcd_ca
+
+- Тип: строка
+
+Доверенный корневой TLS-сертификат для проверки сертификата сервера etcd.
+Может быть путём к файлу, директории или просто строкой с сертификатом в
+формате PEM.
+
+## osd_etcd_client_cert
+
+- Тип: строка
+
+Аналогично [etcd_client_cert](#etcd_client_cert), но только для OSD.
+OSD, клиенты и мониторы должны иметь разные привилегии, поэтому они должны
+использовать разные сертификаты.
+
+## osd_etcd_client_key
+
+- Тип: строка
+
+Аналогично [etcd_client_key](#etcd_client_key), но только для OSD.
+
+## mon_etcd_client_cert
+
+- Тип: строка
+
+Аналогично [etcd_client_cert](#etcd_client_cert), но только для мониторов Vitastor.
+
+## mon_etcd_client_key
+
+- Тип: строка
+
+Аналогично [etcd_client_key](#etcd_client_key), но только для мониторов Vitastor.
+
+## vault_url
+
+- Тип: строка
+
+Базовый адрес Vault.
+
+Клиенты Vitastor поддерживают AES-256-XTS шифрование данных образов с отдельными ключами на
+каждый образ. Данные шифруются клиентами, OSD не имеют доступа к незашифрованным данным.
+
+Ключи шифрования могут храниться в etcd или, для повышенного уровня безопасности, во внешнем
+[HashiCorp Vault](https://developer.hashicorp.com/vault/) или [OpenBao](https://openbao.org/).
+
+Клиенты Vitastor используют [движок секретов v1](https://openbao.org/api-docs/secret/kv/kv-v1/)
+и [TLS-аутентификацию](https://openbao.org/api-docs/auth/cert/) в Vault.
+
+В этом случае, только ID ключей хранятся в etcd.
+
+## vault_secret_api_path
+
+- Тип: строка
+- Значение по умолчанию: /v1/secret/
+
+Путь к API секретов v1 для использования клиентами.
+
+## vault_client_cert
+
+- Тип: строка
+
+Клиентский TLS сертификат для подключений к Vault. Как и [etcd_client_cert](#etcd_client_cert),
+может быть путём к файлу или просто PEM-строкой с сертификатом.
+
+## vault_client_key
+
+- Тип: строка
+
+Закрытый ключ для сертификата vault_client_cert (также путь к файлу или PEM строка).
+
+## vault_ca
+
+- Тип: строка
+
+Доверенный корневой TLS-сертификат для проверки сертификата сервера Vault.
+Может быть путём к файлу, директории или просто строкой с сертификатом в
+формате PEM.
+
+## vault_timeout_ms
+
+- Тип: целое число
+- Значение по умолчанию: 5000
+
+Максимально время выполнения Vault-запросов в миллисекундах.
+
+## vault_error_timeout_sec
+
+- Тип: целое число
+- Значение по умолчанию: 60
+
+Время (в секундах) для ожидания перед повторной попыткой при получении ошибки от Vault.
+
+## vault_refresh_leeway_sec
+
+- Тип: целое число
+- Значение по умолчанию: 60
+
+Зазор времени (в секундах), чтобы обновлять токены Vault чуть раньше их реального
+lease_timeout, на случай "ухода" системных часов.
+
+## max_aes_xts_pool_size
+
+- Тип: целое число
+- Значение по умолчанию: 256
+
+Максимальное количество кэшируемых в памяти OSD контекстов шифрования OpenSSL.
+Вряд ли требует изменения.

docs/config/src/included.en.md

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

docs/config/src/included.ru.md

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

docs/config/src/layout-osd.yml

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

docs/config/src/package.json

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

docs/config/src/security.en.md

@@ -0,0 +1,5 @@
+# Security Parameters
+
+These parameters affect your Vitastor installation security and apply to OSDs, monitors and clients.
+
+Most of them can be set in /etc/vitastor/vitastor.conf and in etcd, but don't support online modification.

docs/config/src/security.ru.md

@@ -0,0 +1,7 @@
+# Параметры безопасности
+
+Данные параметры затрагивают безопасность инсталляций Vitastor и используются
+OSD, мониторами и клиентами.
+
+Большая их часть может задаваться в /etc/vitastor/vitastor.conf и в etcd, но не
+поддерживает онлайн-изменение.

docs/config/src/security.yml

@@ -0,0 +1,131 @@
+- name: etcd_client_cert
+  type: string
+  info: |
+    Client TLS certificate to use for Vitastor client (not OSD and not monitor)
+    etcd https connections. May be path to a file or just a PEM string with certificate.
+    In the latter case, string must begin with "-----BEGIN CERTIFICATE-----".
+  info_ru: |
+    Клиентский TLS сертификат для https-подключений к etcd для клиентов Vitastor
+    (не OSD и не мониторов). Может быть путём к файлу или просто строкой с
+    сертификатом в формате PEM. В последнем случае строка должна начинаться с
+    "-----BEGIN CERTIFICATE-----".
+- name: etcd_client_key
+  type: string
+  info: Private key for etcd_client_cert (also a file or a PEM string).
+  info_ru: Закрытый ключ для сертификата etcd_client_cert (также путь к файлу или PEM строка).
+- name: etcd_ca
+  type: string
+  info: |
+    Trusted TLS CA to verify etcd server certificate. May be path to a file,
+    directory or just a PEM string with certificate.
+  info_ru: |
+    Доверенный корневой TLS-сертификат для проверки сертификата сервера etcd.
+    Может быть путём к файлу, директории или просто строкой с сертификатом в
+    формате PEM.
+- name: osd_etcd_client_cert
+  type: string
+  info: |
+    Same as [etcd_client_cert](#etcd_client_cert), but only for OSDs.
+    OSDs, clients and monitors should have different permissions, so they should
+    use different certificates.
+  info_ru: |
+    Аналогично [etcd_client_cert](#etcd_client_cert), но только для OSD.
+    OSD, клиенты и мониторы должны иметь разные привилегии, поэтому они должны
+    использовать разные сертификаты.
+- name: osd_etcd_client_key
+  type: string
+  info: Same as [etcd_client_key](#etcd_client_key), but only for OSDs.
+  info_ru: Аналогично [etcd_client_key](#etcd_client_key), но только для OSD.
+- name: mon_etcd_client_cert
+  type: string
+  info: Same as [etcd_client_cert](#etcd_client_cert), but only for Vitastor monitors.
+  info_ru: Аналогично [etcd_client_cert](#etcd_client_cert), но только для мониторов Vitastor.
+- name: mon_etcd_client_key
+  type: string
+  info: Same as [etcd_client_key](#etcd_client_key), but only for Vitastor monitors.
+  info_ru: Аналогично [etcd_client_key](#etcd_client_key), но только для мониторов Vitastor.
+- name: vault_url
+  type: string
+  info: |
+    Vault base URL.
+
+    Vitastor clients support AES-256-XTS image data encryption with different per-image keys.
+    Encryption is performed by the client, OSDs don't have access to decrypted data.
+
+    Encryption keys may be stored in etcd or, for the increased security level, in an external
+    [HashiCorp Vault](https://developer.hashicorp.com/vault/) or [OpenBao](https://openbao.org/)
+    instance.
+
+    Vitastor clients use [v1 k/v secrets engine](https://openbao.org/api-docs/secret/kv/kv-v1/)
+    and [TLS authentication engine](https://openbao.org/api-docs/auth/cert/) in Vault.
+
+    In that case, only key IDs are stored in etcd.
+  info_ru: |
+    Базовый адрес Vault.
+
+    Клиенты Vitastor поддерживают AES-256-XTS шифрование данных образов с отдельными ключами на
+    каждый образ. Данные шифруются клиентами, OSD не имеют доступа к незашифрованным данным.
+
+    Ключи шифрования могут храниться в etcd или, для повышенного уровня безопасности, во внешнем
+    [HashiCorp Vault](https://developer.hashicorp.com/vault/) или [OpenBao](https://openbao.org/).
+
+    Клиенты Vitastor используют [движок секретов v1](https://openbao.org/api-docs/secret/kv/kv-v1/)
+    и [TLS-аутентификацию](https://openbao.org/api-docs/auth/cert/) в Vault.
+
+    В этом случае, только ID ключей хранятся в etcd.
+- name: vault_secret_api_path
+  type: string
+  default: /v1/secret/
+  info: Vault v1 secret API mount path to use.
+  info_ru: Путь к API секретов v1 для использования клиентами.
+- name: vault_client_cert
+  type: string
+  info: |
+    Client TLS certificate to use for Vault connections. Just like [etcd_client_cert](#etcd_client_cert),
+    may be path to a file or just a certificate in PEM string.
+  info_ru: |
+    Клиентский TLS сертификат для подключений к Vault. Как и [etcd_client_cert](#etcd_client_cert),
+    может быть путём к файлу или просто PEM-строкой с сертификатом.
+- name: vault_client_key
+  type: string
+  info: Private key for vault_client_cert (also a file or a PEM string).
+  info_ru: Закрытый ключ для сертификата vault_client_cert (также путь к файлу или PEM строка).
+- name: vault_ca
+  type: string
+  info: |
+    Trusted TLS CA to verify Vault server certificate. May be path to a file,
+    directory or just a PEM string with certificate.
+  info_ru: |
+    Доверенный корневой TLS-сертификат для проверки сертификата сервера Vault.
+    Может быть путём к файлу, директории или просто строкой с сертификатом в
+    формате PEM.
+- name: vault_timeout_ms
+  type: int
+  default: 5000
+  info: Timeout for Vault requests in milliseconds.
+  info_ru: Максимально время выполнения Vault-запросов в миллисекундах.
+- name: vault_error_timeout_sec
+  type: int
+  default: 60
+  info: |
+    Time (in seconds) to wait before retrying after receiving an error from Vault.
+  info_ru: |
+    Время (в секундах) для ожидания перед повторной попыткой при получении ошибки от Vault.
+- name: vault_refresh_leeway_sec
+  type: int
+  default: 60
+  info: |
+    Extra time (in seconds) before real Vault token lease_timeout to refresh it, just
+    in case of system clock drift.
+  info_ru: |
+    Зазор времени (в секундах), чтобы обновлять токены Vault чуть раньше их реального
+    lease_timeout, на случай "ухода" системных часов.
+- name: max_aes_xts_pool_size
+  type: int
+  default: 256
+  info: |
+    Maximum number of OpenSSL encryption contexts cached in OSD memory. Probably
+    doesn't require modification.
+  info_ru: |
+    Максимальное количество кэшируемых в памяти OSD контекстов шифрования OpenSSL.
+    Вряд ли требует изменения.

docs/installation/source.en.md

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

docs/installation/source.ru.md

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

docs/intro/features.en.md

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

docs/intro/features.ru.md

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

docs/usage/cli.en.md

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

docs/usage/cli.ru.md

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

mon/antietcd_adapter.js

@@ -18,7 +18,7 @@ class AntiEtcdAdapter
                 cluster = cluster ? (''+(cluster||'')).split(/,+/) : [];
             cluster = Object.keys(cluster.reduce((a, url) =>
             {
-                a[url.toLowerCase().replace(/^(https?:\/\/)/, '').replace(/\/.*$/, '')] = true;
+                a[url.toLowerCase().replace(/^(https?:\/\/)?(.*?)(\/.*)?$/, (m, m1, m2) => (m1||'http://')+m2)] = true;
                 return a;
             }, {}));
             const cfg_port = config.antietcd_port;
@@ -26,7 +26,8 @@ class AntiEtcdAdapter
             is_local['0.0.0.0'] = true;
             is_local['::'] = true;
             is_local[''] = true;
-            const selected = cluster.map(s => s.split(':', 2)).filter(ip => is_local[ip[0]] && (!cfg_port || ip[1] == cfg_port));
+            // split :, 3 -> <schema>:<//ip>:<port>
+            const selected = cluster.map(s => s.split(':', 3)).filter(ip => is_local[ip[1].substr(2)] && (!cfg_port || ip[2] == cfg_port));
             if (selected.length > 1)
             {
                 console.error('More than 1 etcd_address matches local IPs, please specify port');
@@ -35,16 +36,30 @@ class AntiEtcdAdapter
             else if (selected.length == 1)
             {
                 const antietcd_config = {
-                    ip: selected[0][0],
+                    ip: selected[0][1].substr(2),
-                    port: selected[0][1],
+                    port: selected[0][2],
-                    data: config.antietcd_data_file || ((config.antietcd_data_dir || '/var/lib/vitastor') + '/mon_'+selected[0][1]+'.json.gz'),
+                    cert: config.antietcd_cert,
+                    key: config.antietcd_key,
+                    ca: config.etcd_ca,
+                    data: config.antietcd_data_file || ((config.antietcd_data_dir || '/var/lib/vitastor') + '/mon_'+selected[0][2]+'.json.gz'),
                     persist_filter: vitastor_persist_filter({ vitastor_prefix: config.etcd_prefix || '/vitastor' }),
-                    node_id: selected[0][0]+':'+selected[0][1], // node_id = ip:port
+                    node_id: selected[0][1].substr(2)+':'+selected[0][2], // node_id = ip:port
-                    cluster: (cluster.length == 1 ? null : cluster.reduce((a, c) => { a[c] = "http://"+c; return a; }, {})),
+                    cluster: (cluster.length == 1 ? null : cluster.reduce((a, c) => { a[c.replace(/^(https?:\/\/)/, '')] = c; return a; }, {})),
                     cluster_key: (config.etcd_prefix || '/vitastor'),
                     stale_read: 1,
                     log_level: 1,
                 };
+                if (config.use_auth)
+                {
+                    antietcd_config.client_cert_auth = true;
+                    antietcd_config.auth_filter = require('./vitastor_auth_filter.js');
+                    antietcd_config.peer_ca = config.antietcd_server_ca;
+                    if (!config.antietcd_server_ca || config.antietcd_server_ca == config.etcd_ca)
+                    {
+                        console.error('Secure setup requires separate antietcd_server_ca (for signing antietcd server certificates) and etcd_ca (for signing client certificates)');
+                        process.exit(1);
+                    }
+                }
                 for (const key in config)
                 {
                     if (key.substr(0, 9) === 'antietcd_')
@@ -169,7 +184,7 @@ class AntiEtcdAdapter
                     await new Promise(ok => setTimeout(ok, timeout-(Date.now()-prev)));
                 }
                 prev = Date.now();
-                const res = await this.antietcd.api(path.replace(/^\/+/, '').replace(/\/+$/, '').replace(/\/+/g, '_'), body);
+                const res = await this.antietcd.api(path.replace(/^\/+/, '').replace(/\/+$/, '').replace(/\/+/g, '_'), body, { username: 'root' });
                 if (res.error)
                 {
                     console.error('Failed to query antietcd '+path+' (retry '+retry+'/'+retries+'): '+res.error);

mon/etcd_adapter.js

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

mon/etcd_schema.js

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

mon/vitastor_auth_filter.js

@@ -0,0 +1,471 @@
+// AntiEtcd authentication filter for Vitastor
+// (c) Vitaliy Filippov, 2026
+// License: Mozilla Public License 2.0 or Vitastor Network Public License 1.1
+
+// Permissions are based on:
+// 1. Users.
+//    Stored in /vitastor/config/user/<username>.
+//    Has 2 properties:
+//    - type, one of: osd, mon, admin, client.
+//      osd, mon types should be used by OSDs/monitors.
+//      admin should be used for administrative access from vitastor-cli.
+//      client should be used for regular clients.
+//    - groups, a list of group names the user is included in.
+// 2. Images.
+//    Stored in /vitastor/config/inode/<pool>/<inode>. Has the following properties:
+//    - owner (user name)
+//    - owner_group (group name)
+//    - reader_group
+
+const static_perms = {
+    invalid: {
+        keys: {},
+        prefixes: {},
+    },
+    osd: {
+        keys: { '/pg/config': false },
+        prefixes: { '/osd/': true, '/pg/state/': true, '/pg/history/': true, '/pgstats/': true },
+    },
+    mon: {
+        keys: { '/pg/config': true, '/stats': true, '/history/last_clean_pgs': true },
+        prefixes: {
+            '/config/': false, '/osd/': false, '/mon/': true, '/pg/history/': true,
+            '/pgstats/': false, '/inode/stats/': true, '/pool/stats/': true,
+        },
+    },
+    admin: {
+        keys: { '/stats': false },
+        prefixes: {
+            '/config/': true, '/osd/': true, '/index/': true, '/pg/history/': true,
+            '/mon/': false, '/pg/': false, '/pgstats/': false, '/inode/stats/': false, '/pool/stats/': false,
+        },
+    },
+    client: {
+        keys: { '/config/global': false, '/config/node_placement': false, '/config/pools': false, '/pg/config': false },
+        prefixes: { '/osd/stats/': false, '/pg/state/': false, '/index/maxid/': false },
+    },
+};
+
+const api_perms = {
+    osd: { lease_grant: true, lease_revoke: true, lease_keepalive: true },
+    mon: { lease_grant: true, lease_revoke: true, lease_keepalive: true },
+    admin: { maintenance_status: true },
+    client: {},
+};
+
+class VitastorAuthFilter
+{
+    constructor(antietcd)
+    {
+        this.cfg = antietcd.cfg;
+        this.antietcd = antietcd;
+        this.prefix = this.cfg.vitastor_prefix || '/vitastor';
+        this.prefix_parts = this.prefix.split('/');
+    }
+
+    _get(path, decode)
+    {
+        let cur = this.antietcd.etctree.state;
+        path = path instanceof Array ? path : path.split('/');
+        for (const p of path)
+        {
+            if (!cur.children)
+            {
+                return null;
+            }
+            cur = cur.children[p];
+            if (!cur)
+            {
+                return null;
+            }
+        }
+        if (decode)
+        {
+            return this._decode(path, cur.value);
+        }
+        return cur;
+    }
+
+    _decode(path, cur)
+    {
+        if (!cur)
+        {
+            return null;
+        }
+        if (cur)
+        {
+            try
+            {
+                cur = JSON.parse(cur);
+            }
+            catch (e)
+            {
+                console.warn('Invalid JSON in '+(path instanceof Array ? path.join('/') : path)+': '+e);
+            }
+        }
+        return cur;
+    }
+
+    // userInfo: { name: string, type: string, perms: static_perms[type], groups: { [string]: true } }
+    _check_compare(check, userInfo, checked)
+    {
+        let key = String(check.key);
+        if (key.substr(0, this.prefix.length) !== this.prefix)
+        {
+            return false;
+        }
+        key = key.substr(this.prefix.length);
+        if (key in userInfo.perms.keys)
+        {
+            return true;
+        }
+        for (const pfx in userInfo.perms.prefixes)
+        {
+            if (key.substr(0, pfx.length) == pfx)
+            {
+                return true;
+            }
+        }
+        if (userInfo.type == 'client')
+        {
+            // Image permissions
+            if (key.substr(0, 14) == '/config/inode/')
+            {
+                // Allowed to check that a key does not exist
+                if (check.target == 'VERSION' && check.version == 0)
+                {
+                    checked['M'+key] = true;
+                    return true;
+                }
+                else if (check.target == 'MOD')
+                {
+                    const data = this._get(check.key);
+                    if (!data || data.mod_revision != check.mod_revision)
+                    {
+                        // Break check to trigger CAS failure
+                        check.mod_revision = '18446744073709551615'; // UINT64_MAX
+                        return true;
+                    }
+                    const inode = this._decode(check.key, data.value);
+                    if (inode && (inode.owner_group && userInfo.groups[inode.owner_group] ||
+                        inode.owner === userInfo.name))
+                    {
+                        checked['M'+key] = true;
+                        return true;
+                    }
+                }
+                return false;
+            }
+            if (key.substr(0, 13) == '/index/image/')
+            {
+                // Allowed to check that a key does not exist
+                if (check.target == 'VERSION' && check.version == 0)
+                {
+                    checked['M'+key] = true;
+                    return true;
+                }
+                else if (check.target == 'MOD')
+                {
+                    let data = this._get(check.key);
+                    if (!data || data.mod_revision != check.mod_revision)
+                    {
+                        // Break check to trigger CAS failure
+                        check.mod_revision = '18446744073709551615'; // UINT64_MAX
+                        return true;
+                    }
+                    data = this._decode(check.key, data.value);
+                    if (data)
+                    {
+                        const inode = this._get([ ...this.prefix_parts, 'config', 'inode', data.pool_id, data.id ], true);
+                        if (inode && (inode.owner_group && userInfo.groups[inode.owner_group] ||
+                            inode.owner === userInfo.name))
+                        {
+                            checked['M'+key] = true;
+                            return true;
+                        }
+                    }
+                }
+                return false;
+            }
+            if (key.substr(0, 13) == '/index/maxid/')
+            {
+                const pool_id = key.substr(13);
+                const pool_cfg = this._get([ ...this.prefix_parts, 'config', 'pools' ], true);
+                if (!pool_cfg || !pool_cfg[pool_id] || !pool_cfg[pool_id].creator_group || !userInfo.groups[pool_cfg[pool_id].creator_group])
+                {
+                    return false;
+                }
+                if (check.target == 'VERSION' && check.version == 0)
+                {
+                    checked['I'+parseInt(key.substr(13))+'_0'] = true;
+                    return true;
+                }
+                else if (check.target == 'MOD')
+                {
+                    const data = this._get(check.key);
+                    if (!data || data.mod_revision != check.mod_revision)
+                    {
+                        // Break check to trigger CAS failure
+                        check.mod_revision = '18446744073709551615'; // UINT64_MAX
+                        return true;
+                    }
+                    checked['I'+parseInt(key.substr(13))+'_'+data.value] = true;
+                    return true;
+                }
+                return false;
+            }
+        }
+        return false;
+    }
+
+    _check_read(kv, userInfo)
+    {
+        let key = String(kv.key);
+        if (key.substr(0, this.prefix.length) !== this.prefix)
+        {
+            return false;
+        }
+        key = key.substr(this.prefix.length);
+        if (key in userInfo.perms.keys)
+        {
+            return true;
+        }
+        for (const pfx in userInfo.perms.prefixes)
+        {
+            if (key.substr(0, pfx.length) == pfx)
+            {
+                return true;
+            }
+        }
+        if (userInfo.type == 'client')
+        {
+            // Image permissions
+            if (key.substr(0, 14) == '/config/inode/')
+            {
+                const inode = this._decode(kv.key, kv.value);
+                if (inode && (inode.reader_group && userInfo.groups[inode.reader_group] ||
+                    inode.owner_group && userInfo.groups[inode.owner_group] ||
+                    inode.owner === userInfo.name))
+                {
+                    return true;
+                }
+                return false;
+            }
+            if (key.substr(0, 13) == '/index/image/')
+            {
+                const data = this._decode(kv.key, kv.value);
+                const inode = this._get([ ...this.prefix_parts, 'config', 'inode', data.pool_id, data.id ], true);
+                if (inode && (inode.reader_group && userInfo.groups[inode.reader_group] ||
+                    inode.owner_group && userInfo.groups[inode.owner_group] ||
+                    inode.owner === userInfo.name))
+                {
+                    return true;
+                }
+                return false;
+            }
+        }
+        return false;
+    }
+
+    _check_write(put, userInfo, checked)
+    {
+        let key = String(put.key);
+        if (key.substr(0, this.prefix.length) !== this.prefix)
+        {
+            return false;
+        }
+        key = key.substr(this.prefix.length);
+        if (userInfo.perms.keys[key])
+        {
+            return true;
+        }
+        for (const pfx in userInfo.perms.prefixes)
+        {
+            if (userInfo.perms.prefixes[pfx] && key.substr(0, pfx.length) == pfx)
+            {
+                return true;
+            }
+        }
+        if (checked && userInfo.type == 'client')
+        {
+            if (key.substr(0, 13) == '/index/maxid/' &&
+                checked['I'+parseInt(key.substr(13))+'_'+(put.value-1)])
+            {
+                // Allowed to increment maxid
+                return true;
+            }
+            if (checked['M'+key])
+            {
+                // Allowed to modify known images with CAS checks
+                return true;
+            }
+        }
+        return false;
+    }
+
+    _check_req(req, userInfo, checked)
+    {
+        let r;
+        if ((r = (req.request_range || req.requestRange)))
+        {
+            // All range queries are allowed, but responses are filtered - it's simpler
+        }
+        else if ((r = (req.request_put || req.requestPut)))
+        {
+            if (!this._check_write(r, userInfo, checked))
+                return false;
+        }
+        else if ((r = (req.request_delete_range || req.requestDeleteRange)))
+        {
+            if (!r.range_end || r.range_end === r.key)
+            {
+                if (!this._check_write({ key: r.key }, userInfo))
+                    return false;
+            }
+            else
+            {
+                // All keys in range must satisfy prefix
+                r.range_end = String(r.range_end);
+                if (r.key.length != r.range_end.length ||
+                    r.key[r.key.length-1] != '/' ||
+                    r.range_end[r.range_end.length-1] != '0')
+                {
+                    return false;
+                }
+                let key = r.key.substr(this.prefix.length);
+                let found = false;
+                for (const pfx in userInfo.perms.prefixes)
+                {
+                    if (userInfo.perms.prefixes[pfx] && key.substr(0, pfx.length) == pfx)
+                    {
+                        found = true;
+                        break;
+                    }
+                }
+                if (!found)
+                    return false;
+            }
+        }
+        return true;
+    }
+
+    _get_user(username)
+    {
+        if (!username)
+        {
+            return null;
+        }
+        let userInfo = this._get([ ...this.prefix_parts, 'config', 'user', username ], true);
+        if (!userInfo)
+        {
+            userInfo = { type: 'client' };
+        }
+        userInfo.perms = static_perms[userInfo.type] || static_perms['invalid'];
+        userInfo.name = username;
+        if (userInfo.groups instanceof Array)
+        {
+            userInfo.groups = userInfo.groups.reduce((a, c) => { a[c] = true; return a; }, {});
+        }
+        else
+        {
+            userInfo.groups = {};
+        }
+        return userInfo;
+    }
+
+    filter_api(username, api/*, data*/)
+    {
+        if (username === 'root')
+        {
+            return true;
+        }
+        const userInfo = this._get([ ...this.prefix_parts, 'config', 'user', username ], true);
+        return userInfo && api_perms[userInfo.type] && api_perms[userInfo.type][api];
+    }
+
+    filter_txn(username, txn)
+    {
+        if (username === 'root')
+        {
+            return true;
+        }
+        const userInfo = this._get_user(username);
+        if (!userInfo)
+        {
+            return null;
+        }
+        const checked = {};
+        if (txn.compare)
+        {
+            for (const check of txn.compare)
+            {
+                if (!this._check_compare(check, userInfo, checked))
+                    return null;
+            }
+        }
+        // Special transactions:
+        // 1. create image: create config/inode and index/image, increment index/maxid/<pool> (with CAS)
+        // 2. create snapshot: same as create image but also rename previous to @snap
+        if (txn.success)
+        {
+            for (const req of txn.success)
+            {
+                if (!this._check_req(req, userInfo, checked))
+                    return null;
+            }
+        }
+        if (txn.failure)
+        {
+            for (const req of txn.failure)
+            {
+                if (!this._check_req(req, userInfo, null))
+                    return null;
+            }
+        }
+        return txn;
+    }
+
+    filter_txn_response(username, txn, res)
+    {
+        if (!res.responses || username === 'root')
+        {
+            return;
+        }
+        const userInfo = this._get_user(username);
+        if (!userInfo)
+        {
+            for (const resp of res.responses)
+            {
+                if (resp.response_range && resp.response_range.kvs)
+                {
+                    resp.response_range.kvs = [];
+                }
+            }
+            return;
+        }
+        for (const resp of res.responses)
+        {
+            if (resp.response_range && resp.response_range.kvs)
+            {
+                resp.response_range.kvs = resp.response_range.kvs.filter(kv => this._check_read(kv, userInfo));
+            }
+        }
+    }
+
+    filter_watch_message(username, msg)
+    {
+        if (!msg.result || !msg.result.events || username === 'root')
+        {
+            return;
+        }
+        const userInfo = this._get_user(username);
+        if (!userInfo)
+        {
+            msg.result.events = [];
+            return;
+        }
+        msg.result.events = msg.result.events.filter(ev => this._check_read(ev.kv, userInfo));
+    }
+}
+
+module.exports = VitastorAuthFilter;

rpm/vitastor-el10.Dockerfile

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

rpm/vitastor-el10.spec

@@ -16,6 +16,7 @@ BuildRequires:  gf-complete-devel
 BuildRequires:  rdma-core-devel
 BuildRequires:  cmake
 BuildRequires:  libnl3-devel
+BuildRequires:  c-ares-devel
 Requires:       vitastor-osd = %{version}-%{release}
 Requires:       vitastor-mon = %{version}-%{release}
 Requires:       vitastor-client = %{version}-%{release}

rpm/vitastor-el7.Dockerfile

@@ -15,7 +15,7 @@ RUN yum -y --enablerepo=extras install centos-release-scl epel-release yum-utils
 RUN perl -i -pe 's!mirrorlist=!#mirrorlist=!s; s!#\s*baseurl=http://mirror.centos.org!baseurl=http://vault.centos.org!' /etc/yum.repos.d/CentOS-SCLo-scl*.repo
 RUN yum -y install https://vitastor.io/rpms/centos/7/vitastor-release-1.0-1.el7.noarch.rpm
 RUN yum -y install devtoolset-9-gcc-c++ devtoolset-9-libatomic-devel gcc make cmake gperftools-devel \
-    fio rh-nodejs12 jerasure-devel libisa-l-devel gf-complete-devel rdma-core-devel libnl3-devel
+    fio rh-nodejs12 jerasure-devel libisa-l-devel gf-complete-devel rdma-core-devel libnl3-devel c-ares-devel
 RUN yumdownloader --disablerepo=centos-sclo-rh --source fio
 RUN rpm --nomd5 -i fio*.src.rpm
 RUN rm -f /etc/yum.repos.d/CentOS-Media.repo

rpm/vitastor-el7.spec

@@ -17,6 +17,7 @@ BuildRequires:  gf-complete-devel
 BuildRequires:  rdma-core-devel
 BuildRequires:  cmake3
 BuildRequires:  libnl3-devel
+BuildRequires:  c-ares-devel
 Requires:       vitastor-osd = %{version}-%{release}
 Requires:       vitastor-mon = %{version}-%{release}
 Requires:       vitastor-client = %{version}-%{release}

rpm/vitastor-el8.Dockerfile

@@ -13,7 +13,7 @@ RUN dnf -y install centos-release-advanced-virtualization epel-release dnf-plugi
 RUN sed -i 's/^mirrorlist=/#mirrorlist=/; s!#baseurl=.*!baseurl=http://vault.centos.org/centos/8.4.2105/virt/$basearch/$avdir/!; s!^baseurl=.*Source/.*!baseurl=http://vault.centos.org/centos/8.4.2105/virt/Source/advanced-virtualization/!' /etc/yum.repos.d/CentOS-Advanced-Virtualization.repo
 RUN yum -y install https://vitastor.io/rpms/centos/8/vitastor-release-1.0-1.el8.noarch.rpm
 RUN dnf -y install gcc-toolset-9 gcc-toolset-9-gcc-c++ gperftools-devel \
-    fio nodejs rpm-build jerasure-devel libisa-l-devel gf-complete-devel libibverbs-devel libarchive cmake libnl3-devel
+    fio nodejs rpm-build jerasure-devel libisa-l-devel gf-complete-devel libibverbs-devel libarchive cmake libnl3-devel c-ares-devel
 RUN dnf download --source fio
 RUN rpm --nomd5 -i fio*.src.rpm
 RUN cd ~/rpmbuild/SPECS && dnf builddep -y --enablerepo=powertools --spec fio.spec

rpm/vitastor-el8.spec

@@ -16,6 +16,7 @@ BuildRequires:  gf-complete-devel
 BuildRequires:  rdma-core-devel
 BuildRequires:  cmake
 BuildRequires:  libnl3-devel
+BuildRequires:  c-ares-devel
 Requires:       vitastor-osd = %{version}-%{release}
 Requires:       vitastor-mon = %{version}-%{release}
 Requires:       vitastor-client = %{version}-%{release}

rpm/vitastor-el9.Dockerfile

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

rpm/vitastor-el9.spec

@@ -16,6 +16,7 @@ BuildRequires:  gf-complete-devel
 BuildRequires:  rdma-core-devel
 BuildRequires:  cmake
 BuildRequires:  libnl3-devel
+BuildRequires:  c-ares-devel
 Requires:       vitastor-osd = %{version}-%{release}
 Requires:       vitastor-mon = %{version}-%{release}
 Requires:       vitastor-client = %{version}-%{release}

src/CMakeLists.txt

@@ -74,6 +74,14 @@ if (RDMACM_LIBRARIES)
 	add_definitions(-DWITH_RDMACM)
 endif (RDMACM_LIBRARIES)
 
+find_package(OpenSSL REQUIRED)
+if (OPENSSL_FOUND)
+	add_definitions(-DWITH_OPENSSL)
+endif (OPENSSL_FOUND)
+
+pkg_check_modules(CARES REQUIRED libcares)
+include_directories(${CARES_INCLUDE_DIRS})
+
 if (${WITH_SYSTEM_LIBURING})
 	pkg_check_modules(LIBURING REQUIRED liburing>=2.10)
 	include_directories(${LIBURING_INCLUDE_DIRS})

src/blockstore/blockstore_disk.cpp

@@ -83,13 +83,17 @@ void blockstore_disk_t::parse_config(std::map<std::string, std::string> & config
     {
         data_csum_type = BLOCKSTORE_CSUM_CRC32C;
     }
+    else if (config["data_csum_type"] == "xxh3_32")
+    {
+        data_csum_type = BLOCKSTORE_CSUM_XXH3_32;
+    }
     else if (config["data_csum_type"] == "" || config["data_csum_type"] == "none")
     {
         data_csum_type = BLOCKSTORE_CSUM_NONE;
     }
     else
     {
-        throw std::runtime_error("data_csum_type="+config["data_csum_type"]+" is unsupported, only \"crc32c\" and \"none\" are supported");
+        throw std::runtime_error("data_csum_type="+config["data_csum_type"]+" is unsupported, only \"crc32c\", \"xxh3_32\" and \"none\" are supported");
     }
     csum_block_size = parse_size(config["csum_block_size"]);
     discard_on_start = config.find("discard_on_start") != config.end() &&

src/blockstore/blockstore_disk.h

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

src/blockstore/blockstore_heap.cpp

@@ -12,6 +12,7 @@
 #include "blockstore_heap.h"
 #include "../util/allocator.h"
 #include "../util/crc32c.h"
+#include "../util/xxh_x86dispatch.h"
 #include "../util/malloc_or_die.h"
 
 #define BS_HEAP_FREE_MVCC 1
@@ -64,19 +65,19 @@ uint32_t blockstore_heap_t::get_simple_entry_size()
 uint32_t blockstore_heap_t::get_big_entry_size()
 {
     return sizeof(heap_big_write_t) + dsk->clean_entry_bitmap_size*2 +
-        (!dsk->data_csum_type ? 0 : dsk->data_block_size/dsk->csum_block_size * (dsk->data_csum_type & 0xFF));
+        (!dsk->csum_block_size ? 0 : dsk->data_block_size/dsk->csum_block_size * (dsk->data_csum_type & 0xFF));
 }
 
 uint32_t blockstore_heap_t::get_big_intent_entry_size()
 {
     return sizeof(heap_big_intent_t) + dsk->clean_entry_bitmap_size*2 +
-        (!dsk->data_csum_type ? 4 : dsk->data_block_size/dsk->csum_block_size * (dsk->data_csum_type & 0xFF));
+        (!dsk->csum_block_size ? 4 : dsk->data_block_size/dsk->csum_block_size * (dsk->data_csum_type & 0xFF));
 }
 
 uint32_t blockstore_heap_t::get_small_entry_size(uint32_t offset, uint32_t len)
 {
     return sizeof(heap_small_write_t) + dsk->clean_entry_bitmap_size +
-        (!dsk->data_csum_type ? 4 : (dsk->data_csum_type & 0xFF) *
+        (!dsk->csum_block_size ? 4 : (dsk->data_csum_type & 0xFF) *
             ((offset+len+dsk->csum_block_size-1)/dsk->csum_block_size - offset/dsk->csum_block_size));
 }
 
@@ -91,7 +92,7 @@ uint32_t blockstore_heap_t::get_csum_size(heap_entry_t *wr)
 
 uint32_t blockstore_heap_t::get_csum_size(uint32_t entry_type, uint32_t offset, uint32_t len)
 {
-    if (!dsk->data_csum_type)
+    if (!dsk->csum_block_size)
     {
         return 0;
     }
@@ -214,15 +215,24 @@ void heap_entry_t::set_big_location(blockstore_heap_t *heap, uint64_t location)
     big().block_num = location / heap->dsk->data_block_size;
 }
 
-uint32_t heap_entry_t::calc_crc32c()
+uint32_t heap_entry_t::calc_checksum(blockstore_disk_t *dsk)
 {
-    auto old_crc32c = crc32c;
+    auto old_checksum = checksum;
-    crc32c = 0;
+    checksum = 0;
-    uint32_t res = ::crc32c(0, (uint8_t*)this, size);
+    uint32_t res = 0;
-    crc32c = old_crc32c;
+    if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32)
+        res = (uint32_t)XXH3_64bits(this, size);
+    else
+        res = ::crc32c(0, (uint8_t*)this, size);
+    checksum = old_checksum;
     return res;
 }
 
+uint32_t heap_entry_t::calc_checksum(blockstore_heap_t *heap)
+{
+    return calc_checksum(heap->dsk);
+}
+
 uint64_t blockstore_heap_t::get_pg_id(inode_t inode, uint64_t stripe)
 {
     uint64_t pg_num = 0;
@@ -387,12 +397,12 @@ corrupted_object:
                 goto corrupted_object;
             }
             // Verify crc
-            uint32_t expected_crc32c = wr->calc_crc32c();
+            uint32_t expected_checksum = wr->calc_checksum(this);
-            if (wr->crc32c != expected_crc32c)
+            if (wr->checksum != expected_checksum)
             {
-                fprintf(stderr, "Error: entry %jx:%jx v%ju l%ju in metadata block %u at %u is corrupt (crc32c mismatch: expected %08x, got %08x). ",
+                fprintf(stderr, "Error: entry %jx:%jx v%ju l%ju in metadata block %u at %u is corrupt (checksum mismatch: expected %08x, got %08x). ",
                     wr->inode, wr->stripe, wr->version, wr->lsn,
-                    block_num, block_offset, expected_crc32c, wr->crc32c);
+                    block_num, block_offset, expected_checksum, wr->checksum);
                 goto corrupted_object;
             }
             // Verify offset & len
@@ -941,7 +951,11 @@ bool blockstore_heap_t::calc_checksums(heap_entry_t *wr, uint8_t *data, bool set
             len = wr->big_intent().len;
         else
             assert(0);
-        uint32_t real_csum = crc32c(0, data, len);
+        uint32_t real_csum = 0;
+        if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32)
+            real_csum = (uint32_t)XXH3_64bits(data, len);
+        else
+            real_csum = crc32c(0, data, len);
         if (set)
         {
             *wr_csum = real_csum;
@@ -991,11 +1005,26 @@ static uint32_t crc32c_iter(uint32_t prev_crc, const std::function<uint8_t*(uint
     return prev_crc;
 }
 
+static void xxh3_iter(XXH3_state_t* xxh3_state, const std::function<uint8_t*(uint32_t start, uint32_t & len)> & next, uint32_t pos, uint32_t size)
+{
+    uint32_t cur_len = 0;
+    while (size > 0)
+    {
+        uint8_t *data = next(pos, cur_len);
+        assert(data);
+        cur_len = (cur_len < size ? cur_len : size);
+        XXH3_64bits_update(xxh3_state, data, cur_len);
+        pos += cur_len;
+        size -= cur_len;
+    }
+}
+
 bool blockstore_heap_t::calc_block_checksums(uint32_t *block_csums, uint8_t *bitmap,
     uint32_t start, uint32_t end, std::function<uint8_t*(uint32_t start, uint32_t & len)> next,
     bool set, std::function<void(uint32_t, uint32_t, uint32_t)> bad_block_cb)
 {
     bool res = true;
+    XXH3_state_t* xxh3_state = NULL;
     uint32_t pos = start;
     uint32_t block_end = (start/dsk->csum_block_size + 1)*dsk->csum_block_size;
     uint32_t block_crc = 0;
@@ -1012,42 +1041,89 @@ bool blockstore_heap_t::calc_block_checksums(uint32_t *block_csums, uint8_t *bit
                     pos += dsk->bitmap_granularity;
                 // zero padding at the beginning or at the end of the block is not counted
                 if (pos > prev && prev > 0 && pos < block_end)
-                    block_crc = crc32c_pad(block_crc, NULL, 0, pos-prev, 0);
+                {
+                    if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32)
+                    {
+                        if (!xxh3_state)
+                        {
+                            xxh3_state = XXH3_createState();
+                            XXH3_64bits_reset(xxh3_state);
+                        }
+                        uint32_t zeropad = pos-prev;
+                        while (zeropad > 0)
+                        {
+                            uint32_t zerolen = zeropad > 4096 ? 4096 : zeropad;
+                            XXH3_64bits_update(xxh3_state, zero_page, zerolen);
+                            zeropad -= zerolen;
+                        }
+                    }
+                    else
+                        block_crc = crc32c_pad(block_crc, NULL, 0, pos-prev, 0);
+                }
                 prev = pos;
                 while (pos < end && pos < block_end && (bitmap[pos/dsk->bitmap_granularity/8] & (1 << ((pos/dsk->bitmap_granularity) % 8))))
                     pos += dsk->bitmap_granularity;
                 if (pos > prev)
                 {
                     isset = true;
-                    block_crc = crc32c_iter(block_crc, next, prev, pos-prev);
+                    if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32)
+                    {
+                        if (!xxh3_state)
+                        {
+                            xxh3_state = XXH3_createState();
+                            XXH3_64bits_reset(xxh3_state);
+                        }
+                        xxh3_iter(xxh3_state, next, prev, pos-prev);
+                    }
+                    else
+                        block_crc = crc32c_iter(block_crc, next, prev, pos-prev);
                 }
                 prev = pos;
             }
         }
         else
         {
-            block_crc = crc32c_iter(block_crc, next, pos, (end > block_end ? block_end : end)-pos);
+            if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32)
+            {
+                if (!xxh3_state)
+                {
+                    xxh3_state = XXH3_createState();
+                    XXH3_64bits_reset(xxh3_state);
+                }
+                xxh3_iter(xxh3_state, next, pos, (end > block_end ? block_end : end)-pos);
+            }
+            else
+                block_crc = crc32c_iter(block_crc, next, pos, (end > block_end ? block_end : end)-pos);
             pos = (end > block_end ? block_end : end);
             isset = true;
         }
+        if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32 && xxh3_state)
+        {
+            block_crc = (uint32_t)XXH3_64bits_digest(xxh3_state);
+            XXH3_64bits_reset(xxh3_state);
+        }
         if (set)
         {
             *block_csums = block_crc;
         }
         else if (isset && block_crc != *block_csums)
         {
+            res = false;
             if (bad_block_cb)
-            {
                 bad_block_cb(blk_start, *block_csums, block_crc);
-                res = false;
-            }
             else
-                return false;
+                break;
         }
         block_end += dsk->csum_block_size;
         block_crc = 0;
         block_csums++;
     }
+    if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32 && xxh3_state)
+    {
+        block_crc = (uint32_t)XXH3_64bits_digest(xxh3_state);
+        XXH3_freeState(xxh3_state);
+        xxh3_state = NULL;
+    }
     return res;
 }
 
