Implement handshake for AES-256-GCM by capturing TLS secrets (pretty shitty approach actually...)

Seems slightly overcomplicated...

debian/vitastor-buildenv.Dockerfile

@@ -12,20 +12,18 @@ ARG REL=
 WORKDIR /root
 
 RUN set -e -x; \
-    if [ "$REL" = "buster" ]; then \
-        perl -i -pe 's/deb.debian.org/archive.debian.org/' /etc/apt/sources.list; \
-        apt-get update; \
-        apt-get -y install wget; \
-        wget https://vitastor.io/debian/pubkey.gpg -O /etc/apt/trusted.gpg.d/vitastor.gpg; \
-        echo "deb https://vitastor.io/debian $REL main" >> /etc/apt/sources.list; \
-    fi; \
+    perl -i -pe 's/deb.debian.org/archive.debian.org/' /etc/apt/sources.list; \
+    apt-get update; \
+    apt-get -y install wget; \
+    wget https://vitastor.io/debian/pubkey.gpg -O /etc/apt/trusted.gpg.d/vitastor.gpg; \
+    echo "deb https://vitastor.io/debian $REL main" >> /etc/apt/sources.list; \
     grep '^deb ' /etc/apt/sources.list | perl -pe 's/^deb/deb-src/' >> /etc/apt/sources.list; \
     perl -i -pe 's/Types: deb$/Types: deb deb-src/' /etc/apt/sources.list.d/*.sources || true; \
     echo 'APT::Install-Recommends false;' >> /etc/apt/apt.conf; \
     echo 'APT::Install-Suggests false;' >> /etc/apt/apt.conf
 
 RUN apt-get update && \
-    apt-get -y install fio libgoogle-perftools-dev devscripts libjerasure-dev cmake 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

docs/config/security.en.md

@@ -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.

docs/config/security.ru.md

@@ -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.

docs/config/src/security.yml

@@ -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.

docs/usage/nfs.en.md

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

docs/usage/nfs.ru.md

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

mon/stats.js

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

src/CMakeLists.txt

@@ -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})

src/blockstore/blockstore_heap.cpp

@@ -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)) ||

src/blockstore/blockstore_init.cpp

@@ -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)

src/client/CMakeLists.txt

@@ -12,11 +12,11 @@ if (RDMACM_LIBRARIES)
 	set(MSGR_RDMACM "msgr_rdmacm.cpp")
 endif (RDMACM_LIBRARIES)
 add_library(vitastor_common STATIC
-	../util/epoll_manager.cpp etcd_state_client.cpp messenger.cpp ../util/addr_util.cpp ../util/xxh_x86dispatch.c
+	../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_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)

src/client/cluster_client.cpp

@@ -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)
             {

src/client/cluster_client.h

@@ -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;

src/client/http_client.cpp

@@ -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)
 {

src/client/http_client.h

@@ -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

src/client/messenger.cpp

@@ -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)
     {
@@ -339,21 +297,15 @@ 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)
@@ -388,9 +340,9 @@ 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;
+    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())
@@ -420,7 +372,7 @@ void osd_messenger_t::parse_config(const json11::Json & config)
     if (!this->receive_buffer_size || this->receive_buffer_size > 1024*1024*1024)
         this->receive_buffer_size = 65536;
     this->use_sync_send_recv = config["use_sync_send_recv"].bool_value() ||
-        config["use_sync_send_recv"].uint64_value();
+        config["use_sync_send_recv"].uint64_value() || !ringloop;
     this->min_zerocopy_send_size = config["min_zerocopy_send_size"].is_null()
         ? DEFAULT_MIN_ZEROCOPY_SEND_SIZE
         : (int)config["min_zerocopy_send_size"].int64_value();
@@ -642,10 +594,7 @@ 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);
 }
 
