WIP Separate data connections

Prelimilary results:
- CPU usage drops significantly. For example, in T1Q8 128K write test against
  stub_uring_osd with 10G network and Athlon X4 860k CPU it drops from 100% to 30%
- Latency becomes slightly worse. In T1Q1 4K write test in the same environment
  latency increases from 56 to 63 us.
- Small write throughput also becomes slightly worse. In T1Q128 4K write test
  against stub iops decreases from 138k to ~110k (unstable, fluctuates 100k..120k).
  Note that this is without io_uring, of course.

CMakeLists.txt

@@ -2,6 +2,6 @@ cmake_minimum_required(VERSION 2.8)
 
 project(vitastor)
 
-set(VERSION "0.6.10")
+set(VERSION "0.6.11")
 
 add_subdirectory(src)

csi/Makefile

@@ -1,4 +1,4 @@
-VERSION ?= v0.6.10
+VERSION ?= v0.6.11
 
 all: build push
 

csi/deploy/004-csi-nodeplugin.yaml

@@ -49,7 +49,7 @@ spec:
             capabilities:
               add: ["SYS_ADMIN"]
             allowPrivilegeEscalation: true
-          image: vitalif/vitastor-csi:v0.6.10
+          image: vitalif/vitastor-csi:v0.6.11
           args:
             - "--node=$(NODE_ID)"
             - "--endpoint=$(CSI_ENDPOINT)"

csi/deploy/007-csi-provisioner.yaml

@@ -116,7 +116,7 @@ spec:
             privileged: true
             capabilities:
               add: ["SYS_ADMIN"]
-          image: vitalif/vitastor-csi:v0.6.10
+          image: vitalif/vitastor-csi:v0.6.11
           args:
             - "--node=$(NODE_ID)"
             - "--endpoint=$(CSI_ENDPOINT)"

csi/src/config.go

@@ -5,7 +5,7 @@ package vitastor
 
 const (
     vitastorCSIDriverName    = "csi.vitastor.io"
-    vitastorCSIDriverVersion = "0.6.10"
+    vitastorCSIDriverVersion = "0.6.11"
 )
 
 // Config struct fills the parameters of request or user input

debian/changelog

@@ -1,4 +1,4 @@
-vitastor (0.6.10-1) unstable; urgency=medium
+vitastor (0.6.11-1) unstable; urgency=medium
 
   * RDMA support
   * Bugfixes

debian/vitastor.Dockerfile

@@ -33,8 +33,8 @@ RUN set -e -x; \
     mkdir -p /root/packages/vitastor-$REL; \
     rm -rf /root/packages/vitastor-$REL/*; \
     cd /root/packages/vitastor-$REL; \
-    cp -r /root/vitastor vitastor-0.6.10; \
+    cp -r /root/vitastor vitastor-0.6.11; \
-    cd vitastor-0.6.10; \
+    cd vitastor-0.6.11; \
     ln -s /root/fio-build/fio-*/ ./fio; \
     FIO=$(head -n1 fio/debian/changelog | perl -pe 's/^.*\((.*?)\).*$/$1/'); \
     ls /usr/include/linux/raw.h || cp ./debian/raw.h /usr/include/linux/raw.h; \
@@ -47,8 +47,8 @@ RUN set -e -x; \
     rm -rf a b; \
     echo "dep:fio=$FIO" > debian/fio_version; \
     cd /root/packages/vitastor-$REL; \
-    tar --sort=name --mtime='2020-01-01' --owner=0 --group=0 --exclude=debian -cJf vitastor_0.6.10.orig.tar.xz vitastor-0.6.10; \
+    tar --sort=name --mtime='2020-01-01' --owner=0 --group=0 --exclude=debian -cJf vitastor_0.6.11.orig.tar.xz vitastor-0.6.11; \
-    cd vitastor-0.6.10; \
+    cd vitastor-0.6.11; \
     V=$(head -n1 debian/changelog | perl -pe 's/^.*\((.*?)\).*$/$1/'); \
     DEBFULLNAME="Vitaliy Filippov <vitalif@yourcmc.ru>" dch -D $REL -v "$V""$REL" "Rebuild for $REL"; \
     DEB_BUILD_OPTIONS=nocheck dpkg-buildpackage --jobs=auto -sa; \

patches/cinder-vitastor.py

@@ -50,7 +50,7 @@ from cinder.volume import configuration
 from cinder.volume import driver
 from cinder.volume import volume_utils
 
-VERSION = '0.6.10'
+VERSION = '0.6.11'
 
 LOG = logging.getLogger(__name__)
 

rpm/build-tarball.sh

@@ -25,4 +25,4 @@ rm fio
 mv fio-copy fio
 FIO=`rpm -qi fio | perl -e 'while(<>) { /^Epoch[\s:]+(\S+)/ && print "$1:"; /^Version[\s:]+(\S+)/ && print $1; /^Release[\s:]+(\S+)/ && print "-$1"; }'`
 perl -i -pe 's/(Requires:\s*fio)([^\n]+)?/$1 = '$FIO'/' $VITASTOR/rpm/vitastor-el$EL.spec
-tar --transform 's#^#vitastor-0.6.10/#' --exclude 'rpm/*.rpm' -czf $VITASTOR/../vitastor-0.6.10$(rpm --eval '%dist').tar.gz *
+tar --transform 's#^#vitastor-0.6.11/#' --exclude 'rpm/*.rpm' -czf $VITASTOR/../vitastor-0.6.11$(rpm --eval '%dist').tar.gz *

rpm/vitastor-el7.Dockerfile

@@ -34,7 +34,7 @@ ADD . /root/vitastor
 RUN set -e; \
     cd /root/vitastor/rpm; \
     sh build-tarball.sh; \
-    cp /root/vitastor-0.6.10.el7.tar.gz ~/rpmbuild/SOURCES; \
+    cp /root/vitastor-0.6.11.el7.tar.gz ~/rpmbuild/SOURCES; \
     cp vitastor-el7.spec ~/rpmbuild/SPECS/vitastor.spec; \
     cd ~/rpmbuild/SPECS/; \
     rpmbuild -ba vitastor.spec; \

rpm/vitastor-el7.spec

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        0.6.10
+Version:        0.6.11
 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-0.6.10.el7.tar.gz
+Source0:        vitastor-0.6.11.el7.tar.gz
 
 BuildRequires:  liburing-devel >= 0.6
 BuildRequires:  gperftools-devel

rpm/vitastor-el8.Dockerfile

@@ -33,7 +33,7 @@ ADD . /root/vitastor
 RUN set -e; \
     cd /root/vitastor/rpm; \
     sh build-tarball.sh; \
-    cp /root/vitastor-0.6.10.el8.tar.gz ~/rpmbuild/SOURCES; \
+    cp /root/vitastor-0.6.11.el8.tar.gz ~/rpmbuild/SOURCES; \
     cp vitastor-el8.spec ~/rpmbuild/SPECS/vitastor.spec; \
     cd ~/rpmbuild/SPECS/; \
     rpmbuild -ba vitastor.spec; \

rpm/vitastor-el8.spec

@@ -1,11 +1,11 @@
 Name:           vitastor
-Version:        0.6.10
+Version:        0.6.11
 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-0.6.10.el8.tar.gz
+Source0:        vitastor-0.6.11.el8.tar.gz
 
 BuildRequires:  liburing-devel >= 0.6
 BuildRequires:  gperftools-devel

src/CMakeLists.txt

@@ -15,7 +15,7 @@ if("${CMAKE_INSTALL_PREFIX}" MATCHES "^/usr/local/?$")
 	set(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_PREFIX}/${CMAKE_INSTALL_LIBDIR}")
 endif()
 
-add_definitions(-DVERSION="0.6.10")
+add_definitions(-DVERSION="0.6.11")
 add_definitions(-Wall -Wno-sign-compare -Wno-comment -Wno-parentheses -Wno-pointer-arith -fdiagnostics-color=always -I ${CMAKE_SOURCE_DIR}/src)
 if (${WITH_ASAN})
 	add_definitions(-fsanitize=address -fno-omit-frame-pointer)
@@ -88,8 +88,8 @@ if (IBVERBS_LIBRARIES)
 	set(MSGR_RDMA "msgr_rdma.cpp")
 endif (IBVERBS_LIBRARIES)
 add_library(vitastor_common STATIC
-	epoll_manager.cpp etcd_state_client.cpp
+	epoll_manager.cpp etcd_state_client.cpp messenger.cpp addr_util.cpp
-	messenger.cpp msgr_stop.cpp msgr_op.cpp msgr_send.cpp msgr_receive.cpp ringloop.cpp ../json11/json11.cpp
+	msgr_stop.cpp msgr_op.cpp msgr_send.cpp msgr_receive.cpp ringloop.cpp ../json11/json11.cpp
 	http_client.cpp osd_ops.cpp pg_states.cpp timerfd_manager.cpp base64.cpp ${MSGR_RDMA}
 )
 target_compile_options(vitastor_common PUBLIC -fPIC)
@@ -112,6 +112,7 @@ if (${WITH_FIO})
 	add_library(fio_vitastor_sec SHARED
 		fio_sec_osd.cpp
 		rw_blocking.cpp
+		addr_util.cpp
 	)
 	target_link_libraries(fio_vitastor_sec
 		tcmalloc_minimal
@@ -189,11 +190,11 @@ endif (${WITH_QEMU})
 ### Test stubs
 
 # stub_osd, stub_bench, osd_test
-add_executable(stub_osd stub_osd.cpp rw_blocking.cpp)
+add_executable(stub_osd stub_osd.cpp rw_blocking.cpp addr_util.cpp)
 target_link_libraries(stub_osd tcmalloc_minimal)
-add_executable(stub_bench stub_bench.cpp rw_blocking.cpp)
+add_executable(stub_bench stub_bench.cpp rw_blocking.cpp addr_util.cpp)
 target_link_libraries(stub_bench tcmalloc_minimal)
-add_executable(osd_test osd_test.cpp rw_blocking.cpp)
+add_executable(osd_test osd_test.cpp rw_blocking.cpp addr_util.cpp)
 target_link_libraries(osd_test tcmalloc_minimal)
 
