WIP Support TLS certificate generation in make-etcd

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 +702,42 @@ jobs:
           echo ""
         done
 
+  test_snapshot_chain_encrypted:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: ENCRYPTED=1 /root/vitastor/tests/test_snapshot_chain.sh
+    - name: Print logs
+      if: always() && steps.test.outcome == 'failure'
+      run: |
+        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
+          echo "-------- $i --------"
+          cat $i
+          echo ""
+        done
+
+  test_snapshot_chain_enc_gcm:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: TEST_NAME=enc_gcm ENCRYPTED=1 VITASTOR_CFG=',"proto_checksums":"gcm"' /root/vitastor/tests/test_snapshot_chain.sh
+    - name: Print logs
+      if: always() && steps.test.outcome == 'failure'
+      run: |
+        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
+          echo "-------- $i --------"
+          cat $i
+          echo ""
+        done
+
   test_old_snapshot_chain:
     runs-on: ubuntu-latest
     needs: build
@@ -1206,6 +1296,24 @@ jobs:
           echo ""
         done
 
+  test_heal_ec_rdma:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 10
+      run: TEST_NAME=ec_rdma RDMA=1 SCHEME=ec /root/vitastor/tests/test_heal.sh
+    - name: Print logs
+      if: always() && steps.test.outcome == 'failure'
+      run: |
+        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
+          echo "-------- $i --------"
+          cat $i
+          echo ""
+        done
+
   test_checksum:
     runs-on: ubuntu-latest
     needs: build
@@ -1224,6 +1332,24 @@ 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 +2178,39 @@ jobs:
           echo ""
         done
 
+  test_write_encrypted:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: /root/vitastor/tests/test_write_encrypted.sh
+    - name: Print logs
+      if: always() && steps.test.outcome == 'failure'
+      run: |
+        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
+          echo "-------- $i --------"
+          cat $i
+          echo ""
+        done
+
+  test_write_encrypted_ec:
+    runs-on: ubuntu-latest
+    needs: build
+    container: ${{env.TEST_IMAGE}}:${{github.sha}}
+    steps:
+    - name: Run test
+      id: test
+      timeout-minutes: 3
+      run: SCHEME=ec /root/vitastor/tests/test_write_encrypted.sh
+    - name: Print logs
+      if: always() && steps.test.outcome == 'failure'
+      run: |
+        for i in /root/vitastor/testdata/*.log /root/vitastor/testdata/*.txt; do
+          echo "-------- $i --------"
+          cat $i
+          echo ""
+        done
+

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

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

.gitignore

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

CMakeLists.txt

@@ -2,7 +2,7 @@ cmake_minimum_required(VERSION 2.8...3.30)
 
 project(vitastor)
 
-set(VITASTOR_VERSION "3.0.9")
+set(VITASTOR_VERSION "3.0.12")
 
 include(CTest)
 

README-ru.md

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

README.md

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

csi/Makefile

@@ -1,4 +1,4 @@
-VITASTOR_VERSION ?= v3.0.9
+VITASTOR_VERSION ?= v3.0.12
 
 all: build push
 

csi/deploy/004-csi-nodeplugin.yaml

@@ -49,7 +49,7 @@ spec:
             capabilities:
               add: ["SYS_ADMIN"]
             allowPrivilegeEscalation: true
-          image: vitalif/vitastor-csi:v3.0.9
+          image: vitalif/vitastor-csi:v3.0.12
           args:
             - "--node=$(NODE_ID)"
             - "--endpoint=$(CSI_ENDPOINT)"

csi/deploy/007-csi-provisioner.yaml

@@ -121,7 +121,7 @@ spec:
             privileged: true
             capabilities:
               add: ["SYS_ADMIN"]
-          image: vitalif/vitastor-csi:v3.0.9
+          image: vitalif/vitastor-csi:v3.0.12
           args:
             - "--node=$(NODE_ID)"
             - "--endpoint=$(CSI_ENDPOINT)"

csi/src/config.go

@@ -5,7 +5,7 @@ package vitastor
 
 const (
     vitastorCSIDriverName    = "csi.vitastor.io"
-    vitastorCSIDriverVersion = "3.0.9"
+    vitastorCSIDriverVersion = "3.0.12"
 )
 
 // Config struct fills the parameters of request or user input

debian/changelog

@@ -1,4 +1,4 @@
-vitastor (3.0.9-1) unstable; urgency=medium
+vitastor (3.0.12-1) unstable; urgency=medium
 
   * Bugfixes
 

debian/control

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

debian/vitastor-buildenv.Dockerfile

@@ -12,20 +12,18 @@ ARG REL=
 WORKDIR /root
 
 RUN set -e -x; \
-    if [ "$REL" = "buster" ]; then \
-        perl -i -pe 's/deb.debian.org/archive.debian.org/' /etc/apt/sources.list; \
-        apt-get update; \
-        apt-get -y install wget; \
-        wget https://vitastor.io/debian/pubkey.gpg -O /etc/apt/trusted.gpg.d/vitastor.gpg; \
-        echo "deb https://vitastor.io/debian $REL main" >> /etc/apt/sources.list; \
-    fi; \
+    perl -i -pe 's/deb.debian.org/archive.debian.org/' /etc/apt/sources.list; \
+    apt-get update; \
+    apt-get -y install wget; \
+    wget https://vitastor.io/debian/pubkey.gpg -O /etc/apt/trusted.gpg.d/vitastor.gpg; \
+    echo "deb https://vitastor.io/debian $REL main" >> /etc/apt/sources.list; \
     grep '^deb ' /etc/apt/sources.list | perl -pe 's/^deb/deb-src/' >> /etc/apt/sources.list; \
     perl -i -pe 's/Types: deb$/Types: deb deb-src/' /etc/apt/sources.list.d/*.sources || true; \
     echo 'APT::Install-Recommends false;' >> /etc/apt/apt.conf; \
     echo 'APT::Install-Suggests false;' >> /etc/apt/apt.conf
 
 RUN apt-get update && \
-    apt-get -y install fio libgoogle-perftools-dev devscripts libjerasure-dev cmake \
+    apt-get -y install fio libgoogle-perftools-dev devscripts libjerasure-dev cmake libc-ares-dev libisal-crypto-dev \
         libibverbs-dev librdmacm-dev libisal-dev libnl-3-dev libnl-genl-3-dev curl nodejs npm node-nan node-bindings && \
     apt-get -y build-dep fio && \
     apt-get --download-only source fio

docker/Makefile

@@ -1,4 +1,4 @@
-VITASTOR_VERSION ?= v3.0.9
+VITASTOR_VERSION ?= v3.0.12
 
 all: build push
 

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

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

docker/etc/vitastor/docker.conf

@@ -4,7 +4,7 @@
 #
 
 # Desired Vitastor version
-VITASTOR_VERSION=v3.0.9
+VITASTOR_VERSION=v3.0.12
 
 # Additional arguments for all containers
 # For example, you may want to specify a custom logging driver here

docker/etc/vitastor/etcd.conf

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

docs/config.en.md

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

docs/config.ru.md

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

docs/config/layout-osd.en.md

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

docs/config/layout-osd.ru.md

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

docs/config/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_cipher_pool_size](#max_cipher_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_cipher_pool_size
+
+- Type: integer
+- Default: 256
+
+Maximum number of OpenSSL cipher contexts cached in OSD memory, counted separately
+for each cipher and for encryption/decryption. Probably doesn't require modification.

docs/config/security.ru.md

@@ -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_cipher_pool_size](#max_cipher_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_cipher_pool_size
+
+- Тип: целое число
+- Значение по умолчанию: 256
+
+Максимальное количество кэшируемых в памяти OSD контекстов шифра OpenSSL, учитываемое
+отдельно для каждого шифра и для шифрования и расшифровки. Вряд ли требует изменения.

docs/config/src/included.en.md

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

docs/config/src/included.ru.md

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

docs/config/src/layout-osd.yml

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

docs/config/src/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_cipher_pool_size
+  type: int
+  default: 256
+  info: |
+    Maximum number of OpenSSL cipher contexts cached in OSD memory, counted separately
+    for each cipher and for encryption/decryption. Probably doesn't require modification.
+  info_ru: |
+    Максимальное количество кэшируемых в памяти OSD контекстов шифра OpenSSL, учитываемое
+    отдельно для каждого шифра и для шифрования и расшифровки. Вряд ли требует изменения.

docs/installation/docker.en.md

@@ -26,9 +26,9 @@ at Vitastor Kubernetes operator: https://github.com/Antilles7227/vitastor-operat
 The instruction is very simple.
 
 1. Download a Docker image of the desired version: \
-   `docker pull vitalif/vitastor:v3.0.9`
+   `docker pull vitalif/vitastor:v3.0.12`
 2. Install scripts to the host system: \
-   `docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.0.9 install.sh`
+   `docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.0.12 install.sh`
 3. Reload udev rules: \
    `udevadm control --reload-rules`
 4. Enable the vitastor-host service: \

docs/installation/docker.ru.md

@@ -25,9 +25,9 @@ Vitastor можно установить в Docker/Podman. При этом etcd,
 Инструкция по установке максимально простая.
 
 1. Скачайте Docker-образ желаемой версии: \
-   `docker pull vitalif/vitastor:v3.0.9`
+   `docker pull vitalif/vitastor:v3.0.12`
 2. Установите скрипты в хост-систему командой: \
-   `docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.0.9 install.sh`
+   `docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.0.12 install.sh`
 3. Перезагрузите правила udev: \
    `udevadm control --reload-rules`
 4. Включите сервис vitastor-host: \

docs/installation/source.en.md

@@ -15,9 +15,8 @@
 - gcc and g++ 8 or newer, clang 10 or newer, or other compiler with C++11 plus
   designated initializers support from C++20
 - CMake
-- jerasure headers and libraries
-- ISA-L, libibverbs and librdmacm headers and libraries (optional)
-- tcmalloc (google-perftools-dev)
+- jerasure, c-ares headers and libraries
+- ISA-L, libibverbs, librdmacm, libnl3 headers and libraries (optional)
 
 ## Basic instructions
 

docs/installation/source.ru.md

@@ -15,9 +15,8 @@
 - gcc и g++ >= 8, либо clang >= 10, либо другой компилятор с поддержкой C++11 плюс
   назначенных инициализаторов (designated initializers) из C++20
 - CMake
-- Заголовки и библиотеки jerasure
-- Опционально - заголовки и библиотеки ISA-L, libibverbs, librdmacm
-- tcmalloc (google-perftools-dev)
+- Заголовки и библиотеки jerasure, c-ares
+- Опционально - заголовки и библиотеки ISA-L, libibverbs, librdmacm, libnl3
 
 ## Базовая инструкция
 

docs/intro/features.en.md

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

docs/intro/features.ru.md

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

docs/intro/security.ru.md

@@ -0,0 +1,439 @@
+[Документация](../../README-ru.md#документация) → Безопасность
+
+-----
+
+[Read in English](security.en.md)
+
+# Оглавление
+
+⚠️ Предупреждение: детальное описание настроек безопасности достаточно длинное.
+
+Если не боитесь - читайте [Подробное описание](#подробное-описание).
+
+Если хотите просто быстро настроить Vitastor с шифрованием - читайте начало статьи.
+
+- [Быстрая настройка](#быстрая-настройка)
+- 
+
+# Быстрая настройка
+
+
+
+# Пользовательские сценарии
+
+Зачем всё это нужно вам?
+
+
+
+# Подробное описание
+
+Начиная с версии 3.1.0, в Vitastor есть следующие функции:
+1. Шифрование соединений с etcd (TLS)
+2. Шифрование соединений с OSD (AES-GCM) - по выбору либо только заголовков, либо и заголовков, и данных
+3. Сквозное шифрование данных образов (AES-XTS)
+4. Хранения ключей шифрования AES-XTS во внешнем Vault
+5. Контрольных сумм данных на транспортном уровне с секретной "солью"
+6. Аутентификация с помощью TLS (X.509) сертификатов и закрытых ключей
+7. Разграничение прав доступа клиентов к данным etcd
+8. Разграничение прав доступа клиентов к данным самих образов (на стороне OSD)
+
+По умолчанию шифрование, аутентификация и авторизация отключены, но, начиная с 3.1.0,
+используются контрольные суммы данных на транспортном уровне (`proto_checksums=payload`).
+
+## Шифрование соединений с etcd (TLS)
+
+Варианты настройки:
+- Без шифрования (http)
+- С шифрованием (https)
+- С клиентским сертификатом, но при выключенной авторизации (`use_auth=false`) - используется
+  отдельный сертификат и ключ: `etcd_client_cert`, `etcd_client_key`
+- С клиентским сертификатом, при включённой аутентификации на уровне OSD - используется общий
+  сертификат и ключ: для OSD - `osd_cert` и `osd_pkey`, для клиентов - `cert` и `pkey`
+
+## Шифрование соединений с OSD (AES-GCM)
+
+Варианты настройки:
+- Без шифрования и без контрольных сумм: `proto_checksums=none`.
+- Без шифрования, с контрольными суммами данных: `proto_checksums=payload` (можно не указывать,
+  т.к. это значение по умолчанию). При этом контрольные суммы можно отключить на стороне
+  клиента либо использовать более старые версии клиента, не поддерживающие контрольные суммы.
+  Если нужно запретить подключение клиентов без контрольных сумм, можно использовать опцию
+  `force_proto_checksums=payload`.
+- С шифрованием заголовков и контрольными суммами данных: активируется при установленных опциях
+  `cert`, `pkey`, `osd_ca` на стороне клиента и `osd_cert`, `osd_pkey`, `osd_ca`, `client_ca`
+  на стороне OSD, при `proto_checksums=payload`. При этом по умолчанию запрещается
+  отключение контрольных сумм на уровне клиента, то есть используется `force_proto_checksums=payload`.
+- С полным шифрованием всего трафика: аналогично прошлому варианту, но с `proto_checksums=gcm`.
+  Клиенту при этом по умолчанию разрешается понизить уровень защиты до контрольных сумм, но
+  это тоже можно запретить через `force_proto_checksums=gcm`. Данный вариант не является рекомендуемым,
+  так как добавлен в первую очередь для возможной поддержки небезопасных (публичных) сетей и
+  больше всего снижает производительность. В частности, если одновременно использовать полное
+  шифрование трафика и сквозное шифрование образов AES-XTS, то данные будут шифроваться дважды.
+
+Для шифрования используется алгоритм AES-256-GCM и собственный упрощённый протокол согласования
+ключей, полностью аналогичный TLS 1.3 ECDHE.
+
+## Сквозное шифрование данных образов (AES-XTS)
+
+Клиент Vitastor поддерживает шифрование данных каждого образа своим ключом. В этом случае на OSD
+уходят уже зашифрованные данные и сами OSD не видят настоящее содержимое образов. Разные ключи
+в том числе могут иметь разные снимки или клоны одного и того же образа. Например, можно сделать
+базовый образ ВМ (условный Debian Linux) нешифрованным, но наследовать от него шифрованные образы
+клиентских ВМ.
+
+Ключи шифрования образов могут храниться либо в etcd, либо во внешнем Vault. Во втором случае
+в etcd хранятся только ID ключей, а Vitastor вообще не имеет доступа к данным образов. Для
+использования Vault нужно создать образ с опцией `--enc_key vault:ID`, а в конфигурации указать
+опции:
+- vault_url
+- vault_ca
+- vault_client_cert
+- vault_client_key
+
+Ещё раз повторимся, что если AES-XTS используется с полным шифрованием трафика (`proto_checksums=gcm`),
+то данные образов шифруются дважды - сначала AES-XTS, а потом AES-GCM. Можете использовать,
+только если вы совсем параноик :-).
+
+## Производительность шифрования
+
+У вас может возникнуть вопрос - а как быстро всё это прекрасное шифрование работает?
+
+Ответ - скорость сильно зависит от процессора. Складывается она из нескольких вещей:
+
+- 
+
+TODO: vitastor-cli bench.
+
+## Аутентификация по сертификатам
+
+При включённом шифровании клиенты, OSD и мониторы Vitastor аутентифицируются по сертификатам
+как при соединениях с etcd (Antietcd), так и с OSD.
+
+Для OSD и мониторов должны использоваться отдельные сертификаты - либо самоподписанные, либо
+подписанные отдельными CA (`osd_ca` и `mon_ca`). При этом все OSD могут использовать один и
+тот же сертификат и все мониторы тоже могут использовать один и тот же сертификат, так как
+привилегии разных OSD или разных мониторов ничем не отличаются (теоретически можно было бы
+сделать разграничение сертификатов OSD по пулам, но пока что такой необходимости не было).
+
+Также сертификат монитора может быть вообще не нужен, если Antietcd встраивается в сам монитор.
+В этом случае монитор и так имеет доступ ко всем данным etcd прямо в памяти.
+
+Каждый клиент должен иметь свой сертификат, подписанный общим корневым сертификатом
+для клиентов (`client_ca`). Common Name сертификата должно равняться имени пользователя.
+
+## Модель прав доступа
+
+Привилегии пользователей хранятся в данных etcd в ключах `/vitastor/config/user/<имя>`.
+
+У пользователя есть 2 свойства:
+- Тип:
+  - Клиент (`type=client` или не указано) - может читать и модифицировать только явным образом
+    разрешённые образы.
+  - Администратор (`type=admin`) - может читать и модифицировать все образы, а также администрировать
+    кластер: смотреть общую статистику и состояние, создавать и удалять OSD и так далее.
+- Список имён групп, членом которых пользователь является.
+
+У образов есть 3 свойства:
+- Владелец (owner) - имя пользователя, которому разрешено и читать, и менять образ
+- Группа владельцев (owner_group) - имя группы владельцев
+- Группа читатетей (reader_group) - имя группы пользователей, которым разрешено читать образ
+
+У пулов есть 1 свойство:
+- Группа создателей (creator_group) - имя группы пользователей, которым разрешено создавать образы в пуле
+
+## Права доступа к данным etcd
+
+Привилегии реализуются через Antietcd во всех режимах работы. Если используется etcd, то
+Antietcd выступает в роли фильтрующего прокси, при этом он может быть встроен в монитор
+Vitastor или запущен отдельно. В этом случае etcd должен разрешать входящие подключения
+только от Antietcd, а все остальные компоненты должны соединяться с Antietcd.
+
+Если же используется Antietcd, то привилегии реализуются в нём самом.
+
+Если используется встроенный в монитор Antietcd, то привилегии включаются либо параметром
+`use_auth: true`, либо, если этот параметр не указан - включается автоматически, если задан
+любой из параметров `client_ca`, `osd_ca`, `mon_ca`. При этом монитор требует указания
+параметров `client_ca` и `osd_ca`, а если не используется режим проксирования в etcd -
+также `antietcd_server_ca`, чтобы Antietcd мог отличать кластерные соединения от клиентских.
+
+Если используется отдельно стоящий Antietcd, привилегии нужно включать явным образом.
+
+Встроенные привилегии etcd не поддерживаются по причине их многочисленных недоработок:
+- Аутентификация по сертификатам не работает в REST интерфейсе etcd,
+- Привилегии хранятся отдельно от k/v и не могут участвовать в транзакциях,
+- Менять привилегии может только администратор (root)
+- Нет поддержки фильтрации ответов чтения по привилегиям.
+
+Подробный список привилегий на ключи в etcd [смотрите ниже](#привилегии-etcd).
+
+## Права доступа к данным OSD
+
+Регулируется опцией `use_auth`, либо, если она не указана, включается автоматически,
+если используется шифрование, то есть, если заданы опции `osd_ca` и `client_ca`.
+
+OSD аутентифицирует клиентов по сертификатам и разрешает каждому клиенту только
+то, что ему разрешено согласно модели прав доступа.
+
+Подробный список разрешаемых OSD операций [смотрите ниже](#привилегии-osd).
+
+## Права доступа к API
+
+[vitastor-cli serve](../usage/cli.ru.md#serve) также поддерживает клиентскую
+аутентификацию по сертификатам. Принимаются только сертификаты, подписанные
+`client_ca`. В качестве серверного сертификата используется отдельный сертификат
+`server_cert` с ключом `server_key`.
+
+При этом для корректной работы `vitastor-cli serve` он сам должен использовать
+для доступа в Vitastor сертификат (`cert`+`pkey`) пользователя с правами
+администратора (`type=admin`).
+
+Обычным клиентам при доступе к API разрешаются только API-операции с образами,
+доступными им либо на чтение (для чтения), либо на запись (для модификации).
+Все остальные API-вызовы разрешаются только для администраторов.
+
+Подробный список разрешаемых API операций [смотрите ниже](#привилегии-api).
+
+## Привилегии etcd
+
+Ниже все названия ключей приведены без общего префикса `/vitastor`.
+
+Разрешённые операции с ключами в Antietcd для клиентов (`type=client`):
+- Только чтение:
+  - Разрешено всегда:
+    - `/config/global`
+    - `/config/node_placement`
+    - `/config/pools`
+    - `/pg/config`
+    - `/osd/state/*`
+    - `/pg/state/*`
+    - `/index/maxid/*`
+  - Для образов, которые [может читать пользователь](#модель-прав-доступа):
+    - `/config/inode/*`
+    - `/index/image/*`
+    - `/inode/stats/*`
+- Чтение и запись:
+  - Для пулов, в которых может создавать образы пользователь:
+    - `/index/maxid/*`
+  - Для образов, которыми владеет пользователь:
+    - `/config/inode/*`
+    - `/index/image/*`
+
+Разрешённые операции с ключами в Antietcd для администраторов (`type=admin`):
+- Чтение:
+  - `/stats`
+  - `/mon/*`
+  - `/pg/*`
+  - `/pgstats/*`
+  - `/inode/stats/*`
+  - `/pool/stats/*`
+- Чтение и запись:
+  - `/config/*`
+  - `/osd/*`
+  - `/index/*`
+  - `/pg/history/*`
+
+Разрешённые операции с ключами в etcd для OSD:
+- Чтение:
+  - `/pg/config`
+  - `/config/*`
+- Чтение и запись:
+  - `/osd/*`
+  - `/pg/state/*`
+  - `/pg/history/*`
+  - `/pgstats/*`
+
+Разрешённые операции с ключами в etcd для мониторов:
+- Чтение:
+  - `/config/*`
+  - `/osd/*`
+  - `/pgstats/*`
+- Чтение и запись:
+  - `/pg/config`
+  - `/stats`
+  - `/history/last_clean_pgs`
+  - `/mon/*`
+  - `/pg/history/*`
+  - `/inode/stats/*`
+  - `/pool/stats/*`
+
+## Привилегии OSD
+
+Клиентские операции:
+- READ - разрешено для образов, доступных пользователю на чтение.
+- WRITE, DELETE, SCRUB - разрешены для образов, доступных пользователю на запись.
+- SYNC - операция не связана с образом и разрешена всегда.
+- DESCRIBE - операция разрешена только для администраторов (используются командами
+  `vitastor-cli describe` и `fix`).
+- PING - операция разрешена всегда.
+- SHOW_CONFIG - операция разрешена всегда, однако если в ней клиент представляется
+  как OSD, то проверяется, что он использует сертификат, подписанный `osd_ca`.
+- SEC_LIST (листинг) - разрешена другим OSD и администраторам с любыми параметрами,
+  а обычным клиентам разрешена только для запросов, ограниченных образом, доступным
+  пользователю на чтение.
+
+Кластерные операции - разрешаются только другим OSD:
+- SEC_READ
+- SEC_WRITE
+- SEC_WRITE_STABLE
+- SEC_SYNC
+- SEC_STABILIZE
+- SEC_ROLLBACK
+- SEC_DELETE
+- SEC_READ_BMP
+- SEC_LOCK
+
+## Привилегии API
+
+Клиентам (пользователям с `type=client`) разрешаются операции:
+- image/list - для образов, которые пользователь может читать.
+- image/create - для пулов, в которых пользователю разрешено создавать образы, либо
+  для создания снимков образов, которыми пользователь владеет.
+- image/delete, image/flatten, image/modify - для образов, которыми пользователь владеет.
+
+Все остальные операции разрешаются только администраторам (`type=admin`).
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+Таким образом, доступны следующие варианты настройки:
+
+### Mon в роли Etcd proxy
+
+Mon
+- use_antietcd: true
+- etcd_proxy = {
+    urls: [],
+    cert = <antietcd.pem>,
+    key,
+    ca = <etcd.pem>,
+}
+- antietcd_cert = antietcd.pem
+- antietcd_key
+
+etcd
+--client-cert-auth --cert-file=etcd.pem --key-file=etcd.key --trusted-ca-file=antietcd.pem \
+--peer-client-cert-auth --peer-cert-file=etcd.pem --peer-key-file=etcd.key --peer-trusted-ca-file=etcd.pem
+
+### Mon с отдельным Antietcd Proxy
+
+Mon
+- use_antietcd: false
+- etcd_ca = antietcd.pem
+
+Antietcd
+--client_cert_auth 1 --auth_filter vitastor_auth_filter.js --etcd_proxy url1,url2,... \
+--cert antietcd.pem --key antietcd.key --ca client_ca.pem --osd_ca osd_ca.pem \
+--etcd_cert antietcd.pem --etcd_key antietcd.key --etcd_ca etcd.pem
+
+etcd
+--client-cert-auth --cert-file=etcd.pem --key-file=etcd.key --trusted-ca-file=antietcd.pem \
+--peer-client-cert-auth --peer-cert-file=etcd.pem --peer-key-file=etcd.key --peer-trusted-ca-file=etcd.pem
+
+### Mon со встроенным Antietcd
+
+Mon
+- use_antietcd: true
+- use_auth: true
+- antietcd_cert = antietcd.pem
+- antietcd_key
+
+### Отдельный Antietcd
+
+Mon
+- use_antietcd: false
+- etcd_ca = antietcd.pem
+
+Antietcd
+--client_cert_auth 1 --auth_filter vitastor_auth_filter.js
+
+## Варианты настройки
+
+### Настройка по умолчанию
+
+Используются только контрольные суммы данных на транспортном уровне. Соединения с etcd не шифруются.
+Аутентификация и авторизация не используется, любой клиент имеет доступ ко всем данным кластера.
+
+Аналог настройки:
+- proto_checksums: payload
+
+### Полная защита
+
+Везде
+- osd_ca
+- client_ca
+- etcd_ca = antietcd.pem
+
+OSD
+- osd_cert
+- osd_pkey
+
+Клиент
+- cert
+- pkey
+
+### Только защита etcd
+
+- etcd_ca
+- etcd_cert
+- etcd_key
+
+### antietcd и только защита antietcd
+
+- etcd_ca
+- etcd_cert
+- etcd_key
+- use_antietcd: true
+- antietcd_cert = etcd_ca
+- antietcd_key
+- antietcd_ca = etcd_cert
+
+### Полное шифрование протокола, включая данные
+
+Внимание: если включить этот вариант защиты и при этом 
+
+### Только контрольные суммы на транспортном уровне, без шифрования
+
+## Настройка Vault/OpenBao
+
+openssl req -days 3650 -x509 -addext basicConstraints=critical,CA:TRUE,pathlen:1 --addext subjectAltName=DNS:vault \
+    -new -newkey rsa:4096 -nodes -keyout vault.key -out vault.crt
+
+bao status -ca-cert /etc/openbao/tls/vault.crt -address=https://vault:8200
+
+bao operator init -n 1 -t 1 -ca-cert /etc/openbao/tls/vault.crt -address=https://vault:8200
+
+bao operator unseal -ca-cert /etc/openbao/tls/vault.crt -address=https://vault:8200
+
+bao auth enable -ca-cert /etc/openbao/tls/vault.crt -address=https://vault:8200 cert
+
+bao secrets enable -ca-cert /etc/openbao/tls/vault.crt -address=https://vault:8200 -path=secret kv-v1
+
+bao kv put -ca-cert /etc/openbao/tls/vault.crt -address=https://vault:8200 secret/vitastor/testimg3 key=$(openssl rand -hex 64)
+
+cat >testimg3.policy <<EOF
+path "/secret/vitastor/testimg3" {
+  capabilities = ["read"]
+}
+EOF
+
+bao policy write -ca-cert /etc/openbao/tls/vault.crt -address=https://vault:8200 testimg3 testimg3.policy
+
+bao write -ca-cert /etc/openbao/tls/vault.crt -address=https://vault:8200 auth/cert/certs/testimg3 certificate=@testimg3.crt display_name=testimg3 token_ttl=24h token_policies=testimg3
+
+curl --cacert /etc/vitastor/vault.crt --cert testimg3.crt --key testimg3.key --json '{}' https://vault:8200/v1/auth/cert/login
+
+curl --cacert /etc/vitastor/vault.crt --cert testimg3.crt --key testimg3.key -H 'X-Vault-Token: s.Qkrm78BeK7Rqdz5MA3eJZNbu' https://vault:8200/v1/secret/vitastor/testimg3

docs/usage/cli.en.md

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

docs/usage/cli.ru.md

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

docs/usage/nfs.en.md

@@ -262,3 +262,4 @@ Options:
 | `--logfile <FILE>` | log to the specified file                                |
 | `--enforce 1`      | enforce permissions at the server side (no by default)   |
 | `--foreground 1`   | stay in foreground, do not daemonize                     |
+| `--trace`          | trace all NFS requests                                   |

docs/usage/nfs.ru.md

@@ -274,3 +274,4 @@ VitastorFS из GPUDirect.
 | `--logfile <FILE>` | записывать логи в заданный файл                         |
 | `--enforce 1`      | проверять права доступа на стороне сервера (по умолчанию нет) |
 | `--foreground 1`   | не уходить в фон после запуска                          |
+| `--trace`          | логгировать все запросы NFS                             |

mon/antietcd_adapter.js

@@ -3,6 +3,7 @@
 
 const AntiEtcd = require('antietcd');
 
+const vitastor_auth_filter = require('./vitastor_auth_filter.js');
 const vitastor_persist_filter = require('./vitastor_persist_filter.js');
 const { b64, local_ips } = require('./utils.js');
 
@@ -18,7 +19,7 @@ class AntiEtcdAdapter
                 cluster = cluster ? (''+(cluster||'')).split(/,+/) : [];
             cluster = Object.keys(cluster.reduce((a, url) =>
             {
-                a[url.toLowerCase().replace(/^(https?:\/\/)/, '').replace(/\/.*$/, '')] = true;
+                a[url.toLowerCase().replace(/^(https?:\/\/)?(.*?)(\/.*)?$/, (m, m1, m2) => (m1||'http://')+m2)] = true;
                 return a;
             }, {}));
             const cfg_port = config.antietcd_port;
@@ -26,10 +27,16 @@ 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 cluster_local = cluster.map(s =>
+            {
+                const m = /^https?:\/\/(?:\[(.*)\]|([^\[\:]+))(?::(\d+))?$/.exec(s);
+                return [ m[2] || m[1], m[3] || 2379 ];
+            });
+            const selected = cluster_local.filter(ip => is_local[ip[0]] && (!cfg_port || ip[1] == cfg_port));
             if (selected.length > 1)
             {
-                console.error('More than 1 etcd_address matches local IPs, please specify port');
+                console.error('More than 1 etcd_address matches local IPs: '+(selected.join(', '))+', please specify port');
                 process.exit(1);
             }
             else if (selected.length == 1)
@@ -37,14 +44,52 @@ class AntiEtcdAdapter
                 const antietcd_config = {
                     ip: selected[0][0],
                     port: selected[0][1],
+                    cert: config.antietcd_cert,
+                    key: config.antietcd_key,
+                    ca: config.antietcd_ca,
                     data: config.antietcd_data_file || ((config.antietcd_data_dir || '/var/lib/vitastor') + '/mon_'+selected[0][1]+'.json.gz'),
                     persist_filter: vitastor_persist_filter({ vitastor_prefix: config.etcd_prefix || '/vitastor' }),
                     node_id: selected[0][0]+':'+selected[0][1], // node_id = ip:port
-                    cluster: (cluster.length == 1 ? null : cluster.reduce((a, c) => { a[c] = "http://"+c; return a; }, {})),
+                    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.etcd_proxy)
+                {
+                    // Monitor may use the builtin etcd_proxy mode
+                    if (!config.etcd_proxy.urls)
+                    {
+                        console.error('etcd_proxy.urls are empty');
+                        process.exit(1);
+                    }
+                    antietcd_config.etcd_proxy = config.etcd_proxy.urls;
+                    antietcd_config.etcd_cert = config.etcd_proxy.cert;
+                    antietcd_config.etcd_key = config.etcd_proxy.key;
+                    antietcd_config.etcd_ca = config.etcd_proxy.ca;
+                    delete antietcd_config.data;
+                    delete antietcd_config.persist_filter;
+                    delete antietcd_config.cluster;
+                    delete antietcd_config.cluster_key;
+                }
+                const use_auth = config.use_auth || config.use_auth == null && config.client_ca;
+                if (use_auth)
+                {
+                    antietcd_config.client_cert_auth = true;
+                    antietcd_config.auth_filter = vitastor_auth_filter;
+                    antietcd_config.ca = config.client_ca;
+                    antietcd_config.osd_ca = config.osd_ca;
+                    antietcd_config.mon_ca = config.mon_ca;
+                    if (!config.etcd_proxy)
+                    {
+                        antietcd_config.peer_ca = config.antietcd_server_ca;
+                        if (!config.antietcd_server_ca || config.antietcd_server_ca == config.client_ca)
+                        {
+                            console.error('Secure setup requires separate antietcd_server_ca (for signing antietcd server certificates) and client_ca (for signing client certificates)');
+                            process.exit(1);
+                        }
+                    }
+                }
                 for (const key in config)
                 {
                     if (key.substr(0, 9) === 'antietcd_')
@@ -169,7 +214,7 @@ class AntiEtcdAdapter
                     await new Promise(ok => setTimeout(ok, timeout-(Date.now()-prev)));
                 }
                 prev = Date.now();
-                const res = await this.antietcd.api(path.replace(/^\/+/, '').replace(/\/+$/, '').replace(/\/+/g, '_'), body);
+                const res = await this.antietcd.api(path.replace(/^\/+/, '').replace(/\/+$/, '').replace(/\/+/g, '_'), body, { user_type: 'mon' });
                 if (res.error)
                 {
                     console.error('Failed to query antietcd '+path+' (retry '+retry+'/'+retries+'): '+res.error);

mon/etcd_adapter.js

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

mon/etcd_schema.js

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

mon/mon.js

@@ -627,7 +627,7 @@ class Mon
                 if (this.state.pg.history[pool_id] &&
                     this.state.pg.history[pool_id][pg])
                 {
-                    pg_history[pg-1] = this.state.pg.history[pool_id][pg];
+                    pg_history[pg-1] = JSON.parse(JSON.stringify(this.state.pg.history[pool_id][pg]));
                 }
             }
             const real_prev_pgs = [];
@@ -719,7 +719,7 @@ class Mon
                 this.next_recheck_timer = null;
                 this.next_recheck_at = 0;
                 this.schedule_recheck();
-            }, now-this.next_recheck_at);
+            }, (this.next_recheck_at-now)*1000);
         }
     }
 

mon/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vitastor-mon",
-  "version": "3.0.9",
+  "version": "3.0.12",
   "description": "Vitastor SDS monitor service",
   "main": "mon-main.js",
   "scripts": {

mon/pg_utils.js

@@ -84,7 +84,7 @@ function scale_pg_history(prev_pg_history, prev_pgs, new_pgs)
         finish_pg_history(merged_history[1]);
         for (let i = 0; i < new_pg_count; i++)
         {
-            new_pg_history[i] = { ...merged_history[1] };
+            new_pg_history[i] = JSON.parse(JSON.stringify(merged_history[1]));
         }
     }
     // Mark history keys for removed PGs as removed
@@ -102,7 +102,7 @@ function scale_pg_count(prev_pgs, new_pg_count)
     {
         for (let i = prev_pgs.length; i < new_pg_count; i++)
         {
-            prev_pgs[i] = prev_pgs[i % prev_pgs.length];
+            prev_pgs[i] = [ ...prev_pgs[i % prev_pgs.length] ];
         }
     }
     else if (prev_pgs.length > new_pg_count)

mon/scripts/make-etcd

@@ -1,38 +1,278 @@
 #!/usr/bin/node
-// Simple systemd unit generator for etcd
+// Simple Vitastor etcd / antietcd / TLS configurator
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: MIT
 
-// USAGE:
-// 1) Put the same etcd_address into /etc/vitastor/vitastor.conf on all monitor nodes
-// 2) Run ./make-etcd.js. It will create the etcd service on one of specified IPs
-
 const child_process = require('child_process');
 const fs = require('fs');
 const os = require('os');
+const readline = require('readline');
 
 run().catch(e => { console.error(e); process.exit(1); });
 
