WIP Switch back to TCP when RDMA send fails (does not work)

Close TCP connections for RDMA clients
Also check for "Received garbage" during test
2026-05-01 22:26:41 +00:00 · 2026-05-01 22:13:29 +00:00 · 2026-05-01 16:01:28 +03:00 · 2026-05-01 15:57:15 +03:00 · 2026-05-01 15:57:15 +03:00 · 2026-04-29 02:42:23 +03:00
91 changed files with 3187 additions and 1450 deletions
@@ -1278,6 +1278,24 @@ 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
@@ -2,7 +2,7 @@ cmake_minimum_required(VERSION 2.8...3.30)

 project(vitastor)

-set(VITASTOR_VERSION "3.0.9")
+set(VITASTOR_VERSION "3.0.10")

 include(CTest)

@@ -1,4 +1,4 @@
-VITASTOR_VERSION ?= v3.0.9
+VITASTOR_VERSION ?= v3.0.10

 all: build push

@@ -49,7 +49,7 @@ spec:
            capabilities:
              add: ["SYS_ADMIN"]
            allowPrivilegeEscalation: true
-          image: vitalif/vitastor-csi:v3.0.9
+          image: vitalif/vitastor-csi:v3.0.10
          args:
            - "--node=$(NODE_ID)"
            - "--endpoint=$(CSI_ENDPOINT)"
@@ -121,7 +121,7 @@ spec:
            privileged: true
            capabilities:
              add: ["SYS_ADMIN"]
-          image: vitalif/vitastor-csi:v3.0.9
+          image: vitalif/vitastor-csi:v3.0.10
          args:
            - "--node=$(NODE_ID)"
            - "--endpoint=$(CSI_ENDPOINT)"
@@ -5,7 +5,7 @@ package vitastor

 const (
    vitastorCSIDriverName    = "csi.vitastor.io"
-    vitastorCSIDriverVersion = "3.0.9"
+    vitastorCSIDriverVersion = "3.0.10"
 )

 // Config struct fills the parameters of request or user input
@@ -1,4 +1,4 @@
-vitastor (3.0.9-1) unstable; urgency=medium
+vitastor (3.0.10-1) unstable; urgency=medium

  * Bugfixes

@@ -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 libc-ares-dev \
+    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
@@ -1,4 +1,4 @@
-VITASTOR_VERSION ?= v3.0.9
+VITASTOR_VERSION ?= v3.0.10

 all: build push

@@ -4,7 +4,7 @@
 #

 # Desired Vitastor version
-VITASTOR_VERSION=v3.0.9
+VITASTOR_VERSION=v3.0.10

 # Additional arguments for all containers
 # For example, you may want to specify a custom logging driver here
@@ -25,7 +25,7 @@ Most of them can be set in /etc/vitastor/vitastor.conf and in etcd, but don't su
 - [vault_timeout_ms](#vault_timeout_ms)
 - [vault_error_timeout_sec](#vault_error_timeout_sec)
 - [vault_refresh_leeway_sec](#vault_refresh_leeway_sec)
- [max_aes_xts_pool_size](#max_aes_xts_pool_size)
+- [max_cipher_pool_size](#max_cipher_pool_size)

 ## etcd_client_cert

@@ -141,10 +141,10 @@ Time (in seconds) to wait before retrying after receiving an error from Vault.
 Extra time (in seconds) before real Vault token lease_timeout to refresh it, just
 in case of system clock drift.

-## max_aes_xts_pool_size
+## max_cipher_pool_size

 - Type: integer
 - Default: 256

-Maximum number of OpenSSL encryption contexts cached in OSD memory. Probably
-doesn't require modification.
+Maximum number of OpenSSL cipher contexts cached in OSD memory, counted separately
+for each cipher and for encryption/decryption. Probably doesn't require modification.
@@ -27,7 +27,7 @@ OSD, мониторами и клиентами.
 - [vault_timeout_ms](#vault_timeout_ms)
 - [vault_error_timeout_sec](#vault_error_timeout_sec)
 - [vault_refresh_leeway_sec](#vault_refresh_leeway_sec)
- [max_aes_xts_pool_size](#max_aes_xts_pool_size)
+- [max_cipher_pool_size](#max_cipher_pool_size)

 ## etcd_client_cert

@@ -145,10 +145,10 @@ OSD, клиенты и мониторы должны иметь разные п
 Зазор времени (в секундах), чтобы обновлять токены Vault чуть раньше их реального
 lease_timeout, на случай "ухода" системных часов.

-## max_aes_xts_pool_size
+## max_cipher_pool_size

 - Тип: целое число
 - Значение по умолчанию: 256

-Максимальное количество кэшируемых в памяти OSD контекстов шифрования OpenSSL.
-Вряд ли требует изменения.
+Максимальное количество кэшируемых в памяти OSD контекстов шифра OpenSSL, учитываемое
+отдельно для каждого шифра и для шифрования и расшифровки. Вряд ли требует изменения.
@@ -120,12 +120,12 @@
  info_ru: |
    Зазор времени (в секундах), чтобы обновлять токены Vault чуть раньше их реального
    lease_timeout, на случай "ухода" системных часов.
- name: max_aes_xts_pool_size
+- name: max_cipher_pool_size
  type: int
  default: 256
  info: |
-    Maximum number of OpenSSL encryption contexts cached in OSD memory. Probably
-    doesn't require modification.
+    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 шифрования OpenSSL.
-    Вряд ли требует изменения.
+    Максимальное количество кэшируемых в памяти OSD контекстов шифра OpenSSL, учитываемое
+    отдельно для каждого шифра и для шифрования и расшифровки. Вряд ли требует изменения.
@@ -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:v3.0.10`
 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:v3.0.10 install.sh`
 3. Reload udev rules: \
   `udevadm control --reload-rules`
 4. Enable the vitastor-host service: \
@@ -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:v3.0.10`
 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:v3.0.10 install.sh`
 3. Перезагрузите правила udev: \
   `udevadm control --reload-rules`
 4. Включите сервис vitastor-host: \
@@ -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                                   |
@@ -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                             |
@@ -1,6 +1,6 @@
 {
  "name": "vitastor-mon",
-  "version": "3.0.9",
+  "version": "3.0.10",
  "description": "Vitastor SDS monitor service",
  "main": "mon-main.js",
  "scripts": {
@@ -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] = {};
@@ -1,6 +1,6 @@
 {
  "name": "vitastor",
-  "version": "3.0.9",
+  "version": "3.0.10",
  "description": "Low-level native bindings to Vitastor client library",
  "main": "index.js",
  "keywords": [
@@ -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 = '3.0.10'

 LOG = logging.getLogger(__name__)

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        3.0.9
+Version:        3.0.10
 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-3.0.10.el10.tar.gz

 BuildRequires:  gperftools-devel
 BuildRequires:  gcc-c++
@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        3.0.9
+Version:        3.0.10
 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-3.0.10.el7.tar.gz

 BuildRequires:  gperftools-devel
 BuildRequires:  devtoolset-9-gcc-c++
@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        3.0.9
+Version:        3.0.10
 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-3.0.10.el8.tar.gz

 BuildRequires:  gperftools-devel
 BuildRequires:  gcc-toolset-9-gcc-c++
@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        3.0.9
+Version:        3.0.10
 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-3.0.10.el9.tar.gz

 BuildRequires:  gperftools-devel
 BuildRequires:  gcc-c++
@@ -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="3.0.10")
 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})
@@ -69,15 +69,17 @@ 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)
-if (OPENSSL_FOUND)
-	add_definitions(-DWITH_OPENSSL)
-endif (OPENSSL_FOUND)
+add_definitions(-DWITH_OPENSSL)

 pkg_check_modules(CARES REQUIRED libcares)
 include_directories(${CARES_INCLUDE_DIRS})
@@ -344,19 +344,12 @@ corrupted_block:
                    block_num, block_offset, wr->size, sizeof(heap_entry_t));
                goto corrupted_block;
            }
-            if (wr->is_garbage())
-            {
-                // Garbage collection is only performed when writing new entries into the block
-                // because it needs a fake LSN and modified blocks require consecutive modified LSNs
-                // That's why garbage entries may persist on disk
-                if (log_level > 5)
-                {
-                    fprintf(stderr, "Notice: skipping garbage entry %jx:%jx v%ju l%ju in metadata block %u at %u\n",
-                        wr->inode, wr->stripe, wr->version, wr->lsn, block_num, block_offset);
-                }
-                block_offset += wr->size;
-                continue;
-            }
+            // Garbage collection is only performed when writing new entries into the block
+            // because it needs a fake LSN and modified blocks require consecutive modified LSNs
+            // At the same time, further modifications _after_ putting new entries into the block,
+            // but _before_ writing it, may mark some entries in it as garbage. That's why garbage
+            // entries may still be present on disk.
+            wr->entry_type &= ~BS_HEAP_GARBAGE;
            if ((wr->entry_type & BS_HEAP_TYPE) < BS_HEAP_BIG_WRITE ||
                (wr->entry_type & BS_HEAP_TYPE) > BS_HEAP_ROLLBACK ||
                (wr->entry_type & ~(BS_HEAP_TYPE|BS_HEAP_STABLE)) ||
@@ -311,7 +311,7 @@ resume_8:
        uint32_t block_num = recheck_mod[i];
        uint64_t block_offset = bs->dsk.meta_offset + (uint64_t)(block_num+1) * bs->dsk.meta_block_size;
        data = ((ring_data_t*)sqe->user_data);
-        uint8_t *buf = (uint8_t*)malloc_or_die(bs->dsk.meta_block_size);
+        uint8_t *buf = (uint8_t*)memalign_or_die(MEM_ALIGNMENT, bs->dsk.meta_block_size);
        bs->heap->get_meta_block(block_num, buf);
        data->iov = { buf, bs->dsk.meta_block_size };
        data->callback = [this, buf, block_offset](ring_data_t *data)
@@ -12,11 +12,11 @@ if (RDMACM_LIBRARIES)
 	set(MSGR_RDMACM "msgr_rdmacm.cpp")
 endif (RDMACM_LIBRARIES)
 add_library(vitastor_common STATIC
-	../util/epoll_manager.cpp etcd_state_client.cpp messenger.cpp ../util/addr_util.cpp ../util/xxh_x86dispatch.c
-	msgr_encrypt.cpp msgr_stop.cpp msgr_op.cpp msgr_send.cpp msgr_receive.cpp ../util/ringloop.cpp ../../json11/json11.cpp
+	../util/epoll_manager.cpp etcd_state_client.cpp messenger.cpp ../util/addr_util.cpp ../util/xxh_x86dispatch.c ../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 ${OPENSSL_LIBRARIES} ${CARES_LIBRARIES})
+target_link_libraries(vitastor_common pthread ${OPENSSL_LIBRARIES} ${CARES_LIBRARIES} ${ISAL_CRYPTO_LIBRARIES})
 target_compile_options(vitastor_common PUBLIC -fPIC)

 # libvitastor_client.so
@@ -35,6 +35,7 @@ target_link_libraries(vitastor_client
 	${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)
@@ -101,9 +102,9 @@ add_executable(test_cluster_client
 	EXCLUDE_FROM_ALL
 	../test/test_cluster_client.cpp
 	pg_states.cpp osd_ops.cpp cluster_client.cpp cluster_client_list.cpp cluster_client_wb.cpp cluster_client_icache.cpp msgr_op.cpp ../test/mock/messenger.cpp msgr_stop.cpp msgr_encrypt.cpp
-	etcd_state_client.cpp ../util/timerfd_manager.cpp ../util/addr_util.cpp ../util/str_util.cpp ../util/json_util.cpp ../util/xxh_x86dispatch.c ../../json11/json11.cpp
+	etcd_state_client.cpp ../util/timerfd_manager.cpp ../util/addr_util.cpp ../util/str_util.cpp ../util/json_util.cpp ../util/xxh_x86dispatch.c ../util/openssl_util.cpp ../../json11/json11.cpp
 )
-target_link_libraries(test_cluster_client ${OPENSSL_LIBRARIES})
+target_link_libraries(test_cluster_client ${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)
@@ -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); };
@@ -481,7 +481,7 @@ void cluster_client_t::on_load_config_hook(json11::Json::object & etcd_global_co
    }
    // vault
    vault_parse_config();
-    msgr.parse_config(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
@@ -1294,8 +1294,8 @@ void cluster_client_t::slice_rw(cluster_op_t *op)
        unsigned bitmap_mem = object_bitmap_size +
            op->parts.size() * pg_data_size *
            (pool_cfg.data_block_size / pool_cfg.bitmap_granularity / 8
-            // read chain info - 1 byte per block
-            + (op->enc ? op->len/pool_cfg.bitmap_granularity : 0));
+            // 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);
@@ -1452,7 +1452,10 @@ int cluster_client_t::try_send(cluster_op_t *op, int i, std::function<void(osd_o
            op->inflight_count++;
            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
-                + (op->opcode == OSD_OP_READ && op->enc ? pool_cfg.data_block_size/pool_cfg.bitmap_granularity : 0));
+                // 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)
            {
@@ -1597,6 +1600,9 @@ 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;
@@ -1605,15 +1611,10 @@ 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)
@@ -84,7 +84,7 @@ class writeback_cache_t;

 struct inode_cache_t
 {
-    std::vector<inode_t> chain;
+    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;
@@ -12,6 +12,7 @@
 #include "http_client.h"
 #endif
 #include "str_util.h"
+#include "json_util.h"

 etcd_state_client_t::~etcd_state_client_t()
 {
@@ -80,6 +81,46 @@ std::vector<std::string> etcd_state_client_t::get_addresses()
    return addrs;
 }

+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)
@@ -252,17 +293,21 @@ void etcd_state_client_t::parse_config(const json11::Json & config)
    }
    if (this->osd_num)
    {
-        this->etcd_client_cert = config["osd_etcd_client_cert"].string_value();
-        this->etcd_client_key = config["osd_etcd_client_key"].string_value();
+        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();
+    }
+    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();
    }
    this->etcd_ca = config["etcd_ca"].string_value();
    this->etcd_prefix = config["etcd_prefix"].string_value();
-    this->use_auth = config["use_auth"].bool_value();
    if (this->etcd_prefix == "")
    {
        this->etcd_prefix = "/vitastor";
@@ -1339,6 +1384,10 @@ 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
@@ -1354,13 +1403,39 @@ void etcd_state_client_t::parse_state(const etcd_kv_t & kv)
        if (on_change_node_placement_hook)
            on_change_node_placement_hook();
    }
-    else if (use_auth && key.substr(0, etcd_prefix.length()+13) == etcd_prefix+"/config/user/")
+    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())
-            user_info.erase(key.substr(etcd_prefix.length()+13));
+        {
+            if (inf)
+            {
+                inf->type = user_type_t::CLIENT;
+                inf->groups.clear();
+                inf->perm_cache.clear();
+            }
+            user_info.erase(name);
+        }
        else
-            user_info[key.substr(etcd_prefix.length()+13)] = value;
+        {
+            if (!inf)
+            {
+                inf = std::make_shared<user_info_t>();
+                inf->name = name;
+            }
+            inf->type = value["type"] == "admin" ? user_type_t::ADMIN :
+                (value["type"] == "mon" ? user_type_t::MON :
+                (value["type"] == "osd" ? user_type_t::OSD : 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();
+        }
    }
 }

@@ -1384,7 +1459,12 @@ 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;
@@ -9,6 +9,7 @@
 #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
@@ -110,6 +111,32 @@ struct http_url_t
    std::string path;
 };