@@ -887,10 +836,7 @@ 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 +851,6 @@ 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)
 {

src/client/messenger.h

@@ -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"
@@ -44,6 +46,9 @@
 
 #define DEFAULT_MIN_ZEROCOPY_SEND_SIZE 32*1024
 
+#define AES_256_GCM_KEY_SIZE 32
+#define AES_256_GCM_IV_SIZE 12
+
 struct msgr_sendp_t
 {
     osd_op_t *op;
@@ -60,10 +65,12 @@ struct op_aes_xts_decrypt_t;
 void destroy_aes_xts_encrypt(op_aes_xts_encrypt_t *encrypt_ctx);
 void destroy_aes_xts_decrypt(op_aes_xts_decrypt_t *decrypt_ctx);
 
+// Standard TLS record header. We are only interested in the record size
 struct __attribute__((__packed__)) msgr_tls_record_hdr_t
 {
-    uint8_t encrypted;
-    uint32_t size;
+    uint8_t content_type;
+    uint16_t version;
+    uint16_t size;
 };
 
 struct osd_client_t
@@ -88,18 +95,36 @@ 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;
+    int ssl_handshake_pending = 0;
+    msgr_tls_record_hdr_t ssl_read_record;
+    size_t ssl_read_header_size = 0;
+    bool ssl_more_to_buffer = false;
+
+    bool gcm_enabled = false;
+    std::vector<uint8_t> my_secret, peer_secret;
+    std::vector<uint8_t> my_key, peer_key;
+    uint64_t my_iv_ctr, peer_iv_ctr;
+#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 +132,11 @@ 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 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 +162,7 @@ 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 +241,21 @@ struct osd_messenger_t;
 struct rdmacm_connecting_t;
 #endif
 
+class msgr_op_reader_t;
+class msgr_op_writer_t;
+
 struct __attribute__((visibility("default"))) osd_messenger_t
 {
 protected:
+    friend class copy_op_reader_t;
+    friend class ssl_op_reader_t;
+    friend class gcm_op_reader_t;
+    friend class get_op_reader_t;
+    friend class copy_op_writer_t;
+    friend class ssl_op_writer_t;
+    friend class gcm_op_writer_t;
+    friend class get_op_writer_t;
+
     int keepalive_timer_id = -1;
 
     uint32_t receive_buffer_size = 0;
@@ -230,7 +267,7 @@ 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 +290,19 @@ protected:
     robin_hood::unordered_flat_map<rdma_cm_id*, rdmacm_connecting_t*> rdmacm_connecting;
 #endif
 
-#ifdef WITH_OPENSSL
     SSL_CTX *ssl_ctx = NULL;
+    EVP_KDF_CTX *kdf_ctx = NULL;
+    X509 *tls_cert_obj = NULL;
+    X509 *osd_tls_ca_obj = NULL;
+    X509 *client_tls_ca_obj = NULL;
     std::string tls_cn;
 
-    void ssl_init(osd_client_t *cl, bool server_mode);
-    bool ssl_do_handshake(osd_client_t *cl);
-    bool ssl_do_encrypt(osd_client_t *cl);
-    size_t ssl_do_encrypt_to(osd_client_t *cl, uint8_t *buf, size_t size);
-    bool ssl_op_write_buf(osd_client_t *cl, uint8_t *src, size_t src_len, bool skip_csum, size_t & from, size_t & done);
-    size_t ssl_op_copy_to(osd_client_t *cl, uint8_t *dst, size_t dst_len);
-    void ssl_op_get_write_buffers(osd_client_t *cl, std::vector<iovec> & lst);
-#endif
+    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 finalize_tls_handshake(osd_client_t *cl);
+    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;
@@ -272,8 +310,16 @@ 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;
@@ -342,32 +388,32 @@ 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> size_t 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);

src/client/msgr_encrypt.cpp

@@ -3,13 +3,30 @@
 
 #include <assert.h>
 
+#ifdef WITH_ISAL_CRYPTO
+#include <isa-l_crypto/isal_crypto_api.h>
+#endif
+
+#include <mutex>
+
+#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/kdf.h>
+#include <openssl/ssl.h>
+#include <openssl/err.h>
+
+#define MSGR_HSP_HS   1
+#define MSGR_HSP_SEND 2
+#define MSGR_HSP_RECV 4
 
