HTTP "hello, world!"

Seems slightly overcomplicated...

CMakeLists.txt

@@ -2,7 +2,7 @@ cmake_minimum_required(VERSION 2.8...3.30)
 
 project(vitastor)
 
-set(VITASTOR_VERSION "3.0.8")
+set(VITASTOR_VERSION "3.0.9")
 
 include(CTest)
 

csi/Makefile

@@ -1,4 +1,4 @@
-VITASTOR_VERSION ?= v3.0.8
+VITASTOR_VERSION ?= v3.0.9
 
 all: build push
 

csi/deploy/004-csi-nodeplugin.yaml

@@ -49,7 +49,7 @@ spec:
             capabilities:
               add: ["SYS_ADMIN"]
             allowPrivilegeEscalation: true
-          image: vitalif/vitastor-csi:v3.0.8
+          image: vitalif/vitastor-csi:v3.0.9
           args:
             - "--node=$(NODE_ID)"
             - "--endpoint=$(CSI_ENDPOINT)"

csi/deploy/007-csi-provisioner.yaml

@@ -121,7 +121,7 @@ spec:
             privileged: true
             capabilities:
               add: ["SYS_ADMIN"]
-          image: vitalif/vitastor-csi:v3.0.8
+          image: vitalif/vitastor-csi:v3.0.9
           args:
             - "--node=$(NODE_ID)"
             - "--endpoint=$(CSI_ENDPOINT)"

csi/src/config.go

@@ -5,7 +5,7 @@ package vitastor
 
 const (
     vitastorCSIDriverName    = "csi.vitastor.io"
-    vitastorCSIDriverVersion = "3.0.8"
+    vitastorCSIDriverVersion = "3.0.9"
 )
 
 // Config struct fills the parameters of request or user input

debian/changelog

@@ -1,4 +1,4 @@
-vitastor (3.0.8-1) unstable; urgency=medium
+vitastor (3.0.9-1) unstable; urgency=medium
 
   * Bugfixes
 

docker/Makefile

@@ -1,4 +1,4 @@
-VITASTOR_VERSION ?= v3.0.8
+VITASTOR_VERSION ?= v3.0.9
 
 all: build push
 

docker/etc/vitastor/docker.conf

@@ -4,7 +4,7 @@
 #
 
 # Desired Vitastor version
-VITASTOR_VERSION=v3.0.8
+VITASTOR_VERSION=v3.0.9
 
 # Additional arguments for all containers
 # For example, you may want to specify a custom logging driver here

docs/config/osd.en.md

@@ -70,6 +70,7 @@ with an OSD restart or, for some of them, even without restarting by updating co
 - [use_atomic_flag](#use_atomic_flag)
 - [pg_reshard_chunk_size](#pg_reshard_chunk_size)
 - [pg_reshard_chunk_pause_ms](#pg_reshard_chunk_pause_ms)
+- [gc_on_start](#gc_on_start)
 
 ## bind_address
 
@@ -753,3 +754,9 @@ This option sets the maximum number of object is a chunk. Moving 100k objects us
 - Default: 100
 
 This option sets the interval between handling two PG count change chunks.
+
+## gc_on_start
+
+- Type: boolean
+
+Forcibly clean all garbage entries in the new store on every OSD restart.

docs/config/osd.ru.md

@@ -71,6 +71,7 @@
 - [use_atomic_flag](#use_atomic_flag)
 - [pg_reshard_chunk_size](#pg_reshard_chunk_size)
 - [pg_reshard_chunk_pause_ms](#pg_reshard_chunk_pause_ms)
+- [gc_on_start](#gc_on_start)
 
 ## bind_address
 
@@ -793,3 +794,9 @@ pg_minsize OSD во время переключений, что может по
 - Значение по умолчанию: 100
 
 Данная опция задаёт интервал между обработкой двух порций изменения числа PG пулов.
+
+## gc_on_start
+
+- Тип: булево (да/нет)
+
+Принудительно очищать все мусорные записи в новом хранилище при каждом запуске OSD.

docs/config/src/osd.yml

@@ -938,3 +938,7 @@
     This option sets the interval between handling two PG count change chunks.
   info_ru: |
     Данная опция задаёт интервал между обработкой двух порций изменения числа PG пулов.
+- name: gc_on_start
+  type: bool
+  info: Forcibly clean all garbage entries in the new store on every OSD restart.
+  info_ru: Принудительно очищать все мусорные записи в новом хранилище при каждом запуске OSD.

docs/installation/docker.en.md

@@ -26,9 +26,9 @@ at Vitastor Kubernetes operator: https://github.com/Antilles7227/vitastor-operat
 The instruction is very simple.
 
 1. Download a Docker image of the desired version: \
-   `docker pull vitalif/vitastor:v3.0.8`
+   `docker pull vitalif/vitastor:v3.0.9`
 2. Install scripts to the host system: \
-   `docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.0.8 install.sh`
+   `docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.0.9 install.sh`
 3. Reload udev rules: \
    `udevadm control --reload-rules`
 4. Enable the vitastor-host service: \

docs/installation/docker.ru.md

@@ -25,9 +25,9 @@ Vitastor можно установить в Docker/Podman. При этом etcd,
 Инструкция по установке максимально простая.
 
 1. Скачайте Docker-образ желаемой версии: \
-   `docker pull vitalif/vitastor:v3.0.8`
+   `docker pull vitalif/vitastor:v3.0.9`
 2. Установите скрипты в хост-систему командой: \
-   `docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.0.8 install.sh`
+   `docker run --rm -it -v /etc:/host-etc -v /usr/bin:/host-bin vitalif/vitastor:v3.0.9 install.sh`
 3. Перезагрузите правила udev: \
    `udevadm control --reload-rules`
 4. Включите сервис vitastor-host: \

mon/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vitastor-mon",
-  "version": "3.0.8",
+  "version": "3.0.9",
   "description": "Vitastor SDS monitor service",
   "main": "mon-main.js",
   "scripts": {

node-binding/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vitastor",
-  "version": "3.0.8",
+  "version": "3.0.9",
   "description": "Low-level native bindings to Vitastor client library",
   "main": "index.js",
   "keywords": [

patches/cinder-vitastor.py

@@ -50,7 +50,7 @@ from cinder.volume import configuration
 from cinder.volume import driver
 from cinder.volume import volume_utils
 
-VITASTOR_VERSION = '3.0.8'
+VITASTOR_VERSION = '3.0.9'
 
 LOG = logging.getLogger(__name__)
 

rpm/vitastor-el10.spec

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        3.0.8
+Version:        3.0.9
 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-3.0.8.el10.tar.gz
+Source0:        vitastor-3.0.9.el10.tar.gz
 
 BuildRequires:  gperftools-devel
 BuildRequires:  gcc-c++

rpm/vitastor-el7.spec

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        3.0.8
+Version:        3.0.9
 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-3.0.8.el7.tar.gz
+Source0:        vitastor-3.0.9.el7.tar.gz
 
 BuildRequires:  gperftools-devel
 BuildRequires:  devtoolset-9-gcc-c++

rpm/vitastor-el8.spec

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        3.0.8
+Version:        3.0.9
 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-3.0.8.el8.tar.gz
+Source0:        vitastor-3.0.9.el8.tar.gz
 
 BuildRequires:  gperftools-devel
 BuildRequires:  gcc-toolset-9-gcc-c++

rpm/vitastor-el9.spec

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        3.0.8
+Version:        3.0.9
 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-3.0.8.el9.tar.gz
+Source0:        vitastor-3.0.9.el9.tar.gz
 