+enum class user_type_t
+{
+    CLIENT = 0,
+    ADMIN = 1,
+    MON = 2,
+    OSD = 3,
+};
+
+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;

@@ -145,7 +172,6 @@ public:
    int etcd_slow_timeout = 5000;
    int etcd_min_reload_interval = 1000;
    bool infinite_start = true;
-    bool use_auth = false;
    uint64_t global_block_size = DEFAULT_BLOCK_SIZE;
    uint32_t global_bitmap_granularity = DEFAULT_BITMAP_GRANULARITY;
    uint32_t global_immediate_commit = IMMEDIATE_NONE;
@@ -171,7 +197,8 @@ 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;
-    std::map<std::string, json11::Json> user_info;
+    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;
@@ -193,6 +220,8 @@ public:
    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();
+    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);
@@ -19,6 +19,7 @@
 #include <openssl/err.h>
 #include <openssl/pem.h>
 #include <openssl/ssl.h>
+#include "openssl_util.h"
 #endif

 // libc-ares
@@ -163,104 +164,6 @@ void http_ares_cb(void *data, ares_socket_t socket_fd, int readable, int writabl
    });
 }

-#ifdef WITH_OPENSSL
-bool openssl_ctx_add_ca(SSL_CTX *ssl_ctx, const std::string & file_or_pem)
-{
-    BIO *bio = NULL;
-    if (file_or_pem.substr(0, 5) != "-----")
-    {
-        std::string pem = read_file(file_or_pem);
-        bio = BIO_new_mem_buf(pem.data(), pem.size());
-    }
-    else
-        bio = BIO_new_mem_buf(file_or_pem.data(), file_or_pem.size());
-    if (!bio)
-        return false;
-    X509 *x509 = PEM_read_bio_X509(bio, NULL, 0, NULL);
-    bool ok = !!x509;
-    if (x509)
-    {
-        X509_STORE *store = SSL_CTX_get_cert_store(ssl_ctx);
-        X509_STORE_add_cert(store, x509);
-        X509_free(x509);
-    }
-    BIO_free(bio);
-    return ok;
-}
-
-bool openssl_ctx_use_ca(SSL_CTX *ssl_ctx, const std::string & file_or_pem)
-{
-    if (file_or_pem.substr(0, 5) == "-----")
-    {
-        return openssl_ctx_add_ca(ssl_ctx, file_or_pem);
-    }
-    return file_or_pem.empty()
-        ? !!SSL_CTX_set_default_verify_paths(ssl_ctx)
-        : !!SSL_CTX_load_verify_locations(ssl_ctx, file_or_pem.c_str(), NULL);
-}
-
-std::string openssl_get_cn(X509 *x509)
-{
-    X509_NAME* subj = X509_get_subject_name(x509);
-    int pos = X509_NAME_get_index_by_NID(subj, NID_commonName, -1);
-    if (pos != -1)
-    {
-        X509_NAME_ENTRY* cn = X509_NAME_get_entry(subj, pos);
-        ASN1_STRING* str = X509_NAME_ENTRY_get_data(cn);
-        return std::string((const char*)ASN1_STRING_get0_data(str), ASN1_STRING_length(str));
-    }
-    return "";
-}
-
-bool openssl_ctx_use_cert(SSL_CTX *ssl_ctx, const std::string & file_or_pem, std::string & common_name)
-{
-    BIO *bio = NULL;
-    std::string contents;
-    if (file_or_pem.substr(0, 5) == "-----")
-        bio = BIO_new_mem_buf(file_or_pem.data(), file_or_pem.size());
-    else
-    {
-        contents = read_file(file_or_pem);
-        if (!contents.size())
-            return false;
-        bio = BIO_new_mem_buf(contents.data(), contents.size());
-    }
-    if (!bio)
-        return false;
-    X509 *x509 = PEM_read_bio_X509(bio, NULL, 0, NULL);
-    bool ok = !!x509;
-    if (x509)
-    {
-        ok = SSL_CTX_use_certificate(ssl_ctx, x509);
-        if (ok)
-            common_name = openssl_get_cn(x509);
-        X509_free(x509);
-    }
-    BIO_free(bio);
-    return ok;
-}
-
-bool openssl_ctx_use_key(SSL_CTX *ssl_ctx, const std::string & file_or_pem)
-{
-    if (file_or_pem.substr(0, 5) == "-----")
-    {
-        BIO *bio = BIO_new_mem_buf(file_or_pem.data(), file_or_pem.size());
-        if (!bio)
-            return false;
-        EVP_PKEY *pkey = PEM_read_bio_PrivateKey(bio, NULL, NULL, NULL);
-        bool ok = !!pkey;
-        if (pkey)
-        {
-            ok = SSL_CTX_use_PrivateKey(ssl_ctx, pkey);
-            EVP_PKEY_free(pkey);
-        }
-        BIO_free(bio);
-        return ok;
-    }
-    return !!SSL_CTX_use_PrivateKey_file(ssl_ctx, file_or_pem.c_str(), SSL_FILETYPE_PEM);
-}
-#endif
-
 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)
 {
@@ -8,10 +8,6 @@
 #include <functional>
 #include "json11/json11.hpp"

-#ifdef WITH_OPENSSL
-#include <openssl/types.h>
-#endif
-
 #define WS_CONTINUATION 0
 #define WS_TEXT 1
 #define WS_BINARY 2
@@ -73,11 +69,3 @@ void http_close(http_co_t *co);
 void http_destroy(http_co_t *co);

 #pragma GCC visibility pop
-
-#ifdef WITH_OPENSSL
-bool openssl_ctx_add_ca(SSL_CTX *ssl_ctx, const std::string & file_or_pem);
-bool openssl_ctx_use_ca(SSL_CTX *ssl_ctx, const std::string & file_or_pem);
-std::string openssl_get_cn(X509 *x509);
-bool openssl_ctx_use_cert(SSL_CTX *ssl_ctx, const std::string & file_or_pem, std::string & common_name);
-bool openssl_ctx_use_key(SSL_CTX *ssl_ctx, const std::string & file_or_pem);
-#endif
@@ -10,17 +10,11 @@
 #include <stdexcept>

 #include "addr_util.h"
+#include "str_util.h"
 #include "messenger.h"
 #ifdef WITH_RDMA
 #include "msgr_rdma.h"
 #endif
-#include "http_client.h"
-#ifdef WITH_OPENSSL
-#include <openssl/bio.h>
-#include <openssl/err.h>
-#include <openssl/pem.h>
-#include <openssl/ssl.h>
-#endif

 #include <sys/poll.h>

@@ -124,43 +118,7 @@ void msgr_iothread_t::run()

 void osd_messenger_t::init()
 {
-    if (!tls_cert.empty() || !tls_key.empty() || !osd_tls_ca.empty() || !client_tls_ca.empty())
-    {
-        // Initialize TLS context
-        // FIXME: require OpenSSL
-#ifndef WITH_OPENSSL
-        fprintf(stderr, "Vitastor is built without OpenSSL support\n");
-        exit(1);
-#else
-        if (tls_cert.empty() || tls_key.empty() || osd_tls_ca.empty() || osd_num && client_tls_ca.empty())
-        {
-            if (osd_num)
-                fprintf(stderr, "Vitastor OSD TLS requires osd_tls_cert, osd_tls_key, osd_tls_ca, client_tls_ca\n");
-            else
-                fprintf(stderr, "Vitastor client TLS requires tls_cert, tls_key and osd_tls_ca\n");
-            exit(1);
-        }
-        else
-        {
-            ssl_ctx = SSL_CTX_new(TLS_method());
-            SSL_CTX_set_verify(ssl_ctx, SSL_VERIFY_PEER, NULL);
-            bool ok = SSL_CTX_set_min_proto_version(ssl_ctx, TLS1_3_VERSION);
-            ok = ok && openssl_ctx_add_ca(ssl_ctx, osd_tls_ca);
-            if (osd_num)
-            {
-                // OSD uses 2 separate root certificates to distinguish between clients and peer OSDs
-                ok = ok && openssl_ctx_add_ca(ssl_ctx, client_tls_ca);
-            }
-            ok = ok && openssl_ctx_use_cert(ssl_ctx, tls_cert, tls_cn);
-            ok = ok && openssl_ctx_use_key(ssl_ctx, tls_key);
-            if (!ok)
-            {
-                SSL_CTX_free(ssl_ctx);
-                ssl_ctx = NULL;
-            }
-        }
-#endif
-    }
+    init_tls();
 #ifdef WITH_RDMACM
    if (use_rdmacm)
    {
@@ -233,9 +191,12 @@ void osd_messenger_t::init()
        {
            auto cl = cl_it->second;
            cl_it++;
-            if (!cl->osd_num && !cl->in_osd_num || cl->peer_state != PEER_CONNECTED && cl->peer_state != PEER_RDMA)
+            if (!cl->osd_num && !cl->in_osd_num ||
+                cl->peer_state != PEER_CONNECTED &&
+                cl->peer_state != PEER_RDMA_CONNECTING_OUT &&
+                cl->peer_state != PEER_RDMA)
            {
-                // Do not run keepalive on regular clients
+                // Do not run keepalive on regular and unconnected clients
                continue;
            }
            if (cl->ping_time_remaining > 0)
@@ -243,10 +204,21 @@ void osd_messenger_t::init()
                cl->ping_time_remaining--;
                if (!cl->ping_time_remaining)
                {
-                    // Ping timed out, stop the client
-                    fprintf(stderr, "Ping timed out for OSD %ju (client %ju), disconnecting peer\n",
-                        cl->in_osd_num ? cl->in_osd_num : cl->osd_num, cl->client_id);
-                    clients_to_stop.push_back(cl->client_id);
+                    // Ping timed out
+                    if (cl->peer_state == PEER_RDMA_CONNECTING_OUT && cl->peer_fd >= 0)
+                    {
+                        fprintf(stderr, "Ping timed out for OSD %ju over RDMA (client %ju), switching back to TCP\n",
+                            cl->in_osd_num ? cl->in_osd_num : cl->osd_num, cl->client_id);
+                        delete cl->rdma_conn;
+                        cl->rdma_conn = NULL;
+                        cl->peer_state = PEER_CONNECTED;
+                    }
+                    else
+                    {
+                        fprintf(stderr, "Ping timed out for OSD %ju (client %ju), disconnecting peer\n",
+                            cl->in_osd_num ? cl->in_osd_num : cl->osd_num, cl->client_id);
+                        clients_to_stop.push_back(cl->client_id);
+                    }
                }
            }
            else if (cl->idle_time_remaining > 0)
@@ -264,7 +236,7 @@ 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())
@@ -273,14 +245,25 @@ void osd_messenger_t::init()
                            delete op;
                            return;
                        }
-                        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;
+                        auto cl = cl_it->second;
+                        bool failed = (op->reply.hdr.retval != 0);
                        cl->ping_time_remaining = 0;
                        delete op;
-                        if (fail_client_id)
+                        if (failed)
                        {
-                            fprintf(stderr, "Ping failed for OSD %ju (client %ju), disconnecting peer\n", fail_osd_num, fail_client_id);
-                            stop_client(fail_client_id);
+                            auto fail_osd_num = cl->in_osd_num ? cl->in_osd_num : cl->osd_num;
+                            if (cl->peer_state == PEER_RDMA_CONNECTING_OUT && cl->peer_fd >= 0)
+                            {
+                                fprintf(stderr, "Ping failed for OSD %ju over RDMA (client %ju), switching back to TCP\n", fail_osd_num, cl->client_id);
+                                delete cl->rdma_conn;
+                                cl->rdma_conn = NULL;
+                                cl->peer_state = PEER_CONNECTED;
+                            }
+                            else
+                            {
+                                fprintf(stderr, "Ping failed for OSD %ju (client %ju), disconnecting peer\n", fail_osd_num, cl->client_id);
+                                stop_client(cl->client_id);
+                            }
                        }
                    };
                    cl->ping_time_remaining = osd_ping_timeout;
@@ -339,25 +322,52 @@ osd_messenger_t::~osd_messenger_t()
        rdmacm_evch = NULL;
    }
 #endif
-    for (auto encrypt_ctx: encrypt_ctx_pool)
+    for (auto encrypt_ctx: encrypt_xts_pool)
    {
        destroy_aes_xts_encrypt(encrypt_ctx);
    }
-    for (auto decrypt_ctx: decrypt_ctx_pool)
+    for (auto decrypt_ctx: decrypt_xts_pool)
    {
        destroy_aes_xts_decrypt(decrypt_ctx);
    }
-#ifdef WITH_OPENSSL
-    if (ssl_ctx)
-    {
-        SSL_CTX_free(ssl_ctx);
-        ssl_ctx = NULL;
-    }
-#endif
+    destroy_tls();
 }

-void osd_messenger_t::parse_config(const json11::Json & config)
+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;
+    if (config["proto_checksums"].is_null())
+        this->use_proto_checksums = MSGR_CSUM_PAYLOAD;
+    else if (config["proto_checksums"].is_bool())
+        this->use_proto_checksums = config["proto_checksums"].bool_value() ? MSGR_CSUM_FULL : 0;
+    else if (config["proto_checksums"].string_value() != "")
+        this->use_proto_checksums = config["proto_checksums"].string_value() == "full" ? MSGR_CSUM_FULL : MSGR_CSUM_PAYLOAD;
+    else
+        this->use_proto_checksums = 0;
+    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())
    {
@@ -388,55 +398,25 @@ void osd_messenger_t::parse_config(const json11::Json & config)
    if (!this->rdma_max_msg || this->rdma_max_msg > 128*1024*1024)
        this->rdma_max_msg = 129*1024;
 #endif
-    this->max_aes_xts_pool_size = config["max_aes_xts_pool_size"].uint64_value();
-    if (!this->max_aes_xts_pool_size)
-        this->max_aes_xts_pool_size = 256;
-    if (config["proto_checksums"].is_null())
-        this->use_proto_checksums = MSGR_CSUM_PAYLOAD;
-    else if (config["proto_checksums"].is_bool())
-        this->use_proto_checksums = config["proto_checksums"].bool_value() ? MSGR_CSUM_FULL : 0;
-    else if (config["proto_checksums"].string_value() != "")
-        this->use_proto_checksums = config["proto_checksums"].string_value() == "full" ? MSGR_CSUM_FULL : MSGR_CSUM_PAYLOAD;
-    else
-        this->use_proto_checksums = 0;
    if (!osd_num)
    {
-        tls_cert = config["tls_cert"].string_value();
-        tls_key = config["tls_key"].string_value();
-        osd_tls_ca = config["osd_tls_ca"].string_value();
+        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_tls_cert"].string_value();
-        tls_key = config["osd_tls_key"].string_value();
-        osd_tls_ca = config["osd_tls_ca"].string_value();
-        client_tls_ca = config["client_tls_ca"].string_value();
+        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();
    }
    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())
@@ -642,10 +622,7 @@ void osd_messenger_t::handle_connect_epoll(int peer_fd)
        handle_peer_epoll(peer_fd, epoll_events);
    });
    // Check OSD number
-    if (!tls_cert.empty())
-    {
-        ssl_init(cl, false);
-    }
+    init_tls_client(cl);
    check_peer_config(cl);
 }