 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 +38,13 @@ 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 +66,7 @@ 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 +77,12 @@ 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 +115,10 @@ 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 +146,7 @@ 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 +178,7 @@ 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 +190,27 @@ 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 +221,7 @@ 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 +233,35 @@ 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 +296,10 @@ 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 +328,7 @@ 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 +360,65 @@ 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 +450,7 @@ 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 +467,328 @@ 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;
+    }
+}
+
+struct tls_secrets_t
+{
+    std::vector<uint8_t> client_secret;
+    std::vector<uint8_t> server_secret;
+};
+
+// Sadly we have to use a global variable to capture TLS 1.3 secrets
+static std::mutex logged_secrets_mu;
+static std::map<const SSL*, tls_secrets_t> logged_secrets;
+
+static void openssl_key_log(const SSL *ssl, const char *line)
+{
+    // Format: <CLIENT|SERVER>_TRAFFIC_SECRET_0 <server_random> <secret>
+    bool is_client_secret = !strncmp(line, "CLIENT_TRAFFIC_SECRET_0 ", strlen("CLIENT_TRAFFIC_SECRET_0 "));
+    bool is_server_secret = !strncmp(line, "SERVER_TRAFFIC_SECRET_0 ", strlen("SERVER_TRAFFIC_SECRET_0 "));
+    if (!is_client_secret && !is_server_secret)
+        return;
+    const char *hex = strchr(line+strlen("CLIENT_TRAFFIC_SECRET_0 "), ' ');
+    if (!hex)
+        return;
+    hex++;
+    size_t len = strlen(hex);
+    logged_secrets_mu.lock();
+    auto & secrets = logged_secrets[ssl];
+    logged_secrets_mu.unlock();
+    auto & secret = is_client_secret ? secrets.client_secret : secrets.server_secret;
+    secret.resize(len/2);
+    fromhexstr(hex, len, secret.data(), secret.size());
+}
+
+static bool derive_kdf(EVP_KDF_CTX* kdf_ctx, const uint8_t* insecret, size_t insecret_len,
+    const uint8_t* salt, size_t salt_len, const char *label, uint8_t *key, size_t size)
+{
+    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);
+    if (salt)
+        params[n++] = OSSL_PARAM_construct_octet_string("salt", (void*)salt, 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)
+    {
+        ERR_print_errors_fp(stderr);
+        return false;
+    }
+    if (EVP_KDF_derive(kdf_ctx, key, size, NULL) <= 0)
+    {
+        ERR_print_errors_fp(stderr);
+        return false;
+    }
+    return true;
+}
+
+bool osd_messenger_t::derive_aes_keys(osd_client_t *cl, bool update_my, bool update_peer)
+{
+    std::vector<uint8_t> old_my = cl->my_key, old_peer = cl->peer_key;
+    if (!cl->my_secret.size() || !cl->peer_secret.size())
+    {
+        assert(cl->ssl_cli);
+        logged_secrets_mu.lock();
+        auto & secrets = logged_secrets[cl->ssl_cli];
+        cl->my_secret = std::move(cl->is_incoming ? secrets.client_secret : secrets.server_secret);
+        cl->peer_secret = std::move(!cl->is_incoming ? secrets.client_secret : secrets.server_secret);
+        logged_secrets.erase(cl->ssl_cli);
+        logged_secrets_mu.unlock();
+        SSL_free(cl->ssl_cli);
+        cl->ssl_cli = NULL;
+        cl->gcm_enabled = true;
+        cl->write_to_ssl = NULL;
+        cl->read_from_ssl = NULL;
+        if (cl->my_secret.size() < 32 || cl->peer_secret.size() < 32)
+        {
+            fprintf(stderr, "Client %ju error: failed to capture TLS handshake results\n", cl->client_id);
+            return false;
+        }
+    }
+    // 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 && derive_kdf(kdf_ctx, cl->my_secret.data(), cl->my_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 && derive_kdf(kdf_ctx, cl->peer_secret.data(), cl->peer_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())
+    {
+        // Initialize TLS context
+        if (tls_cert.empty() || tls_key.empty() || osd_tls_ca.empty() || osd_num && client_tls_ca.empty())
+        {
+            if (osd_num)
+                fprintf(stderr, "Vitastor OSD TLS requires osd_tls_cert, osd_tls_key, osd_tls_ca, client_tls_ca\n");
+            else
+                fprintf(stderr, "Vitastor client TLS requires tls_cert, tls_key and osd_tls_ca\n");
+            exit(1);
+        }
+        else
+        {
+            ssl_ctx = SSL_CTX_new(TLS_method());
+            if (!ssl_ctx)
+            {
+init_err:
+                fprintf(stderr, "OpenSSL initialization failed: %s\n", ERR_error_string(ERR_get_error(), NULL));