@@ -1406,7 +1482,7 @@ int blockstore_heap_t::add_entry(uint32_t wr_size, uint32_t *modified_block,
     insert_list_item(li);
     li->block_num = block_num;
     new_wr->size = wr_size;
-    new_wr->crc32c = new_wr->calc_crc32c();
+    new_wr->checksum = new_wr->calc_checksum(this);
     return 0;
 }
 
@@ -1466,7 +1542,7 @@ int blockstore_heap_t::add_big_write(object_id oid, heap_entry_t *old_head, bool
             memset(wr->get_ext_bitmap(this), 0, dsk->clean_entry_bitmap_size);
         memset(wr->get_int_bitmap(this), 0, dsk->clean_entry_bitmap_size);
         bitmap_set(wr->get_int_bitmap(this), offset, len, dsk->bitmap_granularity);
-        if (dsk->data_csum_type)
+        if (dsk->csum_block_size)
         {
             memset(wr->get_checksums(this), 0, get_csum_size(wr));
             calc_checksums(wr, (uint8_t*)data, true, offset, len);
@@ -1495,7 +1571,7 @@ int blockstore_heap_t::add_redirect_intent(object_id oid, heap_entry_t **obj_ptr
             memset(wr->get_ext_bitmap(this), 0, dsk->clean_entry_bitmap_size);
         memset(wr->get_int_bitmap(this), 0, dsk->clean_entry_bitmap_size);
         bitmap_set(wr->get_int_bitmap(this), offset, len, dsk->bitmap_granularity);
-        if (dsk->data_csum_type)
+        if (dsk->csum_block_size)
             memset(wr->get_checksums(this), 0, get_csum_size(wr));
         calc_checksums(wr, (uint8_t*)data, true);
         *obj_ptr = wr;
@@ -1533,7 +1609,7 @@ int blockstore_heap_t::add_big_intent(object_id oid, heap_entry_t **obj_ptr, uin
             memcpy(wr->get_ext_bitmap(this), obj->get_ext_bitmap(this), dsk->clean_entry_bitmap_size);
         memcpy(wr->get_int_bitmap(this), obj->get_int_bitmap(this), dsk->clean_entry_bitmap_size);
         bitmap_set(wr->get_int_bitmap(this), offset, len, dsk->bitmap_granularity);
-        if (dsk->data_csum_type)
+        if (dsk->csum_block_size)
         {
             if (checksums)
                 memcpy(wr->get_checksums(this), checksums, get_csum_size(wr));
@@ -1580,7 +1656,7 @@ int blockstore_heap_t::add_compact(heap_entry_t *obj, uint64_t compact_version,
         new_wr->set_big_location(this, compact_location);
         memcpy(new_wr->get_int_bitmap(this), new_int_bitmap, dsk->clean_entry_bitmap_size);
         memcpy(new_wr->get_ext_bitmap(this), new_ext_bitmap, dsk->clean_entry_bitmap_size);
-        if (dsk->data_csum_type && new_csums)
+        if (dsk->csum_block_size && new_csums)
             memcpy(new_wr->get_checksums(this), new_csums, dsk->data_block_size/dsk->csum_block_size*(dsk->data_csum_type & 0xFF));
     });
 }

src/blockstore/blockstore_heap.h

@@ -43,7 +43,7 @@ struct __attribute__((__packed__)) heap_entry_t
 {
     uint16_t size;
     uint16_t entry_type;
-    uint32_t crc32c;
+    uint32_t checksum;
     uint64_t lsn;
     uint64_t inode;
     uint64_t stripe;
@@ -69,7 +69,8 @@ struct __attribute__((__packed__)) heap_entry_t
     uint32_t *get_checksum(blockstore_heap_t *heap);
     uint64_t big_location(blockstore_heap_t *heap);
     void set_big_location(blockstore_heap_t *heap, uint64_t location);
-    uint32_t calc_crc32c();
+    uint32_t calc_checksum(blockstore_heap_t *heap);
+    uint32_t calc_checksum(blockstore_disk_t *dsk);
 };
 
 struct __attribute__((__packed__)) heap_small_write_t
@@ -80,7 +81,7 @@ struct __attribute__((__packed__)) heap_small_write_t
     uint32_t offset;
     uint32_t len;
 
-    // Also includes 1 bitmap and 1 crc32c after the bitmap if checksums are disabled
+    // Also includes 1 bitmap and 1 checksum after the bitmap if block checksums are disabled
 };
 
 struct __attribute__((__packed__)) heap_big_write_t
@@ -98,7 +99,7 @@ struct __attribute__((__packed__)) heap_big_intent_t
     uint32_t offset;
     uint32_t len;
 
-    // Also includes 2 bitmaps and 1 crc32c if checksums are disabled
+    // Also includes 2 bitmaps and 1 checksums if block checksums are disabled
 };
 
 struct __attribute__((__packed__)) heap_list_item_t

src/client/CMakeLists.txt

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

src/client/cluster_client.cpp

@@ -62,6 +62,7 @@ cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd
     st_cli.on_change_node_placement_hook = [this]() { on_change_node_placement_hook(); };
     st_cli.on_load_pgs_hook = [this](bool success) { on_load_pgs_hook(success); };
     st_cli.on_reload_hook = [this]() { st_cli.load_global_config(); };
+    st_cli.on_inode_change_hook = [this](uint64_t inode, bool removed) { on_change_inode_hook(inode, removed); };
 
     st_cli.parse_config(config);
     st_cli.infinite_start = false;
@@ -70,13 +71,11 @@ cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd
         st_cli.infinite_start = config["client_infinite_start"].bool_value();
     }
     st_cli.load_global_config();
-
-    scrap_buffer_size = SCRAP_BUFFER_SIZE;
-    scrap_buffer = malloc_or_die(scrap_buffer_size);
 }
 
 cluster_client_t::~cluster_client_t()
 {
+    vault_destroy();
     if (retry_timeout_id >= 0)
     {
         tfd->clear_timer(retry_timeout_id);
@@ -94,7 +93,6 @@ cluster_client_t::~cluster_client_t()
     {
         ringloop->unregister_consumer(&consumer);
     }
-    free(scrap_buffer);
     delete wb;
     wb = NULL;
 }
@@ -481,6 +479,8 @@ void cluster_client_t::on_load_config_hook(json11::Json::object & etcd_global_co
         self_tree_metrics.clear();
         client_hostname = new_hostname;
     }
+    // vault
+    vault_parse_config();
     msgr.parse_config(config);
     st_cli.parse_config(config);
     st_cli.load_pgs();
@@ -607,6 +607,9 @@ void cluster_client_t::on_change_pool_config_hook()
             pg_counts[pool_item.first] = pool_item.second.real_pg_count;
         }
     }
+    inode_cache.clear();
+    inode_cache_children.clear();
+    vault_keys.clear();
     continue_ops();
 }
 
@@ -673,6 +676,10 @@ bool cluster_client_t::flush()
 {
     if (!ringloop)
     {
+        if (vault_loading)
+        {
+            return false;
+        }
         if (wb->writeback_queue.size())
         {
             wb->start_writebacks(this, 0);
@@ -695,7 +702,7 @@ bool cluster_client_t::flush()
         sync_done = true;
     };
     execute(sync);
-    while (!sync_done)
+    while (!sync_done || vault_loading)
     {
         ringloop->loop();
         if (!sync_done)
@@ -958,10 +965,40 @@ bool cluster_client_t::check_rw(cluster_op_t *op)
     {
         op->flags |= OP_IMMEDIATE_COMMIT;
     }
+    bool searched = false;
+    std::shared_ptr<inode_cache_t> icache;
+    if (op->opcode == OSD_OP_READ || op->opcode == OSD_OP_WRITE)
+    {
+        if (!searched)
+        {
+            icache = inode_cache_get(op->inode);
+            searched = true;
+        }
+        if (icache && icache->has_parent_loop && op->opcode == OSD_OP_READ)
+        {
+            op->retval = -EINVAL;
+            auto cb = std::move(op->callback);
+            cb(op);
+            return false;
+        }
+        if (icache && icache->op_enc)
+        {
+            // Use shared_ptr aliasing to attach op_enc to the inode cache entry
+            op->enc = std::shared_ptr<osd_op_enc_t>(icache, icache->op_enc);
+        }
+        else
+            op->enc.reset();
+    }
+    else
+        op->enc.reset();
     if ((op->opcode == OSD_OP_WRITE || op->opcode == OSD_OP_DELETE) && !(op->flags & OSD_OP_IGNORE_READONLY))
     {
-        auto ino_it = st_cli.inode_config.find(op->inode);
+        if (!searched)
-        if (ino_it != st_cli.inode_config.end() && ino_it->second.readonly)
+        {
+            icache = inode_cache_get(op->inode);
+            searched = true;
+        }
+        if (icache && icache->readonly)
         {
             op->retval = -EROFS;
             auto cb = std::move(op->callback);
@@ -972,33 +1009,39 @@ bool cluster_client_t::check_rw(cluster_op_t *op)
     op->deoptimise_snapshot = false;
     if (enable_writeback && (op->opcode == OSD_OP_READ || op->opcode == OSD_OP_READ_BITMAP || op->opcode == OSD_OP_READ_CHAIN_BITMAP))
     {
-        auto ino_it = st_cli.inode_config.find(op->inode);
+        if (!searched)
-        if (ino_it != st_cli.inode_config.end())
         {
-            int chain_size = 0;
+            icache = inode_cache_get(op->inode);
-            while (ino_it != st_cli.inode_config.end() && ino_it->second.parent_id)
+            searched = true;
+        }
+        if (icache)
+        {
+            for (auto & parent: icache->chain)
             {
-                // Check for loops - FIXME check it in etcd_state_client
+                if (INODE_POOL(parent) == INODE_POOL(op->inode) && wb->has_inode(parent))
-                if (ino_it->second.parent_id == op->inode ||
-                    chain_size > st_cli.inode_config.size())
-                {
-                    op->retval = -EINVAL;
-                    auto cb = std::move(op->callback);
-                    cb(op);
-                    return false;
-                }
-                if (INODE_POOL(ino_it->second.parent_id) == INODE_POOL(ino_it->first) &&
-                    wb->has_inode(ino_it->second.parent_id))
                 {
                     // Deoptimise reads - we have dirty data for one of the parent layer(s).
                     op->deoptimise_snapshot = true;
                     break;
                 }
-                chain_size++;
-                ino_it = st_cli.inode_config.find(ino_it->second.parent_id);
             }
         }
     }
+    if (icache && icache->err_code)
+    {
+        if (icache->err_code == EPERM)
+        {
+            op->retval = -EPERM;
+            auto cb = std::move(op->callback);
+            cb(op);
+            return false;
+        }
+        else if (icache->err_code == EAGAIN)
+        {
+            key_wait_ops.push_back(op);
+            return false;
+        }
+    }
     return true;
 }
 
@@ -1121,31 +1164,33 @@ resume_2:
         // because if some operations were invalid for the new PG count we'd get errors
         if (op->opcode == OSD_OP_READ || op->opcode == OSD_OP_READ_CHAIN_BITMAP)
         {
-            // Check parent inode
+            uint64_t next_inode = 0;
-            auto ino_it = st_cli.inode_config.find(op->cur_inode);
+            auto icache = inode_cache_get(op->cur_inode);
-            // Skip parents from the same pool
+            if (icache)
-            int skipped = 0;
-            while (!op->deoptimise_snapshot &&
-                ino_it != st_cli.inode_config.end() && ino_it->second.parent_id &&
-                INODE_POOL(ino_it->second.parent_id) == INODE_POOL(op->cur_inode))
             {
-                // Check for loops - FIXME check it in etcd_state_client
+                if (icache->has_parent_loop)
-                if (ino_it->second.parent_id == op->inode ||
-                    skipped > st_cli.inode_config.size())
                 {
                     op->retval = -EINVAL;
                     erase_op(op);
                     return 1;
                 }
-                skipped++;
+                if (op->deoptimise_snapshot)
-                ino_it = st_cli.inode_config.find(ino_it->second.parent_id);
+                {
+                    if (icache->chain.size() > 1)
+                        next_inode = icache->chain[1];
+                }
+                else
+                {
+                    if (icache->other_pool_parent_id)
+                        next_inode = icache->other_pool_parent_id;
+                }
             }
-            if (ino_it != st_cli.inode_config.end() &&
+            if (next_inode)
-                ino_it->second.parent_id &&
-                ino_it->second.parent_id != op->inode)
             {
                 // Continue reading from the parent inode
-                op->cur_inode = ino_it->second.parent_id;
+                icache = inode_cache_get(next_inode);
+                op->cur_inode = next_inode;
+                op->enc = (icache && icache->op_enc ? std::shared_ptr<osd_op_enc_t>(icache, icache->op_enc) : nullptr);
                 op->parts.clear();
                 op->done_count = 0;
                 goto resume_0;
@@ -1196,7 +1241,7 @@ resume_2:
     return 0;
 }
 
-static void add_iov(int size, bool skip, cluster_op_t *op, int &iov_idx, size_t &iov_pos, osd_op_buf_list_t &iov, void *scrap, int scrap_len)
+static void add_iov(int size, int skip, cluster_op_t *op, int &iov_idx, size_t &iov_pos, osd_op_buf_list_t &iov)
 {
     int left = size;
     while (left > 0 && iov_idx < op->iov.count)
@@ -1204,7 +1249,7 @@ static void add_iov(int size, bool skip, cluster_op_t *op, int &iov_idx, size_t
         int cur_left = op->iov.buf[iov_idx].iov_len - iov_pos;
         if (cur_left < left)
         {
-            if (!skip)
+            if (skip == 0)
             {
                 iov.push_back((uint8_t*)op->iov.buf[iov_idx].iov_base + iov_pos, cur_left);
             }
@@ -1214,7 +1259,7 @@ static void add_iov(int size, bool skip, cluster_op_t *op, int &iov_idx, size_t
         }
         else
         {
-            if (!skip)
+            if (skip == 0)
             {
                 iov.push_back((uint8_t*)op->iov.buf[iov_idx].iov_base + iov_pos, left);
             }
@@ -1223,16 +1268,10 @@ static void add_iov(int size, bool skip, cluster_op_t *op, int &iov_idx, size_t
         }
     }
     assert(left == 0);
-    if (skip && scrap_len > 0)
+    if (skip == 1)
     {
-        // All skipped ranges are read into the same useless buffer
+        // data read into a NULL buffer will be discarded by messenger
-        left = size;
+        iov.push_back(NULL, size);
-        while (left > 0)
-        {
-            int cur_left = scrap_len < left ? scrap_len : left;
-            iov.push_back(scrap, cur_left);
-            left -= cur_left;
-        }
     }
 }
 
@@ -1252,7 +1291,11 @@ void cluster_client_t::slice_rw(cluster_op_t *op)
         // Allocate memory for the bitmap
         unsigned object_bitmap_size = ((op->len / pool_cfg.bitmap_granularity + 7) / 8);
         object_bitmap_size = (object_bitmap_size < 8 ? 8 : object_bitmap_size);
-        unsigned bitmap_mem = object_bitmap_size + (pool_cfg.data_block_size / pool_cfg.bitmap_granularity / 8 * pg_data_size) * op->parts.size();
+        unsigned bitmap_mem = object_bitmap_size +
+            op->parts.size() * pg_data_size *
+            (pool_cfg.data_block_size / pool_cfg.bitmap_granularity / 8
+            // read chain info - 1 byte per block
+            + (op->enc ? op->len/pool_cfg.bitmap_granularity : 0));
         if (!op->bitmap_buf || op->bitmap_buf_size < bitmap_mem)
         {
             op->bitmap_buf = realloc_or_die(op->bitmap_buf, bitmap_mem);
@@ -1294,10 +1337,10 @@ void cluster_client_t::slice_rw(cluster_op_t *op)
                         {
                             begin = cur;
                             // Just advance iov_idx & iov_pos
-                            add_iov(cur-prev, true, op, iov_idx, iov_pos, op->parts[i].iov, NULL, 0);
+                            add_iov(cur-prev, 2, op, iov_idx, iov_pos, op->parts[i].iov);
                         }
                         else
-                            add_iov(cur-prev, skip_prev, op, iov_idx, iov_pos, op->parts[i].iov, scrap_buffer, scrap_buffer_size);
+                            add_iov(cur-prev, skip_prev ? 1 : 0, op, iov_idx, iov_pos, op->parts[i].iov);
                     }
                     skip_prev = skip;
                     prev = cur;
@@ -1308,11 +1351,11 @@ void cluster_client_t::slice_rw(cluster_op_t *op)
             if (skip_prev)
             {
                 // Just advance iov_idx & iov_pos
-                add_iov(end-prev, true, op, iov_idx, iov_pos, op->parts[i].iov, NULL, 0);
+                add_iov(end-prev, 2, op, iov_idx, iov_pos, op->parts[i].iov);
                 end = prev;
             }
             else
-                add_iov(cur-prev, skip_prev, op, iov_idx, iov_pos, op->parts[i].iov, scrap_buffer, scrap_buffer_size);
+                add_iov(cur-prev, skip_prev ? 1 : 0, op, iov_idx, iov_pos, op->parts[i].iov);
             if (end == begin)
             {
                 op->done_count++;
@@ -1321,7 +1364,7 @@ void cluster_client_t::slice_rw(cluster_op_t *op)
         }
         else if (op->opcode != OSD_OP_READ_BITMAP && op->opcode != OSD_OP_READ_CHAIN_BITMAP && op->opcode != OSD_OP_DELETE)
         {
-            add_iov(end-begin, false, op, iov_idx, iov_pos, op->parts[i].iov, NULL, 0);
+            add_iov(end-begin, 0, op, iov_idx, iov_pos, op->parts[i].iov);
         }
         op->parts[i].parent = op;
         op->parts[i].offset = begin;
@@ -1407,9 +1450,9 @@ int cluster_client_t::try_send(cluster_op_t *op, int i, std::function<void(osd_o
             osd_client_t *cl = peer_it->second;
             part->flags |= PART_SENT|PART_VALID;
             op->inflight_count++;
-            uint64_t pg_bitmap_size = (pool_cfg.data_block_size / pool_cfg.bitmap_granularity / 8) * (
+            uint32_t pg_data_size = (pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks);
-                pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks
+            uint64_t pg_bitmap_size = pg_data_size * (pool_cfg.data_block_size / pool_cfg.bitmap_granularity / 8
-            );
+                + (op->opcode == OSD_OP_READ && op->enc ? pool_cfg.data_block_size/pool_cfg.bitmap_granularity : 0));
             uint64_t meta_rev = 0;
             if (op->opcode != OSD_OP_READ_BITMAP && op->opcode != OSD_OP_DELETE && !op->deoptimise_snapshot)
             {
@@ -1428,6 +1471,7 @@ int cluster_client_t::try_send(cluster_op_t *op, int i, std::function<void(osd_o
                     .inode = op->cur_inode,
                     .offset = part->offset,
                     .len = part->len,
+                    .flags = op->opcode == OSD_OP_READ && op->enc && !op->deoptimise_snapshot ? OSD_OP_RETURN_CHAIN : 0,
                     .meta_revision = meta_rev,
                     .version = op->opcode == OSD_OP_WRITE || op->opcode == OSD_OP_DELETE ? op->version : 0,
                 } },
@@ -1435,6 +1479,7 @@ int cluster_client_t::try_send(cluster_op_t *op, int i, std::function<void(osd_o
                     ? (uint8_t*)op->part_bitmaps + pg_bitmap_size*i : NULL),
                 .bitmap_len = (unsigned)(op->opcode == OSD_OP_READ || op->opcode == OSD_OP_READ_BITMAP || op->opcode == OSD_OP_READ_CHAIN_BITMAP
                     ? pg_bitmap_size : 0),
+                .enc = op->enc,
                 .callback = cb ? cb : [this, part](osd_op_t *op_part)
                 {
                     handle_op_part(part);

src/client/cluster_client.h

@@ -5,6 +5,7 @@
 
 #include "messenger.h"
 #include "etcd_state_client.h"
+#include "../util/robin_hood.h"
 
 #define DEFAULT_CLIENT_MAX_DIRTY_BYTES 32*1024*1024
 #define DEFAULT_CLIENT_MAX_DIRTY_OPS 1024
@@ -71,6 +72,7 @@ protected:
     cluster_op_t *prev = NULL, *next = NULL;
     int prev_wait = 0;
     uint64_t flush_id = 0;
+    std::shared_ptr<osd_op_enc_t> enc;
     friend class cluster_client_t;
     friend class writeback_cache_t;
 };
@@ -80,6 +82,25 @@ struct inode_list_osd_t;
 struct inode_list_pg_t;
 class writeback_cache_t;
 
+struct inode_cache_t
+{
+    std::vector<inode_t> chain;
+    uint8_t *key_data = NULL;
+    osd_op_enc_t *op_enc = NULL;
+    bool readonly = false;
+    bool has_parent_loop = false;
+    inode_t other_pool_parent_id = 0;
+    int err_code = 0;
+
+    ~inode_cache_t();
+};
+
+struct vault_load_key_t
+{
+    int key_state = 0;
+    std::string key;
+};
+
 // FIXME: Split into public and private interfaces
 class __attribute__((visibility("default"))) cluster_client_t
 {
@@ -89,8 +110,8 @@ public:
     timerfd_manager_t *tfd = NULL;
     ring_loop_t *ringloop = NULL;
 
-    std::map<pool_id_t, uint64_t> pg_counts;
+    // config:
-    std::map<pool_pg_num_t, osd_num_t> pg_primary;
+
     // client_max_dirty_* is actually "max unsynced", for the case when immediate_commit is off
     uint64_t client_max_dirty_bytes = 0;
     uint64_t client_max_dirty_ops = 0;
@@ -102,12 +123,23 @@ public:
     uint64_t client_max_writeback_iodepth = 0;
     std::string conf_hostname;
 
+    std::string vault_url;
+    std::string vault_client_cert;
+    std::string vault_client_key;
+    std::string vault_ca;
+    std::string vault_secret_api_path;
+    uint64_t vault_timeout_ms = 0;
+    uint64_t vault_error_timeout_sec = 0;
+    uint64_t vault_refresh_leeway_sec = 0;
+
     int log_level = 0;
     int client_retry_interval = 50; // ms
     int client_eio_retry_interval = 1000; // ms
     bool client_retry_enospc = true;
     int client_wait_up_timeout = 16; // sec (for listings)
 
+    // state:
+
     std::string client_hostname;
     std::map<std::string, int> self_tree_metrics;
     std::map<osd_num_t, int> osd_tree_metrics;
@@ -115,15 +147,28 @@ public:
     int retry_timeout_id = -1;
     int retry_timeout_duration = 0;
     std::vector<cluster_op_t*> offline_ops;
+    std::vector<cluster_op_t*> key_wait_ops;
     cluster_op_t *op_queue_head = NULL, *op_queue_tail = NULL;
     writeback_cache_t *wb = NULL;
     std::set<osd_num_t> dirty_osds;
     uint64_t dirty_bytes = 0, dirty_ops = 0;
 
-    void *scrap_buffer = NULL;
+    // inodes require some extra state for read/write, it's stored here.
-    unsigned scrap_buffer_size = 0;
+    // moreover, robin_hood access is slightly faster than std::map :)
+    robin_hood::unordered_flat_map<inode_t, std::shared_ptr<inode_cache_t>> inode_cache;
+    std::set<std::pair<inode_t, inode_t>> inode_cache_children;
+
+    http_context_t *vault_http_ctx = NULL;
+    http_co_t *vault_http_cli = NULL;
+    bool vault_loading = false;
+    std::string vault_token;
+    bool vault_auth_error = false;
+    timespec vault_token_expire = {};
+    std::vector<std::string> vault_key_load_queue;
+    std::map<std::string, vault_load_key_t> vault_keys;
 
     bool pgs_loaded = false;
+    std::map<pool_id_t, uint64_t> pg_counts;
     ring_consumer_t consumer;
     std::vector<std::function<void(void)>> on_ready_hooks;
     int list_retry_timeout_id = -1;
@@ -163,6 +208,13 @@ protected:
 #endif
     void continue_ops(int time_passed = 0);
 
+    std::shared_ptr<inode_cache_t> inode_cache_get(inode_t ino);
+    void vault_parse_config();
+    bool vault_check_token();
+    void vault_load_keys();
+    void vault_destroy();
+    void vault_parse_secret(const std::string & key_id, const std::string & err, json11::Json data);
+
 protected:
     bool affects_osd(uint64_t inode, uint64_t offset, uint64_t len, osd_num_t osd);
     bool affects_pg(uint64_t inode, uint64_t offset, uint64_t len, pool_id_t pool_id, pg_num_t pg_num);
@@ -173,6 +225,7 @@ protected:
     void on_change_pg_state_hook(pool_id_t pool_id, pg_num_t pg_num, osd_num_t prev_primary);
     void on_change_osd_state_hook(uint64_t peer_osd);
     void on_change_node_placement_hook();
+    void on_change_inode_hook(uint64_t inode, bool removed);
 
     void execute_internal(cluster_op_t *op);
     void execute_cas(cluster_op_t *op);
@@ -189,6 +242,7 @@ protected:
     void erase_op(cluster_op_t *op);
     void calc_wait(cluster_op_t *op);
     void inc_wait(uint64_t opcode, uint64_t flags, cluster_op_t *next, int inc);
+
     void continue_lists();
     bool continue_listing(inode_list_t *lst);
     bool restart_listing(inode_list_t* lst);

src/client/cluster_client_icache.cpp

@@ -0,0 +1,367 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#include <stdexcept>
+#include <assert.h>
+#include "cluster_client_impl.h"
+#include "http_client.h"
+#include "str_util.h"
+
+#define VAULT_KEY_NOT_LOADED 0
+#define VAULT_KEY_LOADING 1
+#define VAULT_KEY_LOADED 2
+#define VAULT_KEY_ERROR 3
+
+inode_cache_t::~inode_cache_t()
+{
+    if (key_data)
+    {
+        free(key_data);
+        key_data = NULL;
+        op_enc = NULL;
+    }
+}
+
+void cluster_client_t::vault_destroy()
+{
+    if (vault_http_ctx)
+    {
+#ifndef __MOCK__
+        http_destroy(vault_http_cli);
+        http_context_destroy(vault_http_ctx);
+        vault_http_cli = NULL;
+        vault_http_ctx = NULL;
+#endif
+    }
+}
+
+void cluster_client_t::vault_parse_config()
+{
+    vault_url = config["vault_url"].string_value();
+    vault_client_cert = config["vault_client_cert"].string_value();
+    vault_client_key = config["vault_client_key"].string_value();
+    vault_ca = config["vault_ca"].string_value();
+    vault_secret_api_path = "/v1/secret/";
+    if (config["vault_secret_api_path"].is_string())
+        vault_secret_api_path = config["vault_secret_api_path"].string_value();
+    vault_timeout_ms = config["vault_timeout_ms"].uint64_value();
+    if (!vault_timeout_ms)
+        vault_timeout_ms = 5000;
+    vault_error_timeout_sec = config["vault_error_timeout_sec"].uint64_value();
+    if (!vault_error_timeout_sec)
+        vault_error_timeout_sec = 60;
+    vault_refresh_leeway_sec = config["vault_refresh_leeway_sec"].uint64_value();
+    if (!vault_refresh_leeway_sec)
+        vault_refresh_leeway_sec = 60;
+}
+
+// FIXME: Rework client API by adding open/close and cache inode information in the "FD" (maybe)
+void cluster_client_t::on_change_inode_hook(uint64_t inode, bool removed)
+{
+    std::vector<inode_t> children = { inode };
+    for (size_t i = 0; i < children.size(); i++)
+    {
+        auto it = inode_cache_children.lower_bound(std::make_pair(children[i], (inode_t)0));
+        while (it != inode_cache_children.end() && it->first == children[i])
+        {
+            children.push_back(it->second);
+            it++;
+        }
+    }
+    for (auto & inode: children)
+    {
+        auto it = inode_cache.find(inode);
+        if (it != inode_cache.end())
+        {
+            auto icache = it->second;
+            for (auto & parent: icache->chain)
+            {
+                inode_cache_children.erase(std::make_pair(parent, inode));
+            }
+            inode_cache.erase(it);
+        }
+    }
+}
+
+std::shared_ptr<inode_cache_t> cluster_client_t::inode_cache_get(inode_t ino)
+{
+    auto icache_it = inode_cache.find(ino);
+    if (icache_it != inode_cache.end())
+    {
+        return icache_it->second;
+    }
+    // Fill inode cache
+    auto ino_it = st_cli.inode_config.find(ino);
+    if (ino_it == st_cli.inode_config.end())
+    {
+        inode_cache[ino] = NULL;
+        return NULL;
+    }
+    auto pool_it = st_cli.pool_config.find(INODE_POOL(ino));
+    if (pool_it == st_cli.pool_config.end())
+    {
+        inode_cache[ino] = NULL;
+        return NULL;
+    }
+    auto & inode_cfg = ino_it->second;
+    auto & pool_cfg = pool_it->second;
+    std::shared_ptr<inode_cache_t> icache = std::make_shared<inode_cache_t>();
+    icache->readonly = inode_cfg.readonly;
+    icache->chain.push_back(ino);
+    std::vector<inode_config_t*> chain_cfg;
+    // FIXME: Allow unencrypted read & write when all chain is encrypted with the same key
+    int enc_key_count = !inode_cfg.enc_key.empty() ? 1 : 0;
+    if (inode_cfg.parent_id)
+    {
+        // Check for loops and cache the chain
+        robin_hood::unordered_flat_set<inode_t> seen;
+        seen.insert(ino);
+        uint64_t parent_id = inode_cfg.parent_id;
+        while (parent_id)
+        {
+            if (seen.find(parent_id) != seen.end())
+            {
+                icache->has_parent_loop = true;
+                break;
+            }
+            seen.insert(parent_id);
+            ino_it = st_cli.inode_config.find(parent_id);
+            if (INODE_POOL(parent_id) == INODE_POOL(ino))
+            {
+                icache->chain.push_back(parent_id);
+                if (ino_it == st_cli.inode_config.end())
+                    chain_cfg.push_back(NULL);
+                else
+                {
+                    chain_cfg.push_back(&ino_it->second);
+                    if (!ino_it->second.enc_key.empty())
+                        enc_key_count++;
+                }
+            }
+            else if (!icache->other_pool_parent_id)
+                icache->other_pool_parent_id = parent_id;
+            if (ino_it == st_cli.inode_config.end())
+                break;
+            parent_id = ino_it->second.parent_id;
+        }
+    }
+    // Check external keys and wait for loading, if required
+    if (enc_key_count)
+    {
+        for (size_t i = 0; i <= chain_cfg.size(); i++)
+        {
+            inode_config_t *cfg = !i ? &inode_cfg : chain_cfg[i-1];
+            if (cfg && cfg->enc_key.substr(0, strlen(VAULT_KEY_PREFIX)) == VAULT_KEY_PREFIX)
+            {
+                auto & ik = vault_keys[inode_cfg.enc_key];
+                if (ik.key_state == VAULT_KEY_ERROR || vault_url.empty())
+                {
+                    icache->err_code = EPERM;
+                    enc_key_count = 0;
+                }
+                else if (ik.key_state == VAULT_KEY_NOT_LOADED)
+                {
+                    ik.key_state = VAULT_KEY_LOADING;
+                    vault_key_load_queue.push_back(inode_cfg.enc_key);
+                    vault_load_keys();
+                    icache->err_code = EAGAIN;
+                    enc_key_count = 0;
+                }
+                else if (ik.key_state == VAULT_KEY_LOADING)
+                {
+                    icache->err_code = EAGAIN;
+                    enc_key_count = 0;
+                }
+                else
+                {
+                    assert(ik.key_state == VAULT_KEY_LOADED);
+                }
+            }
+        }
+    }
+    // Generate encryption key chain, if applicable
+    if (enc_key_count)
+    {
+        uint8_t *key_data = (uint8_t*)malloc_or_die(
+            AES_256_XTS_KEY_SIZE * enc_key_count +
+            sizeof(uint8_t*) * icache->chain.size() +
+            sizeof(osd_op_enc_t)
+        );
+        uint8_t **keys = (uint8_t**)(key_data + AES_256_XTS_KEY_SIZE * enc_key_count);
+        osd_op_enc_t *enc = (osd_op_enc_t*)((uint8_t*)keys + sizeof(uint8_t*)*icache->chain.size());
+        size_t key_pos = 0;
+        for (size_t i = 0; i <= chain_cfg.size(); i++)
+        {
+            inode_config_t *cfg = !i ? &inode_cfg : chain_cfg[i-1];
+            if (cfg && !cfg->enc_key.empty())
+            {
+                const auto & key = cfg->enc_key.substr(0, strlen(VAULT_KEY_PREFIX)) == VAULT_KEY_PREFIX
+                    ? vault_keys.at(cfg->enc_key).key
+                    : cfg->enc_key;
+                assert(key_pos < AES_256_XTS_KEY_SIZE * enc_key_count);
+                assert(key.size() == 2*AES_256_XTS_KEY_SIZE);
+                keys[i] = key_data + key_pos;
+                fromhexstr(key, AES_256_XTS_KEY_SIZE, key_data + key_pos);
+                key_pos += AES_256_XTS_KEY_SIZE;
+            }
+            else
+                keys[i] = NULL;
+        }
+        enc->key_chain = keys;
+        enc->chain_size = icache->chain.size();
+        enc->read_chain_bitmap_pos = pool_cfg.data_block_size/pool_cfg.bitmap_granularity/8;
+        enc->bitmap_granularity = pool_cfg.bitmap_granularity;
+        icache->key_data = key_data;
+        icache->op_enc = enc;
+    }
+    inode_cache[ino] = icache;
+    for (auto & parent: icache->chain)
+    {
+        if (parent != ino)
+            inode_cache_children.insert(std::make_pair(parent, ino));
+    }
+    return icache;
+}
+
+#ifndef __MOCK__
+bool cluster_client_t::vault_check_token()
+{
+    timespec now;
+    clock_gettime(CLOCK_REALTIME, &now);
+    if (!vault_token_expire.tv_sec || vault_token_expire.tv_sec < now.tv_sec)
+    {
+        vault_loading = true;
+        http_json_post(
+            vault_http_cli, vault_url+"/v1/auth/cert/login", json11::Json::object{}, "",
+            (http_options_t){ .timeout = (int)vault_timeout_ms, .keepalive = true },
+            [this](http_message_t *response)
+            {
+                clock_gettime(CLOCK_REALTIME, &vault_token_expire);
+                vault_loading = false;
+                std::string err;
+                json11::Json data;
+                response->parse_json_response(err, data);
+                if (err != "")
+                {
+                    vault_token_expire.tv_sec += vault_error_timeout_sec;
+                    fprintf(stderr, "Vault request failed: %s\n", err.c_str());
+                }
+                else
+                {
+                    uint64_t ttl = data["auth"]["lease_duration"].uint64_value();
+                    vault_token = data["auth"]["client_token"].string_value();
+                    if (vault_token.empty() || !ttl)
+                    {
+                        vault_token_expire.tv_sec += vault_error_timeout_sec;
+                        fprintf(stderr, "No token or lease_duration in Vault response: %s\n", data.dump().c_str());
+                    }
+                    else
+                    {
+                        if (ttl < vault_refresh_leeway_sec)
+                            vault_token_expire.tv_sec += ttl/2;
+                        else
+                            vault_token_expire.tv_sec += ttl - vault_refresh_leeway_sec;
+                    }
+                }
+                vault_load_keys();
+            }
+        );
+        return false;
+    }
+    if (vault_token.empty())
+    {
+        // Auth error happened, mark all loads as failed
+        for (auto & key_id: vault_key_load_queue)
+        {
+            auto & k = vault_keys[key_id];
+            k.key_state = VAULT_KEY_ERROR;
+        }
+        vault_key_load_queue.clear();
+        auto ops = std::move(key_wait_ops);
+        for (cluster_op_t *op: ops)
+            inode_cache.erase(op->inode);
+        for (cluster_op_t *op: ops)
+            execute_internal(op);
+        return false;
+    }
+    return true;
+}
+#endif
+
+void cluster_client_t::vault_load_keys()
+{
+    if (vault_loading || !vault_key_load_queue.size())
+    {
+        return;
+    }
+#ifdef __MOCK__
+    vault_loading = true;
+#else
+    if (!vault_http_ctx)
+    {
+        std::string error;
+        vault_http_ctx = http_context_init(tfd, vault_client_cert, vault_client_key, vault_ca, true, error);
+        if (!vault_http_ctx)
+        {
+            fprintf(stderr, "Failed to initialize HTTP context for Vault: %s\n", error.c_str());
+            exit(1);
+        }
+        vault_http_cli = http_init(vault_http_ctx);
+    }
+    if (!vault_check_token())
+    {
+        return;
+    }
+    std::string key_id = vault_key_load_queue[0];
+    vault_key_load_queue.erase(vault_key_load_queue.begin());
+    vault_loading = true;
+    http_get(
+        vault_http_cli, vault_url+vault_secret_api_path+key_id.substr(strlen(VAULT_KEY_PREFIX)), "X-Vault-Token: "+vault_token+"\r\n",
+        (http_options_t){ .timeout = (int)vault_timeout_ms, .keepalive = true },
+        [this, key_id](http_message_t *response)
+        {
+            vault_loading = false;
+            std::string err;
+            json11::Json data;
+            response->parse_json_response(err, data);
+            vault_parse_secret(key_id, err, data);
+        }
+    );
+#endif
+}
+
+void cluster_client_t::vault_parse_secret(const std::string & key_id, const std::string & err, json11::Json data)
+{
+    vault_loading = false;
+    auto & k = vault_keys[key_id];
+    if (err != "")
+    {
+        k.key_state = VAULT_KEY_ERROR;
+        fprintf(stderr, "Vault %s%s%s request failed: %s\n", vault_url.c_str(),
+            vault_secret_api_path.c_str(), key_id.c_str()+strlen(VAULT_KEY_PREFIX), err.c_str());
+    }
+    else
+    {
+        auto hexkey = data["data"]["key"].string_value();
+        if (hexkey.empty() || !ishexstr(hexkey) || hexkey.size() != 2*AES_256_XTS_KEY_SIZE)
+        {
+            k.key_state = VAULT_KEY_ERROR;
+            fprintf(stderr, "Vault /v1/secret/%s request failed: 'key' is empty or has invalid format\n", key_id.c_str());
+        }
+        else
+        {
+            k.key_state = VAULT_KEY_LOADED;
+            k.key = hexkey;
+        }
+    }
+    if (vault_key_load_queue.empty())
+    {
+        auto ops = std::move(key_wait_ops);
+        for (cluster_op_t *op: ops)
+            inode_cache.erase(op->inode);
+        for (cluster_op_t *op: ops)
+            execute_internal(op);
+    }
+    else
+        vault_load_keys();
+}

src/client/cluster_client_impl.h

@@ -5,7 +5,6 @@
 
 #include "cluster_client.h"
 
-#define SCRAP_BUFFER_SIZE 4*1024*1024
 #define PART_SENT 1
 #define PART_DONE 2
 #define PART_ERROR 4

src/client/etcd_state_client.cpp

@@ -1,7 +1,10 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
+#include <assert.h>
+#include "malloc_or_die.h"
 #include "osd_ops.h"
+#include "msgr_op.h"
 #include "pg_states.h"
 #include "etcd_state_client.h"
 #ifndef __MOCK__
@@ -22,14 +25,19 @@ etcd_state_client_t::~etcd_state_client_t()
     stop_ws_keepalive();
     if (etcd_watch_ws)
     {
-        http_close(etcd_watch_ws);
+        http_destroy(etcd_watch_ws);
         etcd_watch_ws = NULL;
     }
     if (keepalive_client)
     {
-        http_close(keepalive_client);
+        http_destroy(keepalive_client);
         keepalive_client = NULL;
     }
+    if (http_ctx)
+    {
+        http_context_destroy(http_ctx);
+        http_ctx = NULL;
+    }
 #endif
     if (load_pgs_timer_id >= 0)
     {
@@ -72,55 +80,51 @@ std::vector<std::string> etcd_state_client_t::get_addresses()
     return addrs;
 }
 