@@ -656,6 +633,17 @@ void osd_messenger_t::handle_peer_epoll(int peer_fd, int epoll_events)
    if (epoll_events & EPOLLRDHUP)
    {
        // Stop client
+        if (cl->rdma_close_tcp && (cl->peer_state == PEER_RDMA ||
+            cl->peer_state == PEER_RDMA_CONNECTING_IN ||
+            cl->peer_state == PEER_RDMA_CONNECTING_OUT))
+        {
+            // It's allowed to stop the TCP socket during RDMA transition
+            tfd->set_fd_handler(cl->peer_fd, false, NULL);
+            clients_by_fd.erase(cl->peer_fd);
+            close(cl->peer_fd);
+            cl->peer_fd = -1;
+            return;
+        }
        if (log_level > 0)
        {
            fprintf(stderr, "[OSD %ju] client %ju disconnected\n", this->osd_num, cl->client_id);
@@ -735,7 +723,10 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
        // Inform that we're OSD <osd_num>
        payload["osd_num"] = osd_num;
    }
-    auto features = json11::Json::object{ { "check_sequencing", true } };
+    auto features = json11::Json::object{
+        { "check_sequencing", true },
+        { "rdma_close_tcp", true },
+    };
    if (use_proto_checksums)
    {
        features["proto_checksums"] = use_proto_checksums;
@@ -780,6 +771,11 @@ 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);
@@ -824,6 +820,10 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
                cl->proto_csum_status = MSGR_CSUM_PAYLOAD;
        }
 #ifdef WITH_RDMA
+        if (config["features"]["rdma_close_tcp"].bool_value())
+        {
+            cl->rdma_close_tcp = true;
+        }
        if (!use_rdmacm && cl->rdma_conn && config["rdma_address"].is_string())
        {
            msgr_rdma_address_t addr;
@@ -848,9 +848,14 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
                {
                    fprintf(stderr, "Connected to OSD %ju using RDMA\n", cl->osd_num);
                }
-                cl->peer_state = PEER_RDMA;
+                cl->peer_state = (cl->rdma_close_tcp ? PEER_RDMA_CONNECTING_OUT : PEER_RDMA);
                // Add the initial receive request
                init_recv_rdma(cl);
+                // Check the connection by using a ping
+                cl->ping_time_remaining = osd_ping_timeout;
+                cl->idle_time_remaining = 0;
+                delete op;
+                return;
            }
        }
 #endif
@@ -887,10 +892,7 @@ void osd_messenger_t::accept_connections(int listen_fd)
        cl->peer_fd = peer_fd;
        cl->peer_state = PEER_CONNECTED;
        cl->in_buf = (uint8_t*)malloc_or_die(receive_buffer_size);
-        if (!tls_cert.empty())
-        {
-            ssl_init(cl, true);
-        }
+        init_tls_client(cl);
        // Add FD to epoll
        tfd->set_fd_handler(peer_fd, false, [this](int peer_fd, int epoll_events)
        {
@@ -905,26 +907,6 @@ void osd_messenger_t::accept_connections(int listen_fd)
    }
 }

-void osd_messenger_t::ssl_init(osd_client_t *cl, bool server_mode)
-{
-#ifdef WITH_OPENSSL
-    cl->write_to_ssl = BIO_new(BIO_s_mem());
-    cl->read_from_ssl = BIO_new(BIO_s_mem());
-    cl->ssl_cli = SSL_new(ssl_ctx);
-    if (server_mode)
-    {
-        SSL_set_accept_state(cl->ssl_cli);
-    }
-    else
-    {
-        SSL_set_connect_state(cl->ssl_cli);
-    }
-    SSL_set_bio(cl->ssl_cli, cl->write_to_ssl, cl->read_from_ssl);
-    bool ok = ssl_do_handshake(cl);
-    assert(ok);
-#endif
-}
-
 #ifdef WITH_RDMA
 msgr_rdma_context_t* osd_messenger_t::choose_rdma_context(osd_client_t *cl)
 {
@@ -953,6 +935,11 @@ 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;
@@ -12,8 +12,10 @@
 #include <deque>
 #include <vector>

-#ifdef WITH_OPENSSL
 #include <openssl/types.h>
+
+#ifdef WITH_ISAL_CRYPTO
+#include <isa-l_crypto/aes_gcm.h>
 #endif

 #include "../util/xxh_x86dispatch.h"
@@ -23,6 +25,7 @@
 #include "msgr_op.h"
 #include "timerfd_manager.h"
 #include "addr_util.h"
+#include "msgr_handshake.h"
 #include <ringloop.h>

 #define CL_READ_HDR 1
@@ -32,9 +35,10 @@

 #define PEER_CONNECTING 1
 #define PEER_CONNECTED 2
-#define PEER_RDMA_CONNECTING 3
-#define PEER_RDMA 4
-#define PEER_STOPPED 5
+#define PEER_RDMA_CONNECTING_IN 3
+#define PEER_RDMA_CONNECTING_OUT 4
+#define PEER_RDMA 5
+#define PEER_STOPPED 6

 #define MSGR_CSUM_PAYLOAD 1
 #define MSGR_CSUM_FULL 2
@@ -44,6 +48,11 @@

 #define DEFAULT_MIN_ZEROCOPY_SEND_SIZE 32*1024

+#define AES_256_GCM_KEY_SIZE 32
+#define AES_256_GCM_IV_SIZE 12
+
+#define MAX_SIMPLE_PAYLOAD_SIZE 1048576
+
 struct msgr_sendp_t
 {
    osd_op_t *op;
@@ -60,11 +69,7 @@ struct op_aes_xts_decrypt_t;
 void destroy_aes_xts_encrypt(op_aes_xts_encrypt_t *encrypt_ctx);
 void destroy_aes_xts_decrypt(op_aes_xts_decrypt_t *decrypt_ctx);

-struct __attribute__((__packed__)) msgr_tls_record_hdr_t
-{
-    uint8_t encrypted;
-    uint32_t size;
-};
+struct user_info_t;

 struct osd_client_t
 {
@@ -88,18 +93,30 @@ struct osd_client_t
    msgr_rdma_connection_t *rdma_conn = NULL;
 #endif

-#ifdef WITH_OPENSSL
-    SSL *ssl_cli = NULL;
-    BIO *write_to_ssl = NULL;
-    // FIXME: use custom bio to avoid 1 more memory copy?
-    BIO *read_from_ssl = NULL;
    uint8_t *ssl_out_buf = NULL;
    size_t ssl_out_buf_size = 0, ssl_out_buf_cap = 0;
-    bool ssl_handshake_done = false;
-    bool ssl_want_write = false;
+
+    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;
@@ -107,11 +124,12 @@ struct osd_client_t
    iovec read_iov = { 0 };
    msghdr read_msg = { 0 };
    std::vector<iovec> recv_list;
-    size_t recv_list_size = 0;
+    std::vector<int> recv_flags;
    uint64_t read_op_id = 1;
    bool check_sequencing = false;
+    bool rdma_close_tcp = false;
    bool enable_pg_locks = false;
-    op_aes_xts_decrypt_t *decrypt_ctx = NULL;
+    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;
@@ -137,7 +155,7 @@ struct osd_client_t
    size_t send_list_size = 0;
    std::deque<osd_op_t*> send_free_ops;
    std::vector<osd_op_t*> zc_free_list;
-    op_aes_xts_encrypt_t *encrypt_ctx = NULL;
+    op_aes_xts_encrypt_t *xts_enc_ctx = NULL;
    XXH3_state_t* write_csum_state = NULL;

    ~osd_client_t();
@@ -216,9 +234,19 @@ 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;
@@ -230,7 +258,7 @@ protected:
    bool use_sync_send_recv = false;
    int min_zerocopy_send_size = DEFAULT_MIN_ZEROCOPY_SEND_SIZE;
    int iothread_count = 0;
-    int max_aes_xts_pool_size = 256;
+    int max_cipher_pool_size = 256;

    std::string tls_cert;
    std::string tls_key;
@@ -253,18 +281,14 @@ protected:
    robin_hood::unordered_flat_map<rdma_cm_id*, rdmacm_connecting_t*> rdmacm_connecting;
 #endif

-#ifdef WITH_OPENSSL
-    SSL_CTX *ssl_ctx = NULL;
-    std::string tls_cn;
+    bool gcm_enabled = false;
+    msgr_handshake_ctx_i *hs_ctx = NULL;

-    void ssl_init(osd_client_t *cl, bool server_mode);
-    bool ssl_do_handshake(osd_client_t *cl);
-    bool ssl_do_encrypt(osd_client_t *cl);
-    size_t ssl_do_encrypt_to(osd_client_t *cl, uint8_t *buf, size_t size);
-    bool ssl_op_write_buf(osd_client_t *cl, uint8_t *src, size_t src_len, bool skip_csum, size_t & from, size_t & done);
-    size_t ssl_op_copy_to(osd_client_t *cl, uint8_t *dst, size_t dst_len);
-    void ssl_op_get_write_buffers(osd_client_t *cl, std::vector<iovec> & lst);
-#endif
+    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;
@@ -272,16 +296,27 @@ protected:
    // We don't use ringloop->set_immediate here because we may have no ringloop in client :)
    std::deque<osd_op_t*> set_immediate_ops;

-    std::vector<op_aes_xts_encrypt_t*> encrypt_ctx_pool;
-    std::vector<op_aes_xts_decrypt_t*> decrypt_ctx_pool;
+    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;
    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;
@@ -297,7 +332,7 @@ public:
    osd_op_stats_t stats, recovery_stats;

    void init();
-    void parse_config(const json11::Json & config);
+    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);
@@ -306,6 +341,7 @@ 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);
@@ -326,6 +362,7 @@ 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);
@@ -342,32 +379,32 @@ protected:

    bool try_send(osd_client_t *cl);
    void handle_send(int result, bool prev, bool more, osd_client_t *cl);
-    size_t op_copy_to(osd_client_t *cl, uint8_t *dst, size_t dst_len);
-    void op_get_write_buffers(osd_client_t *cl, std::vector<iovec> & lst);
+    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);

    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_copy_from(osd_client_t *cl, uint8_t *src, size_t src_len, size_t & done);
-    void op_get_read_buffers(osd_client_t *cl, std::vector<iovec> & lst);
-    void op_alloc_temp_buffers(osd_op_t *op, int i);
+    bool op_read_from(osd_client_t *cl, msgr_op_reader_t & rdr);
    bool handle_finished_op(osd_client_t *cl);
    void handle_immediate_ops();

-    bool op_encrypted_copy_data_to(osd_client_t* cl, uint8_t *buf, size_t len, size_t from, size_t & done);
-    bool op_decrypted_copy_data_from(osd_client_t* cl, uint8_t *buf, size_t len, size_t from, size_t & done);
+    void op_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);
-    int try_send_rdma_copy(osd_client_t *cl, uint8_t *dst, int dst_len);
    bool init_recv_rdma(osd_client_t *cl);
    void handle_rdma_events(msgr_rdma_context_t *rdma_context);
    msgr_rdma_context_t* choose_rdma_context(osd_client_t *cl);
@@ -3,13 +3,23 @@

 #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()
 {
-#ifdef WITH_OPENSSL
+#ifndef WITH_ISAL_CRYPTO
    if (!(ctx = EVP_CIPHER_CTX_new()))
    {
        ERR_print_errors_fp(stderr);
@@ -21,16 +31,13 @@ op_aes_xts_encrypt_t::op_aes_xts_encrypt_t()
        ERR_print_errors_fp(stderr);
        abort();
    }
-#else
-    fprintf(stderr, "Error: Vitastor is built without encryption support\n");
-    abort();
 #endif
 }

 op_aes_xts_encrypt_t::~op_aes_xts_encrypt_t()
 {
    assert(!encrypted);
-#ifdef WITH_OPENSSL
+#ifndef WITH_ISAL_CRYPTO
    EVP_CIPHER_CTX_free(ctx);
 #endif
    if (tmp)
@@ -52,7 +59,7 @@ void op_aes_xts_encrypt_t::start(uint8_t *key, uint64_t start_offset, size_t blo
        tmp = NULL;
        tmp_size = 0;
    }
-#ifdef WITH_OPENSSL
+#ifndef WITH_ISAL_CRYPTO
    if (EVP_EncryptInit_ex(ctx, NULL, NULL, key, NULL) != 1)
    {
        ERR_print_errors_fp(stderr);
@@ -63,9 +70,12 @@ void op_aes_xts_encrypt_t::start(uint8_t *key, uint64_t start_offset, size_t blo

 void op_aes_xts_encrypt_t::encrypt_block(uint8_t *in, uint8_t *out)
 {
-#ifdef WITH_OPENSSL
    uint8_t iv[16] = { 0 };
    *((uint64_t*)iv) = start_offset + offset - offset%block_size;
+#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);
@@ -98,7 +108,10 @@ void op_aes_xts_encrypt_t::update(uint8_t *in, size_t max_in, uint8_t *out, size
        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)
    {
@@ -126,7 +139,7 @@ void op_aes_xts_encrypt_t::update(uint8_t *in, size_t max_in, uint8_t *out, size
        encrypted = true;
        memcpy(out, tmp, max_out);
        tmp_pos = max_out;
-        done_in += max_in;
+        done_in += max_in-1;
        offset += max_in;
        done_out += max_out;
    }
@@ -158,7 +171,7 @@ void destroy_aes_xts_encrypt(op_aes_xts_encrypt_t *encrypt_ctx)

 op_aes_xts_decrypt_t::op_aes_xts_decrypt_t()
 {
-#ifdef WITH_OPENSSL
+#ifndef WITH_ISAL_CRYPTO
    if (!(ctx = EVP_CIPHER_CTX_new()))
    {
        ERR_print_errors_fp(stderr);
@@ -170,29 +183,27 @@ op_aes_xts_decrypt_t::op_aes_xts_decrypt_t()
        ERR_print_errors_fp(stderr);
        abort();
    }
-#else
-    fprintf(stderr, "Error: Vitastor is built without encryption support\n");
-    abort();
 #endif
 }

 op_aes_xts_decrypt_t::~op_aes_xts_decrypt_t()
 {
    assert(!decrypted);
-#ifdef WITH_OPENSSL
+#ifndef WITH_ISAL_CRYPTO
    EVP_CIPHER_CTX_free(ctx);
 #endif
    if (tmp)
        free(tmp);
 }

-void op_aes_xts_decrypt_t::start(uint8_t **key_chain, size_t chain_size, uint8_t *key_indexes, uint64_t start_offset, size_t block_size)
+void op_aes_xts_decrypt_t::start(uint8_t **key_chain, size_t chain_size, void *key_indexes, uint64_t start_offset, size_t block_size)
 {
    assert(!decrypted);
    this->start_offset = start_offset;
-    this->key_chain = chain_size > 1 ? key_chain : 0;
-    this->chain_size = chain_size > 1 ? chain_size : 0;
-    this->key_indexes = chain_size > 1 ? key_indexes : NULL;
+    this->key_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;
@@ -203,7 +214,7 @@ void op_aes_xts_decrypt_t::start(uint8_t **key_chain, size_t chain_size, uint8_t
        tmp = NULL;
        tmp_size = 0;
    }
-#ifdef WITH_OPENSSL
+#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);
@@ -215,21 +226,35 @@ void op_aes_xts_decrypt_t::start(uint8_t **key_chain, size_t chain_size, uint8_t
 void op_aes_xts_decrypt_t::decrypt_block(uint8_t *in, uint8_t *out)
 {
    uint8_t *key = NULL;
-    if (chain_size)
+    if (chain_size > 1)
    {
-        assert(key_indexes[offset/block_size] < chain_size);
-        key = key_chain[key_indexes[offset/block_size]];
-        if (!key)
-        {
-            if (in != out)
-                memcpy(out, in, block_size);
-            return;
-        }
+        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;
    }
-#ifdef WITH_OPENSSL
    uint8_t iv[16] = { 0 };
    *((uint64_t*)iv) = start_offset + offset - offset%block_size;
-    if (EVP_DecryptInit_ex(ctx, NULL, NULL, key, iv) != 1)
+#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();
@@ -264,7 +289,10 @@ void op_aes_xts_decrypt_t::update(uint8_t *in, size_t max_in, uint8_t *out, size
        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)
    {
@@ -293,7 +321,7 @@ void op_aes_xts_decrypt_t::update(uint8_t *in, size_t max_in, uint8_t *out, size
        if (out)
            memcpy(out, tmp, max_out);
        tmp_pos = max_out;
-        done_in += max_in;
+        done_in += max_in-1;
        offset += max_in;
        done_out += max_out;
    }
@@ -325,101 +353,65 @@ void destroy_aes_xts_decrypt(op_aes_xts_decrypt_t *decrypt_ctx)
    delete decrypt_ctx;
 }