+const help_text = `Initialize a Vitastor cluster (etcd, vitastor.conf and TLS certificates)
+(c) Vitaliy Filippov, 2019+ (MIT)
+
+USAGE:
+1) Create a minimal vitastor.conf with etcd_address, osd_network and (optionally) use_auth.
+   Example: {"etcd_address":["http://10.0.0.10:2379","http://10.0.0.11:2379","http://10.0.0.12:2379"],"use_auth":false,"osd_network":"10.0.0.0/24"}
+   Or: {"etcd_address":["https://10.0.0.10:2379","https://10.0.0.11:2379","https://10.0.0.12:2379"],"use_auth":true,"osd_network":"10.0.0.0/24"}
+2) Run: ${process.argv[1]} [./vitastor.conf]
+   You can run it on etcd/monitor nodes or on an external node.
+   It configures etcd, generates TLS certificates (on the first or external node), copies them
+   to other etcd/monitor nodes, and updates vitastor.conf with TLS options.
+3) If you have OSD-only nodes, run:
+   ${process.argv[1]} --copy-to-osd-node NODE_NAME ./vitastor.conf
+   It copies vitastor.conf and required TLS certificates to that node.
+
+OPTIONS:
+--gen-certs
+    force certificate generation even if it's not the first node
+--no-certs
+    disable certificate generation
+--no-copy
+    do not copy initial certificates to other nodes
+--copy-to-osd-node NODE[,NODE2,...]
+    copy vitastor.conf and TLS certificates required for OSDs to NODES using scp
+--copy-to-mon-node NODE[,NODE2,...]
+    copy vitastor.conf and TLS certificates required for monitor and etcd to NODES using scp
+--copy-to-client-node NODE[,NODE2,...]
+    copy vitastor.conf and TLS certificates required for clients to NODES using scp
+`;
+
 async function run()
 {
-    const config_path = process.argv[2] || '/etc/vitastor/vitastor.conf';
-    if (config_path == '-h' || config_path == '--help')
+    let config_path = '/etc/vitastor/vitastor.conf';
+    let config_dir = '/etc/vitastor/';
+    let gen_certs = 'auto';
+    let copy_initial = true;
+    let copy_to_osd = 
+    for (let i = 2; i < process.argv.length; i++)
     {
-        console.log(
-            'Initialize systemd etcd service for Vitastor\n'+
-            '(c) Vitaliy Filippov, 2019+ (MIT)\n'+
-            '\n'+
-            'USAGE:\n'+
-            '1) Put the same etcd_address into /etc/vitastor/vitastor.conf on all monitor nodes\n'+
-            '2) Run '+process.argv[1]+' [config_path]\n'
-        );
-        process.exit(0);
+        if (arg == '-h' || arg == '--help')
+        {
+            console.log(help_text);
+            process.exit(0);
+        }
+        else if (arg == '--only-certs')
+        {
+            i++;
+            gen_certs = 
+        }
+        else if (arg == '--copy')
+        {
+            i++;
+            copy = process.argv[i];
+            if (copy !== 'ask' && copy !== 'yes' && copy !== 'no')
+            {
+                console.error('--copy should be "ask", "yes" or "no"');
+                process.exit(1);
+            }
+        }
+        else if (arg[0] == '-')
+        {
+            console.error('Unknown option: '+arg[0]);
+            process.exit(1);
+        }
+        else
+        {
+            config_path = arg;
+        }
     }
     if (!fs.existsSync(config_path))
     {
         console.log(config_path+' is missing');
         process.exit(1);
     }
+    const config = JSON.parse(fs.readFileSync(config_path, { encoding: 'utf-8' }));
+    if (!config.etcd_address)
+    {
+        console.log("etcd_address is missing in "+config_path);
+        process.exit(1);
+    }
+    const etcds = (config.etcd_address instanceof Array ? config.etcd_address : (''+config.etcd_address).split(/,/))
+        .map(s => /^(https?):\/\/(\[[^\]]+\]|[^\[\]\:\/]+)(?::(\d+))?/.exec(s.toLowerCase()))
+        .filter(s => s)
+        .map(s => ({
+            scheme: s[1],
+            addr: s[2].indexOf(':') && s[2][0] != '[' ? '['+s[2]+']' : s[2],
+            ip: s[2][0] == '[' ? s[2].substr(1, s[2].length-2) : s[2],
+            port: s[3],
+        }));
+    const tls = etcds.filter(e => e.scheme === 'https').length > 0;
+    const use_auth = tls && config.use_auth;
+    const num = select_local_etcd(etcds);
+    if (tls)
+    {
+        if (gen_certs === 'yes')
+        {
+            gen_certs = true;
+            console.log('Certificate generation is requested explicitly, generating');
+        }
+        else if (gen_certs === 'no')
+        {
+            gen_certs = false;
+            console.log('Certificate generation is disabled explicitly, skipping');
+        }
+        else if (num < 0)
+        {
+            gen_certs = true;
+            console.log('No matching IPs in etcd_address from '+config_path+', only generating certificates');
+        }
+        else if (fs.existsSync("/etc/vitastor/etcd.crt"))
+        {
+            gen_certs = false;
+            console.log('/etc/vitastor/etcd.crt already exists, assuming certificates are already generated');
+        }
+        else if (num === 0)
+        {
+            gen_certs = true;
+            console.log('This is monitor node 1, generating certificates');
+        }
+        else
+        {
+            console.log('This is monitor node '+(num+1)+', /etc/vitastor/etcd.crt does not exist, please copy certificates to this node');
+            process.exit(1);
+        }
+        if (gen_certs)
+        {
+            if (copy === 'ask')
+                copy = await ask_copy('Copy certificates and vitastor.conf to other nodes after generation?');
+            copy = (copy === 'y' || copy === 'yes');
+            await make_certs(dir, copy);
+        }
+        await write_auth_config(config, config_path);
+    }
+    if (num < 0)
+    {
+        console.log('No matching IPs in etcd_address from '+config_path);
+        process.exit(tls && gen_certs ? 0 : 1);
+    }
+    await configure_etcd();
+    process.exit(0);
+}
+
+async function ask_copy(question)
+{
+    const rl = readline.createInterface({
+        input: process.stdin,
+        output: process.stdout,
+        prompt: '> ',
+    });
+    let copy;
+    while (true)
+    {
+        copy = await new Promise(ok => rl.question(question, ok));
+        if (copy != 'y' && copy != 'n' && copy != 'yes' && copy != 'no')
+            console.log('Please type "yes" or "no"');
+        else
+            break;
+    }
+    return copy;
+}
+
+async function make_certs(dir, copy)
+{
+    console.log(`-----
+Generating certificates in ${dir}
+-----
+`);
+    await make_ca("/O=Vitastor etcd CA", dir+"etcd_ca");
+    await make_signed("/CN=Vitastor etcd", dir+"etcd", dir+"etcd_ca", etcds.map(e => "IP:"+e.ip).join(','));
+    if (use_auth)
+    {
+        await make_ca("/O=Vitastor Antietcd CA", dir+"antietcd_ca");
+        await make_signed("/CN=Vitastor Antietcd", dir+"antietcd", dir+"antietcd_ca", etcds.map(e => "IP:"+e.ip).join(','));
+        await make_ca("/CN=Vitastor OSD", dir+"osd");
+        await make_ca("/O=Vitastor Client CA", dir+"client_ca");
+        await make_signed("/CN=admin", dir+"admin", dir+"client_ca");
+    }
+    if (use_auth)
+    {
+        console.log(`-----
+Certificates generated, commands to copy them:
+- Monitor+OSD node:
+  cd ${dir} && scp antietcd_ca.crt antietcd.crt antietcd.key osd.crt osd.key client_ca.crt etcd_ca.crt etcd.crt etcd.key root@NODE:/etc/vitastor/
+- Monitor node:
+  cd ${dir} && scp antietcd_ca.crt antietcd.crt antietcd.key osd.crt client_ca.crt etcd_ca.crt etcd.crt etcd.key root@NODE:/etc/vitastor/
+- OSD node:
+  cd ${dir} && scp antietcd_ca.crt osd.crt osd.key client_ca.crt root@NODE:/etc/vitastor/
+-----
+`);
+    }
+    else
+    {
+        console.log(`-----
+Certificates generated, commands to copy them:
+- Monitor node:
+  cd ${dir} && scp etcd_ca.crt etcd.crt etcd.key root@NODE:/etc/vitastor/
+-----
+`);
+    }
+    if (copy)
+    {
+        const to_copy = use_auth
+            ? [ "antietcd_ca.crt", "antietcd.crt", "antietcd.key", "osd.crt", "osd.key", "client_ca.crt", "etcd_ca.crt", "etcd.crt", "etcd.key" ]
+            : [ "etcd_ca.crt", "etcd.crt", "etcd.key" ];
+        for (const node of etcds)
+        {
+            await system("scp "+dir+to_copy.join(" "+dir)+" root@"+node.ip+"/etc/vitastor/");
+        }
+    }
+    else
+    {
+        console.warn('Certificates generated in /etc/vitastor, please copy them to other nodes');
+    }
+}
+
+async function write_auth_config(config, config_path)
+{
+    const auth = {};
+    if (use_auth)
+    {
+        auth["use_antietcd"] = true;
+        auth["etcd_proxy"] = {
+            urls: etcds.map(e => e.ip+':2381'),
+            cert: "/etc/vitastor/antietcd.crt",
+            key: "/etc/vitastor/antietcd.key",
+            ca: "/etc/vitastor/etcd_ca.crt",
+        };
+        auth["antietcd_cert"] = "/etc/vitastor/antietcd.crt";
+        auth["antietcd_key"] = "/etc/vitastor/antietcd.key";
+        auth["etcd_ca"] = "/etc/vitastor/antietcd_ca.crt";
+        auth["osd_cert"] = "/etc/vitastor/osd.crt";
+        auth["osd_pkey"] = "/etc/vitastor/osd.key";
+        auth["osd_ca"] = "/etc/vitastor/osd.crt";
+        auth["client_ca"] = "/etc/vitastor/client_ca.crt";
+        auth["cert"] = "/etc/vitastor/admin.crt";
+        auth["pkey"] = "/etc/vitastor/admin.key";
+    }
+    else
+    {
+        auth["etcd_ca"] = "/etc/vitastor/etcd.crt";
+    }
+    for (const k in auth)
+    {
+        if ((k in config) && JSON.stringify(auth[k]) != JSON.stringify(config[k]))
+        {
+            // Auth options already overridden with non-default
+            console.log(k+" is already overridden in "+config_path+", skipping config update");
+            return;
+        }
+    }
+    for (const k in auth)
+    {
+        config[k] = auth[k];
+    }
+    console.log(`-----
+Updating ${config_path}
+-----
+`);
+    fs.writeFileSync(config_path, JSON.stringify(config, 0, 4));
+}
+
+async configure_etcd()
+{
     const in_docker = fs.existsSync("/etc/vitastor/etcd.conf") &&
         fs.existsSync("/etc/vitastor/docker.conf");
     if (!in_docker && fs.existsSync("/etc/systemd/system/vitastor-etcd.service"))
@@ -45,30 +285,49 @@ async function run()
         console.log("/etc/systemd/system/etcd.service already exists");
         process.exit(1);
     }
-    const config = JSON.parse(fs.readFileSync(config_path, { encoding: 'utf-8' }));
-    if (!config.etcd_address)
+    const etcd_url = etcds[num].scheme + '://' + etcds[num].addr;
+    const options = {
+        name: 'etcd'+etcds[num].ip.replace(/[^0-9a-z_]/ig, '_'),
+        advertise_client_urls: etcd_url+':'+(use_auth ? 2381 : 2379),
+        listen_client_urls: etcd_url+':'+(use_auth ? 2381 : 2379),
+        initial_advertise_peer_urls: etcd_url+':2380',
+        listen_peer_urls: etcd_url+':2380',
+        initial_cluster_token: 'vitastor-etcd-1',
+        initial_cluster_state: 'new',
+        initial_cluster: etcds.map(e => `etcd${e.ip.replace(/[^0-9a-z_]/ig, '_')}=${e.scheme}://${e.addr}:2380`).join(','),
+        snapshot_count: 10000,
+        max_txn_ops: 100000,
+        max_request_bytes: 104857600,
+        auto_compaction_retention: 10,
+        auto_compaction_mode: 'revision',
+    };
+    if (tls)
     {
-        console.log("etcd_address is missing in "+config_path);
-        process.exit(1);
+        options['cert_file'] = '/etc/vitastor/etcd.crt';
+        options['key_file'] = '/etc/vitastor/etcd.key';
+        if (use_auth)
+        {
+            options['client_cert_auth'] = '1';
+            options['trusted_ca_file'] = '/etc/vitastor/antietcd.crt';
+        }
+        options['peer_cert_file'] = '/etc/vitastor/etcd.crt';
+        options['peer_key_file'] = '/etc/vitastor/etcd.key';
+        if (use_auth)
+        {
+            options['peer_client_cert_auth'] = '1';
+            options['peer_trusted_ca_file'] = '/etc/vitastor/etcd.crt';
+        }
     }
-    const etcds = (config.etcd_address instanceof Array ? config.etcd_address : (''+config.etcd_address).split(/,/))
-        .map(s => (''+s).replace(/^https?:\/\/|(:\d+)?(\/.*)?$/g, '').replace(/^\[(.*)\]$/, '$1').toLowerCase());
-    const num = select_local_etcd(etcds);
-    if (num < 0)
+    let etcd_conf = fs.existsSync("/etc/vitastor/etcd.conf")
+        ? fs.readFileSync("/etc/vitastor/etcd.conf", { encoding: 'utf-8' })
+        : "";
+    for (const k in options)
     {
-        console.log('No matching IPs in etcd_address from '+config_path);
-        process.exit(0);
+        etcd_conf = replace_env(etcd_conf, 'ETCD_'+k.toUpperCase().replace(/-/, '_'), options[k]);
     }
-    const etcd_url = 'http://' + (etcds[num].indexOf(':') >= 0 ? '['+etcds[num]+']' : etcds[num]);
-    const etcd_name = 'etcd'+etcds[num].replace(/[^0-9a-z_]/ig, '_');
-    const etcd_cluster = etcds.map(e => `etcd${e.replace(/[^0-9a-z_]/ig, '_')}=http://${e.indexOf(':') >= 0 ? '['+e+']' : e}:2380`).join(',');
+    fs.writeFileSync("/etc/vitastor/etcd.conf", etcd_conf);
     if (in_docker)
     {
-        let etcd_conf = fs.readFileSync("/etc/vitastor/etcd.conf", { encoding: 'utf-8' });
-        etcd_conf = replace_env(etcd_conf, 'ETCD_NAME', etcd_name);
-        etcd_conf = replace_env(etcd_conf, 'ETCD_IP', etcds[num]);
-        etcd_conf = replace_env(etcd_conf, 'ETCD_INITIAL_CLUSTER', etcd_cluster);
-        fs.writeFileSync("/etc/vitastor/etcd.conf", etcd_conf);
         console.log('etcd for Vitastor configured. Run `systemctl enable --now vitastor-etcd` to start etcd');
         process.exit(0);
     }
@@ -83,12 +342,8 @@ Wants=network-online.target local-fs.target time-sync.target
 [Service]
 Restart=always
 Environment=GOGC=50
-ExecStart=etcd --name ${etcd_name} --data-dir /var/lib/etcd/vitastor \\
-    --snapshot-count 10000 --advertise-client-urls ${etcd_url}:2379 --listen-client-urls ${etcd_url}:2379 \\
-    --initial-advertise-peer-urls ${etcd_url}:2380 --listen-peer-urls ${etcd_url}:2380 \\
-    --initial-cluster-token vitastor-etcd-1 --initial-cluster ${etcd_cluster} \\
-    --initial-cluster-state new --max-txn-ops=100000 --max-request-bytes=104857600 \\
-    --auto-compaction-retention=10 --auto-compaction-mode=revision
+EnvironmentFile=/etc/vitastor/etcd.conf
+ExecStart=etcd --data-dir /var/lib/etcd/vitastor
 WorkingDirectory=/var/lib/etcd/vitastor
 ExecStartPre=+chown -R etcd /var/lib/etcd/vitastor
 User=etcd
@@ -106,7 +361,6 @@ WantedBy=multi-user.target
     // Disable distribution etcd unit and enable our one
     await system(`systemctl disable --now etcd`);
     await system(`systemctl enable --now vitastor-etcd`);
-    process.exit(0);
 }
 
 function replace_env(text, key, value)
@@ -119,16 +373,29 @@ function replace_env(text, key, value)
 function select_local_etcd(etcds)
 {
     const ifaces = os.networkInterfaces();
+    const local = {};
     for (const ifname in ifaces)
+    {
         for (const iface of ifaces[ifname])
-            for (let i = 0; i < etcds.length; i++)
-                if (etcds[i] == iface.address.toLowerCase())
-                    return i;
+        {
+            const addr = iface.address;
+            if (iface.family == 'IPv6')
+                local[addr.toLowerCase()] = local['['+addr.toLowerCase()+']'] = true;
+            else
+                local[addr] = true;
+        }
+    }
+    for (let i = 0; i < etcds.length; i++)
+    {
+        if (local[etcds[i].addr])
+            return i;
+    }
     return -1;
 }
 
 async function system(cmd)
 {
+    console.log('Running '+cmd);
     const cp = child_process.spawn(cmd, { shell: true, stdio: [ 0, 1, 2 ] });
     let finish_cb;
     cp.on('exit', () => finish_cb && finish_cb());
@@ -136,3 +403,19 @@ async function system(cmd)
         await new Promise(ok => finish_cb = ok);
     return cp.exitCode;
 }
+
+async function make_ca(subj, filename)
+{
+    if (await system("openssl req -days 3650 -x509 -subj '"+subj+"' -addext basicConstraints=critical,CA:TRUE,pathlen:1"+
+        " -new -newkey rsa:4096 -nodes -keyout "+filename+".key -out "+filename+".crt"))
+        process.exit(1);
+}
+
+async function make_signed(subj, f, ca, san)
+{
+    if (await system(`openssl req -subj '${subj}' ${san ? "-addext 'subjectAltName="+san+"'" : ""} -nodes -new -keyout ${f}.key -out ${f}.csr`))
+        process.exit(1);
+    if (await system(`openssl x509 -req -days 3650 -CA ${ca}.crt -CAkey ${ca}.key -CAcreateserial -in ${f}.csr -out ${f}.crt`))
+        process.exit(1);
+    fs.unlinkSync(f+".csr");
+}

mon/stats.js

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

mon/vitastor_auth_filter.js

@@ -0,0 +1,539 @@
+// AntiEtcd authentication filter for Vitastor
+// (c) Vitaliy Filippov, 2026
+// License: Mozilla Public License 2.0 or Vitastor Network Public License 1.1
+
+// Permissions are based on:
+// 1. Users.
+//    Stored in /vitastor/config/user/<username>.
+//    Has 2 properties:
+//    - type, one of: admin, client.
+//      admin has full access to all images and also to cluster config.
+//      client has r/w access to owned images and r/o access to images with reader_group.
+//    - groups, a list of group names the user is included in.
+// 2. Images.
+//    Stored in /vitastor/config/inode/<pool>/<inode>. Has the following properties:
+//    - owner (user name)
+//    - owner_group (group name)
+//    - reader_group
+// 3. Certificates.
+//    - osd, mon use their own trusted certificates.
+
+const { X509Certificate } = require('node:crypto');
+
+const static_perms = {
+    invalid: {
+        keys: {},
+        prefixes: {},
+    },
+    osd: {
+        keys: { '/pg/config': false },
+        prefixes: { '/config/': false, '/osd/': true, '/pg/state/': true, '/pg/history/': true, '/pgstats/': true },
+    },
+    mon: {
+        keys: { '/pg/config': true, '/stats': true, '/history/last_clean_pgs': true },
+        prefixes: {
+            '/config/': false, '/osd/': false, '/mon/': true, '/pg/history/': true,
+            '/pgstats/': false, '/inode/stats/': true, '/pool/stats/': true,
+        },
+    },
+    admin: {
+        keys: { '/stats': false },
+        prefixes: {
+            '/config/': true, '/osd/': true, '/index/': true, '/pg/history/': true,
+            '/mon/': false, '/pg/': false, '/pgstats/': false, '/inode/stats/': false, '/pool/stats/': false,
+        },
+    },
+    client: {
+        keys: { '/config/global': false, '/config/node_placement': false, '/config/pools': false, '/pg/config': false },
+        prefixes: { '/osd/state/': false, '/pg/state/': false, '/index/maxid/': false },
+    },
+};
+
+const api_perms = {
+    osd: { lease_grant: true, lease_revoke: true, lease_keepalive: true, maintenance_status: true },
+    mon: { lease_grant: true, lease_revoke: true, lease_keepalive: true, maintenance_status: true },
+    admin: { maintenance_status: true },
+    client: { maintenance_status: true },
+};
+
+class VitastorAuthFilter
+{
+    constructor(antietcd)
+    {
+        this.cfg = antietcd.cfg;
+        this.antietcd = antietcd;
+        this.prefix = this.cfg.vitastor_prefix || '/vitastor';
+        this.prefix_parts = this.prefix.split('/');
+    }
+
+    async init()
+    {
+        if (!this.cfg.cert || !this.cfg.key || !this.cfg.osd_ca || !this.cfg.etcd_proxy && !this.cfg.peer_ca || !this.cfg.client_cert_auth)
+        {
+            throw new Error('Authenticated Vitastor setups require enabled client_cert_auth, cert, key'+
+                ' and separate ca (client CA), osd_ca'+(this.cfg.etcd_proxy ? '' : ', peer_ca')+' and optionally mon_ca');
+        }
+        this.osd_ca = await this.antietcd.readPEM(this.cfg.osd_ca);
+        this.osd_ca_obj = new X509Certificate(this.osd_ca);
+        this.antietcd.tls.ca.push(this.osd_ca);
+        if (this.cfg.mon_ca)
+        {
+            this.mon_ca = await this.antietcd.readPEM(this.cfg.mon_ca);
+            this.mon_ca_obj = new X509Certificate(this.mon_ca_obj);
+            this.antietcd.tls.ca.push(this.mon_ca);
+        }
+    }
+
+    init_context(context, clientCert)
+    {
+        let cert = clientCert;
+        while (cert)
+        {
+            if (cert.fingerprint256 == this.osd_ca_obj.fingerprint256)
+            {
+                context.user_type = 'osd';
+                break;
+            }
+            if (this.mon_ca_obj && cert.fingerprint256 == this.mon_ca_obj.fingerprint256)
+            {
+                context.user_type = 'mon';
+                break;
+            }
+            cert = cert.issuerCertificate;
+        }
+    }
+
+    _get(path, decode)
+    {
+        let cur = this.antietcd.etctree.state;
+        path = path instanceof Array ? path : path.split('/');
+        for (const p of path)
+        {
+            if (!cur.children)
+            {
+                return null;
+            }
+            cur = cur.children[p];
+            if (!cur)
+            {
+                return null;
+            }
+        }
+        if (decode)
+        {
+            return this._decode(path, cur.value);
+        }
+        return cur;
+    }
+
+    _decode(path, cur)
+    {
+        if (!cur)
+        {
+            return null;
+        }
+        if (cur)
+        {
+            try
+            {
+                cur = JSON.parse(cur);
+            }
+            catch (e)
+            {
+                console.warn('Invalid JSON in '+(path instanceof Array ? path.join('/') : path)+': '+e);
+            }
+        }
+        return cur;
+    }
+
+    // userInfo: { name: string, type: string, perms: static_perms[type], groups: { [string]: true } }
+    _check_compare(check, userInfo, checked)
+    {
+        let key = String(check.key);
+        if (key.substr(0, this.prefix.length) !== this.prefix)
+        {
+            return false;
+        }
+        key = key.substr(this.prefix.length);
+        if (key in userInfo.perms.keys)
+        {
+            return true;
+        }
+        for (const pfx in userInfo.perms.prefixes)
+        {
+            if (key.substr(0, pfx.length) == pfx)
+            {
+                return true;
+            }
+        }
+        if (userInfo.type == 'client')
+        {
+            // Image permissions
+            if (key.substr(0, 14) == '/config/inode/')
+            {
+                // Allowed to check that a key does not exist
+                if (check.target == 'VERSION' && check.version == 0)
+                {
+                    checked['M'+key] = true;
+                    return true;
+                }
+                else if (check.target == 'MOD')
+                {
+                    const data = this._get(check.key);
+                    if (!data || data.mod_revision != check.mod_revision)
+                    {
+                        // Break check to trigger CAS failure
+                        check.mod_revision = '18446744073709551615'; // UINT64_MAX
+                        return true;
+                    }
+                    const inode = this._decode(check.key, data.value);
+                    if (inode && (inode.owner_group && userInfo.groups[inode.owner_group] ||
+                        inode.owner === userInfo.name))
+                    {
+                        checked['M'+key] = true;
+                        return true;
+                    }
+                }
+                return false;
+            }
+            if (key.substr(0, 13) == '/index/image/')
+            {
+                // Allowed to check that a key does not exist
+                if (check.target == 'VERSION' && check.version == 0)
+                {
+                    checked['M'+key] = true;
+                    return true;
+                }
+                else if (check.target == 'MOD')
+                {
+                    let data = this._get(check.key);
+                    if (!data || data.mod_revision != check.mod_revision)
+                    {
+                        // Break check to trigger CAS failure
+                        check.mod_revision = '18446744073709551615'; // UINT64_MAX
+                        return true;
+                    }
+                    data = this._decode(check.key, data.value);
+                    if (data)
+                    {
+                        const inode = this._get([ ...this.prefix_parts, 'config', 'inode', data.pool_id, data.id ], true);
+                        if (inode && (inode.owner_group && userInfo.groups[inode.owner_group] ||
+                            inode.owner === userInfo.name))
+                        {
+                            checked['M'+key] = true;
+                            return true;
+                        }
+                    }
+                }
+                return false;
+            }
+            if (key.substr(0, 13) == '/index/maxid/')
+            {
+                const pool_id = key.substr(13);
+                const pool_cfg = this._get([ ...this.prefix_parts, 'config', 'pools' ], true);
+                if (!pool_cfg || !pool_cfg[pool_id] || !pool_cfg[pool_id].creator_group || !userInfo.groups[pool_cfg[pool_id].creator_group])
+                {
+                    return false;
+                }
+                if (check.target == 'VERSION' && check.version == 0)
+                {
+                    checked['I'+parseInt(key.substr(13))+'_0'] = true;
+                    return true;
+                }
+                else if (check.target == 'MOD')
+                {
+                    const data = this._get(check.key);
+                    if (!data || data.mod_revision != check.mod_revision)
+                    {
+                        // Break check to trigger CAS failure
+                        check.mod_revision = '18446744073709551615'; // UINT64_MAX
+                        return true;
+                    }
+                    checked['I'+parseInt(key.substr(13))+'_'+data.value] = true;
+                    return true;
+                }
+                return false;
+            }
+        }
+        return false;
+    }
+
+    _check_read(kv, userInfo)
+    {
+        let key = String(kv.key);
+        if (key.substr(0, this.prefix.length) !== this.prefix)
+        {
+            return false;
+        }
+        key = key.substr(this.prefix.length);
+        if (key in userInfo.perms.keys)
+        {
+            return true;
+        }
+        for (const pfx in userInfo.perms.prefixes)
+        {
+            if (key.substr(0, pfx.length) == pfx)
+            {
+                return true;
+            }
+        }
+        if (userInfo.type == 'client')
+        {
+            // Image permissions
+            if (key.substr(0, 14) == '/config/inode/')
+            {
+                const inode = this._decode(kv.key, kv.value);
+                if (inode && (inode.reader_group && userInfo.groups[inode.reader_group] ||
+                    inode.owner_group && userInfo.groups[inode.owner_group] ||
+                    inode.owner === userInfo.name))
+                {
+                    return true;
+                }
+                return false;
+            }
+            if (key.substr(0, 13) == '/index/image/')
+            {
+                const data = this._decode(kv.key, kv.value);
+                const inode = this._get([ ...this.prefix_parts, 'config', 'inode', data.pool_id, data.id ], true);
+                if (inode && (inode.reader_group && userInfo.groups[inode.reader_group] ||
+                    inode.owner_group && userInfo.groups[inode.owner_group] ||
+                    inode.owner === userInfo.name))
+                {
+                    return true;
+                }
+                return false;
+            }
+            if (key.substr(0, 13) == '/inode/stats/')
+            {
+                const [ pool_id, id ] = key.substr(13).split('/');
+                const inode = this._get([ ...this.prefix_parts, 'config', 'inode', pool_id, id ], true);
+                if (inode && (inode.reader_group && userInfo.groups[inode.reader_group] ||
+                    inode.owner_group && userInfo.groups[inode.owner_group] ||
+                    inode.owner === userInfo.name))
+                {
+                    return true;
+                }
+                return false;
+            }
+        }
+        return false;
+    }
+
+    _check_write(put, userInfo, checked)
+    {
+        let key = String(put.key);
+        if (key.substr(0, this.prefix.length) !== this.prefix)
+        {
+            return false;
+        }
+        key = key.substr(this.prefix.length);
+        if (userInfo.perms.keys[key])
+        {
+            return true;
+        }
+        for (const pfx in userInfo.perms.prefixes)
+        {
+            if (userInfo.perms.prefixes[pfx] && key.substr(0, pfx.length) == pfx)
+            {
+                return true;
+            }
+        }
+        if (checked && userInfo.type == 'client')
+        {
+            if (key.substr(0, 13) == '/index/maxid/' &&
+                checked['I'+parseInt(key.substr(13))+'_'+(put.value-1)])
+            {
+                // Allowed to increment maxid
+                return true;
+            }
+            if (checked['M'+key])
+            {
+                // Allowed to modify known images with CAS checks
+                return true;
+            }
+        }
+        return false;
+    }
+
+    _check_req(req, userInfo, checked)
+    {
+        let r;
+        if ((r = (req.request_range || req.requestRange)))
+        {
+            // All range queries are allowed, but responses are filtered - it's simpler
+        }
+        else if ((r = (req.request_put || req.requestPut)))
+        {
+            if (!this._check_write(r, userInfo, checked))
+                return false;
+        }
+        else if ((r = (req.request_delete_range || req.requestDeleteRange)))
+        {
+            if (!r.range_end || r.range_end === r.key)
+            {
+                if (!this._check_write({ key: r.key }, userInfo))
+                    return false;
+            }
+            else
+            {
+                // All keys in range must satisfy prefix
+                r.range_end = String(r.range_end);
+                if (r.key.length != r.range_end.length ||
+                    r.key[r.key.length-1] != '/' ||
+                    r.range_end[r.range_end.length-1] != '0')
+                {
+                    return false;
+                }
+                let key = r.key.substr(this.prefix.length);
+                let found = false;
+                for (const pfx in userInfo.perms.prefixes)
+                {
+                    if (userInfo.perms.prefixes[pfx] && key.substr(0, pfx.length) == pfx)
+                    {
+                        found = true;
+                        break;
+                    }
+                }
+                if (!found)
+                    return false;
+            }
+        }
+        return true;
+    }
+
+    _get_user(context)
+    {
+        if (context.user_type === 'osd' || context.user_type === 'mon')
+        {
+            return {
+                name: context.user_type,
+                type: context.user_type,
+                perms: static_perms[context.user_type],
+            };
+        }
+        if (!context.username)
+        {
+            return {};
+        }
+        let userInfo = this._get([ ...this.prefix_parts, 'config', 'user', context.username ], true);
+        if (!userInfo)
+        {
+            userInfo = { type: 'client' };
+        }
+        else if (userInfo.type !== 'client' && userInfo.type !== 'admin')
+        {
+            userInfo.type = 'client';
+        }
+        userInfo.perms = static_perms[userInfo.type] || static_perms['invalid'];
+        userInfo.name = context.username;
+        if (userInfo.groups instanceof Array)
+        {
+            userInfo.groups = userInfo.groups.reduce((a, c) => { a[c] = true; return a; }, {});
+        }
+        else
+        {
+            userInfo.groups = {};
+        }
+        return userInfo;
+    }
+
+    filter_api(context, api/*, data*/)
+    {
+        let type = 'client';
+        if (context.user_type === 'osd' || context.user_type === 'mon')
+        {
+            type = context.user_type;
+        }
+        else if (context.username)
+        {
+            const userInfo = this._get([ ...this.prefix_parts, 'config', 'user', context.username ], true);
+            if (userInfo && userInfo.type === 'admin')
+            {
+                type = 'admin';
+            }
+        }
+        return api_perms[type] && api_perms[type][api];
+    }
+
+    filter_txn(context, txn)
+    {
+        const userInfo = this._get_user(context);
+        if (!userInfo)
+        {
+            return null;
+        }
+        const checked = {};
+        if (txn.compare)
+        {
+            for (const check of txn.compare)
+            {
+                if (!this._check_compare(check, userInfo, checked))
+                    return null;
+            }
+        }
+        // Special transactions:
+        // 1. create image: create config/inode and index/image, increment index/maxid/<pool> (with CAS)
+        // 2. create snapshot: same as create image but also rename previous to @snap
+        if (txn.success)
+        {
+            for (const req of txn.success)
+            {
+                if (!this._check_req(req, userInfo, checked))
+                    return null;
+            }
+        }
+        if (txn.failure)
+        {
+            for (const req of txn.failure)
+            {
+                if (!this._check_req(req, userInfo, null))
+                    return null;
+            }
+        }
+        return txn;
+    }
+
+    filter_txn_response(context, txn, res)
+    {
+        if (!res.responses)
+        {
+            return;
+        }
+        const userInfo = this._get_user(context);
+        if (!userInfo)
+        {
+            for (const resp of res.responses)
+            {
+                if (resp.response_range && resp.response_range.kvs)
+                {
+                    resp.response_range.kvs = [];
+                }
+            }
+            return;
+        }
+        for (const resp of res.responses)
+        {
+            if (resp.response_range && resp.response_range.kvs)
+            {
+                resp.response_range.kvs = resp.response_range.kvs.filter(kv => this._check_read(kv, userInfo));
+            }
+        }
+    }
+
+    filter_watch_message(context, msg)
+    {
+        if (!msg.result || !msg.result.events)
+        {
+            return;
+        }
+        const userInfo = this._get_user(context);
+        if (!userInfo)
+        {
+            msg.result.events = [];
+            return;
+        }
+        msg.result.events = msg.result.events.filter(ev => this._check_read(ev.kv, userInfo));
+    }
+}
+
+module.exports = VitastorAuthFilter;

node-binding/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vitastor",
-  "version": "3.0.9",
+  "version": "3.0.12",
   "description": "Low-level native bindings to Vitastor client library",
   "main": "index.js",
   "keywords": [

patches/cinder-vitastor.py

@@ -50,7 +50,7 @@ from cinder.volume import configuration
 from cinder.volume import driver
 from cinder.volume import volume_utils
 
-VITASTOR_VERSION = '3.0.9'
+VITASTOR_VERSION = '3.0.12'
 
 LOG = logging.getLogger(__name__)
 

rpm/vitastor-el10.Dockerfile

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

rpm/vitastor-el10.spec

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        3.0.9
+Version:        3.0.12
 Release:        1%{?dist}
 Summary:        Vitastor, a fast software-defined clustered block storage
 
 License:        Vitastor Network Public License 1.1
 URL:            https://vitastor.io/
-Source0:        vitastor-3.0.9.el10.tar.gz
+Source0:        vitastor-3.0.12.el10.tar.gz
 
 BuildRequires:  gperftools-devel
 BuildRequires:  gcc-c++
@@ -16,6 +16,7 @@ BuildRequires:  gf-complete-devel
 BuildRequires:  rdma-core-devel
 BuildRequires:  cmake
 BuildRequires:  libnl3-devel
+BuildRequires:  c-ares-devel
 Requires:       vitastor-osd = %{version}-%{release}
 Requires:       vitastor-mon = %{version}-%{release}
 Requires:       vitastor-client = %{version}-%{release}

rpm/vitastor-el7.Dockerfile

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

rpm/vitastor-el7.spec

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        3.0.9
+Version:        3.0.12
 Release:        1%{?dist}
 Summary:        Vitastor, a fast software-defined clustered block storage
 
 License:        Vitastor Network Public License 1.1
 URL:            https://vitastor.io/
-Source0:        vitastor-3.0.9.el7.tar.gz
+Source0:        vitastor-3.0.12.el7.tar.gz
 
 BuildRequires:  gperftools-devel
 BuildRequires:  devtoolset-9-gcc-c++
@@ -17,6 +17,7 @@ BuildRequires:  gf-complete-devel
 BuildRequires:  rdma-core-devel
 BuildRequires:  cmake3
 BuildRequires:  libnl3-devel
+BuildRequires:  c-ares-devel
 Requires:       vitastor-osd = %{version}-%{release}
 Requires:       vitastor-mon = %{version}-%{release}
 Requires:       vitastor-client = %{version}-%{release}

rpm/vitastor-el8.Dockerfile

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

rpm/vitastor-el8.spec

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        3.0.9
+Version:        3.0.12
 Release:        1%{?dist}
 Summary:        Vitastor, a fast software-defined clustered block storage
 
 License:        Vitastor Network Public License 1.1
 URL:            https://vitastor.io/
-Source0:        vitastor-3.0.9.el8.tar.gz
+Source0:        vitastor-3.0.12.el8.tar.gz
 