+                exit(1);
+            }
+            // Always use TLS 1.3 with AES-256-GCM
+            SSL_CTX_set_min_proto_version(ssl_ctx, TLS1_3_VERSION);
+            SSL_CTX_set_max_proto_version(ssl_ctx, TLS1_3_VERSION);
+            SSL_CTX_set_ciphersuites(ssl_ctx, "TLS_AES_256_GCM_SHA384");
+            SSL_CTX_set_keylog_callback(ssl_ctx, openssl_key_log);
+            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 && (osd_tls_ca_obj = openssl_load_cert(osd_tls_ca));
+            ok = ok && X509_STORE_add_cert(SSL_CTX_get_cert_store(ssl_ctx), osd_tls_ca_obj);
+            if (osd_num)
+            {
+                // OSD uses 2 separate root certificates to distinguish between clients and peer OSDs
+                ok = ok && (client_tls_ca_obj = openssl_load_cert(client_tls_ca));
+                ok = ok && X509_STORE_add_cert(SSL_CTX_get_cert_store(ssl_ctx), client_tls_ca_obj);
+            }
+            ok = ok && openssl_ctx_use_cert(ssl_ctx, tls_cert, tls_cn);
+            ok = ok && openssl_ctx_use_key(ssl_ctx, tls_key);
+            EVP_KDF *kdf;
+            ok = ok && (kdf = EVP_KDF_fetch(NULL, "hkdf", NULL));
+            ok = ok && (kdf_ctx = EVP_KDF_CTX_new(kdf));
+            if (kdf)
+                EVP_KDF_free(kdf);
+            if (!ok)
+            {
+                SSL_CTX_free(ssl_ctx);
+                ssl_ctx = NULL;
+                goto init_err;
+            }
+        }
+    }
+}
+
+void osd_messenger_t::init_tls_client(osd_client_t *cl)
+{
+    if (!tls_cert.empty())
+    {
+        cl->write_to_ssl = BIO_new(BIO_s_mem());
+        cl->read_from_ssl = BIO_new(BIO_s_mem());
+        cl->ssl_cli = SSL_new(ssl_ctx);
+        cl->ssl_handshake_pending = MSGR_HSP_HS;
+        if (!cl->ssl_cli)
+        {
+            fprintf(stderr, "OpenSSL initialization failed: %s\n", ERR_error_string(ERR_get_error(), NULL));
+            exit(1);
+        }
+        if (cl->is_incoming)
+        {
+            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 = do_tls_handshake(cl);
+        assert(ok);
+    }
+}
+
+bool osd_messenger_t::do_tls_handshake(osd_client_t *cl, bool from_recv)
+{
+    if (!(cl->ssl_handshake_pending & MSGR_HSP_HS))
+        return true;
+    int r = SSL_do_handshake(cl->ssl_cli);
+    if (r > 0)
+    {
+        // Server-side OpenSSL treats handshake as finalized only when receiving
+        // the first message, so we transmit 1 byte after connecting and only then
+        // finalize the handshake
+        cl->ssl_handshake_pending = MSGR_HSP_SEND|MSGR_HSP_RECV;
+        if (cl->write_state == 0 && from_recv)
+        {
+            cl->write_state = CL_WRITE_READY;
+            write_ready_clients.push_back(cl->client_id);
+        }
+    }
+    else
+    {
+        r = SSL_get_error(cl->ssl_cli, r);
+        if (r != 0 && r != SSL_ERROR_WANT_READ && r != SSL_ERROR_WANT_WRITE)
+        {
+            fprintf(stderr, "Client %ju TLS handshake error: %s, stopping client\n", cl->client_id, ERR_error_string(ERR_get_error(), NULL));
+            cl->io_error = true;
+            return false;
+        }
+        if (from_recv && cl->write_state == 0 && openssl_bio_nonempty(cl->read_from_ssl))
+        {
+            cl->write_state = CL_WRITE_READY;
+            write_ready_clients.push_back(cl->client_id);
+        }
+    }
+    return true;
+}
+
+bool osd_messenger_t::finalize_tls_handshake(osd_client_t *cl)
+{
+    if (cl->ssl_handshake_pending)
+        return true;
+    // Capture secrets and switch to direct AES-256-GCM encryption
+    if (!derive_aes_keys(cl, true, true))
+    {
+        cl->io_error = true;
+        return false;
+    }
+    if (cl->read_op)
+    {
+        assert(!cl->read_op_pos);
+        delete cl->read_op;
+        cl->read_op = NULL;
+    }
+    if (cl->write_op)
+    {
+        assert(!cl->write_op_pos);
+        cl->write_ops.insert(cl->write_ops.begin(), cl->write_op);
+        cl->write_op = NULL;
+    }
+    if (cl->write_state == 0)
+    {
+        cl->write_state = CL_WRITE_READY;
+        write_ready_clients.push_back(cl->client_id);
+    }
+    // Switched to direct AES-GCM, stop SSL callers
+    return false;
+}
+
+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 (osd_tls_ca_obj)
+    {
+        X509_free(osd_tls_ca_obj);
+        osd_tls_ca_obj = NULL;
+    }
+    if (client_tls_ca_obj)
+    {
+        X509_free(client_tls_ca_obj);
+        client_tls_ca_obj = NULL;
+    }
+    if (ssl_ctx)
+    {
+        SSL_CTX_free(ssl_ctx);
+        ssl_ctx = NULL;
+    }
+    if (kdf_ctx)
+    {
+        EVP_KDF_CTX_free(kdf_ctx);
+        kdf_ctx = NULL;
     }
 }