-bool osd_messenger_t::op_encrypted_copy_data_to(osd_client_t* cl, uint8_t *enc_buf, size_t enc_len, size_t from, size_t & done)
+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)
 {
-    auto op = cl->write_op;
-    auto & op_pos = cl->write_op_pos;
-    assert(op->req.hdr.opcode == OSD_OP_WRITE);
-    if (!from)
+    if (!cl->xts_enc_ctx)
    {
-        if (!cl->encrypt_ctx)
+        if (encrypt_xts_pool.size())
        {
-            if (encrypt_ctx_pool.size())
-            {
-                cl->encrypt_ctx = encrypt_ctx_pool.back();
-                encrypt_ctx_pool.pop_back();
-            }
-            else
-                cl->encrypt_ctx = new op_aes_xts_encrypt_t();
+            cl->xts_enc_ctx = encrypt_xts_pool.back();
+            encrypt_xts_pool.pop_back();
        }
-        assert(op->enc->key_chain[0]);
-        cl->encrypt_ctx->start(op->enc->key_chain[0], op->req.rw.offset, op->enc->bitmap_granularity);
-    }
-    for (int i = 0; i < op->iov.count; i++)
-    {
-        uint8_t *plain = (uint8_t*)op->iov.buf[i].iov_base;
-        size_t plain_len = op->iov.buf[i].iov_len;
-        while (from < plain_len || cl->encrypt_ctx->has_buffered())
-        {
-            if (done >= enc_len)
-                return false;
-            size_t done_in = 0;
-            size_t done_out = 0;
-            cl->encrypt_ctx->update(plain+from, plain_len-from, enc_buf+done, enc_len-done, done_in, done_out);
-            if (cl->write_csum_state && done_out > 0)
-                XXH3_64bits_update(cl->write_csum_state, enc_buf+done, done_out);
-            done += done_out;
-            op_pos += done_in;
-            from += done_in;
-        }
-        from -= plain_len;
-    }
-    if (cl->encrypt_ctx)
-    {
-        if (encrypt_ctx_pool.size() > max_aes_xts_pool_size)
-            delete cl->encrypt_ctx;
        else
-            encrypt_ctx_pool.push_back(cl->encrypt_ctx);
-        cl->encrypt_ctx = NULL;
+            cl->xts_enc_ctx = new op_aes_xts_encrypt_t();
+        assert(cl->write_op->enc->key_chain[0]);
+        cl->xts_enc_ctx->start(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;
    }
-    return true;
 }

-bool osd_messenger_t::op_decrypted_copy_data_from(osd_client_t* cl, uint8_t *enc_buf, size_t enc_len, size_t from, size_t & done)
+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);
-    auto op = cl->read_op;
-    assert(op->req.hdr.opcode == OSD_OP_READ);
-    for (int i = 0; i < op->iov.count; i++)
+    while (done_plain < plain_len && done_enc < enc_len)
    {
-        uint8_t *plain = (uint8_t*)op->iov.buf[i].iov_base;
-        size_t plain_len = op->iov.buf[i].iov_len;
-        while (from < plain_len)
-        {
-            if (done >= enc_len)
-                return false;
-            size_t done_in = 0;
-            size_t done_out = 0;
-            // plain == NULL means skip output
-            cl->decrypt_ctx->update(enc_buf+done, enc_len-done, plain ? plain+from : NULL, plain_len-from, done_in, done_out);
-            if (cl->read_csum_state && done_in > 0)
-                XXH3_64bits_update(cl->read_csum_state, enc_buf+done, done_in);
-            done += done_in;
-            cl->read_op_pos += done_out;
-            cl->read_op_inline_decrypt_in += done_in;
-            from += done_out;
-            if (!done_out)
-                return false;
-        }
-        from -= plain_len;
+        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;
    }
-    op_decrypt_free(cl);
-    return true;
 }

 void osd_messenger_t::op_decrypt_start(osd_client_t* cl)
 {
-    if (!cl->decrypt_ctx)
+    if (!cl->xts_dec_ctx)
    {
-        if (decrypt_ctx_pool.size())
+        if (decrypt_xts_pool.size())
        {
-            cl->decrypt_ctx = decrypt_ctx_pool.back();
-            decrypt_ctx_pool.pop_back();
+            cl->xts_dec_ctx = decrypt_xts_pool.back();
+            decrypt_xts_pool.pop_back();
        }
        else
-            cl->decrypt_ctx = new op_aes_xts_decrypt_t();
+            cl->xts_dec_ctx = new op_aes_xts_decrypt_t();
        auto & enc = cl->read_op->enc;
-        cl->decrypt_ctx->start(enc->key_chain, enc->chain_size,
+        assert(cl->read_op->req.hdr.opcode == OSD_OP_READ);
+        cl->xts_dec_ctx->start(enc->key_chain, enc->chain_size,
            (cl->read_op->req.rw.flags & OSD_OP_RETURN_CHAIN) ? (uint8_t*)cl->read_op->bitmap + enc->read_chain_bitmap_pos : 0,
            cl->read_op->req.rw.offset, enc->bitmap_granularity);
    }
@@ -451,7 +443,7 @@ void osd_messenger_t::op_decrypt_inline(osd_client_t* cl)
        size_t out_len = op->iov.buf[j].iov_len - from_out;
        size_t done_in = 0;
        size_t done_out = 0;
-        cl->decrypt_ctx->update(in, in_len, out, out_len, done_in, done_out);
+        cl->xts_dec_ctx->update(in, in_len, out, out_len, done_in, done_out);
        if (done_in >= in_len)
        {
            i++;
@@ -468,17 +460,142 @@ void osd_messenger_t::op_decrypt_inline(osd_client_t* cl)
            from_out += done_out;
    }
    assert(j >= op->iov.count);
-    op_decrypt_free(cl);
 }

 void osd_messenger_t::op_decrypt_free(osd_client_t* cl)
 {
-    if (cl->decrypt_ctx)
+    if (cl->xts_dec_ctx)
    {
-        if (decrypt_ctx_pool.size() > max_aes_xts_pool_size)
-            delete cl->decrypt_ctx;
+        if (decrypt_xts_pool.size() > max_cipher_pool_size)
+            delete cl->xts_dec_ctx;
        else
-            decrypt_ctx_pool.push_back(cl->decrypt_ctx);
-        cl->decrypt_ctx = NULL;
+            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->get_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;
    }
 }
@@ -3,17 +3,18 @@

 #include <stdint.h>

+#ifdef WITH_ISAL_CRYPTO
+#include <isa-l_crypto/aes_xts.h>
+#endif
+
 #include "../util/xxh_x86dispatch.h"
-// WITH_OPENSSL is left to possibly support other crypto libraries
-#ifdef WITH_OPENSSL
 #include <openssl/conf.h>
 #include <openssl/evp.h>
 #include <openssl/err.h>
-#endif

 class op_aes_xts_encrypt_t
 {
-#ifdef WITH_OPENSSL
+#ifndef WITH_ISAL_CRYPTO
    EVP_CIPHER_CTX *ctx = NULL;
 #endif
    uint64_t start_offset = 0;
@@ -31,7 +32,6 @@ public:
    op_aes_xts_encrypt_t();
    ~op_aes_xts_encrypt_t();

-    inline bool has_buffered() { return encrypted; };
    void start(uint8_t *key, uint64_t start_offset, size_t block_size);
    void update(uint8_t *in, size_t max_in, uint8_t *out, size_t max_out, size_t & done_in, size_t & done_out);
 };
@@ -40,13 +40,14 @@ void destroy_aes_xts_encrypt(op_aes_xts_encrypt_t *encrypt_ctx);

 class op_aes_xts_decrypt_t
 {
-#ifdef WITH_OPENSSL
+#ifndef WITH_ISAL_CRYPTO
    EVP_CIPHER_CTX *ctx = NULL;
 #endif
    uint64_t start_offset = 0;
    uint8_t **key_chain = NULL;
    size_t chain_size = 0;
-    uint8_t *key_indexes = NULL;
+    void *key_indexes = NULL;
+    int key_index_bytes = 0;
    size_t offset = 0;
    size_t block_size = 0;
    uint8_t *tmp = NULL;
@@ -60,8 +61,7 @@ public:
    op_aes_xts_decrypt_t();
    ~op_aes_xts_decrypt_t();

-    inline bool has_buffered() { return decrypted; };
-    void start(uint8_t **key_chain, size_t chain_size, uint8_t *key_indexes, uint64_t start_offset, size_t block_size);
+    void start(uint8_t **key_chain, size_t chain_size, void *key_indexes, uint64_t start_offset, size_t block_size);
    void update(uint8_t *in, size_t max_in, uint8_t *out, size_t max_out, size_t & done_in, size_t & done_out);
 };

@@ -0,0 +1,765 @@
+// 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 "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;
+    std::vector<uint8_t> out_buf;
+    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;
+    std::vector<uint8_t>& get_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 (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_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_len = out_buf.size();
+    out_buf.resize(out_buf.size() + key_len + sizeof(msgr_handshake_hdr_t));
+    uint8_t *buf = out_buf.data() + old_len;
+    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();
+    copy_to(out_buf, &hdr, sizeof(hdr));
+    copy_to_with_len(out_buf, key, key_len);
+    copy_to(out_buf, 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();
+    copy_to(out_buf, &hdr, sizeof(hdr));
+    copy_to(out_buf, 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);
+}
+
+std::vector<uint8_t>& msgr_handshake_t::get_out()
+{
+    return out_buf;
+}
+
+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;
+}
@@ -0,0 +1,54 @@
+// 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 std::vector<uint8_t> & get_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;
+};
@@ -8,6 +8,7 @@
 osd_op_t::~osd_op_t()
 {
    assert(!bs_op);
+    assert(!op_data);
    if (bitmap_buf)
    {
        free(bitmap_buf);
@@ -26,10 +27,6 @@ osd_op_t::~osd_op_t()
    {
        free(enc_buf);
    }
-    if (op_data)
-    {
-        free(op_data);
-    }
 }

 bool osd_op_t::is_recovery_related()
@@ -181,8 +181,8 @@ 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;
    uint8_t *enc_buf = NULL;
@@ -541,7 +541,7 @@ bool osd_messenger_t::connect_rdma(uint64_t client_id, std::string rdma_address,
            {
                // Remember connection, but switch to RDMA only after sending the configuration response
                cl->rdma_conn = rdma_conn;
-                cl->peer_state = PEER_RDMA_CONNECTING;
+                cl->peer_state = PEER_RDMA_CONNECTING_IN;
                return true;
            }
        }
@@ -568,29 +568,6 @@ static void try_send_rdma_wr(osd_client_t *cl, ibv_sge *sge, int op_sge)
    cl->rdma_conn->cur_send++;
 }

-int osd_messenger_t::try_send_rdma_copy(osd_client_t *cl, uint8_t *dst, int dst_len)
-{
-    int total_dst_len = dst_len;
-    while (dst_len > 0 && (cl->write_op || cl->write_ops.size()))
-    {
-        next_write_op(cl);
-        osd_op_t *op = cl->write_op;
-        size_t copied = op_copy_to(cl, dst, dst_len);
-        if (!copied)
-        {
-            break;
-        }
-        dst += copied;
-        dst_len -= copied;
-        if (!cl->write_op && op->op_type == OSD_OP_IN)
-        {
-            // this is a reply, free the op after sending it
-            cl->send_free_ops.push_back(op);
-        }
-    }
-    return total_dst_len-dst_len;
-}
-
 void osd_messenger_t::try_send_rdma(osd_client_t *cl)
 {
    auto rc = cl->rdma_conn;
@@ -616,7 +593,12 @@ void osd_messenger_t::try_send_rdma(osd_client_t *cl)
        dst_len = (rc->send_out_pos < rc->send_out_size ? rc->send_out_size-rc->send_out_pos : rc->send_done_pos-rc->send_out_pos);
        if (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;
@@ -661,6 +643,7 @@ static void try_recv_rdma_wr(osd_client_t *cl, void *buf)

 bool osd_messenger_t::init_recv_rdma(osd_client_t *cl)
 {
+    return true;
    auto rc = cl->rdma_conn;
    assert(!rc->recv_buf.buf);
    rc->recv_buf.buf = (uint8_t*)malloc_or_die(rc->max_msg * rc->max_recv);
@@ -716,9 +699,14 @@ void osd_messenger_t::handle_rdma_events(msgr_rdma_context_t *rdma_context)
                continue;
            }
            auto rc = cl->rdma_conn;
+            if (!rc)
+            {
+                // Connection destroyed (fallback to TCP)
+                continue;
+            }
            if (wc[i].status != IBV_WC_SUCCESS)
            {
-                fprintf(stderr, "RDMA work request failed for client %ju", client_id);
+                fprintf(stderr, "RDMA work request (%s) failed for client %ju", is_send ? "send" : "recv", client_id);
                if (cl->osd_num)
                {
                    fprintf(stderr, " (OSD %ju)", cl->osd_num);
@@ -729,6 +717,28 @@ void osd_messenger_t::handle_rdma_events(msgr_rdma_context_t *rdma_context)
            }
            if (!is_send)
            {
+                if (cl->peer_state == PEER_RDMA_CONNECTING_OUT)
+                {
+                    // First message received over RDMA - TCP socket can be closed now
+                    if (log_level > 0)
+                    {
+                        fprintf(stderr, "Successfully connected with client %ju using RDMA\n", cl->client_id);
+                    }
+                    cl->peer_state = PEER_RDMA;
+                    if (cl->peer_fd >= 0 && cl->rdma_close_tcp)
+                    {
+                        // TCP socket is not needed anymore
+                        tfd->set_fd_handler(cl->peer_fd, false, NULL);
+                        clients_by_fd.erase(cl->peer_fd);
+                        close(cl->peer_fd);
+                        cl->peer_fd = -1;
+                    }
+                    if (cl->osd_num)
+                    {
+                        osd_peers[cl->osd_num] = cl;
+                        on_connect_peer(cl->osd_num, 0, cl->client_id);
+                    }
+                }
                rc->cur_recv--;
                if (!handle_read_buffer(cl, rc->recv_buffers[rc->next_recv_buf], wc[i].byte_len))
                {
@@ -79,7 +79,7 @@ struct msgr_rdma_connection_t
    msgr_rdma_buf_t recv_buf;
    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();
@@ -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 = (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)
    {
@@ -9,12 +9,8 @@
 #ifdef WITH_RDMA
 #include "msgr_rdma.h"
 #endif
-#ifdef WITH_OPENSSL
-#include <openssl/bio.h>
-#include <openssl/err.h>
+#include <openssl/evp.h>
 #include <openssl/pem.h>
-#include <openssl/ssl.h>
-#endif