 # osd_rmw_test

src/addr_util.cpp

@@ -0,0 +1,60 @@
+#include <arpa/inet.h>
+#include <string.h>
+#include <stdio.h>
+
+#include <stdexcept>
+
+#include "addr_util.h"
+
+bool string_to_addr(std::string str, bool parse_port, int default_port, struct sockaddr *addr)
+{
+    if (parse_port)
+    {
+        int p = str.rfind(':');
+        if (p != std::string::npos && !(str.length() > 0 && str[p-1] == ']')) // "[ipv6]" which contains ':'
+        {
+            char null_byte = 0;
+            int n = sscanf(str.c_str()+p+1, "%d%c", &default_port, &null_byte);
+            if (n != 1 || default_port >= 0x10000)
+                return false;
+            str = str.substr(0, p);
+        }
+    }
+    if (inet_pton(AF_INET, str.c_str(), &((struct sockaddr_in*)addr)->sin_addr) == 1)
+    {
+        addr->sa_family = AF_INET;
+        ((struct sockaddr_in*)addr)->sin_port = htons(default_port);
+        return true;
+    }
+    if (str.length() >= 2 && str[0] == '[' && str[str.length()-1] == ']')
+        str = str.substr(1, str.length()-2);
+    if (inet_pton(AF_INET6, str.c_str(), &((struct sockaddr_in6*)addr)->sin6_addr) == 1)
+    {
+        addr->sa_family = AF_INET6;
+        ((struct sockaddr_in6*)addr)->sin6_port = htons(default_port);
+        return true;
+    }
+    return false;
+}
+
+std::string addr_to_string(const sockaddr &addr)
+{
+    char peer_str[256];
+    bool ok = false;
+    int port;
+    if (addr.sa_family == AF_INET)
+    {
+        ok = !!inet_ntop(AF_INET, &((sockaddr_in*)&addr)->sin_addr, peer_str, 256);
+        port = ntohs(((sockaddr_in*)&addr)->sin_port);
+    }
+    else if (addr.sa_family == AF_INET6)
+    {
+        ok = !!inet_ntop(AF_INET6, &((sockaddr_in6*)&addr)->sin6_addr, peer_str, 256);
+        port = ntohs(((sockaddr_in6*)&addr)->sin6_port);
+    }
+    else
+        throw std::runtime_error("Unknown address family "+std::to_string(addr.sa_family));
+    if (!ok)
+        throw std::runtime_error(std::string("inet_ntop: ") + strerror(errno));
+    return std::string(peer_str)+":"+std::to_string(port);
+}

src/addr_util.h

@@ -0,0 +1,7 @@
+#pragma once
+
+#include <sys/socket.h>
+#include <string>
+
+bool string_to_addr(std::string str, bool parse_port, int default_port, struct sockaddr *addr);
+std::string addr_to_string(const sockaddr &addr);

src/blockstore_impl.cpp

@@ -235,6 +235,12 @@ void blockstore_impl_t::loop()
         {
             throw std::runtime_error(std::string("io_uring_submit: ") + strerror(-ret));
         }
+        for (auto s: journal.submitting_sectors)
+        {
+            // Mark journal sector writes as submitted
+            journal.sector_info[s].submit_id = 0;
+        }
+        journal.submitting_sectors.clear();
         if ((initial_ring_space - ringloop->space_left()) > 0)
         {
             live = true;

src/blockstore_impl.h

@@ -54,6 +54,14 @@
 #define IS_BIG_WRITE(st) (((st) & 0x0F) == BS_ST_BIG_WRITE)
 #define IS_DELETE(st) (((st) & 0x0F) == BS_ST_DELETE)
 
+#define BS_SUBMIT_CHECK_SQES(n) \
+    if (ringloop->space_left() < (n))\
+    {\
+        /* Pause until there are more requests available */\
+        PRIV(op)->wait_for = WAIT_SQE;\
+        return 0;\
+    }
+
 #define BS_SUBMIT_GET_SQE(sqe, data) \
     BS_SUBMIT_GET_ONLY_SQE(sqe); \
     struct ring_data_t *data = ((ring_data_t*)sqe->user_data)
@@ -170,7 +178,7 @@ struct blockstore_op_private_t
     std::vector<fulfill_read_t> read_vec;
 
     // Sync, write
-    uint64_t min_flushed_journal_sector, max_flushed_journal_sector;
+    int min_flushed_journal_sector, max_flushed_journal_sector;
 
     // Write
     struct iovec iov_zerofill[3];
@@ -283,6 +291,10 @@ class blockstore_impl_t
     void open_journal();
     uint8_t* get_clean_entry_bitmap(uint64_t block_loc, int offset);
 
+    // Journaling
+    void prepare_journal_sector_write(int sector, blockstore_op_t *op);
+    void handle_journal_write(ring_data_t *data, uint64_t flush_id);
+
     // Asynchronous init
     int initialized;
     int metadata_buf_size;
@@ -310,21 +322,18 @@ class blockstore_impl_t
 
     // Sync
     int continue_sync(blockstore_op_t *op, bool queue_has_in_progress_sync);
-    void handle_sync_event(ring_data_t *data, blockstore_op_t *op);
     void ack_sync(blockstore_op_t *op);
 
     // Stabilize
     int dequeue_stable(blockstore_op_t *op);
     int continue_stable(blockstore_op_t *op);
     void mark_stable(const obj_ver_id & ov, bool forget_dirty = false);
-    void handle_stable_event(ring_data_t *data, blockstore_op_t *op);
     void stabilize_object(object_id oid, uint64_t max_ver);
 
     // Rollback
     int dequeue_rollback(blockstore_op_t *op);
     int continue_rollback(blockstore_op_t *op);
     void mark_rolled_back(const obj_ver_id & ov);
-    void handle_rollback_event(ring_data_t *data, blockstore_op_t *op);
     void erase_dirty(blockstore_dirty_db_t::iterator dirty_start, blockstore_dirty_db_t::iterator dirty_end, uint64_t clean_loc);
 
     // List

src/blockstore_journal.cpp

@@ -153,22 +153,73 @@ journal_entry* prefill_single_journal_entry(journal_t & journal, uint16_t type,
     return je;
 }
 
-void prepare_journal_sector_write(journal_t & journal, int cur_sector, io_uring_sqe *sqe, std::function<void(ring_data_t*)> cb)
+void blockstore_impl_t::prepare_journal_sector_write(int cur_sector, blockstore_op_t *op)
 {
-    journal.sector_info[cur_sector].dirty = false;
+    // Don't submit the same sector twice in the same batch
-    journal.sector_info[cur_sector].written = true;
+    if (!journal.sector_info[cur_sector].submit_id)
-    journal.sector_info[cur_sector].flush_count++;
+    {
+        io_uring_sqe *sqe = get_sqe();
+        // Caller must ensure availability of an SQE
+        assert(sqe != NULL);
         ring_data_t *data = ((ring_data_t*)sqe->user_data);
+        journal.sector_info[cur_sector].written = true;
+        journal.sector_info[cur_sector].submit_id = ++journal.submit_id;
+        journal.submitting_sectors.push_back(cur_sector);
+        journal.sector_info[cur_sector].flush_count++;
         data->iov = (struct iovec){
             (journal.inmemory
                 ? journal.buffer + journal.sector_info[cur_sector].offset
                 : journal.sector_buf + journal.block_size*cur_sector),
             journal.block_size
         };
-    data->callback = cb;
+        data->callback = [this, flush_id = journal.submit_id](ring_data_t *data) { handle_journal_write(data, flush_id); };
         my_uring_prep_writev(
             sqe, journal.fd, &data->iov, 1, journal.offset + journal.sector_info[cur_sector].offset
         );
+    }
+    journal.sector_info[cur_sector].dirty = false;
+    // But always remember that this operation has to wait until this exact journal write is finished
+    journal.flushing_ops.insert((pending_journaling_t){
+        .flush_id = journal.sector_info[cur_sector].submit_id,
+        .sector = cur_sector,
+        .op = op,
+    });
+    auto priv = PRIV(op);
+    priv->pending_ops++;
+    if (!priv->min_flushed_journal_sector)
+        priv->min_flushed_journal_sector = 1+cur_sector;
+    priv->max_flushed_journal_sector = 1+cur_sector;
+}
+
+void blockstore_impl_t::handle_journal_write(ring_data_t *data, uint64_t flush_id)
+{
+    live = true;
+    if (data->res != data->iov.iov_len)
+    {
+        // FIXME: our state becomes corrupted after a write error. maybe do something better than just die
+        throw std::runtime_error(
+            "journal write failed ("+std::to_string(data->res)+" != "+std::to_string(data->iov.iov_len)+
+            "). in-memory state is corrupted. AAAAAAAaaaaaaaaa!!!111"
+        );
+    }
+    auto fl_it = journal.flushing_ops.upper_bound((pending_journaling_t){ .flush_id = flush_id });
+    if (fl_it != journal.flushing_ops.end() && fl_it->flush_id == flush_id)
+    {
+        journal.sector_info[fl_it->sector].flush_count--;
+    }
+    while (fl_it != journal.flushing_ops.end() && fl_it->flush_id == flush_id)
+    {
+        auto priv = PRIV(fl_it->op);
+        priv->pending_ops--;
+        assert(priv->pending_ops >= 0);
+        if (priv->pending_ops == 0)
+        {
+            release_journal_sectors(fl_it->op);
+            priv->op_state++;
+            ringloop->wakeup();
+        }
+        journal.flushing_ops.erase(fl_it++);
+    }
 }
 
 journal_t::~journal_t()

src/blockstore_journal.h

@@ -4,6 +4,7 @@
 #pragma once
 
 #include "crc32c.h"
+#include <set>
 
 #define MIN_JOURNAL_SIZE 4*1024*1024
 #define JOURNAL_MAGIC 0x4A33
@@ -145,8 +146,21 @@ struct journal_sector_info_t
     uint64_t flush_count;
     bool written;
     bool dirty;
+    uint64_t submit_id;
 };
 
+struct pending_journaling_t
+{
+    uint64_t flush_id;
+    int sector;
+    blockstore_op_t *op;
+};
+
+inline bool operator < (const pending_journaling_t & a, const pending_journaling_t & b)
+{
+    return a.flush_id < b.flush_id || a.flush_id == b.flush_id && a.op < b.op;
+}
+
 struct journal_t
 {
     int fd;
@@ -172,6 +186,9 @@ struct journal_t
     bool no_same_sector_overwrites = false;
     int cur_sector = 0;
     int in_sector_pos = 0;
+    std::vector<int> submitting_sectors;
+    std::set<pending_journaling_t> flushing_ops;
+    uint64_t submit_id = 0;
 
     // Used sector map
     // May use ~ 80 MB per 1 GB of used journal space in the worst case
@@ -200,5 +217,3 @@ struct blockstore_journal_check_t
 };
 
 journal_entry* prefill_single_journal_entry(journal_t & journal, uint16_t type, uint32_t size);
-
-void prepare_journal_sector_write(journal_t & journal, int sector, io_uring_sqe *sqe, std::function<void(ring_data_t*)> cb);

src/blockstore_open.cpp

@@ -306,6 +306,10 @@ static void check_size(int fd, uint64_t *size, uint64_t *sectsize, std::string n
     if (S_ISREG(st.st_mode))
     {
         *size = st.st_size;
+        if (sectsize)
+        {
+            *sectsize = st.st_blksize;
+        }
     }
     else if (S_ISBLK(st.st_mode))
     {