 BuildRequires:  gperftools-devel
 BuildRequires:  gcc-toolset-9-gcc-c++
@@ -16,6 +16,7 @@ BuildRequires:  gf-complete-devel
 BuildRequires:  rdma-core-devel
 BuildRequires:  cmake
 BuildRequires:  libnl3-devel
+BuildRequires:  c-ares-devel
 Requires:       vitastor-osd = %{version}-%{release}
 Requires:       vitastor-mon = %{version}-%{release}
 Requires:       vitastor-client = %{version}-%{release}

rpm/vitastor-el9.Dockerfile

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

rpm/vitastor-el9.spec

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        3.0.9
+Version:        3.0.12
 Release:        1%{?dist}
 Summary:        Vitastor, a fast software-defined clustered block storage
 
 License:        Vitastor Network Public License 1.1
 URL:            https://vitastor.io/
-Source0:        vitastor-3.0.9.el9.tar.gz
+Source0:        vitastor-3.0.12.el9.tar.gz
 
 BuildRequires:  gperftools-devel
 BuildRequires:  gcc-c++
@@ -16,6 +16,7 @@ BuildRequires:  gf-complete-devel
 BuildRequires:  rdma-core-devel
 BuildRequires:  cmake
 BuildRequires:  libnl3-devel
+BuildRequires:  c-ares-devel
 Requires:       vitastor-osd = %{version}-%{release}
 Requires:       vitastor-mon = %{version}-%{release}
 Requires:       vitastor-client = %{version}-%{release}

src/CMakeLists.txt

@@ -20,7 +20,7 @@ if("${CMAKE_INSTALL_PREFIX}" MATCHES "^/usr/local/?$")
 endif()
 set(ENABLE_COVERAGE false CACHE BOOL "Enable code coverage")
 
-add_definitions(-DVITASTOR_VERSION="3.0.9")
+add_definitions(-DVITASTOR_VERSION="3.0.12")
 add_definitions(-D_GNU_SOURCE -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -Wall -Wno-sign-compare -Wno-comment -Wno-parentheses -Wno-pointer-arith -fdiagnostics-color=always -fno-omit-frame-pointer -fvisibility=hidden -I ${CMAKE_SOURCE_DIR}/src)
 add_link_options(-fno-omit-frame-pointer)
 if (${WITH_ASAN})
@@ -69,11 +69,21 @@ pkg_check_modules(ISAL libisal)
 if (ISAL_LIBRARIES)
 	add_definitions(-DWITH_ISAL)
 endif (ISAL_LIBRARIES)
+pkg_check_modules(ISAL_CRYPTO libisal_crypto)
+if (ISAL_CRYPTO_LIBRARIES)
+	add_definitions(-DWITH_ISAL_CRYPTO)
+endif (ISAL_CRYPTO_LIBRARIES)
 pkg_check_modules(RDMACM librdmacm)
 if (RDMACM_LIBRARIES)
 	add_definitions(-DWITH_RDMACM)
 endif (RDMACM_LIBRARIES)
 
+find_package(OpenSSL REQUIRED)
+add_definitions(-DWITH_OPENSSL)
+
+pkg_check_modules(CARES REQUIRED libcares)
+include_directories(${CARES_INCLUDE_DIRS})
+
 if (${WITH_SYSTEM_LIBURING})
 	pkg_check_modules(LIBURING REQUIRED liburing>=2.10)
 	include_directories(${LIBURING_INCLUDE_DIRS})

src/blockstore/blockstore_disk.cpp

@@ -83,19 +83,24 @@ void blockstore_disk_t::parse_config(std::map<std::string, std::string> & config
     {
         data_csum_type = BLOCKSTORE_CSUM_CRC32C;
     }
+    else if (config["data_csum_type"] == "xxh3_32")
+    {
+        data_csum_type = BLOCKSTORE_CSUM_XXH3_32;
+    }
     else if (config["data_csum_type"] == "" || config["data_csum_type"] == "none")
     {
         data_csum_type = BLOCKSTORE_CSUM_NONE;
     }
     else
     {
-        throw std::runtime_error("data_csum_type="+config["data_csum_type"]+" is unsupported, only \"crc32c\" and \"none\" are supported");
+        throw std::runtime_error("data_csum_type="+config["data_csum_type"]+" is unsupported, only \"crc32c\", \"xxh3_32\" and \"none\" are supported");
     }
     csum_block_size = parse_size(config["csum_block_size"]);
     discard_on_start = config.find("discard_on_start") != config.end() &&
         (config["discard_on_start"] == "true" || config["discard_on_start"] == "1" || config["discard_on_start"] == "yes");
     gc_on_start = config.find("gc_on_start") == config.end() ||
         (config["gc_on_start"] == "true" || config["gc_on_start"] == "1" || config["gc_on_start"] == "yes");
+    skip_double_claim = (config["skip_double_claim"] == "true" || config["skip_double_claim"] == "1" || config["skip_double_claim"] == "yes");
     min_discard_size = parse_size(config["min_discard_size"]);
     if (!min_discard_size)
         min_discard_size = 1024*1024;

src/blockstore/blockstore_disk.h

@@ -16,6 +16,7 @@
 #define BLOCKSTORE_CSUM_NONE 0
 // Lower byte of checksum type is its length
 #define BLOCKSTORE_CSUM_CRC32C 0x104
+#define BLOCKSTORE_CSUM_XXH3_32 0x204
 
 #define MOCK_DATA_FD 1000
 #define MOCK_META_FD 1001
@@ -59,6 +60,8 @@ struct blockstore_disk_t
     bool discard_on_start = false;
     // GC on start (new store)
     bool gc_on_start = true;
+    // Skip double claim conflicts on start (new store, temporary until the bug is found)
+    bool skip_double_claim = false;
     uint64_t min_discard_size = 1024*1024;
     uint64_t discard_granularity = 0;
 

src/blockstore/blockstore_flush.cpp

@@ -174,14 +174,18 @@ bool journal_flusher_co::loop()
     else if (wait_state == 19) goto resume_19;
     else if (wait_state == 20) goto resume_20;
     else if (wait_state == 21) goto resume_21;
+    else if (wait_state == 22) goto resume_22;
+    else if (wait_state == 23) goto resume_23;
+    else if (wait_state == 24) goto resume_24;
+    else if (wait_state == 25) goto resume_25;
 resume_0:
     wait_state = 0;
     wait_count = 0;
     cur_oid = {};
     res = bs->heap->get_next_compact(cur_oid);
-    // Advance fsynced_lsn every <journal_trim_interval> intent writes
     if ((bs->intent_write_counter >= bs->journal_trim_interval) && co_id == 0)
     {
+        // Advance fsynced_lsn every <journal_trim_interval> intent writes
         bs->intent_write_counter = 0;
 resume_17:
 resume_18:
@@ -196,6 +200,7 @@ resume_21:
     if (res == ENOENT && flusher->force_start > 0 && co_id == 0 &&
         (!bs->dsk.disable_journal_fsync || !bs->dsk.disable_meta_fsync || !bs->dsk.disable_data_fsync))
     {
+        // When under pressure, do an additional fsync to force entries to be marked compactable
         flusher->active_flushers++;
 resume_14:
 resume_15:
@@ -259,11 +264,9 @@ resume_1:
         if (wr->type() == BS_HEAP_SMALL_WRITE ||
             wr->type() == BS_HEAP_INTENT_WRITE && bs->dsk.csum_block_size > bs->dsk.bitmap_granularity)
         {
-            auto res = bs->prepare_read(read_vec, cur_obj, wr, 0, bs->dsk.data_block_size,
+            bs->prepare_read(read_vec, cur_obj, wr, 0, bs->dsk.data_block_size,
                 wr->type() == BS_HEAP_INTENT_WRITE && bs->dsk.csum_block_size > bs->dsk.bitmap_granularity && !bs->perfect_csum_update
                 ? COPY_BUF_SKIP_CSUM : 0);
-            if (res > 0)
-                copy_count++;
         }
     });
     if (!compact_info.compact_lsn)
@@ -273,6 +276,25 @@ resume_1:
         bs->heap->unlock_entry(cur_oid);
         goto resume_0;
     }
+    flusher->active_flushers++;
+    for (i = 0; i < read_vec.size(); i++)
+    {
+        if ((read_vec[i].copy_flags & COPY_BUF_JOURNAL) &&
+            !(read_vec[i].copy_flags & COPY_BUF_COALESCED))
+        {
+            copy_count++;
+        }
+    }
+    if (copy_count > 0 && !bs->dsk.disable_data_fsync)
+    {
+        init_fsync_data();
+    }
+    if (bs->log_level > 10)
+    {
+        printf("Compacting %jx:%jx v%ju..v%ju / l%ju..l%ju (%d writes)\n", cur_oid.inode, cur_oid.stripe,
+            compact_info.clean_wr->version, compact_info.compact_version,
+            compact_info.clean_wr->lsn, compact_info.compact_lsn, copy_count);
+    }
     mem_or(new_bmp, compact_info.clean_wr->get_int_bitmap(bs->heap), bs->dsk.clean_entry_bitmap_size);
     if (!bitmap_copied)
     {
@@ -291,13 +313,6 @@ resume_1:
         csum_copy.clear();
     }
     clean_loc = compact_info.clean_wr->big_location(bs->heap);
-    flusher->active_flushers++;
-    if (bs->log_level > 10)
-    {
-        printf("Compacting %jx:%jx v%ju..v%ju / l%ju..l%ju (%d writes)\n", cur_oid.inode, cur_oid.stripe,
-            compact_info.clean_wr->version, compact_info.compact_version,
-            compact_info.clean_wr->lsn, compact_info.compact_lsn, copy_count);
-    }
     overwrite_start = overwrite_end = 0;
     if (read_vec.size() > 0)
     {
@@ -336,6 +351,13 @@ resume_3:
     if (res == ENOENT || res == EDOM)
     {
         // Abort compaction
+abort_compact:
+        if (copy_count > 0 && !bs->dsk.disable_data_fsync)
+        {
+            cur_sync->member_count--;
+            if (cur_sync->member_count > 0)
+                bs->ringloop->wakeup();
+        }
         flusher->flushing.erase(cur_oid);
         bs->heap->unlock_entry(cur_oid);
         flusher->active_flushers--;
@@ -349,10 +371,7 @@ resume_4:
         if (res == ENOENT)
         {
             // Abort compaction
-            flusher->flushing.erase(cur_oid);
-            bs->heap->unlock_entry(cur_oid);
-            flusher->active_flushers--;
-            goto resume_0;
+            goto abort_compact;
         }
         if (res == EAGAIN)
         {
@@ -381,14 +400,14 @@ resume_9:
     for (i = 0; i < read_vec.size(); i++)
     {
         if ((read_vec[i].copy_flags & COPY_BUF_JOURNAL) &&
-            !(read_vec[i].copy_flags & COPY_BUF_COALESCED) ||
-            (read_vec[i].copy_flags & COPY_BUF_PADDED)) // FIXME Shit, simplify these flags
+            !(read_vec[i].copy_flags & COPY_BUF_COALESCED))
         {
             assert(read_vec[i].buf);
             await_sqe(10);
             data->iov = (struct iovec){ read_vec[i].buf + (read_vec[i].copy_flags & COPY_BUF_PADDED
                 ? read_vec[i].offset - read_vec[i].disk_offset : 0), (size_t)read_vec[i].len };
             data->callback = simple_callback_w;
+            assert(clean_loc + read_vec[i].offset + data->iov.iov_len <= bs->dsk.block_count*bs->dsk.data_block_size);
             io_uring_prep_writev(sqe, bs->dsk.data_fd, &data->iov, 1, bs->dsk.data_offset + clean_loc + read_vec[i].offset);
             wait_count++;
         }
@@ -399,6 +418,17 @@ resume_11:
         wait_state = 11;
         return false;
     }
+    if (copy_count > 0 && !bs->dsk.disable_data_fsync)
+    {
+resume_22:
+resume_23:
+resume_24:
+resume_25:
+        if (!fsync_data(22))
+        {
+            return false;
+        }
+    }
     // Lock is only needed to prevent freeing the big_write because we overwrite it...
     bs->heap->unlock_entry(cur_oid);
     // Mark the object compacted, but don't free and remove small_writes
@@ -408,12 +438,14 @@ resume_11:
     if (!cur_obj)
     {
         // Abort compaction
+        flusher->active_flushers--;
         flusher->flushing.erase(cur_oid);
         goto resume_0;
     }
     if (!calc_block_checksums())
     {
         // Abort compaction
+        flusher->active_flushers--;
         flusher->flushing.erase(cur_oid);
         goto resume_0;
     }
@@ -422,6 +454,7 @@ resume_11:
     if (res == EBUSY)
     {
         // Abort compaction, object is already overwritten by something else
+        flusher->active_flushers--;
         flusher->flushing.erase(cur_oid);
         goto resume_0;
     }
@@ -586,7 +619,7 @@ int journal_flusher_co::check_and_punch_checksums()
             bs->heap->calc_block_checksums((uint32_t*)(new_csums+csum_off), vec.buf, punch_bmp, vec.offset, vec.offset+vec.len, true, NULL);
         }
     }
-    // Modified, we should add_punch_holes and then write the block to disk
+    // Modified, we should punch_holes and then write the block to disk
     return EBUSY;
 }
 
@@ -699,6 +732,67 @@ resume_1:
     return true;
 }
 
+void journal_flusher_co::init_fsync_data()
+{
+    cur_sync = flusher->data_syncs.begin();
+    if (cur_sync == flusher->data_syncs.end() || cur_sync->ready_count > 0)
+    {
+        cur_sync = flusher->data_syncs.emplace(cur_sync);
+    }
+    cur_sync->member_count++;
+}
+
+bool journal_flusher_co::fsync_data(int wait_base)
+{
+    if (wait_state == wait_base)
+        goto resume_0;
+    else if (wait_state == wait_base+1)
+        goto resume_1;
+    else if (wait_state == wait_base+2)
+        goto resume_2;
+    else if (wait_state == wait_base+3)
+        goto resume_3;
+    cur_sync->ready_count++;
+resume_0:
+    if (cur_sync->ready_count < cur_sync->member_count)
+    {
+        wait_state = wait_base;
+        return false;
+    }
+    if (!cur_sync->sent)
+    {
+        // Sync batch is ready. Do it.
+        await_sqe(1);
+        data->iov = { 0 };
+        data->callback = simple_callback_w;
+        io_uring_prep_fsync(sqe, bs->dsk.data_fd, IORING_FSYNC_DATASYNC);
+        cur_sync->sent = true;
+        wait_count++;
+resume_2:
+        if (wait_count > 0)
+        {
+            wait_state = wait_base+2;
+            return false;
+        }
+        cur_sync->done = true;
+        // Wake up other flushers
+        bs->ringloop->wakeup();
+    }
+resume_3:
+    if (!cur_sync->done)
+    {
+        wait_state = wait_base+3;
+        return false;
+    }
+    cur_sync->done_count++;
+    if (cur_sync->done_count >= cur_sync->member_count)
+    {
+        flusher->data_syncs.erase(cur_sync);
+        cur_sync = flusher->data_syncs.end();
+    }
+    return true;
+}
+
 bool journal_flusher_co::fsync_meta(int wait_base)
 {
     if (wait_state == wait_base)        goto resume_0;

src/blockstore/blockstore_flush.h

@@ -25,6 +25,15 @@ struct flusher_meta_write_t
     std::map<uint64_t, meta_sector_t>::iterator it;
 };
 
+struct flusher_data_sync_t
+{
+    int member_count = 0;
+    int ready_count = 0;
+    int done_count = 0;
+    bool sent = false;
+    bool done = false;
+};
+
 class journal_flusher_t;
 
 // Journal flusher coroutine
@@ -58,6 +67,7 @@ class journal_flusher_co
     int i, res;
     bool read_to_fill_incomplete;
     int copy_count;
+    std::list<flusher_data_sync_t>::iterator cur_sync;
 
     friend class journal_flusher_t;
 
@@ -68,6 +78,8 @@ class journal_flusher_co
     bool calc_block_checksums();
     bool write_meta_block(int wait_base);
     bool read_buffered(int wait_base);
+    void init_fsync_data();
+    bool fsync_data(int wait_base);
     bool fsync_meta(int wait_base);
     bool fsync_buffer(int wait_base);
     bool trim_lsn(int wait_base);
@@ -88,6 +100,7 @@ class journal_flusher_t
 
     robin_hood::unordered_flat_set<object_id> flushing;
     int active_flushers = 0;
+    std::list<flusher_data_sync_t> data_syncs;
     int wanting_meta_fsync = 0;
     bool fsyncing_meta = false;
     int syncing_buffer = 0;

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
@@ -28,6 +29,7 @@
 
 #define IMAP_MALLOC_LOW_BITS ((size_t)0x0F)
 #define IMAP_MAX_LOW 16
+#define POSTPONE_INSERT_COUNT 10
 
 #define list_item_overhead(a) (((a) + sizeof(heap_list_item_t) - sizeof(heap_entry_t) + sizeof(void*) + 15) & ~15)
 
@@ -64,19 +66,19 @@ uint32_t blockstore_heap_t::get_simple_entry_size()
 uint32_t blockstore_heap_t::get_big_entry_size()
 {
     return sizeof(heap_big_write_t) + dsk->clean_entry_bitmap_size*2 +
-        (!dsk->data_csum_type ? 0 : dsk->data_block_size/dsk->csum_block_size * (dsk->data_csum_type & 0xFF));
+        (!dsk->csum_block_size ? 0 : dsk->data_block_size/dsk->csum_block_size * (dsk->data_csum_type & 0xFF));
 }
 
 uint32_t blockstore_heap_t::get_big_intent_entry_size()
 {
     return sizeof(heap_big_intent_t) + dsk->clean_entry_bitmap_size*2 +
-        (!dsk->data_csum_type ? 4 : dsk->data_block_size/dsk->csum_block_size * (dsk->data_csum_type & 0xFF));
+        (!dsk->csum_block_size ? 4 : dsk->data_block_size/dsk->csum_block_size * (dsk->data_csum_type & 0xFF));
 }
 
 uint32_t blockstore_heap_t::get_small_entry_size(uint32_t offset, uint32_t len)
 {
     return sizeof(heap_small_write_t) + dsk->clean_entry_bitmap_size +
-        (!dsk->data_csum_type ? 4 : (dsk->data_csum_type & 0xFF) *
+        (!dsk->csum_block_size ? 4 : (dsk->data_csum_type & 0xFF) *
             ((offset+len+dsk->csum_block_size-1)/dsk->csum_block_size - offset/dsk->csum_block_size));
 }
 
@@ -91,7 +93,7 @@ uint32_t blockstore_heap_t::get_csum_size(heap_entry_t *wr)
 
 uint32_t blockstore_heap_t::get_csum_size(uint32_t entry_type, uint32_t offset, uint32_t len)
 {
-    if (!dsk->data_csum_type)
+    if (!dsk->csum_block_size)
     {
         return 0;
     }
@@ -126,26 +128,26 @@ uint32_t heap_entry_t::get_size(blockstore_heap_t *heap)
     return heap->get_simple_entry_size();
 }
 
-bool heap_entry_t::is_overwrite()
+bool heap_entry_t::is_overwrite() const
 {
     return ((entry_type & ~BS_HEAP_GARBAGE) == (BS_HEAP_BIG_WRITE|BS_HEAP_STABLE) ||
         (entry_type & ~BS_HEAP_GARBAGE) == (BS_HEAP_BIG_INTENT|BS_HEAP_STABLE) ||
         (entry_type & ~BS_HEAP_GARBAGE) == (BS_HEAP_DELETE|BS_HEAP_STABLE));
 }
 
-bool heap_entry_t::is_compactable()
+bool heap_entry_t::is_compactable() const
 {
     return !is_overwrite() && (entry_type & BS_HEAP_STABLE) ||
         (entry_type & ~BS_HEAP_GARBAGE) == BS_HEAP_COMMIT ||
         (entry_type & ~BS_HEAP_GARBAGE) == BS_HEAP_ROLLBACK;
 }
 
-bool heap_entry_t::is_before(heap_entry_t *other)
+bool heap_entry_t::is_before(const heap_entry_t *other) const
 {
     return lsn < other->lsn || lsn == other->lsn && !is_overwrite() && other->is_overwrite();
 }
 
-bool heap_entry_t::is_garbage()
+bool heap_entry_t::is_garbage() const
 {
     return (entry_type & BS_HEAP_GARBAGE);
 }
@@ -214,15 +216,24 @@ void heap_entry_t::set_big_location(blockstore_heap_t *heap, uint64_t location)
     big().block_num = location / heap->dsk->data_block_size;
 }
 
-uint32_t heap_entry_t::calc_crc32c()
+uint32_t heap_entry_t::calc_checksum(blockstore_disk_t *dsk)
 {
-    auto old_crc32c = crc32c;
-    crc32c = 0;
-    uint32_t res = ::crc32c(0, (uint8_t*)this, size);
-    crc32c = old_crc32c;
+    auto old_checksum = checksum;
+    checksum = 0;
+    uint32_t res = 0;
+    if (dsk->data_csum_type == BLOCKSTORE_CSUM_XXH3_32)
+        res = (uint32_t)XXH3_64bits(this, size);
+    else
+        res = ::crc32c(0, (uint8_t*)this, size);
+    checksum = old_checksum;
     return res;
 }
 
+uint32_t heap_entry_t::calc_checksum(blockstore_heap_t *heap)
+{
+    return calc_checksum(heap->dsk);
+}
+
 uint64_t blockstore_heap_t::get_pg_id(inode_t inode, uint64_t stripe)
 {
     uint64_t pg_num = 0;
@@ -334,19 +345,12 @@ corrupted_block:
                     block_num, block_offset, wr->size, sizeof(heap_entry_t));
                 goto corrupted_block;
             }
-            if (wr->is_garbage())
-            {
-                // Garbage collection is only performed when writing new entries into the block
-                // because it needs a fake LSN and modified blocks require consecutive modified LSNs
-                // That's why garbage entries may persist on disk
-                if (log_level > 5)
-                {
-                    fprintf(stderr, "Notice: skipping garbage entry %jx:%jx v%ju l%ju in metadata block %u at %u\n",
-                        wr->inode, wr->stripe, wr->version, wr->lsn, block_num, block_offset);
-                }
-                block_offset += wr->size;
-                continue;
-            }
+            // Garbage collection is only performed when writing new entries into the block
+            // because it needs a fake LSN and modified blocks require consecutive modified LSNs
+            // At the same time, further modifications _after_ putting new entries into the block,
+            // but _before_ writing it, may mark some entries in it as garbage. That's why garbage
+            // entries may still be present on disk.
+            wr->entry_type &= ~BS_HEAP_GARBAGE;
             if ((wr->entry_type & BS_HEAP_TYPE) < BS_HEAP_BIG_WRITE ||
                 (wr->entry_type & BS_HEAP_TYPE) > BS_HEAP_ROLLBACK ||
                 (wr->entry_type & ~(BS_HEAP_TYPE|BS_HEAP_STABLE)) ||
@@ -387,12 +391,12 @@ corrupted_object:
                 goto corrupted_object;
             }
             // Verify crc
-            uint32_t expected_crc32c = wr->calc_crc32c();
-            if (wr->crc32c != expected_crc32c)
+            uint32_t expected_checksum = wr->calc_checksum(this);
+            if (wr->checksum != expected_checksum)
             {
-                fprintf(stderr, "Error: entry %jx:%jx v%ju l%ju in metadata block %u at %u is corrupt (crc32c mismatch: expected %08x, got %08x). ",
+                fprintf(stderr, "Error: entry %jx:%jx v%ju l%ju in metadata block %u at %u is corrupt (checksum mismatch: expected %08x, got %08x). ",
                     wr->inode, wr->stripe, wr->version, wr->lsn,
-                    block_num, block_offset, expected_crc32c, wr->crc32c);
+                    block_num, block_offset, expected_checksum, wr->checksum);
                 goto corrupted_object;
             }
             // Verify offset & len
@@ -415,6 +419,13 @@ corrupted_object:
                     wr->inode, wr->stripe, wr->version, wr->big_intent().offset, wr->big_intent().len);
                 goto corrupted_object;
             }
+            if ((wr->type() == BS_HEAP_BIG_INTENT || wr->type() == BS_HEAP_BIG_WRITE) &&
+                wr->big().block_num >= dsk->block_count)
+            {
+                fprintf(stderr, "Error: big_write or big_intent entry %jx:%jx v%ju block_num is too large: %u > %lu. Metadata is incompatible with current parameters. ",
+                    wr->inode, wr->stripe, wr->version, wr->big_intent().block_num, dsk->block_count);
+                goto corrupted_object;
+            }
             handle_write(block_num, wr);
             block_offset += wr->size;
         }
@@ -441,7 +452,7 @@ int blockstore_heap_t::load_blocks(uint64_t disk_offset, uint64_t size, uint8_t
             next_lsn = wr->lsn;
         }
         entries_loaded++;
-        loaded_list_items.push_back(li);
+        insert_list_items(&li, 1, true);
         modify_alloc(block_num, [&](heap_block_info_t & inf)
         {
             if (!inf.entries.size())
@@ -547,18 +558,26 @@ bool blockstore_heap_t::validate_object(heap_entry_t *obj)
 
 void blockstore_heap_t::finish_load()
 {
-    if (loaded_list_items.size())
+    if (postponed_items.size())
     {
-        // Sort everything and load in correct order
-        std::sort(loaded_list_items.begin(), loaded_list_items.end(), [this](const heap_list_item_t* a, const heap_list_item_t* b)
+        // Sort "postponed" items and load in batches
+        std::sort(postponed_items.begin(), postponed_items.end(), [this](const heap_list_item_t* a, const heap_list_item_t* b)
         {
-            return a->entry.lsn < b->entry.lsn;
+            return a->entry.inode < b->entry.inode || a->entry.inode == b->entry.inode &&
+                (a->entry.stripe < b->entry.stripe || a->entry.stripe == b->entry.stripe &&
+                    !a->entry.is_before(&b->entry)); // object ASC, lsn DESC
         });
-        for (auto & li: loaded_list_items)
+        size_t s = 0, e, n = postponed_items.size();
+        for (e = 1; e <= n; e++)
         {
-            insert_list_item(li);
+            if (e >= n || postponed_items[e]->entry.inode != postponed_items[s]->entry.inode &&
+                postponed_items[e]->entry.stripe != postponed_items[s]->entry.stripe)
+            {
+                insert_list_items(postponed_items.data()+s, e-s, false);
+                s = e;
+            }
         }
-        loaded_list_items.clear();
+        postponed_items.clear();
     }
 }
 
@@ -571,26 +590,11 @@ void blockstore_heap_t::fill_recheck_queue()
             inode_map_iterate(ip.second, [&](heap_list_item_t *li)
             {
                 auto obj = &li->entry;
-                // Add object to recheck queue
-                if (obj->type() == BS_HEAP_INTENT_WRITE || obj->type() == BS_HEAP_BIG_INTENT)
+                // Recheck only the latest intent_write (if after completed_lsn) or a series of small_writes
+                if ((obj->type() == BS_HEAP_INTENT_WRITE || obj->type() == BS_HEAP_BIG_INTENT)
+                    && obj->lsn > completed_lsn || obj->type() == BS_HEAP_SMALL_WRITE)
                 {
-                    // Recheck only the latest intent_write
-                    if (obj->lsn > completed_lsn)
-                    {
-                        // Do not recheck if it's already marked as completed in the superblock
-                        recheck_queue.push_back(obj);
-                    }
-                }
-                else
-                {
-                    // Or recheck a series of small_writes
-                    for (auto wr = obj; wr && wr->type() == BS_HEAP_SMALL_WRITE; wr = prev(wr))
-                    {
-                        if (wr->small().len > 0)
-                        {
-                            recheck_queue.push_back(wr);
-                        }
-                    }
+                    recheck_queue.push_back(obj);
                 }
             });
         }
@@ -600,6 +604,11 @@ void blockstore_heap_t::fill_recheck_queue()
 int blockstore_heap_t::mark_used_blocks()
 {
     int res = 0;
+    std::vector<heap_list_item_t*> used_by;
+    if (dsk->skip_double_claim)
+    {
+        used_by.resize(dsk->block_count);
+    }
     for (auto & pgp: block_index)
     {
         for (auto & ip: pgp.second)
@@ -653,10 +662,34 @@ int blockstore_heap_t::mark_used_blocks()
                     {
                         if (is_data_used(wr->big_location(this)))
                         {
-                            fprintf(stderr, "Error: double-claimed data block %u, second time by %jx:%jx l%ju\n",
-                                wr->big().block_num, wr->inode, wr->stripe, wr->lsn);
-                            res = EDOM;
-                            return;
+                            if (dsk->skip_double_claim)
+                            {
+                                // There is a BUG currently:
+                                // Sometimes (under unknown conditions) deletion entries are removed from the disk
+                                // earlier than previous big_writes.
+                                // Until it's fixed, we provide a way to ignore such objects on start.
+                                auto prev_li = used_by[wr->big().block_num];
+                                assert(prev_li);
+                                // Newer LSN must be trusted. Remove the older object.
+                                fprintf(stderr, "Block %u is double-claimed by entries %jx:%jx l%ju and %jx:%jx l%ju\n",
+                                    wr->big().block_num, prev_li->entry.inode, prev_li->entry.stripe, prev_li->entry.lsn, wr->inode, wr->stripe, wr->lsn);
+                                if (init_erase_double_claim(prev_li, li))
+                                {
+                                    return;
+                                }
+                            }
+                            else
+                            {
+                                fprintf(stderr, "Error: double-claimed data block %u, second time by %jx:%jx l%ju\n",
+                                    wr->big().block_num, wr->inode, wr->stripe, wr->lsn);
+                                res = EDOM;
+                                return;
+                            }
+                        }
+                        if (dsk->skip_double_claim)
+                        {
+                            // Record the object which uses the data block
+                            used_by[wr->big().block_num] = li;
                         }
                         use_data(wr->inode, wr->big_location(this));
                     }
@@ -680,6 +713,116 @@ int blockstore_heap_t::mark_used_blocks()
     return res;
 }
 
+void blockstore_heap_t::init_free_bad_entry(heap_entry_t *wr)
+{
+    if (wr->type() == BS_HEAP_SMALL_WRITE)
+    {
+        free_buffer_area(wr->inode, wr->small().location, wr->small().len);
+    }
+    else if (wr->type() == BS_HEAP_BIG_WRITE || wr->type() == BS_HEAP_BIG_INTENT)
+    {
+        free_data(wr->inode, wr->big_location(this));
+    }
+}
+
+void blockstore_heap_t::init_erase_bad_entry(heap_list_item_t *li)
+{
+    modify_alloc(li->block_num, [&](heap_block_info_t & inf)
+    {
+        for (size_t i = 0; i < inf.entries.size(); i++)
+        {
+            if (inf.entries[i] == li)
+            {
+                inf.entries.erase(inf.entries.begin()+i);
+                break;
+            }
+        }
+        inf.used_space -= li->entry.size;
+        inf.garbage_space -= (li->entry.is_garbage() ? li->entry.size : 0);
+    });
+    recheck_modified_blocks.insert(li->block_num);
+    unlink_list_item(li);
+}
+
+bool blockstore_heap_t::init_erase_double_claim(heap_list_item_t *prev_li, heap_list_item_t *cur_li)
+{
+    bool erase_prev = false;
+    bool erase_cur = false;
+    if (prev_li->entry.lsn < cur_li->entry.lsn)
+    {
+        erase_prev = true;
+        auto latest_li = prev_li;
+        while (latest_li->next)
+        {
+            latest_li = latest_li->next;
+        }
+        if (latest_li->entry.lsn >= cur_li->entry.lsn)
+        {
+            // LSN ranges intersect, erase both
+            erase_cur = true;
+        }
+    }
+    else
+    {
+        erase_cur = true;
+        auto latest_li = cur_li;
+        while (latest_li->next)
+        {
+            latest_li = latest_li->next;
+        }
+        if ((latest_li->entry.inode != prev_li->entry.inode ||
+            latest_li->entry.stripe != prev_li->entry.stripe) &&
+            latest_li->entry.lsn >= prev_li->entry.lsn)
+        {
+            // LSN ranges intersect, erase both
+            erase_prev = true;
+        }
+    }
+    if (erase_prev)
+    {
+        fprintf(stderr, "Erasing object %jx:%jx due to double-claim\n", prev_li->entry.inode, prev_li->entry.stripe);
+        auto erase_li = prev_li;
+        while (erase_li->next)
+        {
+            erase_li = erase_li->next;
+        }
+        bool overwritten = false;
+        while (erase_li)
+        {
+            auto prev_erase_li = erase_li->prev;
+            if (!overwritten)
+            {
+                init_free_bad_entry(&erase_li->entry);
+                overwritten = erase_li->entry.is_overwrite();
+            }
+            init_erase_bad_entry(erase_li);
+            erase_li = prev_erase_li;
+        }
+    }
+    if (erase_cur)
+    {
+        fprintf(stderr, "Erasing object %jx:%jx due to double-claim\n", cur_li->entry.inode, cur_li->entry.stripe);
+        auto erase_li = cur_li->next;
+        while (erase_li)
+        {
+            // Only newer entries are marked as used
+            auto next_erase_li = erase_li->next;
+            init_free_bad_entry(&erase_li->entry);
+            init_erase_bad_entry(erase_li);
+            erase_li = next_erase_li;
+        }
+        erase_li = cur_li;
+        // Older ones are not
+        while (erase_li)
+        {
+            auto prev_erase_li = erase_li->prev;
+            init_erase_bad_entry(erase_li);
+            erase_li = prev_erase_li;
+        }
+    }
+    return erase_cur;
+}
+
 void blockstore_heap_t::recheck_full_gc()
 {
     uint32_t block_num = 0;
@@ -720,80 +863,98 @@ void blockstore_heap_t::recheck_full_gc()
     }
 }
 
-void blockstore_heap_t::recheck_buffer(heap_entry_t *cwr, uint8_t *buf)
+void blockstore_heap_t::recheck_drop_entries(heap_entry_t *obj, heap_entry_t *bad_wr)
 {
-    auto free_entry = [&](heap_list_item_t *li)
+    // write entry is invalid, erase it and all newer entries
+    int bad_count = 1;
+    for (auto wr = obj; wr && wr != bad_wr; wr = prev(wr))
     {
-        uint32_t block_num = li->block_num;
-        auto wr_size = li->entry.size;
-        if (li->entry.is_garbage())
-        {
-            garbage_entries--;
-            garbage_memory -= list_item_overhead(wr_size);
-        }
-        live_entries--;
-        live_memory -= list_item_overhead(wr_size);
-        free(li);
-        modify_alloc(block_num, [&](heap_block_info_t & inf)
-        {
-            inf.used_space -= wr_size;
-            bool found = false;
-            for (auto it = inf.entries.begin(); it != inf.entries.end(); it++)
-            {
-                if (*it == li)
-                {
-                    found = true;
-                    inf.entries.erase(it);
-                    break;
-                }
-            }
-            assert(found);
-        });
-        recheck_modified_blocks.insert(block_num);
-    };
-    if (cwr->is_garbage())
-    {
-        // already freed after rechecking one of the previous small_write entries
-        free_entry(list_item(cwr));
+        bad_count++;
     }
-    else if (!calc_checksums(cwr, buf, false))
+    auto prev_wr = prev(bad_wr);
+    if (prev_wr)
     {
-        // write entry is invalid, erase it and mark newer entries with garbage bit
-        auto & pg_idx = block_index[get_pg_id(cwr->inode, cwr->stripe)];
-        auto & inode_idx = pg_idx[cwr->inode];
-        heap_inode_map_t::iterator li_it;
-        heap_list_item_t *li = NULL;
-        inode_map_get(inode_idx, li_it, li, cwr->stripe);
-        int rolled_back = 1;
-        while (li && cwr != &li->entry)
+        fprintf(stderr, "Notice: %u unfinished %s to %jx:%jx v%ju since good lsn %ju, rolling back\n",
+            bad_count, bad_count > 1 ? "writes" : "write", obj->inode, obj->stripe, obj->version, prev_wr->lsn);
+    }
+    else
+    {
+        fprintf(stderr, "Notice: the whole object %jx:%jx only has unfinished writes, rolling back\n", obj->inode, obj->stripe);
+    }
+    auto li = list_item(obj);
+    while (li && prev_wr != &li->entry)
+    {
+        auto prev = li->prev;
+        assert(li->entry.type() == bad_wr->type());
+        init_erase_bad_entry(li);
+        li = prev;
+    }
+}
+
+void blockstore_heap_t::recheck_start_reads(heap_recheck_state_t *st)
+{
+    if (st->sent_reads >= st->total_reads)
+        return;
+    while (recheck_in_progress < recheck_queue_depth)
+    {
+        auto wr = st->next_wr;
+        st->next_wr = prev(st->next_wr);
+        uint64_t loc = 0, len = 0;
+        bool from_data = false;
+        if (wr->type() == BS_HEAP_SMALL_WRITE)
         {
-            assert(li->entry.entry_type == cwr->entry_type);
-            auto prev = li->prev;
-            li->next = li->prev = NULL;
-            if (!li->entry.is_garbage())
-            {
-                garbage_entries++;
-                garbage_memory += list_item_overhead(li->entry.size);
-                li->entry.set_garbage();
-            }
-            li = prev;
-            rolled_back++;
+            loc = wr->small().location;
+            len = wr->small().len;
         }
-        assert(li);
-        if (li->prev)
+        else if (wr->type() == BS_HEAP_BIG_INTENT)
         {
-            fprintf(stderr, "Notice: %u unfinished %s to %jx:%jx v%ju since lsn %ju, rolling back\n",
-                rolled_back, rolled_back > 1 ? "writes" : "write", cwr->inode, cwr->stripe, li->prev->entry.version, li->entry.lsn);
-            inode_map_replace(inode_idx, li_it, li->prev);
-            li->prev->next = NULL;
+            auto & bi = wr->big_intent();
+            loc = (uint64_t)bi.block_num * dsk->data_block_size + bi.offset;
+            len = bi.len;
+            from_data = true;
         }
         else
         {
-            fprintf(stderr, "Notice: the whole object %jx:%jx only has unfinished writes, rolling back\n",
-                cwr->inode, cwr->stripe);
-            inode_map_erase(pg_idx, inode_idx, li_it, li);
+            assert(wr->type() == BS_HEAP_INTENT_WRITE);
+            auto prev_wr = prev(wr);
+            while (prev_wr && prev_wr->entry_type == wr->entry_type)
+            {
+                // Skip other intent_writes
+                prev_wr = prev(prev_wr);
+            }
+            if (!prev_wr || prev_wr->entry_type != (BS_HEAP_BIG_WRITE | (wr->entry_type & BS_HEAP_STABLE)) &&
+                prev_wr->entry_type != (BS_HEAP_BIG_INTENT | (wr->entry_type & BS_HEAP_STABLE)))
+            {
+                fprintf(stderr, "Error: intent_write entry %jx:%jx v%ju l%ju is not written over a big_write\n",
+                    wr->inode, wr->stripe, wr->version, wr->lsn);
+                exit(1);
+            }
+            loc = wr->small().offset + prev_wr->big_location(this);
+            len = wr->small().len;
+            from_data = true;
         }
-        free_entry(li);
+        uint8_t *buf = (uint8_t*)memalign_or_die(MEM_ALIGNMENT, len);
+        st->sent_reads++;
+        recheck_in_progress++;
+        recheck_pending_reads--;
+        bool is_last = st->sent_reads >= st->total_reads;
+        recheck_cb(from_data, loc, len, buf, [this, st, wr, buf]()
+        {
+            st->checked_reads++;
+            if (!calc_checksums(wr, buf, false))
+                st->bad_wr = !st->bad_wr || st->bad_wr->lsn > wr->lsn ? wr : st->bad_wr;
+            if (st->checked_reads >= st->total_reads)
+            {
+                if (st->bad_wr)
+                    recheck_drop_entries(st->obj, st->bad_wr);
+                recheck_states.erase(st->obj);
+            }
+            free(buf);
+            recheck_in_progress--;
+            recheck_small_writes(NULL, 0);
+        });
+        if (is_last)
+            break;
     }
 }
 