src/client/msgr_encrypt.h

@@ -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);
 };
 

src/client/msgr_op.cpp

@@ -8,6 +8,7 @@
 osd_op_t::~osd_op_t()
 {
     assert(!bs_op);
+    assert(!op_data);
     if (bitmap_buf)
     {
         free(bitmap_buf);
@@ -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()

src/client/msgr_op.h

@@ -182,7 +182,7 @@ struct __attribute__((visibility("default"))) osd_op_t
     void *bitmap = NULL;
     unsigned bitmap_len = 0;
     unsigned bmp_data = 0;
-    void *bitmap_buf = NULL;
+    uint8_t *bitmap_buf = NULL;
     void *rmw_buf = NULL;
     std::shared_ptr<osd_op_enc_t> enc;
     uint8_t *enc_buf = NULL;

src/client/msgr_rdma.cpp

@@ -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;

src/client/msgr_rdma.h

@@ -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();

src/client/msgr_rdmacm.cpp

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

src/client/msgr_stop.cpp

@@ -9,12 +9,10 @@
 #ifdef WITH_RDMA
 #include "msgr_rdma.h"
 #endif
-#ifdef WITH_OPENSSL
 #include <openssl/bio.h>
 #include <openssl/err.h>
 #include <openssl/pem.h>
 #include <openssl/ssl.h>
-#endif
 
 void osd_client_t::cancel_ops()
 {
@@ -86,21 +84,21 @@ 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 +221,10 @@ 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__
@@ -245,7 +246,24 @@ osd_client_t::~osd_client_t()
         XXH3_freeState(write_csum_state);
         write_csum_state = NULL;
     }
-#ifdef WITH_OPENSSL
+    if (enc_ctx)
+    {
+#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_cli)
     {
         SSL_free(ssl_cli);
@@ -258,5 +276,4 @@ osd_client_t::~osd_client_t()
         free(ssl_out_buf);
         ssl_out_buf = NULL;
     }
-#endif
 }