src/blockstore_rollback.cpp

@@ -74,24 +74,17 @@ skip_ov:
     {
         return 0;
     }
-    // There is sufficient space. Get SQEs
+    // There is sufficient space. Check SQEs
-    struct io_uring_sqe *sqe[space_check.sectors_to_write];
+    BS_SUBMIT_CHECK_SQES(space_check.sectors_to_write);
-    for (i = 0; i < space_check.sectors_to_write; i++)
-    {
-        BS_SUBMIT_GET_SQE_DECL(sqe[i]);
-    }
     // Prepare and submit journal entries
-    auto cb = [this, op](ring_data_t *data) { handle_rollback_event(data, op); };
+    int s = 0;
-    int s = 0, cur_sector = -1;
     for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
     {
         if (!journal.entry_fits(sizeof(journal_entry_rollback)) &&
             journal.sector_info[journal.cur_sector].dirty)
         {
-            if (cur_sector == -1)
+            prepare_journal_sector_write(journal.cur_sector, op);
-                PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
+            s++;
-            prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], cb);
-            cur_sector = journal.cur_sector;
         }
         journal_entry_rollback *je = (journal_entry_rollback*)
             prefill_single_journal_entry(journal, JE_ROLLBACK, sizeof(journal_entry_rollback));
@@ -100,12 +93,9 @@ skip_ov:
         je->crc32 = je_crc32((journal_entry*)je);
         journal.crc32_last = je->crc32;
     }
-    prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], cb);
+    prepare_journal_sector_write(journal.cur_sector, op);
+    s++;
     assert(s == space_check.sectors_to_write);
-    if (cur_sector == -1)
-        PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
-    PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
-    PRIV(op)->pending_ops = s;
     PRIV(op)->op_state = 1;
     return 1;
 }
@@ -114,30 +104,23 @@ int blockstore_impl_t::continue_rollback(blockstore_op_t *op)
 {
     if (PRIV(op)->op_state == 2)
         goto resume_2;
-    else if (PRIV(op)->op_state == 3)
+    else if (PRIV(op)->op_state == 4)
-        goto resume_3;
+        goto resume_4;
-    else if (PRIV(op)->op_state == 5)
-        goto resume_5;
     else
         return 1;
 resume_2:
-    // Release used journal sectors
-    release_journal_sectors(op);
-resume_3:
     if (!disable_journal_fsync)
     {
-        io_uring_sqe *sqe;
+        BS_SUBMIT_GET_SQE(sqe, data);
-        BS_SUBMIT_GET_SQE_DECL(sqe);
-        ring_data_t *data = ((ring_data_t*)sqe->user_data);
         my_uring_prep_fsync(sqe, journal.fd, IORING_FSYNC_DATASYNC);
         data->iov = { 0 };
-        data->callback = [this, op](ring_data_t *data) { handle_rollback_event(data, op); };
+        data->callback = [this, op](ring_data_t *data) { handle_write_event(data, op); };
         PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 0;
         PRIV(op)->pending_ops = 1;
-        PRIV(op)->op_state = 4;
+        PRIV(op)->op_state = 3;
         return 1;
     }
-resume_5:
+resume_4:
     obj_ver_id* v;
     int i;
     for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
@@ -196,24 +179,6 @@ void blockstore_impl_t::mark_rolled_back(const obj_ver_id & ov)
     }
 }
 
-void blockstore_impl_t::handle_rollback_event(ring_data_t *data, blockstore_op_t *op)
-{
-    live = true;
-    if (data->res != data->iov.iov_len)
-    {
-        throw std::runtime_error(
-            "write operation failed ("+std::to_string(data->res)+" != "+std::to_string(data->iov.iov_len)+
-            "). in-memory state is corrupted. AAAAAAAaaaaaaaaa!!!111"
-        );
-    }
-    PRIV(op)->pending_ops--;
-    if (PRIV(op)->pending_ops == 0)
-    {
-        PRIV(op)->op_state++;
-        ringloop->wakeup();
-    }
-}
-
 void blockstore_impl_t::erase_dirty(blockstore_dirty_db_t::iterator dirty_start, blockstore_dirty_db_t::iterator dirty_end, uint64_t clean_loc)
 {
     if (dirty_end == dirty_start)

src/blockstore_stable.cpp

@@ -97,25 +97,18 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
     {
         return 0;
     }
-    // There is sufficient space. Get SQEs
+    // There is sufficient space. Check SQEs
-    struct io_uring_sqe *sqe[space_check.sectors_to_write];
+    BS_SUBMIT_CHECK_SQES(space_check.sectors_to_write);
-    for (i = 0; i < space_check.sectors_to_write; i++)
-    {
-        BS_SUBMIT_GET_SQE_DECL(sqe[i]);
-    }
     // Prepare and submit journal entries
-    auto cb = [this, op](ring_data_t *data) { handle_stable_event(data, op); };
+    int s = 0;
-    int s = 0, cur_sector = -1;
     for (i = 0, v = (obj_ver_id*)op->buf; i < op->len; i++, v++)
     {
         // FIXME: Only stabilize versions that aren't stable yet
         if (!journal.entry_fits(sizeof(journal_entry_stable)) &&
             journal.sector_info[journal.cur_sector].dirty)
         {
-            if (cur_sector == -1)
+            prepare_journal_sector_write(journal.cur_sector, op);
-                PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
+            s++;
-            prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], cb);
-            cur_sector = journal.cur_sector;
         }
         journal_entry_stable *je = (journal_entry_stable*)
             prefill_single_journal_entry(journal, JE_STABLE, sizeof(journal_entry_stable));
@@ -124,12 +117,9 @@ int blockstore_impl_t::dequeue_stable(blockstore_op_t *op)
         je->crc32 = je_crc32((journal_entry*)je);
         journal.crc32_last = je->crc32;
     }
-    prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], cb);
+    prepare_journal_sector_write(journal.cur_sector, op);
+    s++;
     assert(s == space_check.sectors_to_write);
-    if (cur_sector == -1)
-        PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
-    PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
-    PRIV(op)->pending_ops = s;
     PRIV(op)->op_state = 1;
     return 1;
 }
@@ -138,30 +128,23 @@ int blockstore_impl_t::continue_stable(blockstore_op_t *op)
 {
     if (PRIV(op)->op_state == 2)
         goto resume_2;
-    else if (PRIV(op)->op_state == 3)
+    else if (PRIV(op)->op_state == 4)
-        goto resume_3;
+        goto resume_4;
-    else if (PRIV(op)->op_state == 5)
-        goto resume_5;
     else
         return 1;
 resume_2:
-    // Release used journal sectors
-    release_journal_sectors(op);
-resume_3:
     if (!disable_journal_fsync)
     {
-        io_uring_sqe *sqe;
+        BS_SUBMIT_GET_SQE(sqe, data);
-        BS_SUBMIT_GET_SQE_DECL(sqe);
-        ring_data_t *data = ((ring_data_t*)sqe->user_data);
         my_uring_prep_fsync(sqe, journal.fd, IORING_FSYNC_DATASYNC);
         data->iov = { 0 };
-        data->callback = [this, op](ring_data_t *data) { handle_stable_event(data, op); };
+        data->callback = [this, op](ring_data_t *data) { handle_write_event(data, op); };
         PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 0;
         PRIV(op)->pending_ops = 1;
-        PRIV(op)->op_state = 4;
+        PRIV(op)->op_state = 3;
         return 1;
     }
-resume_5:
+resume_4:
     // Mark dirty_db entries as stable, acknowledge op completion
     obj_ver_id* v;
     int i;
@@ -257,21 +240,3 @@ void blockstore_impl_t::mark_stable(const obj_ver_id & v, bool forget_dirty)
         unstable_writes.erase(unstab_it);
     }
 }
-
-void blockstore_impl_t::handle_stable_event(ring_data_t *data, blockstore_op_t *op)
-{
-    live = true;
-    if (data->res != data->iov.iov_len)
-    {
-        throw std::runtime_error(
-            "write operation failed ("+std::to_string(data->res)+" != "+std::to_string(data->iov.iov_len)+
-            "). in-memory state is corrupted. AAAAAAAaaaaaaaaa!!!111"
-        );
-    }
-    PRIV(op)->pending_ops--;
-    if (PRIV(op)->pending_ops == 0)
-    {
-        PRIV(op)->op_state++;
-        ringloop->wakeup();
-    }
-}

src/blockstore_sync.cpp

@@ -44,10 +44,8 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op, bool queue_has_in_prog
         if (journal.sector_info[journal.cur_sector].dirty)
         {
             // Write out the last journal sector if it happens to be dirty
-            BS_SUBMIT_GET_ONLY_SQE(sqe);
+            BS_SUBMIT_CHECK_SQES(1);
-            prepare_journal_sector_write(journal, journal.cur_sector, sqe, [this, op](ring_data_t *data) { handle_sync_event(data, op); });
+            prepare_journal_sector_write(journal.cur_sector, op);
-            PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
-            PRIV(op)->pending_ops = 1;
             PRIV(op)->op_state = SYNC_JOURNAL_WRITE_SENT;
             return 1;
         }
@@ -64,7 +62,7 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op, bool queue_has_in_prog
             BS_SUBMIT_GET_SQE(sqe, data);
             my_uring_prep_fsync(sqe, data_fd, IORING_FSYNC_DATASYNC);
             data->iov = { 0 };
-            data->callback = [this, op](ring_data_t *data) { handle_sync_event(data, op); };
+            data->callback = [this, op](ring_data_t *data) { handle_write_event(data, op); };
             PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 0;
             PRIV(op)->pending_ops = 1;
             PRIV(op)->op_state = SYNC_DATA_SYNC_SENT;
@@ -85,24 +83,18 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op, bool queue_has_in_prog
         {
             return 0;
         }
-        // Get SQEs. Don't bother about merging, submit each journal sector as a separate request
+        // Check SQEs. Don't bother about merging, submit each journal sector as a separate request
-        struct io_uring_sqe *sqe[space_check.sectors_to_write];
+        BS_SUBMIT_CHECK_SQES(space_check.sectors_to_write);
-        for (int i = 0; i < space_check.sectors_to_write; i++)
-        {
-            BS_SUBMIT_GET_SQE_DECL(sqe[i]);
-        }
         // Prepare and submit journal entries
         auto it = PRIV(op)->sync_big_writes.begin();
-        int s = 0, cur_sector = -1;
+        int s = 0;
         while (it != PRIV(op)->sync_big_writes.end())
         {
             if (!journal.entry_fits(sizeof(journal_entry_big_write) + clean_entry_bitmap_size) &&
                 journal.sector_info[journal.cur_sector].dirty)
             {
-                if (cur_sector == -1)
+                prepare_journal_sector_write(journal.cur_sector, op);
-                    PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
+                s++;
-                prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], [this, op](ring_data_t *data) { handle_sync_event(data, op); });
-                cur_sector = journal.cur_sector;
             }
             auto & dirty_entry = dirty_db.at(*it);
             journal_entry_big_write *je = (journal_entry_big_write*)prefill_single_journal_entry(
@@ -129,12 +121,9 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op, bool queue_has_in_prog
             journal.crc32_last = je->crc32;
             it++;
         }
-        prepare_journal_sector_write(journal, journal.cur_sector, sqe[s++], [this, op](ring_data_t *data) { handle_sync_event(data, op); });
+        prepare_journal_sector_write(journal.cur_sector, op);
+        s++;
         assert(s == space_check.sectors_to_write);
-        if (cur_sector == -1)
-            PRIV(op)->min_flushed_journal_sector = 1 + journal.cur_sector;
-        PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
-        PRIV(op)->pending_ops = s;
         PRIV(op)->op_state = SYNC_JOURNAL_WRITE_SENT;
         return 1;
     }
@@ -145,7 +134,7 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op, bool queue_has_in_prog
             BS_SUBMIT_GET_SQE(sqe, data);
             my_uring_prep_fsync(sqe, journal.fd, IORING_FSYNC_DATASYNC);
             data->iov = { 0 };
-            data->callback = [this, op](ring_data_t *data) { handle_sync_event(data, op); };
+            data->callback = [this, op](ring_data_t *data) { handle_write_event(data, op); };
             PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 0;
             PRIV(op)->pending_ops = 1;
             PRIV(op)->op_state = SYNC_JOURNAL_SYNC_SENT;
@@ -164,42 +153,6 @@ int blockstore_impl_t::continue_sync(blockstore_op_t *op, bool queue_has_in_prog
     return 1;
 }
 