@@ -816,70 +977,47 @@ bool blockstore_heap_t::recheck_small_writes(std::function<void(bool is_data, ui
         recheck_queue_depth = queue_depth;
     }
     in_recheck = true;
+    while (recheck_pending_reads > 0 && recheck_in_progress < recheck_queue_depth)
+    {
+        for (auto & sp: recheck_states)
+            recheck_start_reads(&sp.second);
+    }
     while (recheck_queue.size() > 0 && recheck_in_progress < recheck_queue_depth)
     {
-        heap_entry_t *wr = recheck_queue.front();
+        heap_entry_t *obj = recheck_queue.front();
         recheck_queue.pop_front();
-        bool from_data = false;
-        uint64_t loc = 0;
-        uint32_t len = 0;
-        if (wr->type() == BS_HEAP_INTENT_WRITE)
+        if (obj->type() == BS_HEAP_SMALL_WRITE && buffer_area)
         {
-            auto prev_wr = prev(wr);
-            while (prev_wr && prev_wr->entry_type == wr->entry_type)
+            // Check this object synchronously
+            heap_entry_t *bad_wr = NULL;
+            for (auto wr = obj; wr && wr->type() == BS_HEAP_SMALL_WRITE; wr = prev(wr))
             {
-                // Skip other intent_writes
-                prev_wr = prev(prev_wr);
+                fprintf(stderr, "Notice: rechecking %jx:%jx l%ju - %u bytes at %ju in buffer area\n",
+                    wr->inode, wr->stripe, wr->lsn, wr->small().len, wr->small().location);
+                if (!calc_checksums(wr, buffer_area + wr->small().location, false))
+                    bad_wr = wr;
             }
-            if (!prev_wr || prev_wr->entry_type != (BS_HEAP_BIG_WRITE | (wr->entry_type & BS_HEAP_STABLE)) &&
-                prev_wr->entry_type != (BS_HEAP_BIG_INTENT | (wr->entry_type & BS_HEAP_STABLE)))
-            {
-                fprintf(stderr, "Error: intent_write entry %jx:%jx v%ju l%ju is not written over a big_write\n",
-                    wr->inode, wr->stripe, wr->version, wr->lsn);
-                exit(1);
-            }
-            loc = wr->small().offset + prev_wr->big_location(this);
-            len = wr->small().len;
-            from_data = true;
-        }
-        else if (wr->type() == BS_HEAP_BIG_INTENT)
-        {
-            auto & bi = wr->big_intent();
-            loc = (uint64_t)bi.block_num * dsk->data_block_size + bi.offset;
-            len = bi.len;
-            from_data = true;
+            if (bad_wr)
+                recheck_drop_entries(obj, bad_wr);
         }
         else
         {
-            assert(wr->type() == BS_HEAP_SMALL_WRITE);
-            loc = wr->small().location;
-            len = wr->small().len;
-        }
-        if (log_level > 5)
-        {
-            fprintf(stderr, "Notice: rechecking %jx:%jx l%ju - %u bytes at %ju in %s area\n",
-                wr->inode, wr->stripe, wr->lsn, len, loc, from_data ? "data" : "buffer");
-        }
-        if (!from_data && buffer_area)
-        {
-            recheck_buffer(wr, buffer_area+loc);
-        }
-        else
-        {
-            recheck_in_progress++;
-            uint8_t *buf = (uint8_t*)memalign_or_die(MEM_ALIGNMENT, len);
-            recheck_cb(from_data, loc, len, buf, [this, wr, buf]()
-            {
-                recheck_buffer(wr, buf);
-                free(buf);
-                recheck_in_progress--;
-                recheck_small_writes(NULL, 0);
-            });
+            // Recheck will be asynchronous. Create state and start it
+            auto & st = recheck_states[obj];
+            st.obj = obj;
+            st.next_wr = obj;
+            st.total_reads = 1;
+            if (obj->type() == BS_HEAP_SMALL_WRITE)
+                for (auto wr = prev(obj); wr && wr->type() == BS_HEAP_SMALL_WRITE; wr = prev(wr))
+                    st.total_reads++;
+            recheck_pending_reads += st.total_reads;
+            recheck_start_reads(&st);
         }
     }
     in_recheck = false;
     if (!recheck_queue.size() && !recheck_in_progress)
     {
+        assert(!recheck_states.size());
         auto cb = std::move(recheck_cb);
         recheck_queue_depth = 0;
         if (cb)
@@ -941,7 +1079,11 @@ 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 +1133,26 @@ 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 +1169,89 @@ 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;
 }
 
@@ -1335,43 +1539,51 @@ int blockstore_heap_t::allocate_entry(uint32_t entry_size, uint32_t *block_num,
     return 0;
 }
 
-void blockstore_heap_t::insert_list_item(heap_list_item_t *li)
+void blockstore_heap_t::insert_list_items(heap_list_item_t** v, size_t count, bool postpone)
 {
-    auto & inode_idx = block_index[get_pg_id(li->entry.inode, li->entry.stripe)][li->entry.inode];
+    auto wr = &v[0]->entry;
+    auto & inode_idx = block_index[get_pg_id(wr->inode, wr->stripe)][wr->inode];
     heap_inode_map_t::iterator li_it;
     heap_list_item_t *old_head = NULL;
     if (inode_idx)
-        inode_map_get(inode_idx, li_it, old_head, li->entry.stripe);
-    if (old_head && !old_head->entry.is_before(&li->entry))
+        inode_map_get(inode_idx, li_it, old_head, wr->stripe);
+    heap_list_item_t *next_li = NULL;
+    heap_list_item_t *prev_li = old_head;
+    int skips = 0;
+    // Merge entry array and inode_idx linked list (both sorted in newest first order)
+    for (size_t i = 0; i < count; i++)
     {
         // BIG_WRITE may be inserted into the middle of the sequence during compaction
         // and it overrides SMALL_WRITEs and COMMITs with the same LSN
         // However, all entries of other types (say DELETE) override previous ones
-        auto next_li = old_head;
-        auto prev_li = old_head->prev;
+        auto li = v[i];
         while (prev_li && !prev_li->entry.is_before(&li->entry))
         {
             next_li = prev_li;
             prev_li = prev_li->prev;
+            skips++;
+        }
+        if (postpone && skips > POSTPONE_INSERT_COUNT)
+        {
+            postponed_items.push_back(li);
+            return;
+        }
+        if (next_li == NULL)
+        {
+            // Replace the latest entry pointer
+            if (old_head)
+                inode_map_replace(inode_idx, li_it, li);
+            else
+                inode_map_put(inode_idx, li);
         }
         // Insert <li> between <next_li> and <prev_li>
+        li->next = next_li;
+        if (next_li)
+            next_li->prev = li;
         li->prev = prev_li;
         if (prev_li)
             prev_li->next = li;
-        next_li->prev = li;
-        li->next = next_li;
-    }
-    else
-    {
-        li->prev = old_head;
-        li->next = NULL;
-        if (old_head)
-        {
-            old_head->next = li;
-            inode_map_replace(inode_idx, li_it, li);
-        }
-        else
-            inode_map_put(inode_idx, li);
+        next_li = li;
     }
 }
 
@@ -1403,10 +1615,10 @@ int blockstore_heap_t::add_entry(uint32_t wr_size, uint32_t *modified_block,
         (explicit_complete ? HEAP_INFLIGHT_EXPLICIT : 0) |
         (new_wr->is_overwrite() ? HEAP_INFLIGHT_COMPACTED : 0) |
         (new_wr->is_compactable() ? HEAP_INFLIGHT_COMPACTABLE : 0));
-    insert_list_item(li);
+    insert_list_items(&li, 1, false);
     li->block_num = block_num;
     new_wr->size = wr_size;
-    new_wr->crc32c = new_wr->calc_crc32c();
+    new_wr->checksum = new_wr->calc_checksum(this);
     return 0;
 }
 
@@ -1466,7 +1678,7 @@ 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 +1707,7 @@ 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 +1745,7 @@ 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 +1792,7 @@ 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));
     });
 }
@@ -1599,6 +1811,7 @@ int blockstore_heap_t::punch_holes(heap_entry_t *wr, uint8_t *new_bitmap, uint8_
     *modified_block = block_num;
     memcpy(wr->get_int_bitmap(this), new_bitmap, dsk->clean_entry_bitmap_size);
     memcpy(wr->get_checksums(this), new_csums, dsk->data_block_size/dsk->csum_block_size*(dsk->data_csum_type & 0xFF));
+    wr->checksum = wr->calc_checksum(dsk);
     return 0;
 }
 
@@ -2397,6 +2610,22 @@ void blockstore_heap_t::apply_inflight(heap_inflight_lsn_t & inflight)
 }
 
 void blockstore_heap_t::remove_list_item(heap_list_item_t *li)
+{
+    if (!li->next)
+    {
+        // The last freed entry must be a deletion
+        assert(!li->prev);
+        assert((li->entry.entry_type & ~BS_HEAP_GARBAGE) == (BS_HEAP_DELETE|BS_HEAP_STABLE));
+    }
+    else if (!li->prev && li->next->entry.entry_type == (BS_HEAP_DELETE|BS_HEAP_STABLE))
+    {
+        // free BS_HEAP_DELETEs when all previous entries are also freed
+        mark_garbage(li->next->block_num, &li->next->entry, UINT32_MAX);
+    }
+    unlink_list_item(li);
+}
+
+void blockstore_heap_t::unlink_list_item(heap_list_item_t *li)
 {
     auto prev = li->prev;
     auto next = li->next;
@@ -2406,25 +2635,20 @@ void blockstore_heap_t::remove_list_item(heap_list_item_t *li)
     }
     if (!next)
     {
-        // The last freed entry must be a deletion
-        assert(!prev);
         auto wr = &li->entry;
-        assert(wr->entry_type == BS_HEAP_DELETE|BS_HEAP_STABLE);
         auto & pg_idx = block_index[get_pg_id(wr->inode, wr->stripe)];
         auto & inode_idx = pg_idx[wr->inode];
         heap_inode_map_t::iterator li_it;
         heap_list_item_t *old_li = NULL;
         inode_map_get(inode_idx, li_it, old_li, wr->stripe);
-        inode_map_erase(pg_idx, inode_idx, li_it, old_li);
+        if (!prev)
+            inode_map_erase(pg_idx, inode_idx, li_it, old_li);
+        else
+            inode_map_replace(inode_idx, li_it, prev);
     }
     else
     {
         next->prev = prev;
-        if (!prev && next->entry.entry_type == (BS_HEAP_DELETE|BS_HEAP_STABLE))
-        {
-            // free BS_HEAP_DELETEs when all previous entries are also freed
-            mark_garbage(next->block_num, &next->entry, UINT32_MAX);
-        }
     }
     if (li->entry.is_garbage())
     {

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;
@@ -57,11 +57,11 @@ struct __attribute__((__packed__)) heap_entry_t
     inline heap_small_write_t& small() { return *(heap_small_write_t*)this; }
     inline heap_big_write_t& big() { return *(heap_big_write_t*)this; }
     inline heap_big_intent_t& big_intent() { return *(heap_big_intent_t*)this; }
-    bool is_garbage();
+    bool is_garbage() const;
     void set_garbage();
-    bool is_overwrite();
-    bool is_compactable();
-    bool is_before(heap_entry_t *other);
+    bool is_overwrite() const;
+    bool is_compactable() const;
+    bool is_before(const heap_entry_t *other) const;
     uint32_t get_size(blockstore_heap_t *heap);
     uint8_t *get_ext_bitmap(blockstore_heap_t *heap);
     uint8_t *get_int_bitmap(blockstore_heap_t *heap);
@@ -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
@@ -158,6 +159,16 @@ struct heap_li_equal
     }
 };
 
+struct heap_recheck_state_t
+{
+    heap_entry_t *obj = NULL;
+    heap_entry_t *next_wr = NULL;
+    size_t total_reads = 0;
+    size_t sent_reads = 0;
+    size_t checked_reads = 0;
+    heap_entry_t *bad_wr = NULL;
+};
+
 using i64hash_t = robin_hood::hash<uint64_t>;
 using heap_inode_map_t = robin_hood::unordered_flat_set<heap_list_item_t*, heap_li_hash, heap_li_equal, 88>;
 using heap_block_index_t = robin_hood::unordered_flat_map<uint64_t,
@@ -208,9 +219,11 @@ class blockstore_heap_t
 
     bool marked_used_blocks = false;
     bool recheck_queue_filled = false;
-    std::vector<heap_list_item_t*> loaded_list_items;
+    std::vector<heap_list_item_t*> postponed_items;
     std::set<uint32_t> recheck_modified_blocks;
     std::deque<heap_entry_t*> recheck_queue;
+    std::map<heap_entry_t*, heap_recheck_state_t> recheck_states;
+    size_t recheck_pending_reads = 0;
     int recheck_in_progress = 0;
     bool in_recheck = false;
     std::function<void(bool is_data, uint64_t offset, uint64_t len, uint8_t* buf, std::function<void()>)> recheck_cb;
@@ -219,7 +232,12 @@ class blockstore_heap_t
     uint64_t get_pg_id(inode_t inode, uint64_t stripe);
     bool validate_object(heap_entry_t *obj);
     void fill_recheck_queue();
+    void recheck_drop_entries(heap_entry_t *obj, heap_entry_t *bad_wr);
+    void recheck_start_reads(heap_recheck_state_t *st);
     int mark_used_blocks();
+    void init_free_bad_entry(heap_entry_t *wr);
+    void init_erase_bad_entry(heap_list_item_t *li);
+    bool init_erase_double_claim(heap_list_item_t *prev_li, heap_list_item_t *cur_li);
     void recheck_full_gc();
     void recheck_buffer(heap_entry_t *cwr, uint8_t *buf);
     void defragment_block(uint32_t block_num);
@@ -227,8 +245,9 @@ class blockstore_heap_t
 
     void gc_block(heap_block_info_t & inf);
     int allocate_entry(uint32_t entry_size, uint32_t *block_num, bool allow_last_free);
-    void insert_list_item(heap_list_item_t *li);
+    void insert_list_items(heap_list_item_t** v, size_t count, bool postpone);
     void remove_list_item(heap_list_item_t *li);
+    void unlink_list_item(heap_list_item_t *li);
     int add_entry(uint32_t wr_size, uint32_t *modified_block, bool allow_last_free,
         bool explicit_complete, std::function<void(heap_entry_t *wr)> fill_entry);
     int add_simple(heap_entry_t *obj, uint64_t version, uint32_t *modified_block, uint32_t entry_type);

src/blockstore/blockstore_impl.cpp

@@ -101,6 +101,7 @@ void blockstore_impl_t::loop()
         unsigned initial_ring_space = ringloop->space_left();
         int op_idx = 0, new_idx = 0;
         bool has_unfinished_writes = false;
+        bool has_unfinished_sync = false;
         for (; op_idx < submit_queue.size(); op_idx++, new_idx++)
         {
             auto op = submit_queue[op_idx];
@@ -138,7 +139,13 @@ void blockstore_impl_t::loop()
             else if (op->opcode == BS_OP_SYNC)
             {
                 // syncs only completed writes, so doesn't have to be blocked by anything
-                wr_st = continue_sync(op);
+                if (!has_unfinished_sync)
+                {
+                    wr_st = continue_sync(op);
+                    has_unfinished_sync = (wr_st != 2);
+                }
+                else
+                    wr_st = 0;
             }
             else if (op->opcode == BS_OP_STABLE || op->opcode == BS_OP_ROLLBACK)
             {
@@ -154,9 +161,7 @@ void blockstore_impl_t::loop()
                     wr_st = 2;
                 }
                 else
-                {
                     wr_st = 0;
-                }
             }
             if (wr_st == 2)
             {

src/blockstore/blockstore_init.cpp

@@ -153,6 +153,14 @@ resume_1:
             );
             exit(1);
         }
+        uint32_t csum = hdr->header_csum;
+        hdr->header_csum = 0;
+        if (crc32c(0, hdr, sizeof(*hdr)) != csum)
+        {
+            printf("Metadata header is corrupt (checksum mismatch).\n");
+            exit(1);
+        }
+        hdr->header_csum = csum;
     }
     bs->heap->start_load(((blockstore_meta_header_v3_t *)bs->meta_superblock)->completed_lsn);
     if (bs->dsk.inmemory_journal)
@@ -241,22 +249,6 @@ resume_4:
     // metadata read finished
     bs->heap->finish_load();
     printf("Metadata entries loaded: %ju, rechecking unfinished writes and garbage entries\n", entries_loaded);
-    if (zero_on_init && !bs->dsk.disable_meta_fsync)
-    {
-        GET_SQE();
-        io_uring_prep_fsync(sqe, bs->dsk.meta_fd, IORING_FSYNC_DATASYNC);
-        last_read_offset = 0;
-        data->iov = { 0 };
-        data->callback = [this](ring_data_t *data) { handle_event(data, -1); };
-        submitted++;
-        bs->ringloop->submit();
-    resume_5:
-        if (submitted > 0)
-        {
-            wait_state = 5;
-            return 1;
-        }
-    }
     // asynchronous recheck
 resume_6:
     wait_state = 6;
@@ -311,7 +303,7 @@ resume_8:
         uint32_t block_num = recheck_mod[i];
         uint64_t block_offset = bs->dsk.meta_offset + (uint64_t)(block_num+1) * bs->dsk.meta_block_size;
         data = ((ring_data_t*)sqe->user_data);
-        uint8_t *buf = (uint8_t*)malloc_or_die(bs->dsk.meta_block_size);
+        uint8_t *buf = (uint8_t*)memalign_or_die(MEM_ALIGNMENT, bs->dsk.meta_block_size);
         bs->heap->get_meta_block(block_num, buf);
         data->iov = { buf, bs->dsk.meta_block_size };
         data->callback = [this, buf, block_offset](ring_data_t *data)
@@ -337,6 +329,22 @@ resume_9:
     }
     free(metadata_buffer);
     metadata_buffer = NULL;
+    if (!bs->dsk.disable_meta_fsync && !bs->readonly)
+    {
+        GET_SQE();
+        io_uring_prep_fsync(sqe, bs->dsk.meta_fd, IORING_FSYNC_DATASYNC);
+        last_read_offset = 0;
+        data->iov = { 0 };
+        data->callback = [this](ring_data_t *data) { handle_event(data, -1); };
+        submitted++;
+        bs->ringloop->submit();
+    resume_5:
+        if (submitted > 0)
+        {
+            wait_state = 5;
+            return 1;
+        }
+    }
     printf("Loading finished. Data used: %ju / %ju bytes (%s / %s)\n",
         bs->heap->get_data_used_space(), bs->dsk.block_count * bs->dsk.data_block_size,
         format_size(bs->heap->get_data_used_space()).c_str(),

src/blockstore/blockstore_stable.cpp

@@ -16,6 +16,7 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
     else if (priv->op_state == 5) goto resume_5;
     assert(!priv->op_state);
     op->retval = 0;
+    PRIV(op)->lsn = 0;
     priv->modified_block = priv->modified_block2 = UINT32_MAX;
     for (priv->stab_pos = 0; priv->stab_pos < op->len; priv->stab_pos++)
     {
@@ -36,6 +37,12 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
                 FINISH_OP(op);
                 return 2;
             }
+            if (res == ENOENT)
+            {
+                op->retval = -ENOENT;
+                FINISH_OP(op);
+                return 2;
+            }
             if (res == ENOSPC)
             {
                 if (!heap->get_to_compact_count())

src/blockstore/blockstore_sync.cpp

@@ -9,6 +9,7 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op)
     if (!PRIV(op)->op_state)
     {
         op->retval = 0;
+        PRIV(op)->lsn = 0;
     }
     int res = do_sync(op, 0);
     if (res == 2)
@@ -104,7 +105,8 @@ int blockstore_impl_t::do_sync(blockstore_op_t *op, int base_state)
         unsynced_data_write_count = unsynced_buffer_write_count = unsynced_meta_write_count = 0;
         return 2;
     }
-    PRIV(op)->modified_block = heap->get_completed_lsn();
+    assert(!PRIV(op)->lsn);
+    PRIV(op)->lsn = heap->get_completed_lsn();
     if (!submit_fsyncs(PRIV(op)->pending_ops))
     {
         PRIV(op)->wait_detail = 1;
@@ -118,6 +120,6 @@ resume_1:
         return 1;
     }
 resume_2:
-    heap->mark_lsn_fsynced(PRIV(op)->modified_block);
+    heap->mark_lsn_fsynced(PRIV(op)->lsn);
     return 2;
 }

src/blockstore/blockstore_write.cpp

@@ -37,6 +37,7 @@ void blockstore_impl_t::prepare_meta_block_write(uint32_t modified_block)
         heap->complete_block_write(modified_block);
         ringloop->wakeup();
     };
+    assert(((uint64_t)modified_block+2)*dsk.meta_block_size <= dsk.meta_area_size);
     io_uring_prep_writev(
         sqe, dsk.meta_fd, &data->iov, 1, dsk.meta_offset + ((uint64_t)modified_block+1)*dsk.meta_block_size
     );
@@ -177,6 +178,7 @@ enospc:
         ring_data_t *data = ((ring_data_t*)sqe->user_data);
         data->iov = (struct iovec){ op->buf, op->len };
         data->callback = [this, op](ring_data_t *data) { handle_write_event(data, op); };
+        assert(loc+op->offset+op->len <= dsk.block_count*dsk.data_block_size);
         io_uring_prep_writev(sqe, dsk.data_fd, &data->iov, 1, dsk.data_offset + loc + op->offset);
         PRIV(op)->pending_ops++;
         write_iodepth++;
@@ -264,6 +266,7 @@ enospc:
             BS_SUBMIT_GET_SQE(sqe2, data2);
             data2->iov = (struct iovec){ op->buf, op->len };
             data2->callback = [this, op](ring_data_t *data) { handle_write_event(data, op); };
+            assert(loc+op->len <= dsk.journal_len);
             io_uring_prep_writev(sqe2, dsk.journal_fd, &data2->iov, 1, dsk.journal_offset + loc);
             PRIV(op)->pending_ops++;
         }
@@ -453,6 +456,7 @@ resume_10:
     BS_SUBMIT_GET_SQE(sqe, data);
     data->iov = (struct iovec){ op->buf, op->len };
     data->callback = [this, op](ring_data_t *data) { handle_write_event(data, op); };
+    assert(PRIV(op)->location + op->offset <= dsk.block_count*dsk.data_block_size);
     io_uring_prep_writev(sqe, dsk.data_fd, &data->iov, 1, dsk.data_offset + PRIV(op)->location + op->offset);
     if (dsk.use_atomic_flag)
         sqe->rw_flags = RWF_ATOMIC;

src/blockstore/ondisk_formats.h

@@ -141,7 +141,7 @@ struct __attribute__((__packed__)) journal_entry
 inline uint32_t je_crc32(journal_entry *je)
 {
     // 0x48674bc7 = crc32(4 zero bytes)
-    return crc32c(0x48674bc7, ((uint8_t*)je)+4, je->size-4);
+    return je->size < 4 ? 0 : crc32c(0x48674bc7, ((uint8_t*)je)+4, je->size-4);
 }
 
 // "VITAstor"

src/blockstore/v1/flush.cpp

@@ -520,6 +520,7 @@ resume_2:
                 await_sqe(15);
                 data->iov = (struct iovec){ it->buf, (size_t)it->len };
                 data->callback = simple_callback_w;
+                assert(clean_loc+it->offset+it->len <= bs->dsk.block_count*bs->dsk.data_block_size);
                 io_uring_prep_writev(
                     sqe, bs->dsk.data_fd, &data->iov, 1, bs->dsk.data_offset + clean_loc + it->offset
                 );
@@ -749,6 +750,7 @@ bool journal_flusher_co::write_meta_block(flusher_meta_write_t & meta_block, int
     await_sqe(0);
     data->iov = (struct iovec){ meta_block.buf, (size_t)bs->dsk.meta_block_size };
     data->callback = simple_callback_w;
+    assert(bs->dsk.meta_block_size + meta_block.sector + bs->dsk.meta_block_size <= bs->dsk.meta_area_size);
     io_uring_prep_writev(
         sqe, bs->dsk.meta_fd, &data->iov, 1, bs->dsk.meta_offset + bs->dsk.meta_block_size + meta_block.sector
     );

src/blockstore/v1/journal.cpp

@@ -193,6 +193,7 @@ void blockstore_impl_t::prepare_journal_sector_write(int cur_sector, blockstore_
             (size_t)journal.block_size
         };
         data->callback = [this, flush_id = journal.submit_id](ring_data_t *data) { handle_journal_write(data, flush_id); };
+        assert(journal.sector_info[cur_sector].offset+journal.block_size <= dsk.journal_len);
         io_uring_prep_writev(
             sqe, dsk.journal_fd, &data->iov, 1, journal.offset + journal.sector_info[cur_sector].offset
         );

src/blockstore/v1/write.cpp

@@ -368,9 +368,9 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
         }
         data->iov.iov_len = op->len + stripe_offset + stripe_end; // to check it in the callback
         data->callback = [this, op](ring_data_t *data) { handle_write_event(data, op); };
-        io_uring_prep_writev(
-            sqe, dsk.data_fd, PRIV(op)->iov_zerofill, vcnt, dsk.data_offset + (loc * dsk.data_block_size) + op->offset - stripe_offset
-        );
+        const uint64_t write_offset = (loc * dsk.data_block_size) + op->offset - stripe_offset;
+        assert(write_offset+op->len+stripe_offset+stripe_end <= dsk.block_count*dsk.data_block_size);
+        io_uring_prep_writev(sqe, dsk.data_fd, PRIV(op)->iov_zerofill, vcnt, dsk.data_offset + write_offset);
         PRIV(op)->pending_ops = 1;
         if (!(dirty_it->second.state & BS_ST_INSTANT))
         {
@@ -495,9 +495,8 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
                 .op = op,
             });
             data2->callback = [this, flush_id = journal.submit_id](ring_data_t *data) { handle_journal_write(data, flush_id); };
-            io_uring_prep_writev(
-                sqe2, dsk.journal_fd, &data2->iov, 1, journal.offset + journal.next_free
-            );
+            assert(journal.next_free+op->len <= dsk.journal_len);
+            io_uring_prep_writev(sqe2, dsk.journal_fd, &data2->iov, 1, journal.offset + journal.next_free);
             PRIV(op)->pending_ops++;
         }
         else

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
-	msgr_stop.cpp msgr_op.cpp msgr_send.cpp msgr_receive.cpp ../util/ringloop.cpp ../../json11/json11.cpp
+	../util/epoll_manager.cpp etcd_state_client.cpp messenger.cpp msgr_iothread.cpp ../util/addr_util.cpp ../util/xxh_x86dispatch.c ../util/openssl_util.cpp
+	msgr_encrypt.cpp msgr_handshake.cpp msgr_stop.cpp msgr_op.cpp msgr_send.cpp msgr_receive.cpp ../util/ringloop.cpp ../../json11/json11.cpp
 	http_client.cpp osd_ops.cpp pg_states.cpp ../util/timerfd_manager.cpp ../util/str_util.cpp ../util/json_util.cpp ${MSGR_RDMA} ${MSGR_RDMACM}
 )
-target_link_libraries(vitastor_common pthread)
+target_link_libraries(vitastor_common pthread ${OPENSSL_LIBRARIES} ${CARES_LIBRARIES} ${ISAL_CRYPTO_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 +34,8 @@ target_link_libraries(vitastor_client
 	${LIBURING_LIBRARIES}
 	${IBVERBS_LIBRARIES}
 	${RDMACM_LIBRARIES}
+	${OPENSSL_LIBRARIES}
+	${ISAL_CRYPTO_LIBRARIES}
 )
 set_target_properties(vitastor_client PROPERTIES VERSION ${VITASTOR_VERSION} SOVERSION 0)
 configure_file(vitastor.pc.in vitastor.pc @ONLY)
@@ -52,9 +55,6 @@ if (${WITH_FIO})
 		../util/rw_blocking.cpp
 		../util/addr_util.cpp
 	)
-	target_link_libraries(fio_vitastor_sec
-		tcmalloc_minimal
-	)
 endif (${WITH_FIO})
 
 # vitastor-nbd
@@ -98,9 +98,10 @@ endif (${WITH_QEMU})
 add_executable(test_cluster_client
 	EXCLUDE_FROM_ALL
 	../test/test_cluster_client.cpp
-	pg_states.cpp osd_ops.cpp cluster_client.cpp cluster_client_list.cpp cluster_client_wb.cpp msgr_op.cpp ../test/mock/messenger.cpp msgr_stop.cpp
-	etcd_state_client.cpp ../util/timerfd_manager.cpp ../util/addr_util.cpp ../util/str_util.cpp ../util/json_util.cpp ../../json11/json11.cpp
+	pg_states.cpp osd_ops.cpp cluster_client.cpp cluster_client_list.cpp cluster_client_wb.cpp cluster_client_icache.cpp msgr_op.cpp ../test/mock/messenger.cpp msgr_stop.cpp msgr_encrypt.cpp
+	etcd_state_client.cpp ../util/timerfd_manager.cpp ../util/addr_util.cpp ../util/str_util.cpp ../util/json_util.cpp ../util/xxh_x86dispatch.c ../util/openssl_util.cpp ../../json11/json11.cpp
 )
+target_link_libraries(test_cluster_client ${LIBURING_LIBRARIES} ${OPENSSL_LIBRARIES} ${ISAL_CRYPTO_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

@@ -51,7 +51,7 @@ cluster_client_t::cluster_client_t(ring_loop_t *ringloop, timerfd_manager_t *tfd
         msgr.stop_client(op->client_id);
         delete op;
     };
-    msgr.parse_config(config);
+    msgr.parse_config(config, true);
 
     st_cli.tfd = tfd;
     st_cli.on_load_config_hook = [this](json11::Json::object & cfg) { on_load_config_hook(cfg); };
@@ -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;
 }
@@ -481,7 +479,9 @@ void cluster_client_t::on_load_config_hook(json11::Json::object & etcd_global_co
         self_tree_metrics.clear();
         client_hostname = new_hostname;
     }
-    msgr.parse_config(config);
+    // vault
+    vault_parse_config();
+    msgr.parse_config(config, false);
     st_cli.parse_config(config);
     st_cli.load_pgs();
 }
@@ -590,7 +590,7 @@ void cluster_client_t::on_change_pool_config_hook()
         {
             if (log_level > 2 && pg_counts[pool_item.first])
             {
-                printf("Pool %u (%s) PG count changed from %lu to %lu\n", pool_item.first, pool_item.second.name.c_str(),
+                fprintf(stderr, "Pool %u (%s) PG count changed from %lu to %lu\n", pool_item.first, pool_item.second.name.c_str(),
                     pg_counts[pool_item.first], pool_item.second.real_pg_count);
             }
             // At this point, all pool operations should have been suspended
@@ -607,6 +607,9 @@ void cluster_client_t::on_change_pool_config_hook()
             pg_counts[pool_item.first] = pool_item.second.real_pg_count;
         }
     }
+    inode_cache.clear();
+    inode_cache_children.clear();
+    vault_keys.clear();
     continue_ops();
 }
 
@@ -673,6 +676,10 @@ bool cluster_client_t::flush()
 {
     if (!ringloop)
     {
+        if (vault_loading)
+        {
+            return false;
+        }
         if (wb->writeback_queue.size())
         {
             wb->start_writebacks(this, 0);
@@ -695,7 +702,7 @@ bool cluster_client_t::flush()
         sync_done = true;
     };
     execute(sync);
-    while (!sync_done)
+    while (!sync_done || vault_loading)
     {
         ringloop->loop();
         if (!sync_done)
@@ -958,10 +965,40 @@ bool cluster_client_t::check_rw(cluster_op_t *op)
     {
         op->flags |= OP_IMMEDIATE_COMMIT;
     }
+    bool searched = false;
+    std::shared_ptr<inode_cache_t> icache;
+    if (op->opcode == OSD_OP_READ || op->opcode == OSD_OP_WRITE)
+    {
+        if (!searched)
+        {
+            icache = inode_cache_get(op->inode);
+            searched = true;
+        }
+        if (icache && icache->has_parent_loop && op->opcode == OSD_OP_READ)
+        {
+            op->retval = -EINVAL;
+            auto cb = std::move(op->callback);
+            cb(op);
+            return false;
+        }
+        if (icache && icache->op_enc)
+        {
+            // Use shared_ptr aliasing to attach op_enc to the inode cache entry
+            op->enc = std::shared_ptr<osd_op_enc_t>(icache, icache->op_enc);
+        }
+        else
+            op->enc.reset();
+    }
+    else
+        op->enc.reset();
     if ((op->opcode == OSD_OP_WRITE || op->opcode == OSD_OP_DELETE) && !(op->flags & OSD_OP_IGNORE_READONLY))
     {
-        auto ino_it = st_cli.inode_config.find(op->inode);
-        if (ino_it != st_cli.inode_config.end() && ino_it->second.readonly)
+        if (!searched)
+        {
+            icache = inode_cache_get(op->inode);
+            searched = true;
+        }
+        if (icache && icache->readonly)
         {
             op->retval = -EROFS;
             auto cb = std::move(op->callback);
@@ -972,33 +1009,39 @@ bool cluster_client_t::check_rw(cluster_op_t *op)
     op->deoptimise_snapshot = false;
     if (enable_writeback && (op->opcode == OSD_OP_READ || op->opcode == OSD_OP_READ_BITMAP || op->opcode == OSD_OP_READ_CHAIN_BITMAP))
     {
-        auto ino_it = st_cli.inode_config.find(op->inode);
-        if (ino_it != st_cli.inode_config.end())
+        if (!searched)
         {
-            int chain_size = 0;
-            while (ino_it != st_cli.inode_config.end() && ino_it->second.parent_id)
+            icache = inode_cache_get(op->inode);
+            searched = true;
+        }
+        if (icache)
+        {
+            for (auto & parent: icache->chain)
             {
-                // Check for loops - FIXME check it in etcd_state_client
-                if (ino_it->second.parent_id == op->inode ||
-                    chain_size > st_cli.inode_config.size())
-                {
-                    op->retval = -EINVAL;
-                    auto cb = std::move(op->callback);
-                    cb(op);
-                    return false;
-                }
-                if (INODE_POOL(ino_it->second.parent_id) == INODE_POOL(ino_it->first) &&
-                    wb->has_inode(ino_it->second.parent_id))
+                if (INODE_POOL(parent) == INODE_POOL(op->inode) && wb->has_inode(parent))
                 {
                     // Deoptimise reads - we have dirty data for one of the parent layer(s).
                     op->deoptimise_snapshot = true;
                     break;
                 }
-                chain_size++;
-                ino_it = st_cli.inode_config.find(ino_it->second.parent_id);
             }
         }
     }
+    if (icache && icache->err_code)
+    {
+        if (icache->err_code == EPERM)
+        {
+            op->retval = -EPERM;
+            auto cb = std::move(op->callback);
+            cb(op);
+            return false;
+        }
+        else if (icache->err_code == EAGAIN)
+        {
+            key_wait_ops.push_back(op);
+            return false;
+        }
+    }
     return true;
 }
 