 BuildRequires:  gperftools-devel
 BuildRequires:  gcc-c++

src/CMakeLists.txt

@@ -20,7 +20,7 @@ if("${CMAKE_INSTALL_PREFIX}" MATCHES "^/usr/local/?$")
 endif()
 set(ENABLE_COVERAGE false CACHE BOOL "Enable code coverage")
 
-add_definitions(-DVITASTOR_VERSION="3.0.8")
+add_definitions(-DVITASTOR_VERSION="3.0.9")
 add_definitions(-D_GNU_SOURCE -D_LARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -Wall -Wno-sign-compare -Wno-comment -Wno-parentheses -Wno-pointer-arith -fdiagnostics-color=always -fno-omit-frame-pointer -fvisibility=hidden -I ${CMAKE_SOURCE_DIR}/src)
 add_link_options(-fno-omit-frame-pointer)
 if (${WITH_ASAN})

src/blockstore/v1/impl.cpp

@@ -877,7 +877,7 @@ uint64_t blockstore_impl_t::get_garbage_entries()
 
 uint64_t blockstore_impl_t::get_garbage_memory()
 {
-    return sizeof(blockstore_dirty_db_t::node_type) * dirty_db.size();
+    return (sizeof(obj_ver_id) + sizeof(dirty_entry) + 32) * dirty_db.size();
 }
 
 } // namespace v1

src/client/http_client.cpp

@@ -164,30 +164,42 @@ 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) == "-----")
     {
-        BIO *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;
+        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);
 }
 
-static std::string openssl_get_cn(X509 *x509)
+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);

src/client/http_client.h

@@ -8,6 +8,10 @@
 #include <functional>
 #include "json11/json11.hpp"
 
+#ifdef WITH_OPENSSL
+#include <openssl/types.h>
+#endif
+
 #define WS_CONTINUATION 0
 #define WS_TEXT 1
 #define WS_BINARY 2
@@ -69,3 +73,11 @@ void http_close(http_co_t *co);
 void http_destroy(http_co_t *co);
 
 #pragma GCC visibility pop
+
+#ifdef WITH_OPENSSL
+bool openssl_ctx_add_ca(SSL_CTX *ssl_ctx, const std::string & file_or_pem);
+bool openssl_ctx_use_ca(SSL_CTX *ssl_ctx, const std::string & file_or_pem);
+std::string openssl_get_cn(X509 *x509);
+bool openssl_ctx_use_cert(SSL_CTX *ssl_ctx, const std::string & file_or_pem, std::string & common_name);
+bool openssl_ctx_use_key(SSL_CTX *ssl_ctx, const std::string & file_or_pem);
+#endif

src/client/messenger.cpp

@@ -14,6 +14,13 @@
 #ifdef WITH_RDMA
 #include "msgr_rdma.h"
 #endif
+#include "http_client.h"
+#ifdef WITH_OPENSSL
+#include <openssl/bio.h>
+#include <openssl/err.h>
+#include <openssl/pem.h>
+#include <openssl/ssl.h>
+#endif
 
 #include <sys/poll.h>
 
@@ -117,6 +124,43 @@ void msgr_iothread_t::run()
 
 void osd_messenger_t::init()
 {
+    if (!tls_cert.empty() || !tls_key.empty() || !osd_tls_ca.empty() || !client_tls_ca.empty())
+    {
+        // Initialize TLS context
+        // FIXME: require OpenSSL
+#ifndef WITH_OPENSSL
+        fprintf(stderr, "Vitastor is built without OpenSSL support\n");
+        exit(1);
+#else
+        if (tls_cert.empty() || tls_key.empty() || osd_tls_ca.empty() || osd_num && client_tls_ca.empty())
+        {
+            if (osd_num)
+                fprintf(stderr, "Vitastor OSD TLS requires osd_tls_cert, osd_tls_key, osd_tls_ca, client_tls_ca\n");
+            else
+                fprintf(stderr, "Vitastor client TLS requires tls_cert, tls_key and osd_tls_ca\n");
+            exit(1);
+        }
+        else
+        {
+            ssl_ctx = SSL_CTX_new(TLS_method());
+            SSL_CTX_set_verify(ssl_ctx, SSL_VERIFY_PEER, NULL);
+            bool ok = SSL_CTX_set_min_proto_version(ssl_ctx, TLS1_3_VERSION);
+            ok = ok && openssl_ctx_add_ca(ssl_ctx, osd_tls_ca);
+            if (osd_num)
+            {
+                // OSD uses 2 separate root certificates to distinguish between clients and peer OSDs
+                ok = ok && openssl_ctx_add_ca(ssl_ctx, client_tls_ca);
+            }
+            ok = ok && openssl_ctx_use_cert(ssl_ctx, tls_cert, tls_cn);
+            ok = ok && openssl_ctx_use_key(ssl_ctx, tls_key);
+            if (!ok)
+            {
+                SSL_CTX_free(ssl_ctx);
+                ssl_ctx = NULL;
+            }
+        }
+#endif
+    }
 #ifdef WITH_RDMACM
     if (use_rdmacm)
     {
@@ -303,6 +347,13 @@ osd_messenger_t::~osd_messenger_t()
     {
         destroy_aes_xts_decrypt(decrypt_ctx);
     }
+#ifdef WITH_OPENSSL
+    if (ssl_ctx)
+    {
+        SSL_CTX_free(ssl_ctx);
+        ssl_ctx = NULL;
+    }
+#endif
 }
 
 void osd_messenger_t::parse_config(const json11::Json & config)
@@ -340,7 +391,27 @@ void osd_messenger_t::parse_config(const json11::Json & config)
     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->use_proto_checksums = config["use_proto_checksums"].is_null() || config["use_proto_checksums"].bool_value();
+    if (config["proto_checksums"].is_null())
+        this->use_proto_checksums = MSGR_CSUM_PAYLOAD;
+    else if (config["proto_checksums"].is_bool())
+        this->use_proto_checksums = config["proto_checksums"].bool_value() ? MSGR_CSUM_FULL : 0;
+    else if (config["proto_checksums"].string_value() != "")
+        this->use_proto_checksums = config["proto_checksums"].string_value() == "full" ? MSGR_CSUM_FULL : MSGR_CSUM_PAYLOAD;
+    else
+        this->use_proto_checksums = 0;
+    if (!osd_num)
+    {
+        tls_cert = config["tls_cert"].string_value();
+        tls_key = config["tls_key"].string_value();
+        osd_tls_ca = config["osd_tls_ca"].string_value();
+    }
+    else
+    {
+        tls_cert = config["osd_tls_cert"].string_value();
+        tls_key = config["osd_tls_key"].string_value();
+        osd_tls_ca = config["osd_tls_ca"].string_value();
+        client_tls_ca = config["client_tls_ca"].string_value();
+    }
     if (!osd_num)
         this->iothread_count = (uint32_t)config["client_iothread_count"].uint64_value();
     else
@@ -571,6 +642,10 @@ void osd_messenger_t::handle_connect_epoll(int peer_fd)
         handle_peer_epoll(peer_fd, epoll_events);
     });
     // Check OSD number
+    if (!tls_cert.empty())
+    {
+        ssl_init(cl, false);
+    }
     check_peer_config(cl);
 }
 