-void blockstore_impl_t::handle_sync_event(ring_data_t *data, blockstore_op_t *op)
-{
-    live = true;
-    if (data->res != data->iov.iov_len)
-    {
-        throw std::runtime_error(
-            "write operation failed ("+std::to_string(data->res)+" != "+std::to_string(data->iov.iov_len)+
-            "). in-memory state is corrupted. AAAAAAAaaaaaaaaa!!!111"
-        );
-    }
-    PRIV(op)->pending_ops--;
-    if (PRIV(op)->pending_ops == 0)
-    {
-        // Release used journal sectors
-        release_journal_sectors(op);
-        // Handle states
-        if (PRIV(op)->op_state == SYNC_DATA_SYNC_SENT)
-        {
-            PRIV(op)->op_state = SYNC_DATA_SYNC_DONE;
-        }
-        else if (PRIV(op)->op_state == SYNC_JOURNAL_WRITE_SENT)
-        {
-            PRIV(op)->op_state = SYNC_JOURNAL_WRITE_DONE;
-        }
-        else if (PRIV(op)->op_state == SYNC_JOURNAL_SYNC_SENT)
-        {
-            PRIV(op)->op_state = SYNC_DONE;
-        }
-        else
-        {
-            throw std::runtime_error("BUG: unexpected sync op state");
-        }
-        ringloop->wakeup();
-    }
-}
-
 void blockstore_impl_t::ack_sync(blockstore_op_t *op)
 {
     // Handle states

src/blockstore_write.cpp

@@ -268,18 +268,8 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
             cancel_all_writes(op, dirty_it, -ENOSPC);
             return 2;
         }
-#ifndef NDEBUG
-        // Double-check to not overwrite anything if easily possible
-        if (inmemory_meta)
-        {
-            uint64_t sector = (loc / (meta_block_size / clean_entry_size)) * meta_block_size;
-            uint64_t pos = (loc % (meta_block_size / clean_entry_size));
-            clean_disk_entry *meta_entry = (clean_disk_entry*)(metadata_buffer + sector + pos*clean_entry_size);
-            assert(!meta_entry->oid.inode);
-        }
-#endif
-        write_iodepth++;
         BS_SUBMIT_GET_SQE(sqe, data);
+        write_iodepth++;
         dirty_it->second.location = loc << block_order;
         dirty_it->second.state = (dirty_it->second.state & ~BS_ST_WORKFLOW_MASK) | BS_ST_SUBMITTED;
 #ifdef BLOCKSTORE_DEBUG
@@ -334,29 +324,21 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
         {
             return 0;
         }
-        write_iodepth++;
+        // There is sufficient space. Check SQE(s)
-        // There is sufficient space. Get SQE(s)
+        BS_SUBMIT_CHECK_SQES(
-        struct io_uring_sqe *sqe1 = NULL;
-        if (immediate_commit != IMMEDIATE_NONE ||
-            !journal.entry_fits(sizeof(journal_entry_small_write) + clean_entry_bitmap_size))
-        {
             // Write current journal sector only if it's dirty and full, or in the immediate_commit mode
-            BS_SUBMIT_GET_SQE_DECL(sqe1);
+            (immediate_commit != IMMEDIATE_NONE ||
-        }
+                !journal.entry_fits(sizeof(journal_entry_small_write) + clean_entry_bitmap_size) ? 1 : 0) +
-        struct io_uring_sqe *sqe2 = NULL;
+            (op->len > 0 ? 1 : 0)
-        if (op->len > 0)
+        );
-        {
+        write_iodepth++;
-            BS_SUBMIT_GET_SQE_DECL(sqe2);
-        }
         // Got SQEs. Prepare previous journal sector write if required
         auto cb = [this, op](ring_data_t *data) { handle_write_event(data, op); };
         if (immediate_commit == IMMEDIATE_NONE)
         {
-            if (sqe1)
+            if (!journal.entry_fits(sizeof(journal_entry_small_write) + clean_entry_bitmap_size))
             {
-                prepare_journal_sector_write(journal, journal.cur_sector, sqe1, cb);
+                prepare_journal_sector_write(journal.cur_sector, op);
-                PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
-                PRIV(op)->pending_ops++;
             }
             else
             {
@@ -390,9 +372,7 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
         journal.crc32_last = je->crc32;
         if (immediate_commit != IMMEDIATE_NONE)
         {
-            prepare_journal_sector_write(journal, journal.cur_sector, sqe1, cb);
+            prepare_journal_sector_write(journal.cur_sector, op);
-            PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
-            PRIV(op)->pending_ops++;
         }
         if (op->len > 0)
         {
@@ -402,7 +382,7 @@ int blockstore_impl_t::dequeue_write(blockstore_op_t *op)
                 // Copy data
                 memcpy(journal.buffer + journal.next_free, op->buf, op->len);
             }
-            ring_data_t *data2 = ((ring_data_t*)sqe2->user_data);
+            BS_SUBMIT_GET_SQE(sqe2, data2);
             data2->iov = (struct iovec){ op->buf, op->len };
             data2->callback = cb;
             my_uring_prep_writev(
@@ -451,13 +431,12 @@ int blockstore_impl_t::continue_write(blockstore_op_t *op)
 resume_2:
     // Only for the immediate_commit mode: prepare and submit big_write journal entry
     {
+        BS_SUBMIT_CHECK_SQES(1);
         auto dirty_it = dirty_db.find((obj_ver_id){
             .oid = op->oid,
             .version = op->version,
         });
         assert(dirty_it != dirty_db.end());
-        io_uring_sqe *sqe = NULL;
-        BS_SUBMIT_GET_SQE_DECL(sqe);
         journal_entry_big_write *je = (journal_entry_big_write*)prefill_single_journal_entry(
             journal, op->opcode == BS_OP_WRITE_STABLE ? JE_BIG_WRITE_INSTANT : JE_BIG_WRITE,
             sizeof(journal_entry_big_write) + clean_entry_bitmap_size
@@ -479,10 +458,7 @@ resume_2:
         memcpy((void*)(je+1), (clean_entry_bitmap_size > sizeof(void*) ? dirty_it->second.bitmap : &dirty_it->second.bitmap), clean_entry_bitmap_size);
         je->crc32 = je_crc32((journal_entry*)je);
         journal.crc32_last = je->crc32;
-        prepare_journal_sector_write(journal, journal.cur_sector, sqe,
+        prepare_journal_sector_write(journal.cur_sector, op);
-            [this, op](ring_data_t *data) { handle_write_event(data, op); });
-        PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
-        PRIV(op)->pending_ops = 1;
         PRIV(op)->op_state = 3;
         return 1;
     }
@@ -597,6 +573,7 @@ void blockstore_impl_t::handle_write_event(ring_data_t *data, blockstore_op_t *o
         );
     }
     PRIV(op)->pending_ops--;
+    assert(PRIV(op)->pending_ops >= 0);
     if (PRIV(op)->pending_ops == 0)
     {
         release_journal_sectors(op);
@@ -614,7 +591,6 @@ void blockstore_impl_t::release_journal_sectors(blockstore_op_t *op)
         uint64_t s = PRIV(op)->min_flushed_journal_sector;
         while (1)
         {
-            journal.sector_info[s-1].flush_count--;
             if (s != (1+journal.cur_sector) && journal.sector_info[s-1].flush_count == 0)
             {
                 // We know for sure that we won't write into this sector anymore
@@ -654,23 +630,19 @@ int blockstore_impl_t::dequeue_del(blockstore_op_t *op)
         return 0;
     }
     write_iodepth++;
-    io_uring_sqe *sqe = NULL;
-    if (immediate_commit != IMMEDIATE_NONE ||
-        (journal_block_size - journal.in_sector_pos) < sizeof(journal_entry_del) &&
-        journal.sector_info[journal.cur_sector].dirty)
-    {
     // Write current journal sector only if it's dirty and full, or in the immediate_commit mode
-        BS_SUBMIT_GET_SQE_DECL(sqe);
+    BS_SUBMIT_CHECK_SQES(
-    }
+        (immediate_commit != IMMEDIATE_NONE ||
-    auto cb = [this, op](ring_data_t *data) { handle_write_event(data, op); };
+            (journal_block_size - journal.in_sector_pos) < sizeof(journal_entry_del) &&
+            journal.sector_info[journal.cur_sector].dirty) ? 1 : 0
+    );
     // Prepare journal sector write
     if (immediate_commit == IMMEDIATE_NONE)
     {
-        if (sqe)
+        if ((journal_block_size - journal.in_sector_pos) < sizeof(journal_entry_del) &&
+            journal.sector_info[journal.cur_sector].dirty)
         {
-            prepare_journal_sector_write(journal, journal.cur_sector, sqe, cb);
+            prepare_journal_sector_write(journal.cur_sector, op);
-            PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
-            PRIV(op)->pending_ops++;
         }
         else
         {
@@ -697,9 +669,7 @@ int blockstore_impl_t::dequeue_del(blockstore_op_t *op)
     dirty_it->second.state = BS_ST_DELETE | BS_ST_SUBMITTED;
     if (immediate_commit != IMMEDIATE_NONE)
     {
-        prepare_journal_sector_write(journal, journal.cur_sector, sqe, cb);
+        prepare_journal_sector_write(journal.cur_sector, op);
-        PRIV(op)->min_flushed_journal_sector = PRIV(op)->max_flushed_journal_sector = 1 + journal.cur_sector;
-        PRIV(op)->pending_ops++;
     }
     if (!PRIV(op)->pending_ops)
     {

src/cli.cpp

@@ -305,6 +305,10 @@ void cli_tool_t::run(json11::Json cfg)
         fprintf(stderr, "unknown command: %s\n", cmd[0].string_value().c_str());
         exit(1);
     }
+    if (action_cb == NULL)
+    {
+        return;
+    }
     color = !cfg["no-color"].bool_value();
     json_output = cfg["json"].bool_value();
     iodepth = cfg["iodepth"].uint64_value();

src/fio_sec_osd.cpp

@@ -28,16 +28,23 @@
 #include <vector>
 #include <unordered_map>
 
+#include "addr_util.h"
 #include "rw_blocking.h"
 #include "osd_ops.h"
 #include "fio_headers.h"
 