-void etcd_state_client_t::etcd_call_oneshot(std::string etcd_address, std::string api, json11::Json payload,
+http_context_t *etcd_state_client_t::get_http_ctx()
+{
+    if (!http_ctx)
+    {
+        std::string error;
+        http_ctx = http_context_init(tfd, etcd_client_cert, etcd_client_key, etcd_ca, true, error);
+        if (!http_ctx)
+        {
+            fprintf(stderr, "Failed to initialize HTTP context: %s\n", error.c_str());
+            exit(1);
+        }
+    }
+    return http_ctx;
+}
+
+void etcd_state_client_t::etcd_call_oneshot(const std::string & etcd_url, const std::string & api, json11::Json payload,
     int timeout, std::function<void(std::string, json11::Json)> callback)
 {
-    std::string etcd_api_path;
+    auto http_cli = http_init(get_http_ctx());
-    int pos = etcd_address.find('/');
+    http_json_post(http_cli, etcd_url+api, payload, "", { .timeout = timeout }, [http_cli, callback](http_message_t *response)
-    if (pos >= 0)
-    {
-        etcd_api_path = etcd_address.substr(pos);
-        etcd_address = etcd_address.substr(0, pos);
-    }
-    std::string req = payload.dump();
-    req = "POST "+etcd_api_path+api+" HTTP/1.1\r\n"
-        "Host: "+etcd_address+"\r\n"
-        "Content-Type: application/json\r\n"
-        "Content-Length: "+std::to_string(req.size())+"\r\n"
-        "Connection: close\r\n"
-        "\r\n"+req;
-    auto http_cli = http_init(tfd);
-    auto cb = [http_cli, callback](const http_response_t *response)
     {
         std::string err;
         json11::Json data;
         response->parse_json_response(err, data);
         callback(err, data);
-        http_close(http_cli);
+        http_destroy(http_cli);
-    };
+    });
-    http_request(http_cli, etcd_address, req, { .timeout = timeout }, cb);
 }
 
-void etcd_state_client_t::etcd_call(std::string api, json11::Json payload, int timeout,
+void etcd_state_client_t::etcd_call(const std::string & api, json11::Json payload, int timeout,
     int retries, int interval, std::function<void(std::string, json11::Json)> callback)
 {
-    if (!etcd_addresses.size() && !etcd_local.size())
+    pick_next_etcd([=]()
     {
-        fprintf(stderr, "etcd_address is missing in Vitastor configuration\n");
+        etcd_call_selected(api, payload, timeout, retries, interval, callback);
-        exit(1);
+    });
-    }
+}
-    pick_next_etcd();
+
-    std::string etcd_address = selected_etcd_address;
+void etcd_state_client_t::etcd_call_selected(const std::string & api, json11::Json payload, int timeout,
-    std::string etcd_api_path;
+    int retries, int interval, std::function<void(std::string, json11::Json)> callback)
-    int pos = etcd_address.find('/');
+{
-    if (pos >= 0)
+    const auto & url = selected_etcd_url;
-    {
-        etcd_api_path = etcd_address.substr(pos);
-        etcd_address = etcd_address.substr(0, pos);
-    }
     std::string req = payload.dump();
-    req = "POST "+etcd_api_path+api+" HTTP/1.1\r\n"
+    req = "POST "+url.path+api+" HTTP/1.1\r\n"
-        "Host: "+etcd_address+"\r\n"
+        "Host: "+url.hostname+"\r\n"
         "Content-Type: application/json\r\n"
         "Content-Length: "+std::to_string(req.size())+"\r\n"
         "Connection: keep-alive\r\n"
@@ -128,15 +132,15 @@ void etcd_state_client_t::etcd_call(std::string api, json11::Json payload, int t
         "\r\n"+req;
     retries--;
     auto cb = [this, api, payload, timeout, retries, interval, callback,
-        cur_addr = selected_etcd_address](const http_response_t *response)
+        cur_addr = url.addr](http_message_t *response)
     {
         std::string err;
         json11::Json data;
         response->parse_json_response(err, data);
         if (err != "")
         {
-            if (cur_addr == selected_etcd_address)
+            if (cur_addr == selected_etcd_url.addr)
-                selected_etcd_address = "";
+                selected_etcd_url = (http_url_t){};
             if (retries > 0)
             {
                 if (this->log_level > 0)
@@ -164,54 +168,58 @@ void etcd_state_client_t::etcd_call(std::string api, json11::Json payload, int t
             callback(err, data);
     };
     if (!keepalive_client)
-    {
+        keepalive_client = http_init(get_http_ctx());
-        keepalive_client = http_init(tfd);
+    http_request(keepalive_client, url.addr, req, { .timeout = timeout, .keepalive = true, .ssl = url.ssl }, cb);
-    }
-    http_request(keepalive_client, etcd_address, req, { .timeout = timeout, .keepalive = true }, cb);
 }
 
-void etcd_state_client_t::add_etcd_url(std::string addr)
+
+void etcd_state_client_t::add_etcd_url(std::string etcd_address)
 {
-    if (addr.length() > 0)
+    if (etcd_address.size() > 0)
     {
-        if (strtolower(addr.substr(0, 7)) == "http://")
-            addr = addr.substr(7);
-        else if (strtolower(addr.substr(0, 8)) == "https://")
-        {
-            fprintf(stderr, "HTTPS is unsupported for etcd. Either use plain HTTP or setup a local proxy for etcd interaction\n");
-            exit(1);
-        }
         if (!local_ips.size())
-            local_ips = getifaddr_list(std::vector<addr_mask_t>(), true);
-        std::string check_addr;
-        int pos = addr.find('/');
-        int pos2 = addr.find(':');
-        if (pos2 >= 0)
-            check_addr = addr.substr(0, pos2);
-        else if (pos >= 0)
-            check_addr = addr.substr(0, pos);
-        else
-            check_addr = addr;
-        if (pos == std::string::npos)
-            addr += "/v3";
-        bool local = false;
-        int i;
-        for (i = 0; i < local_ips.size(); i++)
         {
-            if (local_ips[i] == check_addr)
+            // Fill local_ips
-            {
+            for (auto & ip: getifaddr_list(std::vector<addr_mask_t>(), true))
-                local = true;
+                local_ips.insert(ip);
-                break;
-            }
         }
-        auto & to = local ? this->etcd_local : this->etcd_addresses;
+        std::string etcd_api_path;
+        bool ssl = false;
+        if (etcd_address.substr(0, 8) == "https://")
+        {
+            ssl = true;
+            etcd_address = etcd_address.substr(8);
+        }
+        else if (etcd_address.substr(0, 7) == "http://")
+            etcd_address = etcd_address.substr(7);
+        auto pos = etcd_address.find('/');
+        if (pos != std::string::npos)
+        {
+            etcd_api_path = etcd_address.substr(pos);
+            etcd_address = etcd_address.substr(0, pos);
+        }
+        else
+            etcd_api_path = "/v3";
+        pos = etcd_address.find(':');
+        auto check_addr = (pos != std::string::npos ? etcd_address.substr(0, pos) : etcd_address);
+        bool is_local = local_ips.find(check_addr) != local_ips.end();
+        auto & to = (is_local ? etcd_local : etcd_addresses);
+        check_addr = (ssl ? "https://" : "http://") + etcd_address + etcd_api_path;
+        size_t i;
         for (i = 0; i < to.size(); i++)
         {
-            if (to[i] == addr)
+            if (to[i] == check_addr)
                 break;
         }
         if (i >= to.size())
-            to.push_back(addr);
+        {
+            to.push_back(check_addr);
+            // Check if it's a domain name
+            sockaddr_storage ss;
+            bool is_name = !is_local && !string_to_addr(etcd_address, true, 0, &ss);
+            auto & to_addr = (is_local ? etcd_local_addr_urls : (is_name ? etcd_name_urls : etcd_nonlocal_addr_urls));
+            to_addr.push_back((http_url_t){ .ssl = ssl, .addr = etcd_address, .hostname = etcd_address, .path = etcd_api_path });
+        }
     }
 }
 
@@ -219,6 +227,9 @@ void etcd_state_client_t::parse_config(const json11::Json & config)
 {
     this->etcd_local.clear();
     this->etcd_addresses.clear();
+    this->etcd_local_addr_urls.clear();
+    this->etcd_nonlocal_addr_urls.clear();
+    this->etcd_name_urls.clear();
     if (config["etcd_address"].is_string())
     {
         std::string ea = config["etcd_address"].string_value();
@@ -239,7 +250,19 @@ void etcd_state_client_t::parse_config(const json11::Json & config)
             add_etcd_url(ea.string_value());
         }
     }
+    if (this->osd_num)
+    {
+        this->etcd_client_cert = config["osd_etcd_client_cert"].string_value();
+        this->etcd_client_key = config["osd_etcd_client_key"].string_value();
+    }
+    else
+    {
+        this->etcd_client_cert = config["etcd_client_cert"].string_value();
+        this->etcd_client_key = config["etcd_client_key"].string_value();
+    }
+    this->etcd_ca = config["etcd_ca"].string_value();
     this->etcd_prefix = config["etcd_prefix"].string_value();
+    this->use_auth = config["use_auth"].bool_value();
     if (this->etcd_prefix == "")
     {
         this->etcd_prefix = "/vitastor";
@@ -291,66 +314,130 @@ void etcd_state_client_t::parse_config(const json11::Json & config)
     }
 }
 
-void etcd_state_client_t::pick_next_etcd()
+void etcd_state_client_t::pick_next_etcd(std::function<void()> cb)
-{
-    if (selected_etcd_address != "")
-        return;
-    if (addresses_to_try.size() == 0)
-    {
-        // Prefer local etcd, if any
-        for (int i = 0; i < etcd_local.size(); i++)
-            addresses_to_try.push_back(etcd_local[i]);
-        std::vector<int> ns;
-        for (int i = 0; i < etcd_addresses.size(); i++)
-            ns.push_back(i);
-        if (!rand_initialized)
-        {
-            timespec tv;
-            clock_gettime(CLOCK_REALTIME, &tv);
-            srand48(tv.tv_sec*1000000000 + tv.tv_nsec);
-            rand_initialized = true;
-        }
-        while (ns.size())
-        {
-            int i = lrand48() % ns.size();
-            addresses_to_try.push_back(etcd_addresses[ns[i]]);
-            ns.erase(ns.begin()+i, ns.begin()+i+1);
-        }
-    }
-    selected_etcd_address = addresses_to_try[0];
-    addresses_to_try.erase(addresses_to_try.begin(), addresses_to_try.begin()+1);
-}
-
-void etcd_state_client_t::start_etcd_watcher()
 {
     if (!etcd_addresses.size() && !etcd_local.size())
     {
         fprintf(stderr, "etcd_address is missing in Vitastor configuration\n");
         exit(1);
     }
-    pick_next_etcd();
+    if (selected_etcd_url.addr != "")
-    std::string etcd_address = selected_etcd_address;
-    std::string etcd_api_path;
-    int pos = etcd_address.find('/');
-    if (pos >= 0)
     {
-        etcd_api_path = etcd_address.substr(pos);
+        cb();
-        etcd_address = etcd_address.substr(0, pos);
+        return;
     }
+    if (etcd_urls_to_try.size() != 0)
+    {
+        selected_etcd_url = std::move(etcd_urls_to_try[0]);
+        etcd_urls_to_try.erase(etcd_urls_to_try.begin());
+        cb();
+        return;
+    }
+    on_resolve_queue.push_back(std::move(cb));
+    if (on_resolve_queue.size() > 1)
+    {
+        // Already resolving
+        return;
+    }
+    assert(!resolve_count);
+    local_to_try = 0;
+    for (auto & url: etcd_local_addr_urls)
+    {
+        // Prefer local IPs, if any
+        etcd_urls_to_try.push_back(url);
+        local_to_try++;
+    }
+    for (auto & url: etcd_nonlocal_addr_urls)
+    {
+        etcd_urls_to_try.push_back(url);
+    }
+    resolve_count++;
+    for (auto & url: etcd_name_urls)
+    {
+        resolve_count++;
+        http_resolve(get_http_ctx(), url.ssl, url.addr, [this, url](const std::string & error, const std::vector<std::string>& addresses)
+        {
+            if (error != "")
+                fprintf(stderr, "Error resolving %s: %s\n", url.addr.c_str(), error.c_str());
+            for (auto & addr: addresses)
+            {
+                auto url_copy = url;
+                url_copy.addr = addr;
+                if (local_ips.find(addr) != local_ips.end())
+                {
+                    etcd_urls_to_try.insert(etcd_urls_to_try.begin(), std::move(url_copy));
+                    local_to_try++;
+                }
+                else
+                    etcd_urls_to_try.push_back(std::move(url_copy));
+            }
+            resolve_count--;
+            if (!resolve_count)
+                pick_next_etcd_on_resolve();
+        });
+    }
+    resolve_count--;
+    if (!resolve_count)
+    {
+        pick_next_etcd_on_resolve();
+    }
+}
+
+void etcd_state_client_t::pick_next_etcd_on_resolve()
+{
+    if (!etcd_urls_to_try.size())
+    {
+        fprintf(stderr, "None of etcd_address could be resolved\n");
+        exit(1);
+    }
+    if (!rand_initialized)
+    {
+        timespec tv;
+        clock_gettime(CLOCK_REALTIME, &tv);
+        srand48(tv.tv_sec*1000000000 + tv.tv_nsec);
+        rand_initialized = true;
+    }
+    // Shuffle addresses
+    for (size_t i = etcd_urls_to_try.size()-1; i > local_to_try; i--)
+    {
+        size_t j = local_to_try + lrand48() % (i - local_to_try);
+        if (j != i)
+            std::swap(etcd_urls_to_try[i], etcd_urls_to_try[j]);
+    }
+    selected_etcd_url = std::move(etcd_urls_to_try[0]);
+    etcd_urls_to_try.erase(etcd_urls_to_try.begin());
+    auto cbs = std::move(on_resolve_queue);
+    for (auto cb: cbs)
+    {
+        cb();
+    }
+}
+
+void etcd_state_client_t::start_etcd_watcher()
+{
+    pick_next_etcd([this]()
+    {
+        start_etcd_watcher_selected();
+    });
+}
+
+void etcd_state_client_t::start_etcd_watcher_selected()
+{
+    const auto & url = selected_etcd_url;
     etcd_watches_initialised = 0;
     ws_alive = 1;
-    if (etcd_watch_ws)
-    {
-        http_close(etcd_watch_ws);
-        etcd_watch_ws = NULL;
-    }
     if (this->log_level > 1)
     {
-        fprintf(stderr, "Trying to connect to etcd websocket at %s, watch from revision %ju/%ju/%ju\n", etcd_address.c_str(),
+        fprintf(stderr, "Trying to connect to etcd websocket at %s%s%s (hostname %s), watch from revision %ju/%ju/%ju\n",
+            url.ssl ? "https://" : "http://", url.addr.c_str(), url.path.c_str(), url.hostname.c_str(),
             etcd_watch_revision_config, etcd_watch_revision_osd, etcd_watch_revision_pg);
     }
-    etcd_watch_ws = open_websocket(tfd, etcd_address, etcd_api_path+"/watch", etcd_slow_timeout,
+    if (!etcd_watch_ws)
-        [this, cur_addr = selected_etcd_address](const http_response_t *msg)
+        etcd_watch_ws = http_init(get_http_ctx());
+    else
+        http_close(etcd_watch_ws);
+    open_websocket(etcd_watch_ws, url.addr, url.hostname, url.path+"/watch", { .timeout = etcd_slow_timeout, .ssl = url.ssl },
+        [this, cur_addr = url.addr](http_message_t *msg)
     {
         if (msg->body.length())
         {
@@ -393,7 +480,6 @@ void etcd_state_client_t::start_etcd_watcher()
                                 fprintf(stderr, "Revisions before %ju were compacted by etcd, reloading state\n",
                                     data["result"]["compact_revision"].uint64_value());
                                 http_close(etcd_watch_ws);
-                                etcd_watch_ws = NULL;
                                 etcd_watch_revision_config = etcd_watch_revision_osd = etcd_watch_revision_pg = 0;
                                 on_reload_hook();
                             }
@@ -438,7 +524,7 @@ void etcd_state_client_t::start_etcd_watcher()
                         etcd_watch_revision_pg = watch_rev;
                     else if (watch_id == ETCD_OSD_STATE_WATCH_ID)
                         etcd_watch_revision_osd = watch_rev;
-                    addresses_to_try.clear();
+                    etcd_urls_to_try.clear();
                 }
                 // First gather all changes into a hash to remove multiple overwrites
                 std::map<std::string, etcd_kv_t> changes;
@@ -468,13 +554,8 @@ void etcd_state_client_t::start_etcd_watcher()
         if (msg->eof)
         {
             fprintf(stderr, "Disconnected from etcd %s\n", cur_addr.c_str());
-            if (cur_addr == selected_etcd_address)
+            if (cur_addr == selected_etcd_url.addr)
-                selected_etcd_address = "";
+                selected_etcd_url = (http_url_t){};
-            if (etcd_watch_ws)
-            {
-                http_close(etcd_watch_ws);
-                etcd_watch_ws = NULL;
-            }
             if (etcd_watches_initialised == 0)
             {
                 // Connection not established, retry in <etcd_quick_timeout>
@@ -549,12 +630,7 @@ void etcd_state_client_t::start_ws_keepalive()
             {
                 if (this->log_level > 0)
                 {
-                    fprintf(stderr, "Websocket ping failed, disconnecting from etcd %s\n", selected_etcd_address.c_str());
+                    fprintf(stderr, "Websocket ping failed, disconnecting from etcd %s\n", selected_etcd_url.addr.c_str());
-                }
-                if (etcd_watch_ws)
-                {
-                    http_close(etcd_watch_ws);
-                    etcd_watch_ws = NULL;
                 }
                 start_etcd_watcher();
             }
@@ -942,6 +1018,8 @@ void etcd_state_client_t::parse_state(const etcd_kv_t & kv)
                 pc.used_for_app = "fs:"+pc.used_for_app;
             else
                 pc.used_for_app = pool_item.second["used_for_app"].as_string();
+            // Create group permission
+            pc.creator_group = pool_item.second["creator_group"].string_value();
             // Local Read Configuration
             std::string local_reads = pool_item.second["local_reads"].string_value();
             if (local_reads == "nearest")
@@ -1265,33 +1343,7 @@ void etcd_state_client_t::parse_state(const etcd_kv_t & kv)
             }
             else
             {
-                inode_t parent_inode_num = value["parent_id"].uint64_value();
+                insert_inode_config(deserialize_inode_cfg(inode_num, kv.value, kv.mod_revision));
-                if (parent_inode_num && !(parent_inode_num >> (64-POOL_ID_BITS)))
-                {
-                    uint64_t parent_pool_id = value["parent_pool"].uint64_value();
-                    if (!parent_pool_id)
-                        parent_inode_num |= pool_id << (64-POOL_ID_BITS);
-                    else if (parent_pool_id >= POOL_ID_MAX)
-                    {
-                        fprintf(
-                            stderr, "Inode %ju/%ju parent_pool value is invalid, ignoring parent setting\n",
-                            inode_num >> (64-POOL_ID_BITS), inode_num & (((uint64_t)1 << (64-POOL_ID_BITS)) - 1)
-                        );
-                        parent_inode_num = 0;
-                    }
-                    else
-                        parent_inode_num |= parent_pool_id << (64-POOL_ID_BITS);
-                }
-                insert_inode_config((inode_config_t){
-                    .num = inode_num,
-                    .name = value["name"].string_value(),
-                    .size = value["size"].uint64_value(),
-                    .parent_id = parent_inode_num,
-                    .readonly = value["readonly"].bool_value(),
-                    .deleted = value["deleted"].bool_value(),
-                    .meta = value["meta"],
-                    .mod_revision = kv.mod_revision,
-                });
             }
         }
     }
@@ -1302,6 +1354,14 @@ void etcd_state_client_t::parse_state(const etcd_kv_t & kv)
         if (on_change_node_placement_hook)
             on_change_node_placement_hook();
     }
+    else if (use_auth && key.substr(0, etcd_prefix.length()+13) == etcd_prefix+"/config/user/")
+    {
+        // <etcd_prefix>/config/user/<username>
+        if (!value.is_object())
+            user_info.erase(key.substr(etcd_prefix.length()+13));
+        else
+            user_info[key.substr(etcd_prefix.length()+13)] = value;
+    }
 }
 
 uint32_t etcd_state_client_t::parse_immediate_commit(const std::string & immediate_commit_str, uint32_t default_value)
@@ -1380,6 +1440,10 @@ json11::Json::object etcd_state_client_t::serialize_inode_cfg(inode_config_t *cf
             new_cfg["parent_pool"] = (uint64_t)INODE_POOL(cfg->parent_id);
         new_cfg["parent_id"] = (uint64_t)INODE_NO_POOL(cfg->parent_id);
     }
+    if (!cfg->enc_key.empty())
+    {
+        new_cfg["enc_key"] = cfg->enc_key;
+    }
     if (cfg->readonly)
     {
         new_cfg["readonly"] = true;
@@ -1388,6 +1452,18 @@ json11::Json::object etcd_state_client_t::serialize_inode_cfg(inode_config_t *cf
     {
         new_cfg["deleted"] = true;
     }
+    if (!cfg->owner.empty())
+    {
+        new_cfg["owner"] = cfg->owner;
+    }
+    if (!cfg->owner_group.empty())
+    {
+        new_cfg["owner_group"] = cfg->owner_group;
+    }
+    if (!cfg->reader_group.empty())
+    {
+        new_cfg["reader_group"] = cfg->reader_group;
+    }
     if (cfg->meta.is_object())
     {
         new_cfg["meta"] = cfg->meta;
@@ -1395,6 +1471,53 @@ json11::Json::object etcd_state_client_t::serialize_inode_cfg(inode_config_t *cf
     return new_cfg;
 }
 
+inode_config_t etcd_state_client_t::deserialize_inode_cfg(uint64_t inode_num, json11::Json value, uint64_t mod_revision)
+{
+    inode_t parent_inode_num = value["parent_id"].uint64_value();
+    if (parent_inode_num && !INODE_POOL(parent_inode_num))
+    {
+        uint64_t parent_pool_id = value["parent_pool"].uint64_value();
+        if (!parent_pool_id)
+            parent_inode_num = INODE_WITH_POOL(INODE_POOL(inode_num), parent_inode_num);
+        else if (parent_pool_id >= POOL_ID_MAX)
+        {
+            fprintf(
+                stderr, "Inode %u/%ju parent_pool value is invalid, ignoring parent setting\n",
+                INODE_POOL(inode_num), INODE_NO_POOL(inode_num)
+            );
+            parent_inode_num = 0;
+        }
+        else
+            parent_inode_num |= parent_pool_id << (64-POOL_ID_BITS);
+    }
+    std::string enc_key;
+    if (!value["enc_key"].is_null())
+    {
+        enc_key = value["enc_key"].string_value();
+        if (enc_key.substr(0, strlen(VAULT_KEY_PREFIX)) != VAULT_KEY_PREFIX &&
+            (enc_key.size() != 2*AES_256_XTS_KEY_SIZE || !ishexstr(enc_key)))
+        {
+            enc_key = "";
+            fprintf(stderr, "Inode %u/%ju has invalid enc_key, should be %u bit hex string or Vault key reference\n",
+                INODE_POOL(inode_num), INODE_NO_POOL(inode_num), AES_256_XTS_KEY_SIZE);
+        }
+    }
+    return (inode_config_t){
+        .num = inode_num,
+        .name = value["name"].string_value(),
+        .size = value["size"].uint64_value(),
+        .parent_id = parent_inode_num,
+        .readonly = value["readonly"].bool_value(),
+        .deleted = value["deleted"].bool_value(),
+        .enc_key = std::move(enc_key),
+        .owner = value["owner"].string_value(),
+        .owner_group = value["owner_group"].string_value(),
+        .reader_group = value["reader_group"].string_value(),
+        .meta = value["meta"],
+        .mod_revision = mod_revision,
+    };
+}
+
 int etcd_state_client_t::address_count()
 {
     return etcd_addresses.size() + etcd_local.size();

src/client/etcd_state_client.h

@@ -4,6 +4,7 @@
 #pragma once
 
 #include <set>
+#include <memory>
 
 #include "json11/json11.hpp"
 #include "object_id.h"
@@ -19,6 +20,8 @@
 #define MAX_DATA_BLOCK_SIZE 128*1024*1024
 #define DEFAULT_BITMAP_GRANULARITY 4096
 
+#define VAULT_KEY_PREFIX "vault:"
+
 #ifndef IMMEDIATE_NONE
 #define IMMEDIATE_NONE 0
 #define IMMEDIATE_SMALL 1
@@ -66,6 +69,7 @@ struct pool_config_t
     std::map<pg_num_t, pg_config_t> pg_config;
     uint64_t scrub_interval = 0;
     std::string used_for_app;
+    std::string creator_group;
     int backfillfull = 0;
     int local_reads = 0;
 
@@ -83,6 +87,9 @@ struct inode_config_t
     inode_t parent_id = 0;
     bool readonly = false;
     bool deleted = false;
+    std::string enc_key;
+    // Permissions
+    std::string owner, owner_group, reader_group;
     // Arbitrary metadata
     json11::Json meta;
     // Change revision of the metadata in etcd
@@ -95,23 +102,41 @@ struct inode_watch_t
     inode_config_t cfg = {};
 };
 
+struct http_url_t
+{
+    bool ssl;
+    std::string addr;
+    std::string hostname;
+    std::string path;
+};
+
 struct http_co_t;
+struct http_context_t;
 
 struct __attribute__((visibility("default"))) etcd_state_client_t
 {
 protected:
-    std::vector<std::string> local_ips;
+    std::set<std::string> local_ips;
-    std::vector<std::string> etcd_addresses;
     std::vector<std::string> etcd_local;
-    std::string selected_etcd_address;
+    std::vector<std::string> etcd_addresses;
-    std::vector<std::string> addresses_to_try;
+    std::vector<http_url_t> etcd_local_addr_urls;
+    std::vector<http_url_t> etcd_nonlocal_addr_urls;
+    std::vector<http_url_t> etcd_name_urls;
+    size_t local_to_try = 0;
+    std::vector<http_url_t> etcd_urls_to_try;
+    http_url_t selected_etcd_url;
+    size_t resolve_count = 0;
     std::vector<inode_watch_t*> watches;
+    std::vector<std::function<void()>> on_resolve_queue;
     bool new_pg_config = false;
     int ws_keepalive_timer = -1;
     int ws_alive = 0;
     bool rand_initialized = false;
     void add_etcd_url(std::string);
-    void pick_next_etcd();
+    void pick_next_etcd(std::function<void()> cb);
+    void pick_next_etcd_on_resolve();
+    void etcd_call_selected(const std::string & api, json11::Json payload, int timeout, int retries, int interval, std::function<void(std::string, json11::Json)> callback);
+    void start_etcd_watcher_selected();
 public:
     int etcd_keepalive_timeout = 30;
     int etcd_ws_keepalive_interval = 5;
@@ -120,14 +145,20 @@ public:
     int etcd_slow_timeout = 5000;
     int etcd_min_reload_interval = 1000;
     bool infinite_start = true;
+    bool use_auth = false;
     uint64_t global_block_size = DEFAULT_BLOCK_SIZE;
     uint32_t global_bitmap_granularity = DEFAULT_BITMAP_GRANULARITY;
     uint32_t global_immediate_commit = IMMEDIATE_NONE;
 
+    uint64_t osd_num = 0;
     std::string etcd_prefix;
+    std::string etcd_client_cert;
+    std::string etcd_client_key;
+    std::string etcd_ca;
     int log_level = 0;
     timerfd_manager_t *tfd = NULL;
 
+    http_context_t *http_ctx = NULL;
     http_co_t *etcd_watch_ws = NULL, *keepalive_client = NULL;
     int etcd_watches_initialised = 0;
     uint64_t etcd_watch_revision_config = 0;
@@ -140,6 +171,7 @@ public:
     std::set<osd_num_t> seen_peers;
     std::map<inode_t, inode_config_t> inode_config;
     std::map<std::string, inode_t> inode_by_name;
+    std::map<std::string, json11::Json> user_info;
     json11::Json node_placement;
 
     std::function<void(std::map<std::string, etcd_kv_t> &)> on_change_hook;
@@ -158,10 +190,12 @@ public:
     std::function<void(http_co_t *)> on_start_watcher_hook;
 
     json11::Json::object serialize_inode_cfg(inode_config_t *cfg);
+    inode_config_t deserialize_inode_cfg(uint64_t inode_num, json11::Json value, uint64_t mod_revision);
     etcd_kv_t parse_etcd_kv(const json11::Json & kv_json);
     std::vector<std::string> get_addresses();
-    void etcd_call_oneshot(std::string etcd_address, std::string api, json11::Json payload, int timeout, std::function<void(std::string, json11::Json)> callback);
+    http_context_t *get_http_ctx();
-    void etcd_call(std::string api, json11::Json payload, int timeout, int retries, int interval, std::function<void(std::string, json11::Json)> callback);
+    void etcd_call_oneshot(const std::string & etcd_address, const std::string & api, json11::Json payload, int timeout, std::function<void(std::string, json11::Json)> callback);
+    void etcd_call(const std::string & api, json11::Json payload, int timeout, int retries, int interval, std::function<void(std::string, json11::Json)> callback);
     void etcd_txn(json11::Json txn, int timeout, int retries, int interval, std::function<void(std::string, json11::Json)> callback);
     void etcd_txn_slow(json11::Json txn, std::function<void(std::string, json11::Json)> callback);
     void start_etcd_watcher();

src/client/http_client.h

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

src/client/messenger.cpp

@@ -14,6 +14,13 @@
 #ifdef WITH_RDMA
 #include "msgr_rdma.h"
 #endif
+#include "http_client.h"
+#ifdef WITH_OPENSSL
+#include <openssl/bio.h>
+#include <openssl/err.h>
+#include <openssl/pem.h>
+#include <openssl/ssl.h>
+#endif
 
 #include <sys/poll.h>
 
@@ -117,6 +124,43 @@ void msgr_iothread_t::run()
 
 void osd_messenger_t::init()
 {
+    if (!tls_cert.empty() || !tls_key.empty() || !osd_tls_ca.empty() || !client_tls_ca.empty())
+    {
+        // Initialize TLS context
+        // FIXME: require OpenSSL
+#ifndef WITH_OPENSSL
+        fprintf(stderr, "Vitastor is built without OpenSSL support\n");
+        exit(1);
+#else
+        if (tls_cert.empty() || tls_key.empty() || osd_tls_ca.empty() || osd_num && client_tls_ca.empty())
+        {
+            if (osd_num)
+                fprintf(stderr, "Vitastor OSD TLS requires osd_tls_cert, osd_tls_key, osd_tls_ca, client_tls_ca\n");
+            else
+                fprintf(stderr, "Vitastor client TLS requires tls_cert, tls_key and osd_tls_ca\n");
+            exit(1);
+        }
+        else
+        {
+            ssl_ctx = SSL_CTX_new(TLS_method());
+            SSL_CTX_set_verify(ssl_ctx, SSL_VERIFY_PEER, NULL);
+            bool ok = SSL_CTX_set_min_proto_version(ssl_ctx, TLS1_3_VERSION);
+            ok = ok && openssl_ctx_add_ca(ssl_ctx, osd_tls_ca);
+            if (osd_num)
+            {
+                // OSD uses 2 separate root certificates to distinguish between clients and peer OSDs
+                ok = ok && openssl_ctx_add_ca(ssl_ctx, client_tls_ca);
+            }
+            ok = ok && openssl_ctx_use_cert(ssl_ctx, tls_cert, tls_cn);
+            ok = ok && openssl_ctx_use_key(ssl_ctx, tls_key);
+            if (!ok)
+            {
+                SSL_CTX_free(ssl_ctx);
+                ssl_ctx = NULL;
+            }
+        }
+#endif
+    }
 #ifdef WITH_RDMACM
     if (use_rdmacm)
     {
@@ -294,6 +338,21 @@ osd_messenger_t::~osd_messenger_t()
         rdma_destroy_event_channel(rdmacm_evch);
         rdmacm_evch = NULL;
     }
+#endif
+    for (auto encrypt_ctx: encrypt_ctx_pool)
+    {
+        destroy_aes_xts_encrypt(encrypt_ctx);
+    }
+    for (auto decrypt_ctx: decrypt_ctx_pool)
+    {
+        destroy_aes_xts_decrypt(decrypt_ctx);
+    }
+#ifdef WITH_OPENSSL
+    if (ssl_ctx)
+    {
+        SSL_CTX_free(ssl_ctx);
+        ssl_ctx = NULL;
+    }
 #endif
 }
 
@@ -329,6 +388,30 @@ void osd_messenger_t::parse_config(const json11::Json & config)
     if (!this->rdma_max_msg || this->rdma_max_msg > 128*1024*1024)
         this->rdma_max_msg = 129*1024;
 #endif
+    this->max_aes_xts_pool_size = config["max_aes_xts_pool_size"].uint64_value();
+    if (!this->max_aes_xts_pool_size)
+        this->max_aes_xts_pool_size = 256;
+    if (config["proto_checksums"].is_null())
+        this->use_proto_checksums = MSGR_CSUM_PAYLOAD;
+    else if (config["proto_checksums"].is_bool())
+        this->use_proto_checksums = config["proto_checksums"].bool_value() ? MSGR_CSUM_FULL : 0;
+    else if (config["proto_checksums"].string_value() != "")
+        this->use_proto_checksums = config["proto_checksums"].string_value() == "full" ? MSGR_CSUM_FULL : MSGR_CSUM_PAYLOAD;
+    else
+        this->use_proto_checksums = 0;
+    if (!osd_num)
+    {
+        tls_cert = config["tls_cert"].string_value();
+        tls_key = config["tls_key"].string_value();
+        osd_tls_ca = config["osd_tls_ca"].string_value();
+    }
+    else
+    {
+        tls_cert = config["osd_tls_cert"].string_value();
+        tls_key = config["osd_tls_key"].string_value();
+        osd_tls_ca = config["osd_tls_ca"].string_value();
+        client_tls_ca = config["client_tls_ca"].string_value();
+    }
     if (!osd_num)
         this->iothread_count = (uint32_t)config["client_iothread_count"].uint64_value();
     else
@@ -510,7 +593,7 @@ void osd_messenger_t::try_connect_peer_tcp(osd_num_t peer_osd, const char *peer_
     cl->peer_state = PEER_CONNECTING;
     cl->connect_timeout_id = -1;
     cl->osd_num = peer_osd;
-    cl->in_buf = malloc_or_die(receive_buffer_size);
+    cl->in_buf = (uint8_t*)malloc_or_die(receive_buffer_size);
     clients[client_id] = cl;
     clients_by_fd[peer_fd] = cl;
     tfd->set_fd_handler(peer_fd, true, [this](int peer_fd, int epoll_events)
@@ -559,6 +642,10 @@ void osd_messenger_t::handle_connect_epoll(int peer_fd)
         handle_peer_epoll(peer_fd, epoll_events);
     });
     // Check OSD number
+    if (!tls_cert.empty())
+    {
+        ssl_init(cl, false);
+    }
     check_peer_config(cl);
 }
 
@@ -648,7 +735,12 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
         // Inform that we're OSD <osd_num>
         payload["osd_num"] = osd_num;
     }
-    payload["features"] = json11::Json::object{ { "check_sequencing", true } };
+    auto features = json11::Json::object{ { "check_sequencing", true } };
+    if (use_proto_checksums)
+    {
+        features["proto_checksums"] = use_proto_checksums;
+    }
+    payload["features"] = features;
 #ifdef WITH_RDMA
     if (!use_rdmacm && rdma_contexts.size())
     {
@@ -723,6 +815,14 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
             delete op;
             return;
         }
+        if (use_proto_checksums)
+        {
+            auto peer_csums = config["features"]["proto_checksums"].uint64_value();
+            if (peer_csums == MSGR_CSUM_FULL && use_proto_checksums == MSGR_CSUM_FULL)
+                cl->proto_csum_status = MSGR_CSUM_FULL;
+            else if (peer_csums && use_proto_checksums)
+                cl->proto_csum_status = MSGR_CSUM_PAYLOAD;
+        }
 #ifdef WITH_RDMA
         if (!use_rdmacm && cl->rdma_conn && config["rdma_address"].is_string())
         {
@@ -786,7 +886,11 @@ void osd_messenger_t::accept_connections(int listen_fd)
         cl->peer_port = ntohs(((sockaddr_in*)&addr)->sin_port);
         cl->peer_fd = peer_fd;
         cl->peer_state = PEER_CONNECTED;
-        cl->in_buf = malloc_or_die(receive_buffer_size);
+        cl->in_buf = (uint8_t*)malloc_or_die(receive_buffer_size);
+        if (!tls_cert.empty())
+        {
+            ssl_init(cl, true);
+        }
         // Add FD to epoll
         tfd->set_fd_handler(peer_fd, false, [this](int peer_fd, int epoll_events)
         {
@@ -801,6 +905,26 @@ void osd_messenger_t::accept_connections(int listen_fd)
     }
 }
 
+void osd_messenger_t::ssl_init(osd_client_t *cl, bool server_mode)
+{
+#ifdef WITH_OPENSSL
+    cl->write_to_ssl = BIO_new(BIO_s_mem());
+    cl->read_from_ssl = BIO_new(BIO_s_mem());
+    cl->ssl_cli = SSL_new(ssl_ctx);
+    if (server_mode)
+    {
+        SSL_set_accept_state(cl->ssl_cli);
+    }
+    else
+    {
+        SSL_set_connect_state(cl->ssl_cli);
+    }
+    SSL_set_bio(cl->ssl_cli, cl->write_to_ssl, cl->read_from_ssl);
+    bool ok = ssl_do_handshake(cl);
+    assert(ok);
+#endif
+}
+
 #ifdef WITH_RDMA
 msgr_rdma_context_t* osd_messenger_t::choose_rdma_context(osd_client_t *cl)
 {

src/client/messenger.h

@@ -12,6 +12,11 @@
 #include <deque>
 #include <vector>
 
+#ifdef WITH_OPENSSL
+#include <openssl/types.h>
+#endif
+
+#include "../util/xxh_x86dispatch.h"
 #include "../util/robin_hood.h"
 #include "malloc_or_die.h"
 #include "json11/json11.hpp"
@@ -31,13 +36,14 @@
 #define PEER_RDMA 4
 #define PEER_STOPPED 5
 
+#define MSGR_CSUM_PAYLOAD 1
+#define MSGR_CSUM_FULL 2
+#define MSGR_CSUM_NEG 4
+
 #define VITASTOR_CONFIG_PATH "/etc/vitastor/vitastor.conf"
 
 #define DEFAULT_MIN_ZEROCOPY_SEND_SIZE 32*1024
 
-#define MSGR_SENDP_HDR 1
-#define MSGR_SENDP_FREE 2
-
 struct msgr_sendp_t
 {
     osd_op_t *op;
@@ -49,6 +55,17 @@ struct msgr_rdma_connection_t;
 struct msgr_rdma_context_t;
 #endif
 
+struct op_aes_xts_encrypt_t;
+struct op_aes_xts_decrypt_t;
+void destroy_aes_xts_encrypt(op_aes_xts_encrypt_t *encrypt_ctx);
+void destroy_aes_xts_decrypt(op_aes_xts_decrypt_t *decrypt_ctx);
+
+struct __attribute__((__packed__)) msgr_tls_record_hdr_t
+{
+    uint8_t encrypted;
+    uint32_t size;
+};
+
 struct osd_client_t
 {
     uint64_t client_id = 0;
@@ -65,23 +82,40 @@ struct osd_client_t
     osd_num_t in_osd_num = 0;
     bool is_incoming = false;
 
-    void *in_buf = NULL;
+    uint8_t *in_buf = NULL;
 
 #ifdef WITH_RDMA
     msgr_rdma_connection_t *rdma_conn = NULL;
 #endif
 
+#ifdef WITH_OPENSSL
+    SSL *ssl_cli = NULL;
+    BIO *write_to_ssl = NULL;
+    // FIXME: use custom bio to avoid 1 more memory copy?
+    BIO *read_from_ssl = NULL;
+    uint8_t *ssl_out_buf = NULL;
+    size_t ssl_out_buf_size = 0, ssl_out_buf_cap = 0;
+    bool ssl_handshake_done = false;
+    bool ssl_want_write = false;
+#endif
+
     // Read state
     int read_ready = 0;
     osd_op_t *read_op = NULL;
+    size_t read_op_size = 0;
+    size_t read_op_pos = 0;
     iovec read_iov = { 0 };
     msghdr read_msg = { 0 };
-    int read_remaining = 0;
+    std::vector<iovec> recv_list;
-    int read_state = 0;
+    size_t recv_list_size = 0;
-    osd_op_buf_list_t recv_list;
     uint64_t read_op_id = 1;
     bool check_sequencing = false;
     bool enable_pg_locks = false;
+    op_aes_xts_decrypt_t *decrypt_ctx = NULL;
+    size_t read_op_inline_decrypt_pos = 0;
+    size_t read_op_inline_decrypt_in = 0;
+    int proto_csum_status = 0;
+    XXH3_state_t* read_csum_state = NULL;
 
     // Incoming operations
     std::vector<osd_op_t*> received_ops;
@@ -94,11 +128,17 @@ struct osd_client_t
     std::set<pool_pg_num_t> dirty_pgs;
 