@@ -1119,33 +1162,42 @@ resume_2:
         // Finished successfully
         // Even if the PG count has changed in meanwhile we treat it as success
         // 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_BITMAP || op->opcode == OSD_OP_READ_CHAIN_BITMAP)
+        {
+            // Copy part bitmaps only after finishing all part reads
+            for (auto & part: op->parts)
+                if ((part.flags & (PART_SENT|PART_DONE|PART_VALID)) == (PART_SENT|PART_DONE|PART_VALID))
+                    copy_part_bitmap(op, &part);
+        }
         if (op->opcode == OSD_OP_READ || op->opcode == OSD_OP_READ_CHAIN_BITMAP)
         {
-            // Check parent inode
-            auto ino_it = st_cli.inode_config.find(op->cur_inode);
-            // Skip parents from the same pool
-            int skipped = 0;
-            while (!op->deoptimise_snapshot &&
-                ino_it != st_cli.inode_config.end() && ino_it->second.parent_id &&
-                INODE_POOL(ino_it->second.parent_id) == INODE_POOL(op->cur_inode))
+            uint64_t next_inode = 0;
+            auto icache = inode_cache_get(op->cur_inode);
+            if (icache)
             {
-                // Check for loops - FIXME check it in etcd_state_client
-                if (ino_it->second.parent_id == op->inode ||
-                    skipped > st_cli.inode_config.size())
+                if (icache->has_parent_loop)
                 {
                     op->retval = -EINVAL;
                     erase_op(op);
                     return 1;
                 }
-                skipped++;
-                ino_it = st_cli.inode_config.find(ino_it->second.parent_id);
+                if (op->deoptimise_snapshot)
+                {
+                    if (icache->chain.size() > 1)
+                        next_inode = icache->chain[1];
+                }
+                else
+                {
+                    if (icache->other_pool_parent_id)
+                        next_inode = icache->other_pool_parent_id;
+                }
             }
-            if (ino_it != st_cli.inode_config.end() &&
-                ino_it->second.parent_id &&
-                ino_it->second.parent_id != op->inode)
+            if (next_inode)
             {
                 // Continue reading from the parent inode
-                op->cur_inode = ino_it->second.parent_id;
+                icache = inode_cache_get(next_inode);
+                op->cur_inode = next_inode;
+                op->enc = (icache && icache->op_enc ? std::shared_ptr<osd_op_enc_t>(icache, icache->op_enc) : nullptr);
                 op->parts.clear();
                 op->done_count = 0;
                 goto resume_0;
@@ -1164,7 +1216,7 @@ resume_2:
         erase_op(op);
         return 1;
     }
-    else if (op->retval != 0 && !(op->flags & OP_FLUSH_BUFFER) &&
+    else if (op->retval != 0 && op->opcode != OSD_OP_SYNC && !(op->flags & OP_FLUSH_BUFFER) &&
         op->retval != -EPIPE && (op->retval != -EIO || !client_eio_retry_interval) && (op->retval != -ENOSPC || !client_retry_enospc))
     {
         // Fatal error (neither -EPIPE, -EIO nor -ENOSPC)
@@ -1196,7 +1248,7 @@ resume_2:
     return 0;
 }
 
-static void add_iov(int size, bool skip, cluster_op_t *op, int &iov_idx, size_t &iov_pos, osd_op_buf_list_t &iov, void *scrap, int scrap_len)
+static void add_iov(int size, int skip, cluster_op_t *op, int &iov_idx, size_t &iov_pos, osd_op_buf_list_t &iov)
 {
     int left = size;
     while (left > 0 && iov_idx < op->iov.count)
@@ -1204,7 +1256,7 @@ static void add_iov(int size, bool skip, cluster_op_t *op, int &iov_idx, size_t
         int cur_left = op->iov.buf[iov_idx].iov_len - iov_pos;
         if (cur_left < left)
         {
-            if (!skip)
+            if (skip == 0)
             {
                 iov.push_back((uint8_t*)op->iov.buf[iov_idx].iov_base + iov_pos, cur_left);
             }
@@ -1214,7 +1266,7 @@ static void add_iov(int size, bool skip, cluster_op_t *op, int &iov_idx, size_t
         }
         else
         {
-            if (!skip)
+            if (skip == 0)
             {
                 iov.push_back((uint8_t*)op->iov.buf[iov_idx].iov_base + iov_pos, left);
             }
@@ -1223,16 +1275,10 @@ static void add_iov(int size, bool skip, cluster_op_t *op, int &iov_idx, size_t
         }
     }
     assert(left == 0);
-    if (skip && scrap_len > 0)
+    if (skip == 1)
     {
-        // All skipped ranges are read into the same useless buffer
-        left = size;
-        while (left > 0)
-        {
-            int cur_left = scrap_len < left ? scrap_len : left;
-            iov.push_back(scrap, cur_left);
-            left -= cur_left;
-        }
+        // data read into a NULL buffer will be discarded by messenger
+        iov.push_back(NULL, size);
     }
 }
 
@@ -1252,7 +1298,11 @@ void cluster_client_t::slice_rw(cluster_op_t *op)
         // Allocate memory for the bitmap
         unsigned object_bitmap_size = ((op->len / pool_cfg.bitmap_granularity + 7) / 8);
         object_bitmap_size = (object_bitmap_size < 8 ? 8 : object_bitmap_size);
-        unsigned bitmap_mem = object_bitmap_size + (pool_cfg.data_block_size / pool_cfg.bitmap_granularity / 8 * pg_data_size) * op->parts.size();
+        unsigned bitmap_mem = object_bitmap_size +
+            op->parts.size() * pg_data_size *
+            (pool_cfg.data_block_size / pool_cfg.bitmap_granularity / 8
+            // read chain_info - max 4 bytes per block
+            + (op->enc ? osd_op_rw_t::chain_info_bytes(op->enc->chain_size)*op->len/pool_cfg.bitmap_granularity : 0));
         if (!op->bitmap_buf || op->bitmap_buf_size < bitmap_mem)
         {
             op->bitmap_buf = realloc_or_die(op->bitmap_buf, bitmap_mem);
@@ -1294,10 +1344,10 @@ void cluster_client_t::slice_rw(cluster_op_t *op)
                         {
                             begin = cur;
                             // Just advance iov_idx & iov_pos
-                            add_iov(cur-prev, true, op, iov_idx, iov_pos, op->parts[i].iov, NULL, 0);
+                            add_iov(cur-prev, 2, op, iov_idx, iov_pos, op->parts[i].iov);
                         }
                         else
-                            add_iov(cur-prev, skip_prev, op, iov_idx, iov_pos, op->parts[i].iov, scrap_buffer, scrap_buffer_size);
+                            add_iov(cur-prev, skip_prev ? 1 : 0, op, iov_idx, iov_pos, op->parts[i].iov);
                     }
                     skip_prev = skip;
                     prev = cur;
@@ -1308,11 +1358,11 @@ void cluster_client_t::slice_rw(cluster_op_t *op)
             if (skip_prev)
             {
                 // Just advance iov_idx & iov_pos
-                add_iov(end-prev, true, op, iov_idx, iov_pos, op->parts[i].iov, NULL, 0);
+                add_iov(end-prev, 2, op, iov_idx, iov_pos, op->parts[i].iov);
                 end = prev;
             }
             else
-                add_iov(cur-prev, skip_prev, op, iov_idx, iov_pos, op->parts[i].iov, scrap_buffer, scrap_buffer_size);
+                add_iov(cur-prev, skip_prev ? 1 : 0, op, iov_idx, iov_pos, op->parts[i].iov);
             if (end == begin)
             {
                 op->done_count++;
@@ -1321,7 +1371,7 @@ void cluster_client_t::slice_rw(cluster_op_t *op)
         }
         else if (op->opcode != OSD_OP_READ_BITMAP && op->opcode != OSD_OP_READ_CHAIN_BITMAP && op->opcode != OSD_OP_DELETE)
         {
-            add_iov(end-begin, false, op, iov_idx, iov_pos, op->parts[i].iov, NULL, 0);
+            add_iov(end-begin, 0, op, iov_idx, iov_pos, op->parts[i].iov);
         }
         op->parts[i].parent = op;
         op->parts[i].offset = begin;
@@ -1407,9 +1457,12 @@ int cluster_client_t::try_send(cluster_op_t *op, int i, std::function<void(osd_o
             osd_client_t *cl = peer_it->second;
             part->flags |= PART_SENT|PART_VALID;
             op->inflight_count++;
-            uint64_t pg_bitmap_size = (pool_cfg.data_block_size / pool_cfg.bitmap_granularity / 8) * (
-                pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks
-            );
+            uint32_t pg_data_size = (pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks);
+            uint64_t pg_bitmap_size = pg_data_size * (pool_cfg.data_block_size / pool_cfg.bitmap_granularity / 8
+                // read chain_info - max 4 bytes per block
+                + (op->opcode == OSD_OP_READ && op->enc
+                    ? osd_op_rw_t::chain_info_bytes(op->enc->chain_size)*pool_cfg.data_block_size/pool_cfg.bitmap_granularity
+                    : 0));
             uint64_t meta_rev = 0;
             if (op->opcode != OSD_OP_READ_BITMAP && op->opcode != OSD_OP_DELETE && !op->deoptimise_snapshot)
             {
@@ -1428,6 +1481,7 @@ int cluster_client_t::try_send(cluster_op_t *op, int i, std::function<void(osd_o
                     .inode = op->cur_inode,
                     .offset = part->offset,
                     .len = part->len,
+                    .flags = op->opcode == OSD_OP_READ && op->enc && !op->deoptimise_snapshot ? OSD_OP_RETURN_CHAIN : 0,
                     .meta_revision = meta_rev,
                     .version = op->opcode == OSD_OP_WRITE || op->opcode == OSD_OP_DELETE ? op->version : 0,
                 } },
@@ -1435,6 +1489,7 @@ int cluster_client_t::try_send(cluster_op_t *op, int i, std::function<void(osd_o
                     ? (uint8_t*)op->part_bitmaps + pg_bitmap_size*i : NULL),
                 .bitmap_len = (unsigned)(op->opcode == OSD_OP_READ || op->opcode == OSD_OP_READ_BITMAP || op->opcode == OSD_OP_READ_CHAIN_BITMAP
                     ? pg_bitmap_size : 0),
+                .enc = op->enc,
                 .callback = cb ? cb : [this, part](osd_op_t *op_part)
                 {
                     handle_op_part(part);
@@ -1552,6 +1607,9 @@ static inline void mem_or(void *res, const void *r2, unsigned int len)
     }
 }
 
+// Error priority: others > EPERM > EIO > ENOSPC > ETIMEDOUT > EPIPE
+#define ERR_PRIO(e) (((e) == -EPERM ? 5 : ((e) == -EIO ? 4 : ((e) == -ENOSPC ? 3 : ((e) == -ETIMEDOUT ? 2 : ((e) == -EPIPE ? 1 : (!(e) ? 0 : 10)))))))
+
 void cluster_client_t::handle_op_part(cluster_op_part_t *part)
 {
     cluster_op_t *op = part->parent;
@@ -1560,15 +1618,10 @@ void cluster_client_t::handle_op_part(cluster_op_part_t *part)
     {
         // Operation failed, retry
         part->flags |= PART_ERROR;
-        if (!op->retval || op->retval == -EPIPE ||
-            part->op.reply.hdr.retval == -ENOSPC && op->retval == -ETIMEDOUT ||
-            part->op.reply.hdr.retval == -EIO)
-        {
-            // Error priority: EIO > ENOSPC > ETIMEDOUT > EPIPE
+        if (ERR_PRIO(part->op.reply.hdr.retval) > ERR_PRIO(op->retval))
             op->retval = part->op.reply.hdr.retval;
-        }
         uint64_t stop_client_id = 0;
-        if (op->retval != -EINTR && op->retval != -EIO && op->retval != -ENOSPC)
+        if (op->retval != -EINTR && op->retval != -EIO && op->retval != -ENOSPC && op->retval != -EPERM)
         {
             stop_client_id = part->op.client_id;
             if (op->retval != -EPIPE || log_level > 0)
@@ -1630,13 +1683,6 @@ void cluster_client_t::handle_op_part(cluster_op_part_t *part)
         }
         if (op->inflight_count == 0 && !op->retry_after)
         {
-            // Copy part bitmaps only after finishing all part reads
-            if (op->opcode == OSD_OP_READ || op->opcode == OSD_OP_READ_BITMAP || op->opcode == OSD_OP_READ_CHAIN_BITMAP)
-            {
-                for (auto & part: op->parts)
-                    if ((part.flags & (PART_SENT|PART_VALID|PART_DONE)) == (PART_SENT|PART_VALID|PART_DONE))
-                        copy_part_bitmap(op, &part);
-            }
             if (op->opcode == OSD_OP_SYNC)
                 continue_sync(op);
             else

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;
 };
@@ -80,17 +82,33 @@ struct inode_list_osd_t;
 struct inode_list_pg_t;
 class writeback_cache_t;
 
+struct inode_cache_t
+{
+    std::vector<inode_t> chain; // only parents from the same pool
+    uint8_t *key_data = NULL;
+    osd_op_enc_t *op_enc = NULL;
+    bool readonly = false;
+    bool has_parent_loop = false;
+    inode_t other_pool_parent_id = 0;
+    int err_code = 0;
+
+    ~inode_cache_t();
+};
+
+struct vault_load_key_t
+{
+    int key_state = 0;
+    std::string key;
+};
+
 // FIXME: Split into public and private interfaces
 class __attribute__((visibility("default"))) cluster_client_t
 {
-#ifdef __MOCK__
-public:
-#endif
     timerfd_manager_t *tfd = NULL;
     ring_loop_t *ringloop = NULL;
 
-    std::map<pool_id_t, uint64_t> pg_counts;
-    std::map<pool_pg_num_t, osd_num_t> pg_primary;
+    // config:
+
     // client_max_dirty_* is actually "max unsynced", for the case when immediate_commit is off
     uint64_t client_max_dirty_bytes = 0;
     uint64_t client_max_dirty_ops = 0;
@@ -102,12 +120,23 @@ 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 +144,28 @@ public:
     int retry_timeout_id = -1;
     int retry_timeout_duration = 0;
     std::vector<cluster_op_t*> offline_ops;
+    std::vector<cluster_op_t*> key_wait_ops;
     cluster_op_t *op_queue_head = NULL, *op_queue_tail = NULL;
     writeback_cache_t *wb = NULL;
     std::set<osd_num_t> dirty_osds;
     uint64_t dirty_bytes = 0, dirty_ops = 0;
 
-    void *scrap_buffer = NULL;
-    unsigned scrap_buffer_size = 0;
+    // inodes require some extra state for read/write, it's stored here.
+    // moreover, robin_hood access is slightly faster than std::map :)
+    robin_hood::unordered_flat_map<inode_t, std::shared_ptr<inode_cache_t>> inode_cache;
+    std::set<std::pair<inode_t, inode_t>> inode_cache_children;
+
+    http_context_t *vault_http_ctx = NULL;
+    http_co_t *vault_http_cli = NULL;
+    bool vault_loading = false;
+    std::string vault_token;
+    bool vault_auth_error = false;
+    timespec vault_token_expire = {};
+    std::vector<std::string> vault_key_load_queue;
+    std::map<std::string, vault_load_key_t> vault_keys;
 
     bool pgs_loaded = false;
+    std::map<pool_id_t, uint64_t> pg_counts;
     ring_consumer_t consumer;
     std::vector<std::function<void(void)>> on_ready_hooks;
     int list_retry_timeout_id = -1;
@@ -155,15 +197,16 @@ public:
     void list_inode(inode_t inode, uint64_t min_offset, uint64_t max_offset, int max_parallel_pgs, std::function<void(
         int status, int pgs_left, pg_num_t pg_num, std::set<object_id>&& objects)> pg_callback);
 
-    //inline uint32_t get_bs_bitmap_granularity() { return st_cli.global_bitmap_granularity; }
-    //inline uint64_t get_bs_block_size() { return st_cli.global_block_size; }
-
-#ifndef __MOCK__
 protected:
-#endif
     void continue_ops(int time_passed = 0);
 
-protected:
+    std::shared_ptr<inode_cache_t> inode_cache_get(inode_t ino);
+    void vault_parse_config();
+    bool vault_check_token();
+    void vault_load_keys();
+    void vault_destroy();
+    void vault_parse_secret(const std::string & key_id, const std::string & err, json11::Json data);
+
     bool affects_osd(uint64_t inode, uint64_t offset, uint64_t len, osd_num_t osd);
     bool affects_pg(uint64_t inode, uint64_t offset, uint64_t len, pool_id_t pool_id, pg_num_t pg_num);
 
@@ -173,6 +216,7 @@ 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 +233,7 @@ 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);
@@ -204,4 +249,5 @@ protected:
     osd_num_t select_nearest_osd(const std::vector<osd_num_t> & osds);
 
     friend class writeback_cache_t;
+    friend class cluster_client_test_t;
 };

src/client/cluster_client_icache.cpp

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

src/client/cluster_client_impl.h

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

src/client/cluster_client_wb.cpp

@@ -88,6 +88,11 @@ void writeback_cache_t::copy_write(cluster_op_t *op, int state, uint64_t new_flu
     // ...or just save it for writeback if write buffering is enabled
     if (op->len == 0)
     {
+        // FIXME: OSD_OP_DELETEs are currently only sent by vitastor-cli rm/rm-data and
+        // actually have len=0, because delete is actually a delete of the full object
+        // containing the requested offset, not a "punch hole" operation. But here, writeback
+        // cache assumes it IS a "punch hole" operation. I should select one of these
+        // approaches and fix everything accordingly when I decide to implement TRIM.
         return;
     }
     auto dirty_it = find_dirty(op->inode, op->offset);
@@ -244,12 +249,13 @@ void writeback_cache_t::copy_write(cluster_op_t *op, int state, uint64_t new_flu
             writeback_queue_size--;
         }
     }
-    if (!is_del)
+    if (!is_del && op->len > 0)
     {
         uint64_t pos = 0, len = op->len, iov_idx = 0;
-        while (len > 0 && iov_idx < op->iov.count)
+        while (iov_idx < op->iov.count)
         {
             auto & iov = op->iov.buf[iov_idx];
+            assert(pos + iov.iov_len <= len);
             memcpy(buf + pos, iov.iov_base, iov.iov_len);
             pos += iov.iov_len;
             iov_idx++;
@@ -443,7 +449,7 @@ void writeback_cache_t::start_writebacks(cluster_client_t *cli, int count)
         started++;
         assert(writeback_queue_size > 0);
         writeback_queue_size--;
-        writeback_bytes -= off - from_it->first.stripe;
+        writeback_bytes -= (is_del ? 0 : off - from_it->first.stripe);
         assert(writeback_queue_size > 0 || !writeback_bytes);
         flush_buffers(cli, from_it, to_it);
     }

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__
@@ -9,6 +12,7 @@
 #include "http_client.h"
 #endif
 #include "str_util.h"
+#include "json_util.h"
 
 etcd_state_client_t::~etcd_state_client_t()
 {
@@ -22,14 +26,19 @@ 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 +81,91 @@ 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,
+std::shared_ptr<user_info_t> etcd_state_client_t::get_user(const std::string & username)
+{
+    auto user_it = user_info.find(username);
+    if (user_it != user_info.end())
+    {
+        return user_it->second;
+    }
+    auto inf = std::make_shared<user_info_t>();
+    inf->name = username;
+    return inf;
+}
+
+bool etcd_state_client_t::check_image_perm(const std::shared_ptr<user_info_t> & user_info, inode_t inode_num, bool write)
+{
+    if (user_info->type == user_type_t::ADMIN)
+    {
+        return true;
+    }
+    auto cache_it = user_info->perm_cache.find(inode_num);
+    if (cache_it != user_info->perm_cache.end() &&
+        cache_it->second.mod_revision == user_perm_cache_revision)
+    {
+        return write ? (cache_it->second.perm == user_perm_t::OWNER) : (cache_it->second.perm != user_perm_t::DENY);
+    }
+    auto inode_it = inode_config.find(inode_num);
+    if (inode_it == inode_config.end())
+    {
+        return false;
+    }
+    // FIXME Implement cache reset after reworking etcd interaction to not keep everything in memory
+    auto & perm_item = user_info->perm_cache[inode_num];
+    perm_item.mod_revision = user_perm_cache_revision;
+    perm_item.perm = (user_info->name == inode_it->second.owner || inode_it->second.owner_group != "" &&
+        user_info->groups.find(inode_it->second.owner_group) != user_info->groups.end()
+        ? user_perm_t::OWNER : (inode_it->second.reader_group != "" &&
+            user_info->groups.find(inode_it->second.reader_group) != user_info->groups.end()
+            ? user_perm_t::READER : user_perm_t::DENY));
+    return write ? (perm_item.perm == user_perm_t::OWNER) : (perm_item.perm != user_perm_t::DENY);
+}
+
+http_context_t *etcd_state_client_t::get_http_ctx()
+{
+    if (!http_ctx)
+    {
+        std::string error;
+        http_ctx = http_context_init(tfd, etcd_client_cert, etcd_client_key, etcd_ca, true, error);
+        if (!http_ctx)
+        {
+            fprintf(stderr, "Failed to initialize HTTP context: %s\n", error.c_str());
+            exit(1);
+        }
+    }
+    return http_ctx;
+}
+
+void etcd_state_client_t::etcd_call_oneshot(const std::string & etcd_url, const std::string & api, json11::Json payload,
     int timeout, std::function<void(std::string, json11::Json)> callback)
 {
-    std::string etcd_api_path;
-    int pos = etcd_address.find('/');
-    if (pos >= 0)
-    {
-        etcd_api_path = etcd_address.substr(pos);
-        etcd_address = etcd_address.substr(0, pos);
-    }
-    std::string req = payload.dump();
-    req = "POST "+etcd_api_path+api+" HTTP/1.1\r\n"
-        "Host: "+etcd_address+"\r\n"
-        "Content-Type: application/json\r\n"
-        "Content-Length: "+std::to_string(req.size())+"\r\n"
-        "Connection: close\r\n"
-        "\r\n"+req;
-    auto http_cli = http_init(tfd);
-    auto cb = [http_cli, callback](const http_response_t *response)
+    auto http_cli = http_init(get_http_ctx());
+    http_json_post(http_cli, etcd_url+api, payload, "", { .timeout = timeout }, [http_cli, callback](http_message_t *response)
     {
         std::string err;
         json11::Json data;
         response->parse_json_response(err, data);
         callback(err, data);
-        http_close(http_cli);
-    };
-    http_request(http_cli, etcd_address, req, { .timeout = timeout }, cb);
+        http_destroy(http_cli);
+    });
 }
 
-void etcd_state_client_t::etcd_call(std::string api, json11::Json payload, int timeout,
+void etcd_state_client_t::etcd_call(const std::string & api, json11::Json payload, int timeout,
     int retries, int interval, std::function<void(std::string, json11::Json)> callback)
 {
-    if (!etcd_addresses.size() && !etcd_local.size())
+    pick_next_etcd([=]()
     {
-        fprintf(stderr, "etcd_address is missing in Vitastor configuration\n");
-        exit(1);
-    }
-    pick_next_etcd();
-    std::string etcd_address = selected_etcd_address;
-    std::string etcd_api_path;
-    int pos = etcd_address.find('/');
-    if (pos >= 0)
-    {
-        etcd_api_path = etcd_address.substr(pos);
-        etcd_address = etcd_address.substr(0, pos);
-    }
+        etcd_call_selected(api, payload, timeout, retries, interval, callback);
+    });
+}
+
+void etcd_state_client_t::etcd_call_selected(const std::string & api, json11::Json payload, int timeout,
+    int retries, int interval, std::function<void(std::string, json11::Json)> callback)
+{
+    const auto & url = selected_etcd_url;
     std::string req = payload.dump();
-    req = "POST "+etcd_api_path+api+" HTTP/1.1\r\n"
-        "Host: "+etcd_address+"\r\n"
+    req = "POST "+url.path+api+" HTTP/1.1\r\n"
+        "Host: "+url.hostname+"\r\n"
         "Content-Type: application/json\r\n"
         "Content-Length: "+std::to_string(req.size())+"\r\n"
         "Connection: keep-alive\r\n"
@@ -128,15 +173,15 @@ void etcd_state_client_t::etcd_call(std::string api, json11::Json payload, int t
         "\r\n"+req;
     retries--;
     auto cb = [this, api, payload, timeout, retries, interval, callback,
-        cur_addr = selected_etcd_address](const http_response_t *response)
+        cur_addr = url.addr](http_message_t *response)
     {
         std::string err;
         json11::Json data;
         response->parse_json_response(err, data);
         if (err != "")
         {
-            if (cur_addr == selected_etcd_address)
-                selected_etcd_address = "";
+            if (cur_addr == selected_etcd_url.addr)
+                selected_etcd_url = (http_url_t){};
             if (retries > 0)
             {
                 if (this->log_level > 0)
@@ -164,54 +209,58 @@ void etcd_state_client_t::etcd_call(std::string api, json11::Json payload, int t
             callback(err, data);
     };
     if (!keepalive_client)
-    {
-        keepalive_client = http_init(tfd);
-    }
-    http_request(keepalive_client, etcd_address, req, { .timeout = timeout, .keepalive = true }, cb);
+        keepalive_client = http_init(get_http_ctx());
+    http_request(keepalive_client, url.addr, req, { .timeout = timeout, .keepalive = true, .ssl = url.ssl }, cb);
 }
 
-void etcd_state_client_t::add_etcd_url(std::string addr)
+
+void etcd_state_client_t::add_etcd_url(std::string etcd_address)
 {
-    if (addr.length() > 0)
+    if (etcd_address.size() > 0)
     {
-        if (strtolower(addr.substr(0, 7)) == "http://")
-            addr = addr.substr(7);
-        else if (strtolower(addr.substr(0, 8)) == "https://")
-        {
-            fprintf(stderr, "HTTPS is unsupported for etcd. Either use plain HTTP or setup a local proxy for etcd interaction\n");
-            exit(1);
-        }
         if (!local_ips.size())
-            local_ips = getifaddr_list(std::vector<addr_mask_t>(), true);
-        std::string check_addr;
-        int pos = addr.find('/');
-        int pos2 = addr.find(':');
-        if (pos2 >= 0)
-            check_addr = addr.substr(0, pos2);
-        else if (pos >= 0)
-            check_addr = addr.substr(0, pos);
-        else
-            check_addr = addr;
-        if (pos == std::string::npos)
-            addr += "/v3";
-        bool local = false;
-        int i;
-        for (i = 0; i < local_ips.size(); i++)
         {
-            if (local_ips[i] == check_addr)
-            {
-                local = true;
-                break;
-            }
+            // Fill local_ips
+            for (auto & ip: getifaddr_list(std::vector<addr_mask_t>(), true))
+                local_ips.insert(ip);
         }
-        auto & to = local ? this->etcd_local : this->etcd_addresses;
+        std::string etcd_api_path;
+        bool ssl = false;
+        if (etcd_address.substr(0, 8) == "https://")
+        {
+            ssl = true;
+            etcd_address = etcd_address.substr(8);
+        }
+        else if (etcd_address.substr(0, 7) == "http://")
+            etcd_address = etcd_address.substr(7);
+        auto pos = etcd_address.find('/');
+        if (pos != std::string::npos)
+        {
+            etcd_api_path = etcd_address.substr(pos);
+            etcd_address = etcd_address.substr(0, pos);
+        }
+        else
+            etcd_api_path = "/v3";
+        pos = etcd_address.find(':');
+        auto check_addr = (pos != std::string::npos ? etcd_address.substr(0, pos) : etcd_address);
+        bool is_local = local_ips.find(check_addr) != local_ips.end();
+        auto & to = (is_local ? etcd_local : etcd_addresses);
+        check_addr = (ssl ? "https://" : "http://") + etcd_address + etcd_api_path;
+        size_t i;
         for (i = 0; i < to.size(); i++)
         {
-            if (to[i] == addr)
+            if (to[i] == check_addr)
                 break;
         }
         if (i >= to.size())
-            to.push_back(addr);
+        {
+            to.push_back(check_addr);
+            // Check if it's a domain name
+            sockaddr_storage ss;
+            bool is_name = !is_local && !string_to_addr(etcd_address, true, 0, &ss);
+            auto & to_addr = (is_local ? etcd_local_addr_urls : (is_name ? etcd_name_urls : etcd_nonlocal_addr_urls));
+            to_addr.push_back((http_url_t){ .ssl = ssl, .addr = etcd_address, .hostname = etcd_address, .path = etcd_api_path });
+        }
     }
 }
 
@@ -219,6 +268,9 @@ void etcd_state_client_t::parse_config(const json11::Json & config)
 {
     this->etcd_local.clear();
     this->etcd_addresses.clear();
+    this->etcd_local_addr_urls.clear();
+    this->etcd_nonlocal_addr_urls.clear();
+    this->etcd_name_urls.clear();
     if (config["etcd_address"].is_string())
     {
         std::string ea = config["etcd_address"].string_value();
@@ -239,6 +291,22 @@ void etcd_state_client_t::parse_config(const json11::Json & config)
             add_etcd_url(ea.string_value());
         }
     }
+    if (this->etcd_client_cert != "")
+    {
+        this->etcd_client_cert = config["etcd_client_cert"].string_value();
+        this->etcd_client_key = config["etcd_client_key"].string_value();
+    }
+    else if (this->osd_num)
+    {
+        this->etcd_client_cert = config["osd_cert"].string_value();
+        this->etcd_client_key = config["osd_pkey"].string_value();
+    }
+    else
+    {
+        this->etcd_client_cert = config["cert"].string_value();
+        this->etcd_client_key = config["pkey"].string_value();
+    }
+    this->etcd_ca = config["etcd_ca"].string_value();
     this->etcd_prefix = config["etcd_prefix"].string_value();
     if (this->etcd_prefix == "")
     {
@@ -291,66 +359,130 @@ void etcd_state_client_t::parse_config(const json11::Json & config)
     }
 }
 