@@ -663,7 +738,7 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
     auto features = json11::Json::object{ { "check_sequencing", true } };
     if (use_proto_checksums)
     {
-        features["proto_checksums"] = true;
+        features["proto_checksums"] = use_proto_checksums;
     }
     payload["features"] = features;
 #ifdef WITH_RDMA
@@ -740,9 +815,13 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
             delete op;
             return;
         }
-        if (use_proto_checksums && config["features"]["proto_checksums"].bool_value())
+        if (use_proto_checksums)
         {
-            cl->proto_csum_status = MSGR_PEER_CSUM_IN|MSGR_PEER_CSUM_OUT;
+            auto peer_csums = config["features"]["proto_checksums"].uint64_value();
+            if (peer_csums == MSGR_CSUM_FULL && use_proto_checksums == MSGR_CSUM_FULL)
+                cl->proto_csum_status = MSGR_CSUM_FULL;
+            else if (peer_csums && use_proto_checksums)
+                cl->proto_csum_status = MSGR_CSUM_PAYLOAD;
         }
 #ifdef WITH_RDMA
         if (!use_rdmacm && cl->rdma_conn && config["rdma_address"].is_string())
@@ -808,6 +887,10 @@ 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);
+        }
         // Add FD to epoll
         tfd->set_fd_handler(peer_fd, false, [this](int peer_fd, int epoll_events)
         {
@@ -822,6 +905,26 @@ void osd_messenger_t::accept_connections(int listen_fd)
     }
 }
 
+void osd_messenger_t::ssl_init(osd_client_t *cl, bool server_mode)
+{
+#ifdef WITH_OPENSSL
+    cl->write_to_ssl = BIO_new(BIO_s_mem());
+    cl->read_from_ssl = BIO_new(BIO_s_mem());
+    cl->ssl_cli = SSL_new(ssl_ctx);
+    if (server_mode)
+    {
+        SSL_set_accept_state(cl->ssl_cli);
+    }
+    else
+    {
+        SSL_set_connect_state(cl->ssl_cli);
+    }
+    SSL_set_bio(cl->ssl_cli, cl->write_to_ssl, cl->read_from_ssl);
+    bool ok = ssl_do_handshake(cl);
+    assert(ok);
+#endif
+}
+
 #ifdef WITH_RDMA
 msgr_rdma_context_t* osd_messenger_t::choose_rdma_context(osd_client_t *cl)
 {

src/client/messenger.h

@@ -12,6 +12,10 @@
 #include <deque>
 #include <vector>
 
+#ifdef WITH_OPENSSL
+#include <openssl/types.h>
+#endif
+
 #include "../util/xxh_x86dispatch.h"
 #include "../util/robin_hood.h"
 #include "malloc_or_die.h"
@@ -32,8 +36,9 @@
 #define PEER_RDMA 4
 #define PEER_STOPPED 5
 
-#define MSGR_PEER_CSUM_IN 1
-#define MSGR_PEER_CSUM_OUT 2
+#define MSGR_CSUM_PAYLOAD 1
+#define MSGR_CSUM_FULL 2
+#define MSGR_CSUM_NEG 4
 
 #define VITASTOR_CONFIG_PATH "/etc/vitastor/vitastor.conf"
 
@@ -55,6 +60,12 @@ 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);
 
+struct __attribute__((__packed__)) msgr_tls_record_hdr_t
+{
+    uint8_t encrypted;
+    uint32_t size;
+};
+
 struct osd_client_t
 {
     uint64_t client_id = 0;
@@ -77,6 +88,17 @@ 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;
+#endif
+
     // Read state
     int read_ready = 0;
     osd_op_t *read_op = NULL;
@@ -210,6 +232,11 @@ protected:
     int iothread_count = 0;
     int max_aes_xts_pool_size = 256;
 
+    std::string tls_cert;
+    std::string tls_key;
+    std::string osd_tls_ca;
+    std::string client_tls_ca;
+
 #ifdef WITH_RDMA
     bool use_rdma = true;
     bool use_rdmacm = false;
@@ -226,6 +253,19 @@ protected:
     robin_hood::unordered_flat_map<rdma_cm_id*, rdmacm_connecting_t*> rdmacm_connecting;
 #endif
 
+#ifdef WITH_OPENSSL
+    SSL_CTX *ssl_ctx = NULL;
+    std::string tls_cn;
+
+    void ssl_init(osd_client_t *cl, bool server_mode);
+    bool ssl_do_handshake(osd_client_t *cl);
+    bool ssl_do_encrypt(osd_client_t *cl);
+    size_t ssl_do_encrypt_to(osd_client_t *cl, uint8_t *buf, size_t size);
+    bool ssl_op_write_buf(osd_client_t *cl, uint8_t *src, size_t src_len, bool skip_csum, size_t & from, size_t & done);
+    size_t ssl_op_copy_to(osd_client_t *cl, uint8_t *dst, size_t dst_len);
+    void ssl_op_get_write_buffers(osd_client_t *cl, std::vector<iovec> & lst);
+#endif
+
     std::vector<msgr_iothread_t*> iothreads;
     std::vector<uint64_t> read_ready_clients;
     std::vector<uint64_t> write_ready_clients;
@@ -252,7 +292,7 @@ public:
     std::vector<addr_mask_t> osd_cluster_network_masks;
     std::vector<std::string> all_osd_networks;
     std::vector<addr_mask_t> all_osd_network_masks;
-    bool use_proto_checksums = true;
+    int use_proto_checksums = 0;
     // op statistics
     osd_op_stats_t stats, recovery_stats;
 
@@ -304,6 +344,9 @@ protected:
     void handle_send(int result, bool prev, bool more, osd_client_t *cl);
     size_t op_copy_to(osd_client_t *cl, uint8_t *dst, size_t dst_len);
     void op_get_write_buffers(osd_client_t *cl, std::vector<iovec> & lst);
+    void next_write_op(osd_client_t *cl);
+    bool op_write_buf(osd_client_t *cl, uint8_t *src, size_t src_len, uint8_t *dst, size_t dst_len, bool skip_csum, size_t & from, size_t & done);
+    bool op_copy_data_to(osd_client_t *cl, uint8_t *dst, size_t dst_len, size_t & from, size_t & done);
 
     void handle_read(int result, osd_client_t *cl);
     bool handle_read_buffer(osd_client_t *cl, uint8_t *curbuf, size_t bufsize);

src/client/msgr_rdma.cpp

@@ -187,6 +187,8 @@ std::vector<msgr_rdma_context_t*> msgr_rdma_context_t::create_all(const std::vec
     ibv_device **raw_dev_list = NULL;
     ibv_device **dev_list = NULL;
     ibv_device *single_list[2] = {};
+    int up_ports = 0;
+    int single_port_num = 0;
 
     raw_dev_list = dev_list = ibv_get_device_list(NULL);
     if (!dev_list || !*dev_list)
@@ -221,6 +223,7 @@ std::vector<msgr_rdma_context_t*> msgr_rdma_context_t::create_all(const std::vec
         dev_list = single_list;
     }
 
+retry:
     for (int i = 0; dev_list[i]; ++i)
     {
         auto dev = dev_list[i];
@@ -258,6 +261,9 @@ std::vector<msgr_rdma_context_t*> msgr_rdma_context_t::create_all(const std::vec
                 fprintf(stderr, "RDMA device %s port %d GID %d does not exist\n", ibv_get_device_name(dev), port_num, sel_gid_index);
                 continue;
             }
+            up_ports++;
+            single_port_num = port_num;
+            single_list[0] = dev;
             uint32_t port_mtu = sel_mtu ? sel_mtu : ibv_mtu_to_bytes(portinfo.active_mtu);
 #ifdef IBV_ADVISE_MR_ADVICE_PREFETCH_NO_FAULT
             if (sel_gid_index < 0)
@@ -298,6 +304,14 @@ cleanup_dev:
         ibv_close_device(context);
     }
 