     // Write state
+    std::deque<osd_op_t *> write_ops;
+    osd_op_t *write_op = NULL;
+    size_t write_op_pos = 0;
     msghdr write_msg = { 0 };
     int write_state = 0;
-    std::vector<iovec> send_list, next_send_list;
+    std::vector<iovec> send_list;
-    std::vector<msgr_sendp_t> outbox, next_outbox;
+    size_t send_list_size = 0;
+    std::deque<osd_op_t*> send_free_ops;
     std::vector<osd_op_t*> zc_free_list;
+    op_aes_xts_encrypt_t *encrypt_ctx = NULL;
+    XXH3_state_t* write_csum_state = NULL;
 
     ~osd_client_t();
     void cancel_ops();
@@ -190,6 +230,12 @@ protected:
     bool use_sync_send_recv = false;
     int min_zerocopy_send_size = DEFAULT_MIN_ZEROCOPY_SEND_SIZE;
     int iothread_count = 0;
+    int max_aes_xts_pool_size = 256;
+
+    std::string tls_cert;
+    std::string tls_key;
+    std::string osd_tls_ca;
+    std::string client_tls_ca;
 
 #ifdef WITH_RDMA
     bool use_rdma = true;
@@ -207,12 +253,28 @@ protected:
     robin_hood::unordered_flat_map<rdma_cm_id*, rdmacm_connecting_t*> rdmacm_connecting;
 #endif
 
+#ifdef WITH_OPENSSL
+    SSL_CTX *ssl_ctx = NULL;
+    std::string tls_cn;
+
+    void ssl_init(osd_client_t *cl, bool server_mode);
+    bool ssl_do_handshake(osd_client_t *cl);
+    bool ssl_do_encrypt(osd_client_t *cl);
+    size_t ssl_do_encrypt_to(osd_client_t *cl, uint8_t *buf, size_t size);
+    bool ssl_op_write_buf(osd_client_t *cl, uint8_t *src, size_t src_len, bool skip_csum, size_t & from, size_t & done);
+    size_t ssl_op_copy_to(osd_client_t *cl, uint8_t *dst, size_t dst_len);
+    void ssl_op_get_write_buffers(osd_client_t *cl, std::vector<iovec> & lst);
+#endif
+
     std::vector<msgr_iothread_t*> iothreads;
     std::vector<uint64_t> read_ready_clients;
     std::vector<uint64_t> write_ready_clients;
     // We don't use ringloop->set_immediate here because we may have no ringloop in client :)
     std::deque<osd_op_t*> set_immediate_ops;
 
+    std::vector<op_aes_xts_encrypt_t*> encrypt_ctx_pool;
+    std::vector<op_aes_xts_decrypt_t*> decrypt_ctx_pool;
+
 public:
     timerfd_manager_t *tfd = NULL;
     ring_loop_t *ringloop = NULL;
@@ -230,6 +292,7 @@ public:
     std::vector<addr_mask_t> osd_cluster_network_masks;
     std::vector<std::string> all_osd_networks;
     std::vector<addr_mask_t> all_osd_network_masks;
+    int use_proto_checksums = 0;
     // op statistics
     osd_op_stats_t stats, recovery_stats;
 
@@ -279,17 +342,32 @@ protected:
 
     bool try_send(osd_client_t *cl);
     void handle_send(int result, bool prev, bool more, osd_client_t *cl);
+    size_t op_copy_to(osd_client_t *cl, uint8_t *dst, size_t dst_len);
+    void op_get_write_buffers(osd_client_t *cl, std::vector<iovec> & lst);
+    void next_write_op(osd_client_t *cl);
+    bool op_write_buf(osd_client_t *cl, uint8_t *src, size_t src_len, uint8_t *dst, size_t dst_len, bool skip_csum, size_t & from, size_t & done);
+    bool op_copy_data_to(osd_client_t *cl, uint8_t *dst, size_t dst_len, size_t & from, size_t & done);
 
-    bool handle_read(int result, osd_client_t *cl);
+    void handle_read(int result, osd_client_t *cl);
-    bool handle_read_buffer(osd_client_t *cl, void *curbuf, int remain);
+    bool handle_read_buffer(osd_client_t *cl, uint8_t *curbuf, size_t bufsize);
-    bool handle_finished_read(osd_client_t *cl);
+    bool handle_hdr(osd_client_t *cl);
-    void handle_op_hdr(osd_client_t *cl);
+    bool allocate_op_buffers(osd_client_t *cl);
-    bool handle_reply_hdr(osd_client_t *cl);
+    bool allocate_reply_buffers(osd_client_t *cl, osd_op_t *op);
-    void handle_reply_ready(osd_op_t *op);
+    bool op_copy_from(osd_client_t *cl, uint8_t *src, size_t src_len, size_t & done);
+    void op_get_read_buffers(osd_client_t *cl, std::vector<iovec> & lst);
+    void op_alloc_temp_buffers(osd_op_t *op, int i);
+    bool handle_finished_op(osd_client_t *cl);
     void handle_immediate_ops();
 
+    bool op_encrypted_copy_data_to(osd_client_t* cl, uint8_t *buf, size_t len, size_t from, size_t & done);
+    bool op_decrypted_copy_data_from(osd_client_t* cl, uint8_t *buf, size_t len, size_t from, size_t & done);
+    void op_decrypt_start(osd_client_t* cl);
+    void op_decrypt_inline(osd_client_t* cl);
+    void op_decrypt_free(osd_client_t* cl);
+
 #ifdef WITH_RDMA
     void try_send_rdma(osd_client_t *cl);
+    int try_send_rdma_copy(osd_client_t *cl, uint8_t *dst, int dst_len);
     bool init_recv_rdma(osd_client_t *cl);
     void handle_rdma_events(msgr_rdma_context_t *rdma_context);
     msgr_rdma_context_t* choose_rdma_context(osd_client_t *cl);

src/client/msgr_encrypt.cpp

@@ -0,0 +1,484 @@
+// Copyright (c) Vitaliy Filippov, 2026+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#include <assert.h>
+
+#include "etcd_state_client.h"
+#include "messenger.h"
+#include "msgr_encrypt.h"
+
+op_aes_xts_encrypt_t::op_aes_xts_encrypt_t()
+{
+#ifdef WITH_OPENSSL
+    if (!(ctx = EVP_CIPHER_CTX_new()))
+    {
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+    EVP_CIPHER_CTX_set_padding(ctx, 0);
+    if (EVP_EncryptInit_ex(ctx, EVP_aes_256_xts(), NULL, NULL, NULL) != 1)
+    {
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+#else
+    fprintf(stderr, "Error: Vitastor is built without encryption support\n");
+    abort();
+#endif
+}
+
+op_aes_xts_encrypt_t::~op_aes_xts_encrypt_t()
+{
+    assert(!encrypted);
+#ifdef WITH_OPENSSL
+    EVP_CIPHER_CTX_free(ctx);
+#endif
+    if (tmp)
+        free(tmp);
+}
+
+void op_aes_xts_encrypt_t::start(uint8_t *key, uint64_t start_offset, size_t block_size)
+{
+    assert(!encrypted);
+    this->start_offset = start_offset;
+    this->key = key;
+    this->block_size = block_size;
+    this->offset = 0;
+    this->encrypted = false;
+    this->tmp_pos = 0;
+    if (tmp && tmp_size != block_size)
+    {
+        free(tmp);
+        tmp = NULL;
+        tmp_size = 0;
+    }
+#ifdef WITH_OPENSSL
+    if (EVP_EncryptInit_ex(ctx, NULL, NULL, key, NULL) != 1)
+    {
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+#endif
+}
+
+void op_aes_xts_encrypt_t::encrypt_block(uint8_t *in, uint8_t *out)
+{
+#ifdef WITH_OPENSSL
+    uint8_t iv[16] = { 0 };
+    *((uint64_t*)iv) = start_offset + offset - offset%block_size;
+    if (EVP_EncryptInit_ex(ctx, NULL, NULL, NULL, iv) != 1)
+    {
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+    int actual_out = 0;
+    if (EVP_EncryptUpdate(ctx, out, &actual_out, in, block_size) != 1)
+    {
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+    assert(actual_out == block_size);
+#endif
+}
+
+void op_aes_xts_encrypt_t::update(uint8_t *in, size_t max_in, uint8_t *out, size_t max_out, size_t & done_in, size_t & done_out)
+{
+    // Fucking AES-XTS implementations (all of them) don't have streaming support,
+    // crafting IV to resume encryption is slow, so we have to accumulate a full block
+    // and encrypt it at once :-(
+    // And then we have to support consuming it in parts because it's simpler for the
+    // higher layers.
+    if (encrypted)
+    {
+        // Copy accumulated and encrypted output
+        assert(tmp);
+        if (max_out > block_size - tmp_pos)
+            max_out = block_size - tmp_pos;
+        memcpy(out, tmp + tmp_pos, max_out);
+        done_out += max_out;
+        tmp_pos += max_out;
+        if (tmp_pos >= block_size)
+            encrypted = false;
+    }
+    else if (max_in < block_size - offset%block_size)
+    {
+        // Just accumulate input
+        if (!tmp)
+        {
+            tmp = (uint8_t*)malloc_or_die(block_size);
+            tmp_size = block_size;
+        }
+        memcpy(tmp + offset%block_size, in, max_in);
+        done_in += max_in;
+        offset += max_in;
+    }
+    else if (max_out < block_size)
+    {
+        // Accumulate and encrypt input in <tmp>, then copy part of it to <out>
+        if (!tmp)
+        {
+            tmp = (uint8_t*)malloc_or_die(block_size);
+            tmp_size = block_size;
+        }
+        max_in = block_size - offset%block_size;
+        memcpy(tmp + offset%block_size, in, max_in);
+        encrypt_block(tmp, tmp);
+        encrypted = true;
+        memcpy(out, tmp, max_out);
+        tmp_pos = max_out;
+        done_in += max_in;
+        offset += max_in;
+        done_out += max_out;
+    }
+    else if (!(offset%block_size))
+    {
+        // Full block - simplest case
+        encrypt_block(in, out);
+        done_in += block_size;
+        offset += block_size;
+        done_out += block_size;
+    }
+    else
+    {
+        // Accumulate input and encrypt directly to <output>
+        assert(tmp);
+        max_in = block_size - offset%block_size;
+        memcpy(tmp + offset%block_size, in, max_in);
+        encrypt_block(tmp, out);
+        done_in += max_in;
+        offset += max_in;
+        done_out += block_size;
+    }
+}
+
+void destroy_aes_xts_encrypt(op_aes_xts_encrypt_t *encrypt_ctx)
+{
+    delete encrypt_ctx;
+}
+
+op_aes_xts_decrypt_t::op_aes_xts_decrypt_t()
+{
+#ifdef WITH_OPENSSL
+    if (!(ctx = EVP_CIPHER_CTX_new()))
+    {
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+    EVP_CIPHER_CTX_set_padding(ctx, 0);
+    if (EVP_DecryptInit_ex(ctx, EVP_aes_256_xts(), NULL, NULL, NULL) != 1)
+    {
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+#else
+    fprintf(stderr, "Error: Vitastor is built without encryption support\n");
+    abort();
+#endif
+}
+
+op_aes_xts_decrypt_t::~op_aes_xts_decrypt_t()
+{
+    assert(!decrypted);
+#ifdef WITH_OPENSSL
+    EVP_CIPHER_CTX_free(ctx);
+#endif
+    if (tmp)
+        free(tmp);
+}
+
+void op_aes_xts_decrypt_t::start(uint8_t **key_chain, size_t chain_size, uint8_t *key_indexes, uint64_t start_offset, size_t block_size)
+{
+    assert(!decrypted);
+    this->start_offset = start_offset;
+    this->key_chain = chain_size > 1 ? key_chain : 0;
+    this->chain_size = chain_size > 1 ? chain_size : 0;
+    this->key_indexes = chain_size > 1 ? key_indexes : NULL;
+    assert(chain_size <= 1 || key_indexes != NULL);
+    this->block_size = block_size;
+    this->offset = 0;
+    this->tmp_pos = 0;
+    if (tmp && tmp_size != block_size)
+    {
+        free(tmp);
+        tmp = NULL;
+        tmp_size = 0;
+    }
+#ifdef WITH_OPENSSL
+    if (chain_size == 1 && key_chain[0] && EVP_DecryptInit_ex(ctx, NULL, NULL, key_chain[0], NULL) != 1)
+    {
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+#endif
+}
+
+void op_aes_xts_decrypt_t::decrypt_block(uint8_t *in, uint8_t *out)
+{
+    uint8_t *key = NULL;
+    if (chain_size)
+    {
+        assert(key_indexes[offset/block_size] < chain_size);
+        key = key_chain[key_indexes[offset/block_size]];
+        if (!key)
+        {
+            if (in != out)
+                memcpy(out, in, block_size);
+            return;
+        }
+    }
+#ifdef WITH_OPENSSL
+    uint8_t iv[16] = { 0 };
+    *((uint64_t*)iv) = start_offset + offset - offset%block_size;
+    if (EVP_DecryptInit_ex(ctx, NULL, NULL, key, iv) != 1)
+    {
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+    int actual_out = 0;
+    if (EVP_DecryptUpdate(ctx, out, &actual_out, in, block_size) != 1)
+    {
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+    assert(actual_out == block_size);
+#endif
+}
+
+// out may be NULL, in this case all input is still decrypted to calculate checksums,
+// but part of it is skipped and not copied to out
+void op_aes_xts_decrypt_t::update(uint8_t *in, size_t max_in, uint8_t *out, size_t max_out, size_t & done_in, size_t & done_out)
+{
+    // Fucking AES-XTS implementations (all of them) don't have streaming support,
+    // crafting IV to resume decryption is slow, so we have to accumulate a full block
+    // and decrypt it at once :-(
+    // And then we have to support consuming it in parts because clients sometimes need
+    // fragmented output.
+    if (decrypted)
+    {
+        // Copy accumulated and decrypted output
+        assert(tmp);
+        if (max_out > block_size - tmp_pos)
+            max_out = block_size - tmp_pos;
+        if (out)
+            memcpy(out, tmp + tmp_pos, max_out);
+        done_out += max_out;
+        tmp_pos += max_out;
+        if (tmp_pos >= block_size)
+            decrypted = false;
+    }
+    else if (max_in < block_size - offset%block_size)
+    {
+        // Just accumulate input
+        if (!tmp)
+        {
+            tmp = (uint8_t*)malloc_or_die(block_size);
+            tmp_size = block_size;
+        }
+        memcpy(tmp + offset%block_size, in, max_in);
+        done_in += max_in;
+        offset += max_in;
+    }
+    else if (max_out < block_size || !out)
+    {
+        // Accumulate and decrypt input in <tmp>, then copy part of it to <out>
+        if (!tmp)
+        {
+            tmp = (uint8_t*)malloc_or_die(block_size);
+            tmp_size = block_size;
+        }
+        max_in = block_size - offset%block_size;
+        memcpy(tmp + offset%block_size, in, max_in);
+        decrypt_block(tmp, tmp);
+        decrypted = true;
+        if (out)
+            memcpy(out, tmp, max_out);
+        tmp_pos = max_out;
+        done_in += max_in;
+        offset += max_in;
+        done_out += max_out;
+    }
+    else if (!(offset%block_size))
+    {
+        // Full block - simplest case
+        if (out)
+            decrypt_block(in, out);
+        done_in += block_size;
+        offset += block_size;
+        done_out += block_size;
+    }
+    else
+    {
+        // Accumulate input and decrypt directly to <output>
+        assert(tmp);
+        max_in = block_size - offset%block_size;
+        memcpy(tmp + offset%block_size, in, max_in);
+        assert(out);
+        decrypt_block(tmp, out);
+        done_in += max_in;
+        offset += max_in;
+        done_out += block_size;
+    }
+}
+
+void destroy_aes_xts_decrypt(op_aes_xts_decrypt_t *decrypt_ctx)
+{
+    delete decrypt_ctx;
+}
+
+bool osd_messenger_t::op_encrypted_copy_data_to(osd_client_t* cl, uint8_t *enc_buf, size_t enc_len, size_t from, size_t & done)
+{
+    auto op = cl->write_op;
+    auto & op_pos = cl->write_op_pos;
+    assert(op->req.hdr.opcode == OSD_OP_WRITE);
+    if (!from)
+    {
+        if (!cl->encrypt_ctx)
+        {
+            if (encrypt_ctx_pool.size())
+            {
+                cl->encrypt_ctx = encrypt_ctx_pool.back();
+                encrypt_ctx_pool.pop_back();
+            }
+            else
+                cl->encrypt_ctx = new op_aes_xts_encrypt_t();
+        }
+        assert(op->enc->key_chain[0]);
+        cl->encrypt_ctx->start(op->enc->key_chain[0], op->req.rw.offset, op->enc->bitmap_granularity);
+    }
+    for (int i = 0; i < op->iov.count; i++)
+    {
+        uint8_t *plain = (uint8_t*)op->iov.buf[i].iov_base;
+        size_t plain_len = op->iov.buf[i].iov_len;
+        while (from < plain_len || cl->encrypt_ctx->has_buffered())
+        {
+            if (done >= enc_len)
+                return false;
+            size_t done_in = 0;
+            size_t done_out = 0;
+            cl->encrypt_ctx->update(plain+from, plain_len-from, enc_buf+done, enc_len-done, done_in, done_out);
+            if (cl->write_csum_state && done_out > 0)
+                XXH3_64bits_update(cl->write_csum_state, enc_buf+done, done_out);
+            done += done_out;
+            op_pos += done_in;
+            from += done_in;
+        }
+        from -= plain_len;
+    }
+    if (cl->encrypt_ctx)
+    {
+        if (encrypt_ctx_pool.size() > max_aes_xts_pool_size)
+            delete cl->encrypt_ctx;
+        else
+            encrypt_ctx_pool.push_back(cl->encrypt_ctx);
+        cl->encrypt_ctx = NULL;
+    }
+    return true;
+}
+
+bool osd_messenger_t::op_decrypted_copy_data_from(osd_client_t* cl, uint8_t *enc_buf, size_t enc_len, size_t from, size_t & done)
+{
+    op_decrypt_start(cl);
+    auto op = cl->read_op;
+    assert(op->req.hdr.opcode == OSD_OP_READ);
+    for (int i = 0; i < op->iov.count; i++)
+    {
+        uint8_t *plain = (uint8_t*)op->iov.buf[i].iov_base;
+        size_t plain_len = op->iov.buf[i].iov_len;
+        while (from < plain_len)
+        {
+            if (done >= enc_len)
+                return false;
+            size_t done_in = 0;
+            size_t done_out = 0;
+            // plain == NULL means skip output
+            cl->decrypt_ctx->update(enc_buf+done, enc_len-done, plain ? plain+from : NULL, plain_len-from, done_in, done_out);
+            if (cl->read_csum_state && done_in > 0)
+                XXH3_64bits_update(cl->read_csum_state, enc_buf+done, done_in);
+            done += done_in;
+            cl->read_op_pos += done_out;
+            cl->read_op_inline_decrypt_in += done_in;
+            from += done_out;
+            if (!done_out)
+                return false;
+        }
+        from -= plain_len;
+    }
+    op_decrypt_free(cl);
+    return true;
+}
+
+void osd_messenger_t::op_decrypt_start(osd_client_t* cl)
+{
+    if (!cl->decrypt_ctx)
+    {
+        if (decrypt_ctx_pool.size())
+        {
+            cl->decrypt_ctx = decrypt_ctx_pool.back();
+            decrypt_ctx_pool.pop_back();
+        }
+        else
+            cl->decrypt_ctx = new op_aes_xts_decrypt_t();
+        auto & enc = cl->read_op->enc;
+        cl->decrypt_ctx->start(enc->key_chain, enc->chain_size,
+            (cl->read_op->req.rw.flags & OSD_OP_RETURN_CHAIN) ? (uint8_t*)cl->read_op->bitmap + enc->read_chain_bitmap_pos : 0,
+            cl->read_op->req.rw.offset, enc->bitmap_granularity);
+    }
+}
+
+void osd_messenger_t::op_decrypt_inline(osd_client_t* cl)
+{
+    op_decrypt_start(cl);
+    osd_op_t *op = cl->read_op;
+    size_t from_in = cl->read_op_inline_decrypt_in;
+    int i = 0;
+    while (i < op->iov.count && from_in >= op->iov.buf[i].iov_len)
+    {
+        from_in -= op->iov.buf[i].iov_len;
+        i++;
+    }
+    size_t from_out = cl->read_op_inline_decrypt_pos - OSD_PACKET_SIZE - op->reply.rw.bitmap_len;
+    int j = 0;
+    while (j < op->iov.count && from_out >= op->iov.buf[j].iov_len)
+    {
+        from_out -= op->iov.buf[j].iov_len;
+        j++;
+    }
+    while (i < op->iov.count && j < op->iov.count)
+    {
+        uint8_t *in = (uint8_t*)op->iov.buf[i].iov_base + from_in;
+        size_t in_len = op->iov.buf[i].iov_len - from_in;
+        uint8_t *out = (uint8_t*)op->iov.buf[j].iov_base + from_out;
+        size_t out_len = op->iov.buf[j].iov_len - from_out;
+        size_t done_in = 0;
+        size_t done_out = 0;
+        cl->decrypt_ctx->update(in, in_len, out, out_len, done_in, done_out);
+        if (done_in >= in_len)
+        {
+            i++;
+            from_in = 0;
+        }
+        else
+            from_in += done_in;
+        if (done_out >= out_len)
+        {
+            j++;
+            from_out = 0;
+        }
+        else
+            from_out += done_out;
+    }
+    assert(j >= op->iov.count);
+    op_decrypt_free(cl);
+}
+
+void osd_messenger_t::op_decrypt_free(osd_client_t* cl)
+{
+    if (cl->decrypt_ctx)
+    {
+        if (decrypt_ctx_pool.size() > max_aes_xts_pool_size)
+            delete cl->decrypt_ctx;
+        else
+            decrypt_ctx_pool.push_back(cl->decrypt_ctx);
+        cl->decrypt_ctx = NULL;
+    }
+}

src/client/msgr_encrypt.h

@@ -0,0 +1,68 @@
+// Copyright (c) Vitaliy Filippov, 2026+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#include <stdint.h>
+
+#include "../util/xxh_x86dispatch.h"
+// WITH_OPENSSL is left to possibly support other crypto libraries
+#ifdef WITH_OPENSSL
+#include <openssl/conf.h>
+#include <openssl/evp.h>
+#include <openssl/err.h>
+#endif
+
+class op_aes_xts_encrypt_t
+{
+#ifdef WITH_OPENSSL
+    EVP_CIPHER_CTX *ctx = NULL;
+#endif
+    uint64_t start_offset = 0;
+    uint8_t *key = NULL;
+    size_t offset = 0;
+    size_t block_size = 0;
+    uint8_t *tmp = NULL;
+    size_t tmp_size = 0;
+    size_t tmp_pos = 0;
+    bool encrypted = false;
+
+    void encrypt_block(uint8_t *in, uint8_t *out);
+
+public:
+    op_aes_xts_encrypt_t();
+    ~op_aes_xts_encrypt_t();
+
+    inline bool has_buffered() { return encrypted; };
+    void start(uint8_t *key, uint64_t start_offset, size_t block_size);
+    void update(uint8_t *in, size_t max_in, uint8_t *out, size_t max_out, size_t & done_in, size_t & done_out);
+};
+
+void destroy_aes_xts_encrypt(op_aes_xts_encrypt_t *encrypt_ctx);
+
+class op_aes_xts_decrypt_t
+{
+#ifdef WITH_OPENSSL
+    EVP_CIPHER_CTX *ctx = NULL;
+#endif
+    uint64_t start_offset = 0;
+    uint8_t **key_chain = NULL;
+    size_t chain_size = 0;
+    uint8_t *key_indexes = NULL;
+    size_t offset = 0;
+    size_t block_size = 0;
+    uint8_t *tmp = NULL;
+    size_t tmp_size = 0;
+    size_t tmp_pos = 0;
+    bool decrypted = false;
+
+    void decrypt_block(uint8_t *in, uint8_t *out);
+
+public:
+    op_aes_xts_decrypt_t();
+    ~op_aes_xts_decrypt_t();
+
+    inline bool has_buffered() { return decrypted; };
+    void start(uint8_t **key_chain, size_t chain_size, uint8_t *key_indexes, uint64_t start_offset, size_t block_size);
+    void update(uint8_t *in, size_t max_in, uint8_t *out, size_t max_out, size_t & done_in, size_t & done_out);
+};
+
+void destroy_aes_xts_decrypt(op_aes_xts_decrypt_t *decrypt_ctx);

src/client/msgr_op.cpp

@@ -22,6 +22,10 @@ osd_op_t::~osd_op_t()
         // So we don't reuse it, but free it every time
         free(buf);
     }
+    if (enc_buf)
+    {
+        free(enc_buf);
+    }
     if (op_data)
     {
         free(op_data);

src/client/msgr_op.h

@@ -3,6 +3,8 @@
 
 #pragma once
 
+#include <memory>
+
 #include <sys/uio.h>
 #include <stdint.h>
 #include <stdio.h>
@@ -16,6 +18,8 @@
 
 #define OSD_OP_INLINE_BUF_COUNT 16
 
+#define AES_256_XTS_KEY_SIZE 64
+
 // Kind of a vector with small-list-optimisation
 struct osd_op_buf_list_t
 {
@@ -152,6 +156,19 @@ struct blockstore_op_t;
 
 struct osd_primary_op_data_t;
 
+struct osd_op_enc_t
+{
+    // Keys may contain more information in the future, like encryption algorithm and key ID
+    // In this case, key_chain will become inode_key_t* with inode_key_t also being a structure
+    // Currently all keys are required to be 512 bit (64 byte) long, for AES-256-XTS
+    // Raw pointers are convenient for messenger code; external users may use shared_ptr aliasing
+    // to implement complex freeing of osd_op_enc_t along with their external inode cache info
+    uint8_t** key_chain = NULL;
+    size_t chain_size = 0;
+    uint32_t read_chain_bitmap_pos = 0;
+    uint32_t bitmap_granularity = 0;
+};
+
 struct __attribute__((visibility("default"))) osd_op_t
 {
     timespec tv_begin = { 0 }, tv_end = { 0 };
@@ -167,6 +184,9 @@ struct __attribute__((visibility("default"))) osd_op_t
     unsigned bmp_data = 0;
     void *bitmap_buf = NULL;
     void *rmw_buf = NULL;
+    std::shared_ptr<osd_op_enc_t> enc;
+    uint8_t *enc_buf = NULL;
+    uint64_t csum = 0; // network layer checksum
     osd_primary_op_data_t* op_data = NULL;
     std::function<void(osd_op_t*)> callback;
 
@@ -176,4 +196,5 @@ struct __attribute__((visibility("default"))) osd_op_t
     void cancel();
 
     bool is_recovery_related();
+    uint64_t calc_data_checksum();
 };

src/client/msgr_rdma.cpp

@@ -187,6 +187,8 @@ std::vector<msgr_rdma_context_t*> msgr_rdma_context_t::create_all(const std::vec
     ibv_device **raw_dev_list = NULL;
     ibv_device **dev_list = NULL;
     ibv_device *single_list[2] = {};
+    int up_ports = 0;
+    int single_port_num = 0;
 
     raw_dev_list = dev_list = ibv_get_device_list(NULL);
     if (!dev_list || !*dev_list)
@@ -221,6 +223,7 @@ std::vector<msgr_rdma_context_t*> msgr_rdma_context_t::create_all(const std::vec
         dev_list = single_list;
     }
 
+retry:
     for (int i = 0; dev_list[i]; ++i)
     {
         auto dev = dev_list[i];
@@ -258,6 +261,9 @@ std::vector<msgr_rdma_context_t*> msgr_rdma_context_t::create_all(const std::vec
                 fprintf(stderr, "RDMA device %s port %d GID %d does not exist\n", ibv_get_device_name(dev), port_num, sel_gid_index);
                 continue;
             }
+            up_ports++;
+            single_port_num = port_num;
+            single_list[0] = dev;
             uint32_t port_mtu = sel_mtu ? sel_mtu : ibv_mtu_to_bytes(portinfo.active_mtu);
 #ifdef IBV_ADVISE_MR_ADVICE_PREFETCH_NO_FAULT
             if (sel_gid_index < 0)
@@ -298,6 +304,14 @@ cleanup_dev:
         ibv_close_device(context);
     }
 
+    if (!ret.size() && up_ports == 1 && dev_list != single_list)
+    {
+        // Auto-select the only available device/port if there is only one
+        dev_list = single_list;
+        sel_port_num = single_port_num;
+        goto retry;
+    }
+
 cleanup:
     if (raw_dev_list)
         ibv_free_device_list(raw_dev_list);
@@ -554,23 +568,24 @@ static void try_send_rdma_wr(osd_client_t *cl, ibv_sge *sge, int op_sge)
     cl->rdma_conn->cur_send++;
 }
 
-static int try_send_rdma_copy(osd_client_t *cl, uint8_t *dst, int dst_len)
+int osd_messenger_t::try_send_rdma_copy(osd_client_t *cl, uint8_t *dst, int dst_len)
 {
-    auto rc = cl->rdma_conn;
     int total_dst_len = dst_len;
-    while (dst_len > 0 && rc->send_pos < cl->send_list.size())
+    while (dst_len > 0 && (cl->write_op || cl->write_ops.size()))
     {
-        iovec & iov = cl->send_list[rc->send_pos];
+        next_write_op(cl);
-        uint32_t len = (uint32_t)(iov.iov_len-rc->send_buf_pos < dst_len
+        osd_op_t *op = cl->write_op;
-            ? iov.iov_len-rc->send_buf_pos : dst_len);
+        size_t copied = op_copy_to(cl, dst, dst_len);
-        memcpy(dst, (uint8_t*)iov.iov_base+rc->send_buf_pos, len);
+        if (!copied)
-        dst += len;
-        dst_len -= len;
-        rc->send_buf_pos += len;
-        if (rc->send_buf_pos >= iov.iov_len)
         {
-            rc->send_pos++;
+            break;
-            rc->send_buf_pos = 0;
+        }
+        dst += copied;
+        dst_len -= copied;
+        if (!cl->write_op && op->op_type == OSD_OP_IN)
+        {
+            // this is a reply, free the op after sending it
+            cl->send_free_ops.push_back(op);
         }
     }
     return total_dst_len-dst_len;
@@ -617,6 +632,7 @@ void osd_messenger_t::try_send_rdma(osd_client_t *cl)
             };
             try_send_rdma_wr(cl, &sge, 1);
             rc->send_sizes.push_back(copied);
+            cl->send_free_ops.push_back(NULL); // end marker
         }
     }
 }
@@ -713,9 +729,6 @@ void osd_messenger_t::handle_rdma_events(msgr_rdma_context_t *rdma_context)
             }
             if (!is_send)
             {
-                // Reset OSD ping state - client is obviously alive
-                cl->ping_time_remaining = 0;
-                cl->idle_time_remaining = osd_idle_timeout;
                 rc->cur_recv--;
                 if (!handle_read_buffer(cl, rc->recv_buffers[rc->next_recv_buf], wc[i].byte_len))
                 {
@@ -727,54 +740,27 @@ void osd_messenger_t::handle_rdma_events(msgr_rdma_context_t *rdma_context)
             else
             {
                 rc->cur_send--;
-                uint64_t sent_size = rc->send_sizes.at(0);
+                // byte_len is not filled for send operations
-                rc->send_sizes.erase(rc->send_sizes.begin(), rc->send_sizes.begin()+1);
+                uint64_t sent_size = rc->send_sizes.front();
+                rc->send_sizes.pop_front();
                 rc->send_done_pos += sent_size;
                 rc->send_out_full = false;
                 if (rc->send_done_pos == rc->send_out_size)
                     rc->send_done_pos = 0;
                 assert(rc->send_done_pos < rc->send_out_size);
-                int send_pos = 0, send_buf_pos = 0;
+                while (cl->send_free_ops.front())
-                while (sent_size > 0)
                 {
-                    if (sent_size >= cl->send_list.at(send_pos).iov_len)
+                    delete cl->send_free_ops.front();
-                    {
+                    cl->send_free_ops.pop_front();
-                        sent_size -= cl->send_list[send_pos].iov_len;
-                        send_pos++;
-                    }
-                    else
-                    {
-                        send_buf_pos = sent_size;
-                        sent_size = 0;
-                    }
                 }
-                assert(rc->send_pos >= send_pos);
+                cl->send_free_ops.pop_front();
-                if (rc->send_pos == send_pos)
+                if ((cl->proto_csum_status & MSGR_CSUM_NEG) && !cl->write_op && !cl->write_ops.size())
                 {
-                    rc->send_buf_pos -= send_buf_pos;
+                    // Checksums negotiated, enable
-                }
+                    cl->proto_csum_status = cl->proto_csum_status & (~MSGR_CSUM_NEG);
-                rc->send_pos -= send_pos;
-                for (int i = 0; i < send_pos; i++)
-                {
-                    if (cl->outbox[i].flags & MSGR_SENDP_FREE)
-                    {
-                        // Reply fully sent
-                        delete cl->outbox[i].op;
-                    }
-                }
-                if (send_pos > 0)
-                {
-                    cl->send_list.erase(cl->send_list.begin(), cl->send_list.begin()+send_pos);
-                    cl->outbox.erase(cl->outbox.begin(), cl->outbox.begin()+send_pos);
-                }
-                if (send_buf_pos > 0)
-                {
-                    cl->send_list[0].iov_base = (uint8_t*)cl->send_list[0].iov_base + send_buf_pos;
-                    cl->send_list[0].iov_len -= send_buf_pos;
                 }
                 try_send_rdma(cl);
             }
         }
     } while (event_count > 0);
-    handle_immediate_ops();
 }

src/client/msgr_rdma.h

@@ -8,8 +8,11 @@
 #include <infiniband/verbs.h>
 #include <string>
 #include <vector>
+#include <deque>
 #include "addr_util.h"
 
+struct osd_op_t;
+
 struct msgr_rdma_address_t
 {
     ibv_gid gid;
@@ -72,9 +75,9 @@ struct msgr_rdma_connection_t
     int cur_send = 0, cur_recv = 0;
     int send_pos = 0, send_buf_pos = 0;
     int next_recv_buf = 0;
-    std::vector<void*> recv_buffers;
+    std::vector<uint8_t*> recv_buffers;
     msgr_rdma_buf_t recv_buf;
-    std::vector<uint64_t> send_sizes;
+    std::deque<uint64_t> send_sizes;
     msgr_rdma_buf_t send_out;
     int send_out_pos = 0, send_done_pos = 0, send_out_size = 0;
     bool send_out_full = false;

src/client/msgr_rdmacm.cpp

@@ -495,7 +495,7 @@ void osd_messenger_t::rdmacm_established(rdma_cm_event *ev)
     cl->peer_state = PEER_RDMA;
     cl->connect_timeout_id = -1;
     cl->osd_num = peer_osd;
-    cl->in_buf = malloc_or_die(receive_buffer_size);
+    cl->in_buf = (uint8_t*)malloc_or_die(receive_buffer_size);
     cl->rdma_conn = rc;
     clients[conn->client_id] = cl;
     if (conn->timeout_id >= 0)

src/client/msgr_receive.cpp

@@ -1,6 +1,8 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
 
+#define _XOPEN_SOURCE
+#include <limits.h>
 #include "messenger.h"
 
 void osd_messenger_t::read_requests()
@@ -15,7 +17,11 @@ void osd_messenger_t::read_requests()
             continue;
         }
         auto cl = cl_it->second;
-        if (cl->read_remaining < receive_buffer_size)
+        if (cl->read_op && cl->read_op_size-(cl->read_op_pos-OSD_PACKET_SIZE) >= receive_buffer_size)
+        {
+            op_get_read_buffers(cl, cl->recv_list);
+        }
+        if (!cl->recv_list.size())
         {
             cl->read_iov.iov_base = cl->in_buf;
             cl->read_iov.iov_len = receive_buffer_size;
@@ -25,10 +31,11 @@ void osd_messenger_t::read_requests()
         else
         {
             cl->read_iov.iov_base = 0;
-            cl->read_iov.iov_len = cl->read_remaining;
+            cl->read_iov.iov_len = 0;
-            cl->read_msg.msg_iov = cl->recv_list.get_iovec();
+            cl->read_msg.msg_iov = cl->recv_list.data();
-            cl->read_msg.msg_iovlen = cl->recv_list.get_size();
+            cl->read_msg.msg_iovlen = cl->recv_list.size();
         }
+        assert(!cl->read_op || cl->read_op_pos < OSD_PACKET_SIZE || cl->read_op_size >= (cl->read_op_pos-OSD_PACKET_SIZE));
         cl->refs++;
         if (ringloop && !use_sync_send_recv)
         {
@@ -50,7 +57,7 @@ void osd_messenger_t::read_requests()
             }
             ring_data_t* data = ((ring_data_t*)sqe->user_data);
             data->callback = [this, cl](ring_data_t *data) { handle_read(data->res, cl); };
-            io_uring_prep_recvmsg(sqe, cl->peer_fd, &cl->read_msg, 0);
+            io_uring_prep_recvmsg(sqe, cl->peer_fd, &cl->read_msg, cl->recv_list.size() ? MSG_WAITALL : 0);
             if (iothread)
             {
                 iothread->add_sqe(sqe_local);
@@ -68,16 +75,15 @@ void osd_messenger_t::read_requests()
         }
     }
     read_ready_clients.clear();
+    handle_immediate_ops();
 }
 
-bool osd_messenger_t::handle_read(int result, osd_client_t *cl)
+void osd_messenger_t::handle_read(int result, osd_client_t *cl)
 {
-    bool ret = false;
-    cl->read_msg.msg_iovlen = 0;
     cl->refs--;
     if (cl->peer_state == PEER_RDMA)
     {
-        return true;
+        return;
     }
     if (cl->peer_state == PEER_STOPPED)
     {
@@ -85,7 +91,7 @@ bool osd_messenger_t::handle_read(int result, osd_client_t *cl)
         {
             destroy_client(cl);
         }
-        return false;
+        return;
     }
     if (result <= 0 && result != -EAGAIN && result != -EINTR)
     {
@@ -95,9 +101,56 @@ bool osd_messenger_t::handle_read(int result, osd_client_t *cl)
             fprintf(stderr, "Client %ju socket read error: %d (%s). Disconnecting client\n", cl->client_id, -result, strerror(-result));
         }
         stop_client(cl->client_id);
-        return false;
+out_wakeup:
+        if (set_immediate_ops.size())
+            ringloop->wakeup();
+        return;
     }
-    if (result == -EAGAIN || result == -EINTR || result < cl->read_iov.iov_len)
+    bool full_read = false;
+    if (result > 0)
+    {
+        if (cl->read_iov.iov_base == cl->in_buf)
+        {
+            full_read = result >= cl->read_iov.iov_len;
+            if (!handle_read_buffer(cl, cl->in_buf, result))
+                goto out_wakeup;
+        }
+        else
+        {
+            // Reset OSD ping state
+            cl->ping_time_remaining = 0;
+            cl->idle_time_remaining = osd_idle_timeout;
+            // Long data
+            size_t i = 0;
+            while (i < cl->recv_list.size() && result >= cl->recv_list[i].iov_len)
+            {
+                if (cl->read_csum_state && cl->recv_list[i].iov_len > 0 &&
+                    i != cl->recv_list.size()-1) // skip the checksum itself
+                {
+                    XXH3_64bits_update(cl->read_csum_state, cl->recv_list[i].iov_base, cl->recv_list[i].iov_len);
+                }
+                result -= cl->recv_list[i].iov_len;
+                i++;
+            }
+            if (i < cl->recv_list.size())
+            {
+                cl->recv_list[i].iov_base += result;
+                cl->recv_list[i].iov_len -= result;
+            }
+            else
+            {
+                full_read = true;
+            }
+            cl->recv_list.erase(cl->recv_list.begin(), cl->recv_list.begin()+i);
+            if (!cl->recv_list.size())
+            {
+                if (!handle_finished_op(cl))
+                    goto out_wakeup;
+            }
+        }
+    }
+    cl->read_msg.msg_iovlen = 0;
+    if (result == -EAGAIN || result == -EINTR || !full_read)
     {
         cl->read_ready--;
         if (cl->read_ready > 0)
@@ -107,37 +160,7 @@ bool osd_messenger_t::handle_read(int result, osd_client_t *cl)
     {
         read_ready_clients.push_back(cl->client_id);
     }
-    if (result > 0)
+    goto out_wakeup;
-    {
-        if (cl->read_iov.iov_base == cl->in_buf)
-        {
-            if (!handle_read_buffer(cl, cl->in_buf, result))
-            {
-                handle_immediate_ops();
-                return false;
-            }
-        }
-        else
-        {
-            // Long data
-            cl->read_remaining -= result;
-            cl->recv_list.eat(result);
-            if (cl->recv_list.done >= cl->recv_list.count)
-            {
-                if (!handle_finished_read(cl))
-                {
-                    handle_immediate_ops();
-                    return false;
-                }
-            }
-        }
-        if (result >= cl->read_iov.iov_len)
-        {
-            ret = true;
-        }
-    }
-    handle_immediate_ops();
-    return ret;
 }
 
 void osd_messenger_t::handle_immediate_ops()
@@ -162,113 +185,109 @@ void osd_messenger_t::handle_immediate_ops()
     }
 }
 