-void etcd_state_client_t::pick_next_etcd()
-{
-    if (selected_etcd_address != "")
-        return;
-    if (addresses_to_try.size() == 0)
-    {
-        // Prefer local etcd, if any
-        for (int i = 0; i < etcd_local.size(); i++)
-            addresses_to_try.push_back(etcd_local[i]);
-        std::vector<int> ns;
-        for (int i = 0; i < etcd_addresses.size(); i++)
-            ns.push_back(i);
-        if (!rand_initialized)
-        {
-            timespec tv;
-            clock_gettime(CLOCK_REALTIME, &tv);
-            srand48(tv.tv_sec*1000000000 + tv.tv_nsec);
-            rand_initialized = true;
-        }
-        while (ns.size())
-        {
-            int i = lrand48() % ns.size();
-            addresses_to_try.push_back(etcd_addresses[ns[i]]);
-            ns.erase(ns.begin()+i, ns.begin()+i+1);
-        }
-    }
-    selected_etcd_address = addresses_to_try[0];
-    addresses_to_try.erase(addresses_to_try.begin(), addresses_to_try.begin()+1);
-}
-
-void etcd_state_client_t::start_etcd_watcher()
+void etcd_state_client_t::pick_next_etcd(std::function<void()> cb)
 {
     if (!etcd_addresses.size() && !etcd_local.size())
     {
         fprintf(stderr, "etcd_address is missing in Vitastor configuration\n");
         exit(1);
     }
-    pick_next_etcd();
-    std::string etcd_address = selected_etcd_address;
-    std::string etcd_api_path;
-    int pos = etcd_address.find('/');
-    if (pos >= 0)
+    if (selected_etcd_url.addr != "")
     {
-        etcd_api_path = etcd_address.substr(pos);
-        etcd_address = etcd_address.substr(0, pos);
+        cb();
+        return;
     }
+    if (etcd_urls_to_try.size() != 0)
+    {
+        selected_etcd_url = std::move(etcd_urls_to_try[0]);
+        etcd_urls_to_try.erase(etcd_urls_to_try.begin());
+        cb();
+        return;
+    }
+    on_resolve_queue.push_back(std::move(cb));
+    if (on_resolve_queue.size() > 1)
+    {
+        // Already resolving
+        return;
+    }
+    assert(!resolve_count);
+    local_to_try = 0;
+    for (auto & url: etcd_local_addr_urls)
+    {
+        // Prefer local IPs, if any
+        etcd_urls_to_try.push_back(url);
+        local_to_try++;
+    }
+    for (auto & url: etcd_nonlocal_addr_urls)
+    {
+        etcd_urls_to_try.push_back(url);
+    }
+    resolve_count++;
+    for (auto & url: etcd_name_urls)
+    {
+        resolve_count++;
+        http_resolve(get_http_ctx(), url.ssl, url.addr, [this, url](const std::string & error, const std::vector<std::string>& addresses)
+        {
+            if (error != "")
+                fprintf(stderr, "Error resolving %s: %s\n", url.addr.c_str(), error.c_str());
+            for (auto & addr: addresses)
+            {
+                auto url_copy = url;
+                url_copy.addr = addr;
+                if (local_ips.find(addr) != local_ips.end())
+                {
+                    etcd_urls_to_try.insert(etcd_urls_to_try.begin(), std::move(url_copy));
+                    local_to_try++;
+                }
+                else
+                    etcd_urls_to_try.push_back(std::move(url_copy));
+            }
+            resolve_count--;
+            if (!resolve_count)
+                pick_next_etcd_on_resolve();
+        });
+    }
+    resolve_count--;
+    if (!resolve_count)
+    {
+        pick_next_etcd_on_resolve();
+    }
+}
+
+void etcd_state_client_t::pick_next_etcd_on_resolve()
+{
+    if (!etcd_urls_to_try.size())
+    {
+        fprintf(stderr, "None of etcd_address could be resolved\n");
+        exit(1);
+    }
+    if (!rand_initialized)
+    {
+        timespec tv;
+        clock_gettime(CLOCK_REALTIME, &tv);
+        srand48(tv.tv_sec*1000000000 + tv.tv_nsec);
+        rand_initialized = true;
+    }
+    // Shuffle addresses
+    for (size_t i = etcd_urls_to_try.size()-1; i > local_to_try; i--)
+    {
+        size_t j = local_to_try + lrand48() % (i - local_to_try);
+        if (j != i)
+            std::swap(etcd_urls_to_try[i], etcd_urls_to_try[j]);
+    }
+    selected_etcd_url = std::move(etcd_urls_to_try[0]);
+    etcd_urls_to_try.erase(etcd_urls_to_try.begin());
+    auto cbs = std::move(on_resolve_queue);
+    for (auto cb: cbs)
+    {
+        cb();
+    }
+}
+
+void etcd_state_client_t::start_etcd_watcher()
+{
+    pick_next_etcd([this]()
+    {
+        start_etcd_watcher_selected();
+    });
+}
+
+void etcd_state_client_t::start_etcd_watcher_selected()
+{
+    const auto & url = selected_etcd_url;
     etcd_watches_initialised = 0;
     ws_alive = 1;
-    if (etcd_watch_ws)
-    {
-        http_close(etcd_watch_ws);
-        etcd_watch_ws = NULL;
-    }
     if (this->log_level > 1)
     {
-        fprintf(stderr, "Trying to connect to etcd websocket at %s, watch from revision %ju/%ju/%ju\n", etcd_address.c_str(),
+        fprintf(stderr, "Trying to connect to etcd websocket at %s%s%s (hostname %s), watch from revision %ju/%ju/%ju\n",
+            url.ssl ? "https://" : "http://", url.addr.c_str(), url.path.c_str(), url.hostname.c_str(),
             etcd_watch_revision_config, etcd_watch_revision_osd, etcd_watch_revision_pg);
     }
-    etcd_watch_ws = open_websocket(tfd, etcd_address, etcd_api_path+"/watch", etcd_slow_timeout,
-        [this, cur_addr = selected_etcd_address](const http_response_t *msg)
+    if (!etcd_watch_ws)
+        etcd_watch_ws = http_init(get_http_ctx());
+    else
+        http_close(etcd_watch_ws);
+    open_websocket(etcd_watch_ws, url.addr, url.hostname, url.path+"/watch", { .timeout = etcd_slow_timeout, .ssl = url.ssl },
+        [this, cur_addr = url.addr](http_message_t *msg)
     {
         if (msg->body.length())
         {
@@ -393,7 +525,6 @@ 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 +569,7 @@ 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 +599,8 @@ void etcd_state_client_t::start_etcd_watcher()
         if (msg->eof)
         {
             fprintf(stderr, "Disconnected from etcd %s\n", cur_addr.c_str());
-            if (cur_addr == selected_etcd_address)
-                selected_etcd_address = "";
-            if (etcd_watch_ws)
-            {
-                http_close(etcd_watch_ws);
-                etcd_watch_ws = NULL;
-            }
+            if (cur_addr == selected_etcd_url.addr)
+                selected_etcd_url = (http_url_t){};
             if (etcd_watches_initialised == 0)
             {
                 // Connection not established, retry in <etcd_quick_timeout>
@@ -549,12 +675,7 @@ void etcd_state_client_t::start_ws_keepalive()
             {
                 if (this->log_level > 0)
                 {
-                    fprintf(stderr, "Websocket ping failed, disconnecting from etcd %s\n", selected_etcd_address.c_str());
-                }
-                if (etcd_watch_ws)
-                {
-                    http_close(etcd_watch_ws);
-                    etcd_watch_ws = NULL;
+                    fprintf(stderr, "Websocket ping failed, disconnecting from etcd %s\n", selected_etcd_url.addr.c_str());
                 }
                 start_etcd_watcher();
             }
@@ -654,12 +775,8 @@ void etcd_state_client_t::load_pgs()
     json11::Json::array txn = {
         json11::Json::object {
             { "request_range", json11::Json::object {
-                { "key", base64_encode(etcd_prefix+"/config/pools") },
-            } }
-        },
-        json11::Json::object {
-            { "request_range", json11::Json::object {
-                { "key", base64_encode(etcd_prefix+"/config/pgs") },
+                { "key", base64_encode(etcd_prefix+"/config/") },
+                { "range_end", base64_encode(etcd_prefix+"/config0") },
             } }
         },
         json11::Json::object {
@@ -667,12 +784,6 @@ void etcd_state_client_t::load_pgs()
                 { "key", base64_encode(etcd_prefix+"/pg/config") },
             } }
         },
-        json11::Json::object {
-            { "request_range", json11::Json::object {
-                { "key", base64_encode(etcd_prefix+"/config/inode/") },
-                { "range_end", base64_encode(etcd_prefix+"/config/inode0") },
-            } }
-        },
         json11::Json::object {
             { "request_range", json11::Json::object {
                 { "key", base64_encode(etcd_prefix+"/pg/history/") },
@@ -942,6 +1053,8 @@ 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")
@@ -1185,7 +1298,6 @@ void etcd_state_client_t::parse_state(const etcd_kv_t & kv)
                 if (i >= pg_state_bit_count)
                 {
                     fprintf(stderr, "Unexpected pool %u PG %u state keyword in etcd: %s\n", pool_id, pg_num, e.dump().c_str());
-                    return;
                 }
             }
             if (!cur_primary || !value["state"].is_array() || !state ||
@@ -1194,7 +1306,6 @@ void etcd_state_client_t::parse_state(const etcd_kv_t & kv)
                 (state & PG_INCOMPLETE) && state != PG_INCOMPLETE && state != (PG_INCOMPLETE|PG_HAS_INVALID))
             {
                 fprintf(stderr, "Unexpected pool %u PG %u state in etcd: primary=%ju, state=%s\n", pool_id, pg_num, cur_primary, value["state"].dump().c_str());
-                return;
             }
             pg_cfg.cur_primary = cur_primary;
             pg_cfg.cur_state = state;
@@ -1261,37 +1372,15 @@ void etcd_state_client_t::parse_state(const etcd_kv_t & kv)
                 {
                     on_inode_change_hook(inode_num, true);
                 }
+                if (this->inode_config.find(inode_num) != this->inode_config.end())
+                {
+                    user_perm_cache_revision = kv.mod_revision;
+                }
                 this->inode_config.erase(inode_num);
             }
             else
             {
-                inode_t parent_inode_num = value["parent_id"].uint64_value();
-                if (parent_inode_num && !(parent_inode_num >> (64-POOL_ID_BITS)))
-                {
-                    uint64_t parent_pool_id = value["parent_pool"].uint64_value();
-                    if (!parent_pool_id)
-                        parent_inode_num |= pool_id << (64-POOL_ID_BITS);
-                    else if (parent_pool_id >= POOL_ID_MAX)
-                    {
-                        fprintf(
-                            stderr, "Inode %ju/%ju parent_pool value is invalid, ignoring parent setting\n",
-                            inode_num >> (64-POOL_ID_BITS), inode_num & (((uint64_t)1 << (64-POOL_ID_BITS)) - 1)
-                        );
-                        parent_inode_num = 0;
-                    }
-                    else
-                        parent_inode_num |= parent_pool_id << (64-POOL_ID_BITS);
-                }
-                insert_inode_config((inode_config_t){
-                    .num = inode_num,
-                    .name = value["name"].string_value(),
-                    .size = value["size"].uint64_value(),
-                    .parent_id = parent_inode_num,
-                    .readonly = value["readonly"].bool_value(),
-                    .deleted = value["deleted"].bool_value(),
-                    .meta = value["meta"],
-                    .mod_revision = kv.mod_revision,
-                });
+                insert_inode_config(deserialize_inode_cfg(inode_num, kv.value, kv.mod_revision));
             }
         }
     }
@@ -1302,6 +1391,38 @@ void etcd_state_client_t::parse_state(const etcd_kv_t & kv)
         if (on_change_node_placement_hook)
             on_change_node_placement_hook();
     }
+    else if (key.substr(0, etcd_prefix.length()+13) == etcd_prefix+"/config/user/")
+    {
+        // <etcd_prefix>/config/user/<username>
+        auto name = key.substr(etcd_prefix.length()+13);
+        auto & inf = user_info[name];
+        if (!value.is_object())
+        {
+            if (inf)
+            {
+                inf->type = user_type_t::CLIENT;
+                inf->groups.clear();
+                inf->perm_cache.clear();
+            }
+            user_info.erase(name);
+        }
+        else
+        {
+            if (!inf)
+            {
+                inf = std::make_shared<user_info_t>();
+                inf->name = name;
+            }
+            inf->type = value["type"] == "admin" ? user_type_t::ADMIN : user_type_t::CLIENT;
+            inf->groups.clear();
+            for (auto & group: value["groups"].array_items())
+            {
+                if (group.string_value() != "")
+                    inf->groups.insert(group.string_value());
+            }
+            inf->perm_cache.clear();
+        }
+    }
 }
 
 uint32_t etcd_state_client_t::parse_immediate_commit(const std::string & immediate_commit_str, uint32_t default_value)
@@ -1324,7 +1445,12 @@ uint32_t etcd_state_client_t::parse_scheme(const std::string & scheme)
 
 void etcd_state_client_t::insert_inode_config(const inode_config_t & cfg)
 {
-    this->inode_config[cfg.num] = cfg;
+    auto & cfg_ref = this->inode_config[cfg.num];
+    if (cfg_ref.mod_revision != cfg.mod_revision)
+    {
+        user_perm_cache_revision = cfg.mod_revision;
+    }
+    cfg_ref = cfg;
     if (cfg.name != "")
     {
         this->inode_by_name[cfg.name] = cfg.num;
@@ -1380,6 +1506,10 @@ 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 +1518,18 @@ 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 +1537,53 @@ json11::Json::object etcd_state_client_t::serialize_inode_cfg(inode_config_t *cf
     return new_cfg;
 }
 
+inode_config_t etcd_state_client_t::deserialize_inode_cfg(uint64_t inode_num, json11::Json value, uint64_t mod_revision)
+{
+    inode_t parent_inode_num = value["parent_id"].uint64_value();
+    if (parent_inode_num && !INODE_POOL(parent_inode_num))
+    {
+        uint64_t parent_pool_id = value["parent_pool"].uint64_value();
+        if (!parent_pool_id)
+            parent_inode_num = INODE_WITH_POOL(INODE_POOL(inode_num), parent_inode_num);
+        else if (parent_pool_id >= POOL_ID_MAX)
+        {
+            fprintf(
+                stderr, "Inode %u/%ju parent_pool value is invalid, ignoring parent setting\n",
+                INODE_POOL(inode_num), INODE_NO_POOL(inode_num)
+            );
+            parent_inode_num = 0;
+        }
+        else
+            parent_inode_num |= parent_pool_id << (64-POOL_ID_BITS);
+    }
+    std::string enc_key;
+    if (!value["enc_key"].is_null())
+    {
+        enc_key = value["enc_key"].string_value();
+        if (enc_key.substr(0, strlen(VAULT_KEY_PREFIX)) != VAULT_KEY_PREFIX &&
+            (enc_key.size() != 2*AES_256_XTS_KEY_SIZE || !ishexstr(enc_key)))
+        {
+            enc_key = "";
+            fprintf(stderr, "Inode %u/%ju has invalid enc_key, should be %u bit hex string or Vault key reference\n",
+                INODE_POOL(inode_num), INODE_NO_POOL(inode_num), AES_256_XTS_KEY_SIZE);
+        }
+    }
+    return (inode_config_t){
+        .num = inode_num,
+        .name = value["name"].string_value(),
+        .size = value["size"].uint64_value(),
+        .parent_id = parent_inode_num,
+        .readonly = value["readonly"].bool_value(),
+        .deleted = value["deleted"].bool_value(),
+        .enc_key = std::move(enc_key),
+        .owner = value["owner"].string_value(),
+        .owner_group = value["owner_group"].string_value(),
+        .reader_group = value["reader_group"].string_value(),
+        .meta = value["meta"],
+        .mod_revision = mod_revision,
+    };
+}
+
 int etcd_state_client_t::address_count()
 {
     return etcd_addresses.size() + etcd_local.size();

src/client/etcd_state_client.h

@@ -4,10 +4,12 @@
 #pragma once
 
 #include <set>
+#include <memory>
 
 #include "json11/json11.hpp"
 #include "object_id.h"
 #include "timerfd_manager.h"
+#include "../util/robin_hood.h"
 
 #define ETCD_CONFIG_WATCH_ID 1
 #define ETCD_OSD_STATE_WATCH_ID 2
@@ -19,6 +21,8 @@
 #define MAX_DATA_BLOCK_SIZE 128*1024*1024
 #define DEFAULT_BITMAP_GRANULARITY 4096
 
+#define VAULT_KEY_PREFIX "vault:"
+
 #ifndef IMMEDIATE_NONE
 #define IMMEDIATE_NONE 0
 #define IMMEDIATE_SMALL 1
@@ -66,6 +70,7 @@ struct pool_config_t
     std::map<pg_num_t, pg_config_t> pg_config;
     uint64_t scrub_interval = 0;
     std::string used_for_app;
+    std::string creator_group;
     int backfillfull = 0;
     int local_reads = 0;
 
@@ -83,6 +88,9 @@ struct inode_config_t
     inode_t parent_id = 0;
     bool readonly = false;
     bool deleted = false;
+    std::string enc_key;
+    // Permissions
+    std::string owner, owner_group, reader_group;
     // Arbitrary metadata
     json11::Json meta;
     // Change revision of the metadata in etcd
@@ -95,23 +103,65 @@ struct inode_watch_t
     inode_config_t cfg = {};
 };
 
+struct http_url_t
+{
+    bool ssl;
+    std::string addr;
+    std::string hostname;
+    std::string path;
+};
+
+enum class user_type_t
+{
+    CLIENT = 0,
+    ADMIN = 1,
+};
+
+struct user_perm_t
+{
+    enum class perm_type_t: uint8_t;
+    constexpr static perm_type_t DENY = (perm_type_t)0;
+    constexpr static perm_type_t READER = (perm_type_t)1;
+    constexpr static perm_type_t OWNER = (perm_type_t)2;
+    uint64_t mod_revision = 0;
+    perm_type_t perm = DENY;
+};
+
+struct user_info_t
+{
+    std::string name;
+    user_type_t type;
+    robin_hood::unordered_flat_set<std::string> groups;
+    robin_hood::unordered_flat_map<inode_t, user_perm_t> perm_cache;
+};
+
 struct http_co_t;
+struct http_context_t;
 
 struct __attribute__((visibility("default"))) etcd_state_client_t
 {
 protected:
-    std::vector<std::string> local_ips;
-    std::vector<std::string> etcd_addresses;
+    std::set<std::string> local_ips;
     std::vector<std::string> etcd_local;
-    std::string selected_etcd_address;
-    std::vector<std::string> addresses_to_try;
+    std::vector<std::string> etcd_addresses;
+    std::vector<http_url_t> etcd_local_addr_urls;
+    std::vector<http_url_t> etcd_nonlocal_addr_urls;
+    std::vector<http_url_t> etcd_name_urls;
+    size_t local_to_try = 0;
+    std::vector<http_url_t> etcd_urls_to_try;
+    http_url_t selected_etcd_url;
+    size_t resolve_count = 0;
     std::vector<inode_watch_t*> watches;
+    std::vector<std::function<void()>> on_resolve_queue;
     bool new_pg_config = false;
     int ws_keepalive_timer = -1;
     int ws_alive = 0;
     bool rand_initialized = false;
     void add_etcd_url(std::string);
-    void pick_next_etcd();
+    void pick_next_etcd(std::function<void()> cb);
+    void pick_next_etcd_on_resolve();
+    void etcd_call_selected(const std::string & api, json11::Json payload, int timeout, int retries, int interval, std::function<void(std::string, json11::Json)> callback);
+    void start_etcd_watcher_selected();
 public:
     int etcd_keepalive_timeout = 30;
     int etcd_ws_keepalive_interval = 5;
@@ -124,10 +174,15 @@ public:
     uint32_t global_bitmap_granularity = DEFAULT_BITMAP_GRANULARITY;
     uint32_t global_immediate_commit = IMMEDIATE_NONE;
 
+    uint64_t osd_num = 0;
     std::string etcd_prefix;
+    std::string etcd_client_cert;
+    std::string etcd_client_key;
+    std::string etcd_ca;
     int log_level = 0;
     timerfd_manager_t *tfd = NULL;
 
+    http_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 +195,8 @@ 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;
+    robin_hood::unordered_flat_map<std::string, std::shared_ptr<user_info_t>> user_info;
+    uint64_t user_perm_cache_revision = 0;
     json11::Json node_placement;
 
     std::function<void(std::map<std::string, etcd_kv_t> &)> on_change_hook;
@@ -158,10 +215,14 @@ public:
     std::function<void(http_co_t *)> on_start_watcher_hook;
 
     json11::Json::object serialize_inode_cfg(inode_config_t *cfg);
+    inode_config_t deserialize_inode_cfg(uint64_t inode_num, json11::Json value, uint64_t mod_revision);
     etcd_kv_t parse_etcd_kv(const json11::Json & kv_json);
     std::vector<std::string> get_addresses();
-    void etcd_call_oneshot(std::string etcd_address, std::string api, json11::Json payload, int timeout, std::function<void(std::string, json11::Json)> callback);
-    void etcd_call(std::string api, json11::Json payload, int timeout, int retries, int interval, std::function<void(std::string, json11::Json)> callback);
+    std::shared_ptr<user_info_t> get_user(const std::string & username);
+    bool check_image_perm(const std::shared_ptr<user_info_t> & user_info, inode_t inode_num, bool write);
+    http_context_t *get_http_ctx();
+    void etcd_call_oneshot(const std::string & etcd_address, const std::string & api, json11::Json payload, int timeout, std::function<void(std::string, json11::Json)> callback);
+    void etcd_call(const std::string & api, json11::Json payload, int timeout, int retries, int interval, std::function<void(std::string, json11::Json)> callback);
     void etcd_txn(json11::Json txn, int timeout, int retries, int interval, std::function<void(std::string, json11::Json)> callback);
     void etcd_txn_slow(json11::Json txn, std::function<void(std::string, json11::Json)> callback);
     void start_etcd_watcher();

src/client/http_client.h

@@ -17,14 +17,19 @@
 
 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 +46,26 @@ struct http_response_t
 // Opened websocket or keepalive HTTP connection
 struct http_co_t;
 
-http_co_t* http_init(timerfd_manager_t *tfd);
-http_co_t* open_websocket(timerfd_manager_t *tfd, const std::string & host, const std::string & path,
-    int timeout, std::function<void(const http_response_t *msg)> on_message);
+http_context_t* http_context_init(timerfd_manager_t *tfd, const std::string & ssl_cert, const std::string & ssl_key,
+    const std::string & ssl_ca, bool verify_peer, std::string & error);
+std::string http_context_get_ssl_cn(http_context_t *ctx);
+void http_resolve(http_context_t *ctx, bool ssl, std::string host,
+    std::function<void(const std::string & error, const std::vector<std::string> & addrs)> cb);
+void http_context_destroy(http_context_t *ctx);
+http_co_t* http_init(http_context_t *ctx = NULL);
+void open_websocket(http_co_t *handler, const std::string & addr, const std::string & hostname, const std::string & path,
+    const http_options_t & options, std::function<void(http_message_t *msg)> on_message);
 void http_request(http_co_t *handler, const std::string & host, const std::string & request,
-    const http_options_t & options, std::function<void(const http_response_t *response)> response_callback);
+    const http_options_t & options, std::function<void(http_message_t *response)> response_callback);
+void http_get(http_co_t *handler, const std::string & url, const std::string & headers,
+    const http_options_t & options, std::function<void(http_message_t *response)> response_callback);
+void http_json_post(http_co_t *handler, const std::string & url, json11::Json body, const std::string & headers,
+    const http_options_t & options, std::function<void(http_message_t *response)> response_callback);
 void http_post_message(http_co_t *handler, uint8_t type, const std::string & msg);
+void http_serve(http_co_t *handler, int peer_fd, const http_options_t & options,
+    std::function<void(http_message_t *msg)> request_callback);
+void http_reply(http_co_t *handler, const std::string & reply);
 void http_close(http_co_t *co);
+void http_destroy(http_co_t *co);
+
+#pragma GCC visibility pop

src/client/messenger.cpp

@@ -10,113 +10,15 @@
 #include <stdexcept>
 
 #include "addr_util.h"
+#include "str_util.h"
 #include "messenger.h"
 #ifdef WITH_RDMA
 #include "msgr_rdma.h"
 #endif
 
-#include <sys/poll.h>
-
-msgr_iothread_t::msgr_iothread_t():
-    ring(RINGLOOP_DEFAULT_SIZE, true),
-    thread(&msgr_iothread_t::run, this)
-{
-    eventfd = ring.register_eventfd();
-    if (eventfd < 0)
-    {
-        throw std::runtime_error(std::string("failed to register eventfd: ") + strerror(-eventfd));
-    }
-}
-
-msgr_iothread_t::~msgr_iothread_t()
-{
-    stop();
-}
-
-void msgr_iothread_t::add_sqe(io_uring_sqe & sqe)
-{
-    mu.lock();
-    queue.push_back((iothread_sqe_t){ .sqe = sqe, .data = std::move(*(ring_data_t*)sqe.user_data) });
-    if (queue.size() == 1)
-    {
-        cond.notify_all();
-    }
-    mu.unlock();
-}
-
-void msgr_iothread_t::stop()
-{
-    mu.lock();
-    if (stopped)
-    {
-        mu.unlock();
-        return;
-    }
-    stopped = true;
-    if (outer_loop_data)
-    {
-        outer_loop_data->callback = [](ring_data_t*){};
-    }
-    cond.notify_all();
-    close(eventfd);
-    mu.unlock();
-    thread.join();
-}
-
-void msgr_iothread_t::add_to_ringloop(ring_loop_t *outer_loop)
-{
-    assert(!this->outer_loop || this->outer_loop == outer_loop);
-    io_uring_sqe *sqe = outer_loop->get_sqe();
-    assert(sqe != NULL);
-    this->outer_loop = outer_loop;
-    this->outer_loop_data = ((ring_data_t*)sqe->user_data);
-    io_uring_prep_poll_add(sqe, eventfd, POLLIN);
-    outer_loop_data->callback = [this](ring_data_t *data)
-    {
-        if (data->res < 0)
-        {
-            throw std::runtime_error(std::string("eventfd poll failed: ") + strerror(-data->res));
-        }
-        outer_loop_data = NULL;
-        if (stopped)
-        {
-            return;
-        }
-        add_to_ringloop(this->outer_loop);
-        ring.loop();
-    };
-}
-
-void msgr_iothread_t::run()
-{
-    while (true)
-    {
-        {
-            std::unique_lock<std::mutex> lk(mu);
-            while (!stopped && !queue.size())
-                cond.wait(lk);
-            if (stopped)
-                return;
-            int i = 0;
-            for (; i < queue.size(); i++)
-            {
-                io_uring_sqe *sqe = ring.get_sqe();
-                if (!sqe)
-                    break;
-                ring_data_t *data = ((ring_data_t*)sqe->user_data);
-                *data = std::move(queue[i].data);
-                *sqe = queue[i].sqe;
-                sqe->user_data = (uint64_t)data;
-            }
-            queue.erase(queue.begin(), queue.begin()+i);
-        }
-        // We only want to offload sendmsg/recvmsg. Callbacks will be called in main thread
-        ring.submit();
-    }
-}
-
 void osd_messenger_t::init()
 {
+    init_tls();
 #ifdef WITH_RDMACM
     if (use_rdmacm)
     {
@@ -173,12 +75,7 @@ void osd_messenger_t::init()
     }
     if (ringloop && iothread_count > 0)
     {
-        for (int i = 0; i < iothread_count; i++)
-        {
-            auto iot = new msgr_iothread_t();
-            iothreads.push_back(iot);
-            iot->add_to_ringloop(ringloop);
-        }
+        init_iothreads();
     }
     keepalive_timer_id = tfd->set_timer(1000, true, [this](int)
     {
@@ -220,7 +117,7 @@ void osd_messenger_t::init()
                             .opcode = OSD_OP_PING,
                         },
                     };
-                    op->callback = [this, cl](osd_op_t *op)
+                    op->callback = [this](osd_op_t *op)
                     {
                         auto cl_it = clients.find(op->client_id);
                         if (cl_it == clients.end())
@@ -229,6 +126,7 @@ void osd_messenger_t::init()
                             delete op;
                             return;
                         }
+                        auto cl = cl_it->second;
                         uint64_t fail_client_id = (op->reply.hdr.retval != 0 ? op->client_id : 0);
                         auto fail_osd_num = cl->in_osd_num ? cl->in_osd_num : cl->osd_num;
                         cl->ping_time_remaining = 0;
@@ -271,14 +169,7 @@ osd_messenger_t::~osd_messenger_t()
     {
         stop_client(clients.begin()->first, true);
     }
-    if (iothreads.size())
-    {
-        for (auto iot: iothreads)
-        {
-            delete iot;
-        }
-        iothreads.clear();
-    }
+    destroy_iothreads();
 #ifdef WITH_RDMA
     for (auto rdma_context: rdma_contexts)
     {
@@ -295,10 +186,67 @@ osd_messenger_t::~osd_messenger_t()
         rdmacm_evch = NULL;
     }
 #endif
+    for (auto encrypt_ctx: encrypt_xts_pool)
+    {
+        destroy_aes_xts_encrypt(encrypt_ctx);
+    }
+    for (auto decrypt_ctx: decrypt_xts_pool)
+    {
+        destroy_aes_xts_decrypt(decrypt_ctx);
+    }
+    destroy_tls();
 }
 
-void osd_messenger_t::parse_config(const json11::Json & config)
+static int parse_proto_checksums(const json11::Json & val, int default_value)
 {
+    if (val.is_string())
+    {
+        const auto & str = val.string_value();
+        if (str == "full")
+            return MSGR_CSUM_FULL;
+        else if (str == "payload")
+            return MSGR_CSUM_PAYLOAD;
+        else if (str == "gcm")
+            return MSGR_CSUM_GCM;
+        else if (str == "none")
+            return 0;
+        else if (str == "")
+            return default_value;
+    }
+    else if (val.is_null())
+        return default_value;
+    fprintf(stderr, "proto_checksums should be \"full\", \"payload\", \"gcm\", \"none\""
+        ", \"\" or null (default), but it is: %s\n", val.dump().c_str());
+    exit(1);
+}
+
+void osd_messenger_t::parse_config(const json11::Json & config, bool init)
+{
+    this->max_cipher_pool_size = config["max_cipher_pool_size"].uint64_value();
+    if (!this->max_cipher_pool_size)
+        this->max_cipher_pool_size = 256;
+    this->receive_buffer_size = (uint32_t)config["tcp_header_buffer_size"].uint64_value();
+    if (!this->receive_buffer_size || this->receive_buffer_size > 1024*1024*1024)
+        this->receive_buffer_size = 65536;
+    this->min_zerocopy_send_size = config["min_zerocopy_send_size"].is_null()
+        ? DEFAULT_MIN_ZEROCOPY_SEND_SIZE
+        : (int)config["min_zerocopy_send_size"].int64_value();
+    this->peer_connect_interval = config["peer_connect_interval"].uint64_value();
+    if (!this->peer_connect_interval)
+        this->peer_connect_interval = 5;
+    this->peer_connect_timeout = config["peer_connect_timeout"].uint64_value();
+    if (!this->peer_connect_timeout)
+        this->peer_connect_timeout = 5;
+    this->osd_idle_timeout = config["osd_idle_timeout"].uint64_value();
+    if (!this->osd_idle_timeout)
+        this->osd_idle_timeout = 5;
+    this->osd_ping_timeout = config["osd_ping_timeout"].uint64_value();
+    if (!this->osd_ping_timeout)
+        this->osd_ping_timeout = 5;
+    this->log_level = config["log_level"].uint64_value();
+    // All other parameters are only set on init
+    if (!init)
+        return;
 #ifdef WITH_RDMA
     if (!config["use_rdma"].is_null())
     {
@@ -329,31 +277,27 @@ 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
+    if (!osd_num)
+    {
+        tls_cert = config["cert"].string_value();
+        tls_key = config["pkey"].string_value();
+        osd_tls_ca = config["osd_ca"].string_value();
+    }
+    else
+    {
+        tls_cert = config["osd_cert"].string_value();
+        tls_key = config["osd_pkey"].string_value();
+        osd_tls_ca = config["osd_ca"].string_value();
+        client_tls_ca = config["client_ca"].string_value();
+    }
+    this->use_proto_checksums = parse_proto_checksums(config["proto_checksums"], MSGR_CSUM_PAYLOAD);
+    this->force_proto_checksums = parse_proto_checksums(config["force_proto_checksums"], tls_cert != "" ? MSGR_CSUM_PAYLOAD : 0);
     if (!osd_num)
         this->iothread_count = (uint32_t)config["client_iothread_count"].uint64_value();
     else
         this->iothread_count = (uint32_t)config["osd_iothread_count"].uint64_value();
-    this->receive_buffer_size = (uint32_t)config["tcp_header_buffer_size"].uint64_value();
-    if (!this->receive_buffer_size || this->receive_buffer_size > 1024*1024*1024)
-        this->receive_buffer_size = 65536;
     this->use_sync_send_recv = config["use_sync_send_recv"].bool_value() ||
-        config["use_sync_send_recv"].uint64_value();
-    this->min_zerocopy_send_size = config["min_zerocopy_send_size"].is_null()
-        ? DEFAULT_MIN_ZEROCOPY_SEND_SIZE
-        : (int)config["min_zerocopy_send_size"].int64_value();
-    this->peer_connect_interval = config["peer_connect_interval"].uint64_value();
-    if (!this->peer_connect_interval)
-        this->peer_connect_interval = 5;
-    this->peer_connect_timeout = config["peer_connect_timeout"].uint64_value();
-    if (!this->peer_connect_timeout)
-        this->peer_connect_timeout = 5;
-    this->osd_idle_timeout = config["osd_idle_timeout"].uint64_value();
-    if (!this->osd_idle_timeout)
-        this->osd_idle_timeout = 5;
-    this->osd_ping_timeout = config["osd_ping_timeout"].uint64_value();
-    if (!this->osd_ping_timeout)
-        this->osd_ping_timeout = 5;
-    this->log_level = config["log_level"].uint64_value();
+        config["use_sync_send_recv"].uint64_value() || !ringloop;
     // OSD public & cluster networks
     this->osd_networks.clear();
     if (config["osd_network"].is_string())
@@ -510,7 +454,7 @@ 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 +503,7 @@ void osd_messenger_t::handle_connect_epoll(int peer_fd)
         handle_peer_epoll(peer_fd, epoll_events);
     });
     // Check OSD number
+    init_tls_client(cl);
     check_peer_config(cl);
 }
 
@@ -648,7 +593,12 @@ 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())
     {
@@ -688,6 +638,11 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
             err = true;
             fprintf(stderr, "Failed to get config from OSD %ju (retval=%jd), disconnecting peer\n", cl->osd_num, op->reply.hdr.retval);
         }
+        else if (cl->gcm_enabled && !cl->hs_result.peer_is_osd)
+        {
+            err = true;
+            fprintf(stderr, "Client %ju is not authenticated as an OSD, disconnecting peer\n", cl->client_id);
+        }
         else
         {
             config = json11::Json::parse(std::string((char*)op->buf), json_err);
@@ -715,6 +670,21 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
                 err = !check_config_hook(cl, config);
             }
         }
+        if (!err && use_proto_checksums)
+        {
+            auto peer_csums = config["features"]["proto_checksums"].uint64_value();
+            if (peer_csums == MSGR_CSUM_GCM && use_proto_checksums == MSGR_CSUM_GCM && cl->gcm_enabled)
+                cl->proto_csum_status = MSGR_CSUM_GCM;
+            else 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;
+            if (cl->proto_csum_status < force_proto_checksums)
+            {
+                fprintf(stderr, "Error: OSD %ju use_proto_checksums security level is lower than force_proto_checksums\n", cl->osd_num);
+                err = true;
+            }
+        }
         if (err)
         {
             osd_num_t peer_osd = cl->osd_num;
@@ -786,7 +756,8 @@ 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);
+        init_tls_client(cl);
         // Add FD to epoll
         tfd->set_fd_handler(peer_fd, false, [this](int peer_fd, int epoll_events)
         {
@@ -829,6 +800,11 @@ bool osd_messenger_t::is_use_rdmacm()
 }
 #endif
 
+bool osd_messenger_t::is_encryption_enabled()
+{
+    return tls_cert != "" || tls_key != "" || osd_tls_ca != "";
+}
+
 json11::Json::object osd_messenger_t::read_config(const json11::Json & config)
 {
     json11::Json::object file_config;

src/client/messenger.h

@@ -12,12 +12,20 @@
 #include <deque>
 #include <vector>
 
+#include <openssl/types.h>
+
+#ifdef WITH_ISAL_CRYPTO
+#include <isa-l_crypto/aes_gcm.h>
+#endif
+
+#include "../util/xxh_x86dispatch.h"
 #include "../util/robin_hood.h"
 #include "malloc_or_die.h"
 #include "json11/json11.hpp"
 #include "msgr_op.h"
 #include "timerfd_manager.h"
 #include "addr_util.h"
+#include "msgr_handshake.h"
 #include <ringloop.h>
 
 #define CL_READ_HDR 1
@@ -31,12 +39,19 @@
 #define PEER_RDMA 4
 #define PEER_STOPPED 5
 
+#define MSGR_CSUM_PAYLOAD 1
+#define MSGR_CSUM_FULL 2
+#define MSGR_CSUM_GCM 4
+#define MSGR_CSUM_NEG 8
+
 #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
+#define AES_256_GCM_KEY_SIZE 32
+#define AES_256_GCM_IV_SIZE 12
+
+#define MAX_SIMPLE_PAYLOAD_SIZE 1048576
 
 struct msgr_sendp_t
 {
@@ -49,6 +64,13 @@ 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 user_info_t;
+
 struct osd_client_t
 {
     uint64_t client_id = 0;
@@ -65,23 +87,49 @@ 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
 
+    bool gcm_enabled = false;
+    msgr_handshake_i *hs = NULL;
+    msgr_handshake_result_t hs_result;
+    std::shared_ptr<user_info_t> user_info;
+    std::vector<uint8_t> my_key, peer_key;
+    uint64_t my_iv_ctr = 0, peer_iv_ctr = 0;
+#ifdef WITH_ISAL_CRYPTO
+    isal_gcm_key_data my_key_isal, peer_key_isal;
+    isal_gcm_context_data *enc_ctx = NULL;
+    isal_gcm_context_data *dec_ctx = NULL;
+#else
+    EVP_CIPHER_CTX *enc_ctx = NULL;
+    EVP_CIPHER_CTX *dec_ctx = NULL;
+#endif
+    uint8_t enc_tag[16];
+    size_t enc_tag_size = 0;
+    uint8_t dec_tag[16];
+    size_t dec_tag_size = 0;
+
     // Read state
+    bool io_error = false;
     int read_ready = 0;
     osd_op_t *read_op = NULL;
+    size_t read_op_size = 0;
+    size_t read_op_pos = 0;
     iovec read_iov = { 0 };
     msghdr read_msg = { 0 };
-    int read_remaining = 0;
-    int read_state = 0;
-    osd_op_buf_list_t recv_list;
+    std::vector<iovec> recv_list;
+    std::vector<int> recv_flags;
     uint64_t read_op_id = 1;
     bool check_sequencing = false;
     bool enable_pg_locks = false;
+    op_aes_xts_decrypt_t *xts_dec_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 +142,17 @@ 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<msgr_sendp_t> outbox, next_outbox;
+    std::vector<iovec> send_list;
+    size_t send_list_size = 0;
+    std::deque<osd_op_t*> send_free_ops;
     std::vector<osd_op_t*> zc_free_list;
+    op_aes_xts_encrypt_t *xts_enc_ctx = NULL;
+    XXH3_state_t* write_csum_state = NULL;
 
     ~osd_client_t();
     void cancel_ops();
@@ -129,43 +183,7 @@ struct osd_op_stats_t
     uint64_t subop_stat_count[OSD_OP_MAX+1] = { 0 };
 };
 
-#include <mutex>
-#include <condition_variable>
-#include <thread>
-
-#ifdef __MOCK__
 class msgr_iothread_t;
-#else
-struct iothread_sqe_t
-{
-    io_uring_sqe sqe;
-    ring_data_t data;
-};
-
-class msgr_iothread_t
-{
-protected:
-    ring_loop_t ring;
-    ring_loop_t *outer_loop = NULL;
-    ring_data_t *outer_loop_data = NULL;
-    int eventfd = -1;
-    bool stopped = false;
-    std::mutex mu;
-    std::condition_variable cond;
-    std::vector<iothread_sqe_t> queue;
-    std::thread thread;
-
-    void run();
-public:
-
-    msgr_iothread_t();
-    ~msgr_iothread_t();
-
-    void add_sqe(io_uring_sqe & sqe);
-    void stop();
-    void add_to_ringloop(ring_loop_t *outer_loop);
-};
-#endif
 
 #ifdef WITH_RDMA
 struct rdma_event_channel;
@@ -176,9 +194,19 @@ struct osd_messenger_t;
 struct rdmacm_connecting_t;
 #endif
 
+class msgr_op_reader_t;
+class msgr_op_writer_t;
+
 struct __attribute__((visibility("default"))) osd_messenger_t
 {
 protected:
+    friend class copy_op_reader_t;
+    friend class gcm_op_reader_t;
+    friend class get_op_reader_t;
+    friend class copy_op_writer_t;
+    friend class gcm_op_writer_t;
+    friend class get_op_writer_t;
+
     int keepalive_timer_id = -1;
 
     uint32_t receive_buffer_size = 0;
@@ -190,6 +218,12 @@ protected:
     bool use_sync_send_recv = false;
     int min_zerocopy_send_size = DEFAULT_MIN_ZEROCOPY_SEND_SIZE;
     int iothread_count = 0;
+    int max_cipher_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;
@@ -207,19 +241,42 @@ protected:
     robin_hood::unordered_flat_map<rdma_cm_id*, rdmacm_connecting_t*> rdmacm_connecting;
 #endif
 