+    if (!ret.size() && up_ports == 1 && dev_list != single_list)
+    {
+        // Auto-select the only available device/port if there is only one
+        dev_list = single_list;
+        sel_port_num = single_port_num;
+        goto retry;
+    }
+
 cleanup:
     if (raw_dev_list)
         ibv_free_device_list(raw_dev_list);
@@ -559,11 +573,7 @@ int osd_messenger_t::try_send_rdma_copy(osd_client_t *cl, uint8_t *dst, int dst_
     int total_dst_len = dst_len;
     while (dst_len > 0 && (cl->write_op || cl->write_ops.size()))
     {
-        if (!cl->write_op)
-        {
-            cl->write_op = cl->write_ops.front();
-            cl->write_ops.pop_front();
-        }
+        next_write_op(cl);
         osd_op_t *op = cl->write_op;
         size_t copied = op_copy_to(cl, dst, dst_len);
         if (!copied)
@@ -744,10 +754,10 @@ void osd_messenger_t::handle_rdma_events(msgr_rdma_context_t *rdma_context)
                     cl->send_free_ops.pop_front();
                 }
                 cl->send_free_ops.pop_front();
-                if (cl->proto_csum_status == MSGR_PEER_CSUM_IN && !cl->write_op && !cl->write_ops.size())
+                if ((cl->proto_csum_status & MSGR_CSUM_NEG) && !cl->write_op && !cl->write_ops.size())
                 {
                     // Checksums negotiated, enable
-                    cl->proto_csum_status = MSGR_PEER_CSUM_IN|MSGR_PEER_CSUM_OUT;
+                    cl->proto_csum_status = cl->proto_csum_status & (~MSGR_CSUM_NEG);
                 }
                 try_send_rdma(cl);
             }

src/client/msgr_receive.cpp

@@ -203,7 +203,7 @@ bool osd_messenger_t::handle_read_buffer(osd_client_t *cl, uint8_t *curbuf, size
             cl->read_op_size = 0;
             cl->read_op_inline_decrypt_in = 0;
             cl->read_op_inline_decrypt_pos = (size_t)-1;
-            if (cl->proto_csum_status == (MSGR_PEER_CSUM_IN|MSGR_PEER_CSUM_OUT))
+            if (cl->proto_csum_status == MSGR_CSUM_FULL || cl->proto_csum_status == MSGR_CSUM_PAYLOAD)
             {
                 if (!cl->read_csum_state)
                     cl->read_csum_state = XXH3_createState();
@@ -236,6 +236,10 @@ bool osd_messenger_t::handle_read_buffer(osd_client_t *cl, uint8_t *curbuf, size
 
 bool osd_messenger_t::handle_hdr(osd_client_t *cl)
 {
+    if (cl->proto_csum_status == MSGR_CSUM_FULL)
+    {
+        XXH3_64bits_update(cl->read_csum_state, cl->read_op->req.buf, OSD_PACKET_SIZE);
+    }
     if (cl->read_op->req.hdr.magic == SECONDARY_OSD_REPLY_MAGIC)
     {
         auto req_it = cl->sent_ops.find(cl->read_op->req.hdr.id);
@@ -333,7 +337,8 @@ bool osd_messenger_t::allocate_op_buffers(osd_client_t *cl)
         }
         cl->read_op_size = cur_op->req.show_conf.json_len;
     }
-    if (cl->proto_csum_status == (MSGR_PEER_CSUM_IN|MSGR_PEER_CSUM_OUT))
+    if (cl->proto_csum_status == MSGR_CSUM_FULL ||
+        cl->read_op_size > 0 && cl->proto_csum_status == MSGR_CSUM_PAYLOAD)
     {
         cl->read_op_size += 8;
     }
@@ -391,7 +396,8 @@ bool osd_messenger_t::allocate_reply_buffers(osd_client_t *cl, osd_op_t *op)
         free(op->buf);
         op->buf = malloc_or_die(op->reply.describe.result_bytes);
     }
-    if (cl->proto_csum_status == (MSGR_PEER_CSUM_IN|MSGR_PEER_CSUM_OUT))
+    if (cl->proto_csum_status == MSGR_CSUM_FULL ||
+        cl->read_op_size > 0 && cl->proto_csum_status == MSGR_CSUM_PAYLOAD)
     {
         cl->read_op_size += 8;
     }
@@ -429,10 +435,6 @@ bool osd_messenger_t::op_copy_from(osd_client_t *cl, uint8_t *src, size_t src_le
             from -= dst_len;
         return true;
     };
-    if (cl->read_csum_state && !from)
-    {
-        XXH3_64bits_update(cl->read_csum_state, (op->op_type == OSD_OP_IN ? op->req.buf : op->reply.buf), OSD_PACKET_SIZE);
-    }
     if (op->op_type == OSD_OP_IN)
     {
         if (op->req.hdr.opcode == OSD_OP_SEC_WRITE ||
@@ -520,7 +522,8 @@ bool osd_messenger_t::op_copy_from(osd_client_t *cl, uint8_t *src, size_t src_le
                 return true;
         }
     }
-    if (cl->proto_csum_status == (MSGR_PEER_CSUM_IN|MSGR_PEER_CSUM_OUT))
+    if (cl->proto_csum_status == MSGR_CSUM_FULL ||
+        cl->read_op_size > 0 && cl->proto_csum_status == MSGR_CSUM_PAYLOAD)
     {
         if (!op_read_buf((uint8_t*)&op->csum, 8, true))
             return true;
@@ -642,7 +645,8 @@ void osd_messenger_t::op_get_read_buffers(osd_client_t *cl, std::vector<iovec> &
                 return;
         }
     }
-    if (cl->proto_csum_status == (MSGR_PEER_CSUM_IN|MSGR_PEER_CSUM_OUT))
+    if (cl->proto_csum_status == MSGR_CSUM_FULL ||
+        cl->read_op_size > 0 && cl->proto_csum_status == MSGR_CSUM_PAYLOAD)
     {
         if (!op_read_buf((uint8_t*)&op->csum, 8))
             return;
@@ -676,7 +680,8 @@ void osd_messenger_t::op_alloc_temp_buffers(osd_op_t *op, int i)
 bool osd_messenger_t::handle_finished_op(osd_client_t *cl)
 {
     osd_op_t *op = cl->read_op;
-    if (cl->proto_csum_status == (MSGR_PEER_CSUM_IN|MSGR_PEER_CSUM_OUT))
+    if (cl->proto_csum_status == MSGR_CSUM_FULL ||
+        cl->read_op_size > 0 && cl->proto_csum_status == MSGR_CSUM_PAYLOAD)
     {
         uint64_t real_csum = XXH3_64bits_digest(cl->read_csum_state);
         if (op->csum != real_csum)

src/client/msgr_send.cpp

@@ -7,6 +7,13 @@
 
 #include "messenger.h"
 
+#ifdef WITH_OPENSSL
+#include <openssl/bio.h>
+#include <openssl/err.h>
+#include <openssl/pem.h>
+#include <openssl/ssl.h>
+#endif
+
 void osd_messenger_t::outbox_push(osd_op_t *cur_op)
 {
     assert(cur_op->client_id);
@@ -116,25 +123,103 @@ void osd_messenger_t::measure_exec(osd_op_t *cur_op)
     }
 }
 