+struct op_buf_t
+{
+    osd_any_op_t buf;
+    io_u* fio_op;
+};
+
 struct sec_data
 {
     int connect_fd;
     /* block_size = 1 << block_order (128KB by default) */
     uint64_t block_order = 17, block_size = 1 << 17;
-    std::unordered_map<uint64_t, io_u*> queue;
+    std::unordered_map<uint64_t, op_buf_t*> queue;
     bool last_sync = false;
     /* The list of completed io_u structs. */
     std::vector<io_u*> completed;
@@ -52,6 +59,7 @@ struct sec_options
     int single_primary = 0;
     int trace = 0;
     int block_order = 17;
+    int zerocopy_send = 0;
 };
 
 static struct fio_option options[] = {
@@ -102,6 +110,16 @@ static struct fio_option options[] = {
         .category = FIO_OPT_C_ENGINE,
         .group  = FIO_OPT_G_FILENAME,
     },
+    {
+        .name   = "zerocopy_send",
+        .lname  = "Use zero-copy send",
+        .type   = FIO_OPT_BOOL,
+        .off1   = offsetof(struct sec_options, zerocopy_send),
+        .help   = "Use zero-copy send (MSG_ZEROCOPY)",
+        .def    = "0",
+        .category = FIO_OPT_C_ENGINE,
+        .group  = FIO_OPT_G_FILENAME,
+    },
     {
         .name = NULL,
     },
@@ -152,17 +170,14 @@ static int sec_init(struct thread_data *td)
     bsd->block_order = o->block_order == 0 ? 17 : o->block_order;
     bsd->block_size = 1 << o->block_order;
 
-    struct sockaddr_in addr;
+    sockaddr addr;
-    int r;
+    if (!string_to_addr(std::string(o->host ? o->host : "127.0.0.1"), false, o->port > 0 ? o->port : 11203, &addr))
-    if ((r = inet_pton(AF_INET, o->host ? o->host : "127.0.0.1", &addr.sin_addr)) != 1)
     {
-        fprintf(stderr, "server address: %s%s\n", o->host ? o->host : "127.0.0.1", r == 0 ? " is not valid" : ": no ipv4 support");
+        fprintf(stderr, "server address: %s is not valid\n", o->host ? o->host : "127.0.0.1");
         return 1;
     }
-    addr.sin_family = AF_INET;
-    addr.sin_port = htons(o->port ? o->port : 11203);
 
-    bsd->connect_fd = socket(AF_INET, SOCK_STREAM, 0);
+    bsd->connect_fd = socket(addr.sa_family, SOCK_STREAM, 0);
     if (bsd->connect_fd < 0)
     {
         perror("socket");
@@ -175,6 +190,14 @@ static int sec_init(struct thread_data *td)
     }
     int one = 1;
     setsockopt(bsd->connect_fd, SOL_TCP, TCP_NODELAY, &one, sizeof(one));
+    if (o->zerocopy_send)
+    {
+        if (setsockopt(bsd->connect_fd, SOL_SOCKET, SO_ZEROCOPY, &one, sizeof(one)) < 0)
+        {
+            perror("setsockopt zerocopy");
+            return 1;
+        }
+    }
 
     // FIXME: read config (block size) from OSD
 
@@ -195,7 +218,9 @@ static enum fio_q_status sec_queue(struct thread_data *td, struct io_u *io)
     }
 
     io->engine_data = bsd;
-    osd_any_op_t op = { 0 };
+    op_buf_t *op_buf = new op_buf_t;
+    op_buf->fio_op = io;
+    osd_any_op_t &op = op_buf->buf;
 
     op.hdr.magic = SECONDARY_OSD_OP_MAGIC;
     op.hdr.id = n;
@@ -271,19 +296,18 @@ static enum fio_q_status sec_queue(struct thread_data *td, struct io_u *io)
     io->error = 0;
     bsd->inflight++;
     bsd->op_n++;
-    bsd->queue[n] = io;
+    bsd->queue[n] = op_buf;
 
     iovec iov[2] = { { .iov_base = op.buf, .iov_len = OSD_PACKET_SIZE } };
     int iovcnt = 1, wtotal = OSD_PACKET_SIZE;
     if (io->ddir == DDIR_WRITE)
     {
-        iov[1] = { .iov_base = io->xfer_buf, .iov_len = io->xfer_buflen };
+        iov[iovcnt++] = { .iov_base = io->xfer_buf, .iov_len = io->xfer_buflen };
         wtotal += io->xfer_buflen;
-        iovcnt++;
     }
-    if (writev_blocking(bsd->connect_fd, iov, iovcnt) != wtotal)
+    if (sendv_blocking(bsd->connect_fd, iov, iovcnt, opt->zerocopy_send ? MSG_ZEROCOPY : 0) != wtotal)
     {
-        perror("writev");
+        perror("sendmsg");
         exit(1);
     }
 
@@ -312,7 +336,8 @@ static int sec_getevents(struct thread_data *td, unsigned int min, unsigned int
             fprintf(stderr, "bad reply: op id %lx missing in local queue\n", reply.hdr.id);
             exit(1);
         }
-        io_u* io = it->second;
+        io_u* io = it->second->fio_op;
+        delete it->second;
         bsd->queue.erase(it);
         if (io->ddir == DDIR_READ)
         {
@@ -321,7 +346,23 @@ static int sec_getevents(struct thread_data *td, unsigned int min, unsigned int
                 fprintf(stderr, "Short read: retval = %ld instead of %llu\n", reply.hdr.retval, io->xfer_buflen);
                 exit(1);
             }
-            read_blocking(bsd->connect_fd, io->xfer_buf, io->xfer_buflen);
+            // Support bitmap
+            uint64_t bitmap = 0;
+            int iovcnt = 0;
+            iovec iov[2];
+            if (reply.sec_rw.attr_len > 0)
+            {
+                if (reply.sec_rw.attr_len <= 8)
+                    iov[iovcnt++] = { .iov_base = &bitmap, .iov_len = reply.sec_rw.attr_len };
+                else
+                    iov[iovcnt++] = { .iov_base = (void*)(bitmap = (uint64_t)malloc(reply.sec_rw.attr_len)), .iov_len = reply.sec_rw.attr_len };
+            }
+            iov[iovcnt++] = { .iov_base = io->xfer_buf, .iov_len = io->xfer_buflen };
+            readv_blocking(bsd->connect_fd, iov, iovcnt);
+            if (reply.sec_rw.attr_len > 8)
+            {
+                free((void*)bitmap);
+            }
         }
         else if (io->ddir == DDIR_WRITE)
         {

src/http_client.cpp

@@ -15,13 +15,13 @@
 
 #include <stdexcept>
 
+#include "addr_util.h"
 #include "json11/json11.hpp"
 #include "http_client.h"
 #include "timerfd_manager.h"
 
 #define READ_BUFFER_SIZE 9000
 
-static int extract_port(std::string & host);
 static std::string trim(const std::string & in);
 static std::string ws_format_frame(int type, uint64_t size);
 static bool ws_parse_frame(std::string & buf, int & type, std::string & res);
@@ -185,19 +185,15 @@ http_co_t::~http_co_t()
 void http_co_t::start_connection()
 {
     stackin();
-    int port = extract_port(host);
+    struct sockaddr addr;
-    struct sockaddr_in addr;
+    if (!string_to_addr(host.c_str(), 1, 80, &addr))
-    int r;
-    if ((r = inet_pton(AF_INET, host.c_str(), &addr.sin_addr)) != 1)
     {
         parsed.error_code = ENXIO;
         stackout();
         end();
         return;
     }
-    addr.sin_family = AF_INET;
+    peer_fd = socket(addr.sa_family, SOCK_STREAM, 0);
-    addr.sin_port = htons(port ? port : 80);
-    peer_fd = socket(AF_INET, SOCK_STREAM, 0);
     if (peer_fd < 0)
     {
         parsed.error_code = errno;
@@ -219,7 +215,7 @@ void http_co_t::start_connection()
     }
     epoll_events = 0;
     // Finally call connect
-    r = ::connect(peer_fd, (sockaddr*)&addr, sizeof(addr));
+    int r = ::connect(peer_fd, (sockaddr*)&addr, sizeof(addr));
     if (r < 0 && errno != EINPROGRESS)
     {
         parsed.error_code = errno;
@@ -759,22 +755,6 @@ std::vector<std::string> getifaddr_list(json11::Json mask_cfg, bool include_v6)
     return addresses;
 }
 
-static int extract_port(std::string & host)
-{
-    int port = 0;
-    int pos = 0;
-    if ((pos = host.find(':')) >= 0)
-    {
-        port = strtoull(host.c_str() + pos + 1, NULL, 10);
-        if (port >= 0x10000)
-        {
-            port = 0;
-        }
-        host = host.substr(0, pos);
-    }
-    return port;
-}
-
 std::string strtolower(const std::string & in)
 {
     std::string s = in;

src/http_client.h

@@ -45,7 +45,7 @@ struct websocket_t
 
 void parse_http_headers(std::string & res, http_response_t *parsed);
 
-std::vector<std::string> getifaddr_list(json11::Json mask_cfg = json11::Json(), bool include_v6 = false);
+std::vector<std::string> getifaddr_list(json11::Json mask_cfg = json11::Json(), bool include_v6 = true);
 
 uint64_t stoull_full(const std::string & str, int base = 10);
 

src/messenger.cpp

@@ -4,10 +4,13 @@
 #include <unistd.h>
 #include <fcntl.h>
 #include <sys/socket.h>
+#include <sys/random.h>
 #include <sys/epoll.h>
 #include <netinet/tcp.h>
 #include <stdexcept>
 
+#include "base64.h"
+#include "addr_util.h"
 #include "messenger.h"
 
 void osd_messenger_t::init()
@@ -193,7 +196,7 @@ void osd_messenger_t::connect_peer(uint64_t peer_osd, json11::Json peer_state)
     try_connect_peer(peer_osd);
 }
 
-void osd_messenger_t::try_connect_peer(uint64_t peer_osd)
+void osd_messenger_t::try_connect_peer(osd_num_t peer_osd)
 {
     auto wp_it = wanted_peers.find(peer_osd);
     if (wp_it == wanted_peers.end() || wp_it->second.connecting ||
@@ -214,43 +217,75 @@ void osd_messenger_t::try_connect_peer(uint64_t peer_osd)
     wp.cur_addr = wp.address_list[wp.address_index].string_value();
     wp.cur_port = wp.port;
     wp.connecting = true;
-    try_connect_peer_addr(peer_osd, wp.cur_addr.c_str(), wp.cur_port);
+    try_connect_peer_addr(peer_osd, wp.cur_addr.c_str(), wp.cur_port, NULL, [this](osd_num_t peer_osd, int peer_fd)
+    {
+        if (peer_fd >= 0)
+            osd_peer_fds[peer_osd] = peer_fd;
+        on_connect_peer(peer_osd, peer_fd);
+    });
 }
 