+    bool gcm_enabled = false;
+    msgr_handshake_ctx_i *hs_ctx = NULL;
+
+    void init_tls();
+    void destroy_tls();
+    void init_tls_client(osd_client_t *cl);
+    bool do_tls_handshake(osd_client_t *cl, bool from_recv = false);
+    bool derive_aes_keys(osd_client_t *cl, bool update_my, bool update_peer);
+
     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_xts_pool;
+    std::vector<op_aes_xts_decrypt_t*> decrypt_xts_pool;
+
+#ifdef WITH_ISAL_CRYPTO
+    std::vector<isal_gcm_context_data*> encrypt_gcm_pool;
+    std::vector<isal_gcm_context_data*> decrypt_gcm_pool;
+#else
+    std::vector<EVP_CIPHER_CTX*> encrypt_gcm_pool;
+    std::vector<EVP_CIPHER_CTX*> decrypt_gcm_pool;
+#endif
+
 public:
     timerfd_manager_t *tfd = NULL;
-    ring_loop_t *ringloop = NULL;
+    ring_loop_i *ringloop = NULL;
     bool has_sendmsg_zc = false;
-    // osd_num_t is only for logging and asserts
     uint64_t next_client_id = 1;
-    osd_num_t osd_num;
+    // osd_num = 0 for client messenger, osd_num > 0 for OSD messenger
+    osd_num_t osd_num = 0;
+    uint32_t clean_entry_bitmap_size = 0;
+    uint32_t bs_block_size = 0;
+    uint32_t max_write_request_size = 0;
     robin_hood::unordered_flat_map<uint64_t, osd_client_t*> clients;
     robin_hood::unordered_flat_map<uint64_t, osd_client_t*> osd_peers;
     robin_hood::unordered_flat_map<int, osd_client_t*> clients_by_fd;
@@ -230,11 +287,14 @@ 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;
+    int force_proto_checksums = 0;
     // op statistics
     osd_op_stats_t stats, recovery_stats;
 
     void init();
-    void parse_config(const json11::Json & config);
+    void init_iothreads();
+    void parse_config(const json11::Json & config, bool init);
     void connect_peer(uint64_t osd_num, json11::Json peer_state);
     void stop_client(uint64_t client_id, bool force_delete = false);
     void destroy_client(osd_client_t *cl);
@@ -243,9 +303,11 @@ public:
     std::function<void(osd_num_t)> repeer_pgs;
     std::function<void(osd_num_t)> break_pg_locks;
     std::function<bool(osd_client_t*, json11::Json)> check_config_hook;
+    std::function<void(osd_client_t*)> handshake_hook;
     void read_requests();
     void send_replies();
     void accept_connections(int listen_fd);
+    void destroy_iothreads();
     ~osd_messenger_t();
 
     static json11::Json::object read_config(const json11::Json & config);
@@ -263,6 +325,7 @@ public:
     rdma_cm_id *rdmacm_listen(const std::string & bind_address, int rdmacm_port, int *bound_port, int log_level);
     void rdmacm_destroy_listener(rdma_cm_id *listener);
 #endif
+    bool is_encryption_enabled();
 
     void inc_op_stats(osd_op_stats_t & stats, uint64_t opcode, timespec & tv_begin, timespec & tv_end, uint64_t len);
     void measure_exec(osd_op_t *cur_op);
@@ -279,20 +342,36 @@ protected:
 
     bool try_send(osd_client_t *cl);
     void handle_send(int result, bool prev, bool more, osd_client_t *cl);
+    bool op_write_to(osd_client_t *cl, msgr_op_writer_t & wr);
+    void next_write_op(osd_client_t *cl);
+    bool op_write_buf(osd_client_t *cl, uint8_t *src, size_t src_len, uint8_t *dst, size_t dst_len, bool skip_csum, size_t & from, size_t & done);
+    bool op_copy_data_to(osd_client_t *cl, uint8_t *dst, size_t dst_len, size_t & from, size_t & done);
+    size_t copy_ops_to(osd_client_t *cl, uint8_t *dst, size_t dst_len);
+    template<typename T> size_t copy_ops_to_with(osd_client_t *cl, uint8_t *dst, size_t dst_len);
 
-    bool handle_read(int result, osd_client_t *cl);
-    bool handle_read_buffer(osd_client_t *cl, void *curbuf, int remain);
-    bool handle_finished_read(osd_client_t *cl);
-    void handle_op_hdr(osd_client_t *cl);
-    bool handle_reply_hdr(osd_client_t *cl);
-    void handle_reply_ready(osd_op_t *op);
+    void handle_read(int result, osd_client_t *cl);
+    bool handle_read_buffer(osd_client_t *cl, uint8_t *curbuf, size_t bufsize);
+    template<typename T> bool handle_buffer_with(osd_client_t *cl, uint8_t *curbuf, size_t bufsize);
+    bool handle_hdr(osd_client_t *cl);
+    bool allocate_op_buffers(osd_client_t *cl);
+    bool allocate_reply_buffers(osd_client_t *cl, osd_op_t *op);
+    bool op_read_from(osd_client_t *cl, msgr_op_reader_t & rdr);
+    bool handle_finished_op(osd_client_t *cl);
     void handle_immediate_ops();
 
+    void op_encrypted_copy_buf(osd_client_t *cl, uint8_t *enc_buf, size_t enc_len, uint8_t *plain, size_t plain_len, size_t & done_plain, size_t & done_enc);
+    void op_encrypt_free(osd_client_t* cl);
+    void op_decrypted_copy_buf(osd_client_t *cl, uint8_t *enc_buf, size_t enc_len, uint8_t *plain, size_t plain_len, size_t & done_plain, size_t & done_enc);
+    void op_decrypt_start(osd_client_t* cl);
+    void op_decrypt_inline(osd_client_t* cl);
+    void op_decrypt_free(osd_client_t* cl);
+
 #ifdef WITH_RDMA
     void try_send_rdma(osd_client_t *cl);
     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);
+    void destroy_rdma_conn(msgr_rdma_connection_t *rdma_conn);
 #endif
 #ifdef WITH_RDMACM
     void handle_rdmacm_events();

src/client/msgr_encrypt.cpp

@@ -0,0 +1,661 @@
+// Copyright (c) Vitaliy Filippov, 2026+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#include <assert.h>
+
+#ifdef WITH_ISAL_CRYPTO
+#include <isa-l_crypto/isal_crypto_api.h>
+#endif
+
+#include "str_util.h"
+#include "etcd_state_client.h"
+#include "messenger.h"
+#include "msgr_encrypt.h"
+#include "http_client.h"
+#include "openssl_util.h"
+
+#include <openssl/ssl.h>
+#include <openssl/err.h>
+
+op_aes_xts_encrypt_t::op_aes_xts_encrypt_t()
+{
+#ifndef WITH_ISAL_CRYPTO
+    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();
+    }
+#endif
+}
+
+op_aes_xts_encrypt_t::~op_aes_xts_encrypt_t()
+{
+    assert(!encrypted);
+#ifndef WITH_ISAL_CRYPTO
+    EVP_CIPHER_CTX_free(ctx);
+#endif
+    if (tmp)
+        free(tmp);
+}
+
+void op_aes_xts_encrypt_t::start(osd_client_t *cl, uint8_t *key, uint64_t start_offset, size_t block_size)
+{
+    assert(!encrypted);
+    this->cl = cl;
+    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;
+    }
+#ifndef WITH_ISAL_CRYPTO
+    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)
+{
+    uint8_t iv[16] = { 0 };
+    *((uint64_t*)iv) = start_offset + offset - offset%block_size;
+#ifdef WITH_ISAL_CRYPTO
+    int r = isal_aes_xts_enc_256(key+32, key, iv, block_size, in, out);
+    assert(r == 0 || r == ISAL_CRYPTO_ERR_XTS_SAME_KEYS);
+#else
+    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
+}
+
+static inline void copy_or_gcm(osd_client_t *cl, uint8_t *out, uint8_t *in, size_t n)
+{
+    if (cl->proto_csum_status != MSGR_CSUM_GCM)
+        memcpy(out, in, n);
+    else
+    {
+#ifdef WITH_ISAL_CRYPTO
+        int r = isal_aes_gcm_enc_256_update(&cl->my_key_isal, cl->enc_ctx, out, in, n);
+        assert(!r);
+#else
+        int actual_out;
+        if (EVP_EncryptUpdate(cl->enc_ctx, out, &actual_out, in, n) != 1)
+        {
+            fprintf(stderr, "EncryptUpdate error: ");
+            ERR_print_errors_fp(stderr);
+            abort();
+        }
+        assert(actual_out == n);
+#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;
+        copy_or_gcm(cl, out, tmp + tmp_pos, max_out);
+        done_out += max_out;
+        tmp_pos += max_out;
+        if (tmp_pos >= block_size)
+        {
+            encrypted = false;
+            done_in += 1;
+        }
+    }
+    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;
+        copy_or_gcm(cl, out, tmp, max_out);
+        tmp_pos = max_out;
+        done_in += max_in-1;
+        offset += max_in;
+        done_out += max_out;
+    }
+    else if (!(offset%block_size))
+    {
+        // Full block - simplest case
+        encrypt_block(in, out);
+        if (cl->proto_csum_status == MSGR_CSUM_GCM)
+            copy_or_gcm(cl, out, out, block_size);
+        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);
+        if (cl->proto_csum_status == MSGR_CSUM_GCM)
+            copy_or_gcm(cl, out, out, block_size);
+        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()
+{
+#ifndef WITH_ISAL_CRYPTO
+    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();
+    }
+#endif
+}
+
+op_aes_xts_decrypt_t::~op_aes_xts_decrypt_t()
+{
+    assert(!decrypted);
+#ifndef WITH_ISAL_CRYPTO
+    EVP_CIPHER_CTX_free(ctx);
+#endif
+    if (tmp)
+        free(tmp);
+}
+
+void op_aes_xts_decrypt_t::start(osd_client_t *cl, uint8_t **key_chain, size_t chain_size, void *key_indexes, uint64_t start_offset, size_t block_size)
+{
+    assert(!decrypted);
+    this->cl = cl;
+    this->start_offset = start_offset;
+    this->key_chain = key_chain;
+    this->chain_size = chain_size;
+    this->key_indexes = key_indexes;
+    this->key_index_bytes = osd_op_rw_t::chain_info_bytes(chain_size);
+    assert(chain_size <= 1 || key_indexes != NULL);
+    this->block_size = block_size;
+    this->offset = 0;
+    this->tmp_pos = 0;
+    if (tmp && tmp_size != block_size)
+    {
+        free(tmp);
+        tmp = NULL;
+        tmp_size = 0;
+    }
+#ifndef WITH_ISAL_CRYPTO
+    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 > 1)
+    {
+        uint32_t key_index = key_index_bytes == 1
+            ? ((uint8_t*)key_indexes)[offset/block_size]
+            : (key_index_bytes == 2
+                ? ((uint16_t*)key_indexes)[offset/block_size]
+                : (key_index_bytes == 4
+                    ? ((uint32_t*)key_indexes)[offset/block_size]
+                    : UINT32_MAX));
+        assert(key_index < chain_size);
+        key = key_chain[key_index];
+    }
+    else
+    {
+        key = key_chain[0];
+    }
+    if (!key)
+    {
+        if (in != out)
+            memcpy(out, in, block_size);
+        return;
+    }
+    uint8_t iv[16] = { 0 };
+    *((uint64_t*)iv) = start_offset + offset - offset%block_size;
+#ifdef WITH_ISAL_CRYPTO
+    int r = isal_aes_xts_dec_256(key+32, key, iv, block_size, in, out);
+    assert(r == 0 || r == ISAL_CRYPTO_ERR_XTS_SAME_KEYS);
+#else
+    if (EVP_DecryptInit_ex(ctx, NULL, NULL, chain_size == 1 ? 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
+}
+
+static inline void gcm_dec(osd_client_t *cl, uint8_t *out, uint8_t *in, size_t n)
+{
+#ifdef WITH_ISAL_CRYPTO
+    int r = isal_aes_gcm_dec_256_update(&cl->peer_key_isal, cl->dec_ctx, out, in, n);
+    assert(!r);
+#else
+    int actual_out;
+    if (EVP_DecryptUpdate(cl->dec_ctx, out, &actual_out, in, n) != 1)
+    {
+        fprintf(stderr, "DecryptUpdate error: ");
+        ERR_print_errors_fp(stderr);
+        abort();
+    }
+    assert(actual_out == n);
+#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;
+            done_in += 1;
+        }
+    }
+    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);
+        if (cl->proto_csum_status == MSGR_CSUM_GCM)
+            gcm_dec(cl, tmp, tmp, block_size);
+        decrypt_block(tmp, tmp);
+        decrypted = true;
+        if (out)
+            memcpy(out, tmp, max_out);
+        tmp_pos = max_out;
+        done_in += max_in-1;
+        offset += max_in;
+        done_out += max_out;
+    }
+    else if (!(offset%block_size))
+    {
+        // Full block - simplest case
+        if (cl->proto_csum_status == MSGR_CSUM_GCM)
+        {
+            if (!tmp)
+            {
+                tmp = (uint8_t*)malloc_or_die(block_size);
+                tmp_size = block_size;
+            }
+            gcm_dec(cl, tmp, in, block_size);
+            if (out)
+                decrypt_block(tmp, out);
+        }
+        else
+            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);
+        if (cl->proto_csum_status == MSGR_CSUM_GCM)
+            gcm_dec(cl, tmp, tmp, block_size);
+        decrypt_block(tmp, out);
+        done_in += max_in;
+        offset += max_in;
+        done_out += block_size;
+    }
+}
+
+void destroy_aes_xts_decrypt(op_aes_xts_decrypt_t *decrypt_ctx)
+{
+    delete decrypt_ctx;
+}
+
+void osd_messenger_t::op_encrypted_copy_buf(osd_client_t *cl, uint8_t *enc_buf, size_t enc_len, uint8_t *plain, size_t plain_len, size_t & done_plain, size_t & done_enc)
+{
+    if (!cl->xts_enc_ctx)
+    {
+        if (encrypt_xts_pool.size())
+        {
+            cl->xts_enc_ctx = encrypt_xts_pool.back();
+            encrypt_xts_pool.pop_back();
+        }
+        else
+            cl->xts_enc_ctx = new op_aes_xts_encrypt_t();
+        assert(cl->write_op->enc->key_chain[0]);
+        cl->xts_enc_ctx->start(cl, cl->write_op->enc->key_chain[0], cl->write_op->req.rw.offset, cl->write_op->enc->bitmap_granularity);
+    }
+    while (done_plain < plain_len && done_enc < enc_len)
+    {
+        size_t done_in = 0;
+        size_t done_out = 0;
+        cl->xts_enc_ctx->update(plain+done_plain, plain_len-done_plain, enc_buf+done_enc, enc_len-done_enc, done_in, done_out);
+        if (cl->write_csum_state && done_out > 0)
+            XXH3_64bits_update(cl->write_csum_state, enc_buf+done_enc, done_out);
+        done_enc += done_out;
+        cl->write_op_pos += done_in;
+        done_plain += done_in;
+    }
+}
+
+void osd_messenger_t::op_decrypted_copy_buf(osd_client_t *cl, uint8_t *enc_buf, size_t enc_len, uint8_t *plain, size_t plain_len, size_t & done_plain, size_t & done_enc)
+{
+    op_decrypt_start(cl);
+    while (done_plain < plain_len && done_enc < enc_len)
+    {
+        size_t done_in = 0;
+        size_t done_out = 0;
+        // plain == NULL means skip output
+        cl->xts_dec_ctx->update(enc_buf+done_enc, enc_len-done_enc, plain ? plain+done_plain : NULL, plain_len-done_plain, done_in, done_out);
+        if (cl->read_csum_state && done_in > 0)
+            XXH3_64bits_update(cl->read_csum_state, enc_buf+done_enc, done_in);
+        done_enc += done_in;
+        cl->read_op_pos += done_out;
+        cl->read_op_inline_decrypt_in += done_in;
+        done_plain += done_out;
+    }
+}
+
+void osd_messenger_t::op_decrypt_start(osd_client_t* cl)
+{
+    if (!cl->xts_dec_ctx)
+    {
+        if (decrypt_xts_pool.size())
+        {
+            cl->xts_dec_ctx = decrypt_xts_pool.back();
+            decrypt_xts_pool.pop_back();
+        }
+        else
+            cl->xts_dec_ctx = new op_aes_xts_decrypt_t();
+        auto & enc = cl->read_op->enc;
+        assert(cl->read_op->req.hdr.opcode == OSD_OP_READ);
+        cl->xts_dec_ctx->start(cl, 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->xts_dec_ctx->update(in, in_len, out, out_len, done_in, done_out);
+        if (done_in >= in_len)
+        {
+            i++;
+            from_in = 0;
+        }
+        else
+            from_in += done_in;
+        if (done_out >= out_len)
+        {
+            j++;
+            from_out = 0;
+        }
+        else
+            from_out += done_out;
+    }
+    assert(j >= op->iov.count);
+}
+
+void osd_messenger_t::op_decrypt_free(osd_client_t* cl)
+{
+    if (cl->xts_dec_ctx)
+    {
+        if (decrypt_xts_pool.size() > max_cipher_pool_size)
+            delete cl->xts_dec_ctx;
+        else
+            decrypt_xts_pool.push_back(cl->xts_dec_ctx);
+        cl->xts_dec_ctx = NULL;
+    }
+}
+
+void osd_messenger_t::op_encrypt_free(osd_client_t* cl)
+{
+    if (cl->xts_enc_ctx)
+    {
+        if (encrypt_xts_pool.size() > max_cipher_pool_size)
+            delete cl->xts_enc_ctx;
+        else
+            encrypt_xts_pool.push_back(cl->xts_enc_ctx);
+        cl->xts_enc_ctx = NULL;
+    }
+}
+
+bool osd_messenger_t::derive_aes_keys(osd_client_t *cl, bool update_my, bool update_peer)
+{
+    if (!cl->hs_result.shared_secret.size())
+    {
+        cl->hs_result = cl->hs->get_result();
+        if (handshake_hook)
+        {
+            handshake_hook(cl);
+        }
+    }
+    std::vector<uint8_t> old_my = cl->my_key, old_peer = cl->peer_key;
+    // Both keys include AES key and iv + xxhash3 secret
+    const auto len = AES_256_GCM_KEY_SIZE + AES_256_GCM_IV_SIZE + XXH_SECRET_DEFAULT_SIZE;
+    cl->my_key.resize(len);
+    cl->peer_key.resize(len);
+    bool ok = true;
+    if (update_my || !old_my.size())
+    {
+        ok = ok && hs_ctx->derive_kdf(cl->hs_result.shared_secret.data(), cl->hs_result.shared_secret.size(),
+            old_my.size() ? old_my.data() : NULL, old_my.size(),
+            cl->is_incoming ? "server key" : "client key", cl->my_key.data(), len);
+#ifdef WITH_ISAL_CRYPTO
+        if (ok)
+            isal_aes_gcm_pre_256(cl->my_key.data(), &cl->my_key_isal);
+#endif
+        cl->my_iv_ctr = 0;
+    }
+    if (update_peer || !old_peer.size())
+    {
+        ok = ok && hs_ctx->derive_kdf(cl->hs_result.shared_secret.data(), cl->hs_result.shared_secret.size(),
+            old_peer.size() ? old_peer.data() : NULL, old_peer.size(),
+            !cl->is_incoming ? "server key" : "client key", cl->peer_key.data(), len);
+#ifdef WITH_ISAL_CRYPTO
+        if (ok)
+            isal_aes_gcm_pre_256(cl->peer_key.data(), &cl->peer_key_isal);
+#endif
+        cl->peer_iv_ctr = 0;
+    }
+    return ok;
+}
+
+void osd_messenger_t::init_tls()
+{
+    if (!tls_cert.empty() || !tls_key.empty() || !osd_tls_ca.empty() || !client_tls_ca.empty())
+    {
+        if (tls_cert.empty() || tls_key.empty() || osd_tls_ca.empty() || osd_num && client_tls_ca.empty())
+        {
+            if (osd_num)
+                fprintf(stderr, "Vitastor transport encryption requires osd_cert, osd_pkey, osd_ca, client_ca options for OSDs\n");
+            else
+                fprintf(stderr, "Vitastor transport encryption requires cert, pkey and osd_ca options\n");
+            exit(1);
+        }
+        else
+        {
+#ifndef __MOCK__
+            gcm_enabled = true;
+            hs_ctx = msgr_handshake_ctx_i::create_ctx();
+            if (!hs_ctx->init(tls_cert, tls_key, osd_tls_ca, client_tls_ca))
+            {
+                fprintf(stderr, "Error: %s\n", hs_ctx->get_error().c_str());
+                exit(1);
+            }
+#endif
+        }
+    }
+}
+
+void osd_messenger_t::init_tls_client(osd_client_t *cl)
+{
+    if (gcm_enabled)
+    {
+        cl->gcm_enabled = true;
+        cl->hs = hs_ctx->create();
+        cl->hs->init(cl->is_incoming);
+        if (cl->hs->out_size())
+        {
+            if (cl->write_state == 0)
+            {
+                cl->write_state = CL_WRITE_READY;
+                write_ready_clients.push_back(cl->client_id);
+            }
+        }
+    }
+}
+
+void osd_messenger_t::destroy_tls()
+{
+#ifdef WITH_ISAL_CRYPTO
+    for (isal_gcm_context_data *ctx: encrypt_gcm_pool)
+    {
+        free(ctx);
+    }
+    for (isal_gcm_context_data *ctx: decrypt_gcm_pool)
+    {
+        free(ctx);
+    }
+#else
+    for (EVP_CIPHER_CTX *ctx: encrypt_gcm_pool)
+    {
+        EVP_CIPHER_CTX_free(ctx);
+    }
+    for (EVP_CIPHER_CTX *ctx: decrypt_gcm_pool)
+    {
+        EVP_CIPHER_CTX_free(ctx);
+    }
+#endif
+    if (hs_ctx)
+    {
+        delete hs_ctx;
+        hs_ctx = NULL;
+    }
+}

src/client/msgr_encrypt.h

@@ -0,0 +1,72 @@
+// Copyright (c) Vitaliy Filippov, 2026+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#include <stdint.h>
+
+#ifdef WITH_ISAL_CRYPTO
+#include <isa-l_crypto/aes_xts.h>
+#endif
+
+#include "../util/xxh_x86dispatch.h"
+#include <openssl/conf.h>
+#include <openssl/evp.h>
+#include <openssl/err.h>
+
+struct osd_client_t;
+
+class op_aes_xts_encrypt_t
+{
+#ifndef WITH_ISAL_CRYPTO
+    EVP_CIPHER_CTX *ctx = NULL;
+#endif
+    osd_client_t *cl = NULL;
+    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();
+
+    void start(osd_client_t *cl, 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
+{
+#ifndef WITH_ISAL_CRYPTO
+    EVP_CIPHER_CTX *ctx = NULL;
+#endif
+    osd_client_t *cl = NULL;
+    uint64_t start_offset = 0;
+    uint8_t **key_chain = NULL;
+    size_t chain_size = 0;
+    void *key_indexes = NULL;
+    int key_index_bytes = 0;
+    size_t offset = 0;
+    size_t block_size = 0;
+    uint8_t *tmp = NULL;
+    size_t tmp_size = 0;
+    size_t tmp_pos = 0;
+    bool decrypted = false;
+
+    void decrypt_block(uint8_t *in, uint8_t *out);
+
+public:
+    op_aes_xts_decrypt_t();
+    ~op_aes_xts_decrypt_t();
+
+    void start(osd_client_t *cl, uint8_t **key_chain, size_t chain_size, void *key_indexes, uint64_t start_offset, size_t block_size);
+    void update(uint8_t *in, size_t max_in, uint8_t *out, size_t max_out, size_t & done_in, size_t & done_out);
+};
+
+void destroy_aes_xts_decrypt(op_aes_xts_decrypt_t *decrypt_ctx);