 void osd_client_t::cancel_ops()
 {
@@ -86,21 +82,21 @@ void osd_messenger_t::stop_client(uint64_t client_id, bool force_delete)
            fprintf(stderr, "[OSD %ju] Stopping client %ju (regular client)\n", osd_num, client_id);
        }
    }
-    if (cl->encrypt_ctx)
+    if (cl->xts_enc_ctx)
    {
-        if (encrypt_ctx_pool.size() > max_aes_xts_pool_size)
-            destroy_aes_xts_encrypt(cl->encrypt_ctx);
+        if (encrypt_xts_pool.size() > max_cipher_pool_size)
+            destroy_aes_xts_encrypt(cl->xts_enc_ctx);
        else
-            encrypt_ctx_pool.push_back(cl->encrypt_ctx);
-        cl->encrypt_ctx = NULL;
+            encrypt_xts_pool.push_back(cl->xts_enc_ctx);
+        cl->xts_enc_ctx = NULL;
    }
-    if (cl->decrypt_ctx)
+    if (cl->xts_dec_ctx)
    {
-        if (decrypt_ctx_pool.size() > max_aes_xts_pool_size)
-            destroy_aes_xts_decrypt(cl->decrypt_ctx);
+        if (decrypt_xts_pool.size() > max_cipher_pool_size)
+            destroy_aes_xts_decrypt(cl->xts_dec_ctx);
        else
-            decrypt_ctx_pool.push_back(cl->decrypt_ctx);
-        cl->decrypt_ctx = NULL;
+            decrypt_xts_pool.push_back(cl->xts_dec_ctx);
+        cl->xts_dec_ctx = NULL;
    }
    // First set state to STOPPED so another stop_client() call doesn't try to free it again
    cl->refs++;
@@ -223,7 +219,10 @@ osd_client_t::~osd_client_t()
    {
        if (op)
        {
-            delete op;
+            if (!((size_t)op & 7))
+                delete op;
+            else
+                free((void*)((size_t)op & ~(size_t)7));
        }
    }
 #ifndef __MOCK__
@@ -245,18 +244,37 @@ osd_client_t::~osd_client_t()
        XXH3_freeState(write_csum_state);
        write_csum_state = NULL;
    }
-#ifdef WITH_OPENSSL
-    if (ssl_cli)
+    if (enc_ctx)
    {
-        SSL_free(ssl_cli);
-        ssl_cli = NULL;
-        write_to_ssl = NULL;
-        read_from_ssl = NULL;
+#ifdef WITH_ISAL_CRYPTO
+        free(enc_ctx);
+#else
+        EVP_CIPHER_CTX_free(enc_ctx);
+#endif
+        enc_ctx = NULL;
+    }
+    if (dec_ctx)
+    {
+#ifdef WITH_ISAL_CRYPTO
+        free(dec_ctx);
+#else
+        EVP_CIPHER_CTX_free(dec_ctx);
+#endif
+        dec_ctx = NULL;
    }
    if (ssl_out_buf)
    {
        free(ssl_out_buf);
        ssl_out_buf = NULL;
    }
-#endif
+    if (hs)
+    {
+        delete hs;
+        hs = NULL;
+    }
+    if (hs_result.peer_cert)
+    {
+        X509_free(hs_result.peer_cert);
+        hs_result.peer_cert = NULL;
+    }
 }
@@ -231,12 +231,25 @@ struct __attribute__((__packed__)) osd_op_rw_t
    uint32_t len;
    // flags
    // OSD_OP_RETURN_CHAIN for chained reads: return parent number in chain for each block
+    // read_chain size comes after bitmap, takes 0 bytes / 1 byte / 2 byte / 4 byte per each block,
+    // depending on the number of parent inodes (0 parents = 0 bytes, up to 255 parents = 1 byte, etc)
    uint32_t flags;
    // inode metadata revision for chained reads
    uint64_t meta_revision;
    // object version for atomic "CAS" (compare-and-set) writes
    // writes and deletes fail with -EINTR if object version differs from (version-1)
    uint64_t version;
+
+    static inline size_t chain_info_bytes(size_t chain_size)
+    {
+        if (chain_size <= 1)
+            return 0;
+        if (chain_size <= 256)
+            return 1;
+        if (chain_size <= 65536)
+            return 2;
+        return 4;
+    }
 };

 struct __attribute__((__packed__)) osd_reply_rw_t
@@ -6,7 +6,7 @@ includedir=${prefix}/@CMAKE_INSTALL_INCLUDEDIR@

 Name: Vitastor
 Description: Vitastor client library
-Version: 3.0.9
+Version: 3.0.10
 Libs: -L${libdir} -lvitastor_client
 Cflags: -I${includedir}

@@ -259,9 +259,9 @@ static const char* help_text =
    "  Start HTTP server able to handle CLI commands over a REST API. Options:\n"
    "  --bind_address ADDR  Specify server IP address or addresses, separated by space. Default is 127.0.0.1.\n"
    "  --port 8080          Specify server port.\n"
-    "  --ssl_cert FILE      Path to server SSL certificate file (PEM format).\n"
-    "  --ssl_key FILE       Path to server SSL private key file.\n"
-    "  --ssl_ca FILE        Path to file with SSL CA certificates used to validate client connections.\n"
+    "  --server_cert FILE   Path to server TLS certificate file (PEM format).\n"
+    "  --server_key FILE    Path to server TLS private key file.\n"
+    "  --client_ca FILE     Path to file with TLS CA certificates used to validate client connections.\n"
    "\n"
    "Use vitastor-cli --help <command> for command details or vitastor-cli --help --all for all details.\n"
    "\n"
@@ -27,12 +27,7 @@ struct cli_result_t
    json11::Json data;
 };

-struct cli_user_t
-{
-    std::string name;
-    std::string type;
-    std::set<std::string> groups;
-};
+struct user_info_t;

 class cli_tool_t
 {
@@ -45,7 +40,7 @@ public:
    bool is_command_line = false;
    bool color = false;

-    std::unique_ptr<cli_user_t> user; // for http mode
+    std::shared_ptr<user_info_t> user; // for http mode

    ring_loop_t *ringloop = NULL;
    epoll_manager_t *epmgr = NULL;
@@ -9,7 +9,7 @@
 bool cli_tool_t::check_image_perm(const inode_config_t & cfg, bool write)
 {
    return !user ||
-        user->type == "admin" ||
+        user->type == user_type_t::ADMIN ||
        user->name == cfg.owner ||
        cfg.owner_group != "" && user->groups.find(cfg.owner_group) != user->groups.end() ||
        !write && cfg.reader_group != "" && user->groups.find(cfg.reader_group) != user->groups.end();
@@ -1,9 +1,7 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 (see README.md for details)

-#ifdef WITH_OPENSSL
 #include <openssl/rand.h>
-#endif

 #include <ctype.h>
 #include "cli.h"
@@ -123,7 +121,7 @@ struct image_creator_t

    bool check_pool_permission()
    {
-        if (!parent->user || parent->user->type == "admin")
+        if (!parent->user || parent->user->type == user_type_t::ADMIN)
        {
            return true;
        }
@@ -628,14 +626,12 @@ std::function<bool(cli_result_t &)> cli_tool_t::start_create(json11::Json cfg)
    if (!cfg["enc_key"].is_null())
    {
        image_creator->set_key = true;
-#ifdef WITH_OPENSSL
        if (image_creator->enc_key == "random")
        {
            uint8_t newkey[64];
            RAND_bytes(newkey, 64);
            image_creator->enc_key = tohexstr(newkey, 64);
        }
-#endif
        else
        {
            image_creator->enc_key = cfg["enc_key"].string_value();
@@ -147,7 +147,7 @@ resume_1:
            inode_t inode_num = INODE_WITH_POOL(pool_id, only_inode_num);
            uint64_t used_size = kv.value["raw_used"].uint64_value();
            auto stat_it = stats.find(inode_num);
-            if (parent->user && parent->user->type != "admin" && stat_it == stats.end())
+            if (parent->user && parent->user->type != user_type_t::ADMIN && stat_it == stats.end())
            {
                continue;
            }
@@ -286,6 +286,8 @@ struct rm_inode_t
                    .data = data,
                };
            }
+            // Wakeup callers (otherwise they wake up only on next loop())
+            parent->ringloop->wakeup();
        }
        in_continue = false;
    }
@@ -76,6 +76,7 @@ struct cli_serve_t
    int port = 0;
    int listen_backlog = 0;
    bool ssl = false;
+    bool use_auth = false;
    std::vector<int> listen_fds;
    http_context_t *http_ctx = NULL;
    std::set<cli_serve_conn_t*> connections;
@@ -111,19 +112,34 @@ struct cli_serve_t
        listen_backlog = options["listen_backlog"].uint64_value();
        if (!listen_backlog)
            listen_backlog = 128;
-        ssl = json_is_true(options["ssl"]);
-        if (ssl)
        {
-            std::string ssl_cert = options["ssl_cert"].string_value();
-            std::string ssl_key = options["ssl_key"].string_value();
-            std::string ssl_ca = options["ssl_ca"].string_value();
-            std::string error;
-            http_ctx = http_context_init(parent->epmgr->tfd, ssl_cert, ssl_key, ssl_ca, ssl_ca != "", error);
-            if (error != "")
+            std::string tls_cert = (parent->cli->config.find("server_cert") != parent->cli->config.end()
+                ? parent->cli->config["server_cert"].string_value() : "");
+            std::string tls_key = (parent->cli->config.find("server_key") != parent->cli->config.end()
+                ? parent->cli->config["server_key"].string_value() : "");
+            std::string tls_ca = (parent->cli->config.find("client_ca") != parent->cli->config.end()
+                ? parent->cli->config["client_ca"].string_value() : "");
+            if (tls_cert != "" || tls_key != "" || tls_ca != "")
            {
-                result = (cli_result_t){ .err = EINVAL, .text = error };
-                state = 100;
-                return;
+                ssl = true;
+                if (tls_cert == "" || tls_key == "")
+                {
+                    result = (cli_result_t){ .err = EINVAL, .text = "server_cert and server_key are required to serve HTTPS" };
+                    state = 100;
+                    return;
+                }
+                // use_auth is enabled by default when client_ca is set
+                use_auth = (parent->cli->config["use_auth"].is_null()
+                    ? (tls_ca != "")
+                    : json_is_true(parent->cli->config["use_auth"]));
+                std::string error;
+                http_ctx = http_context_init(parent->epmgr->tfd, tls_cert, tls_key, tls_ca, tls_ca != "", error);
+                if (error != "")
+                {
+                    result = (cli_result_t){ .err = EINVAL, .text = error };
+                    state = 100;
+                    return;
+                }
            }
        }
        for (auto & bind_address: bind_addresses)
@@ -344,17 +360,9 @@ struct cli_serve_t
        conn->request_path = std::move(req_line[1]);
        conn->request_body = std::move(msg->body);
        conn->response_type = "";
-        if (parent->cli->st_cli.use_auth)
+        if (use_auth)
        {
-            auto user = std::make_unique<cli_user_t>();
-            user->name = msg->headers["_tls_common_name"];
-            auto user_it = parent->cli->st_cli.user_info.find(user->name);
-            auto userinfo = user_it == parent->cli->st_cli.user_info.end() ? user_it->second : json11::Json();
-            user->type = user->name == "root" ? "admin" : userinfo["type"].string_value();
-            for (auto & gr: userinfo["groups"].array_items())
-            {
-                user->groups.insert(gr.string_value());
-            }
+            conn->p->user = parent->cli->st_cli.get_user(msg->headers["_tls_common_name"]);
        }
        auto ctype = msg->headers["content-type"];
        if (conn->request_method != "GET" && conn->request_method != "POST")
@@ -383,10 +391,10 @@ struct cli_serve_t
            {
                conn->response_type = "application/json";
                conn->result = { .text = openapi_description };
-                if (parent->cli->st_cli.use_auth)
+                if (use_auth)
                {
                    // Filter available paths by privileges
-                    if (conn->p->user->type == "client")
+                    if (conn->p->user->type == user_type_t::CLIENT)
                    {
                        std::string error;
                        auto openapi = json11::Json::parse(openapi_description, error).object_items();
@@ -403,7 +411,7 @@ struct cli_serve_t
                        conn->response_type = "application/json";
                        conn->result = { .text = json11::Json(openapi).dump() };
                    }
-                    else if (conn->p->user->type != "admin")
+                    else if (conn->p->user->type != user_type_t::ADMIN)
                    {
                        conn->response_type = "";
                        conn->result = { .err = EACCES, .text = "Access denied" };
@@ -418,7 +426,7 @@ struct cli_serve_t
            {
                conn->result = { .err = ENOSYS, .text = "method /"+uri[0]+" only allows POST requests" };
            }
-            else if (parent->cli->st_cli.use_auth && conn->p->user->type == "client" && !cmd_it->second.allow_client)
+            else if (use_auth && conn->p->user->type == user_type_t::CLIENT && !cmd_it->second.allow_client)
            {
                conn->result = { .err = EACCES, .text = "Access denied" };
            }
@@ -121,6 +121,7 @@ static const char* help_text =
    "  --logfile <FILE>  log to the specified file\n"
    "  --enforce 1       enforce permissions at the server side (default is disabled)\n"
    "  --foreground 1    stay in foreground, do not daemonize\n"
+    "  --trace           trace all NFS requests\n"
    "\n"
    "NFS proxy is stateless if you use immediate_commit=all in your cluster and if\n"
    "you do not use client_enable_writeback=true, so you can freely use multiple\n"
@@ -158,7 +159,7 @@ json11::Json::object nfs_proxy_t::parse_args(int narg, const char *args[])
        {
            const char *opt = args[i]+2;
            cfg[str_replace(opt, "-", "_")] = !strcmp(opt, "json") || !strcmp(opt, "block") ||
-                !strcmp(opt, "dry-run") || !strcmp(opt, "recalc-stats") ||
+                !strcmp(opt, "dry-run") || !strcmp(opt, "recalc-stats") || !strcmp(opt, "trace") ||
                !strcmp(opt, "include-empty") || !strcmp(opt, "no-rm") || i == narg-1 ? "1" : args[++i];
        }
        else
@@ -14,6 +14,7 @@
 #include "http_client.h"
 #include "str_util.h"
 #include "json_util.h"
+#include "openssl_util.h"

 osd_t::osd_t(const json11::Json & config, ring_loop_t *ringloop)
 {
@@ -67,6 +68,11 @@ osd_t::osd_t(const json11::Json & config, ring_loop_t *ringloop)
    msgr.repeer_pgs = [this](osd_num_t peer_osd) { repeer_pgs(peer_osd); };
    msgr.break_pg_locks = [this](osd_num_t peer_osd) { break_pg_locks(peer_osd); };
    msgr.check_config_hook = [this](osd_client_t *cl, json11::Json conf) { return check_peer_config(cl, conf); };
+    msgr.handshake_hook = [this](osd_client_t *cl)
+    {
+        if (!cl->hs_result.peer_is_osd)
+            cl->user_info = st_cli.get_user(openssl_get_cn(cl->hs_result.peer_cert));
+    };
    msgr.init();

    init_cluster();
@@ -170,9 +176,13 @@ void osd_t::parse_config(bool init)
        bs->parse_config(bs_cfg);
    }
    st_cli.parse_config(config);
-    msgr.parse_config(config);
+    msgr.parse_config(config, init);
    if (init)
    {
+        // use_auth is enabled by default when encryption is enabled
+        use_auth = (config["use_auth"].is_null()
+            ? msgr.is_encryption_enabled()
+            : json_is_true(config["use_auth"]));
        // Vital Blockstore parameters
        bs_block_size = config["block_size"].uint64_value();
        if (!bs_block_size)
@@ -181,6 +191,9 @@ void osd_t::parse_config(bool init)
        if (!bs_bitmap_granularity)
            bs_bitmap_granularity = DEFAULT_BITMAP_GRANULARITY;
        clean_entry_bitmap_size = bs_block_size / bs_bitmap_granularity / 8;
+        msgr.bs_block_size = bs_block_size;
+        msgr.clean_entry_bitmap_size = clean_entry_bitmap_size;
+        msgr.max_write_request_size = MAX_DATA_BLOCK_SIZE; // will be changed after pool config
        // immediate_commit
        if (config["immediate_commit"] == "all")
            immediate_commit = IMMEDIATE_ALL;
@@ -30,6 +30,8 @@
 #define OSD_RECOVERING 0x10
 #define OSD_SCRUBBING 0x20