-bool osd_messenger_t::handle_read_buffer(osd_client_t *cl, void *curbuf, int remain)
+bool osd_messenger_t::handle_read_buffer(osd_client_t *cl, uint8_t *curbuf, size_t bufsize)
 {
+    // Reset OSD ping state
+    cl->ping_time_remaining = 0;
+    cl->idle_time_remaining = osd_idle_timeout;
     // Compose operation(s) from the buffer
-    while (remain > 0)
+    size_t done = 0;
+    while (done < bufsize)
     {
         if (!cl->read_op)
         {
             cl->read_op = new osd_op_t;
             cl->read_op->client_id = cl->client_id;
             cl->read_op->op_type = OSD_OP_IN;
-            cl->recv_list.push_back(cl->read_op->req.buf, OSD_PACKET_SIZE);
+            cl->read_op_pos = 0;
-            cl->read_remaining = OSD_PACKET_SIZE;
+            cl->read_op_size = 0;
-            cl->read_state = CL_READ_HDR;
+            cl->read_op_inline_decrypt_in = 0;
-        }
+            cl->read_op_inline_decrypt_pos = (size_t)-1;
-        while (cl->recv_list.done < cl->recv_list.count && remain > 0)
+            if (cl->proto_csum_status == MSGR_CSUM_FULL || cl->proto_csum_status == MSGR_CSUM_PAYLOAD)
-        {
-            iovec* cur = cl->recv_list.get_iovec();
-            if (cur->iov_len > remain)
             {
-                memcpy(cur->iov_base, curbuf, remain);
+                if (!cl->read_csum_state)
-                cl->read_remaining -= remain;
+                    cl->read_csum_state = XXH3_createState();
-                cur->iov_len -= remain;
+                XXH3_64bits_reset(cl->read_csum_state);
-                cur->iov_base = (uint8_t*)cur->iov_base + remain;
-                remain = 0;
-            }
-            else
-            {
-                memcpy(cur->iov_base, curbuf, cur->iov_len);
-                curbuf = (uint8_t*)curbuf + cur->iov_len;
-                cl->read_remaining -= cur->iov_len;
-                remain -= cur->iov_len;
-                cur->iov_len = 0;
-                cl->recv_list.done++;
             }
         }
-        if (cl->recv_list.done >= cl->recv_list.count)
+        if (cl->read_op_pos < OSD_PACKET_SIZE)
         {
-            if (!handle_finished_read(cl))
+            int len = OSD_PACKET_SIZE - cl->read_op_pos;
+            if (len > bufsize-done)
+                len = bufsize-done;
+            memcpy(cl->read_op->req.buf + cl->read_op_pos, curbuf+done, len);
+            done += len;
+            cl->read_op_pos += len;
+            if (cl->read_op_pos < OSD_PACKET_SIZE)
+                return true;
+            if (!handle_hdr(cl))
             {
+                stop_client(cl->client_id);
                 return false;
             }
         }
-    }
+        if (!op_copy_from(cl, curbuf, bufsize, done))
-    return true;
-}
-
-bool osd_messenger_t::handle_finished_read(osd_client_t *cl)
-{
-    // Reset OSD ping state
-    cl->ping_time_remaining = 0;
-    cl->idle_time_remaining = osd_idle_timeout;
-    cl->recv_list.reset();
-    if (cl->read_state == CL_READ_HDR)
-    {
-        if (cl->read_op->req.hdr.magic == SECONDARY_OSD_REPLY_MAGIC)
-            return handle_reply_hdr(cl);
-        else if (cl->read_op->req.hdr.magic == SECONDARY_OSD_OP_MAGIC)
         {
-            if (cl->check_sequencing)
-            {
-                if (cl->read_op->req.hdr.id != cl->read_op_id)
-                {
-                    fprintf(stderr, "Warning: operation sequencing is broken on client %ju: expected num %ju, got %ju, stopping client\n", cl->client_id, cl->read_op_id, cl->read_op->req.hdr.id);
-                    stop_client(cl->client_id);
-                    return false;
-                }
-                cl->read_op_id++;
-            }
-            handle_op_hdr(cl);
-        }
-        else
-        {
-            fprintf(stderr, "Received garbage: magic=%jx id=%ju opcode=%jx from client %ju\n", cl->read_op->req.hdr.magic, cl->read_op->req.hdr.id, cl->read_op->req.hdr.opcode, cl->client_id);
-            stop_client(cl->client_id);
             return false;
         }
     }
-    else if (cl->read_state == CL_READ_DATA)
+    return true;
+}
+
+bool osd_messenger_t::handle_hdr(osd_client_t *cl)
+{
+    if (cl->proto_csum_status == MSGR_CSUM_FULL)
     {
-        // Operation is ready
+        XXH3_64bits_update(cl->read_csum_state, cl->read_op->req.buf, OSD_PACKET_SIZE);
-        cl->received_ops.push_back(cl->read_op);
-        set_immediate_ops.push_back(cl->read_op);
-        cl->read_op = NULL;
-        cl->read_state = 0;
     }
-    else if (cl->read_state == CL_READ_REPLY_DATA)
+    if (cl->read_op->req.hdr.magic == SECONDARY_OSD_REPLY_MAGIC)
     {
-        // Reply is ready
+        auto req_it = cl->sent_ops.find(cl->read_op->req.hdr.id);
-        handle_reply_ready(cl->read_op);
+        if (req_it == cl->sent_ops.end())
-        cl->read_op = NULL;
+        {
-        cl->read_state = 0;
+            // Command out of sync. Drop connection
+            fprintf(stderr, "Client %ju command out of sync: id %ju\n", cl->client_id, cl->read_op->req.hdr.id);
+            return false;
+        }
+        osd_op_t *op = req_it->second;
+        memcpy(op->reply.buf, cl->read_op->req.buf, OSD_PACKET_SIZE);
+        if (!allocate_reply_buffers(cl, op))
+        {
+            return false;
+        }
+        cl->sent_ops.erase(req_it);
+        delete cl->read_op;
+        cl->read_op = op;
+    }
+    else if (cl->read_op->req.hdr.magic == SECONDARY_OSD_OP_MAGIC)
+    {
+        if (cl->check_sequencing)
+        {
+            if (cl->read_op->req.hdr.id != cl->read_op_id)
+            {
+                fprintf(stderr, "Warning: operation sequencing is broken on client %d: expected num %ju, got %ju, stopping client\n", cl->peer_fd, cl->read_op_id, cl->read_op->req.hdr.id);
+                return false;
+            }
+            cl->read_op_id++;
+        }
+        if (!allocate_op_buffers(cl))
+        {
+            return false;
+        }
     }
     else
     {
-        assert(0);
+        fprintf(stderr, "Received garbage: magic=%jx id=%ju opcode=%jx from client %ju\n", cl->read_op->req.hdr.magic, cl->read_op->req.hdr.id, cl->read_op->req.hdr.opcode, cl->client_id);
+        return false;
     }
     return true;
 }
 
-void osd_messenger_t::handle_op_hdr(osd_client_t *cl)
+bool osd_messenger_t::allocate_op_buffers(osd_client_t *cl)
 {
     osd_op_t *cur_op = cl->read_op;
-    if (cur_op->req.hdr.opcode == OSD_OP_SEC_READ)
+    cl->read_op_size = 0;
-    {
+    if (cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE ||
-        cl->read_remaining = 0;
-    }
-    else if (cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE ||
         cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE)
     {
         if (cur_op->req.sec_rw.attr_len > 0)
@@ -277,14 +296,12 @@ void osd_messenger_t::handle_op_hdr(osd_client_t *cl)
                 cur_op->bitmap = cur_op->rmw_buf = malloc_or_die(cur_op->req.sec_rw.attr_len);
             else
                 cur_op->bitmap = &cur_op->bmp_data;
-            cl->recv_list.push_back(cur_op->bitmap, cur_op->req.sec_rw.attr_len);
         }
         if (cur_op->req.sec_rw.len > 0)
         {
             cur_op->buf = memalign_or_die(MEM_ALIGNMENT, cur_op->req.sec_rw.len);
-            cl->recv_list.push_back(cur_op->buf, cur_op->req.sec_rw.len);
         }
-        cl->read_remaining = cur_op->req.sec_rw.len + cur_op->req.sec_rw.attr_len;
+        cl->read_op_size = cur_op->req.sec_rw.len + cur_op->req.sec_rw.attr_len;
     }
     else if (cur_op->req.hdr.opcode == OSD_OP_SEC_STABILIZE ||
         cur_op->req.hdr.opcode == OSD_OP_SEC_ROLLBACK)
@@ -292,27 +309,24 @@ void osd_messenger_t::handle_op_hdr(osd_client_t *cl)
         if (cur_op->req.sec_stab.len > 0)
         {
             cur_op->buf = memalign_or_die(MEM_ALIGNMENT, cur_op->req.sec_stab.len);
-            cl->recv_list.push_back(cur_op->buf, cur_op->req.sec_stab.len);
         }
-        cl->read_remaining = cur_op->req.sec_stab.len;
+        cl->read_op_size = cur_op->req.sec_stab.len;
     }
     else if (cur_op->req.hdr.opcode == OSD_OP_SEC_READ_BMP)
     {
         if (cur_op->req.sec_read_bmp.len > 0)
         {
             cur_op->buf = memalign_or_die(MEM_ALIGNMENT, cur_op->req.sec_read_bmp.len);
-            cl->recv_list.push_back(cur_op->buf, cur_op->req.sec_read_bmp.len);
         }
-        cl->read_remaining = cur_op->req.sec_read_bmp.len;
+        cl->read_op_size = cur_op->req.sec_read_bmp.len;
     }
     else if (cur_op->req.hdr.opcode == OSD_OP_WRITE)
     {
         if (cur_op->req.rw.len > 0)
         {
             cur_op->buf = memalign_or_die(MEM_ALIGNMENT, cur_op->req.rw.len);
-            cl->recv_list.push_back(cur_op->buf, cur_op->req.rw.len);
         }
-        cl->read_remaining = cur_op->req.rw.len;
+        cl->read_op_size = cur_op->req.rw.len;
     }
     else if (cur_op->req.hdr.opcode == OSD_OP_SHOW_CONFIG)
     {
@@ -320,44 +334,20 @@ void osd_messenger_t::handle_op_hdr(osd_client_t *cl)
         {
             cur_op->buf = malloc_or_die(cur_op->req.show_conf.json_len+1);
             ((uint8_t*)cur_op->buf)[cur_op->req.show_conf.json_len] = 0;
-            cl->recv_list.push_back(cur_op->buf, cur_op->req.show_conf.json_len);
         }
-        cl->read_remaining = cur_op->req.show_conf.json_len;
+        cl->read_op_size = cur_op->req.show_conf.json_len;
     }
-    /*else if (cur_op->req.hdr.opcode == OSD_OP_READ ||
+    if (cl->proto_csum_status == MSGR_CSUM_FULL ||
-        cur_op->req.hdr.opcode == OSD_OP_SCRUB ||
+        cl->read_op_size > 0 && cl->proto_csum_status == MSGR_CSUM_PAYLOAD)
-        cur_op->req.hdr.opcode == OSD_OP_DESCRIBE)
     {
-        cl->read_remaining = 0;
+        cl->read_op_size += 8;
-    }*/
-    if (cl->read_remaining > 0)
-    {
-        // Read data
-        cl->read_state = CL_READ_DATA;
-    }
-    else
-    {
-        // Operation is ready
-        cl->received_ops.push_back(cur_op);
-        set_immediate_ops.push_back(cur_op);
-        cl->read_op = NULL;
-        cl->read_state = 0;
     }
+    return true;
 }
 
-bool osd_messenger_t::handle_reply_hdr(osd_client_t *cl)
+bool osd_messenger_t::allocate_reply_buffers(osd_client_t *cl, osd_op_t *op)
 {
-    auto req_it = cl->sent_ops.find(cl->read_op->req.hdr.id);
+    cl->read_op_size = 0;
-    if (req_it == cl->sent_ops.end())
-    {
-        // Command out of sync. Drop connection
-        fprintf(stderr, "Client %ju command out of sync: id %ju\n", cl->client_id, cl->read_op->req.hdr.id);
-        stop_client(cl->client_id);
-        return false;
-    }
-    osd_op_t *op = req_it->second;
-    memcpy(op->reply.buf, cl->read_op->req.buf, OSD_PACKET_SIZE);
-    cl->sent_ops.erase(req_it);
     if (op->reply.hdr.opcode == OSD_OP_SEC_READ || op->reply.hdr.opcode == OSD_OP_READ)
     {
         // Read data. In this case we assume that the buffer is preallocated by the caller (!)
@@ -368,97 +358,368 @@ bool osd_messenger_t::handle_reply_hdr(osd_client_t *cl)
             // Check reply length to not overflow the buffer
             fprintf(stderr, "Client %ju read reply of different length: expected %u+%u, got %jd+%u\n",
                 cl->client_id, expected_size, op->bitmap_len, op->reply.hdr.retval, bmp_len);
-            cl->sent_ops[op->req.hdr.id] = op;
-            stop_client(cl->client_id);
             return false;
         }
         if (bmp_len > 0)
         {
             assert(op->bitmap);
-            cl->recv_list.push_back(op->bitmap, bmp_len);
+            cl->read_op_size += bmp_len;
-            cl->read_remaining += bmp_len;
         }
         if (op->reply.hdr.retval > 0)
         {
             assert(op->iov.count > 0);
-            cl->recv_list.append(op->iov);
+            cl->read_op_size += op->reply.hdr.retval;
-            cl->read_remaining += op->reply.hdr.retval;
         }
-        if (cl->read_remaining == 0)
-        {
-            goto reuse;
-        }
-        delete cl->read_op;
-        cl->read_op = op;
-        cl->read_state = CL_READ_REPLY_DATA;
     }
     else if (op->reply.hdr.opcode == OSD_OP_SEC_LIST && op->reply.hdr.retval > 0)
     {
         assert(!op->iov.count);
-        delete cl->read_op;
+        cl->read_op_size = sizeof(obj_ver_id) * op->reply.hdr.retval;
-        cl->read_op = op;
+        op->buf = memalign_or_die(MEM_ALIGNMENT, cl->read_op_size);
-        cl->read_state = CL_READ_REPLY_DATA;
-        cl->read_remaining = sizeof(obj_ver_id) * op->reply.hdr.retval;
-        op->buf = memalign_or_die(MEM_ALIGNMENT, cl->read_remaining);
-        cl->recv_list.push_back(op->buf, cl->read_remaining);
     }
     else if (op->reply.hdr.opcode == OSD_OP_SEC_READ_BMP && op->reply.hdr.retval > 0)
     {
         assert(!op->iov.count);
-        delete cl->read_op;
+        cl->read_op_size = op->reply.hdr.retval;
-        cl->read_op = op;
-        cl->read_state = CL_READ_REPLY_DATA;
-        cl->read_remaining = op->reply.hdr.retval;
         free(op->buf);
-        op->buf = memalign_or_die(MEM_ALIGNMENT, cl->read_remaining);
+        op->buf = memalign_or_die(MEM_ALIGNMENT, cl->read_op_size);
-        cl->recv_list.push_back(op->buf, cl->read_remaining);
     }
     else if (op->reply.hdr.opcode == OSD_OP_SHOW_CONFIG && op->reply.hdr.retval > 0)
     {
-        delete cl->read_op;
+        cl->read_op_size = op->reply.hdr.retval;
-        cl->read_op = op;
-        cl->read_state = CL_READ_REPLY_DATA;
-        cl->read_remaining = op->reply.hdr.retval;
         free(op->buf);
         op->buf = malloc_or_die(op->reply.hdr.retval);
-        cl->recv_list.push_back(op->buf, op->reply.hdr.retval);
     }
     else if (op->reply.hdr.opcode == OSD_OP_DESCRIBE && op->reply.describe.result_bytes > 0)
     {
-        delete cl->read_op;
+        cl->read_op_size = op->reply.describe.result_bytes;
-        cl->read_op = op;
-        cl->read_state = CL_READ_REPLY_DATA;
-        cl->read_remaining = op->reply.describe.result_bytes;
         free(op->buf);
         op->buf = malloc_or_die(op->reply.describe.result_bytes);
-        cl->recv_list.push_back(op->buf, op->reply.describe.result_bytes);
     }
-    else
+    if (cl->proto_csum_status == MSGR_CSUM_FULL ||
+        cl->read_op_size > 0 && cl->proto_csum_status == MSGR_CSUM_PAYLOAD)
     {
-reuse:
+        cl->read_op_size += 8;
-        // It's fine to reuse cl->read_op for the next reply
-        handle_reply_ready(op);
-        cl->recv_list.push_back(cl->read_op->req.buf, OSD_PACKET_SIZE);
-        cl->read_remaining = OSD_PACKET_SIZE;
-        cl->read_state = CL_READ_HDR;
     }
     return true;
 }
 
-void osd_messenger_t::handle_reply_ready(osd_op_t *op)
+bool osd_messenger_t::op_copy_from(osd_client_t *cl, uint8_t *src, size_t src_len, size_t & done)
 {
-    // Measure subop latency
+    osd_op_t *op = cl->read_op;
-    timespec tv_end;
+    size_t from = cl->read_op_pos-OSD_PACKET_SIZE;
-    clock_gettime(CLOCK_REALTIME, &tv_end);
+    auto op_read_buf = [&](uint8_t *dst, size_t dst_len, bool skip_csum = false)
-    stats.subop_stat_count[op->req.hdr.opcode]++;
-    if (!stats.subop_stat_count[op->req.hdr.opcode])
     {
-        stats.subop_stat_count[op->req.hdr.opcode]++;
+        if (from < dst_len)
-        stats.subop_stat_sum[op->req.hdr.opcode] = 0;
+        {
+            size_t n = dst_len-from;
+            if (n > src_len-done)
+                n = src_len-done;
+            if (cl->read_csum_state && !skip_csum)
+            {
+                // it may be skipped if !dst but checksum is still calculated
+                XXH3_64bits_update(cl->read_csum_state, src+done, n);
+            }
+            if (dst)
+                memcpy(dst+from, src+done, n);
+            else
+                assert(!this->osd_num); // NULL buffers are only used by clients
+            done += n;
+            cl->read_op_pos += n;
+            from += n;
+            if (from < dst_len)
+                return false;
+            from = 0;
+        }
+        else
+            from -= dst_len;
+        return true;
+    };
+    if (op->op_type == OSD_OP_IN)
+    {
+        if (op->req.hdr.opcode == OSD_OP_SEC_WRITE ||
+            op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE)
+        {
+            if (!op_read_buf((uint8_t*)op->bitmap, op->req.sec_rw.attr_len))
+                return true;
+            if (!op_read_buf((uint8_t*)op->buf, op->req.sec_rw.len))
+                return true;
+        }
+        else if (op->req.hdr.opcode == OSD_OP_SEC_STABILIZE ||
+            op->req.hdr.opcode == OSD_OP_SEC_ROLLBACK)
+        {
+            if (!op_read_buf((uint8_t*)op->buf, op->req.sec_stab.len))
+                return true;
+        }
+        else if (op->req.hdr.opcode == OSD_OP_SEC_READ_BMP)
+        {
+            if (!op_read_buf((uint8_t*)op->buf, op->req.sec_read_bmp.len))
+                return true;
+        }
+        else if (op->req.hdr.opcode == OSD_OP_WRITE)
+        {
+            if (!op_read_buf((uint8_t*)op->buf, op->req.rw.len))
+                return true;
+        }
+        else if (op->req.hdr.opcode == OSD_OP_SHOW_CONFIG)
+        {
+            if (!op_read_buf((uint8_t*)op->buf, op->req.show_conf.json_len))
+                return true;
+        }
     }
-    stats.subop_stat_sum[op->req.hdr.opcode] += (
+    else
-        (tv_end.tv_sec - op->tv_begin.tv_sec)*1000000 +
+    {
-        (tv_end.tv_nsec - op->tv_begin.tv_nsec)/1000
+        if (op->reply.hdr.opcode == OSD_OP_SEC_READ)
-    );
+        {
-    set_immediate_ops.push_back(op);
+            if (op->reply.sec_rw.attr_len > 0)
+            {
+                if (!op_read_buf((uint8_t*)op->bitmap, op->reply.sec_rw.attr_len))
+                    return true;
+            }
+            if (op->reply.hdr.retval > 0)
+            {
+                for (int i = 0; i < op->iov.count; i++)
+                    if (!op_read_buf((uint8_t*)op->iov.buf[i].iov_base, op->iov.buf[i].iov_len))
+                        return true;
+            }
+        }
+        else if (op->reply.hdr.opcode == OSD_OP_READ)
+        {
+            if (op->reply.rw.bitmap_len > 0)
+            {
+                if (!op_read_buf((uint8_t*)op->bitmap, op->reply.rw.bitmap_len))
+                    return true;
+            }
+            if (op->reply.hdr.retval > 0)
+            {
+                if (op->enc)
+                {
+                    if (!op_decrypted_copy_data_from(cl, src, src_len, from, done))
+                        return true;
+                }
+                else
+                {
+                    for (int i = 0; i < op->iov.count; i++)
+                        if (!op_read_buf((uint8_t*)op->iov.buf[i].iov_base, op->iov.buf[i].iov_len))
+                            return true;
+                }
+            }
+        }
+        else if (op->reply.hdr.opcode == OSD_OP_SEC_LIST && op->reply.hdr.retval > 0)
+        {
+            if (!op_read_buf((uint8_t*)op->buf, sizeof(obj_ver_id) * op->reply.hdr.retval))
+                return true;
+        }
+        else if ((op->reply.hdr.opcode == OSD_OP_SEC_READ_BMP ||
+            op->reply.hdr.opcode == OSD_OP_SHOW_CONFIG) && op->reply.hdr.retval > 0)
+        {
+            if (!op_read_buf((uint8_t*)op->buf, op->reply.hdr.retval))
+                return true;
+        }
+        else if (op->reply.hdr.opcode == OSD_OP_DESCRIBE && op->reply.describe.result_bytes > 0)
+        {
+            if (!op_read_buf((uint8_t*)op->buf, op->reply.describe.result_bytes))
+                return true;
+        }
+    }
+    if (cl->proto_csum_status == MSGR_CSUM_FULL ||
+        cl->read_op_size > 0 && cl->proto_csum_status == MSGR_CSUM_PAYLOAD)
+    {
+        if (!op_read_buf((uint8_t*)&op->csum, 8, true))
+            return true;
+    }
+    return handle_finished_op(cl);
+}
+
+void osd_messenger_t::op_get_read_buffers(osd_client_t *cl, std::vector<iovec> & lst)
+{
+    osd_op_t *op = cl->read_op;
+    size_t from = cl->read_op_pos-OSD_PACKET_SIZE;
+    size_t done = 0;
+    auto op_read_buf = [&](uint8_t *dst, size_t dst_len)
+    {
+        if (lst.size() >= IOV_MAX)
+            return false;
+        if (from < dst_len)
+        {
+            lst.push_back((iovec){ .iov_base = dst+from, .iov_len = dst_len-from });
+            cl->read_op_pos += dst_len-from;
+            done += dst_len-from;
+            from = 0;
+        }
+        else
+            from -= dst_len;
+        return true;
+    };
+    if (op->op_type == OSD_OP_IN)
+    {
+        if (op->req.hdr.opcode == OSD_OP_SEC_WRITE ||
+            op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE)
+        {
+            if (!op_read_buf((uint8_t*)op->bitmap, op->req.sec_rw.attr_len))
+                return;
+            if (!op_read_buf((uint8_t*)op->buf, op->req.sec_rw.len))
+                return;
+        }
+        else if (op->req.hdr.opcode == OSD_OP_SEC_STABILIZE ||
+            op->req.hdr.opcode == OSD_OP_SEC_ROLLBACK)
+        {
+            if (!op_read_buf((uint8_t*)op->buf, op->req.sec_stab.len))
+                return;
+        }
+        else if (op->req.hdr.opcode == OSD_OP_SEC_READ_BMP)
+        {
+            if (!op_read_buf((uint8_t*)op->buf, op->req.sec_read_bmp.len))
+                return;
+        }
+        else if (op->req.hdr.opcode == OSD_OP_WRITE)
+        {
+            if (!op_read_buf((uint8_t*)op->buf, op->req.rw.len))
+                return;
+        }
+        else if (op->req.hdr.opcode == OSD_OP_SHOW_CONFIG)
+        {
+            if (!op_read_buf((uint8_t*)op->buf, op->req.show_conf.json_len))
+                return;
+        }
+    }
+    else
+    {
+        if (op->reply.hdr.opcode == OSD_OP_SEC_READ)
+        {
+            if (op->reply.sec_rw.attr_len > 0)
+            {
+                if (!op_read_buf((uint8_t*)op->bitmap, op->reply.sec_rw.attr_len))
+                    return;
+            }
+            if (op->reply.hdr.retval > 0)
+            {
+                for (int i = 0; i < op->iov.count; i++)
+                    if (!op_read_buf((uint8_t*)op->iov.buf[i].iov_base, op->iov.buf[i].iov_len))
+                        return;
+            }
+        }
+        else if (op->reply.hdr.opcode == OSD_OP_READ)
+        {
+            if (op->reply.rw.bitmap_len > 0)
+            {
+                if (!op_read_buf((uint8_t*)op->bitmap, op->reply.rw.bitmap_len))
+                    return;
+            }
+            if (op->reply.hdr.retval > 0)
+            {
+                if (op->enc)
+                {
+                    cl->read_op_inline_decrypt_pos = cl->read_op_pos;
+                    cl->read_op_pos = cl->read_op_inline_decrypt_in + OSD_PACKET_SIZE + op->reply.rw.bitmap_len;
+                    from = cl->read_op_inline_decrypt_in;
+                }
+                for (int i = 0; i < op->iov.count; i++)
+                {
+                    if (!op->iov.buf[i].iov_base)
+                    {
+                        // When we recvmsg directly into the operation without copying,
+                        // we need some place for all buffers, so we allocate temporary
+                        // buffers for all skipped parts
+                        op_alloc_temp_buffers(op, i);
+                    }
+                    if (!op_read_buf((uint8_t*)op->iov.buf[i].iov_base, op->iov.buf[i].iov_len))
+                        return;
+                }
+            }
+        }
+        else if (op->reply.hdr.opcode == OSD_OP_SEC_LIST && op->reply.hdr.retval > 0)
+        {
+            if (!op_read_buf((uint8_t*)op->buf, sizeof(obj_ver_id) * op->reply.hdr.retval))
+                return;
+        }
+        else if ((op->reply.hdr.opcode == OSD_OP_SEC_READ_BMP ||
+            op->reply.hdr.opcode == OSD_OP_SHOW_CONFIG) && op->reply.hdr.retval > 0)
+        {
+            if (!op_read_buf((uint8_t*)op->buf, op->reply.hdr.retval))
+                return;
+        }
+        else if (op->reply.hdr.opcode == OSD_OP_DESCRIBE && op->reply.describe.result_bytes > 0)
+        {
+            if (!op_read_buf((uint8_t*)op->buf, op->reply.describe.result_bytes))
+                return;
+        }
+    }
+    if (cl->proto_csum_status == MSGR_CSUM_FULL ||
+        cl->read_op_size > 0 && cl->proto_csum_status == MSGR_CSUM_PAYLOAD)
+    {
+        if (!op_read_buf((uint8_t*)&op->csum, 8))
+            return;
+    }
+}
+
+void osd_messenger_t::op_alloc_temp_buffers(osd_op_t *op, int i)
+{
+    size_t total_skip = 0;
+    for (int j = i; j < op->iov.count; j++)
+    {
+        if (!op->iov.buf[j].iov_base)
+        {
+            total_skip += op->iov.buf[j].iov_len;
+        }
+    }
+    assert(total_skip);
+    assert(!op->rmw_buf);
+    op->rmw_buf = malloc_or_die(total_skip);
+    total_skip = 0;
+    for (int j = i; j < op->iov.count; j++)
+    {
+        if (!op->iov.buf[j].iov_base)
+        {
+            op->iov.buf[j].iov_base = (uint8_t*)op->rmw_buf + total_skip;
+            total_skip += op->iov.buf[j].iov_len;
+        }
+    }
+}
+
+bool osd_messenger_t::handle_finished_op(osd_client_t *cl)
+{
+    osd_op_t *op = cl->read_op;
+    if (cl->proto_csum_status == MSGR_CSUM_FULL ||
+        cl->read_op_size > 0 && cl->proto_csum_status == MSGR_CSUM_PAYLOAD)
+    {
+        uint64_t real_csum = XXH3_64bits_digest(cl->read_csum_state);
+        if (op->csum != real_csum)
+        {
+            fprintf(stderr, "Client %ju checksum mismatch for received data: expected %016jx, got %016jx, disconnecting client\n",
+                cl->client_id, op->csum, real_csum);
+            stop_client(cl->client_id);
+            return false;
+        }
+    }
+    if (op->op_type == OSD_OP_IN)
+    {
+        // Operation is ready
+        cl->received_ops.push_back(op);
+    }
+    else
+    {
+        // Inline decryption
+        if (cl->read_op_inline_decrypt_pos != (size_t)-1)
+        {
+            op_decrypt_inline(cl);
+            cl->read_op_inline_decrypt_pos = (size_t)-1;
+        }
+        // Measure subop (outbound op) latency
+        timespec tv_end;
+        clock_gettime(CLOCK_REALTIME, &tv_end);
+        stats.subop_stat_count[op->req.hdr.opcode]++;
+        if (!stats.subop_stat_count[op->req.hdr.opcode])
+        {
+            stats.subop_stat_count[op->req.hdr.opcode]++;
+            stats.subop_stat_sum[op->req.hdr.opcode] = 0;
+        }
+        stats.subop_stat_sum[op->req.hdr.opcode] += (
+            (tv_end.tv_sec - op->tv_begin.tv_sec)*1000000 +
+            (tv_end.tv_nsec - op->tv_begin.tv_nsec)/1000
+        );
+    }
+    set_immediate_ops.push_back(op);
+    cl->read_op = NULL;
+    return true;
 }

src/client/msgr_send.cpp

@@ -7,6 +7,13 @@
 
 #include "messenger.h"
 
+#ifdef WITH_OPENSSL
+#include <openssl/bio.h>
+#include <openssl/err.h>
+#include <openssl/pem.h>
+#include <openssl/ssl.h>
+#endif
+
 void osd_messenger_t::outbox_push(osd_op_t *cur_op)
 {
     assert(cur_op->client_id);
@@ -21,6 +28,7 @@ void osd_messenger_t::outbox_push(osd_op_t *cur_op)
     {
         clock_gettime(CLOCK_REALTIME, &cur_op->tv_begin);
         cur_op->req.hdr.id = ++cl->send_op_id;
+        cl->sent_ops[cur_op->req.hdr.id] = cur_op;
     }
     else
     {
@@ -37,77 +45,9 @@ void osd_messenger_t::outbox_push(osd_op_t *cur_op)
         }
         // Can't be not found because client IDs are unique
         assert(found);
-    }
-    auto & to_send_list = cl->write_msg.msg_iovlen ? cl->next_send_list : cl->send_list;
-    auto & to_outbox = cl->write_msg.msg_iovlen ? cl->next_outbox : cl->outbox;
-    if (cur_op->op_type == OSD_OP_IN)
-    {
         measure_exec(cur_op);
-        to_send_list.push_back((iovec){ .iov_base = cur_op->reply.buf, .iov_len = OSD_PACKET_SIZE });
-    }
-    else
-    {
-        to_send_list.push_back((iovec){ .iov_base = cur_op->req.buf, .iov_len = OSD_PACKET_SIZE });
-        cl->sent_ops[cur_op->req.hdr.id] = cur_op;
-    }
-    to_outbox.push_back((msgr_sendp_t){ .op = cur_op, .flags = MSGR_SENDP_HDR });
-    // Bitmap
-    if (cur_op->op_type == OSD_OP_IN &&
-        cur_op->req.hdr.opcode == OSD_OP_SEC_READ &&
-        cur_op->reply.sec_rw.attr_len > 0)
-    {
-        to_send_list.push_back((iovec){
-            .iov_base = cur_op->bitmap,
-            .iov_len = cur_op->reply.sec_rw.attr_len,
-        });
-        to_outbox.push_back((msgr_sendp_t){ .op = cur_op, .flags = 0 });
-    }
-    else if (cur_op->op_type == OSD_OP_OUT &&
-        (cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE || cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE) &&
-        cur_op->req.sec_rw.attr_len > 0)
-    {
-        to_send_list.push_back((iovec){
-            .iov_base = cur_op->bitmap,
-            .iov_len = cur_op->req.sec_rw.attr_len,
-        });
-        to_outbox.push_back((msgr_sendp_t){ .op = cur_op, .flags = 0 });
-    }
-    // Operation data
-    if ((cur_op->op_type == OSD_OP_IN
-        ? (cur_op->req.hdr.opcode == OSD_OP_READ ||
-        cur_op->req.hdr.opcode == OSD_OP_SEC_READ ||
-        cur_op->req.hdr.opcode == OSD_OP_SEC_LIST ||
-        cur_op->req.hdr.opcode == OSD_OP_SHOW_CONFIG ||
-        cur_op->req.hdr.opcode == OSD_OP_DESCRIBE)
-        : (cur_op->req.hdr.opcode == OSD_OP_WRITE ||
-        cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE ||
-        cur_op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE ||
-        cur_op->req.hdr.opcode == OSD_OP_SEC_STABILIZE ||
-        cur_op->req.hdr.opcode == OSD_OP_SEC_ROLLBACK ||
-        cur_op->req.hdr.opcode == OSD_OP_SHOW_CONFIG)) && cur_op->iov.count > 0)
-    {
-        for (int i = 0; i < cur_op->iov.count; i++)
-        {
-            if (cur_op->iov.buf[i].iov_len > 0)
-            {
-                assert(cur_op->iov.buf[i].iov_base);
-                to_send_list.push_back(cur_op->iov.buf[i]);
-                to_outbox.push_back((msgr_sendp_t){ .op = cur_op, .flags = 0 });
-            }
-        }
-    }
-    if (cur_op->req.hdr.opcode == OSD_OP_SEC_READ_BMP)
-    {
-        if (cur_op->op_type == OSD_OP_IN && cur_op->reply.hdr.retval > 0)
-            to_send_list.push_back((iovec){ .iov_base = cur_op->buf, .iov_len = (size_t)cur_op->reply.hdr.retval });
-        else if (cur_op->op_type == OSD_OP_OUT && cur_op->req.sec_read_bmp.len > 0)
-            to_send_list.push_back((iovec){ .iov_base = cur_op->buf, .iov_len = (size_t)cur_op->req.sec_read_bmp.len });
-        to_outbox.push_back((msgr_sendp_t){ .op = cur_op, .flags = 0 });
-    }
-    if (cur_op->op_type == OSD_OP_IN)
-    {
-        to_outbox[to_outbox.size()-1].flags |= MSGR_SENDP_FREE;
     }
+    cl->write_ops.push_back(cur_op);
 #ifdef WITH_RDMA
     if (cl->peer_state == PEER_RDMA)
     {
@@ -118,7 +58,7 @@ void osd_messenger_t::outbox_push(osd_op_t *cur_op)
     if (!ringloop)
     {
         // FIXME: It's worse because it doesn't allow batching
-        while (cl->outbox.size())
+        while (cl->write_op || cl->write_ops.size())
         {
             try_send(cl);
         }
@@ -183,13 +123,105 @@ void osd_messenger_t::measure_exec(osd_op_t *cur_op)
     }
 }
 
+bool osd_messenger_t::ssl_do_handshake(osd_client_t *cl)
+{
+    int r = SSL_do_handshake(cl->ssl_cli);
+    if (r > 0)
+    {
+        cl->ssl_handshake_done = true;
+    }
+    else
+    {
+        r = SSL_get_error(cl->ssl_cli, r);
+        if (r != SSL_ERROR_WANT_WRITE && r != SSL_ERROR_WANT_READ)
+        {
+            fprintf(stderr, "Client %ju TLS handshake error: %s, stopping client\n", cl->client_id, ERR_error_string(ERR_get_error(), NULL));
+            stop_client(cl->client_id);
+            return false;
+        }
+        if (r == SSL_ERROR_WANT_WRITE)
+        {
+            cl->ssl_want_write = true;
+        }
+    }
+    return true;
+}
+
+bool osd_messenger_t::ssl_do_encrypt(osd_client_t *cl)
+{
+    if (cl->send_list.size() >= IOV_MAX || !cl->ssl_want_write)
+        return false;
+    size_t prev_size = cl->ssl_out_buf_size;
+    while (true)
+    {
+        size_t min_cap = cl->ssl_out_buf_size*2;
+        if (min_cap < 16384)
+            min_cap = 16384;
+        if (cl->ssl_out_buf_cap < min_cap)
+        {
+            uint8_t *old_buf = cl->ssl_out_buf;
+            uint8_t *old_end = old_buf + cl->ssl_out_buf_cap;
+            cl->ssl_out_buf = (uint8_t*)realloc_or_die(cl->ssl_out_buf, min_cap);
+            cl->ssl_out_buf_cap = min_cap;
+            for (auto & iov: cl->send_list)
+            {
+                if (iov.iov_base >= old_buf && iov.iov_base < old_end)
+                    iov.iov_base = cl->ssl_out_buf + ((uint8_t*)iov.iov_base - old_buf);
+            }
+        }
+        bool full_read = (ssl_do_encrypt_to(cl, cl->ssl_out_buf+cl->ssl_out_buf_size,
+            cl->ssl_out_buf_cap-cl->ssl_out_buf_size) == cl->ssl_out_buf_cap-cl->ssl_out_buf_size);
+        if (!full_read)
+            break;
+    }
+    if (cl->ssl_out_buf_size > prev_size)
+        cl->send_list.push_back((iovec){ .iov_base = cl->ssl_out_buf+prev_size, .iov_len = cl->ssl_out_buf_size-prev_size });
+    return true;
+}
+
+size_t osd_messenger_t::ssl_do_encrypt_to(osd_client_t *cl, uint8_t *buf, size_t size)
+{
+    if (size < sizeof(msgr_tls_record_hdr_t))
+        return 0;
+    int r = BIO_read(cl->read_from_ssl, buf+sizeof(msgr_tls_record_hdr_t), size-sizeof(msgr_tls_record_hdr_t));
+    if (r > 0)
+    {
+        if (r < size-sizeof(msgr_tls_record_hdr_t))
+            cl->ssl_want_write = false;
+        msgr_tls_record_hdr_t *hdr = (msgr_tls_record_hdr_t*)buf;
+        hdr->encrypted = 1;
+        hdr->size = r;
+        return r+sizeof(msgr_tls_record_hdr_t);
+    }
+    return 0;
+}
+
 bool osd_messenger_t::try_send(osd_client_t *cl)
 {
-    if (!cl->send_list.size() || cl->write_msg.msg_iovlen > 0 || cl->peer_state == PEER_STOPPED || cl->peer_fd < 0)
+    if (!cl->write_op && !cl->write_ops.size() && !cl->ssl_want_write ||
+        cl->write_msg.msg_iovlen > 0 || cl->peer_state == PEER_STOPPED || cl->peer_fd < 0)
     {
         return true;
     }
     assert(cl->peer_state != PEER_RDMA);
+    if (cl->ssl_cli && !cl->ssl_handshake_done)
+    {
+        bool ok = ssl_do_encrypt(cl);
+        assert(ok && cl->send_list.size() > 0);
+    }
+    else
+    {
+        while ((cl->write_op || cl->write_ops.size()) && cl->send_list.size() < IOV_MAX)
+        {
+            next_write_op(cl);
+            osd_op_t *op = cl->write_op;
+            op_get_write_buffers(cl, cl->send_list);
+            if (!cl->write_op && op->op_type == OSD_OP_IN)
+            {
+                cl->send_free_ops.push_back(op);
+            }
+        }
+    }
     if (ringloop && !use_sync_send_recv)
     {
         auto iothread = iothreads.size() ? iothreads[cl->peer_fd % iothreads.size()] : NULL;
@@ -202,20 +234,24 @@ bool osd_messenger_t::try_send(osd_client_t *cl)
             data_local = {};
         }
         if (!sqe)
+        {
             return false;
+        }
+        cl->send_list_size = 0;
+        for (auto & iov: cl->send_list)
+        {
+            cl->send_list_size += iov.iov_len;
+        }
         cl->write_msg.msg_iov = cl->send_list.data();
         cl->write_msg.msg_iovlen = cl->send_list.size() < IOV_MAX ? cl->send_list.size() : IOV_MAX;
         cl->refs++;
         ring_data_t* data = ((ring_data_t*)sqe->user_data);
         data->callback = [this, cl](ring_data_t *data) { handle_send(data->res, data->prev, data->more, cl); };
         bool use_zc = has_sendmsg_zc && min_zerocopy_send_size >= 0;
-        if (use_zc && min_zerocopy_send_size > 0)
+        if (use_zc && min_zerocopy_send_size > 0 &&
+            cl->send_list_size/cl->write_msg.msg_iovlen < min_zerocopy_send_size)
         {
-            size_t avg_size = 0;
+            use_zc = false;
-            for (size_t i = 0; i < cl->write_msg.msg_iovlen; i++)
-                avg_size += cl->write_msg.msg_iov[i].iov_len;
-            if (avg_size/cl->write_msg.msg_iovlen < min_zerocopy_send_size)
-                use_zc = false;
         }
         if (use_zc)
         {
@@ -246,6 +282,21 @@ bool osd_messenger_t::try_send(osd_client_t *cl)
     return true;
 }
 