src/client/msgr_handshake.cpp

@@ -0,0 +1,810 @@
+// Copyright (c) Vitaliy Filippov, 2026+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#include <stdint.h>
+#include <assert.h>
+
+#include <string>
+#include <vector>
+#include <memory>
+
+#include <openssl/conf.h>
+#include <openssl/evp.h>
+#include <openssl/obj_mac.h>
+#include <openssl/ec.h>
+#include <openssl/kdf.h>
+#include <openssl/bio.h>
+#include <openssl/pem.h>
+#include <openssl/err.h>
+
+#include "msgr_handshake.h"
+#include "malloc_or_die.h"
+#include "openssl_util.h"
+#include "str_util.h"
+
+#define AES_256_GCM_KEY_SIZE 32
+#define AES_256_GCM_IV_SIZE 12
+
+// TLS 1.3-like handshake
+
+// 1. Client->server: EC public key
+// 2. Server->client: EC public key, encrypted certificate and digital signature of the handshake
+// 3. Client->server: encrypted certificate and digital signature of the handshake
+
+// "vitaECDH" interleaved
+#define MSGR_HS_MAGIC 0x4861447443694576l
+#define MSGR_HS_MAX_LEN 131072
+
+#define MSGR_HS_SERVER_INIT 0
+#define MSGR_HS_CLIENT_INIT 1
+#define MSGR_HS_SERVER_REPLY 2
+#define MSGR_HS_CLIENT_REPLY 3
+#define MSGR_HS_DONE 100
+#define MSGR_HS_ERROR 101
+
+struct __attribute__((__packed__)) msgr_handshake_hdr_t
+{
+    uint32_t msg_len;
+    uint64_t magic;
+    uint32_t type;
+};
+
+class msgr_handshake_ctx_t: public msgr_handshake_ctx_i
+{
+    friend class msgr_handshake_t;
+
+    EVP_PKEY_CTX *pctx = NULL;
+    EVP_PKEY *params = NULL;
+    X509_STORE *ca = NULL;
+    X509 *osd_ca = NULL;
+    X509 *client_ca = NULL;
+    std::string my_cert_pem;
+    X509 *my_cert = NULL;
+    EVP_PKEY *my_pubkey = NULL;
+    const EVP_MD *md = NULL;
+    EVP_PKEY *my_privkey = NULL;
+    EVP_KDF_CTX* kdf_ctx = NULL;
+    std::string error;
+
+    bool on_error(const std::string & prefix);
+
+public:
+    ~msgr_handshake_ctx_t();
+    msgr_handshake_i* create() override;
+    bool init(const std::string & pem_cert, const std::string & pem_key,
+        const std::string & pem_osd_ca, const std::string & pem_client_ca) override;
+    std::string get_error() override;
+    bool derive_kdf(const uint8_t* insecret, size_t insecret_len,
+        const uint8_t* salt, size_t salt_len, const char *label, uint8_t *outsecret, size_t outsize) override;
+};
+
+class msgr_handshake_t: public msgr_handshake_i
+{
+    msgr_handshake_ctx_t *ctx = NULL;
+    bool is_server = false;
+
+    std::vector<uint8_t> full_handshake;
+    std::vector<uint8_t> in_buf;
+    uint8_t *out_buf = NULL;
+    size_t out_buf_size = 0;
+    std::string error;
+
+    int state = 0;
+
+    EVP_PKEY *ec_key = NULL;
+    X509 *peer_cert = NULL;
+    bool peer_is_osd = false;
+    std::vector<uint8_t> shared_secret;
+    std::vector<uint8_t> hs_key, peer_hs_key;
+
+    msgr_handshake_hdr_t *cur_hdr = NULL;
+    uint8_t *cur_buf = NULL;
+    size_t cur_left = 0;
+
+    bool on_error(const std::string & prefix);
+    bool derive_shared_secret(EVP_PKEY *peer_ec_key);
+    bool derive_hs_keys(const uint8_t *encoded_peer_key, size_t encoded_key_len);
+    bool sign(std::vector<uint8_t> & out);
+    bool verify(const uint8_t *signature, size_t signature_len);
+    bool encrypt(const uint8_t* src, size_t len, uint8_t* dest);
+    bool decrypt(const uint8_t* src, size_t & len, uint8_t* dest);
+    bool make_client_init();
+    bool make_server_reply();
+    bool make_client_reply();
+    bool verify_peer(const uint8_t *peer_cert_pem, size_t peer_cert_len);
+    ssize_t start_msg(uint8_t* src, size_t len, uint32_t expected_type);
+    bool read_with_len(const uint8_t* & dst, uint32_t & dst_len);
+    bool handle_client_init();
+    bool handle_server_reply();
+    bool handle_client_reply();
+    bool handle_peer_cert(const uint8_t *key, uint32_t key_len);
+    void complete();
+
+public:
+    // Workflow: create -> init -> handle -> get_result/get_error -> destruct
+    msgr_handshake_t(msgr_handshake_ctx_t *ctx): ctx(ctx) {}
+    ~msgr_handshake_t();
+    bool init(bool server_mode) override;
+    ssize_t handle(uint8_t* in_buf, size_t in_size) override;
+    bool done() override;
+    uint8_t *get_out() override;
+    size_t out_size() override;
+    void eat_out(size_t n) override;
+    void reset_out() override;
+    msgr_handshake_result_t get_result() override;
+    std::string get_error() override;
+};
+
+msgr_handshake_ctx_i* msgr_handshake_ctx_i::create_ctx()
+{
+    return new msgr_handshake_ctx_t();
+}
+
+msgr_handshake_i* msgr_handshake_ctx_t::create()
+{
+    return new msgr_handshake_t(this);
+}
+
+bool msgr_handshake_ctx_t::init(const std::string & pem_cert, const std::string & pem_key,
+    const std::string & pem_osd_ca, const std::string & pem_client_ca)
+{
+    if (pem_cert.substr(0, 5) == "-----")
+        my_cert_pem = pem_cert;
+    else
+    {
+        my_cert_pem = read_file(pem_cert);
+        if (my_cert_pem.empty())
+        {
+            error = "Failed to load certificate file";
+            return false;
+        }
+    }
+    {
+        BIO *bio = BIO_new_mem_buf(my_cert_pem.data(), my_cert_pem.size());
+        if (!bio)
+            return on_error("BIO_new_mem_buf: ");
+        my_cert = PEM_read_bio_X509(bio, NULL, 0, NULL);
+        BIO_free(bio);
+        if (!my_cert)
+            return on_error("Failed to load certificate: ");
+    }
+    if (!(my_pubkey = X509_get0_pubkey(my_cert)))
+        return on_error("X509_get0_pubkey: ");
+    if (!(md = EVP_get_digestbynid(NID_sha384)))
+        return on_error("EVP_get_digestbynid SHA384: ");
+    if (!(my_privkey = openssl_load_key(pem_key)))
+        return on_error("Failed to load private key: ");
+    if (!(ca = X509_STORE_new()))
+        return on_error("X509_STORE_CTX_new: ");
+    if (!(osd_ca = openssl_load_cert(pem_osd_ca)))
+        return on_error("Failed to load OSD CA certificate: ");
+    if (X509_STORE_add_cert(ca, osd_ca) <= 0)
+        return on_error("X509_STORE_add_cert OSD CA: ");
+    if (!pem_client_ca.empty() && !(client_ca = openssl_load_cert(pem_client_ca)))
+        return on_error("Failed to load client CA certificate: ");
+    if (client_ca && X509_STORE_add_cert(ca, client_ca) <= 0)
+        return on_error("X509_STORE_add_cert client CA: ");
+    if (!(pctx = EVP_PKEY_CTX_new_id(EVP_PKEY_EC, NULL)))
+        return on_error("EVP_PKEY_CTX_new_id: ");
+    if (EVP_PKEY_paramgen_init(pctx) <= 0)
+        return on_error("EVP_PKEY_paramgen_init: ");
+    if (EVP_PKEY_CTX_set_ec_paramgen_curve_nid(pctx, /*NID_X9_62_prime256v1*/NID_secp384r1) <= 0)
+        return on_error("EVP_PKEY_CTX_set_ec_paramgen_curve_nid: ");
+    if (EVP_PKEY_paramgen(pctx, &params) <= 0)
+        return on_error("EVP_PKEY_paramgen: ");
+    EVP_KDF *kdf = EVP_KDF_fetch(NULL, "hkdf", NULL);
+    if (!kdf)
+        return on_error("EVP_KDF_fetch: ");
+    kdf_ctx = EVP_KDF_CTX_new(kdf);
+    EVP_KDF_free(kdf);
+    return true;
+}
+
+std::string msgr_handshake_ctx_t::get_error()
+{
+    return error;
+}
+
+msgr_handshake_ctx_t::~msgr_handshake_ctx_t()
+{
+    if (pctx)
+        EVP_PKEY_CTX_free(pctx);
+    if (params)
+        EVP_PKEY_free(params);
+    if (ca)
+        X509_STORE_free(ca);
+    if (osd_ca)
+        X509_free(osd_ca);
+    if (client_ca)
+        X509_free(client_ca);
+    my_pubkey = NULL;
+    if (my_cert)
+        X509_free(my_cert);
+    if (my_privkey)
+        EVP_PKEY_free(my_privkey);
+    if (kdf_ctx)
+        EVP_KDF_CTX_free(kdf_ctx);
+}
+
+bool msgr_handshake_ctx_t::on_error(const std::string & prefix)
+{
+    error = prefix+ERR_error_string(ERR_get_error(), NULL);
+    return false;
+}
+
+bool msgr_handshake_ctx_t::derive_kdf(const uint8_t* insecret, size_t insecret_len,
+    const uint8_t* salt, size_t salt_len, const char *label, uint8_t *outsecret, size_t outsize)
+{
+    OSSL_PARAM params[5];
+    int n = 0;
+    params[n++] = OSSL_PARAM_construct_utf8_string("digest", (char*)"sha384", (size_t)7);
+    params[n++] = OSSL_PARAM_construct_octet_string("key", (void*)insecret, insecret_len);
+    params[n++] = OSSL_PARAM_construct_octet_string("info", (void*)label, strlen(label)+1);
+    params[n++] = OSSL_PARAM_construct_octet_string("salt", (salt ? (void*)salt : (void*)""), salt_len);
+    params[n++] = OSSL_PARAM_construct_end();
+    assert(n <= sizeof(params)/sizeof(OSSL_PARAM));
+    if (EVP_KDF_CTX_set_params(kdf_ctx, params) <= 0)
+        return false;
+    if (EVP_KDF_derive(kdf_ctx, outsecret, outsize, NULL) <= 0)
+        return false;
+    return true;
+}
+
+msgr_handshake_t::~msgr_handshake_t()
+{
+    if (out_buf)
+        free(out_buf);
+    if (peer_cert)
+        X509_free(peer_cert);
+    if (ec_key)
+        EVP_PKEY_free(ec_key);
+}
+
+bool msgr_handshake_t::on_error(const std::string & prefix)
+{
+    error = prefix+ERR_error_string(ERR_get_error(), NULL);
+    state = MSGR_HS_ERROR;
+    return false;
+}
+
+bool msgr_handshake_t::init(bool server_mode)
+{
+    this->ctx = ctx;
+    std::unique_ptr<EVP_PKEY_CTX, decltype(&EVP_PKEY_CTX_free)> kctx(EVP_PKEY_CTX_new(ctx->params, NULL), EVP_PKEY_CTX_free);
+    if (!kctx)
+        return on_error("EVP_PKEY_CTX_new with EC params: ");
+    if (EVP_PKEY_keygen_init(kctx.get()) <= 0)
+        return on_error("EVP_PKEY_keygen_init: ");
+    if (EVP_PKEY_keygen(kctx.get(), &ec_key) <= 0)
+        return on_error("EVP_PKEY_keygen: ");
+    if (!server_mode)
+    {
+        // Send initial message - only the EC public key
+        if (!make_client_init())
+            return false;
+    }
+    this->is_server = server_mode;
+    this->state = server_mode ? MSGR_HS_SERVER_INIT : MSGR_HS_CLIENT_INIT;
+    return true;
+}
+
+static void copy_to(std::vector<uint8_t> & buf, const void* src, uint32_t len)
+{
+    size_t old_size = buf.size();
+    buf.resize(buf.size() + len);
+    memcpy(buf.data() + old_size, src, len);
+}
+
+static void copy_to_raw(uint8_t* & buf, const void* src, size_t len)
+{
+    memcpy(buf, src, len);
+    buf += len;
+}
+
+static void copy_to_with_len(std::vector<uint8_t> & buf, const void* src, uint32_t len)
+{
+    copy_to(buf, &len, sizeof(len));
+    copy_to(buf, src, len);
+}
+
+bool msgr_handshake_t::derive_shared_secret(EVP_PKEY *peer_ec_key)
+{
+    EVP_PKEY_CTX *dh_ctx = NULL;
+    if (!(dh_ctx = EVP_PKEY_CTX_new(ec_key, NULL)))
+        return on_error("EVP_PKEY_CTX_new for ECDH: ");
+    if (!EVP_PKEY_derive_init(dh_ctx))
+    {
+        EVP_PKEY_CTX_free(dh_ctx);
+        return on_error("EVP_PKEY_derive_init: ");
+    }
+    if (!EVP_PKEY_derive_set_peer(dh_ctx, peer_ec_key))
+    {
+        EVP_PKEY_CTX_free(dh_ctx);
+        return on_error("EVP_PKEY_derive_set_peer: ");
+    }
+    size_t len = 0;
+    if (!EVP_PKEY_derive(dh_ctx, NULL, &len))
+    {
+        EVP_PKEY_CTX_free(dh_ctx);
+        return on_error("EVP_PKEY_derive get length: ");
+    }
+    shared_secret.resize(len);
+    assert(len == 48);
+    if (!EVP_PKEY_derive(dh_ctx, shared_secret.data(), &len))
+    {
+        EVP_PKEY_CTX_free(dh_ctx);
+        return on_error("EVP_PKEY_derive: ");
+    }
+    assert(len == shared_secret.size());
+    shared_secret.resize(len);
+    EVP_PKEY_CTX_free(dh_ctx);
+    return true;
+}
+
+bool msgr_handshake_t::derive_hs_keys(const uint8_t *encoded_peer_key, size_t encoded_key_len)
+{
+    std::unique_ptr<EVP_PKEY, decltype(&EVP_PKEY_free)> peer_ec_key(EVP_PKEY_new(), EVP_PKEY_free);
+    if (!peer_ec_key)
+        return on_error("EVP_PKEY_new: ");
+    if (EVP_PKEY_copy_parameters(peer_ec_key.get(), ec_key) <= 0)
+        return on_error("EVP_PKEY_copy_parameters: ");
+    if (EVP_PKEY_set1_encoded_public_key(peer_ec_key.get(), encoded_peer_key, encoded_key_len) <= 0)
+        return on_error("Invalid handshake peer key: ");
+    if (!derive_shared_secret(peer_ec_key.get()))
+        return false;
+    hs_key.resize(AES_256_GCM_KEY_SIZE + AES_256_GCM_IV_SIZE);
+    peer_hs_key.resize(AES_256_GCM_KEY_SIZE + AES_256_GCM_IV_SIZE);
+    if (!ctx->derive_kdf(shared_secret.data(), shared_secret.size(),
+        NULL, 0, (state == MSGR_HS_SERVER_INIT ? "server hs key" : "client hs key"),
+        hs_key.data(), hs_key.size()))
+        return on_error("derive_kdf: ");
+    if (!ctx->derive_kdf(shared_secret.data(), shared_secret.size(),
+        NULL, 0, (state != MSGR_HS_SERVER_INIT ? "server hs key" : "client hs key"),
+        peer_hs_key.data(), peer_hs_key.size()))
+        return on_error("derive_kdf: ");
+    return true;
+}
+
+bool msgr_handshake_t::sign(std::vector<uint8_t> & out)
+{
+    std::unique_ptr<EVP_MD_CTX, decltype(&EVP_MD_CTX_free)> md_ctx(EVP_MD_CTX_new(), EVP_MD_CTX_free);
+    if (!md_ctx)
+        return on_error("EVP_MD_CTX_create: ");
+    if (EVP_DigestSignInit(md_ctx.get(), NULL, ctx->md, NULL, ctx->my_privkey) <= 0)
+        return on_error("EVP_DigestSignInit: ");
+    if (EVP_DigestSignUpdate(md_ctx.get(), full_handshake.data(), full_handshake.size()) <= 0)
+        return on_error("EVP_DigestSignUpdate: ");
+    size_t siglen = 0;
+    if (EVP_DigestSignFinal(md_ctx.get(), NULL, &siglen) <= 0)
+        return on_error("EVP_DigestSignFinal get length: ");
+    size_t oldsize = out.size();
+    out.resize(oldsize + siglen);
+    if (EVP_DigestSignFinal(md_ctx.get(), out.data() + oldsize, &siglen) <= 0)
+        return on_error("EVP_DigestSignFinal: ");
+    out.resize(oldsize + siglen);
+    return true;
+}
+
+bool msgr_handshake_t::verify(const uint8_t *signature, size_t signature_len)
+{
+    std::unique_ptr<EVP_MD_CTX, decltype(&EVP_MD_CTX_free)> md_ctx(EVP_MD_CTX_new(), EVP_MD_CTX_free);
+    if (!md_ctx)
+        return on_error("EVP_MD_CTX_create: ");
+    if (EVP_DigestVerifyInit(md_ctx.get(), NULL, ctx->md, NULL, X509_get0_pubkey(peer_cert)) <= 0)
+        return on_error("EVP_DigestVerifyInit: ");
+    if (EVP_DigestVerifyUpdate(md_ctx.get(), full_handshake.data(), full_handshake.size()) <= 0)
+        return on_error("EVP_DigestVerifyUpdate: ");
+    if (EVP_DigestVerifyFinal(md_ctx.get(), signature, signature_len) <= 0)
+        return false;
+    return true;
+}
+
+bool msgr_handshake_t::encrypt(const uint8_t* src, size_t len, uint8_t* dest)
+{
+    std::unique_ptr<EVP_CIPHER_CTX, decltype(&EVP_CIPHER_CTX_free)> enc_ctx(EVP_CIPHER_CTX_new(), EVP_CIPHER_CTX_free);
+    if (!ctx)
+        return on_error("EVP_CIPHER_CTX_new: ");
+    if (EVP_EncryptInit_ex(enc_ctx.get(), EVP_aes_256_gcm(), NULL, hs_key.data(), hs_key.data() + AES_256_GCM_KEY_SIZE) <= 0)
+        return on_error("EVP_EncryptInit AES-256-GCM: ");
+    int actual_out;
+    if (EVP_EncryptUpdate(enc_ctx.get(), dest, &actual_out, src, len) <= 0)
+        return on_error("EVP_EncryptUpdate: ");
+    assert(actual_out == len);
+    if (EVP_EncryptFinal_ex(enc_ctx.get(), NULL, &actual_out) <= 0)
+        return on_error("EVP_EncryptFinal: ");
+    if (EVP_CIPHER_CTX_ctrl(enc_ctx.get(), EVP_CTRL_GCM_GET_TAG, 16, dest+len) <= 0)
+        return on_error("EVP_CTRL_GCM_GET_TAG: ");
+    (*(uint64_t*)(hs_key.data() + AES_256_GCM_KEY_SIZE))++; // change IV
+    return true;
+}
+
+bool msgr_handshake_t::decrypt(const uint8_t *src, size_t & len, uint8_t* dest)
+{
+    if (len <= 16) // only tag?!
+    {
+        len = 0;
+        error = "Handshake decryption failed";
+        state = MSGR_HS_ERROR;
+        return false;
+    }
+    std::unique_ptr<EVP_CIPHER_CTX, decltype(&EVP_CIPHER_CTX_free)> dec_ctx(EVP_CIPHER_CTX_new(), EVP_CIPHER_CTX_free);
+    if (!ctx)
+        return on_error("EVP_CIPHER_CTX_new: ");
+    if (EVP_DecryptInit_ex(dec_ctx.get(), EVP_aes_256_gcm(), NULL, peer_hs_key.data(), peer_hs_key.data() + AES_256_GCM_KEY_SIZE) <= 0)
+        return on_error("EVP_DecryptInit AES-256-GCM: ");
+    int actual_out;
+    len -= 16;
+    if (EVP_DecryptUpdate(dec_ctx.get(), dest, &actual_out, src, len) <= 0)
+        return on_error("EVP_DecryptUpdate: ");
+    assert(actual_out == len);
+    if (EVP_CIPHER_CTX_ctrl(dec_ctx.get(), EVP_CTRL_GCM_SET_TAG, 16, (void*)(src+len)) <= 0)
+        return on_error("EVP_CTRL_GCM_SET_TAG: ");
+    if (EVP_DecryptFinal_ex(dec_ctx.get(), NULL, &actual_out) <= 0)
+    {
+        error = "Handshake decryption failed";
+        state = MSGR_HS_ERROR;
+        return false;
+    }
+    (*(uint64_t*)(peer_hs_key.data() + AES_256_GCM_KEY_SIZE))++; // change IV
+    return true;
+}
+
+bool msgr_handshake_t::make_client_init()
+{
+    uint8_t *key = NULL;
+    size_t key_len = EVP_PKEY_get1_encoded_public_key(ec_key, &key);
+    if (!key_len)
+        return on_error("EVP_PKEY_get1_encoded_public_key: ");
+    const size_t old_out_size = out_buf_size;
+    out_buf_size += key_len + sizeof(msgr_handshake_hdr_t);
+    out_buf = (uint8_t*)realloc_or_die(out_buf, out_buf_size);
+    uint8_t *buf = out_buf + old_out_size;
+    msgr_handshake_hdr_t *hdr = (msgr_handshake_hdr_t *)buf;
+    hdr->msg_len = key_len + sizeof(msgr_handshake_hdr_t);
+    hdr->magic = MSGR_HS_MAGIC;
+    hdr->type = MSGR_HS_CLIENT_INIT;
+    memcpy(buf + sizeof(msgr_handshake_hdr_t), key, key_len);
+    copy_to(full_handshake, &hdr->type, sizeof(hdr->type));
+    copy_to_with_len(full_handshake, key, key_len);
+    OPENSSL_free(key);
+    return true;
+}
+
+bool msgr_handshake_t::make_server_reply()
+{
+    uint8_t *key = NULL;
+    size_t key_len = EVP_PKEY_get1_encoded_public_key(ec_key, &key);
+    if (!key_len)
+        return on_error("EVP_PKEY_get1_encoded_public_key: ");
+    // Append type, key and raw certificate to signed data and sign it
+    msgr_handshake_hdr_t hdr = { .magic = MSGR_HS_MAGIC, .type = MSGR_HS_SERVER_REPLY };
+    copy_to(full_handshake, &hdr.type, sizeof(hdr.type));
+    copy_to_with_len(full_handshake, key, key_len);
+    copy_to_with_len(full_handshake, ctx->my_cert_pem.data(), ctx->my_cert_pem.size());
+    std::vector<uint8_t> signature;
+    if (!sign(signature))
+    {
+        OPENSSL_free(key);
+        return false;
+    }
+    // Encrypt certificate and signature
+    std::vector<uint8_t> encrypt_data;
+    copy_to_with_len(encrypt_data, ctx->my_cert_pem.data(), ctx->my_cert_pem.size());
+    copy_to_with_len(encrypt_data, signature.data(), signature.size());
+    encrypt_data.resize(encrypt_data.size()+16);
+    if (!encrypt(encrypt_data.data(), encrypt_data.size()-16, encrypt_data.data()))
+    {
+        OPENSSL_free(key);
+        return false;
+    }
+    // Construct message
+    hdr.msg_len = sizeof(msgr_handshake_hdr_t) + 4 + key_len + encrypt_data.size();
+    out_buf = (uint8_t*)realloc_or_die(out_buf, (out_buf_size += hdr.msg_len));
+    uint8_t *cur = out_buf + out_buf_size - hdr.msg_len;
+    copy_to_raw(cur, &hdr, sizeof(hdr));
+    copy_to_raw(cur, &key_len, 4);
+    copy_to_raw(cur, key, key_len);
+    copy_to_raw(cur, encrypt_data.data(), encrypt_data.size());
+    OPENSSL_free(key);
+    return true;
+}
+
+bool msgr_handshake_t::make_client_reply()
+{
+    // Append type and raw certificate to signed data and sign it
+    msgr_handshake_hdr_t hdr = { .magic = MSGR_HS_MAGIC, .type = MSGR_HS_CLIENT_REPLY };
+    copy_to(full_handshake, &hdr.type, sizeof(hdr.type));
+    copy_to_with_len(full_handshake, ctx->my_cert_pem.data(), ctx->my_cert_pem.size());
+    std::vector<uint8_t> signature;
+    if (!sign(signature))
+        return false;
+    // Encrypt certificate and signature
+    std::vector<uint8_t> encrypt_data;
+    copy_to_with_len(encrypt_data, ctx->my_cert_pem.data(), ctx->my_cert_pem.size());
+    copy_to_with_len(encrypt_data, signature.data(), signature.size());
+    encrypt_data.resize(encrypt_data.size()+16);
+    if (!encrypt(encrypt_data.data(), encrypt_data.size()-16, encrypt_data.data()))
+        return false;
+    // Construct message
+    hdr.msg_len = sizeof(msgr_handshake_hdr_t) + encrypt_data.size();
+    out_buf = (uint8_t*)realloc_or_die(out_buf, (out_buf_size += hdr.msg_len));
+    uint8_t *cur = out_buf + out_buf_size - hdr.msg_len;
+    copy_to_raw(cur, &hdr, sizeof(hdr));
+    copy_to_raw(cur, encrypt_data.data(), encrypt_data.size());
+    return true;
+}
+
+bool msgr_handshake_t::verify_peer(const uint8_t *peer_cert_pem, size_t peer_cert_len)
+{
+    BIO *bio = BIO_new_mem_buf(peer_cert_pem, peer_cert_len);
+    if (!bio)
+        return on_error("BIO_new_mem_buf: ");
+    peer_cert = PEM_read_bio_X509(bio, NULL, 0, NULL);
+    BIO_free(bio);
+    if (!peer_cert)
+    {
+        error = "Invalid peer certificate";
+        state = MSGR_HS_ERROR;
+        return false;
+    }
+    std::unique_ptr<X509_STORE_CTX, decltype(&X509_STORE_CTX_free)> ca_ctx(X509_STORE_CTX_new(), X509_STORE_CTX_free);
+    if (!ca_ctx)
+        return on_error("X509_STORE_CTX_new: ");
+    if (X509_STORE_CTX_init(ca_ctx.get(), ctx->ca, peer_cert, NULL) <= 0)
+        return on_error("X509_STORE_CTX_init: ");
+    // Maybe use X509_VERIFY_PARAM_set_auth_level(X509_STORE_CTX_get0_param(ca_ctx.get()), 2) ?
+    X509_STORE_CTX_set_default(ca_ctx.get(), is_server ? "ssl_client" : "ssl_server");
+    if (X509_verify_cert(ca_ctx.get()) <= 0)
+    {
+        error = "Peer certificate verification failed: ";
+        error += X509_verify_cert_error_string(X509_STORE_CTX_get_error(ca_ctx.get()));
+        state = MSGR_HS_ERROR;
+        return false;
+    }
+    peer_is_osd = (X509_verify(peer_cert, X509_get0_pubkey(ctx->osd_ca)) > 0);
+    if (!is_server && !peer_is_osd)
+    {
+        error = "Peer is not an OSD";
+        state = MSGR_HS_ERROR;
+        return false;
+    }
+    return true;
+}
+
+ssize_t msgr_handshake_t::start_msg(uint8_t* src, size_t len, uint32_t expected_type)
+{
+    size_t orig_len = len;
+    size_t to_buffer = (len < sizeof(msgr_handshake_hdr_t)-in_buf.size()
+        ? len : sizeof(msgr_handshake_hdr_t)-in_buf.size());
+    in_buf.insert(in_buf.end(), src, src+to_buffer);
+    len -= to_buffer;
+    src += to_buffer;
+    if (in_buf.size() < sizeof(msgr_handshake_hdr_t))
+        return 0;
+    cur_hdr = (msgr_handshake_hdr_t *)in_buf.data();
+    if (cur_hdr->magic != MSGR_HS_MAGIC ||
+        cur_hdr->type != expected_type ||
+        cur_hdr->msg_len <= sizeof(msgr_handshake_hdr_t) ||
+        cur_hdr->msg_len >= MSGR_HS_MAX_LEN)
+    {
+        error = "Invalid handshake packet magic, type or size";
+        state = MSGR_HS_ERROR;
+        return -1;
+    }
+    to_buffer = (len < cur_hdr->msg_len-in_buf.size()
+        ? len : cur_hdr->msg_len-in_buf.size());
+    in_buf.insert(in_buf.end(), src, src+to_buffer);
+    cur_hdr = (msgr_handshake_hdr_t *)in_buf.data();
+    len -= to_buffer;
+    src += to_buffer;
+    if (in_buf.size() < cur_hdr->msg_len)
+        return 0;
+    cur_left = cur_hdr->msg_len - sizeof(msgr_handshake_hdr_t);
+    cur_buf = in_buf.data() + sizeof(msgr_handshake_hdr_t);
+    return orig_len - len;
+}
+
+bool msgr_handshake_t::read_with_len(const uint8_t* & dst, uint32_t & dst_len)
+{
+    if (cur_left < 4)
+    {
+        error = "Handshake packet too short";
+        state = MSGR_HS_ERROR;
+        return false;
+    }
+    dst_len = *(uint32_t*)cur_buf;
+    cur_buf += 4;
+    cur_left -= 4;
+    if (cur_left < dst_len)
+    {
+        error = "Handshake packet too short";
+        state = MSGR_HS_ERROR;
+        return false;
+    }
+    dst = cur_buf;
+    cur_buf += dst_len;
+    cur_left -= dst_len;
+    return true;
+}
+
+bool msgr_handshake_t::handle_client_init()
+{
+    // Derive shared secret and handshake keys
+    if (!derive_hs_keys(cur_buf, cur_left))
+        return false;
+    // Add type and key to full_handshake
+    copy_to(full_handshake, &cur_hdr->type, sizeof(cur_hdr->type));
+    copy_to_with_len(full_handshake, cur_buf, cur_left);
+    in_buf.clear();
+    return true;
+}
+
+// Decrypt and check peer certificate
+bool msgr_handshake_t::handle_peer_cert(const uint8_t *key, uint32_t key_len)
+{
+    if (!decrypt(cur_buf, cur_left, cur_buf))
+        return false;
+    const uint8_t *peer_cert_pem = NULL;
+    uint32_t peer_cert_len = 0;
+    if (!read_with_len(peer_cert_pem, peer_cert_len))
+        return false;
+    // Parse and verify certificate
+    if (!verify_peer(peer_cert_pem, peer_cert_len))
+        return false;
+    // Verify signature
+    copy_to(full_handshake, &cur_hdr->type, sizeof(cur_hdr->type));
+    if (key)
+        copy_to_with_len(full_handshake, key, key_len);
+    copy_to_with_len(full_handshake, peer_cert_pem, peer_cert_len);
+    const uint8_t *signature = NULL;
+    uint32_t signature_len = 0;
+    if (!read_with_len(signature, signature_len))
+        return false;
+    if (!verify(signature, signature_len))
+        return false;
+    return true;
+}
+
+bool msgr_handshake_t::handle_server_reply()
+{
+    // Derive shared secret and handshake keys
+    const uint8_t *key = NULL;
+    uint32_t key_len = 0;
+    if (!read_with_len(key, key_len))
+        return false;
+    if (!derive_hs_keys(key, key_len))
+        return false;
+    // Decrypt and check peer certificate
+    if (!handle_peer_cert(key, key_len))
+        return false;
+    in_buf.clear();
+    return true;
+}
+
+bool msgr_handshake_t::handle_client_reply()
+{
+    // Decrypt and check peer certificate
+    if (!handle_peer_cert(NULL, 0))
+        return false;
+    in_buf.clear();
+    return true;
+}
+
+void msgr_handshake_t::complete()
+{
+    state = MSGR_HS_DONE;
+    hs_key.clear();
+    peer_hs_key.clear();
+    full_handshake.clear();
+}
+
+ssize_t msgr_handshake_t::handle(uint8_t* in_buf, size_t in_size)
+{
+    if (state == MSGR_HS_SERVER_INIT)
+    {
+        ssize_t r = start_msg(in_buf, in_size, MSGR_HS_CLIENT_INIT);
+        if (r < 0)
+            return r;
+        if (r == 0)
+            return in_size;
+        if (!handle_client_init())
+            return -1;
+        // Send encrypted & signed response
+        if (!make_server_reply())
+            return -1;
+        state = MSGR_HS_SERVER_REPLY;
+        return r;
+    }
+    else if (state == MSGR_HS_CLIENT_INIT)
+    {
+        ssize_t r = start_msg(in_buf, in_size, MSGR_HS_SERVER_REPLY);
+        if (r < 0)
+            return r;
+        if (r == 0)
+            return in_size;
+        if (!handle_server_reply())
+            return -1;
+        // Verification passed, send certificate to the server
+        if (!make_client_reply())
+            return -1;
+        // Finished!
+        complete();
+        return r;
+    }
+    else if (state == MSGR_HS_SERVER_REPLY)
+    {
+        ssize_t r = start_msg(in_buf, in_size, MSGR_HS_CLIENT_REPLY);
+        if (r < 0)
+            return r;
+        if (r == 0)
+            return in_size;
+        if (!handle_client_reply())
+            return -1;
+        // Verification passed
+        // Finished!
+        complete();
+        return r;
+    }
+    else if (state == MSGR_HS_DONE)
+    {
+        return 0;
+    }
+    else if (state != MSGR_HS_ERROR)
+    {
+        error = "Unexpected handshake state: "+std::to_string(state);
+    }
+    return -1;
+}
+
+bool msgr_handshake_t::done()
+{
+    return (state == MSGR_HS_DONE);
+}
+
+uint8_t *msgr_handshake_t::get_out()
+{
+    return out_buf;
+}
+
+size_t msgr_handshake_t::out_size()
+{
+    return out_buf_size;
+}
+
+void msgr_handshake_t::eat_out(size_t n)
+{
+    if (n >= out_buf_size)
+    {
+        free(out_buf);
+        out_buf = NULL;
+        out_buf_size = 0;
+    }
+    else
+    {
+        memmove(out_buf, out_buf + n, out_buf_size - n);
+        out_buf_size -= n;
+    }
+}
+
+void msgr_handshake_t::reset_out()
+{
+    out_buf = NULL;
+    out_buf_size = 0;
+}
+
+msgr_handshake_result_t msgr_handshake_t::get_result()
+{
+    if (state != MSGR_HS_DONE)
+        return msgr_handshake_result_t{};
+    X509_up_ref(peer_cert);
+    return msgr_handshake_result_t{
+        .peer_cert = peer_cert,
+        .peer_is_osd = peer_is_osd,
+        .shared_secret = shared_secret,
+    };
+}
+
+std::string msgr_handshake_t::get_error()
+{
+    return error;
+}

src/client/msgr_handshake.h

@@ -0,0 +1,57 @@
+// Copyright (c) Vitaliy Filippov, 2026+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#pragma once
+
+#include <stdint.h>
+
+#include <string>
+#include <vector>
+
+#include <openssl/types.h>
+
+#define AES_256_GCM_KEY_SIZE 32
+#define AES_256_GCM_IV_SIZE 12
+#define AES_256_GCM_MAX_IV_CTR ((uint64_t)1 << 32)
+
+// TLS 1.3-like handshake
+
+// 1. Client->server: EC public key
+// 2. Server->client: EC public key, encrypted certificate and digital signature of the handshake
+// 3. Client->server: encrypted certificate and digital signature of the handshake
+
+struct msgr_handshake_result_t
+{
+    X509 *peer_cert = NULL;
+    bool peer_is_osd = false;
+    std::vector<uint8_t> shared_secret;
+};
+
+class msgr_handshake_i
+{
+public:
+    // Workflow: create -> init -> handle_msg -> get_result/get_error -> destruct
+    virtual ~msgr_handshake_i() = default;
+    virtual bool init(bool server_mode) = 0;
+    virtual ssize_t handle(uint8_t* in_buf, size_t in_size) = 0;
+    virtual bool done() = 0;
+    virtual uint8_t *get_out() = 0;
+    virtual size_t out_size() = 0;
+    virtual void eat_out(size_t n) = 0;
+    virtual void reset_out() = 0;
+    virtual msgr_handshake_result_t get_result() = 0;
+    virtual std::string get_error() = 0;
+};
+
+class msgr_handshake_ctx_i
+{
+public:
+    static msgr_handshake_ctx_i* create_ctx();
+    virtual ~msgr_handshake_ctx_i() = default;
+    virtual msgr_handshake_i* create() = 0;
+    virtual bool init(const std::string & pem_cert, const std::string & pem_key,
+        const std::string & pem_osd_ca, const std::string & pem_client_ca) = 0;
+    virtual std::string get_error() = 0;
+    virtual bool derive_kdf(const uint8_t* insecret, size_t insecret_len,
+        const uint8_t* salt, size_t salt_len, const char *label, uint8_t *outsecret, size_t outsize) = 0;
+};

src/client/msgr_iothread.cpp

@@ -0,0 +1,129 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#include <stdexcept>
+#include <sys/poll.h>
+#include <unistd.h>
+
+#include "messenger.h"
+#include "msgr_iothread.h"
+
+msgr_iothread_t::msgr_iothread_t():
+    ring(RINGLOOP_DEFAULT_SIZE, true),
+    thread(&msgr_iothread_t::run, this)
+{
+    eventfd = ring.register_eventfd();
+    if (eventfd < 0)
+    {
+        throw std::runtime_error(std::string("failed to register eventfd: ") + strerror(-eventfd));
+    }
+}
+
+msgr_iothread_t::~msgr_iothread_t()
+{
+    stop();
+}
+
+void msgr_iothread_t::add_sqe(io_uring_sqe & sqe)
+{
+    mu.lock();
+    queue.push_back((iothread_sqe_t){ .sqe = sqe, .data = std::move(*(ring_data_t*)sqe.user_data) });
+    if (queue.size() == 1)
+    {
+        cond.notify_all();
+    }
+    mu.unlock();
+}
+
+void msgr_iothread_t::stop()
+{
+    mu.lock();
+    if (stopped)
+    {
+        mu.unlock();
+        return;
+    }
+    stopped = true;
+    if (outer_loop_data)
+    {
+        outer_loop_data->callback = [](ring_data_t*){};
+    }
+    cond.notify_all();
+    close(eventfd);
+    mu.unlock();
+    thread.join();
+}
+
+void msgr_iothread_t::add_to_ringloop(ring_loop_i *outer_loop)
+{
+    assert(!this->outer_loop || this->outer_loop == outer_loop);
+    io_uring_sqe *sqe = outer_loop->get_sqe();
+    assert(sqe != NULL);
+    this->outer_loop = outer_loop;
+    this->outer_loop_data = ((ring_data_t*)sqe->user_data);
+    io_uring_prep_poll_add(sqe, eventfd, POLLIN);
+    outer_loop_data->callback = [this](ring_data_t *data)
+    {
+        if (data->res < 0)
+        {
+            throw std::runtime_error(std::string("eventfd poll failed: ") + strerror(-data->res));
+        }
+        outer_loop_data = NULL;
+        if (stopped)
+        {
+            return;
+        }
+        add_to_ringloop(this->outer_loop);
+        ring.loop();
+    };
+}
+
+void msgr_iothread_t::run()
+{
+    while (true)
+    {
+        {
+            std::unique_lock<std::mutex> lk(mu);
+            while (!stopped && !queue.size())
+                cond.wait(lk);
+            if (stopped)
+                return;
+            int i = 0;
+            for (; i < queue.size(); i++)
+            {
+                io_uring_sqe *sqe = ring.get_sqe();
+                if (!sqe)
+                    break;
+                ring_data_t *data = ((ring_data_t*)sqe->user_data);
+                *data = std::move(queue[i].data);
+                *sqe = queue[i].sqe;
+                sqe->user_data = (uint64_t)data;
+            }
+            queue.erase(queue.begin(), queue.begin()+i);
+        }
+        // We only want to offload sendmsg/recvmsg. Callbacks will be called in main thread
+        ring.submit();
+    }
+}
+
+void osd_messenger_t::init_iothreads()
+{
+    for (int i = 0; i < iothread_count; i++)
+    {
+        auto iot = new msgr_iothread_t();
+        iothreads.push_back(iot);
+        iot->add_to_ringloop(ringloop);
+    }
+}
+
+void osd_messenger_t::destroy_iothreads()
+{
+    if (iothreads.size())
+    {
+        for (auto iot: iothreads)
+        {
+            delete iot;
+        }
+        iothreads.clear();
+    }
+}

src/client/msgr_iothread.h

@@ -0,0 +1,38 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#include <mutex>
+#include <condition_variable>
+#include <thread>
+
+#include "ringloop.h"
+
+struct iothread_sqe_t
+{
+    io_uring_sqe sqe;
+    ring_data_t data;
+};
+
+class msgr_iothread_t
+{
+protected:
+    ring_loop_t ring;
+    ring_loop_i *outer_loop = NULL;
+    ring_data_t *outer_loop_data = NULL;
+    int eventfd = -1;
+    bool stopped = false;
+    std::mutex mu;
+    std::condition_variable cond;
+    std::vector<iothread_sqe_t> queue;
+    std::thread thread;
+
+    void run();
+public:
+
+    msgr_iothread_t();
+    ~msgr_iothread_t();
+
+    void add_sqe(io_uring_sqe & sqe);
+    void stop();
+    void add_to_ringloop(ring_loop_i *outer_loop);
+};

src/client/msgr_op.cpp

@@ -8,6 +8,7 @@
 osd_op_t::~osd_op_t()
 {
     assert(!bs_op);
+    assert(!op_data);
     if (bitmap_buf)
     {
         free(bitmap_buf);
@@ -22,10 +23,6 @@ osd_op_t::~osd_op_t()
         // So we don't reuse it, but free it every time
         free(buf);
     }
-    if (op_data)
-    {
-        free(op_data);
-    }
 }
 
 bool osd_op_t::is_recovery_related()

src/client/msgr_op.h

@@ -3,6 +3,8 @@
 
 #pragma once
 
+#include <memory>
+
 #include <sys/uio.h>
 #include <stdint.h>
 #include <stdio.h>
@@ -16,6 +18,8 @@
 
 #define OSD_OP_INLINE_BUF_COUNT 16
 
+#define AES_256_XTS_KEY_SIZE 64
+
 // Kind of a vector with small-list-optimisation
 struct osd_op_buf_list_t
 {
@@ -152,11 +156,25 @@ struct blockstore_op_t;
 
 struct osd_primary_op_data_t;
 
+struct osd_op_enc_t
+{
+    // Keys may contain more information in the future, like encryption algorithm and key ID
+    // In this case, key_chain will become inode_key_t* with inode_key_t also being a structure
+    // Currently all keys are required to be 512 bit (64 byte) long, for AES-256-XTS
+    // Raw pointers are convenient for messenger code; external users may use shared_ptr aliasing
+    // to implement complex freeing of osd_op_enc_t along with their external inode cache info
+    uint8_t** key_chain = NULL;
+    size_t chain_size = 0;
+    uint32_t read_chain_bitmap_pos = 0;
+    uint32_t bitmap_granularity = 0;
+};
+
 struct __attribute__((visibility("default"))) osd_op_t
 {
     timespec tv_begin = { 0 }, tv_end = { 0 };
     uint64_t op_type = OSD_OP_IN;
     uint64_t client_id = 0;
+    osd_num_t osd_num = 0;
     osd_any_op_t req;
     osd_any_reply_t reply;
     blockstore_op_t *bs_op = NULL;
@@ -164,9 +182,11 @@ struct __attribute__((visibility("default"))) osd_op_t
     // bitmap, bitmap_len, bmp_data are only meaningful for reads
     void *bitmap = NULL;
     unsigned bitmap_len = 0;
-    unsigned bmp_data = 0;
-    void *bitmap_buf = NULL;
+    size_t bmp_data = 0;
+    uint8_t *bitmap_buf = NULL;
     void *rmw_buf = NULL;
+    std::shared_ptr<osd_op_enc_t> enc;
+    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);
@@ -554,28 +568,6 @@ static void try_send_rdma_wr(osd_client_t *cl, ibv_sge *sge, int op_sge)
     cl->rdma_conn->cur_send++;
 }
 
-static int try_send_rdma_copy(osd_client_t *cl, uint8_t *dst, int dst_len)
-{
-    auto rc = cl->rdma_conn;
-    int total_dst_len = dst_len;
-    while (dst_len > 0 && rc->send_pos < cl->send_list.size())
-    {
-        iovec & iov = cl->send_list[rc->send_pos];
-        uint32_t len = (uint32_t)(iov.iov_len-rc->send_buf_pos < dst_len
-            ? iov.iov_len-rc->send_buf_pos : dst_len);
-        memcpy(dst, (uint8_t*)iov.iov_base+rc->send_buf_pos, len);
-        dst += len;
-        dst_len -= len;
-        rc->send_buf_pos += len;
-        if (rc->send_buf_pos >= iov.iov_len)
-        {
-            rc->send_pos++;
-            rc->send_buf_pos = 0;
-        }
-    }
-    return total_dst_len-dst_len;
-}
-
 void osd_messenger_t::try_send_rdma(osd_client_t *cl)
 {
     auto rc = cl->rdma_conn;
@@ -598,17 +590,24 @@ void osd_messenger_t::try_send_rdma(osd_client_t *cl)
     while (!rc->send_out_full && copied > 0 && rc->cur_send < rc->max_send)
     {
         dst = (uint8_t*)rc->send_out.buf + rc->send_out_pos;
-        dst_len = (rc->send_out_pos < rc->send_out_size ? rc->send_out_size-rc->send_out_pos : rc->send_done_pos-rc->send_out_pos);
+        dst_len = (rc->send_out_pos >= rc->send_done_pos
+            ? rc->send_out_size-rc->send_out_pos
+            : rc->send_done_pos-rc->send_out_pos);
         if (dst_len > rc->max_msg)
             dst_len = rc->max_msg;
-        copied = try_send_rdma_copy(cl, dst, dst_len);
+        copied = copy_ops_to(cl, dst, dst_len);
+        if (cl->io_error)
+        {
+            stop_client(cl->client_id);
+            return;
+        }
         if (copied > 0)
         {
             rc->send_out_pos += copied;
             if (rc->send_out_pos == rc->send_out_size)
                 rc->send_out_pos = 0;
             assert(rc->send_out_pos < rc->send_out_size);
-            if (rc->send_out_pos >= rc->send_done_pos)
+            if (rc->send_out_pos == rc->send_done_pos)
                 rc->send_out_full = true;
             ibv_sge sge = {
                 .addr = (uintptr_t)dst,
@@ -617,6 +616,7 @@ 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 +713,6 @@ 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 +724,43 @@ void osd_messenger_t::handle_rdma_events(msgr_rdma_context_t *rdma_context)
             else
             {
                 rc->cur_send--;
-                uint64_t sent_size = rc->send_sizes.at(0);
-                rc->send_sizes.erase(rc->send_sizes.begin(), rc->send_sizes.begin()+1);
+                // byte_len is not filled for send operations
+                uint64_t sent_size = rc->send_sizes.front();
+                rc->send_sizes.pop_front();
                 rc->send_done_pos += sent_size;
                 rc->send_out_full = false;
                 if (rc->send_done_pos == rc->send_out_size)
                     rc->send_done_pos = 0;
                 assert(rc->send_done_pos < rc->send_out_size);
-                int send_pos = 0, send_buf_pos = 0;
-                while (sent_size > 0)
+                while (osd_op_t *op = cl->send_free_ops.front())
                 {
-                    if (sent_size >= cl->send_list.at(send_pos).iov_len)
-                    {
-                        sent_size -= cl->send_list[send_pos].iov_len;
-                        send_pos++;
-                    }
+                    if (!((size_t)op & 7))
+                        delete op;
                     else
-                    {
-                        send_buf_pos = sent_size;
-                        sent_size = 0;
-                    }
+                        free((void*)((size_t)op & ~(size_t)7));
+                    cl->send_free_ops.pop_front();
                 }
-                assert(rc->send_pos >= send_pos);
-                if (rc->send_pos == send_pos)
+                cl->send_free_ops.pop_front();
+                if ((cl->proto_csum_status & MSGR_CSUM_NEG) && !cl->write_op && !cl->write_ops.size())
                 {
-                    rc->send_buf_pos -= send_buf_pos;
-                }
-                rc->send_pos -= send_pos;
-                for (int i = 0; i < send_pos; i++)
-                {
-                    if (cl->outbox[i].flags & MSGR_SENDP_FREE)
-                    {
-                        // Reply fully sent
-                        delete cl->outbox[i].op;
-                    }
-                }
-                if (send_pos > 0)
-                {
-                    cl->send_list.erase(cl->send_list.begin(), cl->send_list.begin()+send_pos);
-                    cl->outbox.erase(cl->outbox.begin(), cl->outbox.begin()+send_pos);
-                }
-                if (send_buf_pos > 0)
-                {
-                    cl->send_list[0].iov_base = (uint8_t*)cl->send_list[0].iov_base + send_buf_pos;
-                    cl->send_list[0].iov_len -= send_buf_pos;
+                    // Checksums negotiated, enable
+                    cl->proto_csum_status = cl->proto_csum_status & (~MSGR_CSUM_NEG);
                 }
                 try_send_rdma(cl);
             }
         }
     } while (event_count > 0);
-    handle_immediate_ops();
+}
+
+void osd_messenger_t::destroy_rdma_conn(msgr_rdma_connection_t *rdma_conn)
+{
+    if (rdma_conn->cmid)
+    {
+        auto rdma_it = rdmacm_connections.find(rdma_conn->cmid);
+        if (rdma_it != rdmacm_connections.end() && rdma_it->second->rdma_conn == rdma_conn)
+        {
+            rdmacm_connections.erase(rdma_it);
+        }
+    }
+    delete rdma_conn;
 }

src/client/msgr_rdma.h

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

src/client/msgr_rdmacm.cpp

@@ -19,6 +19,7 @@ struct rdmacm_connecting_t
     int tcp_port = 0;
     int timeout_ms = 0;
     int timeout_id = -1;
+    bool is_incoming = false;
     msgr_rdma_context_t *rdma_context = NULL;
 };
 
@@ -292,6 +293,7 @@ void osd_messenger_t::rdmacm_accept(rdma_cm_event *ev)
     conn->client_id = next_client_id++;
     conn->parsed_addr = *(sockaddr_storage*)rdma_get_peer_addr(ev->id);
     conn->rdma_context = rdma_context;
+    conn->is_incoming = true;
     rdmacm_set_conn_timeout(conn);
     rdmacm_connecting[ev->id] = conn;
     fprintf(stderr, "[OSD %ju] new client %ju: connection from %s via RDMA-CM\n", this->osd_num, conn->client_id,
@@ -492,11 +494,13 @@ void osd_messenger_t::rdmacm_established(rdma_cm_event *ev)
     cl->peer_addr = conn->parsed_addr;
     cl->peer_port = conn->rdmacm_port;
     cl->client_id = conn->client_id;
+    cl->is_incoming = conn->is_incoming;
     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;
+    init_tls_client(cl);
     clients[conn->client_id] = cl;
     if (conn->timeout_id >= 0)
     {