+#define SELF_CLIENT 0
+
 #define MAX_AUTOSYNC_INTERVAL 3600
 #define DEFAULT_AUTOSYNC_INTERVAL 5
 #define DEFAULT_AUTOSYNC_WRITES 128
@@ -155,6 +157,7 @@ class osd_t

    etcd_state_client_t st_cli;
    osd_messenger_t msgr;
+    bool use_auth = false;
    int etcd_failed_attempts = 0;
    std::string etcd_lease_id;
    json11::Json self_state;
@@ -428,6 +428,21 @@ void osd_t::on_change_pool_config_hook()
    {
        apply_pg_locks_localize_only();
    }
+    msgr.max_write_request_size = 0;
+    for (auto & pc: st_cli.pool_config)
+    {
+        auto & pool_cfg = pc.second;
+        uint32_t pg_data_size = (pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pool_cfg.pg_size-pool_cfg.parity_chunks);
+        uint32_t pg_block_size = pool_cfg.data_block_size * pg_data_size;
+        if (msgr.max_write_request_size < pg_block_size)
+        {
+            msgr.max_write_request_size = pg_block_size;
+        }
+    }
+    if (!msgr.max_write_request_size)
+    {
+        msgr.max_write_request_size = MAX_DATA_BLOCK_SIZE;
+    }
 }

 void osd_t::apply_pg_locks_localize_only()
@@ -4,7 +4,8 @@
 #include "osd.h"

 #define FLUSH_BATCH 512
-#define SELF_CLIENT 0
+
+static_assert(FLUSH_BATCH <= MAX_SIMPLE_PAYLOAD_SIZE / sizeof(obj_ver_id));

 void osd_t::submit_pg_flush_ops(pg_t & pg)
 {
@@ -9,8 +9,6 @@
 #include "str_util.h"
 #include "osd.h"

-#define SELF_CLIENT 0
-
 // Peering loop
 void osd_t::handle_peers()
 {
@@ -71,6 +71,21 @@ bool osd_t::prepare_primary_rw(osd_op_t *cur_op)
        finish_op(cur_op, -EINVAL);
        return false;
    }
+    if (use_auth && cur_op->client_id != SELF_CLIENT)
+    {
+        osd_client_t *cl = msgr.clients.at(cur_op->client_id);
+        if (cl->hs_result.peer_is_osd)
+        {
+            // OSDs are not allowed to execute "primary" operations
+            finish_op(cur_op, -EPERM);
+            return false;
+        }
+        if (!st_cli.check_image_perm(cl->user_info, cur_op->req.rw.inode, (cur_op->req.hdr.opcode != OSD_OP_READ)))
+        {
+            finish_op(cur_op, -EPERM);
+            return false;
+        }
+    }
    int stripe_count = (cur_op->req.hdr.opcode == OSD_OP_SCRUB ? 0 :
        (pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 1 : pg_it->second.pg_size));
    int chain_size = 0;
@@ -96,7 +111,7 @@ bool osd_t::prepare_primary_rw(osd_op_t *cur_op)
            if (inode_it->second.parent_id == cur_op->req.rw.inode ||
                inode_it->second.parent_id == inode_it->second.num ||
                chain_size > st_cli.inode_config.size() ||
-                chain_size > 255)
+                chain_size > UINT32_MAX)
            {
                printf("Inode %ju from pool %u has too many parents, returning EINVAL in response to read\n",
                    INODE_NO_POOL(cur_op->req.rw.inode), INODE_POOL(cur_op->req.rw.inode));
@@ -111,7 +126,7 @@ bool osd_t::prepare_primary_rw(osd_op_t *cur_op)
            // Add the original inode
            chain_size++;
            chain_info_len = (cur_op->req.rw.flags & OSD_OP_RETURN_CHAIN
-                ? (cur_op->req.rw.len / bs_bitmap_granularity)
+                ? osd_op_rw_t::chain_info_bytes(chain_size) * (cur_op->req.rw.len / bs_bitmap_granularity)
                : 0);
        }
    }
@@ -130,9 +145,7 @@ bool osd_t::prepare_primary_rw(osd_op_t *cur_op)
            stripe_count * clean_entry_bitmap_size +
            // - 'missing' flags for chained reads
            (pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 0 : pg_it->second.pg_size)
-        ) +
-        // read chain info
-        chain_info_len
+        )
    );
    void *data_buf = (uint8_t*)op_data + sizeof(osd_primary_op_data_t);
    op_data->pg_num = pg_num;
@@ -140,6 +153,7 @@ bool osd_t::prepare_primary_rw(osd_op_t *cur_op)
    op_data->oid = oid;
    op_data->stripes = (osd_rmw_stripe_t*)data_buf;
    op_data->stripe_count = stripe_count;
+    op_data->chain_info = NULL;
    data_buf = (uint8_t*)data_buf + sizeof(osd_rmw_stripe_t) * stripe_count;
    cur_op->op_data = op_data;
    if (cur_op->req.hdr.opcode != OSD_OP_SCRUB)
@@ -147,10 +161,17 @@ bool osd_t::prepare_primary_rw(osd_op_t *cur_op)
        split_stripes(pg_data_size, bs_block_size, (uint32_t)(cur_op->req.rw.offset - oid.stripe), cur_op->req.rw.len, op_data->stripes);
        // Resulting bitmaps have to survive op_data and be freed with the op itself
        assert(!cur_op->bitmap_buf);
-        cur_op->bitmap_buf = calloc_or_die(1, clean_entry_bitmap_size * stripe_count);
+        cur_op->bitmap_buf = (uint8_t*)calloc_or_die(1, clean_entry_bitmap_size*stripe_count + chain_info_len);
+        uint8_t *buf = cur_op->bitmap_buf;
        for (int i = 0; i < stripe_count; i++)
        {
-            op_data->stripes[i].bmp_buf = (uint8_t*)cur_op->bitmap_buf + clean_entry_bitmap_size * i;
+            op_data->stripes[i].bmp_buf = buf;
+            buf += clean_entry_bitmap_size;
+            if (i == pg_data_size-1 && chain_info_len)
+            {
+                op_data->chain_info = buf;
+                buf += chain_info_len;
+            }
        }
    }
    op_data->chain_size = chain_size;
@@ -164,11 +185,6 @@ bool osd_t::prepare_primary_rw(osd_op_t *cur_op)
        data_buf = (uint8_t*)data_buf + chain_size * stripe_count * clean_entry_bitmap_size;
        op_data->missing_flags = (uint8_t*)data_buf;
        data_buf = (uint8_t*)data_buf + chain_size * (pool_cfg.scheme == POOL_SCHEME_REPLICATED ? 0 : pg_it->second.pg_size);
-        if (chain_info_len)
-        {
-            op_data->chain_info = (uint8_t*)data_buf;
-            data_buf = (uint8_t*)data_buf + chain_info_len;
-        }
        // Copy chain
        int chain_num = 0;
        op_data->read_chain[chain_num] = cur_op->req.rw.inode;
@@ -319,6 +335,7 @@ resume_2:
    }
    cur_op->reply.rw.version = op_data->fact_ver;
    cur_op->reply.rw.bitmap_len = (pg ? pg->pg_data_size : 1) * clean_entry_bitmap_size;
+    cur_op->bitmap = op_data->stripes[0].bmp_buf;
    if (op_data->degraded)
    {
        // Reconstruct missing stripes
@@ -331,7 +348,6 @@ resume_2:
        {
            reconstruct_stripes_ec(stripes, pg->pg_size, pg->pg_data_size, clean_entry_bitmap_size);
        }
-        cur_op->iov.push_back(op_data->stripes[0].bmp_buf, cur_op->reply.rw.bitmap_len);
        for (int role = 0; role < pg->pg_size; role++)
        {
            if (stripes[role].req_end != 0)
@@ -346,7 +362,6 @@ resume_2:
    }
    else
    {
-        cur_op->iov.push_back(op_data->stripes[0].bmp_buf, cur_op->reply.rw.bitmap_len);
        cur_op->iov.push_back(cur_op->buf, cur_op->req.rw.len);
    }
    finish_op(cur_op, cur_op->req.rw.len);
@@ -56,7 +56,7 @@ struct osd_primary_op_data_t
            int chain_size;
            osd_chain_read_t *chain_reads;
            int chain_read_count;
-            uint8_t *chain_info;
+            void *chain_info;
        };
    };
 };
@@ -582,11 +582,10 @@ void osd_t::send_chained_read_results(pg_t *pg, osd_op_t *cur_op)
        }
    }
    // Send bitmap
+    cur_op->bitmap = op_data->stripes[0].bmp_buf;
    cur_op->reply.rw.bitmap_len = (pg ? pg->pg_data_size : 1) * clean_entry_bitmap_size;
-    cur_op->iov.push_back(op_data->stripes[0].bmp_buf, cur_op->reply.rw.bitmap_len);
    if (cur_op->req.rw.flags & OSD_OP_RETURN_CHAIN)
    {
-        cur_op->iov.push_back(op_data->chain_info, (cur_op->req.rw.len / bs_bitmap_granularity));
        cur_op->reply.rw.bitmap_len += (cur_op->req.rw.len / bs_bitmap_granularity);
    }
    // And finally compose the result
@@ -596,6 +595,7 @@ void osd_t::send_chained_read_results(pg_t *pg, osd_op_t *cur_op)
    int prev = (cur_op->req.rw.offset - op_data->oid.stripe) / bs_bitmap_granularity;
    int end = prev + cur_op->req.rw.len/bs_bitmap_granularity;
    int cur = prev;
+    size_t key_index_bytes = osd_op_rw_t::chain_info_bytes(op_data->chain_size);
    while (cur <= end)
    {
        bool has_bit = false;
@@ -607,7 +607,14 @@ void osd_t::send_chained_read_results(pg_t *pg, osd_op_t *cur_op)
                if (has_bit)
                {
                    if (op_data->chain_info)
-                        op_data->chain_info[cur] = pos;
+                    {
+                        if (key_index_bytes == 1)
+                            ((uint8_t*)op_data->chain_info)[cur] = pos;
+                        else if (key_index_bytes == 2)
+                            ((uint16_t*)op_data->chain_info)[cur] = pos;
+                        else
+                            ((uint32_t*)op_data->chain_info)[cur] = pos;
+                    }
                    break;
                }
            }
@@ -91,6 +91,21 @@ static void scan_lists(std::vector<unclean_list_t> & lists, uint64_t limit, desc
 // Describe unclean objects
 void osd_t::continue_primary_describe(osd_op_t *cur_op)
 {
+    if (use_auth)
+    {
+        osd_client_t *cl = msgr.clients.at(cur_op->client_id);
+        if (cl->hs_result.peer_is_osd)
+        {
+            // OSDs are not allowed to execute "primary" operations
+            finish_op(cur_op, -EPERM);
+            return;
+        }
+        if (cl->user_info->type != user_type_t::ADMIN)
+        {
+            finish_op(cur_op, -EPERM);
+            return;
+        }
+    }
    auto & desc = cur_op->req.describe;
    if (!desc.object_state)
        desc.object_state = ~desc.object_state;
@@ -3,8 +3,6 @@

 #include "osd_primary.h"

-#define SELF_CLIENT 0
-
 void osd_t::autosync()
 {
    if (immediate_commit != IMMEDIATE_ALL && !autosync_op)
@@ -83,6 +81,8 @@ void osd_t::finish_op(osd_op_t *cur_op, int retval)
            rm_inflight(pg);
        }
        assert(!cur_op->op_data->subops);
+        free(cur_op->op_data);
+        cur_op->op_data = NULL;
    }
    cur_op->reply.hdr.magic = SECONDARY_OSD_REPLY_MAGIC;
    cur_op->reply.hdr.id = cur_op->req.hdr.id;
@@ -8,6 +8,16 @@ void osd_t::continue_primary_sync(osd_op_t *cur_op)
 {
    if (!cur_op->op_data)
    {
+        if (use_auth && cur_op->client_id != SELF_CLIENT)
+        {
+            osd_client_t *cl = msgr.clients.at(cur_op->client_id);
+            if (cl->hs_result.peer_is_osd)
+            {
+                // OSDs are not allowed to execute "primary" operations
+                finish_op(cur_op, -EPERM);
+                return;
+            }
+        }
        cur_op->op_data = (osd_primary_op_data_t*)calloc_or_die(1, sizeof(osd_primary_op_data_t));
    }
    osd_primary_op_data_t *op_data = cur_op->op_data;
@@ -3,8 +3,6 @@

 #include "osd_primary.h"

-#define SELF_CLIENT 0
-
 void osd_t::scrub_list(pool_pg_num_t pg_id, osd_num_t role_osd, object_id min_oid)
 {
    pool_id_t pool_id = pg_id.pool_id;
@@ -413,7 +411,7 @@ void osd_t::submit_scrub_subops(osd_op_t *cur_op)
        }
    }
    assert(!cur_op->bitmap_buf);
-    cur_op->bitmap_buf = calloc_or_die(1, clean_entry_bitmap_size * op_data->stripe_count);
+    cur_op->bitmap_buf = (uint8_t*)calloc_or_die(1, clean_entry_bitmap_size * op_data->stripe_count);
    for (int i = 0; i < op_data->stripe_count; i++)
    {
        op_data->stripes[i].bmp_buf = (uint8_t*)cur_op->bitmap_buf + clean_entry_bitmap_size * i;
@@ -1,12 +1,13 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 (see README.md for details)