+void osd_messenger_t::next_write_op(osd_client_t *cl)
+{
+    if (!cl->write_op)
+    {
+        cl->write_op = cl->write_ops.front();
+        cl->write_ops.pop_front();
+        if (cl->proto_csum_status == MSGR_CSUM_FULL || cl->proto_csum_status == MSGR_CSUM_PAYLOAD)
+        {
+            if (!cl->write_csum_state)
+                cl->write_csum_state = XXH3_createState();
+            XXH3_64bits_reset(cl->write_csum_state);
+        }
+    }
+}
+
 void osd_messenger_t::send_replies()
 {
     for (int i = 0; i < write_ready_clients.size(); i++)
@@ -266,6 +317,7 @@ void osd_messenger_t::handle_send(int result, bool prev, bool more, osd_client_t
     if (!prev)
     {
         cl->write_msg.msg_iovlen = 0;
+        cl->send_list.clear();
     }
     if (!more)
     {
@@ -298,57 +350,32 @@ void osd_messenger_t::handle_send(int result, bool prev, bool more, osd_client_t
                 cl->zc_free_list.erase(cl->zc_free_list.begin(), cl->zc_free_list.begin()+i+1);
             return;
         }
-        int done = 0;
+        if (cl->send_list_size > result)
-        while (result > 0 && done < cl->send_list.size())
         {
-            iovec & iov = cl->send_list[done];
+            fprintf(stderr, "Client %ju socket write error: expected to send "
-            if (iov.iov_len <= result)
+                "%zu bytes with MSG_WAITALL but sent %u. Disconnecting client\n", cl->client_id, cl->send_list_size, result);
-            {
+            stop_client(cl->client_id);
-                if (cl->outbox[done].flags & MSGR_SENDP_FREE)
+            return;
-                {
+        }
-                    // Reply fully sent
+        for (auto op: cl->send_free_ops)
-                    if (more)
+        {
-                        cl->zc_free_list.push_back(cl->outbox[done].op);
+            if (more)
-                    else
+                cl->zc_free_list.push_back(op);
-                        delete cl->outbox[done].op;
-                }
-                result -= iov.iov_len;
-                done++;
-            }
             else
-            {
+                delete op;
-                iov.iov_len -= result;
-                iov.iov_base = (uint8_t*)iov.iov_base + result;
-                break;
-            }
         }
+        cl->ssl_out_buf_size = 0;
         if (more)
-        {
-            int expected = cl->send_list.size() < IOV_MAX ? cl->send_list.size() : IOV_MAX;
-            if (done != expected)
-            {
-                fprintf(stderr, "Client %ju socket write error: expected to send "
-                    "%d iovecs with MSG_WAITALL but sent %d. Disconnecting client\n", cl->client_id, expected, done);
-                stop_client(cl->client_id);
-                return;
-            }
             cl->zc_free_list.push_back(NULL); // end marker
-        }
+        cl->send_free_ops.clear();
-        if (done > 0)
+        cl->write_state = cl->write_op || cl->write_ops.size() ? CL_WRITE_READY : 0;
+        if ((cl->proto_csum_status & MSGR_CSUM_NEG) && !cl->write_op && !cl->write_ops.size())
         {
-            cl->send_list.erase(cl->send_list.begin(), cl->send_list.begin()+done);
+            // Checksums negotiated, enable
-            cl->outbox.erase(cl->outbox.begin(), cl->outbox.begin()+done);
+            cl->proto_csum_status = cl->proto_csum_status & (~MSGR_CSUM_NEG);
         }
-        if (cl->next_send_list.size())
-        {
-            cl->send_list.insert(cl->send_list.end(), cl->next_send_list.begin(), cl->next_send_list.end());
-            cl->outbox.insert(cl->outbox.end(), cl->next_outbox.begin(), cl->next_outbox.end());
-            cl->next_send_list.clear();
-            cl->next_outbox.clear();
-        }
-        cl->write_state = cl->outbox.size() > 0 ? CL_WRITE_READY : 0;
 #ifdef WITH_RDMA
-        if (cl->rdma_conn && !cl->outbox.size() && cl->peer_state == PEER_RDMA_CONNECTING)
+        if (cl->rdma_conn && !cl->write_op && !cl->write_ops.size() && cl->peer_state == PEER_RDMA_CONNECTING)
         {
             // FIXME: Ignore pings during RDMA state transition
             if (log_level > 0)
@@ -366,3 +393,406 @@ void osd_messenger_t::handle_send(int result, bool prev, bool more, osd_client_t
         write_ready_clients.push_back(cl->client_id);
     }
 }
+
+static inline bool op_write_headers(osd_op_t *op, std::function<bool(uint8_t*, size_t, bool)> op_write_buf, bool skip_hdr_csum)
+{
+    if (!op_write_buf((op->op_type == OSD_OP_IN ? op->reply.buf : op->req.buf), OSD_PACKET_SIZE, skip_hdr_csum))
+    {
+        return false;
+    }
+    // Bitmap
+    if (op->op_type == OSD_OP_IN &&
+        op->req.hdr.opcode == OSD_OP_SEC_READ &&
+        op->reply.sec_rw.attr_len > 0)
+    {
+        if (!op_write_buf((uint8_t*)op->bitmap, op->reply.sec_rw.attr_len, false))
+            return false;
+    }
+    else if (op->op_type == OSD_OP_OUT &&
+        (op->req.hdr.opcode == OSD_OP_SEC_WRITE || op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE) &&
+        op->req.sec_rw.attr_len > 0)
+    {
+        if (!op_write_buf((uint8_t*)op->bitmap, op->req.sec_rw.attr_len, false))
+            return false;
+    }
+    if (op->req.hdr.opcode == OSD_OP_SEC_READ_BMP)
+    {
+        if (op->op_type == OSD_OP_IN && op->reply.hdr.retval > 0)
+        {
+            if (!op_write_buf((uint8_t*)op->buf, (size_t)op->reply.hdr.retval, false))
+                return false;
+        }
+        else if (op->op_type == OSD_OP_OUT && op->req.sec_read_bmp.len > 0)
+        {
+            if (!op_write_buf((uint8_t*)op->buf, (size_t)op->req.sec_read_bmp.len, false))
+                return false;
+        }
+    }
+    return true;
+}
+
+static inline bool op_has_data(osd_op_t *op)
+{
+    return (op->op_type == OSD_OP_IN
+        ? (op->req.hdr.opcode == OSD_OP_READ ||
+        op->req.hdr.opcode == OSD_OP_SEC_READ ||
+        op->req.hdr.opcode == OSD_OP_SEC_LIST ||
+        op->req.hdr.opcode == OSD_OP_SHOW_CONFIG ||
+        op->req.hdr.opcode == OSD_OP_DESCRIBE)
+        : (op->req.hdr.opcode == OSD_OP_WRITE ||
+        op->req.hdr.opcode == OSD_OP_SEC_WRITE ||
+        op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE ||
+        op->req.hdr.opcode == OSD_OP_SEC_STABILIZE ||
+        op->req.hdr.opcode == OSD_OP_SEC_ROLLBACK ||
+        op->req.hdr.opcode == OSD_OP_SHOW_CONFIG)) && op->iov.count > 0;
+}
+
+static inline bool op_has_data_for_ssl(osd_op_t *op)
+{
+    return (op->op_type == OSD_OP_IN
+        ? (op->req.hdr.opcode == OSD_OP_SEC_LIST ||
+        op->req.hdr.opcode == OSD_OP_SHOW_CONFIG ||
+        op->req.hdr.opcode == OSD_OP_DESCRIBE)
+        : (op->req.hdr.opcode == OSD_OP_SEC_STABILIZE ||
+        op->req.hdr.opcode == OSD_OP_SEC_ROLLBACK ||
+        op->req.hdr.opcode == OSD_OP_SHOW_CONFIG)) && op->iov.count > 0;
+}
+
+static inline bool op_has_data_for_nonssl(osd_op_t *op)
+{
+    return (op->op_type == OSD_OP_IN
+        ? (op->req.hdr.opcode == OSD_OP_READ ||
+        op->req.hdr.opcode == OSD_OP_SEC_READ)
+        : (op->req.hdr.opcode == OSD_OP_WRITE ||
+        op->req.hdr.opcode == OSD_OP_SEC_WRITE ||
+        op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE)) && op->iov.count > 0;
+}
+
+bool osd_messenger_t::ssl_op_write_buf(osd_client_t *cl, uint8_t *src, size_t src_len, bool skip_csum, size_t & from, size_t & done)
+{
+    if (from < src_len)
+    {
+        size_t n = src_len-from;
+        int ok = SSL_write_ex(cl->ssl_cli, src+from, n, &n);
+        if (ok)
+        {
+            cl->ssl_want_write = true;
+            if (cl->write_csum_state && !skip_csum)
+                XXH3_64bits_update(cl->write_csum_state, src+from, n);
+            done += n;
+            cl->write_op_pos += n;
+            from += n;
+            if (from < src_len)
+                return false;
+            from = 0;
+        }
+        else
+        {
+            int res = SSL_get_error(cl->ssl_cli, ok);
+            if (res == SSL_ERROR_WANT_WRITE || res == 0)
+                cl->ssl_want_write = true;
+            else if (res == SSL_ERROR_ZERO_RETURN)
+            {
+                fprintf(stderr, "Client %ju TLS disconnected\n", cl->client_id);
+                stop_client(cl->client_id);
+            }
+            else if (res != SSL_ERROR_WANT_READ)
+            {
+                fprintf(stderr, "Client %ju TLS write error: %s. Disconnecting client\n", cl->client_id, ERR_error_string(ERR_get_error(), NULL));
+                stop_client(cl->client_id);
+            }
+            return false;
+        }
+    }
+    else
+        from -= src_len;
+    return true;
+}
+
+bool osd_messenger_t::op_write_buf(osd_client_t *cl, uint8_t *src, size_t src_len, uint8_t *dst, size_t dst_len, bool skip_csum, size_t & from, size_t & done)
+{
+    if (from < src_len)
+    {
+        size_t n = src_len-from;
+        if (n > dst_len-done)
+            n = dst_len-done;
+        if (cl->write_csum_state && !skip_csum)
+            XXH3_64bits_update(cl->write_csum_state, src+from, n);
+        memcpy(dst+done, src+from, n);
+        done += n;
+        cl->write_op_pos += n;
+        from += n;
+        if (from < src_len)
+            return false;
+        from = 0;
+    }
+    else
+        from -= src_len;
+    return true;
+}
+
+size_t osd_messenger_t::ssl_op_copy_to(osd_client_t *cl, uint8_t *dst, size_t dst_len)
+{
+    size_t done = 0;
+    size_t from = cl->write_op_pos;
+    // Encrypt headers and data except read/write data
+    auto to_ssl = [&](uint8_t *src, size_t src_len, bool skip_csum)
+    {
+        return ssl_op_write_buf(cl, src, src_len, skip_csum, from, done);
+    };
+    int i = 0;
+    bool full_hdr = false;
+    bool full_op = false;
+    do
+    {
+        if (!full_hdr)
+            full_hdr = op_write_headers(cl->write_op, to_ssl, cl->proto_csum_status != MSGR_CSUM_FULL);
+        if (full_hdr && op_has_data_for_ssl(cl->write_op))
+        {
+            for (; i < cl->write_op->iov.count; i++)
+                if (!ssl_op_write_buf(cl, (uint8_t*)cl->write_op->iov.buf[i].iov_base, cl->write_op->iov.buf[i].iov_len, false, from, done))
+                    break;
+            full_op = true;
+        }
+        if (cl->peer_state == PEER_STOPPED)
+            return 0;
+        if (cl->ssl_want_write)
+        {
+            auto ssl_written = ssl_do_encrypt_to(cl, dst+done, dst_len-done);
+            if (!ssl_written)
+                return done;
+            done += ssl_written;
+        }
+    } while (!full_op);
+    // Non-TLS-encrypted operation data
+    if (op_has_data_for_nonssl(cl->write_op))
+    {
+        if (!op_copy_data_to(cl, dst, dst_len, from, done))
+            return done;
+    }
+    // TLS-encrypted checksum (uh oh...)
+    if (cl->proto_csum_status == MSGR_CSUM_FULL ||
+        cl->proto_csum_status == MSGR_CSUM_PAYLOAD && cl->write_op_pos > OSD_PACKET_SIZE)
+    {
+        if (!from)
+            cl->write_op->csum = XXH3_64bits_digest(cl->write_csum_state);
+        if (!ssl_op_write_buf(cl, (uint8_t*)&cl->write_op->csum, 8, true, from, done))
+            return done;
+        if (cl->ssl_want_write)
+        {
+            auto ssl_written = ssl_do_encrypt_to(cl, dst+done, dst_len-done);
+            if (!ssl_written)
+                return done;
+            done += ssl_written;
+        }
+    }
+    cl->write_op = NULL;
+    cl->write_op_pos = 0;
+    return done;
+}
+
+size_t osd_messenger_t::op_copy_to(osd_client_t *cl, uint8_t *dst, size_t dst_len)
+{
+    if (cl->ssl_cli)
+    {
+        return ssl_op_copy_to(cl, dst, dst_len);
+    }
+    size_t done = 0;
+    size_t from = cl->write_op_pos;
+    auto op_write_buf = [&](uint8_t *src, size_t src_len, bool skip_csum)
+    {
+        return this->op_write_buf(cl, src, src_len, dst, dst_len, skip_csum, from, done);
+    };
+    // Header
+    if (!op_write_headers(cl->write_op, op_write_buf, cl->proto_csum_status != MSGR_CSUM_FULL))
+    {
+        return done;
+    }
+    // Operation data
+    if (op_has_data(cl->write_op))
+    {
+        if (!op_copy_data_to(cl, dst, dst_len, from, done))
+            return done;
+    }
+    if (cl->proto_csum_status == MSGR_CSUM_FULL ||
+        cl->proto_csum_status == MSGR_CSUM_PAYLOAD && cl->write_op_pos > OSD_PACKET_SIZE)
+    {
+        if (!from)
+            cl->write_op->csum = XXH3_64bits_digest(cl->write_csum_state);
+        if (!op_write_buf((uint8_t*)&cl->write_op->csum, 8, true))
+            return done;
+    }
+    cl->write_op = NULL;
+    cl->write_op_pos = 0;
+    return done;
+}
+
+bool osd_messenger_t::op_copy_data_to(osd_client_t *cl, uint8_t *dst, size_t dst_len, size_t & from, size_t & done)
+{
+    if (cl->write_op->enc)
+    {
+        if (!op_encrypted_copy_data_to(cl, dst, dst_len, from, done))
+        {
+            return false;
+        }
+    }
+    else
+    {
+        for (int i = 0; i < cl->write_op->iov.count; i++)
+        {
+            auto & iov = cl->write_op->iov.buf[i];
+            if (!op_write_buf(cl, (uint8_t*)iov.iov_base, iov.iov_len, dst, dst_len, false, from, done))
+                return false;
+        }
+    }
+    return true;
+}
+
+void osd_messenger_t::op_get_write_buffers(osd_client_t *cl, std::vector<iovec> & lst)
+{
+    size_t from = cl->write_op_pos;
+    auto op_write_buf = [&](uint8_t *src, size_t src_len, bool skip_csum)
+    {
+        if (lst.size() >= IOV_MAX)
+            return false;
+        if (from < src_len)
+        {
+            if (cl->write_csum_state && !skip_csum)
+                XXH3_64bits_update(cl->write_csum_state, src+from, src_len-from);
+            lst.push_back((iovec){ .iov_base = src+from, .iov_len = src_len-from });
+            cl->write_op_pos += src_len-from;
+            from = 0;
+        }
+        else
+            from -= src_len;
+        return true;
+    };
+    // Header
+    if (!op_write_headers(cl->write_op, op_write_buf, cl->proto_csum_status != MSGR_CSUM_FULL))
+    {
+        return;
+    }
+    // Operation data
+    if (op_has_data(cl->write_op))
+    {
+        if (cl->write_op->enc)
+        {
+            if (lst.size() >= IOV_MAX)
+                return;
+            // No way except to allocate a temporary buffer and encrypt data to it
+            assert(cl->write_op->req.hdr.opcode == OSD_OP_WRITE);
+            size_t remsize = cl->write_op->req.rw.len - from + (from % 16);
+            assert(remsize > 0);
+            assert(!cl->write_op->enc_buf);
+            cl->write_op->enc_buf = (uint8_t*)malloc_or_die(remsize);
+            size_t done = 0;
+            bool end = op_encrypted_copy_data_to(cl, cl->write_op->enc_buf, remsize, from, done);
+            assert(end);
+            lst.push_back((iovec){ .iov_base = cl->write_op->enc_buf, .iov_len = remsize });
+        }
+        else
+        {
+            for (int i = 0; i < cl->write_op->iov.count; i++)
+            {
+                if (!op_write_buf((uint8_t*)cl->write_op->iov.buf[i].iov_base, cl->write_op->iov.buf[i].iov_len, false))
+                    return;
+            }
+        }
+    }
+    if (cl->proto_csum_status == MSGR_CSUM_FULL ||
+        cl->proto_csum_status == MSGR_CSUM_PAYLOAD && cl->write_op_pos > OSD_PACKET_SIZE)
+    {
+        if (!from)
+            cl->write_op->csum = XXH3_64bits_digest(cl->write_csum_state);
+        if (!op_write_buf((uint8_t*)&cl->write_op->csum, 8, true))
+            return;
+    }
+    cl->write_op = NULL;
+    cl->write_op_pos = 0;
+}
+
+void osd_messenger_t::ssl_op_get_write_buffers(osd_client_t *cl, std::vector<iovec> & lst)
+{
+    size_t done = 0;
+    size_t from = cl->write_op_pos;
+    auto op_write_buf = [&](uint8_t *src, size_t src_len, bool skip_csum)
+    {
+        if (lst.size() >= IOV_MAX)
+            return false;
+        if (from < src_len)
+        {
+            if (cl->write_csum_state && !skip_csum)
+                XXH3_64bits_update(cl->write_csum_state, src+from, src_len-from);
+            lst.push_back((iovec){ .iov_base = src+from, .iov_len = src_len-from });
+            cl->write_op_pos += src_len-from;
+            from = 0;
+        }
+        else
+            from -= src_len;
+        return true;
+    };
+    // Encrypt headers and data except read/write data
+    auto to_ssl = [&](uint8_t *src, size_t src_len, bool skip_csum)
+    {
+        return ssl_op_write_buf(cl, src, src_len, skip_csum, from, done);
+    };
+    int i = 0;
+    bool full_hdr = false;
+    bool full_op = false;
+    do
+    {
+        if (!full_hdr)
+            full_hdr = op_write_headers(cl->write_op, to_ssl, cl->proto_csum_status != MSGR_CSUM_FULL);
+        if (full_hdr && op_has_data_for_ssl(cl->write_op))
+        {
+            for (; i < cl->write_op->iov.count; i++)
+                if (!ssl_op_write_buf(cl, (uint8_t*)cl->write_op->iov.buf[i].iov_base, cl->write_op->iov.buf[i].iov_len, false, from, done))
+                    break;
+            full_op = true;
+        }
+        if (cl->peer_state == PEER_STOPPED)
+            return;
+        if (!ssl_do_encrypt(cl))
+            return;
+    } while (!full_op);
+    // Non-TLS-encrypted operation data
+    if (op_has_data_for_nonssl(cl->write_op))
+    {
+        if (cl->write_op->enc)
+        {
+            if (lst.size() >= IOV_MAX)
+                return;
+            // No way except to allocate a temporary buffer and encrypt data to it
+            assert(cl->write_op->req.hdr.opcode == OSD_OP_WRITE);
+            size_t remsize = cl->write_op->req.rw.len - from + (from % 16);
+            assert(remsize > 0);
+            assert(!cl->write_op->enc_buf);
+            cl->write_op->enc_buf = (uint8_t*)malloc_or_die(remsize);
+            size_t done = 0;
+            bool end = op_encrypted_copy_data_to(cl, cl->write_op->enc_buf, remsize, from, done);
+            assert(end);
+            lst.push_back((iovec){ .iov_base = cl->write_op->enc_buf, .iov_len = remsize });
+        }
+        else
+        {
+            for (int i = 0; i < cl->write_op->iov.count; i++)
+            {
+                if (!op_write_buf((uint8_t*)cl->write_op->iov.buf[i].iov_base, cl->write_op->iov.buf[i].iov_len, false))
+                    return;
+            }
+        }
+    }
+    // TLS-encrypted checksum (uh oh...)
+    if (cl->proto_csum_status == MSGR_CSUM_FULL ||
+        cl->proto_csum_status == MSGR_CSUM_PAYLOAD && cl->write_op_pos > OSD_PACKET_SIZE)
+    {
+        if (!from)
+            cl->write_op->csum = XXH3_64bits_digest(cl->write_csum_state);
+        if (!ssl_op_write_buf(cl, (uint8_t*)&cl->write_op->csum, 8, true, from, done))
+            return;
+        if (!ssl_do_encrypt(cl))
+            return;
+    }
+    cl->write_op = NULL;
+    cl->write_op_pos = 0;
+}

src/client/msgr_stop.cpp

@@ -5,9 +5,16 @@
 #include <assert.h>
 
 #include "messenger.h"
+#include "../util/xxh_x86dispatch.h"
 #ifdef WITH_RDMA
 #include "msgr_rdma.h"
 #endif
+#ifdef WITH_OPENSSL
+#include <openssl/bio.h>
+#include <openssl/err.h>
+#include <openssl/pem.h>
+#include <openssl/ssl.h>
+#endif
 
 void osd_client_t::cancel_ops()
 {
@@ -79,6 +86,22 @@ void osd_messenger_t::stop_client(uint64_t client_id, bool force_delete)
             fprintf(stderr, "[OSD %ju] Stopping client %ju (regular client)\n", osd_num, client_id);
         }
     }
+    if (cl->encrypt_ctx)
+    {
+        if (encrypt_ctx_pool.size() > max_aes_xts_pool_size)
+            destroy_aes_xts_encrypt(cl->encrypt_ctx);
+        else
+            encrypt_ctx_pool.push_back(cl->encrypt_ctx);
+        cl->encrypt_ctx = NULL;
+    }
+    if (cl->decrypt_ctx)
+    {
+        if (decrypt_ctx_pool.size() > max_aes_xts_pool_size)
+            destroy_aes_xts_decrypt(cl->decrypt_ctx);
+        else
+            decrypt_ctx_pool.push_back(cl->decrypt_ctx);
+        cl->decrypt_ctx = NULL;
+    }
     // First set state to STOPPED so another stop_client() call doesn't try to free it again
     cl->refs++;
     int prev_state = cl->peer_state;
@@ -189,6 +212,13 @@ osd_client_t::~osd_client_t()
     }
     // Cancel outbound ops
     cancel_ops();
+    for (osd_op_t *op: send_free_ops)
+    {
+        if (op)
+        {
+            delete op;
+        }
+    }
     for (osd_op_t *op: zc_free_list)
     {
         if (op)
@@ -204,5 +234,29 @@ osd_client_t::~osd_client_t()
         rdma_conn = NULL;
     }
 #endif
+#endif
+    if (read_csum_state)
+    {
+        XXH3_freeState(read_csum_state);
+        read_csum_state = NULL;
+    }
+    if (write_csum_state)
+    {
+        XXH3_freeState(write_csum_state);
+        write_csum_state = NULL;
+    }
+#ifdef WITH_OPENSSL
+    if (ssl_cli)
+    {
+        SSL_free(ssl_cli);
+        ssl_cli = NULL;
+        write_to_ssl = NULL;
+        read_from_ssl = NULL;
+    }
+    if (ssl_out_buf)
+    {
+        free(ssl_out_buf);
+        ssl_out_buf = NULL;
+    }
 #endif
 }

src/client/osd_ops.h

@@ -37,6 +37,7 @@
 
 #define OSD_OP_RECOVERY_RELATED     (uint32_t)1
 #define OSD_OP_IGNORE_PG_LOCK       (uint32_t)2
+#define OSD_OP_RETURN_CHAIN         (uint32_t)4
 
 // Memory alignment for direct I/O (usually 512 bytes)
 #ifndef DIRECT_IO_ALIGNMENT
@@ -228,9 +229,10 @@ struct __attribute__((__packed__)) osd_op_rw_t
     uint64_t offset;
     // length. 0 means to read all bitmaps of the specified range, but no data.
     uint32_t len;
-    // flags (for future)
+    // flags
+    // OSD_OP_RETURN_CHAIN for chained reads: return parent number in chain for each block
     uint32_t flags;
-    // inode metadata revision
+    // inode metadata revision for chained reads
     uint64_t meta_revision;
     // object version for atomic "CAS" (compare-and-set) writes
     // writes and deletes fail with -EINTR if object version differs from (version-1)

src/cmd/CMakeLists.txt

@@ -2,11 +2,19 @@ cmake_minimum_required(VERSION 2.8...3.30)
 
 project(vitastor)
 
+set(OPENAPI_JSON_H "${CMAKE_CURRENT_BINARY_DIR}/openapi.json.h")
+add_custom_command(
+	OUTPUT ${OPENAPI_JSON_H}
+	COMMAND ${CMAKE_COMMAND} -E echo const char* openapi_description = R\\\"json\\\( > ${OPENAPI_JSON_H}
+	COMMAND ${CMAKE_COMMAND} -E cat ${CMAKE_CURRENT_SOURCE_DIR}/openapi.json >> ${OPENAPI_JSON_H}
+	COMMAND ${CMAKE_COMMAND} -E echo "\\)json\\\"\\;" >> ${OPENAPI_JSON_H}
+	DEPENDS openapi.json
+)
+
 # libvitastor_cli.a
 add_library(vitastor_cli STATIC
 	cli_common.cpp
 	cli_alloc_osd.cpp
-	cli_status.cpp
 	cli_describe.cpp
 	cli_fix.cpp
 	cli_ls.cpp
@@ -14,6 +22,7 @@ add_library(vitastor_cli STATIC
 	cli_dd.cpp
 	cli_modify.cpp
 	cli_modify_osd.cpp
+	cli_modify_user.cpp
 	cli_osd_tree.cpp
 	cli_pg_ls.cpp
 	cli_flatten.cpp
@@ -27,8 +36,13 @@ add_library(vitastor_cli STATIC
 	cli_pool_ls.cpp
 	cli_pool_modify.cpp
 	cli_pool_rm.cpp
+	cli_serve.cpp
+	cli_status.cpp
+	cli_user_ls.cpp
+	${OPENAPI_JSON_H}
 )
 target_compile_options(vitastor_cli PUBLIC -fPIC)
+target_include_directories(vitastor_cli PRIVATE ${CMAKE_CURRENT_BINARY_DIR})
 
 # vitastor-cli
 add_executable(vitastor-cli

src/cmd/cli.cpp

@@ -37,24 +37,42 @@ static const char* help_text =
     "  --sort FIELD    Sort by specified field (name, size, used_size, <read|write|delete>_<iops|bps|lat|queue>)\n"
     "  -r|--reverse    Sort in descending order\n"
     "  -n|--count N    Only list first N items\n"
+    "  --ids ID1,ID2   Only list images with specified full IDs\n"
     "  --tree          Show image snapshot/clone tree\n"
     "\n"
-    "vitastor-cli create -s|--size <size> [-p|--pool <id|name>] [--parent <parent_name>[@<snapshot>]] <name>\n"
+    "vitastor-cli create -s|--size SIZE [OPTIONS] <name>\n"
-    "  Create an image. You may use K/M/G/T suffixes for <size>. If --parent is specified,\n"
+    "  Create an image. Options:\n"
-    "  a copy-on-write image clone is created. Parent must be a snapshot (readonly image).\n"
+    "  -s|--size SIZE      New image size in bytes or with a K/M/G/T unit suffix.\n"
-    "  Pool must be specified if there is more than one pool.\n"
+    "  -p|--pool POOL      Specify pool for the new image (may be omitted if there is only 1 pool).\n"
+    "  --parent PARENT     Create a copy-on-write image clone based on PARENT (or PARENT@SNAPSHOT).\n"
+    "                      If parent is not a snapshot, it must be a read-only image.\n"
+    "  --enc-key random    Generate a new random AES-256-XTS encryption key for the new image.\n"
+    "  --enc-key HEX       Set a specified AES-256-XTS key (64 bytes in hex) for the new image.\n"
+    "  --enc-key vault:ID  Use an encryption key from an external Vault secret with specified ID.\n"
+    "  --owner username    Set owner (default is the current user from TLS certificate).\n"
+    "  --owner_group name  Set owner group name.\n"
+    "  --reader_group rdr  Set reader group name.\n"
     "\n"
-    "vitastor-cli create --snapshot <snapshot> [-p|--pool <id|name>] <image>\n"
+    "vitastor-cli create --snapshot <snapshot> [OPTIONS] <image>\n"
-    "vitastor-cli snap-create [-p|--pool <id|name>] <image>@<snapshot>\n"
+    "vitastor-cli snap-create [OPTIONS] <image>@<snapshot>\n"
-    "  Create a snapshot of image <name>. May be used live if only a single writer is active.\n"
+    "  Create a snapshot of image <image>. May be used live if only a single writer is active.\n"
+    "  Options:\n"
+    "  -p|--pool POOL    Move image to pool POOL, leaving the snapshot in the old pool.\n"
+    "  --enc-key random  Change image encryption key to a new random AES-256-XTS key.\n"
+    "  --enc-key KEY     Change image encryption key to a specified key, Vault key or to an empty key.\n"
+    "                    By default, the image retains its old key when taking a snapshot.\n"
     "\n"
     "vitastor-cli modify <name> [--rename <new-name>] [--resize <size>] [--readonly | --readwrite] [-f|--force] [--down-ok]\n"
     "  Rename, resize image or change its readonly status. Images with children can't be made read-write.\n"
     "  If the new size is smaller than the old size, extra data will be purged.\n"
     "  You should resize file system in the image, if present, before shrinking it.\n"
-    "  --deleted 1|0  Set/clear 'deleted image' flag (set automatically during unfinished deletes).\n"
+    "  --deleted 1|0       Set/clear 'deleted image' flag (set automatically during unfinished deletes).\n"
-    "  -f|--force     Proceed with shrinking or setting readwrite flag even if the image has children.\n"
+    "  -f|--force          Proceed with shrinking or setting readwrite flag even if the image has children.\n"
-    "  --down-ok      Proceed with shrinking even if some data will be left on unavailable OSDs.\n"
+    "  --down-ok           Proceed with shrinking even if some data will be left on unavailable OSDs.\n"
+    "  --enc-key HEX       Change image encryption key (allowed only with --force).\n"
+    "  --owner username    Change image owner.\n"
+    "  --owner_group name  Change image owner group name.\n"
+    "  --reader_group rdr  Change image reader group name.\n"
     "\n"
     "vitastor-cli dd [iimg=<image> | if=<file>] [oimg=<image> | of=<file>] [bs=1M]\n"
     "    [count=N] [seek/oseek=N] [skip/iseek=M] [iodepth=N] [status=progress]\n"
@@ -193,6 +211,7 @@ static const char* help_text =
     "    --used_for_app s3:<name>      Mark pool as used for S3 location with name <name>\n"
     "    --pg_stripe_size <number>     Increase object grouping stripe\n"
     "    --max_osd_combinations 10000  Maximum number of random combinations for LP solver input\n"
+    "    --creator_group <group>       User group allowed to create images in this pool.\n"
     "    --wait                        Wait for the new pool to come online\n"
     "    -f|--force                    Do not check that cluster has enough OSDs to create the pool\n"
     "  Examples:\n"
@@ -204,7 +223,7 @@ static const char* help_text =
     "    [-s|--pg_size <number>] [--pg_minsize <number>] [-n|--pg_count <count>]\n"
     "    [--failure_domain <level>] [--root_node <node>] [--osd_tags <tags>] [--used_for_app <type>:<name>]\n"
     "    [--max_osd_combinations <number>] [--primary_affinity_tags <tags>] [--scrub_interval <time>]\n"
-    "    [--level_placement <rules>] [--raw_placement <rules>]\n"
+    "    [--level_placement <rules>] [--raw_placement <rules>] [--creator_group <group>]\n"
     "  Non-modifiable parameters (changing them WILL lead to data loss):\n"
     "    [--block_size <size>] [--bitmap_granularity <size>]\n"
     "    [--immediate_commit <all|small|none>] [--pg_stripe_size <size>]\n"
@@ -217,7 +236,7 @@ static const char* help_text =
     "vitastor-cli rm-pool|pool-rm [--force] <id|name>\n"
     "  Remove a pool. Refuses to remove pools with images without --force.\n"
     "\n"
-    "vitastor-cli ls-pools|pool-ls|ls-pool|pools [-l] [--detail] [--sort FIELD] [-r] [-n N] [--stats] [<glob> ...]\n"
+    "vitastor-cli ls-pools|pool-ls|ls-pool|pools [-l] [--detail] [--sort FIELD] [-r] [-n N] [<glob> ...]\n"
     "  List pools (only matching <glob> patterns if passed).\n"
     "  -l|--long       Also report I/O statistics\n"
     "  --detail        Use list format (not table), show all details\n"
@@ -225,6 +244,25 @@ static const char* help_text =
     "  -r|--reverse    Sort in descending order\n"
     "  -n|--count N    Only list first N items\n"
     "\n"
+    "vitastor-cli ls-users|user-ls|ls-user|list-users [<name> ...]\n"
+    "  List users (only with specified names if passed).\n"
+    "\n"
+    "vitastor-cli modify-user --type <type> --groups group1,group2,... <username>\n"
+    "  Create or update user permissions. User names match CN of their certificates.\n"
+    "  --type TYPE      Set user type: client, admin, mon or osd. Default is client.\n"
+    "  --groups GROUPS  Set user's groups.\n"
+    "\n"
+    "vitastor-cli rm-user|remove-user|delete-user <username>\n"
+    "  Remove a user.\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"
+    "  --ssl_cert FILE      Path to server SSL certificate file (PEM format).\n"
+    "  --ssl_key FILE       Path to server SSL private key file.\n"
+    "  --ssl_ca FILE        Path to file with SSL 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"
     "\n"
     "GLOBAL OPTIONS:\n"
@@ -319,27 +357,24 @@ static json11::Json::object parse_args(int narg, const char *args[])
     return cfg;
 }
 
-static int run(cli_tool_t *p, json11::Json::object cfg)
+std::function<bool(cli_result_t &)> cli_tool_t::start(json11::Json::object cfg, cli_result_t & result)
 {
-    cli_result_t result = {};
-    p->is_command_line = true;
-    p->parse_config(cfg);
     json11::Json::array cmd = cfg["command"].array_items();
     cfg.erase("command");
     std::function<bool(cli_result_t &)> action_cb;
     if (!cmd.size())
     {
-        result = { .err = EINVAL, .text = "command is missing" };
+        result = { .err = EOPNOTSUPP, .text = "command is missing" };
     }
     else if (cmd[0] == "status")
     {
         // Show cluster status
-        action_cb = p->start_status(cfg);
+        action_cb = start_status(cfg);
     }
     else if (cmd[0] == "df")
     {
         // Show pool space stats
-        action_cb = p->start_pool_ls(cfg);
+        action_cb = start_pool_ls(cfg);
     }
     else if (cmd[0] == "ls")
     {
@@ -349,7 +384,7 @@ static int run(cli_tool_t *p, json11::Json::object cfg)
             cmd.erase(cmd.begin(), cmd.begin()+1);
             cfg["names"] = cmd;
         }
-        action_cb = p->start_ls(cfg);
+        action_cb = start_ls(cfg);
     }
     else if (cmd[0] == "snap-create")
     {
@@ -364,7 +399,7 @@ static int run(cli_tool_t *p, json11::Json::object cfg)
         {
             cfg["image"] = name.substr(0, pos);
             cfg["snapshot"] = name.substr(pos + 1);
-            action_cb = p->start_create(cfg);
+            action_cb = start_create(cfg);
         }
     }
     else if (cmd[0] == "create")
@@ -374,7 +409,7 @@ static int run(cli_tool_t *p, json11::Json::object cfg)
         {
             cfg["image"] = cmd[1];
         }
-        action_cb = p->start_create(cfg);
+        action_cb = start_create(cfg);
     }
     else if (cmd[0] == "modify")
     {
@@ -383,12 +418,12 @@ static int run(cli_tool_t *p, json11::Json::object cfg)
         {
             cfg["image"] = cmd[1];
         }
-        action_cb = p->start_modify(cfg);
+        action_cb = start_modify(cfg);
     }
     else if (cmd[0] == "rm-data")
     {
         // Delete inode data
-        action_cb = p->start_rm_data(cfg);
+        action_cb = start_rm_data(cfg);
     }
     else if (cmd[0] == "rm-osd")
     {
@@ -398,7 +433,7 @@ static int run(cli_tool_t *p, json11::Json::object cfg)
             cmd.erase(cmd.begin(), cmd.begin()+1);
             cfg["osd_id"] = cmd;
         }
-        action_cb = p->start_rm_osd(cfg);
+        action_cb = start_rm_osd(cfg);
     }
     else if (cmd[0] == "merge-data")
     {
@@ -409,7 +444,7 @@ static int run(cli_tool_t *p, json11::Json::object cfg)
             if (cmd.size() > 2)
                 cfg["to"] = cmd[2];
         }
-        action_cb = p->start_merge(cfg);
+        action_cb = start_merge(cfg);
     }
     else if (cmd[0] == "flatten")
     {
@@ -418,7 +453,7 @@ static int run(cli_tool_t *p, json11::Json::object cfg)
         {
             cfg["image"] = cmd[1];
         }
-        action_cb = p->start_flatten(cfg);
+        action_cb = start_flatten(cfg);
     }
     else if (cmd[0] == "dd")
     {
@@ -432,16 +467,31 @@ static int run(cli_tool_t *p, json11::Json::object cfg)
                 cfg[arg.substr(0, p)] = arg.substr(p+1);
             }
         }
-        action_cb = p->start_dd(cfg);
+        action_cb = start_dd(cfg);
     }
     else if (cmd[0] == "rm")
     {
         // Remove multiple snapshots and rebase their children
-        if (cfg["exact"].bool_value() || cfg["matching"].bool_value())
+        if (cfg["names"].is_array())
+        {
+            cfg["globs"] = cfg["names"];
+            cfg.erase("names");
+            cfg["exact"] = true;
+            cfg["matching"] = false;
+            action_cb = start_rm_wildcard(cfg);
+        }
+        else if (cfg["matching"].is_array())
+        {
+            cfg["globs"] = cfg["matching"];
+            cfg["exact"] = false;
+            cfg["matching"] = true;
+            action_cb = start_rm_wildcard(cfg);
+        }
+        else if (cfg["exact"].bool_value() || cfg["matching"].bool_value())
         {
             cmd.erase(cmd.begin(), cmd.begin()+1);
             cfg["globs"] = cmd;
-            action_cb = p->start_rm_wildcard(cfg);
+            action_cb = start_rm_wildcard(cfg);
         }
         else
         {
@@ -451,41 +501,41 @@ static int run(cli_tool_t *p, json11::Json::object cfg)
                 if (cmd.size() > 2)
                     cfg["to"] = cmd[2];
             }
-            action_cb = p->start_rm(cfg);
+            action_cb = start_rm(cfg);
         }
     }
     else if (cmd[0] == "describe")
     {
         // Describe unclean objects
-        action_cb = p->start_describe(cfg);
+        action_cb = start_describe(cfg);
     }
     else if (cmd[0] == "fix")
     {
         // Fix inconsistent objects (by deleting some copies)
-        action_cb = p->start_fix(cfg);
+        action_cb = start_fix(cfg);
     }
     else if (cmd[0] == "alloc-osd")
     {
         // Allocate a new OSD number
-        action_cb = p->start_alloc_osd(cfg);
+        action_cb = start_alloc_osd(cfg);
     }
     else if (cmd[0] == "osd-tree")
     {
         // Print OSD tree
-        action_cb = p->start_osd_tree(cfg);
+        cfg["as_tree"] = true;
+        action_cb = start_osd_tree(cfg);
     }
     else if (cmd[0] == "osds" || cmd[0] == "ls-osds" || cmd[0] == "ls-osd" || cmd[0] == "osd-ls")
     {
         // Print OSD list
-        cfg["flat"] = true;
+        action_cb = start_osd_tree(cfg);
-        action_cb = p->start_osd_tree(cfg);
     }
     else if (cmd[0] == "modify-osd")
     {
         // Modify OSD configuration
         if (cmd.size() > 1)
             cfg["osd_num"] = cmd[1];
-        action_cb = p->start_modify_osd(cfg);
+        action_cb = start_modify_osd(cfg);
     }
     else if (cmd[0] == "pg-list" || cmd[0] == "pg-ls" || cmd[0] == "list-pg" || cmd[0] == "ls-pg" || cmd[0] == "ls-pgs" || cmd[0] == "pgs")
     {
@@ -495,7 +545,7 @@ static int run(cli_tool_t *p, json11::Json::object cfg)
             cmd.erase(cmd.begin(), cmd.begin()+1);
             cfg["pg_state"] = cmd;
         }
-        action_cb = p->start_pg_list(cfg);
+        action_cb = start_pg_list(cfg);
     }
     else if (cmd[0] == "create-pool" || cmd[0] == "pool-create")
     {
@@ -504,16 +554,16 @@ static int run(cli_tool_t *p, json11::Json::object cfg)
         {
             cfg["name"] = cmd[1];
         }
-        action_cb = p->start_pool_create(cfg);
+        action_cb = start_pool_create(cfg);
     }
     else if (cmd[0] == "modify-pool" || cmd[0] == "pool-modify")
     {
         // Modify existing pool
         if (cmd.size() > 1)
         {
-            cfg["old_name"] = cmd[1];
+            cfg["pool"] = cmd[1];
         }
-        action_cb = p->start_pool_modify(cfg);
+        action_cb = start_pool_modify(cfg);
     }
     else if (cmd[0] == "rm-pool" || cmd[0] == "pool-rm")
     {
@@ -522,7 +572,7 @@ static int run(cli_tool_t *p, json11::Json::object cfg)
         {
             cfg["pool"] = cmd[1];
         }
-        action_cb = p->start_pool_rm(cfg);
+        action_cb = start_pool_rm(cfg);
     }
     else if (cmd[0] == "ls-pool" || cmd[0] == "pool-ls" || cmd[0] == "ls-pools" || cmd[0] == "pools")
     {
@@ -533,12 +583,55 @@ static int run(cli_tool_t *p, json11::Json::object cfg)
             cmd.erase(cmd.begin(), cmd.begin()+1);
             cfg["names"] = cmd;
         }
-        action_cb = p->start_pool_ls(cfg);
+        action_cb = start_pool_ls(cfg);
+    }
+    else if (cmd[0] == "user-ls" || cmd[0] == "ls-user" || cmd[0] == "ls-users" || cmd[0] == "list-users")
+    {
+        // List users
+        if (cmd.size() > 1)
+        {
+            cmd.erase(cmd.begin(), cmd.begin()+1);
+            cfg["names"] = cmd;
+        }
+        action_cb = start_user_ls(cfg);
+    }
+    else if (cmd[0] == "modify-user" || cmd[0] == "user-modify")
+    {
+        // Create/update user
+        if (cmd.size() > 1)
+        {
+            cfg["name"] = cmd[1];
+        }
+        action_cb = start_modify_user(cfg);
+    }
+    else if (cmd[0] == "rm-user" || cmd[0] == "remove-user" || cmd[0] == "delete-user")
+    {
+        // Remove user
+        if (cmd.size() > 1)
+        {
+            cfg["name"] = cmd[1];
+        }
+        cfg["remove"] = true;
+        action_cb = start_modify_user(cfg);
+    }
+    else if (cmd[0] == "serve")
+    {
+        // Start HTTP server
+        action_cb = start_serve(cfg);
     }
     else
     {
-        result = { .err = EINVAL, .text = "unknown command: "+cmd[0].string_value() };
+        result = { .err = EOPNOTSUPP, .text = "unknown command: "+cmd[0].string_value() };
     }
+    return action_cb;
+}
+
+static int run(cli_tool_t *p, json11::Json::object cfg)
+{
+    cli_result_t result = {};
+    p->is_command_line = true;
+    p->parse_config(cfg);
+    auto action_cb = p->start(cfg, result);
     if (action_cb != NULL)
     {
         // Create client
@@ -550,6 +643,7 @@ static int run(cli_tool_t *p, json11::Json::object cfg)
         {
             result = r;
             action_cb = NULL;
+            p->ringloop->submit();
         });
         // Loop until it completes
         while (action_cb != NULL)

src/cmd/cli.h

@@ -9,6 +9,7 @@
 #include "object_id.h"
 #include "ringloop.h"
 #include <functional>
+#include <set>
 
 struct rm_inode_t;
 struct snap_merger_t;
@@ -26,6 +27,13 @@ struct cli_result_t
     json11::Json data;
 };
 