-void osd_messenger_t::try_connect_peer_addr(osd_num_t peer_osd, const char *peer_host, int peer_port)
+static std::string urandom_str(int bytes)
+{
+    std::string str;
+    str.resize(bytes);
+    char *buf = (char*)str.data();
+    while (bytes > 0)
+    {
+        int r = getrandom(buf, bytes, 0);
+        if (r < 0)
+            throw std::runtime_error(std::string("getrandom: ") + strerror(errno));
+        buf += r;
+        bytes -= r;
+    }
+    return str;
+}
+
+void osd_messenger_t::try_connect_peer_addr(osd_num_t peer_osd, const char *peer_host, int peer_port,
+    osd_client_t *meta_cl, std::function<void(osd_num_t, int)> connect_callback)
 {
     assert(peer_osd != this->osd_num);
-    struct sockaddr_in addr;
+    struct sockaddr addr;
-    int r;
+    if (!meta_cl)
-    if ((r = inet_pton(AF_INET, peer_host, &addr.sin_addr)) != 1)
     {
-        on_connect_peer(peer_osd, -EINVAL);
+        if (!string_to_addr(peer_host, 0, peer_port, &addr))
+        {
+            connect_callback(peer_osd, -EINVAL);
             return;
         }
-    addr.sin_family = AF_INET;
-    addr.sin_port = htons(peer_port ? peer_port : 11203);
-    int peer_fd = socket(AF_INET, SOCK_STREAM, 0);
-    if (peer_fd < 0)
-    {
-        on_connect_peer(peer_osd, -errno);
-        return;
     }
+    else
+    {
+        addr = meta_cl->peer_addr;
+    }
+    int peer_fd = socket(addr.sa_family, SOCK_STREAM, 0);
+    if (peer_fd >= 0)
+    {
         fcntl(peer_fd, F_SETFL, fcntl(peer_fd, F_GETFL, 0) | O_NONBLOCK);
-    r = connect(peer_fd, (sockaddr*)&addr, sizeof(addr));
+        int r = connect(peer_fd, (sockaddr*)&addr, sizeof(addr));
         if (r < 0 && errno != EINPROGRESS)
         {
             close(peer_fd);
-        on_connect_peer(peer_osd, -errno);
+            peer_fd = -1;
+        }
+    }
+    if (peer_fd < 0)
+    {
+        connect_callback(peer_osd, -errno);
         return;
     }
     clients[peer_fd] = new osd_client_t();
     clients[peer_fd]->peer_addr = addr;
-    clients[peer_fd]->peer_port = peer_port;
+    clients[peer_fd]->peer_port = ((struct sockaddr_in*)&addr)->sin_port;
     clients[peer_fd]->peer_fd = peer_fd;
     clients[peer_fd]->peer_state = PEER_CONNECTING;
     clients[peer_fd]->connect_timeout_id = -1;
+    clients[peer_fd]->connect_callback = connect_callback;
     clients[peer_fd]->osd_num = peer_osd;
     clients[peer_fd]->in_buf = malloc_or_die(receive_buffer_size);
+    clients[peer_fd]->data_for = meta_cl ? addr_to_string(meta_cl->peer_addr) : "";
+    clients[peer_fd]->data_connection_cookie = meta_cl
+        ? meta_cl->data_connection_cookie : base64_encode(urandom_str(12));
     tfd->set_fd_handler(peer_fd, true, [this](int peer_fd, int epoll_events)
     {
         // Either OUT (connected) or HUP
@@ -260,10 +295,12 @@ void osd_messenger_t::try_connect_peer_addr(osd_num_t peer_osd, const char *peer
     {
         clients[peer_fd]->connect_timeout_id = tfd->set_timer(1000*peer_connect_timeout, false, [this, peer_fd](int timer_id)
         {
-            osd_num_t peer_osd = clients.at(peer_fd)->osd_num;
+            auto cl = clients.at(peer_fd);
+            auto connect_callback = cl->connect_callback;
+            cl->connect_callback = NULL;
+            osd_num_t peer_osd = cl->osd_num;
             stop_client(peer_fd, true);
-            on_connect_peer(peer_osd, -EPIPE);
+            connect_callback(peer_osd, -EPIPE);
-            return;
         });
     }
 }
@@ -285,8 +322,10 @@ void osd_messenger_t::handle_connect_epoll(int peer_fd)
     }
     if (result != 0)
     {
+        auto connect_callback = cl->connect_callback;
+        cl->connect_callback = NULL;
         stop_client(peer_fd, true);
-        on_connect_peer(peer_osd, -result);
+        connect_callback(peer_osd, -result);
         return;
     }
     int one = 1;
@@ -366,6 +405,11 @@ void osd_messenger_t::on_connect_peer(osd_num_t peer_osd, int peer_fd)
 
 void osd_messenger_t::check_peer_config(osd_client_t *cl)
 {
+    json11::Json::object payload;
+    if (cl->data_connection_cookie != "")
+    {
+        payload["data_cookie"] = cl->data_connection_cookie;
+    }
     osd_op_t *op = new osd_op_t();
     op->op_type = OSD_OP_OUT;
     op->peer_fd = cl->peer_fd;
@@ -378,24 +422,33 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
             },
         },
     };
+    if (cl->data_for == "")
+    {
 #ifdef WITH_RDMA
         if (rdma_context)
         {
             cl->rdma_conn = msgr_rdma_connection_t::create(rdma_context, rdma_max_send, rdma_max_recv, rdma_max_sge, rdma_max_msg);
             if (cl->rdma_conn)
             {
-            json11::Json payload = json11::Json::object {
+                payload["connect_rdma"] = cl->rdma_conn->addr.to_string();
-                { "connect_rdma", cl->rdma_conn->addr.to_string() },
+                payload["rdma_max_msg"] = cl->rdma_conn->max_msg;
-                { "rdma_max_msg", cl->rdma_conn->max_msg },
+            }
-            };
+        }
-            std::string payload_str = payload.dump();
+#endif
+    }
+    else
+    {
+        // Mark it as a data connection
+        payload["data_for"] = cl->data_for;
+    }
+    if (payload.size())
+    {
+        std::string payload_str = json11::Json(payload).dump();
         op->req.show_conf.json_len = payload_str.size();
         op->buf = malloc_or_die(payload_str.size());
         op->iov.push_back(op->buf, payload_str.size());
         memcpy(op->buf, payload_str.c_str(), payload_str.size());
     }
-    }
-#endif
     op->callback = [this, cl](osd_op_t *op)
     {
         std::string json_err;
@@ -428,18 +481,30 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
                     cl->osd_num, config["protocol_version"].uint64_value(), OSD_PROTOCOL_VERSION
                 );
             }
+            else if (cl->data_for != "" && config["data_for"] != cl->data_for)
+            {
+                err = true;
+                fprintf(
+                    stderr, "OSD %lu does not support separate data connections."
+                    " Proceeding with a single connection\n", cl->osd_num
+                );
+            }
         }
         if (err)
         {
             osd_num_t peer_osd = cl->osd_num;
+            auto connect_callback = cl->connect_callback;
+            cl->connect_callback = NULL;
             stop_client(op->peer_fd);
-            on_connect_peer(peer_osd, -1);
+            connect_callback(peer_osd, -EINVAL);
             delete op;
             return;
         }
 #ifdef WITH_RDMA
-        if (config["rdma_address"].is_string())
+        if (rdma_context && cl->rdma_conn && config["rdma_address"].is_string())
         {
+            // Prevent creating data connection - we are trying RDMA
+            cl->data_connection_cookie = "";
             msgr_rdma_address_t addr;
             if (!msgr_rdma_address_t::from_string(config["rdma_address"].string_value().c_str(), &addr) ||
                 cl->rdma_conn->connect(&addr) != 0)
@@ -452,8 +517,10 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
                 cl->rdma_conn = NULL;
                 // FIXME: Keep TCP connection in this case
                 osd_num_t peer_osd = cl->osd_num;
+                auto connect_callback = cl->connect_callback;
+                cl->connect_callback = NULL;
                 stop_client(cl->peer_fd);
-                on_connect_peer(peer_osd, -1);
+                connect_callback(peer_osd, -EPIPE);
                 delete op;
                 return;
             }
@@ -475,8 +542,37 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
             }
         }
 #endif
+        if (cl->data_connection_cookie != "")
+        {
+            // Try to open second connection to the same address
+            try_connect_peer_addr(cl->osd_num, NULL, 0, cl, [this, peer_fd = cl->peer_fd](osd_num_t data_peer, int data_peer_fd)
+            {
+                auto cl_it = clients.find(peer_fd);
+                if (cl_it != clients.end())
+                {
+                    // Proceed with or without the data connection
+                    auto cl = cl_it->second;
+                    if (data_peer_fd >= 0)
+                    {
+                        cl->data_connection_fd = data_peer_fd;
+                        auto data_cl = clients.at(data_peer_fd);
+                        data_cl->meta_connection_fd = cl->peer_fd;
+                    }
                     osd_peer_fds[cl->osd_num] = cl->peer_fd;
                     on_connect_peer(cl->osd_num, cl->peer_fd);
+                }
+                else if (data_peer_fd >= 0)
+                {
+                    stop_client(data_peer_fd);
+                }
+            });
+        }
+        else
+        {
+            auto connect_callback = cl->connect_callback;
+            cl->connect_callback = NULL;
+            connect_callback(cl->osd_num, cl->peer_fd);
+        }
         delete op;
     };
     outbox_push(op);
@@ -485,24 +581,24 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
 void osd_messenger_t::accept_connections(int listen_fd)
 {
     // Accept new connections
-    sockaddr_in addr;
+    sockaddr addr;
     socklen_t peer_addr_size = sizeof(addr);
     int peer_fd;
-    while ((peer_fd = accept(listen_fd, (sockaddr*)&addr, &peer_addr_size)) >= 0)
+    while ((peer_fd = accept(listen_fd, &addr, &peer_addr_size)) >= 0)
     {
         assert(peer_fd != 0);
-        char peer_str[256];
+        fprintf(stderr, "[OSD %lu] new client %d: connection from %s\n", this->osd_num, peer_fd,
-        fprintf(stderr, "[OSD %lu] new client %d: connection from %s port %d\n", this->osd_num, peer_fd,
+            addr_to_string(addr).c_str());
-            inet_ntop(AF_INET, &addr.sin_addr, peer_str, 256), ntohs(addr.sin_port));
         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));
         clients[peer_fd] = new osd_client_t();
         clients[peer_fd]->peer_addr = addr;
-        clients[peer_fd]->peer_port = ntohs(addr.sin_port);
+        clients[peer_fd]->peer_port = ntohs(((sockaddr_in*)&addr)->sin_port);
         clients[peer_fd]->peer_fd = peer_fd;
         clients[peer_fd]->peer_state = PEER_CONNECTED;
         clients[peer_fd]->in_buf = malloc_or_die(receive_buffer_size);
+        clients_by_addr[addr_to_string(addr)] = peer_fd;
         // Add FD to epoll
         tfd->set_fd_handler(peer_fd, false, [this](int peer_fd, int epoll_events)
         {

src/messenger.h

@@ -49,7 +49,7 @@ struct osd_client_t
 {
     int refs = 0;
 
-    sockaddr_in peer_addr;
+    sockaddr peer_addr;
     int peer_port;
     int peer_fd;
     int peer_state;
@@ -57,6 +57,10 @@ struct osd_client_t
     int ping_time_remaining = 0;
     int idle_time_remaining = 0;
     osd_num_t osd_num = 0;
+    std::function<void(osd_num_t, int)> connect_callback;
+
+    int data_connection_fd = -1, meta_connection_fd = -1;
+    std::string data_connection_cookie, data_for;
 
     void *in_buf = NULL;
 