+bool osd_messenger_t::ssl_do_handshake(osd_client_t *cl)
+{
+    int r = SSL_do_handshake(cl->ssl_cli);
+    if (r > 0)
+    {
+        cl->ssl_handshake_done = true;
+    }
+    else
+    {
+        r = SSL_get_error(cl->ssl_cli, r);
+        if (r != SSL_ERROR_WANT_WRITE && r != SSL_ERROR_WANT_READ)
+        {
+            fprintf(stderr, "Client %ju TLS handshake error: %s, stopping client\n", cl->client_id, ERR_error_string(ERR_get_error(), NULL));
+            stop_client(cl->client_id);
+            return false;
+        }
+        if (r == SSL_ERROR_WANT_WRITE)
+        {
+            cl->ssl_want_write = true;
+        }
+    }
+    return true;
+}
+
+bool osd_messenger_t::ssl_do_encrypt(osd_client_t *cl)
+{
+    if (cl->send_list.size() >= IOV_MAX || !cl->ssl_want_write)
+        return false;
+    size_t prev_size = cl->ssl_out_buf_size;
+    while (true)
+    {
+        size_t min_cap = cl->ssl_out_buf_size*2;
+        if (min_cap < 16384)
+            min_cap = 16384;
+        if (cl->ssl_out_buf_cap < min_cap)
+        {
+            uint8_t *old_buf = cl->ssl_out_buf;
+            uint8_t *old_end = old_buf + cl->ssl_out_buf_cap;
+            cl->ssl_out_buf = (uint8_t*)realloc_or_die(cl->ssl_out_buf, min_cap);
+            cl->ssl_out_buf_cap = min_cap;
+            for (auto & iov: cl->send_list)
+            {
+                if (iov.iov_base >= old_buf && iov.iov_base < old_end)
+                    iov.iov_base = cl->ssl_out_buf + ((uint8_t*)iov.iov_base - old_buf);
+            }
+        }
+        bool full_read = (ssl_do_encrypt_to(cl, cl->ssl_out_buf+cl->ssl_out_buf_size,
+            cl->ssl_out_buf_cap-cl->ssl_out_buf_size) == cl->ssl_out_buf_cap-cl->ssl_out_buf_size);
+        if (!full_read)
+            break;
+    }
+    if (cl->ssl_out_buf_size > prev_size)
+        cl->send_list.push_back((iovec){ .iov_base = cl->ssl_out_buf+prev_size, .iov_len = cl->ssl_out_buf_size-prev_size });
+    return true;
+}
+
+size_t osd_messenger_t::ssl_do_encrypt_to(osd_client_t *cl, uint8_t *buf, size_t size)
+{
+    if (size < sizeof(msgr_tls_record_hdr_t))
+        return 0;
+    int r = BIO_read(cl->read_from_ssl, buf+sizeof(msgr_tls_record_hdr_t), size-sizeof(msgr_tls_record_hdr_t));
+    if (r > 0)
+    {
+        if (r < size-sizeof(msgr_tls_record_hdr_t))
+            cl->ssl_want_write = false;
+        msgr_tls_record_hdr_t *hdr = (msgr_tls_record_hdr_t*)buf;
+        hdr->encrypted = 1;
+        hdr->size = r;
+        return r+sizeof(msgr_tls_record_hdr_t);
+    }
+    return 0;
+}
+
 bool osd_messenger_t::try_send(osd_client_t *cl)
 {
-    if (!cl->write_op && !cl->write_ops.size() || cl->write_msg.msg_iovlen > 0 || cl->peer_state == PEER_STOPPED || cl->peer_fd < 0)
+    if (!cl->write_op && !cl->write_ops.size() && !cl->ssl_want_write ||
+        cl->write_msg.msg_iovlen > 0 || cl->peer_state == PEER_STOPPED || cl->peer_fd < 0)
     {
         return true;
     }
     assert(cl->peer_state != PEER_RDMA);
-    while ((cl->write_op || cl->write_ops.size()) && cl->send_list.size() < IOV_MAX)
+    if (cl->ssl_cli && !cl->ssl_handshake_done)
     {
-        if (!cl->write_op)
+        bool ok = ssl_do_encrypt(cl);
+        assert(ok && cl->send_list.size() > 0);
+    }
+    else
+    {
+        while ((cl->write_op || cl->write_ops.size()) && cl->send_list.size() < IOV_MAX)
         {
-            cl->write_op = cl->write_ops.front();
-            cl->write_ops.pop_front();
-        }
-        osd_op_t *op = cl->write_op;
-        op_get_write_buffers(cl, cl->send_list);
-        if (!cl->write_op && op->op_type == OSD_OP_IN)
-        {
-            cl->send_free_ops.push_back(op);
+            next_write_op(cl);
+            osd_op_t *op = cl->write_op;
+            op_get_write_buffers(cl, cl->send_list);
+            if (!cl->write_op && op->op_type == OSD_OP_IN)
+            {
+                cl->send_free_ops.push_back(op);
+            }
         }
     }
     if (ringloop && !use_sync_send_recv)
@@ -197,6 +282,21 @@ bool osd_messenger_t::try_send(osd_client_t *cl)
     return true;
 }
 
+void osd_messenger_t::next_write_op(osd_client_t *cl)
+{
+    if (!cl->write_op)
+    {
+        cl->write_op = cl->write_ops.front();
+        cl->write_ops.pop_front();
+        if (cl->proto_csum_status == MSGR_CSUM_FULL || cl->proto_csum_status == MSGR_CSUM_PAYLOAD)
+        {
+            if (!cl->write_csum_state)
+                cl->write_csum_state = XXH3_createState();
+            XXH3_64bits_reset(cl->write_csum_state);
+        }
+    }
+}
+
 void osd_messenger_t::send_replies()
 {
     for (int i = 0; i < write_ready_clients.size(); i++)
@@ -264,14 +364,15 @@ void osd_messenger_t::handle_send(int result, bool prev, bool more, osd_client_t
             else
                 delete op;
         }
+        cl->ssl_out_buf_size = 0;
         if (more)
             cl->zc_free_list.push_back(NULL); // end marker
         cl->send_free_ops.clear();
         cl->write_state = cl->write_op || cl->write_ops.size() ? CL_WRITE_READY : 0;
-        if (cl->proto_csum_status == MSGR_PEER_CSUM_IN && !cl->write_op && !cl->write_ops.size())
+        if ((cl->proto_csum_status & MSGR_CSUM_NEG) && !cl->write_op && !cl->write_ops.size())
         {
             // Checksums negotiated, enable
-            cl->proto_csum_status = MSGR_PEER_CSUM_IN|MSGR_PEER_CSUM_OUT;
+            cl->proto_csum_status = cl->proto_csum_status & (~MSGR_CSUM_NEG);
         }
 #ifdef WITH_RDMA
         if (cl->rdma_conn && !cl->write_op && !cl->write_ops.size() && cl->peer_state == PEER_RDMA_CONNECTING)
@@ -293,36 +394,37 @@ void osd_messenger_t::handle_send(int result, bool prev, bool more, osd_client_t
     }
 }
 