+struct cli_user_t
+{
+    std::string name;
+    std::string type;
+    std::set<std::string> groups;
+};
+
 class cli_tool_t
 {
 public:
@@ -37,6 +45,8 @@ public:
     bool is_command_line = false;
     bool color = false;
 
+    std::unique_ptr<cli_user_t> user; // for http mode
+
     ring_loop_t *ringloop = NULL;
     epoll_manager_t *epmgr = NULL;
     cluster_client_t *cli = NULL;
@@ -46,18 +56,24 @@ public:
     json11::Json etcd_result;
 
     void parse_config(json11::Json::object & cfg);
+    void parse_api_opts(json11::Json::object & cfg);
     json11::Json parse_tags(std::string tags);
 
+    json11::Json::object format_image(const inode_config_t & cfg);
     void change_parent(inode_t cur, inode_t new_parent, cli_result_t *result);
     inode_config_t* get_inode_cfg(const std::string & name);
 
+    bool check_image_perm(const inode_config_t & cfg, bool write);
+
     friend struct rm_inode_t;
     friend struct snap_merger_t;
     friend struct snap_flattener_t;
     friend struct snap_remover_t;
 
+    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_create(json11::Json);
+    std::function<bool(cli_result_t &)> start_dd(json11::Json);
     std::function<bool(cli_result_t &)> start_describe(json11::Json);
     std::function<bool(cli_result_t &)> start_fix(json11::Json);
     std::function<bool(cli_result_t &)> start_flatten(json11::Json);
@@ -65,6 +81,7 @@ public:
     std::function<bool(cli_result_t &)> start_merge(json11::Json);
     std::function<bool(cli_result_t &)> start_modify(json11::Json);
     std::function<bool(cli_result_t &)> start_modify_osd(json11::Json);
+    std::function<bool(cli_result_t &)> start_modify_user(json11::Json);
     std::function<bool(cli_result_t &)> start_osd_tree(json11::Json);
     std::function<bool(cli_result_t &)> start_pg_list(json11::Json);
     std::function<bool(cli_result_t &)> start_pool_create(json11::Json);
@@ -75,8 +92,9 @@ public:
     std::function<bool(cli_result_t &)> start_rm_data(json11::Json);
     std::function<bool(cli_result_t &)> start_rm_osd(json11::Json);
     std::function<bool(cli_result_t &)> start_rm_wildcard(json11::Json);
+    std::function<bool(cli_result_t &)> start_serve(json11::Json);
     std::function<bool(cli_result_t &)> start_status(json11::Json);
-    std::function<bool(cli_result_t &)> start_dd(json11::Json);
+    std::function<bool(cli_result_t &)> start_user_ls(json11::Json);
 
     // Should be called like loop_and_wait(start_status(), <completion callback>)
     void loop_and_wait(std::function<bool(cli_result_t &)> loop_cb, std::function<void(const cli_result_t &)> complete_cb);

src/cmd/cli_common.cpp

@@ -6,6 +6,62 @@
 #include "cluster_client.h"
 #include "cli.h"
 
+bool cli_tool_t::check_image_perm(const inode_config_t & cfg, bool write)
+{
+    return !user ||
+        user->type == "admin" ||
+        user->name == cfg.owner ||
+        cfg.owner_group != "" && user->groups.find(cfg.owner_group) != user->groups.end() ||
+        !write && cfg.reader_group != "" && user->groups.find(cfg.reader_group) != user->groups.end();
+}
+
+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 img = json11::Json::object {
+        { "name", cfg.name },
+        { "size", cfg.size },
+        { "inode_id", cfg.num },
+        { "inode_num", INODE_NO_POOL(cfg.num) },
+        { "pool_id", (uint64_t)INODE_POOL(cfg.num) },
+        { "pool_name", good_pool ? pool_it->second.name : "? (ID:"+std::to_string(INODE_POOL(cfg.num))+")" },
+        { "readonly", cfg.readonly },
+        { "deleted", cfg.deleted },
+    };
+    if (cfg.owner != "")
+    {
+        img["owner"] = cfg.owner;
+    }
+    if (cfg.owner_group != "")
+    {
+        img["owner_group"] = cfg.owner_group;
+    }
+    if (cfg.reader_group != "")
+    {
+        img["reader_group"] = cfg.reader_group;
+    }
+    if (!cfg.enc_key.empty())
+    {
+        img["encrypted"] = true;
+        // Only show Vault key IDs
+        if (cfg.enc_key.substr(0, strlen(VAULT_KEY_PREFIX)) == VAULT_KEY_PREFIX)
+            img["enc_key_id"] = cfg.enc_key;
+    }
+    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())
+        {
+            img["parent_name"] = parent_it->second.name;
+        }
+        img["parent_inode_id"] = cfg.parent_id;
+        img["parent_inode_num"] = INODE_NO_POOL(cfg.parent_id);
+        img["parent_pool_id"] = (uint64_t)INODE_POOL(cfg.parent_id);
+    }
+    return img;
+}
+
 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);
@@ -101,6 +157,16 @@ inode_config_t* cli_tool_t::get_inode_cfg(const std::string & name)
     return NULL;
 }
 
+void cli_tool_t::parse_api_opts(json11::Json::object & cfg)
+{
+    iodepth = cfg["iodepth"].uint64_value();
+    if (!iodepth)
+        iodepth = 32;
+    parallel_osds = cfg["parallel_osds"].uint64_value();
+    if (!parallel_osds)
+        parallel_osds = 4;
+}
+
 void cli_tool_t::parse_config(json11::Json::object & cfg)
 {
     for (auto kv_it = cfg.begin(); kv_it != cfg.end();)
@@ -121,15 +187,10 @@ void cli_tool_t::parse_config(json11::Json::object & cfg)
     else
         color = isatty(1);
     json_output = cfg["json"].bool_value();
-    iodepth = cfg["iodepth"].uint64_value();
-    if (!iodepth)
-        iodepth = 32;
-    parallel_osds = cfg["parallel_osds"].uint64_value();
-    if (!parallel_osds)
-        parallel_osds = 4;
     log_level = cfg["log_level"].int64_value();
     progress = cfg["progress"].uint64_value() ? true : false;
     list_first = cfg["wait_list"].uint64_value() ? true : false;
+    parse_api_opts(cfg);
 }
 
 struct cli_result_looper_t
@@ -153,7 +214,6 @@ void cli_tool_t::loop_and_wait(std::function<bool(cli_result_t &)> loop_cb, std:
             ringloop->unregister_consumer(&looper->consumer);
             looper->loop_cb = NULL;
             looper->complete_cb(looper->result);
-            ringloop->submit();
             delete looper;
             return;
         }

src/cmd/cli_create.cpp

@@ -1,8 +1,13 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 (see README.md for details)
 
+#ifdef WITH_OPENSSL
+#include <openssl/rand.h>
+#endif
+
 #include <ctype.h>
 #include "cli.h"
+#include "http_client.h"
 #include "cluster_client.h"
 #include "str_util.h"
 
@@ -29,11 +34,15 @@ struct image_creator_t
     uint64_t size = 0;
     bool force = false;
     bool force_size = false;
+    std::string enc_key;
+    bool set_key = false;
+    std::string new_owner, new_owner_group, new_reader_group;
 
     pool_id_t old_pool_id = 0;
     inode_t new_parent_id = 0;
     inode_t new_id = 0, old_id = 0;
-    uint64_t max_id_mod_rev = 0, cfg_mod_rev = 0, idx_mod_rev = 0;
+    uint64_t max_id_mod_rev = 0, idx_mod_rev = 0;
+    inode_config_t cur_cfg;
     inode_config_t new_cfg;
 
     int state = 0;
@@ -64,7 +73,8 @@ struct image_creator_t
         }
         if (new_pool_id)
         {
-            if (pools.find(new_pool_id) == pools.end())
+            auto pool_it = pools.find(new_pool_id);
+            if (pool_it == pools.end())
             {
                 result = (cli_result_t){ .err = ENOENT, .text = "Pool "+std::to_string(new_pool_id)+" does not exist" };
                 state = 100;
@@ -111,6 +121,23 @@ struct image_creator_t
             create_snapshot();
     }
 
+    bool check_pool_permission()
+    {
+        if (!parent->user || parent->user->type == "admin")
+        {
+            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() ||
+            (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" };
+            state = 100;
+            return false;
+        }
+        return true;
+    }
+
     void create_image()
     {
         if (state == 2)
@@ -150,6 +177,10 @@ struct image_creator_t
             state = 100;
             return;
         }
+        if (!check_pool_permission())
+        {
+            return;
+        }
         if (!size && !force_size)
         {
             result = (cli_result_t){ .err = EINVAL, .text = "Image size is missing" };
@@ -194,15 +225,11 @@ resume_3:
         // 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);
+        auto img = parent->format_image(new_cfg);
         result = (cli_result_t){
             .err = 0,
             .text = "Image "+image_name+" created",
-            .data = json11::Json::object {
+            .data = img,
-                { "name", image_name },
-                { "pool", new_pool_name },
-                { "parent", new_parent },
-                { "size", size },
-            }
         };
         state = 100;
     }
@@ -233,7 +260,7 @@ resume_3:
         }
         do
         {
-            // In addition to next_id, get: size, old_id, old_pool_id, new_parent, cfg_mod_rev, idx_mod_rev
+            // In addition to next_id, get: cur_cfg, old_id, old_pool_id, size, idx_mod_rev
 resume_2:
 resume_3:
             get_image_details();
@@ -245,11 +272,22 @@ resume_3:
                 state = 100;
                 return;
             }
+            if (!parent->check_image_perm(cur_cfg, true))
+            {
+                result = (cli_result_t){ .err = EACCES, .text = "Image permission denied" };
+                state = 100;
+                return;
+            }
             if (!new_pool_id)
             {
                 // Create snapshot in the same pool by default
                 new_pool_id = old_pool_id;
             }
+            // Verify pool permissions if the pool is different from the original
+            if (new_pool_id != old_pool_id && !check_pool_permission())
+            {
+                return;
+            }
             attempt_create();
             state = 4;
 resume_4:
@@ -272,13 +310,23 @@ resume_4:
         // 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);
+        {
+            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,
             .text = "Snapshot "+image_name+"@"+new_snap+" created",
             .data = json11::Json::object {
-                { "name", image_name+"@"+new_snap },
+                { "inode_id", INODE_WITH_POOL(new_pool_id, new_id) },
-                { "pool", (uint64_t)new_pool_id },
+                { "inode_num", new_id },
-                { "parent", new_parent },
+                { "name", image_name },
+                { "pool_id", (uint64_t)new_pool_id },
+                { "pool_name", new_pool_name },
+                { "parent_name", image_name+"@"+new_snap },
+                { "parent_inode_id", INODE_WITH_POOL(old_pool_id, old_id) },
+                { "parent_inode_num", old_id },
+                { "parent_pool_id", (uint64_t)old_pool_id },
                 { "size", size },
             }
         };
@@ -323,17 +371,6 @@ resume_4:
             goto resume_2;
         else if (state == 3)
             goto resume_3;
-        if (!new_pool_id)
-        {
-            for (auto & ic: parent->cli->st_cli.inode_config)
-            {
-                if (ic.second.name == image_name)
-                {
-                    new_pool_id = INODE_POOL(ic.first);
-                    break;
-                }
-            }
-        }
         parent->etcd_txn(json11::Json::object { { "success", json11::Json::array {
             get_next_id(),
             json11::Json::object {
@@ -357,7 +394,7 @@ resume_2:
         extract_next_id(parent->etcd_result["responses"][0]);
         old_id = 0;
         old_pool_id = 0;
-        cfg_mod_rev = idx_mod_rev = 0;
+        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)
@@ -366,9 +403,8 @@ resume_2:
                 {
                     old_id = INODE_NO_POOL(ic.first);
                     old_pool_id = INODE_POOL(ic.first);
+                    cur_cfg = ic.second;
                     size = ic.second.size;
-                    new_parent_id = ic.second.parent_id;
-                    cfg_mod_rev = ic.second.mod_revision;
                     break;
                 }
             }
@@ -412,16 +448,14 @@ resume_3:
             }
             {
                 auto kv = parent->cli->st_cli.parse_etcd_kv(parent->etcd_result["responses"][0]["response_range"]["kvs"][0]);
-                size = kv.value["size"].uint64_value();
+                cur_cfg = parent->cli->st_cli.deserialize_inode_cfg(INODE_WITH_POOL(old_pool_id, old_id), kv.value, kv.mod_revision);
-                new_parent_id = kv.value["parent_id"].uint64_value();
+                size = cur_cfg.size;
-                uint64_t parent_pool_id = kv.value["parent_pool"].uint64_value();
-                if (new_parent_id)
-                {
-                    new_parent_id = INODE_WITH_POOL(parent_pool_id ? parent_pool_id : old_pool_id, new_parent_id);
-                }
-                cfg_mod_rev = kv.mod_revision;
             }
         }
+        if (!new_pool_id)
+        {
+            new_pool_id = old_pool_id;
+        }
     }
 
     void attempt_create()
@@ -434,6 +468,27 @@ resume_3:
             .readonly = false,
             .meta = new_meta,
         };
+        if (set_key)
+        {
+            new_cfg.enc_key = enc_key;
+        }
+        else if (new_snap != "")
+        {
+            new_cfg.enc_key = cur_cfg.enc_key;
+        }
+        new_cfg.owner = http_context_get_ssl_cn(parent->cli->st_cli.get_http_ctx());
+        if (!new_owner.empty())
+        {
+            new_cfg.owner = new_owner;
+        }
+        if (!new_owner_group.empty())
+        {
+            new_cfg.owner_group = new_owner_group;
+        }
+        if (!new_reader_group.empty())
+        {
+            new_cfg.reader_group = new_reader_group;
+        }
         json11::Json::array checks = json11::Json::array {
             json11::Json::object {
                 { "target", "VERSION" },
@@ -500,16 +555,12 @@ resume_3:
         };
         if (new_snap != "")
         {
-            inode_config_t snap_cfg = {
+            inode_config_t snap_cfg = cur_cfg;
-                .num = INODE_WITH_POOL(old_pool_id, old_id),
+            snap_cfg.name = image_name+"@"+new_snap;
-                .name = image_name+"@"+new_snap,
+            snap_cfg.readonly = true;
-                .size = size,
-                .parent_id = new_parent_id,
-                .readonly = true,
-            };
             checks.push_back(json11::Json::object {
                 { "target", "MOD" },
-                { "mod_revision", cfg_mod_rev },
+                { "mod_revision", cur_cfg.mod_revision },
                 { "key", base64_encode(
                     parent->cli->st_cli.etcd_prefix+"/config/inode/"+
                     std::to_string(old_pool_id)+"/"+std::to_string(old_id)
@@ -554,8 +605,16 @@ std::function<bool(cli_result_t &)> cli_tool_t::start_create(json11::Json cfg)
     auto image_creator = new image_creator_t();
     image_creator->parent = this;
     image_creator->image_name = cfg["image"].string_value();
-    image_creator->new_pool_id = cfg["pool"].uint64_value();
+    if (!cfg["pool"].is_null())
-    image_creator->new_pool_name = cfg["pool"].string_value();
+    {
+        image_creator->new_pool_id = cfg["pool"].uint64_value();
+        image_creator->new_pool_name = cfg["pool"].string_value();
+    }
+    else
+    {
+        image_creator->new_pool_id = cfg["pool_id"].uint64_value();
+        image_creator->new_pool_name = cfg["pool_name"].string_value();
+    }
     image_creator->force = cfg["force"].bool_value();
     image_creator->force_size = cfg["force_size"].bool_value();
     if (cfg["image_meta"].is_object())
@@ -566,6 +625,35 @@ std::function<bool(cli_result_t &)> cli_tool_t::start_create(json11::Json cfg)
     {
         image_creator->new_snap = cfg["snapshot"].string_value();
     }
+    if (!cfg["enc_key"].is_null())
+    {
+        image_creator->set_key = true;
+#ifdef WITH_OPENSSL
+        if (image_creator->enc_key == "random")
+        {
+            uint8_t newkey[64];
+            RAND_bytes(newkey, 64);
+            image_creator->enc_key = tohexstr(newkey, 64);
+        }
+#endif
+        else
+        {
+            image_creator->enc_key = cfg["enc_key"].string_value();
+            if (image_creator->enc_key != "" &&
+                image_creator->enc_key.substr(0, strlen(VAULT_KEY_PREFIX)) != VAULT_KEY_PREFIX &&
+                (!ishexstr(image_creator->enc_key) || image_creator->enc_key.size() != 128))
+            {
+                return [](cli_result_t & result)
+                {
+                    result = (cli_result_t){ .err = EINVAL, .text = "Encryption key is not a 512-bit hex string, not \"\" and not \"random\"" };
+                    return true;
+                };
+            }
+        }
+    }
+    image_creator->new_owner = cfg["owner"].string_value();
+    image_creator->new_owner_group = cfg["owner_group"].string_value();
+    image_creator->new_reader_group = cfg["reader_group"].string_value();
     image_creator->new_parent = cfg["parent"].string_value();
     if (!cfg["size"].is_null())
     {

src/cmd/cli_dd.cpp

@@ -864,7 +864,7 @@ resume_2:
         // Copy data
         if (iinfo.in_seekable && iseek >= iinfo.in_size)
         {
-            result = (cli_result_t){ .err = -EINVAL, .text = "Input seek position is beyond end of input" };
+            result = (cli_result_t){ .err = EINVAL, .text = "Input seek position is beyond end of input" };
             goto close_end;
         }
         if (!iinfo.iwatch && !iinfo.in_seekable && iseek)

src/cmd/cli_describe.cpp

@@ -57,12 +57,24 @@ struct cli_describe_t
 
     void parse_options(json11::Json cfg)
     {
-        only_pool = cfg["pool"].uint64_value();
+        uint64_t pool_id;
-        if (!only_pool && cfg["pool"].is_string())
+        std::string pool_name;
+        if (!cfg["pool"].is_null())
+        {
+            pool_id = cfg["pool"].uint64_value();
+            pool_name = pool_id ? "" : cfg["pool"].string_value();
+        }
+        else
+        {
+            pool_id = cfg["pool_id"].uint64_value();
+            pool_name = pool_id ? "" : cfg["pool_name"].string_value();
+        }
+        only_pool = pool_id;
+        if (!only_pool && pool_name != "")
         {
             for (auto & pp: parent->cli->st_cli.pool_config)
             {
-                if (pp.second.name == cfg["pool"].string_value())
+                if (pp.second.name == pool_name)
                 {
                     only_pool = pp.first;
                     break;
@@ -106,6 +118,22 @@ struct cli_describe_t
             if (cfg["object_state"].string_value().find("misplaced") != std::string::npos)
                 object_state |= OBJ_MISPLACED;
         }
+        else if (!object_state && cfg["object_state"].is_array())
+        {
+            for (auto & st: cfg["object_state"].array_items())
+            {
+                if (st == "inconsistent")
+                    object_state |= OBJ_INCONSISTENT;
+                else if (st == "corrupted")
+                    object_state |= OBJ_CORRUPTED;
+                else if (st == "incomplete")
+                    object_state |= OBJ_INCOMPLETE;
+                else if (st == "degraded")
+                    object_state |= OBJ_DEGRADED;
+                else if (st == "misplaced")
+                    object_state |= OBJ_MISPLACED;
+            }
+        }
     }
 
     void loop()

src/cmd/cli_flatten.cpp

@@ -35,6 +35,12 @@ struct snap_flattener_t
             state = 100;
             return;
         }
+        if (!parent->check_image_perm(*target_cfg, true))
+        {
+            result = (cli_result_t){ .err = EACCES, .text = "Image permission denied" };
+            state = 100;
+            return;
+        }
         target_id = target_cfg->num;
         std::vector<inode_t> chain_list;
         inode_config_t *cur = target_cfg;

src/cmd/cli_ls.cpp

@@ -17,6 +17,7 @@ struct image_lister_t
     std::string list_pool_name;
     std::string sort_field;
     std::set<std::string> only_names;
+    std::vector<uint64_t> only_ids;
     bool reverse = false;
     bool exact = false;
     bool tree = false;
@@ -52,35 +53,19 @@ struct image_lister_t
                 return;
             }
         }
-        for (auto & ic: parent->cli->st_cli.inode_config)
+        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();
+        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();
+        for (auto it = begin_it; it != end_it; it++)
         {
-            if (list_pool_id && INODE_POOL(ic.second.num) != list_pool_id)
+            if (!parent->check_image_perm(it->second, false))
             {
                 continue;
             }
-            auto pool_it = parent->cli->st_cli.pool_config.find(INODE_POOL(ic.second.num));
+            stats[it->second.num] = parent->format_image(it->second);
-            bool good_pool = pool_it != parent->cli->st_cli.pool_config.end();
-            auto item = json11::Json::object {
-                { "name", ic.second.name },
-                { "size", ic.second.size },
-                { "used_size", 0 },
-                { "readonly", ic.second.readonly },
-                { "pool_id", (uint64_t)INODE_POOL(ic.second.num) },
-                { "pool_name", good_pool ? pool_it->second.name : "? (ID:"+std::to_string(INODE_POOL(ic.second.num))+")" },
-                { "inode_num", INODE_NO_POOL(ic.second.num) },
-                { "inode_id", ic.second.num },
-                { "deleted", ic.second.deleted },
-            };
-            if (ic.second.parent_id)
-            {
-                auto p_it = parent->cli->st_cli.inode_config.find(ic.second.parent_id);
-                item["parent_name"] = p_it != parent->cli->st_cli.inode_config.end()
-                    ? p_it->second.name : "";
-                item["parent_pool_id"] = (uint64_t)INODE_POOL(ic.second.parent_id);
-                item["parent_inode_num"] = INODE_NO_POOL(ic.second.parent_id);
-                item["parent_inode_id"] = ic.second.parent_id;
-            }
-            stats[ic.second.num] = item;
         }
     }
 
@@ -127,6 +112,7 @@ resume_1:
             state = 100;
             return;
         }
+        // FIXME: Do not always read everything
         space_info = parent->etcd_result;
         std::map<pool_id_t, uint64_t> pool_pg_real_size;
         for (auto & kv_item: space_info["responses"][0]["response_range"]["kvs"].array_items())
@@ -160,6 +146,11 @@ resume_1:
             }
             inode_t inode_num = INODE_WITH_POOL(pool_id, only_inode_num);
             uint64_t used_size = kv.value["raw_used"].uint64_value();
+            auto stat_it = stats.find(inode_num);
+            if (parent->user && parent->user->type != "admin" && stat_it == stats.end())
+            {
+                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())
@@ -168,7 +159,6 @@ resume_1:
                 used_size = used_size / (pool_pg_real_size[pool_id] ? pool_pg_real_size[pool_id] : 1)
                     * (pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks);
             }
-            auto stat_it = stats.find(inode_num);
             if (stat_it == stats.end())
             {
                 stats[inode_num] = json11::Json::object {
@@ -202,20 +192,33 @@ resume_1:
     json11::Json::array to_list()
     {
         json11::Json::array list;
-        for (auto & kv: stats)
+        if (only_ids.size())
         {
-            if (!only_names.size())
+            for (auto & id: only_ids)
             {
-                list.push_back(kv.second);
+                if (stats.find(id) != stats.end())
-            }
-            else
-            {
-                for (auto & glob: only_names)
                 {
-                    if (exact ? (kv.second["name"].string_value() == glob) : stupid_glob(kv.second["name"].string_value(), glob))
+                    list.push_back(stats[id]);
+                }
+            }
+        }
+        else
+        {
+            for (auto & kv: stats)
+            {
+                if (!only_names.size())
+                {
+                    list.push_back(kv.second);
+                }
+                else
+                {
+                    for (auto & glob: only_names)
                     {
-                        list.push_back(kv.second);
+                        if (exact ? (kv.second["name"].string_value() == glob) : stupid_glob(kv.second["name"].string_value(), glob))
-                        break;
+                        {
+                            list.push_back(kv.second);
+                            break;
+                        }
                     }
                 }
             }
@@ -371,7 +374,7 @@ resume_1:
             }
         }
         cols.push_back(json11::Json::object{
-            { "key", "ro" },
+            { "key", "flags" },
             { "title", "FLAGS" },
             { "right", true },
         });
@@ -399,8 +402,14 @@ resume_1:
                 kv.second["delete_q"] = format_q(kv.second["delete_queue"].number_value());
             }
             kv.second["size_fmt"] = format_size(kv.second["size"].uint64_value());
-            kv.second["ro"] = kv.second["deleted"].bool_value() ? "DEL" :
+            std::string flags;
-                (kv.second["readonly"].bool_value() ? "RO" : "-");
+            if (kv.second["deleted"].bool_value())
+                flags += "DEL";
+            if (kv.second["readonly"].bool_value())
+                flags += (flags.empty() ? "RO" : ",RO");
+            if (kv.second["encrypted"].bool_value())
+                flags += (flags.empty() ? "ENC" : ",ENC");
+            kv.second["flags"] = flags;
         }
         result.text = print_table(tree ? to_tree(to_list()) : to_list(), cols, parent->color);
         state = 100;
@@ -570,17 +579,30 @@ std::function<bool(cli_result_t &)> cli_tool_t::start_ls(json11::Json cfg)
     lister->parent = this;
     lister->exact = cfg["exact"].bool_value();
     lister->tree = cfg["tree"].bool_value();
-    lister->list_pool_id = cfg["pool"].uint64_value();
+    if (!cfg["pool"].is_null())
-    lister->list_pool_name = lister->list_pool_id ? "" : cfg["pool"].as_string();
+    {
+        lister->list_pool_id = cfg["pool"].uint64_value();
+        lister->list_pool_name = lister->list_pool_id ? "" : cfg["pool"].as_string();
+    }
+    else
+    {
+        lister->list_pool_id = cfg["pool_id"].uint64_value();
+        lister->list_pool_name = lister->list_pool_id ? "" : cfg["pool_name"].string_value();
+    }
     lister->show_stats = cfg["long"].bool_value();
     lister->show_delete = cfg["del"].bool_value();
     lister->sort_field = cfg["sort"].string_value() != "" ? cfg["sort"].string_value() : "name";
     lister->reverse = cfg["reverse"].bool_value();
     lister->max_count = cfg["count"].uint64_value();
+    if (cfg["names"].is_string())
+        lister->only_names.insert(cfg["names"].string_value());
     for (auto & item: cfg["names"].array_items())
-    {
         lister->only_names.insert(item.string_value());
-    }
+    if (cfg["ids"].is_string())
+        for (auto & item: explode(",", cfg["ids"].string_value(), true))
+            lister->only_ids.push_back(stoull_full(item));
+    for (auto & item: cfg["ids"].array_items())
+        lister->only_ids.push_back(item.uint64_value());
     return [lister](cli_result_t & result)
     {
         lister->loop();

src/cmd/cli_merge.cpp

@@ -374,7 +374,7 @@ struct snap_merger_t
         result = (cli_result_t){ .text = "Done, layers from "+from_name+" to "+to_name+" merged into "+target_name, .data = json11::Json::object {
             { "from", from_name },
             { "to", to_name },
-            { "into", target_name },
+            { "target", target_name },
         }};
         state = 100;
     resume_100:

src/cmd/cli_modify.cpp

@@ -17,6 +17,9 @@ struct image_changer_t
     bool force_size = false, inc_size = false;
     bool set_readonly = false, set_readwrite = false, force = false;
     bool set_deleted = false, new_deleted = false;
+    bool set_key = false;
+    std::string enc_key;
+    json11::Json new_owner, new_owner_group, new_reader_group;
     bool down_ok = false;
     // interval between fsyncs
     int fsync_interval = 128;
@@ -74,6 +77,12 @@ struct image_changer_t
             state = 100;
             return;
         }
+        if (!parent->check_image_perm(cfg, true))
+        {
+            result = (cli_result_t){ .err = EACCES, .text = "Image permission denied" };
+            state = 100;
+            return;
+        }
         for (auto & ic: parent->cli->st_cli.inode_config)
         {
             if (ic.second.parent_id == inode_num)
@@ -88,10 +97,7 @@ struct image_changer_t
             (!new_size && !force_size || cfg.size == new_size || cfg.size >= new_size && inc_size) &&
             (new_name == "" || new_name == image_name))
         {
-            result = (cli_result_t){ .err = 0, .text = "No change", .data = json11::Json::object {
+            result = (cli_result_t){ .err = 0, .text = "No change", .data = parent->format_image(cfg) };
-                { "error_code", 0 },
-                { "error_text", "No change" },
-            }};
             state = 100;
             return;
         }
@@ -152,6 +158,36 @@ resume_1:
         {
             cfg.name = new_name;
         }
+        if (new_owner.is_string())
+        {
+            cfg.owner = new_owner.string_value();
+        }
+        if (new_owner_group.is_string())
+        {
+            cfg.owner_group = new_owner_group.string_value();
+        }
+        if (new_reader_group.is_string())
+        {
+            cfg.reader_group = new_reader_group.string_value();
+        }
+        if (set_key)
+        {
+            if (!force)
+            {
+                result = (cli_result_t){ .err = EINVAL, .text = "Changing image encryption key is only allowed with --force" };
+                state = 100;
+                return;
+            }
+            if (enc_key != "" &&
+                enc_key.substr(0, strlen(VAULT_KEY_PREFIX)) != VAULT_KEY_PREFIX &&
+                (!ishexstr(enc_key) || enc_key.size() != 128))
+            {
+                result = (cli_result_t){ .err = EINVAL, .text = "Encryption key is not a 512-bit hex string and not \"\"" };
+                state = 100;
+                return;
+            }
+            cfg.enc_key = enc_key;
+        }
         {
             std::string cur_cfg_key = base64_encode(parent->cli->st_cli.etcd_prefix+
                 "/config/inode/"+std::to_string(INODE_POOL(inode_num))+
@@ -235,12 +271,7 @@ resume_2:
         result = (cli_result_t){
             .err = 0,
             .text = "Image "+image_name+" modified",
-            .data = json11::Json::object {
+            .data = parent->format_image(cfg)
-                { "name", image_name },
-                { "inode", INODE_NO_POOL(inode_num) },
-                { "pool", (uint64_t)INODE_POOL(inode_num) },
-                { "size", new_size },
-            }
         };
         state = 100;
     }
@@ -261,9 +292,14 @@ std::function<bool(cli_result_t &)> cli_tool_t::start_modify(json11::Json cfg)
     changer->set_deleted = !cfg["deleted"].is_null();
     changer->new_deleted = json_is_true(cfg["deleted"]);
     changer->fsync_interval = cfg["fsync_interval"].uint64_value();
+    changer->enc_key = cfg["enc_key"].string_value();
+    changer->set_key = cfg["enc_key"].is_string();
     if (!changer->fsync_interval)
         changer->fsync_interval = 128;
     changer->down_ok = cfg["down_ok"].bool_value();
+    changer->new_owner = cfg["owner"];
+    changer->new_owner_group = cfg["owner_group"];
+    changer->new_reader_group = cfg["reader_group"];
     // FIXME Check that the image doesn't have children when shrinking
     return [changer](cli_result_t & result)
     {

src/cmd/cli_modify_user.cpp

@@ -0,0 +1,161 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+#include "cli.h"
+#include "cluster_client.h"
+#include "str_util.h"
+
+// Create/update/delete a user
+struct cli_modify_user_t
+{
+    cli_tool_t *parent;
+
+    std::string user_name;
+    std::string user_type;
+    json11::Json groups;
+    bool del = false;
+
+    int state = 0;
+    cli_result_t result;
+    etcd_kv_t kv;
+    json11::Json::object new_cfg;
+
+    bool is_done()
+    {
+        return state == 100;
+    }
+
+    void loop()
+    {
+        if (state == 1)
+            goto resume_1;
+        else if (state == 2)
+            goto resume_2;
+        if (user_type != "client" && user_type != "admin" && user_type != "mon" && user_type != "osd")
+        {
+            result = (cli_result_t){ .err = EINVAL, .text = "Unknown user type: "+user_type };
+            state = 100;
+            return;
+        }
+        if (groups.is_string())
+        {
+            groups = groups == "" ? std::vector<std::string>() : explode(",", groups.string_value(), true);
+        }
+        else if (groups.is_array())
+        {
+            for (auto & gr: groups.array_items())
+            {
+                if (!gr.is_string())
+                {
+                    result = (cli_result_t){ .err = EINVAL, .text = "Group names must be strings" };
+                    state = 100;
+                    return;
+                }
+            }
+        }
+        else if (!groups.is_null())
+        {
+            result = (cli_result_t){ .err = EINVAL, .text = "Group names must be strings" };
+            state = 100;
+            return;
+        }
+        if (user_name == "")
+        {
+            result = (cli_result_t){ .err = EINVAL, .text = "User name must not be empty" };
+            state = 100;
+            return;
+        }
+        parent->etcd_txn(json11::Json::object {
+            { "success", json11::Json::array { json11::Json::object {
+                { "request_range", json11::Json::object {
+                    { "key", base64_encode(parent->cli->st_cli.etcd_prefix+"/config/user/"+user_name) },
+                } },
+            } } }
+        });
+        state = 1;
+resume_1:
+        if (parent->waiting > 0)
+            return;
+        if (parent->etcd_err.err)
+        {
+            result = parent->etcd_err;
+            state = 100;
+            return;
+        }
+        kv = parent->cli->st_cli.parse_etcd_kv(parent->etcd_result["responses"][0]["response_range"]["kvs"][0]);
+        while (true)
+        {
+            new_cfg = kv.value.object_items();
+            if (!kv.mod_revision && del)
+            {
+                result = (cli_result_t){ .err = ENOENT, .text = "User "+user_name+" does not exist" };
+                state = 100;
+                break;
+            }
+            if (!groups.is_null())
+                new_cfg["groups"] = groups;
+            if (user_type != "")
+                new_cfg["type"] = user_type;
+            if (!new_cfg["type"].is_string())
+                new_cfg["type"] = "client";
+            parent->etcd_txn(json11::Json::object {
+                { "compare", json11::Json::array { json11::Json::object {
+                    { "key", base64_encode(parent->cli->st_cli.etcd_prefix+"/config/user/"+user_name) },
+                    { "target", kv.mod_revision ? "MOD" : "VERSION" },
+                    { kv.mod_revision ? "mod_revision" : "version", kv.mod_revision },
+                } } },
+                { "success", json11::Json::array {
+                    del ? json11::Json::object { { "request_delete_range", json11::Json::object {
+                        { "key", base64_encode(parent->cli->st_cli.etcd_prefix+"/config/user/"+user_name) },
+                    } } } : json11::Json::object { { "request_put", json11::Json::object {
+                        { "key", base64_encode(parent->cli->st_cli.etcd_prefix+"/config/user/"+user_name) },
+                        { "value", base64_encode(json11::Json(new_cfg).dump()) }
+                    } } },
+                } },
+                { "failure", json11::Json::array { json11::Json::object {
+                    { "request_range", json11::Json::object {
+                        { "key", base64_encode(parent->cli->st_cli.etcd_prefix+"/config/user/"+user_name) },
+                    } },
+                } } },
+            });
+            state = 2;
+resume_2:
+            if (parent->waiting > 0)
+                return;
+            if (parent->etcd_err.err)
+            {
+                result = parent->etcd_err;
+                state = 100;
+                return;
+            }
+            if (parent->etcd_result["succeeded"].bool_value())
+                break;
+            kv = parent->cli->st_cli.parse_etcd_kv(parent->etcd_result["responses"][0]["response_range"]["kvs"][0]);
+        }
+        state = 100;
+        result.text = del ? "User "+user_name+" removed" : "User "+user_name+" modified";
+        new_cfg["name"] = user_name;
+        result.data = del ? json11::Json::object{ { "ok", true } } : new_cfg;
+    }
+};
+
+std::function<bool(cli_result_t &)> cli_tool_t::start_modify_user(json11::Json cfg)
+{
+    auto creator = new cli_modify_user_t();
+    creator->parent = this;
+    creator->user_name = cfg["name"].string_value();
+    creator->user_type = cfg["type"].string_value();
+    creator->groups = cfg["groups"];
+    creator->del = cfg["remove"].bool_value();
+    return [creator](cli_result_t & result)
+    {
+        creator->loop();
+        if (creator->is_done())
+        {
+            result = creator->result;
+            delete creator;
+            return true;
+        }
+        return false;
+    };
+}

src/cmd/cli_osd_tree.cpp

@@ -41,7 +41,7 @@ struct osd_tree_printer_t
 {
     cli_tool_t *parent;
     json11::Json cfg;
-    bool flat = false;
+    bool as_tree = false;
     bool show_stats = false;
 
     int state = 0;
@@ -209,11 +209,14 @@ resume_1:
             for (int i = 1; i < node_seq.size(); i++)
             {
                 auto & node = placement_tree->nodes.at(node_seq[i]);
-                fmt_items.push_back(json11::Json::object{
+                if (as_tree)
-                    { "type", node.level },
+                {
-                    { "name", node.name },
+                    fmt_items.push_back(json11::Json::object{
-                    { "parent", node.parent },
+                        { "type", node.level },
-                });
+                        { "name", node.name },
+                        { "parent", node.parent },
+                    });
+                }
                 for (uint64_t osd_num: node.child_osds)
                 {
                     auto & osd = placement_tree->osds.at(osd_num);
@@ -221,17 +224,22 @@ resume_1:
                         { "type", "osd" },
                         { "name", osd.num },
                         { "parent", node.name },
-                        { "up", osd.up ? "up" : "down" },
+                        { "up", osd.up },
                         { "size", osd.size },
                         { "free", osd.free },
                         { "reweight", osd.reweight },
                         { "noout", osd.noout },
                         { "tags", osd.tags },
-                        { "block", (uint64_t)osd.block_size },
+                        { "data_block_size", (uint64_t)osd.block_size },
-                        { "bitmap", (uint64_t)osd.bitmap_granularity },
+                        { "bitmap_granularity", (uint64_t)osd.bitmap_granularity },
-                        { "commit", osd.immediate_commit == IMMEDIATE_NONE ? "none" : (osd.immediate_commit == IMMEDIATE_ALL ? "all" : "small") },
+                        { "immediate_commit", osd.immediate_commit == IMMEDIATE_NONE ? "none" : (osd.immediate_commit == IMMEDIATE_ALL ? "all" : "small") },
-                        { "op_stats", osd_stats[osd_num]["op_stats"] },
                     };
+                    if (show_stats)
+                    {
+                        json_osd["op_stats"] = osd_stats[osd_num]["op_stats"];
+                        json_osd["subop_stats"] = osd_stats[osd_num]["subop_stats"];
+                        json_osd["recovery_stats"] = osd_stats[osd_num]["recovery_stats"];
+                    }
                     if (osd_stats[osd_num]["slow_ops_primary"].uint64_value() > 0)
                     {
                         json_osd["slow_ops_primary"] = osd_stats[osd_num]["slow_ops_primary"];
@@ -249,7 +257,7 @@ resume_1:
         for (int i = 1; i < node_seq.size(); i++)
         {
             auto & node = placement_tree->nodes.at(node_seq[i]);
-            if (!flat)
+            if (as_tree)
             {
                 fmt_items.push_back(json11::Json::object{
                     { "type", str_repeat("  ", indents[i]) + node.level },
@@ -257,7 +265,7 @@ resume_1:
                 });
             }
             std::string parent = node.name;
-            if (flat)
+            if (!as_tree)
             {
                 auto cur = &placement_tree->nodes.at(node.name);
                 while (cur->parent != "" && cur->parent != node.name)
@@ -270,7 +278,7 @@ resume_1:
             {
                 auto & osd = placement_tree->osds.at(osd_num);
                 auto fmt = json11::Json::object{
-                    { "type", (flat ? "osd" : str_repeat("  ", indents[i]+1) + "osd") },
+                    { "type", (!as_tree ? "osd" : str_repeat("  ", indents[i]+1) + "osd") },
                     { "name", osd.num },
                     { "parent", parent },
                     { "up", osd.up ? "up" : "down" },
@@ -300,7 +308,7 @@ resume_1:
             }
         }
         json11::Json::array cols;
-        if (!flat)
+        if (as_tree)
         {
             cols.push_back(json11::Json::object{
                 { "key", "type" },
@@ -309,9 +317,9 @@ resume_1:
         }
         cols.push_back(json11::Json::object{
             { "key", "name" },
-            { "title", flat ? "OSD" : "NAME" },
+            { "title", !as_tree ? "OSD" : "NAME" },
         });
-        if (flat)
+        if (!as_tree)
         {
             cols.push_back(json11::Json::object{
                 { "key", "parent" },
@@ -414,7 +422,7 @@ std::function<bool(cli_result_t &)> cli_tool_t::start_osd_tree(json11::Json cfg)
     auto osd_tree_printer = new osd_tree_printer_t();
     osd_tree_printer->parent = this;
     osd_tree_printer->cfg = cfg;
-    osd_tree_printer->flat = cfg["flat"].bool_value();
+    osd_tree_printer->as_tree = cfg["as_tree"].bool_value();
     osd_tree_printer->show_stats = cfg["long"].bool_value();
     return [osd_tree_printer](cli_result_t & result)
     {