@@ -148,6 +152,7 @@ public:
     osd_num_t osd_num;
     uint64_t next_subop_id = 1;
     std::map<int, osd_client_t*> clients;
+    std::map<std::string, int> clients_by_addr;
     std::map<osd_num_t, osd_wanted_peer_t> wanted_peers;
     std::map<uint64_t, int> osd_peer_fds;
     // op statistics
@@ -157,6 +162,7 @@ public:
     void parse_config(const json11::Json & config);
     void connect_peer(uint64_t osd_num, json11::Json peer_state);
     void stop_client(int peer_fd, bool force = false, bool force_delete = false);
+    void break_data_client_pair(osd_client_t *cl);
     void outbox_push(osd_op_t *cur_op);
     std::function<void(osd_op_t*)> exec_op;
     std::function<void(osd_num_t)> repeer_pgs;
@@ -174,7 +180,8 @@ public:
 
 protected:
     void try_connect_peer(uint64_t osd_num);
-    void try_connect_peer_addr(osd_num_t peer_osd, const char *peer_host, int peer_port);
+    void try_connect_peer_addr(osd_num_t peer_osd, const char *peer_host, int peer_port,
+        osd_client_t *meta_cl, std::function<void(osd_num_t, int)> connect_callback);
     void handle_peer_epoll(int peer_fd, int epoll_events);
     void handle_connect_epoll(int peer_fd);
     void on_connect_peer(osd_num_t peer_osd, int peer_fd);

src/msgr_stop.cpp

@@ -4,6 +4,7 @@
 #include <unistd.h>
 #include <assert.h>
 
+#include "addr_util.h"
 #include "messenger.h"
 
 void osd_messenger_t::cancel_osd_ops(osd_client_t *cl)
@@ -58,7 +59,8 @@ void osd_messenger_t::stop_client(int peer_fd, bool force, bool force_delete)
     {
         if (cl->osd_num)
         {
-            fprintf(stderr, "[OSD %lu] Stopping client %d (OSD peer %lu)\n", osd_num, peer_fd, cl->osd_num);
+            fprintf(stderr, "[OSD %lu] Stopping client %d (OSD %speer %lu)\n",
+                osd_num, peer_fd, cl->meta_connection_fd >= 0 ? " data" : "", cl->osd_num);
         }
         else
         {
@@ -68,7 +70,7 @@ void osd_messenger_t::stop_client(int peer_fd, bool force, bool force_delete)
     // First set state to STOPPED so another stop_client() call doesn't try to free it again
     cl->refs++;
     cl->peer_state = PEER_STOPPED;
-    if (cl->osd_num)
+    if (cl->osd_num && cl->meta_connection_fd < 0)
     {
         // ...and forget OSD peer
         osd_peer_fds.erase(cl->osd_num);
@@ -99,11 +101,19 @@ void osd_messenger_t::stop_client(int peer_fd, bool force, bool force_delete)
     }
 #endif
     if (cl->osd_num)
+    {
+        if (cl->meta_connection_fd < 0)
         {
             // Then repeer PGs because cancel_op() callbacks can try to perform
             // some actions and we need correct PG states to not do something silly
             repeer_pgs(cl->osd_num);
         }
+        else
+        {
+            // FIXME Try to re-establish data connection
+            // Only when the connection is outbound, but here it's always outbound
+        }
+    }
     // Then cancel all operations
     if (cl->read_op)
     {
@@ -128,6 +138,7 @@ void osd_messenger_t::stop_client(int peer_fd, bool force, bool force_delete)
         delete cl->rdma_conn;
     }
 #endif
+    clients_by_addr.erase(addr_to_string(cl->peer_addr));
 #endif
     // Find the item again because it can be invalidated at this point
     it = clients.find(peer_fd);
@@ -135,9 +146,40 @@ void osd_messenger_t::stop_client(int peer_fd, bool force, bool force_delete)
     {
         clients.erase(it);
     }
+    // Break metadata/data connection pair
+    if (cl->data_connection_fd >= 0)
+    {
+        // No sense to keep data connection when metadata connection is stopped
+        auto dc_it = clients.find(cl->data_connection_fd);
+        cl->data_connection_fd = -1;
+        if (dc_it != clients.end() && dc_it->second->meta_connection_fd == cl->peer_fd)
+        {
+            stop_client(dc_it->second->peer_fd);
+        }
+    }
+    break_data_client_pair(cl);
+    // Refcount and delete
     cl->refs--;
     if (cl->refs <= 0 || force_delete)
     {
         delete cl;
     }
 }
+
+void osd_messenger_t::break_data_client_pair(osd_client_t *cl)
+{
+    if (cl->meta_connection_fd >= 0)
+    {
+        auto dc_it = clients.find(cl->meta_connection_fd);
+        if (dc_it != clients.end() && dc_it->second->data_connection_fd == cl->peer_fd)
+            dc_it->second->data_connection_fd = -1;
+        cl->meta_connection_fd = -1;
+    }
+    if (cl->data_connection_fd >= 0)
+    {
+        auto dc_it = clients.find(cl->data_connection_fd);
+        if (dc_it != clients.end() && dc_it->second->meta_connection_fd == cl->peer_fd)
+            dc_it->second->meta_connection_fd = -1;
+        cl->data_connection_fd = -1;
+    }
+}

src/osd.cpp

@@ -7,6 +7,7 @@
 #include <netinet/tcp.h>
 #include <arpa/inet.h>
 
+#include "addr_util.h"
 #include "blockstore_impl.h"
 #include "osd_primary.h"
 #include "osd.h"
@@ -156,14 +157,6 @@ void osd_t::parse_config(const json11::Json & config)
 
 void osd_t::bind_socket()
 {
-    listen_fd = socket(AF_INET, SOCK_STREAM, 0);
-    if (listen_fd < 0)
-    {
-        throw std::runtime_error(std::string("socket: ") + strerror(errno));
-    }
-    int enable = 1;
-    setsockopt(listen_fd, SOL_SOCKET, SO_REUSEADDR, &enable, sizeof(enable));
-
     if (config["osd_network"].is_string() ||
         config["osd_network"].is_array())
     {
@@ -173,7 +166,7 @@ void osd_t::bind_socket()
         else
             for (auto v: config["osd_network"].array_items())
                 mask.push_back(v.string_value());
-        auto matched_addrs = getifaddr_list(mask, false);
+        auto matched_addrs = getifaddr_list(mask);
         if (matched_addrs.size() > 1)
         {
             fprintf(stderr, "More than 1 address matches requested network(s): %s\n", json11::Json(matched_addrs).dump().c_str());
@@ -192,17 +185,21 @@ void osd_t::bind_socket()
 
     // FIXME Support multiple listening sockets
 
-    sockaddr_in addr;
+    sockaddr addr;
-    int r;
+    if (!string_to_addr(bind_address, 0, bind_port, &addr))
-    if ((r = inet_pton(AF_INET, bind_address.c_str(), &addr.sin_addr)) != 1)
     {
-        close(listen_fd);
+        throw std::runtime_error("bind address "+bind_address+" is not valid");
-        throw std::runtime_error("bind address "+bind_address+(r == 0 ? " is not valid" : ": no ipv4 support"));
     }
-    addr.sin_family = AF_INET;
 
-    addr.sin_port = htons(bind_port);
+    listen_fd = socket(addr.sa_family, SOCK_STREAM, 0);
-    if (bind(listen_fd, (sockaddr*)&addr, sizeof(addr)) < 0)
+    if (listen_fd < 0)
+    {
+        throw std::runtime_error(std::string("socket: ") + strerror(errno));
+    }
+    int enable = 1;
+    setsockopt(listen_fd, SOL_SOCKET, SO_REUSEADDR, &enable, sizeof(enable));
+
+    if (bind(listen_fd, &addr, sizeof(addr)) < 0)
     {
         close(listen_fd);
         throw std::runtime_error(std::string("bind: ") + strerror(errno));
@@ -215,7 +212,7 @@ void osd_t::bind_socket()
             close(listen_fd);
             throw std::runtime_error(std::string("getsockname: ") + strerror(errno));
         }
-        listening_port = ntohs(addr.sin_port);
+        listening_port = ntohs(((sockaddr_in*)&addr)->sin_port);
     }
     else
     {

src/osd.h

@@ -102,7 +102,7 @@ class osd_t
     bool no_rebalance = false;
     bool no_recovery = false;
     std::string bind_address;
-    int bind_port, listen_backlog;
+    int bind_port, listen_backlog = 128;
     // FIXME: Implement client queue depth limit
     int client_queue_depth = 128;
     bool allow_test_ops = false;

src/osd_secondary.cpp

@@ -178,6 +178,37 @@ void osd_t::exec_show_config(osd_op_t *cur_op)
             }
         }
     }
+    else
+    {
+#endif
+        if (req_json["data_for"].is_string())
+        {
+            auto cli = msgr.clients.at(cur_op->peer_fd);
+            auto md_it = msgr.clients_by_addr.find(req_json["data_for"].string_value());
+            if (md_it != msgr.clients_by_addr.end())
+            {
+                int md_peer_fd = md_it->second;
+                auto md_it = msgr.clients.find(md_peer_fd);
+                if (md_it != msgr.clients.end() && md_it->second->data_connection_cookie != "" &&
+                    req_json["data_cookie"].string_value() == md_it->second->data_connection_cookie)
+                {
+                    // Break previous metadata/data connections for both FDs, if present
+                    msgr.break_data_client_pair(cli);
+                    msgr.break_data_client_pair(md_it->second);
+                    // And setup the new pair
+                    cli->meta_connection_fd = md_it->second->peer_fd;
+                    md_it->second->data_connection_fd = cli->peer_fd;
+                    wire_config["data_for"] = req_json["data_for"];
+                }
+            }
+        }
+        else if (req_json["data_cookie"].is_string())
+        {
+            auto cli = msgr.clients.at(cur_op->peer_fd);
+            cli->data_connection_cookie = req_json["data_cookie"].string_value();
+        }
+#ifdef WITH_RDMA
+    }
 #endif
     if (cur_op->buf)
         free(cur_op->buf);

src/osd_test.cpp

@@ -16,6 +16,7 @@
 
 #include <stdexcept>
 
+#include "addr_util.h"
 #include "osd_ops.h"
 #include "rw_blocking.h"
 #include "test_pattern.h"
@@ -133,17 +134,14 @@ int main(int narg, char *args[])
 
 int connect_osd(const char *osd_address, int osd_port)
 {
-    struct sockaddr_in addr;
+    struct sockaddr addr;
-    int r;
+    if (!string_to_addr(osd_address, 0, osd_port, &addr))
-    if ((r = inet_pton(AF_INET, osd_address, &addr.sin_addr)) != 1)
     {
-        fprintf(stderr, "server address: %s%s\n", osd_address, r == 0 ? " is not valid" : ": no ipv4 support");
+        fprintf(stderr, "server address: %s is not valid\n", osd_address);
         return -1;
     }
-    addr.sin_family = AF_INET;
-    addr.sin_port = htons(osd_port);
 
-    int connect_fd = socket(AF_INET, SOCK_STREAM, 0);
+    int connect_fd = socket(addr.sa_family, SOCK_STREAM, 0);
     if (connect_fd < 0)
     {
         perror("socket");

src/rw_blocking.cpp

@@ -4,6 +4,8 @@
 #include <errno.h>
 #include <stdlib.h>
 #include <stdio.h>
+#include <sys/types.h>
+#include <sys/socket.h>
 
 #include "rw_blocking.h"
 