-static inline bool op_write_headers(osd_op_t *op, std::function<bool(uint8_t*, size_t)> op_write_buf)
+static inline bool op_write_headers(osd_op_t *op, std::function<bool(uint8_t*, size_t, bool)> op_write_buf, bool skip_hdr_csum)
 {
-    // Header
-    if (!op_write_buf((op->op_type == OSD_OP_IN ? op->reply.buf : op->req.buf), OSD_PACKET_SIZE))
+    if (!op_write_buf((op->op_type == OSD_OP_IN ? op->reply.buf : op->req.buf), OSD_PACKET_SIZE, skip_hdr_csum))
+    {
         return false;
+    }
     // Bitmap
     if (op->op_type == OSD_OP_IN &&
         op->req.hdr.opcode == OSD_OP_SEC_READ &&
         op->reply.sec_rw.attr_len > 0)
     {
-        if (!op_write_buf((uint8_t*)op->bitmap, op->reply.sec_rw.attr_len))
+        if (!op_write_buf((uint8_t*)op->bitmap, op->reply.sec_rw.attr_len, false))
             return false;
     }
     else if (op->op_type == OSD_OP_OUT &&
         (op->req.hdr.opcode == OSD_OP_SEC_WRITE || op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE) &&
         op->req.sec_rw.attr_len > 0)
     {
-        if (!op_write_buf((uint8_t*)op->bitmap, op->req.sec_rw.attr_len))
+        if (!op_write_buf((uint8_t*)op->bitmap, op->req.sec_rw.attr_len, false))
             return false;
     }
     if (op->req.hdr.opcode == OSD_OP_SEC_READ_BMP)
     {
         if (op->op_type == OSD_OP_IN && op->reply.hdr.retval > 0)
         {
-            if (!op_write_buf((uint8_t*)op->buf, (size_t)op->reply.hdr.retval))
+            if (!op_write_buf((uint8_t*)op->buf, (size_t)op->reply.hdr.retval, false))
                 return false;
         }
         else if (op->op_type == OSD_OP_OUT && op->req.sec_read_bmp.len > 0)
         {
-            if (!op_write_buf((uint8_t*)op->buf, (size_t)op->req.sec_read_bmp.len))
+            if (!op_write_buf((uint8_t*)op->buf, (size_t)op->req.sec_read_bmp.len, false))
                 return false;
         }
     }
@@ -345,20 +447,38 @@ static inline bool op_has_data(osd_op_t *op)
         op->req.hdr.opcode == OSD_OP_SHOW_CONFIG)) && op->iov.count > 0;
 }
 
-size_t osd_messenger_t::op_copy_to(osd_client_t *cl, uint8_t *dst, size_t dst_len)
+static inline bool op_has_data_for_ssl(osd_op_t *op)
 {
-    size_t done = 0;
-    size_t from = cl->write_op_pos;
-    auto op_write_buf = [&](uint8_t *src, size_t src_len, bool skip_csum = false)
+    return (op->op_type == OSD_OP_IN
+        ? (op->req.hdr.opcode == OSD_OP_SEC_LIST ||
+        op->req.hdr.opcode == OSD_OP_SHOW_CONFIG ||
+        op->req.hdr.opcode == OSD_OP_DESCRIBE)
+        : (op->req.hdr.opcode == OSD_OP_SEC_STABILIZE ||
+        op->req.hdr.opcode == OSD_OP_SEC_ROLLBACK ||
+        op->req.hdr.opcode == OSD_OP_SHOW_CONFIG)) && op->iov.count > 0;
+}
+
+static inline bool op_has_data_for_nonssl(osd_op_t *op)
+{
+    return (op->op_type == OSD_OP_IN
+        ? (op->req.hdr.opcode == OSD_OP_READ ||
+        op->req.hdr.opcode == OSD_OP_SEC_READ)
+        : (op->req.hdr.opcode == OSD_OP_WRITE ||
+        op->req.hdr.opcode == OSD_OP_SEC_WRITE ||
+        op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE)) && op->iov.count > 0;
+}
+
+bool osd_messenger_t::ssl_op_write_buf(osd_client_t *cl, uint8_t *src, size_t src_len, bool skip_csum, size_t & from, size_t & done)
+{
+    if (from < src_len)
     {
-        if (from < src_len)
+        size_t n = src_len-from;
+        int ok = SSL_write_ex(cl->ssl_cli, src+from, n, &n);
+        if (ok)
         {
-            size_t n = src_len-from;
-            if (n > dst_len-done)
-                n = dst_len-done;
+            cl->ssl_want_write = true;
             if (cl->write_csum_state && !skip_csum)
                 XXH3_64bits_update(cl->write_csum_state, src+from, n);
-            memcpy(dst+done, src+from, n);
             done += n;
             cl->write_op_pos += n;
             from += n;
@@ -367,39 +487,135 @@ size_t osd_messenger_t::op_copy_to(osd_client_t *cl, uint8_t *dst, size_t dst_le
             from = 0;
         }
         else
-            from -= src_len;
-        return true;
-    };
-    if (cl->proto_csum_status == (MSGR_PEER_CSUM_IN|MSGR_PEER_CSUM_OUT) && !from)
-    {
-        if (!cl->write_csum_state)
-            cl->write_csum_state = XXH3_createState();
-        XXH3_64bits_reset(cl->write_csum_state);
+        {
+            int res = SSL_get_error(cl->ssl_cli, ok);
+            if (res == SSL_ERROR_WANT_WRITE || res == 0)
+                cl->ssl_want_write = true;
+            else if (res == SSL_ERROR_ZERO_RETURN)
+            {
+                fprintf(stderr, "Client %ju TLS disconnected\n", cl->client_id);
+                stop_client(cl->client_id);
+            }
+            else if (res != SSL_ERROR_WANT_READ)
+            {
+                fprintf(stderr, "Client %ju TLS write error: %s. Disconnecting client\n", cl->client_id, ERR_error_string(ERR_get_error(), NULL));
+                stop_client(cl->client_id);
+            }
+            return false;
+        }
     }
-    if (!op_write_headers(cl->write_op, op_write_buf))
+    else
+        from -= src_len;
+    return true;
+}
+
+bool osd_messenger_t::op_write_buf(osd_client_t *cl, uint8_t *src, size_t src_len, uint8_t *dst, size_t dst_len, bool skip_csum, size_t & from, size_t & done)
+{
+    if (from < src_len)
+    {
+        size_t n = src_len-from;
+        if (n > dst_len-done)
+            n = dst_len-done;
+        if (cl->write_csum_state && !skip_csum)
+            XXH3_64bits_update(cl->write_csum_state, src+from, n);
+        memcpy(dst+done, src+from, n);
+        done += n;
+        cl->write_op_pos += n;
+        from += n;
+        if (from < src_len)
+            return false;
+        from = 0;
+    }
+    else
+        from -= src_len;
+    return true;
+}
+
+size_t osd_messenger_t::ssl_op_copy_to(osd_client_t *cl, uint8_t *dst, size_t dst_len)
+{
+    size_t done = 0;
+    size_t from = cl->write_op_pos;
+    // Encrypt headers and data except read/write data
+    auto to_ssl = [&](uint8_t *src, size_t src_len, bool skip_csum)
+    {
+        return ssl_op_write_buf(cl, src, src_len, skip_csum, from, done);
+    };
+    int i = 0;
+    bool full_hdr = false;
+    bool full_op = false;
+    do
+    {
+        if (!full_hdr)
+            full_hdr = op_write_headers(cl->write_op, to_ssl, cl->proto_csum_status != MSGR_CSUM_FULL);
+        if (full_hdr && op_has_data_for_ssl(cl->write_op))
+        {
+            for (; i < cl->write_op->iov.count; i++)
+                if (!ssl_op_write_buf(cl, (uint8_t*)cl->write_op->iov.buf[i].iov_base, cl->write_op->iov.buf[i].iov_len, false, from, done))
+                    break;
+            full_op = true;
+        }
+        if (cl->peer_state == PEER_STOPPED)
+            return 0;
+        if (cl->ssl_want_write)
+        {
+            auto ssl_written = ssl_do_encrypt_to(cl, dst+done, dst_len-done);
+            if (!ssl_written)
+                return done;
+            done += ssl_written;
+        }
+    } while (!full_op);
+    // Non-TLS-encrypted operation data
+    if (op_has_data_for_nonssl(cl->write_op))
+    {
+        if (!op_copy_data_to(cl, dst, dst_len, from, done))
+            return done;
+    }
+    // TLS-encrypted checksum (uh oh...)
+    if (cl->proto_csum_status == MSGR_CSUM_FULL ||
+        cl->proto_csum_status == MSGR_CSUM_PAYLOAD && cl->write_op_pos > OSD_PACKET_SIZE)
+    {
+        if (!from)
+            cl->write_op->csum = XXH3_64bits_digest(cl->write_csum_state);
+        if (!ssl_op_write_buf(cl, (uint8_t*)&cl->write_op->csum, 8, true, from, done))
+            return done;
+        if (cl->ssl_want_write)
+        {
+            auto ssl_written = ssl_do_encrypt_to(cl, dst+done, dst_len-done);
+            if (!ssl_written)
+                return done;
+            done += ssl_written;
+        }
+    }
+    cl->write_op = NULL;
+    cl->write_op_pos = 0;
+    return done;
+}
+
+size_t osd_messenger_t::op_copy_to(osd_client_t *cl, uint8_t *dst, size_t dst_len)
+{
+    if (cl->ssl_cli)
+    {
+        return ssl_op_copy_to(cl, dst, dst_len);
+    }
+    size_t done = 0;
+    size_t from = cl->write_op_pos;
+    auto op_write_buf = [&](uint8_t *src, size_t src_len, bool skip_csum)
+    {
+        return this->op_write_buf(cl, src, src_len, dst, dst_len, skip_csum, from, done);
+    };
+    // Header
+    if (!op_write_headers(cl->write_op, op_write_buf, cl->proto_csum_status != MSGR_CSUM_FULL))
     {
         return done;
     }
     // Operation data
     if (op_has_data(cl->write_op))
     {
-        if (cl->write_op->enc)
-        {
-            if (!op_encrypted_copy_data_to(cl, dst, dst_len, from, done))
-            {
-                return done;
-            }
-        }
-        else
-        {
-            for (int i = 0; i < cl->write_op->iov.count; i++)
-            {
-                if (!op_write_buf((uint8_t*)cl->write_op->iov.buf[i].iov_base, cl->write_op->iov.buf[i].iov_len))
-                    return done;
-            }
-        }
+        if (!op_copy_data_to(cl, dst, dst_len, from, done))
+            return done;
     }
-    if (cl->write_csum_state)
+    if (cl->proto_csum_status == MSGR_CSUM_FULL ||
+        cl->proto_csum_status == MSGR_CSUM_PAYLOAD && cl->write_op_pos > OSD_PACKET_SIZE)
     {
         if (!from)
             cl->write_op->csum = XXH3_64bits_digest(cl->write_csum_state);
@@ -411,10 +627,31 @@ size_t osd_messenger_t::op_copy_to(osd_client_t *cl, uint8_t *dst, size_t dst_le
     return done;
 }
 