src/client/osd_ops.h

@@ -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

src/cmd/cli_create.cpp

@@ -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"
@@ -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();

src/cmd/cli_rm_data.cpp

@@ -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;
     }

src/nfs/nfs_proxy.cpp

@@ -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

src/osd/osd_primary.cpp

@@ -96,7 +96,7 @@ 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 +111,7 @@ 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 +130,7 @@ 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 +138,7 @@ 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 +146,17 @@ 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 +170,6 @@ 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 +320,7 @@ 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 +333,6 @@ 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 +347,6 @@ 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);

src/osd/osd_primary.h

@@ -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;
         };
     };
 };

src/osd/osd_primary_chain.cpp

@@ -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;
                 }
             }

src/osd/osd_primary_subops.cpp

@@ -83,6 +83,8 @@ 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;

src/osd/osd_scrub.cpp

@@ -413,7 +413,7 @@ 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;

src/test/CMakeLists.txt

@@ -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

src/test/http_hello.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;
-}

src/test/test_cluster_client.cpp

@@ -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;
 }

src/test/test_heap.cpp

@@ -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 };

src/util/openssl_util.cpp

@@ -0,0 +1,119 @@
+// 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;
+}
+
+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)
+{
+    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);
+}
+
+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;
+}

src/util/openssl_util.h

@@ -0,0 +1,18 @@
+// 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);
+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);

src/util/str_util.cpp

@@ -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;

src/util/str_util.h

@@ -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);

src/util/xxhash.h

@@ -34,6 +34,8 @@ extern "C" {
 # define XXH_NOESCAPE
 #endif
 
+#define XXH_SECRET_DEFAULT_SIZE 192
+
 typedef enum {
     XXH_OK = 0,
     XXH_ERROR

tests/common.sh

@@ -27,6 +27,7 @@ ETCD_COUNT=${ETCD_COUNT:-1}
 ANTIETCD=${ANTIETCD}
 USE_RAMDISK=${USE_RAMDISK}
 ETCD_SCHEME=${ETCD_SCHEME:-http}
+OSD_TLS=${OSD_TLS:-1}
 
 RAMDISK=/run/user/$(id -u)
 findmnt $RAMDISK >/dev/null || (sudo mkdir -p $RAMDISK && sudo mount -t tmpfs tmpfs $RAMDISK)
@@ -124,6 +125,23 @@ VITASTOR_CFG='"etcd_address":"'$ETCD_URL'"'"$VITASTOR_CFG"
 if [[ "$ETCD_SCHEME" = "https" ]]; then
     VITASTOR_CFG="$VITASTOR_CFG"',"etcd_ca":"'$(pwd)'/testdata/etcd.crt"'
 fi
+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_tls_cert":"'$(pwd)'/testdata/osd.crt"'
+    VITASTOR_CFG="$VITASTOR_CFG"',"osd_tls_key":"'$(pwd)'/testdata/osd.key"'
+    VITASTOR_CFG="$VITASTOR_CFG"',"osd_tls_ca":"'$(pwd)'/testdata/osd.crt"'
+    VITASTOR_CFG="$VITASTOR_CFG"',"client_tls_ca":"'$(pwd)'/testdata/client_ca.crt"'
+    VITASTOR_CFG="$VITASTOR_CFG"',"tls_cert":"'$(pwd)'/testdata/cli.crt"'
+    VITASTOR_CFG="$VITASTOR_CFG"',"tls_key":"'$(pwd)'/testdata/cli.key"'
+fi
 echo "{$VITASTOR_CFG}" > ./testdata/vitastor.conf
 VITASTOR_CFG=./testdata/vitastor.conf
 VITASTOR_CLI="build/src/cmd/vitastor-cli --config_path $VITASTOR_CFG"