+#include "json11/json11.hpp"
+
 #include "osd.h"
 #ifdef WITH_RDMA
 #include "msgr_rdma.h"
 #endif
-
-#include "json11/json11.hpp"
+#include "openssl_util.h"

 void osd_t::secondary_op_callback(osd_op_t *op)
 {
@@ -112,6 +113,31 @@ bool osd_t::sec_check_pg_lock(osd_num_t primary_osd, const object_id &oid, uint3

 void osd_t::exec_secondary_real(osd_op_t *cur_op)
 {
+    osd_client_t *cl = msgr.clients.at(cur_op->client_id);
+    if (use_auth && !cl->hs_result.peer_is_osd)
+    {
+        // Non-OSDs are not allowed to execute "secondary" operations except LIST
+        bool allowed = false;
+        if (cur_op->req.hdr.opcode == OSD_OP_SEC_LIST)
+        {
+            if (cl->user_info->type == user_type_t::ADMIN)
+            {
+                // Admin is allowed to execute arbitrary listings
+                allowed = true;
+            }
+            else if (cl->user_info->type == user_type_t::CLIENT && cur_op->req.sec_list.min_inode &&
+                cur_op->req.sec_list.min_inode == cur_op->req.sec_list.max_inode)
+            {
+                // Clients are only allowed to execute listings for readable inodes
+                allowed = st_cli.check_image_perm(cl->user_info, cur_op->req.sec_list.min_inode, false);
+            }
+        }
+        if (!allowed)
+        {
+            finish_op(cur_op, -EPERM);
+            return;
+        }
+    }
    if (cur_op->req.hdr.opcode == OSD_OP_SEC_LIST &&
        (cur_op->req.sec_list.flags & OSD_LIST_PRIMARY))
    {
@@ -128,7 +154,6 @@ void osd_t::exec_secondary_real(osd_op_t *cur_op)
        exec_sec_lock(cur_op);
        return;
    }
-    osd_client_t *cl = msgr.clients.at(cur_op->client_id);
    cur_op->bs_op = new blockstore_op_t();
    cur_op->bs_op->callback = [this, cur_op](blockstore_op_t* bs_op) { secondary_op_callback(cur_op); };
    cur_op->bs_op->opcode = (cur_op->req.hdr.opcode == OSD_OP_SEC_READ ? BS_OP_READ
@@ -348,6 +373,13 @@ void osd_t::exec_show_config(osd_op_t *cur_op)
        cl->read_op_id = cur_op->req.hdr.id + 1;
    }
    auto features = json11::Json::object{ { "pg_locks", true } };
+#ifdef WITH_RDMA
+    if (msgr.is_rdma_enabled() && req_json["features"]["rdma_close_tcp"].bool_value())
+    {
+        cl->rdma_close_tcp = true;
+        features["rdma_close_tcp"] = true;
+    }
+#endif
    if (msgr.use_proto_checksums)
    {
        auto peer_csums = req_json["features"]["proto_checksums"].uint64_value();
@@ -62,14 +62,6 @@ target_link_libraries(test_cas
 	vitastor_client
 )

-# http_hello
-add_executable(http_hello
-	http_hello.cpp
-)
-target_link_libraries(http_hello
-	vitastor_client
-)
-
 # test_crc32
 add_executable(test_crc32
 	test_crc32.cpp
@@ -1,61 +0,0 @@
-#include <sys/types.h>
-#include <sys/socket.h>
-#include <netinet/in.h>
-#include <netinet/tcp.h>
-#include <arpa/inet.h>
-#include <string.h>
-#include <stdio.h>
-#include <unistd.h>
-#include <fcntl.h>
-#include <errno.h>
-#include <stdlib.h>
-
-#include <stdexcept>
-
-#include "http_client.h"
-#include "ringloop.h"
-#include "epoll_manager.h"
-#include "addr_util.h"
-
-int main(int narg, char *args[])
-{
-    ring_consumer_t looper;
-    ring_loop_t *ringloop = new ring_loop_t(RINGLOOP_DEFAULT_SIZE);
-    epoll_manager_t *epmgr = new epoll_manager_t(ringloop);
-    // Accept new connections
-    int listen_fd = create_and_bind_socket("0.0.0.0", 8085, 128, NULL);
-    fcntl(listen_fd, F_SETFL, fcntl(listen_fd, F_GETFL, 0) | O_NONBLOCK);
-    std::string error;
-    auto http_ctx = http_context_init(epmgr->tfd, "", "", "", false, error);
-    epmgr->set_fd_handler(listen_fd, false, [http_ctx](int listen_fd, int events)
-    {
-        sockaddr_storage addr;
-        socklen_t peer_addr_size = sizeof(addr);
-        int peer_fd;
-        while ((peer_fd = accept(listen_fd, (sockaddr*)&addr, &peer_addr_size)) >= 0)
-        {
-            assert(peer_fd != 0);
-            fcntl(peer_fd, F_SETFL, fcntl(peer_fd, F_GETFL, 0) | O_NONBLOCK);
-            int one = 1;
-            setsockopt(peer_fd, SOL_TCP, TCP_NODELAY, &one, sizeof(one));
-            auto co = http_init(http_ctx);
-            http_serve(co, peer_fd, (http_options_t){}, [co](http_message_t *msg)
-            {
-                if (msg->eof)
-                {
-                    http_destroy(co);
-                    return;
-                }
-                http_reply(co, "HTTP/1.1 200 OK\r\nConnection: keep-alive\r\nContent-Length: 13\r\n\r\nHello, world!");
-            });
-        }
-    });
-    while (true)
-    {
-        ringloop->loop();
-        ringloop->wait();
-    }
-    delete epmgr;
-    delete ringloop;
-    return 0;
-}
@@ -26,7 +26,7 @@ void osd_messenger_t::outbox_push(osd_op_t *cur_op)
    cl->sent_ops[cur_op->req.hdr.id] = cur_op;
 }

-void osd_messenger_t::parse_config(const json11::Json & config)
+void osd_messenger_t::parse_config(const json11::Json & config, bool init)
 {
 }

@@ -39,7 +39,7 @@ int main(int narg, char *args[])
    msgr->repeer_pgs = [](osd_num_t) {};
    msgr->exec_op = [msgr](osd_op_t *op) { stub_exec_op(msgr, op); };
    json11::Json config = json11::Json::object { { "log_level", 1 } };
-    msgr->parse_config(config);
+    msgr->parse_config(config, true);
    // Accept new connections
    int listen_fd = create_and_bind_socket("0.0.0.0", 11203, 128, NULL);
    fcntl(listen_fd, F_SETFL, fcntl(listen_fd, F_GETFL, 0) | O_NONBLOCK);
@@ -1,9 +1,7 @@
 // Copyright (c) Vitaliy Filippov, 2019+
 // License: VNPL-1.1 (see README.md for details)

-#ifdef WITH_OPENSSL
 #include <openssl/rand.h>
-#endif

 #include <stdio.h>
 #include <stdlib.h>
@@ -556,7 +554,6 @@ void test_writeback_merge()
    printf("[ok] writeback merge test\n");
 }

-#ifdef WITH_OPENSSL
 void test_msgr_encrypt()
 {
    const size_t sz = 1048576;
@@ -595,7 +592,7 @@ void test_msgr_encrypt()
    enc->start(key, 4096 * 114, 4096);
    in_pos = out_pos = 0;
    enc->update(src+4096, 4096, crypt2, 4095, in_pos, out_pos);
-    assert(in_pos == 4096);
+    assert(in_pos == 4095);
    assert(out_pos == 4095);
    enc->update(src+4096+in_pos, 4096-in_pos, crypt2+out_pos, 4096-out_pos, in_pos, out_pos);
    assert(in_pos == 4096);
@@ -626,10 +623,10 @@ void test_msgr_encrypt()
    assert(in_pos == 3000);
    assert(out_pos == 0);
    dec->update(crypt+4096+3000, 3000, decrypt, 500, in_pos, out_pos);
-    assert(in_pos == 4096);
+    assert(in_pos == 4095);
    assert(out_pos == 500);
    dec->update(crypt+4096+in_pos, 6000-in_pos, decrypt+out_pos, 3000, in_pos, out_pos);
-    assert(in_pos == 4096);
+    assert(in_pos == 4095);
    assert(out_pos == 3500);
    dec->update(crypt+4096+in_pos, 6000-in_pos, decrypt+out_pos, 1000, in_pos, out_pos);
    assert(in_pos == 4096);
@@ -638,10 +635,10 @@ void test_msgr_encrypt()
    assert(in_pos == 6000);
    assert(out_pos == 4096);
    dec->update(crypt+4096+in_pos, 8192-in_pos, decrypt+out_pos, 4500-out_pos, in_pos, out_pos);
-    assert(in_pos == 8192);
+    assert(in_pos == 8191);
    assert(out_pos == 4500);
    dec->update(crypt+4096+in_pos, 8192-in_pos, decrypt+out_pos, 7500-out_pos, in_pos, out_pos);
-    assert(in_pos == 8192);
+    assert(in_pos == 8191);
    assert(out_pos == 7500);
    dec->update(crypt+4096+in_pos, 8192-in_pos, decrypt+out_pos, 8192-out_pos, in_pos, out_pos);
    assert(in_pos == 8192);
@@ -725,7 +722,6 @@ void test_msgr_decrypt_chain()
    free(src);
    printf("[ok] msgr aes-xts chained decrypt\n");
 }
-#endif

 void test_vault()
 {
@@ -794,10 +790,8 @@ int main(int narg, char *args[])
    test2();
    test_writeback();
    test_writeback_merge();
-#ifdef WITH_OPENSSL
    test_msgr_encrypt();
    test_msgr_decrypt_chain();
-#endif
    test_vault();
    return 0;
 }
@@ -1893,7 +1893,8 @@ void test_big_intent_csums()
        heap.finish_recheck();

        auto mod = heap.get_recheck_modified_blocks();
-        assert(mod.size() == 0);
+        assert(mod.size() == 1);
+        assert(mod[0] == 0);

        // read object 1 - big_intent should be there
        object_id oid = { .inode = INODE_WITH_POOL(1, 1), .stripe = 0 };
@@ -0,0 +1,114 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#include "openssl_util.h"
+#include "str_util.h"
+
+#include <openssl/ssl.h>
+
+X509 *openssl_load_cert(const std::string & file_or_pem)
+{
+    std::string pem;
+    BIO *bio = NULL;
+    if (file_or_pem.substr(0, 5) != "-----")
+    {
+        pem = read_file(file_or_pem);
+        bio = BIO_new_mem_buf(pem.data(), pem.size());
+    }
+    else
+        bio = BIO_new_mem_buf(file_or_pem.data(), file_or_pem.size());
+    if (!bio)
+        return NULL;
+    X509 *x509 = PEM_read_bio_X509(bio, NULL, 0, NULL);
+    BIO_free(bio);
+    return x509;
+}
+
+EVP_PKEY *openssl_load_key(const std::string & file_or_pem)
+{
+    std::string pem;
+    BIO *bio = NULL;
+    if (file_or_pem.substr(0, 5) != "-----")
+    {
+        pem = read_file(file_or_pem);
+        bio = BIO_new_mem_buf(pem.data(), pem.size());
+    }
+    else
+        bio = BIO_new_mem_buf(file_or_pem.data(), file_or_pem.size());
+    if (!bio)
+        return NULL;
+    EVP_PKEY *pkey = PEM_read_bio_PrivateKey(bio, NULL, NULL, NULL);
+    BIO_free(bio);
+    return pkey;
+}
+
+bool openssl_ctx_add_ca(SSL_CTX *ssl_ctx, const std::string & file_or_pem)
+{
+    X509 *cert = openssl_load_cert(file_or_pem);
+    bool ok = !!cert;
+    if (cert)
+    {
+        X509_STORE *store = SSL_CTX_get_cert_store(ssl_ctx);
+        X509_STORE_add_cert(store, cert);
+        X509_free(cert);
+    }
+    return ok;
+}
+
+bool openssl_ctx_use_ca(SSL_CTX *ssl_ctx, const std::string & file_or_pem)
+{
+    if (file_or_pem.substr(0, 5) == "-----")
+    {
+        return openssl_ctx_add_ca(ssl_ctx, file_or_pem);
+    }
+    return file_or_pem.empty()
+        ? !!SSL_CTX_set_default_verify_paths(ssl_ctx)
+        : !!SSL_CTX_load_verify_locations(ssl_ctx, file_or_pem.c_str(), NULL);
+}
+
+std::string openssl_get_cn(X509 *x509)
+{
+    X509_NAME* subj = X509_get_subject_name(x509);
+    int pos = X509_NAME_get_index_by_NID(subj, NID_commonName, -1);
+    if (pos != -1)
+    {
+        X509_NAME_ENTRY* cn = X509_NAME_get_entry(subj, pos);
+        ASN1_STRING* str = X509_NAME_ENTRY_get_data(cn);
+        return std::string((const char*)ASN1_STRING_get0_data(str), ASN1_STRING_length(str));
+    }
+    return "";
+}
+
+bool openssl_ctx_use_cert(SSL_CTX *ssl_ctx, const std::string & file_or_pem, std::string & common_name)
+{
+    X509 *cert = openssl_load_cert(file_or_pem);
+    bool ok = false;
+    if (cert)
+    {
+        common_name = openssl_get_cn(cert);
+        ok = SSL_CTX_use_certificate(ssl_ctx, cert);
+        X509_free(cert);
+    }
+    return ok;
+}
+
+bool openssl_ctx_use_key(SSL_CTX *ssl_ctx, const std::string & file_or_pem)
+{
+    EVP_PKEY *pkey = openssl_load_key(file_or_pem);
+    bool ok = false;
+    if (pkey)
+    {
+        ok = SSL_CTX_use_PrivateKey(ssl_ctx, pkey);
+        EVP_PKEY_free(pkey);
+    }
+    return ok;
+}
+
+bool openssl_bio_nonempty(BIO *bio)
+{
+    // SSL_ERROR_WANT_WRITE is absolutely non-informative with memory BIO, it basically never happens
+    // So we have to check memory BIO for outstanding data
+    char *bio_buf = NULL;
+    size_t bio_sz = BIO_get_mem_data(bio, &bio_buf);
+    return bio_sz > 0;
+}
@@ -0,0 +1,19 @@
+// Copyright (c) Vitaliy Filippov, 2019+
+// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
+
+#pragma once
+
+#include <string>
+
+#ifdef WITH_OPENSSL
+#include <openssl/types.h>
+#endif
+
+X509 *openssl_load_cert(const std::string & file_or_pem);
+EVP_PKEY *openssl_load_key(const std::string & file_or_pem);
+bool openssl_ctx_add_ca(SSL_CTX *ssl_ctx, const std::string & file_or_pem);
+bool openssl_ctx_use_ca(SSL_CTX *ssl_ctx, const std::string & file_or_pem);
+std::string openssl_get_cn(X509 *x509);
+bool openssl_ctx_use_cert(SSL_CTX *ssl_ctx, const std::string & file_or_pem, std::string & common_name);
+bool openssl_ctx_use_key(SSL_CTX *ssl_ctx, const std::string & file_or_pem);
+bool openssl_bio_nonempty(BIO *bio);
@@ -535,12 +535,12 @@ std::string urldecode(const std::string & orig)
    return res;
 }

-size_t fromhexstr(const std::string & from, size_t bytes, uint8_t *to)
+size_t fromhexstr(const char *from, size_t from_len, uint8_t *to, size_t to_len)
 {
-    if (bytes > from.size()/2)
-        bytes = from.size()/2;
+    if (to_len > from_len/2)
+        to_len = from_len/2;
    size_t i = 0;
-    while (i < bytes)
+    while (i < to_len)
    {
        uint8_t x = fromhexchar(from[2*i], 16);
        uint8_t y = fromhexchar(from[2*i+1], 16);
@@ -552,6 +552,11 @@ size_t fromhexstr(const std::string & from, size_t bytes, uint8_t *to)
    return i;
 }

+size_t fromhexstr(const std::string & from, size_t bytes, uint8_t *to)
+{
+    return fromhexstr(from.data(), from.size(), to, bytes);
+}
+
 std::string tohexstr(const uint8_t *from, size_t bytes)
 {
    std::string res;
@@ -35,6 +35,7 @@ std::string realpath_str(std::string path, bool nofail = true);
 std::string format_datetime(uint64_t unixtime);
 bool is_zero(void *buf, size_t size);
 std::string urldecode(const std::string & orig);
+size_t fromhexstr(const char *from, size_t from_len, uint8_t *to, size_t to_len);
 size_t fromhexstr(const std::string & from, size_t bytes, uint8_t *to);
 std::string tohexstr(const uint8_t *from, size_t bytes);
 bool ishexstr(const std::string & str);
@@ -34,6 +34,8 @@ extern "C" {
 # define XXH_NOESCAPE
 #endif