+bool osd_messenger_t::op_copy_data_to(osd_client_t *cl, uint8_t *dst, size_t dst_len, size_t & from, size_t & done)
+{
+    if (cl->write_op->enc)
+    {
+        if (!op_encrypted_copy_data_to(cl, dst, dst_len, from, done))
+        {
+            return false;
+        }
+    }
+    else
+    {
+        for (int i = 0; i < cl->write_op->iov.count; i++)
+        {
+            auto & iov = cl->write_op->iov.buf[i];
+            if (!op_write_buf(cl, (uint8_t*)iov.iov_base, iov.iov_len, dst, dst_len, false, from, done))
+                return false;
+        }
+    }
+    return true;
+}
+
 void osd_messenger_t::op_get_write_buffers(osd_client_t *cl, std::vector<iovec> & lst)
 {
     size_t from = cl->write_op_pos;
-    auto op_write_buf = [&](uint8_t *src, size_t src_len, bool skip_csum = false)
+    auto op_write_buf = [&](uint8_t *src, size_t src_len, bool skip_csum)
     {
         if (lst.size() >= IOV_MAX)
             return false;
@@ -430,13 +667,8 @@ void osd_messenger_t::op_get_write_buffers(osd_client_t *cl, std::vector<iovec>
             from -= src_len;
         return true;
     };
-    if (cl->proto_csum_status == (MSGR_PEER_CSUM_IN|MSGR_PEER_CSUM_OUT) && !from)
-    {
-        if (!cl->write_csum_state)
-            cl->write_csum_state = XXH3_createState();
-        XXH3_64bits_reset(cl->write_csum_state);
-    }
-    if (!op_write_headers(cl->write_op, op_write_buf))
+    // Header
+    if (!op_write_headers(cl->write_op, op_write_buf, cl->proto_csum_status != MSGR_CSUM_FULL))
     {
         return;
     }
@@ -462,12 +694,13 @@ void osd_messenger_t::op_get_write_buffers(osd_client_t *cl, std::vector<iovec>
         {
             for (int i = 0; i < cl->write_op->iov.count; i++)
             {
-                if (!op_write_buf((uint8_t*)cl->write_op->iov.buf[i].iov_base, cl->write_op->iov.buf[i].iov_len))
+                if (!op_write_buf((uint8_t*)cl->write_op->iov.buf[i].iov_base, cl->write_op->iov.buf[i].iov_len, false))
                     return;
             }
         }
     }
-    if (cl->write_csum_state)
+    if (cl->proto_csum_status == MSGR_CSUM_FULL ||
+        cl->proto_csum_status == MSGR_CSUM_PAYLOAD && cl->write_op_pos > OSD_PACKET_SIZE)
     {
         if (!from)
             cl->write_op->csum = XXH3_64bits_digest(cl->write_csum_state);
@@ -477,3 +710,89 @@ void osd_messenger_t::op_get_write_buffers(osd_client_t *cl, std::vector<iovec>
     cl->write_op = NULL;
     cl->write_op_pos = 0;
 }
+
+void osd_messenger_t::ssl_op_get_write_buffers(osd_client_t *cl, std::vector<iovec> & lst)
+{
+    size_t done = 0;
+    size_t from = cl->write_op_pos;
+    auto op_write_buf = [&](uint8_t *src, size_t src_len, bool skip_csum)
+    {
+        if (lst.size() >= IOV_MAX)
+            return false;
+        if (from < src_len)
+        {
+            if (cl->write_csum_state && !skip_csum)
+                XXH3_64bits_update(cl->write_csum_state, src+from, src_len-from);
+            lst.push_back((iovec){ .iov_base = src+from, .iov_len = src_len-from });
+            cl->write_op_pos += src_len-from;
+            from = 0;
+        }
+        else
+            from -= src_len;
+        return true;
+    };
+    // Encrypt headers and data except read/write data
+    auto to_ssl = [&](uint8_t *src, size_t src_len, bool skip_csum)
+    {
+        return ssl_op_write_buf(cl, src, src_len, skip_csum, from, done);
+    };
+    int i = 0;
+    bool full_hdr = false;
+    bool full_op = false;
+    do
+    {
+        if (!full_hdr)
+            full_hdr = op_write_headers(cl->write_op, to_ssl, cl->proto_csum_status != MSGR_CSUM_FULL);
+        if (full_hdr && op_has_data_for_ssl(cl->write_op))
+        {
+            for (; i < cl->write_op->iov.count; i++)
+                if (!ssl_op_write_buf(cl, (uint8_t*)cl->write_op->iov.buf[i].iov_base, cl->write_op->iov.buf[i].iov_len, false, from, done))
+                    break;
+            full_op = true;
+        }
+        if (cl->peer_state == PEER_STOPPED)
+            return;
+        if (!ssl_do_encrypt(cl))
+            return;
+    } while (!full_op);
+    // Non-TLS-encrypted operation data
+    if (op_has_data_for_nonssl(cl->write_op))
+    {
+        if (cl->write_op->enc)
+        {
+            if (lst.size() >= IOV_MAX)
+                return;
+            // No way except to allocate a temporary buffer and encrypt data to it
+            assert(cl->write_op->req.hdr.opcode == OSD_OP_WRITE);
+            size_t remsize = cl->write_op->req.rw.len - from + (from % 16);
+            assert(remsize > 0);
+            assert(!cl->write_op->enc_buf);
+            cl->write_op->enc_buf = (uint8_t*)malloc_or_die(remsize);
+            size_t done = 0;
+            bool end = op_encrypted_copy_data_to(cl, cl->write_op->enc_buf, remsize, from, done);
+            assert(end);
+            lst.push_back((iovec){ .iov_base = cl->write_op->enc_buf, .iov_len = remsize });
+        }
+        else
+        {
+            for (int i = 0; i < cl->write_op->iov.count; i++)
+            {
+                if (!op_write_buf((uint8_t*)cl->write_op->iov.buf[i].iov_base, cl->write_op->iov.buf[i].iov_len, false))
+                    return;
+            }
+        }
+    }
+    // TLS-encrypted checksum (uh oh...)
+    if (cl->proto_csum_status == MSGR_CSUM_FULL ||
+        cl->proto_csum_status == MSGR_CSUM_PAYLOAD && cl->write_op_pos > OSD_PACKET_SIZE)
+    {
+        if (!from)
+            cl->write_op->csum = XXH3_64bits_digest(cl->write_csum_state);
+        if (!ssl_op_write_buf(cl, (uint8_t*)&cl->write_op->csum, 8, true, from, done))
+            return;
+        if (!ssl_do_encrypt(cl))
+            return;
+    }
+    cl->write_op = NULL;
+    cl->write_op_pos = 0;
+}