src/cmd/cli_pg_ls.cpp

@@ -282,10 +282,16 @@ std::function<bool(cli_result_t &)> cli_tool_t::start_pg_list(json11::Json cfg)
 {
     auto pg_lister = new pg_lister_t();
     pg_lister->parent = this;
-    if (cfg["pool"].uint64_value())
+    if (!cfg["pool"].is_null())
+    {
         pg_lister->pool_id = cfg["pool"].uint64_value();
+        pg_lister->pool_name = pg_lister->pool_id ? "" : cfg["pool"].string_value();
+    }
     else
-        pg_lister->pool_name = cfg["pool"].string_value();
+    {
+        pg_lister->pool_id = cfg["pool_id"].uint64_value();
+        pg_lister->pool_name = pg_lister->pool_id ? "" : cfg["pool_name"].string_value();
+    }
     for (auto & st: cfg["pg_state"].array_items())
         pg_lister->pg_state.push_back(st.string_value());
     if (cfg["pg_state"].is_string())

src/cmd/cli_pool_cfg.cpp

@@ -91,7 +91,7 @@ std::string validate_pool_config(json11::Json::object & new_cfg, json11::Json ol
         }
         else if (key == "name" || key == "scheme" || key == "immediate_commit" ||
             key == "failure_domain" || key == "root_node" || key == "scrub_interval" || key == "used_for_app" ||
-            key == "used_for_fs" || key == "raw_placement" || key == "local_reads")
+            key == "used_for_fs" || key == "raw_placement" || key == "local_reads" || key == "creator_group")
         {
             if (!value.is_string())
             {

src/cmd/cli_pool_create.cpp

@@ -213,7 +213,7 @@ resume_3:
                     if (failure_domain != "osd")
                         pool_err += "\n- different parent '"+failure_domain+"' nodes";
                     result = (cli_result_t){
-                        .err = EINVAL,
+                        .err = EBUSY,
                         .text = pool_err,
                     };
                     state = 100;

src/cmd/cli_pool_ls.cpp

@@ -206,7 +206,7 @@ resume_1:
                 { "space_efficiency", pool_stats[pool_cfg.id]["space_efficiency"].number_value() },
                 { "pg_real_size", pool_stats[pool_cfg.id]["pg_real_size"].uint64_value() },
                 { "osd_count", (uint64_t)pg_per_osd.size() },
-                { "backfillfull", pool_cfg.backfillfull },
+                { "backfillfull", !!pool_cfg.backfillfull },
             };
         }
         // Include full pool config
@@ -546,6 +546,10 @@ resume_3:
                 { "write_fmt", "Write" },
                 { "delete_fmt", "Delete" },
             };
+            if (sort_field == "osd_tags" || sort_field == "primary_affinity_tags")
+            {
+                sort_field += "_fmt";
+            }
             auto list = to_list();
             size_t title_len = 0;
             for (auto & item: list)
@@ -666,15 +670,12 @@ std::function<bool(cli_result_t &)> cli_tool_t::start_pool_ls(json11::Json cfg)
     lister->show_stats = cfg["long"].bool_value();
     lister->detailed = cfg["detail"].bool_value();
     lister->sort_field = cfg["sort"].string_value();
-    if ((lister->sort_field == "osd_tags") ||
-        (lister->sort_field == "primary_affinity_tags" ))
-        lister->sort_field = lister->sort_field + "_fmt";
     lister->reverse = cfg["reverse"].bool_value();
     lister->max_count = cfg["count"].uint64_value();
+    if (cfg["names"].is_string())
+        lister->only_names.insert(cfg["names"].string_value());
     for (auto & item: cfg["names"].array_items())
-    {
         lister->only_names.insert(item.string_value());
-    }
     return [lister](cli_result_t & result)
     {
         lister->loop();

src/cmd/cli_pool_modify.cpp

@@ -39,16 +39,21 @@ struct pool_changer_t
             goto resume_1;
         else if (state == 2)
             goto resume_2;
-        pool_id = stoull_full(cfg["old_name"].string_value());
+        if (!cfg["pool"].is_null())
-        if (!pool_id)
         {
-            pool_name = cfg["old_name"].string_value();
+            pool_id = cfg["pool"].uint64_value();
-            if (pool_name == "")
+            pool_name = pool_id ? "" : cfg["pool"].string_value();
-            {
+        }
-                result = (cli_result_t){ .err = ENOENT, .text = "Pool ID or name is required to modify it" };
+        else
-                state = 100;
+        {
-                return;
+            pool_id = cfg["id"].uint64_value();
-            }
+            pool_name = pool_id ? "" : cfg["old_name"].string_value();
+        }
+        if (!pool_id && pool_name == "")
+        {
+            result = (cli_result_t){ .err = ENOENT, .text = "Pool ID or name is required to modify it" };
+            state = 100;
+            return;
         }
 resume_0:
         // Get pools from etcd
@@ -179,7 +184,7 @@ resume_2:
         result = (cli_result_t){
             .err = 0,
             .text = "Pool "+pool_name+" updated",
-            .data = new_pools,
+            .data = new_pools[std::to_string(pool_id)],
         };
         state = 100;
     }

src/cmd/cli_pool_rm.cpp

@@ -196,7 +196,9 @@ resume_3:
         result = (cli_result_t){
             .err = 0,
             .text = "Pool "+pool_name+" deleted",
-            .data = new_pools
+            .data = json11::Json::object{
+                {"ok", true},
+            },
         };
         state = 100;
     }
@@ -207,8 +209,16 @@ std::function<bool(cli_result_t &)> cli_tool_t::start_pool_rm(json11::Json cfg)
     auto pool_remover = new pool_remover_t();
     pool_remover->parent = this;
 
-    pool_remover->pool_id = cfg["pool"].uint64_value();
+    if (!cfg["pool"].is_null())
-    pool_remover->pool_name = pool_remover->pool_id ? "" : cfg["pool"].as_string();
+    {
+        pool_remover->pool_id = cfg["pool"].uint64_value();
+        pool_remover->pool_name = pool_remover->pool_id ? "" : cfg["pool"].as_string();
+    }
+    else
+    {
+        pool_remover->pool_id = cfg["pool_id"].uint64_value();
+        pool_remover->pool_name = pool_remover->pool_id ? "" : cfg["pool_name"].as_string();
+    }
 
     pool_remover->force = !cfg["force"].is_null();
 

src/cmd/cli_rm.cpp

@@ -269,10 +269,17 @@ resume_100:
             {
                 char buf[1024];
                 snprintf(buf, 1024, "Parent inode of layer %s (id 0x%jx) not found", cur->name.c_str(), cur->parent_id);
+                result = (cli_result_t){ .err = ENOENT, .text = buf };
                 state = 100;
                 return;
             }
             cur = &it->second;
+            if (!parent->check_image_perm(*cur, true))
+            {
+                result = (cli_result_t){ .err = EACCES, .text = "Image permission denied" };
+                state = 100;
+                return;
+            }
             chain_list.push_back(cur->num);
         }
         if (cur->num != from_cfg->num)
@@ -297,6 +304,12 @@ resume_100:
             auto it = sources.find(ic.second.parent_id);
             if (it != sources.end() && sources.find(ic.second.num) == sources.end())
             {
+                if (!parent->check_image_perm(ic.second, true))
+                {
+                    result = (cli_result_t){ .err = EACCES, .text = "Image permission denied" };
+                    state = 100;
+                    return;
+                }
                 merge_children.push_back(ic.second.num);
                 if (ic.second.readonly || writers_stopped)
                 {
@@ -451,6 +464,7 @@ resume_100:
         inode_config_t new_cfg = *child_cfg;
         new_cfg.deleted = false;
         new_cfg.num = target_cfg->num;
+        new_cfg.enc_key = target_cfg->enc_key;
         new_cfg.parent_id = new_parent;
         json11::Json::array cmp = json11::Json::array {
             json11::Json::object {

src/cmd/cli_rm_data.cpp

@@ -264,7 +264,7 @@ struct rm_inode_t
             {
                 // Error
                 result = (cli_result_t){
-                    .err = EIO,
+                    .err = EBUSY,
                     .text = "Failed: some blocks were not removed",
                     .data = data,
                 };

src/cmd/cli_rm_osd.cpp

@@ -181,6 +181,7 @@ struct rm_osd_t
             if (dry_run || is_dataloss && !force_dataloss || is_warning && !force_warning)
             {
                 result.err = is_dataloss && !force_dataloss || is_warning && !force_warning ? EBUSY : 0;
+                result.data = json11::Json::object{ { "pool_effects", pool_effects } };
                 state = 100;
                 return;
             }
@@ -332,6 +333,9 @@ struct rm_osd_t
         ids = (osd_ids.size() > 1 ? "OSDs " : "OSD ")+ids+(osd_ids.size() > 1 ? " are" : " is")+" removed from etcd";
         state = 100;
         result.text = (result.text != "" ? ids+"\n"+result.text : ids);
+        result.data = json11::Json::object {
+            {"deleted_osds", osd_ids},
+        };
         result.err = 0;
     }
 

src/cmd/cli_rm_wildcard.cpp

@@ -195,6 +195,8 @@ std::function<bool(cli_result_t &)> cli_tool_t::start_rm_wildcard(json11::Json c
     auto wildcard_remover = new wildcard_remover_t();
     wildcard_remover->parent = this;
     wildcard_remover->cfg = cfg;
+    if (cfg["globs"].is_string())
+        wildcard_remover->globs.push_back(cfg["globs"].string_value());
     for (auto & glob: cfg["globs"].array_items())
         wildcard_remover->globs.push_back(glob.string_value());
     wildcard_remover->exact = cfg["exact"].bool_value();

src/cmd/cli_serve.cpp

@@ -0,0 +1,491 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+#include <signal.h>
+#include <sys/socket.h>
+#include <netinet/tcp.h>
+#include <stdexcept>
+#include "cli.h"
+#include "cluster_client.h"
+#include "epoll_manager.h"
+#include "http_client.h"
+#include "str_util.h"
+#include "json_util.h"
+#include "addr_util.h"
+#include "openapi.json.h"
+
+struct cli_serve_conn_t
+{
+    int peer_fd = 0;
+    std::string peer_addr;
+    http_co_t *co = NULL;
+    cli_tool_t *p = NULL;
+    cli_result_t result;
+    bool keepalive = false;
+    bool closed = false;
+    timespec request_time;
+    std::string request_method;
+    std::string request_path;
+    std::string request_body;
+    std::string response_type;
+    std::function<bool(cli_result_t &)> action_cb;
+};
+
+struct cli_serve_path_t
+{
+    std::string cmd;
+    bool allow_get;
+    bool allow_client;
+};
+
+// Serve vitastor-cli commands over HTTP in JSON format
+struct cli_serve_t
+{
+    std::map<std::string, cli_serve_path_t> cmd_paths = {
+        {"data/delete",   {"rm-data",     false, false}},
+        {"data/describe", {"describe",    true,  false}},
+        {"data/fix",      {"fix",         false, false}},
+        {"data/merge",    {"merge-data",  false, false}},
+        {"image/create",  {"create",      false, true}},
+        {"image/delete",  {"rm",          false, true}},
+        {"image/flatten", {"flatten",     false, true}},
+        {"image/list",    {"ls",          true,  true}},
+        {"image/modify",  {"modify",      false, true}},
+        {"osd/alloc",     {"alloc-osd",   false, false}},
+        {"osd/delete",    {"rm-osd",      false, false}},
+        {"osd/list",      {"ls-osd",      true,  false}},
+        {"osd/modify",    {"modify-osd",  false, false}},
+        {"pg/list",       {"ls-pgs",      true,  false}},
+        {"pool/create",   {"create-pool", false, false}},
+        {"pool/delete",   {"rm-pool",     false, false}},
+        {"pool/list",     {"pools",       true,  false}},
+        {"pool/modify",   {"modify-pool", false, false}},
+        {"user/delete",   {"remove-user", false, false}},
+        {"user/list",     {"ls-user",     false, false}},
+        {"user/modify",   {"modify-user", false, false}},
+        {"status",        {"status",      true,  false}},
+    };
+
+    cli_tool_t *parent = NULL;
+    json11::Json options;
+    cli_result_t result;
+
+    bool log_body = false;
+    bool stop = false;
+    std::vector<std::string> bind_addresses;
+    int port = 0;
+    int listen_backlog = 0;
+    bool ssl = false;
+    std::vector<int> listen_fds;
+    http_context_t *http_ctx = NULL;
+    std::set<cli_serve_conn_t*> connections;
+
+    int state = 0;
+
+    bool is_done()
+    {
+        return state == 100;
+    }
+
+    void loop()
+    {
+        if (state == 1)
+            goto resume_1;
+        else if (state == 2)
+            goto resume_2;
+        else if (state == 100)
+            return;
+        if (options["bind_address"].is_string())
+            bind_addresses = explode(" ", options["bind_address"].string_value(), true);
+        else
+            bind_addresses.push_back("127.0.0.1");
+        port = options["port"].uint64_value();
+        if (!port)
+            port = 8080;
+        else if (port < 0 || port > 65535)
+        {
+            result = (cli_result_t){ .err = EINVAL, .text = "HTTP port can't be larger than 65536" };
+            state = 100;
+            return;
+        }
+        listen_backlog = options["listen_backlog"].uint64_value();
+        if (!listen_backlog)
+            listen_backlog = 128;
+        ssl = json_is_true(options["ssl"]);
+        if (ssl)
+        {
+            std::string ssl_cert = options["ssl_cert"].string_value();
+            std::string ssl_key = options["ssl_key"].string_value();
+            std::string ssl_ca = options["ssl_ca"].string_value();
+            std::string error;
+            http_ctx = http_context_init(parent->epmgr->tfd, ssl_cert, ssl_key, ssl_ca, ssl_ca != "", error);
+            if (error != "")
+            {
+                result = (cli_result_t){ .err = EINVAL, .text = error };
+                state = 100;
+                return;
+            }
+        }
+        for (auto & bind_address: bind_addresses)
+        {
+            int listen_fd = create_and_bind_socket(bind_address, port, listen_backlog, NULL);
+            fcntl(listen_fd, F_SETFL, fcntl(listen_fd, F_GETFL, 0) | O_NONBLOCK);
+            parent->epmgr->set_fd_handler(listen_fd, false, [this](int fd, int events)
+            {
+                accept_connections(fd);
+            });
+            listen_fds.push_back(listen_fd);
+        }
+    resume_1:
+        if (!stop)
+        {
+            state = 1;
+            return;
+        }
+        for (auto conn: connections)
+        {
+            http_close(conn->co);
+        }
+    resume_2:
+        // Wait for all connections to finish
+        if (connections.size() > 0)
+        {
+            state = 2;
+            return;
+        }
+        if (http_ctx)
+        {
+            http_context_destroy(http_ctx);
+            http_ctx = NULL;
+        }
+        state = 100;
+    }
+
+    void accept_connections(int listen_fd)
+    {
+        sockaddr_storage addr;
+        socklen_t peer_addr_size = sizeof(addr);
+        int peer_fd;
+        while ((peer_fd = accept(listen_fd, (sockaddr*)&addr, &peer_addr_size)) >= 0)
+        {
+            auto peer_addr_str = addr_to_string(addr);
+            assert(peer_fd != 0);
+            timespec ts;
+            clock_gettime(CLOCK_REALTIME, &ts);
+            printf("[%s.%03ju] New connection %d from %s\n", format_datetime(ts.tv_sec).c_str(), (uint64_t)ts.tv_nsec/1000000,
+                peer_fd, peer_addr_str.c_str());
+            fcntl(peer_fd, F_SETFL, fcntl(peer_fd, F_GETFL, 0) | O_NONBLOCK);
+            int one = 1;
+            setsockopt(peer_fd, SOL_TCP, TCP_NODELAY, &one, sizeof(one));
+            cli_serve_conn_t *conn = new cli_serve_conn_t;
+            conn->peer_fd = peer_fd;
+            conn->peer_addr = peer_addr_str;
+            conn->co = http_init(http_ctx);
+            http_serve(conn->co, peer_fd, (http_options_t){ .ssl = ssl }, [this, conn](http_message_t *msg)
+            {
+                process_request(conn, msg);
+            });
+            connections.insert(conn);
+            // Try to accept next connection
+            peer_addr_size = sizeof(addr);
+        }
+        if (peer_fd == -1 && errno != EAGAIN)
+        {
+            throw std::runtime_error(std::string("accept: ") + strerror(errno));
+        }
+    }
+
+    int map_to_http(int err, std::string *text)
+    {
+        int code = 0;
+        if (err == EINVAL)
+        {
+            code = 400;
+            if (text)
+                *text = "Bad Request";
+        }
+        else if (err == EACCES)
+        {
+            code = 403;
+            if (text)
+                *text = "Forbidden";
+        }
+        else if (err == EOPNOTSUPP)
+        {
+            code = 404;
+            if (text)
+                *text = "Not Found";
+        }
+        else if (err == ENOSYS)
+        {
+            code = 405;
+            if (text)
+                *text = "Method Not Allowed";
+        }
+        else if (err == EAGAIN)
+        {
+            code = 409;
+            if (text)
+                *text = "Update Conflict";
+        }
+        else if (err == ENOTEMPTY || err == EEXIST || err == ENOENT || err == EBUSY)
+        {
+            code = 412;
+            if (text)
+                *text = "Precondition Failed";
+        }
+        else /*if (err == EIO || err == EBADF)*/
+        {
+            code = 500;
+            if (text)
+                *text = "Internal Server Error";
+        }
+        return code;
+    }
+
+    std::string cli_http_response(cli_serve_conn_t *conn)
+    {
+        timespec now;
+        clock_gettime(CLOCK_REALTIME, &now);
+        int code = 200;
+        std::string response;
+        if (conn->result.err)
+        {
+            std::string status_line;
+            code = map_to_http(conn->result.err, &status_line);
+            response = "HTTP/1.1 "+std::to_string(code)+" "+status_line+"\r\n";
+        }
+        else
+        {
+            response = "HTTP/1.1 200 OK\r\n";
+        }
+        response += (conn->keepalive
+            ? "Connection: keep-alive\r\n"
+            : "Connection: close\r\n");
+        std::string body;
+        if (!conn->result.data.is_null())
+        {
+            response += "Content-Type: application/json\r\n";
+            body = conn->result.data.dump();
+        }
+        else
+        {
+            if (!conn->response_type.empty())
+                response += "Content-Type: "+conn->response_type+"\r\n";
+            else
+                response += "Content-Type: text/plain; charset=utf-8\r\n";
+            body = conn->result.text;
+        }
+        response += "Content-Length: "+std::to_string(body.size())+"\r\n\r\n";
+        response += body;
+        if (conn->request_method.find("\n") != std::string::npos)
+            conn->request_method = str_replace(conn->request_method, "\n", "%0a");
+        if (conn->request_method.find(" ") != std::string::npos)
+            conn->request_method = str_replace(conn->request_method, " ", "%20");
+        if (conn->request_path.find("\n") != std::string::npos)
+            conn->request_path = str_replace(conn->request_path, "\n", "%0a");
+        if (conn->request_path.find(" ") != std::string::npos)
+            conn->request_path = str_replace(conn->request_path, " ", "%20");
+        uint64_t response_time = (now.tv_sec-conn->request_time.tv_sec)*1000 + (now.tv_nsec-conn->request_time.tv_nsec)/1000000;
+        printf("[%s.%03ju] %s %s %s %d %.03f sec\n",
+            format_datetime(now.tv_sec).c_str(), (uint64_t)now.tv_nsec/1000000,
+            conn->peer_addr.c_str(), conn->request_method.c_str(), conn->request_path.c_str(), code,
+            response_time/1000.0);
+        if (log_body)
+        {
+            if (conn->request_body.find("\n") != std::string::npos)
+                conn->request_body = str_replace(conn->request_body, "\n", " ");
+            if (conn->request_body.size())
+                printf("  %s\n", conn->request_body.c_str());
+            printf("  %s\n", body.c_str());
+        }
+        return response;
+    }
+
+    void process_request(cli_serve_conn_t *conn, http_message_t *msg)
+    {
+        timespec ts;
+        clock_gettime(CLOCK_REALTIME, &ts);
+        if (!msg->error.empty())
+        {
+            // connection is closed
+            fprintf(stderr, "[%s.%03ju] Connection %d closed: %s\n", format_datetime(ts.tv_sec).c_str(),
+                (uint64_t)ts.tv_nsec/1000000, conn->peer_fd, msg->error.c_str());
+            if (conn->p)
+                conn->closed = true;
+            else
+            {
+                connections.erase(conn);
+                http_destroy(conn->co);
+                delete conn;
+            }
+            return;
+        }
+        conn->keepalive = msg->headers.find("connection") != msg->headers.end() &&
+            msg->headers.at("connection") == "keep-alive";
+        conn->p = new cli_tool_t;
+        conn->p->iodepth = parent->iodepth;
+        conn->p->parallel_osds = parent->parallel_osds;
+        conn->p->json_output = true;
+        conn->p->ringloop = parent->ringloop;
+        conn->p->epmgr = parent->epmgr;
+        conn->p->cli = parent->cli;
+        conn->p->is_command_line = false;
+        // Parse request
+        auto req_line = explode(" ", msg->status_line, true);
+        if (req_line.size() < 2)
+        {
+            if (req_line[0] == "")
+                req_line[0] = "-";
+            req_line.push_back("-");
+        }
+        conn->request_time = ts;
+        conn->request_method = std::move(req_line[0]);
+        conn->request_path = std::move(req_line[1]);
+        conn->request_body = std::move(msg->body);
+        conn->response_type = "";
+        if (parent->cli->st_cli.use_auth)
+        {
+            auto user = std::make_unique<cli_user_t>();
+            user->name = msg->headers["_tls_common_name"];
+            auto user_it = parent->cli->st_cli.user_info.find(user->name);
+            auto userinfo = user_it == parent->cli->st_cli.user_info.end() ? user_it->second : json11::Json();
+            user->type = user->name == "root" ? "admin" : userinfo["type"].string_value();
+            for (auto & gr: userinfo["groups"].array_items())
+            {
+                user->groups.insert(gr.string_value());
+            }
+        }
+        auto ctype = msg->headers["content-type"];
+        if (conn->request_method != "GET" && conn->request_method != "POST")
+        {
+            conn->result = { .err = ENOSYS, .text = "Unsupported request method "+conn->request_method };
+        }
+        else if (ctype != (conn->request_method == "GET" ? "" : "application/json"))
+        {
+            conn->result = { .err = EINVAL, .text = "Unsupported Content-Type: "+ctype+" for "+conn->request_method+" requests" };
+        }
+        else
+        {
+            auto uri = explode("?", conn->request_path, true);
+            uri[0] = trim(uri[0], "/");
+            auto cmd_it = cmd_paths.find(uri[0]);
+            if (uri[0] == "")
+            {
+                std::string text = "Supported APIs:\n\n- GET /openapi\n";
+                for (auto & pp: cmd_paths)
+                {
+                    text += (pp.second.allow_get ? "- GET" : "- POST") + (" /" + pp.first) + "\n";
+                }
+                conn->result = { .text = text };
+            }
+            else if (uri[0] == "openapi")
+            {
+                conn->response_type = "application/json";
+                conn->result = { .text = openapi_description };
+                if (parent->cli->st_cli.use_auth)
+                {
+                    // Filter available paths by privileges
+                    if (conn->p->user->type == "client")
+                    {
+                        std::string error;
+                        auto openapi = json11::Json::parse(openapi_description, error).object_items();
+                        json11::Json::object paths;
+                        for (auto & kv: openapi["paths"].object_items())
+                        {
+                            auto cmd_it = cmd_paths.find(kv.first.substr(1));
+                            if (cmd_it != cmd_paths.end() && cmd_it->second.allow_client)
+                            {
+                                paths[kv.first] = kv.second;
+                            }
+                        }
+                        openapi["paths"] = paths;
+                        conn->response_type = "application/json";
+                        conn->result = { .text = json11::Json(openapi).dump() };
+                    }
+                    else if (conn->p->user->type != "admin")
+                    {
+                        conn->response_type = "";
+                        conn->result = { .err = EACCES, .text = "Access denied" };
+                    }
+                }
+            }
+            else if (cmd_it == cmd_paths.end())
+            {
+                conn->result = { .err = EOPNOTSUPP, .text = "unknown command: "+uri[0] };
+            }
+            else if (conn->request_method == "GET" && !cmd_it->second.allow_get)
+            {
+                conn->result = { .err = ENOSYS, .text = "method /"+uri[0]+" only allows POST requests" };
+            }
+            else if (parent->cli->st_cli.use_auth && conn->p->user->type == "client" && !cmd_it->second.allow_client)
+            {
+                conn->result = { .err = EACCES, .text = "Access denied" };
+            }
+            else
+            {
+                std::string error;
+                json11::Json::object cfg;
+                if (conn->request_method == "POST")
+                {
+                    cfg = json11::Json::parse(conn->request_body, error).object_items();
+                }
+                else
+                {
+                    // Parse URI
+                    cfg = parse_uri_params(uri[1]);
+                }
+                if (error != "")
+                {
+                    conn->result = { .err = EINVAL, .text = "Invalid JSON in body: "+error };
+                }
+                else
+                {
+                    cfg["command"] = json11::Json::array{cmd_it->second.cmd};
+                    conn->p->parse_api_opts(cfg);
+                    conn->action_cb = conn->p->start(cfg, conn->result);
+                }
+            }
+        }
+        if (!conn->action_cb)
+        {
+            http_reply(conn->co, cli_http_response(conn));
+            delete conn->p;
+            conn->p = NULL;
+            return;
+        }
+        conn->p->loop_and_wait(conn->action_cb, [this, conn](const cli_result_t & r)
+        {
+            conn->result = r;
+            conn->action_cb = NULL;
+            delete conn->p;
+            conn->p = NULL;
+            if (!conn->closed)
+                http_reply(conn->co, cli_http_response(conn));
+            else
+            {
+                connections.erase(conn);
+                http_destroy(conn->co);
+                delete conn;
+            }
+        });
+    }
+};
+
+std::function<bool(cli_result_t &)> cli_tool_t::start_serve(json11::Json cfg)
+{
+    auto server = new cli_serve_t();
+    server->parent = this;
+    server->options = cfg;
+    return [server](cli_result_t & result)
+    {
+        server->loop();
+        if (server->is_done())
+        {
+            result = server->result;
+            delete server;
+            return true;
+        }
+        return false;
+    };
+}

src/cmd/cli_status.cpp

@@ -18,6 +18,7 @@ struct status_printer_t
 {
     cli_tool_t *parent;
 
+    cli_result_t result;
     int state = 0;
     json11::Json::array mon_members;
     json11::Json agg_stats;
@@ -89,7 +90,7 @@ resume_2:
             return;
         if (parent->etcd_err.err)
         {
-            fprintf(stderr, "%s\n", parent->etcd_err.text.c_str());
+            result = parent->etcd_err;
             state = 100;
             return;
         }
@@ -107,7 +108,8 @@ resume_2:
             if (etcd_states[i]["error"].is_null())
             {
                 etcd_alive++;
-                etcd_db_size = etcd_states[i]["dbSize"].uint64_value();
+                uint64_t db_size = etcd_states[i]["dbSize"].uint64_value();
+                etcd_db_size = db_size > etcd_db_size ? db_size : etcd_db_size;
             }
         }
         int mon_count = 0;
@@ -265,7 +267,7 @@ resume_2:
                     obj_n = agg_stats["object_counts"][str].uint64_value() * parent->cli->st_cli.global_block_size;
                 json_status[str+"_data"] = obj_n;
             }
-            printf("%s\n", json11::Json(json_status).dump().c_str());
+            result.data = json11::Json(json_status);
             state = 100;
             return;
         }
@@ -406,7 +408,7 @@ std::function<bool(cli_result_t &)> cli_tool_t::start_status(json11::Json cfg)
         printer->loop();
         if (printer->is_done())
         {
-            result = { .err = 0 };
+            result = printer->result;
             delete printer;
             return true;
         }

src/cmd/cli_user_ls.cpp

@@ -0,0 +1,128 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 (see README.md for details)
+
+#include <algorithm>
+#include "cli.h"
+#include "cluster_client.h"
+#include "str_util.h"
+#include "json_util.h"
+
+// List users
+struct user_lister_t
+{
+    cli_tool_t *parent;
+    std::vector<std::string> only_names;
+
+    int state = 0;
+    cli_result_t result;
+    json11::Json::array users;
+
+    bool is_done()
+    {
+        return state == 100;
+    }
+
+    void loop()
+    {
+        if (state == 1)
+            goto resume_1;
+        if (state == 100)
+            return;
+        {
+            json11::Json::array select;
+            if (!only_names.size())
+            {
+                select.push_back(json11::Json::object {
+                    { "request_range", json11::Json::object {
+                        { "key", base64_encode(
+                            parent->cli->st_cli.etcd_prefix+"/config/user/"
+                        ) },
+                        { "range_end", base64_encode(
+                            parent->cli->st_cli.etcd_prefix+"/config/user0"
+                        ) },
+                    } },
+                });
+            }
+            else
+            {
+                for (auto & name: only_names)
+                {
+                    select.push_back(json11::Json::object {
+                        { "request_range", json11::Json::object {
+                            { "key", base64_encode(parent->cli->st_cli.etcd_prefix+"/config/user/"+name) },
+                        } }
+                    });
+                }
+            }
+            parent->etcd_txn(json11::Json::object {
+                { "success", select },
+            });
+        }
+        state = 1;
+resume_1:
+        if (parent->waiting > 0)
+            return;
+        if (parent->etcd_err.err)
+        {
+            result = parent->etcd_err;
+            state = 100;
+            return;
+        }
+        for (auto & response: parent->etcd_result["responses"].array_items())
+        {
+            for (auto & kv_item: response["response_range"]["kvs"].array_items())
+            {
+                auto kv = parent->cli->st_cli.parse_etcd_kv(kv_item);
+                auto user = kv.value.object_items();
+                user["name"] = kv.key.substr(parent->cli->st_cli.etcd_prefix.size()+13);
+                if (!parent->json_output)
+                    user["groups_fmt"] = implode(",", user["groups"].array_items());
+                users.push_back(std::move(user));
+            }
+        }
+        if (parent->json_output)
+        {
+            // JSON output
+            result.data = users;
+            state = 100;
+            return;
+        }
+        // Table output: name, type, groups
+        json11::Json::array cols;
+        cols.push_back(json11::Json::object{
+            { "key", "name" },
+            { "title", "NAME" },
+        });
+        cols.push_back(json11::Json::object{
+            { "key", "type" },
+            { "title", "TYPE" },
+        });
+        cols.push_back(json11::Json::object{
+            { "key", "groups_fmt" },
+            { "title", "GROUPS" },
+        });
+        result.text = print_table(users, cols, parent->color);
+        state = 100;
+    }
+};
+
+std::function<bool(cli_result_t &)> cli_tool_t::start_user_ls(json11::Json cfg)
+{
+    auto lister = new user_lister_t();
+    lister->parent = this;
+    if (cfg["names"].is_string())
+        lister->only_names.push_back(cfg["names"].string_value());
+    for (auto & item: cfg["names"].array_items())
+        lister->only_names.push_back(item.string_value());
+    return [lister](cli_result_t & result)
+    {
+        lister->loop();
+        if (lister->is_done())
+        {
+            result = lister->result;
+            delete lister;
+            return true;
+        }
+        return false;
+    };
+}

src/disk_tool/CMakeLists.txt

@@ -7,7 +7,7 @@ add_executable(vitastor-disk
 	disk_tool.cpp disk_simple_offsets.cpp
 	disk_tool_discard.cpp disk_tool_journal.cpp disk_tool_meta.cpp disk_tool_prepare.cpp disk_tool_resize.cpp
 	disk_tool_resize_auto.cpp disk_tool_udev.cpp disk_tool_utils.cpp disk_tool_upgrade.cpp
-	../util/crc32c.c ../util/str_util.cpp ../util/json_util.cpp ../../json11/json11.cpp ../util/rw_blocking.cpp ../util/allocator.cpp ../util/ringloop.cpp
+	../util/crc32c.c ../util/xxh_x86dispatch.c ../util/str_util.cpp ../util/json_util.cpp ../../json11/json11.cpp ../util/rw_blocking.cpp ../util/allocator.cpp ../util/ringloop.cpp
 	../blockstore/blockstore_disk.cpp ../blockstore/blockstore_heap.cpp ../blockstore/multilist.cpp
 )
 target_link_libraries(vitastor-disk

src/disk_tool/disk_tool.h

@@ -136,8 +136,8 @@ struct disk_tool_t
     void choose_journal_block(uint32_t je_size);
     int resize_rebuild_journal();
     int resize_write_new_journal();
-    void remap_big_write(heap_entry_t *wr);
+    void remap_big_write(blockstore_heap_t *heap, heap_entry_t *wr);
-    void remap_small_write(heap_entry_t *wr);
+    void remap_small_write(blockstore_heap_t *heap, heap_entry_t *wr);
     void fill_old_clean_entry(blockstore_heap_t *heap, heap_entry_t *big_wr);
     void fill_old_journal_entry(blockstore_heap_t *heap, heap_entry_t *wr);
     int resize_rebuild_meta();
@@ -175,7 +175,6 @@ struct disk_tool_t
 void disk_tool_simple_offsets(json11::Json cfg, bool json_output);
 
 uint64_t sscanf_json(const char *fmt, const json11::Json & str);
-void fromhexstr(const std::string & from, int bytes, uint8_t *to);
 int disable_cache(const std::string & dev);
 uint64_t get_atomic_write_size(const std::string & dev);
 uint64_t get_device_size(const std::string & dev, bool should_exist = false);

src/disk_tool/disk_tool_journal.cpp

@@ -4,6 +4,7 @@
 #include <unistd.h>
 
 #include "disk_tool.h"
+#include "str_util.h"
 #include "malloc_or_die.h"
 
 int disk_tool_t::dump_journal()

src/disk_tool/disk_tool_meta.cpp

@@ -4,6 +4,7 @@
 #include "disk_tool.h"
 #include "rw_blocking.h"
 #include "json_util.h"
+#include "str_util.h"
 #include "malloc_or_die.h"
 
 int disk_tool_t::process_meta(std::function<void(blockstore_meta_header_v3_t *)> hdr_fn,
@@ -751,7 +752,7 @@ close_err0:
                 {
                     *wr->get_checksum(&heap) = sscanf_json("%jx", write_entry["data_crc32c"]);
                 }
-                wr->crc32c = wr->calc_crc32c();
+                wr->checksum = wr->calc_checksum(&heap);
                 assert((uint8_t*)wr + wr->size == new_meta_buf + meta_offset + used_space);
             }
         }
@@ -794,7 +795,7 @@ close_err:
             fromhexstr(meta_entry["ext_bitmap"].string_value(), new_clean_entry_bitmap_size, wr->get_ext_bitmap(&heap));
             if (new_meta_hdr->data_csum_type != 0)
                 fromhexstr(meta_entry["data_csum"].string_value(), new_data_csum_size, wr->get_checksums(&heap));
-            wr->crc32c = wr->calc_crc32c();
+            wr->checksum = wr->calc_checksum(&heap);
             assert((uint8_t*)wr + wr->size == new_meta_buf + meta_offset + used_space);
             auto j_it = journal_by_object.find(oid);
             if (j_it != journal_by_object.end())
@@ -860,7 +861,7 @@ close_err:
                         assert(0);
                     }
                     wr->size = wr->get_size(&heap);
-                    wr->crc32c = wr->calc_crc32c();
+                    wr->checksum = wr->calc_checksum(&heap);
                     assert((uint8_t*)wr + wr->size == new_meta_buf + meta_offset + used_space);
                 }
             }

src/disk_tool/disk_tool_resize.cpp

@@ -531,7 +531,7 @@ int disk_tool_t::resize_write_new_journal()
     return 0;
 }
 
-void disk_tool_t::remap_big_write(heap_entry_t *wr)
+void disk_tool_t::remap_big_write(blockstore_heap_t *heap, heap_entry_t *wr)
 {
     uint64_t block_num = wr->big().block_num;
     auto remap_it = data_remap.find(block_num);
@@ -544,10 +544,10 @@ void disk_tool_t::remap_big_write(heap_entry_t *wr)
     }
     block_num += data_idx_diff;
     wr->big().block_num = block_num;
-    wr->crc32c = wr->calc_crc32c();
+    wr->checksum = wr->calc_checksum(heap);
 }
 
-void disk_tool_t::remap_small_write(heap_entry_t *wr)
+void disk_tool_t::remap_small_write(blockstore_heap_t *heap, heap_entry_t *wr)
 {
     if (new_meta_format == BLOCKSTORE_META_FORMAT_HEAP && wr->small().len > 0)
     {
@@ -559,7 +559,7 @@ void disk_tool_t::remap_small_write(heap_entry_t *wr)
         memcpy(new_journal_ptr, buffer_area+wr->small().location, wr->small().len);
         wr->small().location = new_journal_ptr-new_journal_buf;
         new_journal_ptr += wr->small().len;
-        wr->crc32c = wr->calc_crc32c();
+        wr->checksum = wr->calc_checksum(heap);
     }
 }
 
@@ -673,11 +673,11 @@ int disk_tool_t::resize_rebuild_meta()
             {
                 if (wr->type() == BS_HEAP_BIG_WRITE || wr->type() == BS_HEAP_BIG_INTENT)
                 {
-                    remap_big_write(wr);
+                    remap_big_write(heap, wr);
                 }
                 else if (wr->type() == BS_HEAP_SMALL_WRITE)
                 {
-                    remap_small_write(wr);
+                    remap_small_write(heap, wr);
                 }
                 else if (wr->type() != BS_HEAP_DELETE && new_meta_format != BLOCKSTORE_META_FORMAT_HEAP)
                 {
@@ -786,7 +786,7 @@ int disk_tool_t::resize_rebuild_meta()
                     memcpy(((uint8_t*)wr) + sizeof(heap_big_write_t) + new_clean_entry_bitmap_size, bitmap+new_clean_entry_bitmap_size, new_clean_entry_bitmap_size);
                     memcpy(((uint8_t*)wr) + sizeof(heap_big_write_t) + 2*new_clean_entry_bitmap_size, bitmap+2*new_clean_entry_bitmap_size, new_data_csum_size);
                 }
-                wr->crc32c = wr->calc_crc32c();
+                wr->checksum = wr->calc_checksum(&dsk);
                 new_meta_pos += wr->size;
             }
             else

src/disk_tool/disk_tool_utils.cpp

@@ -22,28 +22,6 @@ uint64_t sscanf_json(const char *fmt, const json11::Json & str)
     return value;
 }
 
-static int fromhex(char c)
-{
-    if (c >= '0' && c <= '9')
-        return (c-'0');
-    else if (c >= 'a' && c <= 'f')
-        return (c-'a'+10);
-    else if (c >= 'A' && c <= 'F')
-        return (c-'A'+10);
-    return -1;
-}
-
-void fromhexstr(const std::string & from, int bytes, uint8_t *to)
-{
-    for (int i = 0; i < from.size() && i < bytes; i++)
-    {
-        int x = fromhex(from[2*i]), y = fromhex(from[2*i+1]);
-        if (x < 0 || y < 0)
-            break;
-        to[i] = x*16 + y;
-    }
-}
-
 // returns 1 = check error, 0 = write through, -1 = write back
 // (similar to 1 = warning, -1 = error, 0 = success in disable_cache)
 static int check_queue_cache(std::string dev, std::string parent_dev)

src/osd/CMakeLists.txt

@@ -15,6 +15,7 @@ target_link_libraries(vitastor-osd
 	${ISAL_LIBRARIES}
 	${IBVERBS_LIBRARIES}
 	${RDMACM_LIBRARIES}
+	${OPENSSL_LIBRARIES}
 )
 
 # osd_rmw_test

src/osd/osd.cpp

@@ -162,6 +162,7 @@ void osd_t::parse_config(bool init)
         if (!osd_num)
             throw std::runtime_error("osd_num is required in the configuration");
         msgr.osd_num = osd_num;
+        st_cli.osd_num = osd_num;
     }
     if (bs)
     {