+#define XXH_SECRET_DEFAULT_SIZE 192
+
 typedef enum {
    XXH_OK = 0,
    XXH_ERROR
@@ -27,6 +27,8 @@ ETCD_COUNT=${ETCD_COUNT:-1}
 ANTIETCD=${ANTIETCD}
 USE_RAMDISK=${USE_RAMDISK}
 ETCD_SCHEME=${ETCD_SCHEME:-http}
+OSD_TLS=${OSD_TLS:-1}
+RDMA=${RDMA:-0}

 RAMDISK=/run/user/$(id -u)
 findmnt $RAMDISK >/dev/null || (sudo mkdir -p $RAMDISK && sudo mount -t tmpfs tmpfs $RAMDISK)
@@ -124,6 +126,26 @@ VITASTOR_CFG='"etcd_address":"'$ETCD_URL'"'"$VITASTOR_CFG"
 if [[ "$ETCD_SCHEME" = "https" ]]; then
    VITASTOR_CFG="$VITASTOR_CFG"',"etcd_ca":"'$(pwd)'/testdata/etcd.crt"'
 fi
+VITASTOR_CFG="$VITASTOR_CFG"',"osd_network":"'$ETCD_IP'/32"'
+VITASTOR_CFG="$VITASTOR_CFG"',"use_rdma":"'$RDMA'"'
+if [[ "$OSD_TLS" = "1" ]]; then
+    cd ./testdata
+    openssl req -days 3650 -x509 -new -newkey rsa:4096 -nodes -keyout client_ca.key -out client_ca.crt \
+        -subj '/C=RU/ST=Russia/L=Moscow/O=VitastorClientCA'
+    openssl req -days 3650 -x509 -new -newkey rsa:4096 -nodes -keyout osd.key -out osd.crt \
+        -subj '/C=RU/ST=Russia/L=Moscow/O=VitastorOSD' -addext "extendedKeyUsage = serverAuth, clientAuth"
+    openssl req -subj '/CN=test' -nodes -new -keyout cli.key -out cli.csr -addext "extendedKeyUsage = clientAuth"
+    openssl x509 -req -days 3650 -CA client_ca.crt -CAkey client_ca.key -CAcreateserial -in cli.csr -out cli.crt
+    rm cli.csr
+    cd $(dirname $0)/..
+    VITASTOR_CFG="$VITASTOR_CFG"',"osd_cert":"'$(pwd)'/testdata/osd.crt"'
+    VITASTOR_CFG="$VITASTOR_CFG"',"osd_pkey":"'$(pwd)'/testdata/osd.key"'
+    VITASTOR_CFG="$VITASTOR_CFG"',"osd_ca":"'$(pwd)'/testdata/osd.crt"'
+    VITASTOR_CFG="$VITASTOR_CFG"',"client_ca":"'$(pwd)'/testdata/client_ca.crt"'
+    VITASTOR_CFG="$VITASTOR_CFG"',"cert":"'$(pwd)'/testdata/cli.crt"'
+    VITASTOR_CFG="$VITASTOR_CFG"',"pkey":"'$(pwd)'/testdata/cli.key"'
+    VITASTOR_CFG="$VITASTOR_CFG"',"use_auth":false'
+fi
 echo "{$VITASTOR_CFG}" > ./testdata/vitastor.conf
 VITASTOR_CFG=./testdata/vitastor.conf
 VITASTOR_CLI="build/src/cmd/vitastor-cli --config_path $VITASTOR_CFG"
@@ -40,7 +40,7 @@ start_osd_on()
 {
    local i=$1
    local dev=$2
-    build/src/osd/vitastor-osd --osd_num $i --bind_address $ETCD_IP $NO_SAME $OSD_ARGS \
+    build/src/osd/vitastor-osd --osd_num $i $NO_SAME $OSD_ARGS \
        $(build/src/disk_tool/vitastor-disk simple-offsets --format options $OFFSET_ARGS $dev $OFFSET_ARGS 2>/dev/null) \
        >>./testdata/osd$i.log 2>&1 &
    eval OSD${i}_PID=$!
@@ -119,6 +119,14 @@ if [[ $OSD_COUNT -gt 0 ]]; then
    wait_up 120
 fi

+if [[ "$RDMA" = 1 ]]; then
+    for i in $(seq 1 $OSD_COUNT); do
+        if ! grep 'RDMA initialized successfully' ./testdata/osd$i.log; then
+            format_error "OSD $i failed to initialize RDMA, please try to enable rdma-rxe on the host"
+        fi
+    done
+fi
+
 try_reweight()
 {
    osd=$1
@@ -100,6 +100,7 @@ PG_SIZE=2 ./test_heal.sh
 TEST_NAME=local_read POOLCFG='"local_reads":"random",' ./test_heal.sh
 SCHEME=ec ./test_heal.sh
 ANTIETCD=1 ./test_heal.sh
+TEST_NAME=ec_rdma RDMA=1 SCHEME=ec ./test_heal.sh

 ./test_checksum.sh
 TEST_NAME=xxhash OSD_ARGS="--data_csum_type xxh3_32" ./test_checksum.sh
@@ -62,7 +62,7 @@ if ! diff -q ./testdata/bin/read.bin ./testdata/bin/mirror.bin; then
    format_error Data lost during self-heal
 fi

-if grep -qP 'Checksum mismatch|BUG' ./testdata/osd*.log; then
+if grep -qP 'Checksum mismatch|BUG|Received garbage' ./testdata/osd*.log; then
    format_error Checksum mismatches or BUGs detected during test
 fi

@@ -9,7 +9,7 @@ OSD_COUNT=5
 OSD_ARGS="$OSD_ARGS"
 for i in $(seq 1 $OSD_COUNT); do
    dd if=/dev/zero of=./testdata/bin/test_osd$i.bin bs=1024 count=1 seek=$((OSD_SIZE*1024-1))
-    build/src/osd/vitastor-osd --log_level 10 --osd_num $i --bind_address 127.0.0.1 --etcd_stats_interval 5 $OSD_ARGS \
+    build/src/osd/vitastor-osd --log_level 10 --osd_num $i --etcd_stats_interval 5 $OSD_ARGS \
        $(build/src/disk_tool/vitastor-disk simple-offsets --format options ./testdata/bin/test_osd$i.bin $OFFSET_ARGS 2>/dev/null) >>./testdata/osd$i.log 2>&1 &
    eval OSD${i}_PID=$!
 done
@@ -70,7 +70,7 @@ done
 $ETCDCTL del --prefix /vitastor/osd/state/

 for i in $(seq 1 $OSD_COUNT); do
-    build/src/osd/vitastor-osd --osd_num $i --bind_address 127.0.0.1 $NO_SAME $OSD_ARGS \
+    build/src/osd/vitastor-osd --osd_num $i $NO_SAME $OSD_ARGS \
        --meta_format $meta_format \
        --data_device ./testdata/bin/test_osd$i.bin \
        --meta_offset 0 \
@@ -21,7 +21,7 @@ $VITASTOR_FIO -bs=4k -direct=1 -iodepth=1 -fsync=1 \
 # Kill OSD 2, start OSD 1

 kill $OSD2_PID
-build/src/osd/vitastor-osd --osd_num 1 --bind_address 127.0.0.1 $NO_SAME $OSD_ARGS \
+build/src/osd/vitastor-osd --osd_num 1 $NO_SAME $OSD_ARGS \
    $(build/src/disk_tool/vitastor-disk simple-offsets --format options --device ./testdata/bin/test_osd2.bin $OFFSET_ARGS 2>/dev/null) >>./testdata/osd2.log 2>&1 &
 sleep 2

@@ -15,7 +15,7 @@ OSD_COUNT=3
 OSD_ARGS="$OSD_ARGS"
 OFFSET_ARGS="$OFFSET_ARGS"
 for i in $(seq 1 $OSD_COUNT); do
-    build/src/osd/vitastor-osd --osd_num $i --bind_address 127.0.0.1 $OSD_ARGS \
+    build/src/osd/vitastor-osd --osd_num $i $OSD_ARGS \
        $(build/src/disk_tool/vitastor-disk simple-offsets --format options ./testdata/bin/test_osd$i.bin $OFFSET_ARGS 2>/dev/null) >>./testdata/osd$i.log 2>&1 &
    eval OSD${i}_PID=$!
 done
Author	SHA1	Message	Date
Vitaliy Filippov	32721046b8	WIP Switch back to TCP when RDMA send fails (does not work)	2026-05-01 22:26:41 +00:00
Vitaliy Filippov	f8de201625	Close TCP connections for RDMA clients	2026-05-01 22:13:29 +00:00
Vitaliy Filippov	e7f30c4939	Also check for "Received garbage" during test	2026-05-01 16:01:28 +03:00
Vitaliy Filippov	8e441f4173	Add a CI test with RDMA	2026-05-01 15:57:15 +03:00
Vitaliy Filippov	80a05c052a	Add request size validation to prevent OOMDoS	2026-05-01 15:57:15 +03:00
Vitaliy Filippov	a972c13571	Implement OSD-side authorization for operations	2026-04-29 02:42:23 +03:00
Vitaliy Filippov	9372df4beb	Remove tls_ from option names	2026-04-27 15:41:00 +03:00
Vitaliy Filippov	24c27386b6	Remove TLS support (superseded by direct GCM)	2026-04-27 15:41:00 +03:00
Vitaliy Filippov	394de5e9aa	Implement TLS 1.3-like handshake manually	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	be61e76df0	Extract openssl-related code, wire ssl implementation back	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	5cad126987	Coalesce entries in send_out_buf	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	afa50a05b8	Support isa-l_crypto for AES-XTS too	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	9ff833178e	Fix xts+rdma encrypt errors	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	b152baa733	Fix "use-after-realloc" warning	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	8dbbf4097a	Support isa-l_crypto for AES-GCM	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	17cb875e10	Remove WITH_OPENSSL from all files except http_client, always require OpenSSL	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	03e15338e5	Use pools for GCM contexts	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	dccc549e33	Try to run tests with AES-GCM	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	440a8bc33a	Implement direct AES-256-GCM with a static key for benchmark	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	78a21542da	Make sure to send all TLS data before continuing	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	75ba9509ad	Parse standard TLS record headers	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	8b58fe9808	Omit msgr_tls_record_hdr_t for non-tls data	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	331f345a78	Allow 2 and 4 byte per block chain_info (allow more than 255 snapshots with encryption)	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	d0a4ac914b	Implement OSD TLS support	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	e42b812323	Allow to skip checksums for headers	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	ac5c753735	Implement protocol-level checksums (xxhash3)	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	f9eb497105	Include xxhash3 x86dispatch	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	3489c99640	Do not use scrap_buffer in the client (it would block protocol checksum support)	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	a21dcd4480	Support TLS CN authentication and per-image permissions in vitastor-cli serve	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	340849248b	Add VitastorAuthFilter	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	ee462c4375	Implement vitastor-cli ls-user, modify-user, remove-user commands	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	0053a0d825	Add image owner/owner_group/reader_group support (for antietcd VitastorAuthFilter)	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	cdeecead8d	Add security parameter documentation	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	c0e34b19c9	Support inline (string PEM) certificates and pkeys	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	e8abe99205	Show encryption keys (only IDs) in the listing	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	8225b37f53	Support storing image encryption keys in Vault	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	62b29b03c5	Prefer local etcd addresses and correctly cycle over them even when they need resolving Seems slightly overcomplicated...	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	ae2f570b58	Support DNS resolving via libc-ares	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	d7f854ab3e	Batch handle_immediate_ops more	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	29640d1aac	Add vitastor-cli create & modify --enc-key parameter	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	66a016f8a1	Support reading from snapshots encrypted with different keys	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	34d01f44db	Support decryption with multiple keys	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	3956982112	Allow to return chain_info in response to reads	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	713b17e574	Add basic AES-XTS client-side encryption support	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	4efa48c9ca	Rework msgr send/receive to allow encryption support	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	26d5e1ed00	Move fromhexstr() to str_util	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	9ba3444cfa	Add openapi description	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	96c0001127	Slightly fix API return and input types	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	56a14cc0bd	Implement vitastor-cli serve command to serve simple HTTP API	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	6ca1b9eca9	Implement HTTP server support O_o	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	876dbd730a	Rename http_response_t to http_message_t	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	4398ee0377	Extract common HTTP context	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	2c673a90bd	Support xxhash 32-bit checksums (data_csum_type=xxh3_32)	2026-04-27 15:24:44 +03:00
Vitaliy Filippov	a8371498e1	Detect block checksums using csum_block_size, not data_csum_type	2026-04-27 15:24:43 +03:00
Vitaliy Filippov	2a093cbfb5	Add client certificate support	2026-04-27 15:24:43 +03:00
Vitaliy Filippov	094a181491	Do not re-initialize TLS context every connection	2026-04-27 15:24:43 +03:00
Vitaliy Filippov	fe30783135	Add https support to antietcd	2026-04-27 15:24:43 +03:00
Vitaliy Filippov	e9a6c58418	Implement etcd SSL support via OpenSSL Maybe I should remove all of this and use libwebsockets :)	2026-04-27 15:24:43 +03:00
Vitaliy Filippov	4acfe149cb	Release 3.0.10 Important bug fixes (new store): - Fix OSDs possibly refusing to start with "write metadata failed at offset xxx: Invalid argument" (fix buffer alignment during initial garbage collection) - Rollback change from 3.0.4 - on-disk garbage entries are not skipped on start again. This change doesn't have any impact normally, but OSDs originally running 3.0.0-3.0.2 and then upgraded to 3.0.9 may hit a bug where 3.0.9 refuses to start due to entries marked as garbage too early and flushed to disk in 3.0.0-3.0.2. Other changes: - Auto-select the only RDMA device/port if there is only one - Rollback one 3.0.9 change - there was no actual use-after-free :) (the problem was only relevant to an unstable development version) - Fix `vitastor-nfs --trace` option - Fix an unintended 1 second sleep in vitastor-cli rm-data - Fix inode statistics not being cleared for a deleted pool - Fix print to stdout in client	2026-04-27 13:53:48 +03:00
Vitaliy Filippov	008ed5b269	Rollback `25ecca7625` - there was no actual use-after-free :)	2026-04-22 01:35:38 +03:00
Vitaliy Filippov	4fffe0f032	Fix vitastor-nfs trace option	2026-04-20 21:51:47 +03:00
Vitaliy Filippov	a76d5ccc0d	Wakeup callers in rm-data	2026-04-17 13:53:32 +03:00
Vitaliy Filippov	8ed1e180e0	Clear inode_stats in mon	2026-04-14 02:44:46 +03:00
Vitaliy Filippov	8832fc3b14	Fix print to stdout in client	2026-04-14 02:43:50 +03:00
Vitaliy Filippov	0134934c99	Do not skip garbage entries on start (rollback change from 3.0.4)	2026-04-11 12:05:02 +03:00
Vitaliy Filippov	2e36f292bd	Fix buffer alignment during metadata clearing on init	2026-04-10 21:56:12 +03:00
Vitaliy Filippov	bcc6419760	Auto-select the only RDMA device/port if there is only one	2026-04-08 15:46:10 +03:00