src/client/msgr_stop.cpp

@@ -9,6 +9,12 @@
 #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()
 {
@@ -239,4 +245,18 @@ osd_client_t::~osd_client_t()
         XXH3_freeState(write_csum_state);
         write_csum_state = NULL;
     }
+#ifdef WITH_OPENSSL
+    if (ssl_cli)
+    {
+        SSL_free(ssl_cli);
+        ssl_cli = NULL;
+        write_to_ssl = NULL;
+        read_from_ssl = NULL;
+    }
+    if (ssl_out_buf)
+    {
+        free(ssl_out_buf);
+        ssl_out_buf = NULL;
+    }
+#endif
 }

src/client/vitastor.pc.in

@@ -6,7 +6,7 @@ includedir=${prefix}/@CMAKE_INSTALL_INCLUDEDIR@
 
 Name: Vitastor
 Description: Vitastor client library
-Version: 3.0.8
+Version: 3.0.9
 Libs: -L${libdir} -lvitastor_client
 Cflags: -I${includedir}
 

src/osd/osd.cpp

@@ -8,7 +8,6 @@
 #include <netinet/tcp.h>
 #include <arpa/inet.h>
 
-#include "xxhash.h"
 #include "addr_util.h"
 #include "osd_primary.h"
 #include "osd.h"
@@ -16,46 +15,8 @@
 #include "str_util.h"
 #include "json_util.h"
 
-void xxhash_run_test(XXH3_state_t *st, void *buf, size_t size)
-{
-    timespec tv, tv1;
-    clock_gettime(CLOCK_REALTIME, &tv);
-    uint64_t res = 1;
-    uint64_t done = 0;
-    while (true)
-    {
-        XXH3_64bits_reset(st);
-        XXH3_64bits_update(st, buf, size);
-        res *= XXH3_64bits_digest(st);
-        done++;
-        if (!(done % 100000))
-        {
-            clock_gettime(CLOCK_REALTIME, &tv1);
-            if ((tv1.tv_sec-tv.tv_sec)*1000 + (tv1.tv_nsec-tv.tv_nsec)/1000000 > 1000)
-                break;
-        }
-    }
-    double sec = (tv1.tv_sec-tv.tv_sec) + (tv1.tv_nsec-tv.tv_nsec)/1000000000.0;
-    fprintf(stderr, "xxhash3: %lu %zu byte blocks in %.2f sec = %.1f MB/s\n",
-        done, size, sec, done*size/sec/1024/1024);
-}
-
-void xxhash_bench()
-{
-    std::vector<uint8_t> buf;
-    buf.resize(128*1024);
-    for (size_t i = 0; i < buf.size(); i++)
-        buf[i] = ((i+37)*3) % 255;
-    XXH3_state_t *st = XXH3_createState();
-    xxhash_run_test(st, buf.data(), 4096);
-    xxhash_run_test(st, buf.data(), buf.size());
-    XXH3_freeState(st);
-}
-
 osd_t::osd_t(const json11::Json & config, ring_loop_t *ringloop)
 {
-    xxhash_bench();
-
     zero_buffer_size = 1<<20;
     zero_buffer = malloc_or_die(zero_buffer_size);
     memset(zero_buffer, 0, zero_buffer_size);

src/osd/osd_secondary.cpp

@@ -348,10 +348,17 @@ 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 } };
-    if (req_json["features"]["proto_checksums"].bool_value() && msgr.use_proto_checksums)
+    if (msgr.use_proto_checksums)
     {
-        cl->proto_csum_status = MSGR_PEER_CSUM_IN;
-        features["proto_checksums"] = true;
+        auto peer_csums = req_json["features"]["proto_checksums"].uint64_value();
+        if (peer_csums == MSGR_CSUM_FULL || peer_csums == MSGR_CSUM_PAYLOAD)
+        {
+            if (msgr.use_proto_checksums == MSGR_CSUM_FULL && peer_csums == MSGR_CSUM_FULL)
+                cl->proto_csum_status = MSGR_CSUM_FULL|MSGR_CSUM_NEG;
+            else
+                cl->proto_csum_status = MSGR_CSUM_PAYLOAD|MSGR_CSUM_NEG;
+            features["proto_checksums"] = msgr.use_proto_checksums;
+        }
     }
     // Expose sensitive configuration values so peers can check them
     json11::Json::object wire_config = json11::Json::object {

src/test/CMakeLists.txt

@@ -41,7 +41,7 @@ add_executable(test_heap
 	../blockstore/multilist.cpp
 	../blockstore/blockstore_heap.cpp
 	../util/crc32c.c
-	../util/xxhash.c
+	../util/xxh_x86dispatch.c
 	../util/allocator.cpp
 	../blockstore/blockstore_disk.cpp
 	../util/str_util.cpp
@@ -62,6 +62,14 @@ 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

@@ -0,0 +1,61 @@
+#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;
+}