@@ -123,3 +125,41 @@ int writev_blocking(int fd, iovec *iov, int iovcnt)
     }
     return done;
 }
+
+int sendv_blocking(int fd, iovec *iov, int iovcnt, int flags)
+{
+    struct msghdr msg = { 0 };
+    int v = 0;
+    int done = 0;
+    while (v < iovcnt)
+    {
+        msg.msg_iov = iov+v;
+        msg.msg_iovlen = iovcnt-v;
+        ssize_t r = sendmsg(fd, &msg, flags);
+        if (r < 0)
+        {
+            if (errno != EAGAIN && errno != EPIPE)
+            {
+                perror("sendmsg");
+                exit(1);
+            }
+            continue;
+        }
+        done += r;
+        while (v < iovcnt)
+        {
+            if (iov[v].iov_len > r)
+            {
+                iov[v].iov_len -= r;
+                iov[v].iov_base += r;
+                break;
+            }
+            else
+            {
+                r -= iov[v].iov_len;
+                v++;
+            }
+        }
+    }
+    return done;
+}

src/rw_blocking.h

@@ -10,3 +10,4 @@ int read_blocking(int fd, void *read_buf, size_t remaining);
 int write_blocking(int fd, void *write_buf, size_t remaining);
 int readv_blocking(int fd, iovec *iov, int iovcnt);
 int writev_blocking(int fd, iovec *iov, int iovcnt);
+int sendv_blocking(int fd, iovec *iov, int iovcnt, int flags);

src/stub_bench.cpp

@@ -21,6 +21,7 @@
 
 #include <stdexcept>
 
+#include "addr_util.h"
 #include "rw_blocking.h"
 #include "osd_ops.h"
 
@@ -66,16 +67,14 @@ int main(int narg, char *args[])
 
 int connect_stub(const char *server_address, int server_port)
 {
-    struct sockaddr_in addr;
+    struct sockaddr addr;
-    int r;
+    if (!string_to_addr(server_address, 0, server_port, &addr))
-    if ((r = inet_pton(AF_INET, server_address, &addr.sin_addr)) != 1)
     {
-        fprintf(stderr, "server address: %s%s\n", server_address, r == 0 ? " is not valid" : ": no ipv4 support");
+        fprintf(stderr, "server address: %s is not valid\n", server_address);
         return -1;
     }
-    addr.sin_family = AF_INET;
+
-    addr.sin_port = htons(server_port);
+    int connect_fd = socket(addr.sa_family, SOCK_STREAM, 0);
-    int connect_fd = socket(AF_INET, SOCK_STREAM, 0);
     if (connect_fd < 0)
     {
         perror("socket");

src/stub_osd.cpp

@@ -37,10 +37,11 @@
 
 #include <stdexcept>
 
+#include "addr_util.h"
 #include "rw_blocking.h"
 #include "osd_ops.h"
 
-int bind_stub(const char *bind_address, int bind_port);
+int bind_stub(std::string bind_address, int bind_port);
 
 void run_stub(int peer_fd);
 
@@ -48,13 +49,13 @@ int main(int narg, char *args[])
 {
     int listen_fd = bind_stub("0.0.0.0", 11203);
     // Accept new connections
-    sockaddr_in addr;
+    sockaddr addr;
     socklen_t peer_addr_size = sizeof(addr);
     int peer_fd;
     while (1)
     {
         printf("stub_osd: waiting for 1 client\n");
-        peer_fd = accept(listen_fd, (sockaddr*)&addr, &peer_addr_size);
+        peer_fd = accept(listen_fd, &addr, &peer_addr_size);
         if (peer_fd == -1)
         {
             if (errno == EAGAIN)
@@ -62,9 +63,8 @@ int main(int narg, char *args[])
             else
                 throw std::runtime_error(std::string("accept: ") + strerror(errno));
         }
-        char peer_str[256];
+        printf("stub_osd: new client %d: connection from %s\n", peer_fd,
-        printf("stub_osd: new client %d: connection from %s port %d\n", peer_fd,
+            addr_to_string(addr).c_str());
-            inet_ntop(AF_INET, &addr.sin_addr, peer_str, 256), ntohs(addr.sin_port));
         int one = 1;
         setsockopt(peer_fd, SOL_TCP, TCP_NODELAY, &one, sizeof(one));
         run_stub(peer_fd);
@@ -76,11 +76,17 @@ int main(int narg, char *args[])
     return 0;
 }
 
-int bind_stub(const char *bind_address, int bind_port)
+int bind_stub(std::string bind_address, int bind_port)
 {
     int listen_backlog = 128;
 
-    int listen_fd = socket(AF_INET, SOCK_STREAM, 0);
+    sockaddr addr;
+    if (!string_to_addr(bind_address, 0, bind_port, &addr))
+    {
+        throw std::runtime_error("bind address "+bind_address+" is not valid");
+    }
+
+    int listen_fd = socket(addr.sa_family, SOCK_STREAM, 0);
     if (listen_fd < 0)
     {
         throw std::runtime_error(std::string("socket: ") + strerror(errno));
@@ -88,17 +94,7 @@ int bind_stub(const char *bind_address, int bind_port)
     int enable = 1;
     setsockopt(listen_fd, SOL_SOCKET, SO_REUSEADDR, &enable, sizeof(enable));
 
-    sockaddr_in addr;
+    if (bind(listen_fd, &addr, sizeof(addr)) < 0)
-    int r;
-    if ((r = inet_pton(AF_INET, bind_address, &addr.sin_addr)) != 1)
-    {
-        close(listen_fd);
-        throw std::runtime_error("bind address "+std::string(bind_address)+(r == 0 ? " is not valid" : ": no ipv4 support"));
-    }
-    addr.sin_family = AF_INET;
-    addr.sin_port = htons(bind_port);
-
-    if (bind(listen_fd, (sockaddr*)&addr, sizeof(addr)) < 0)
     {
         close(listen_fd);
         throw std::runtime_error(std::string("bind: ") + strerror(errno));

src/stub_uring_osd.cpp

@@ -20,11 +20,12 @@
 
 #include <stdexcept>
 
+#include "addr_util.h"
 #include "ringloop.h"
 #include "epoll_manager.h"
 #include "messenger.h"
 
-int bind_stub(const char *bind_address, int bind_port);
+int bind_stub(std::string bind_address, int bind_port);
 
 void stub_exec_op(osd_messenger_t *msgr, osd_op_t *op);
 
@@ -66,11 +67,17 @@ int main(int narg, char *args[])
     return 0;
 }
 
-int bind_stub(const char *bind_address, int bind_port)
+int bind_stub(std::string bind_address, int bind_port)
 {
     int listen_backlog = 128;
 
-    int listen_fd = socket(AF_INET, SOCK_STREAM, 0);
+    sockaddr addr;
+    if (!string_to_addr(bind_address, 0, bind_port, &addr))
+    {
+        throw std::runtime_error("bind address "+bind_address+" is not valid");
+    }
+
+    int listen_fd = socket(addr.sa_family, SOCK_STREAM, 0);
     if (listen_fd < 0)
     {
         throw std::runtime_error(std::string("socket: ") + strerror(errno));
@@ -78,17 +85,7 @@ int bind_stub(const char *bind_address, int bind_port)
     int enable = 1;
     setsockopt(listen_fd, SOL_SOCKET, SO_REUSEADDR, &enable, sizeof(enable));
 
-    sockaddr_in addr;
+    if (bind(listen_fd, &addr, sizeof(addr)) < 0)
-    int r;
-    if ((r = inet_pton(AF_INET, bind_address, &addr.sin_addr)) != 1)
-    {
-        close(listen_fd);
-        throw std::runtime_error("bind address "+std::string(bind_address)+(r == 0 ? " is not valid" : ": no ipv4 support"));
-    }
-    addr.sin_family = AF_INET;
-    addr.sin_port = htons(bind_port);
-
-    if (bind(listen_fd, (sockaddr*)&addr, sizeof(addr)) < 0)
     {
         close(listen_fd);
         throw std::runtime_error(std::string("bind: ") + strerror(errno));
@@ -113,8 +110,10 @@ void stub_exec_op(osd_messenger_t *msgr, osd_op_t *op)
     if (op->req.hdr.opcode == OSD_OP_SEC_READ)
     {
         op->reply.hdr.retval = op->req.sec_rw.len;
-        op->buf = malloc(op->req.sec_rw.len);
+        op->buf = memalign_or_die(MEM_ALIGNMENT, op->req.sec_rw.len);
         op->iov.push_back(op->buf, op->req.sec_rw.len);
+        op->reply.sec_rw.attr_len = 4;
+        op->bitmap = op->buf;
     }
     else if (op->req.hdr.opcode == OSD_OP_SEC_WRITE || op->req.hdr.opcode == OSD_OP_SEC_WRITE_STABLE)
     {

src/vitastor.pc.in

@@ -6,7 +6,7 @@ includedir=${prefix}/@CMAKE_INSTALL_INCLUDEDIR@
 
 Name: Vitastor
 Description: Vitastor client library
-Version: 0.6.10
+Version: 0.6.11
 Libs: -L${libdir} -lvitastor_client
 Cflags: -I${includedir}
 

tests/run_3osds.sh

@@ -5,6 +5,7 @@
 OSD_SIZE=${OSD_SIZE:-1024}
 PG_COUNT=${PG_COUNT:-1}
 PG_SIZE=${PG_SIZE:-3}
+PG_MINSIZE=${PG_MINSIZE:-2}
 OSD_COUNT=${OSD_COUNT:-3}
 SCHEME=${SCHEME:-ec}
 
@@ -25,9 +26,9 @@ if [ -n "$GLOBAL_CONF" ]; then
 fi
 
 if [ "$SCHEME" = "replicated" ]; then
-    $ETCDCTL put /vitastor/config/pools '{"1":{"name":"testpool","scheme":"replicated","pg_size":'$PG_SIZE',"pg_minsize":'$((PG_SIZE-1))',"pg_count":'$PG_COUNT',"failure_domain":"osd"}}'
+    $ETCDCTL put /vitastor/config/pools '{"1":{"name":"testpool","scheme":"replicated","pg_size":'$PG_SIZE',"pg_minsize":'$PG_MINSIZE',"pg_count":'$PG_COUNT',"failure_domain":"osd"}}'
 else
-    $ETCDCTL put /vitastor/config/pools '{"1":{"name":"testpool","scheme":"xor","pg_size":'$PG_SIZE',"pg_minsize":'$((PG_SIZE-1))',"parity_chunks":1,"pg_count":'$PG_COUNT',"failure_domain":"osd"}}'
+    $ETCDCTL put /vitastor/config/pools '{"1":{"name":"testpool","scheme":"xor","pg_size":'$PG_SIZE',"pg_minsize":'$PG_MINSIZE',"parity_chunks":1,"pg_count":'$PG_COUNT',"failure_domain":"osd"}}'
 fi
 
 sleep 2

tests/test_minsize_1.sh

@@ -0,0 +1,17 @@
+#!/bin/bash -ex
+
+PG_MINSIZE=1
+SCHEME=replicated
+
+. `dirname $0`/run_3osds.sh
+
+kill -INT $OSD1_PID
+kill -INT $OSD2_PID
+
+sleep 5
+
+if ! ($ETCDCTL get /vitastor/pg/state/1/ --prefix --print-value-only | jq -s -e '[ .[] | select(.state == ["active", "degraded"]) ] | length == '$PG_COUNT); then
+    format_error "FAILED: $PG_COUNT PG(s) NOT ACTIVE+DEGRADED"
+fi
+
+format_green OK