block/export: convert vhost-user-blk server to block export API
authorStefan Hajnoczi <stefanha@redhat.com>
Thu, 24 Sep 2020 15:15:47 +0000 (16:15 +0100)
committerStefan Hajnoczi <stefanha@redhat.com>
Fri, 23 Oct 2020 12:42:16 +0000 (13:42 +0100)
Use the new QAPI block exports API instead of defining our own QOM
objects.

This is a large change because the lifecycle of VuBlockDev needs to
follow BlockExportDriver. QOM properties are replaced by QAPI options
objects.

VuBlockDev is renamed VuBlkExport and contains a BlockExport field.
Several fields can be dropped since BlockExport already has equivalents.

The file names and meson build integration will be adjusted in a future
patch. libvhost-user should probably be built as a static library that
is linked into QEMU instead of as a .c file that results in duplicate
compilation.

The new command-line syntax is:

  $ qemu-storage-daemon \
      --blockdev file,node-name=drive0,filename=test.img \
      --export vhost-user-blk,node-name=drive0,id=export0,unix-socket=/tmp/vhost-user-blk.sock

Note that unix-socket is optional because we may wish to accept chardevs
too in the future.

Markus noted that supported address families are not explicit in the
QAPI schema. It is unlikely that support for more address families will
be added since file descriptor passing is required and few address
families support it. If a new address family needs to be added, then the
QAPI 'features' syntax can be used to advertize them.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Acked-by: Markus Armbruster <armbru@redhat.com>
Message-id: 20200924151549.913737-12-stefanha@redhat.com
[Skip test on big-endian host architectures because this device doesn't
support them yet (as already mentioned in a code comment).
--Stefan]
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
block/export/export.c
block/export/meson.build
block/export/vhost-user-blk-server.c
block/export/vhost-user-blk-server.h
block/meson.build
qapi/block-export.json
util/vhost-user-server.c

index f2c00d13bffd735de434ab4251c0766d4c5d3b76..bd7cac241f28b70a0030a26967476f765fc90069 100644 (file)
@@ -17,6 +17,9 @@
 #include "sysemu/block-backend.h"
 #include "block/export.h"
 #include "block/nbd.h"
+#if CONFIG_LINUX
+#include "block/export/vhost-user-blk-server.h"
+#endif
 #include "qapi/error.h"
 #include "qapi/qapi-commands-block-export.h"
 #include "qapi/qapi-events-block-export.h"
@@ -24,6 +27,9 @@
 
 static const BlockExportDriver *blk_exp_drivers[] = {
     &blk_exp_nbd,
+#if CONFIG_LINUX
+    &blk_exp_vhost_user_blk,
+#endif
 };
 
 /* Only accessed from the main thread */
index 558ef35d3829059cdbc567919972db48059a4e2a..ef3a9576f7792409c672d56068f814f4d8de02b7 100644 (file)
@@ -1 +1,2 @@
 block_ss.add(files('export.c'))
+block_ss.add(when: 'CONFIG_LINUX', if_true: files('vhost-user-blk-server.c', '../../contrib/libvhost-user/libvhost-user.c'))
index faefcfcaea1517adfa825ae2b14fa6013075fbba..3e5bd6caee8e1417eb50f50fc77a37a1866ee635 100644 (file)
@@ -11,6 +11,9 @@
  */
 #include "qemu/osdep.h"
 #include "block/block.h"
+#include "contrib/libvhost-user/libvhost-user.h"
+#include "standard-headers/linux/virtio_blk.h"
+#include "util/vhost-user-server.h"
 #include "vhost-user-blk-server.h"
 #include "qapi/error.h"
 #include "qom/object_interfaces.h"
@@ -24,7 +27,7 @@ struct virtio_blk_inhdr {
     unsigned char status;
 };
 
-typedef struct VuBlockReq {
+typedef struct VuBlkReq {
     VuVirtqElement elem;
     int64_t sector_num;
     size_t size;
@@ -32,9 +35,19 @@ typedef struct VuBlockReq {
     struct virtio_blk_outhdr out;
     VuServer *server;
     struct VuVirtq *vq;
-} VuBlockReq;
+} VuBlkReq;
 
-static void vu_block_req_complete(VuBlockReq *req)
+/* vhost user block device */
+typedef struct {
+    BlockExport export;
+    VuServer vu_server;
+    uint32_t blk_size;
+    QIOChannelSocket *sioc;
+    struct virtio_blk_config blkcfg;
+    bool writable;
+} VuBlkExport;
+
+static void vu_blk_req_complete(VuBlkReq *req)
 {
     VuDev *vu_dev = &req->server->vu_dev;
 
@@ -45,14 +58,9 @@ static void vu_block_req_complete(VuBlockReq *req)
     free(req);
 }
 
-static VuBlockDev *get_vu_block_device_by_server(VuServer *server)
-{
-    return container_of(server, VuBlockDev, vu_server);
-}
-
 static int coroutine_fn
-vu_block_discard_write_zeroes(VuBlockReq *req, struct iovec *iov,
-                              uint32_t iovcnt, uint32_t type)
+vu_blk_discard_write_zeroes(BlockBackend *blk, struct iovec *iov,
+                            uint32_t iovcnt, uint32_t type)
 {
     struct virtio_blk_discard_write_zeroes desc;
     ssize_t size = iov_to_buf(iov, iovcnt, 0, &desc, sizeof(desc));
@@ -61,16 +69,14 @@ vu_block_discard_write_zeroes(VuBlockReq *req, struct iovec *iov,
         return -EINVAL;
     }
 
-    VuBlockDev *vdev_blk = get_vu_block_device_by_server(req->server);
     uint64_t range[2] = { le64_to_cpu(desc.sector) << 9,
                           le32_to_cpu(desc.num_sectors) << 9 };
     if (type == VIRTIO_BLK_T_DISCARD) {
-        if (blk_co_pdiscard(vdev_blk->backend, range[0], range[1]) == 0) {
+        if (blk_co_pdiscard(blk, range[0], range[1]) == 0) {
             return 0;
         }
     } else if (type == VIRTIO_BLK_T_WRITE_ZEROES) {
-        if (blk_co_pwrite_zeroes(vdev_blk->backend,
-                                 range[0], range[1], 0) == 0) {
+        if (blk_co_pwrite_zeroes(blk, range[0], range[1], 0) == 0) {
             return 0;
         }
     }
@@ -78,22 +84,15 @@ vu_block_discard_write_zeroes(VuBlockReq *req, struct iovec *iov,
     return -EINVAL;
 }
 
-static int coroutine_fn vu_block_flush(VuBlockReq *req)
+static void coroutine_fn vu_blk_virtio_process_req(void *opaque)
 {
-    VuBlockDev *vdev_blk = get_vu_block_device_by_server(req->server);
-    BlockBackend *backend = vdev_blk->backend;
-    return blk_co_flush(backend);
-}
-
-static void coroutine_fn vu_block_virtio_process_req(void *opaque)
-{
-    VuBlockReq *req = opaque;
+    VuBlkReq *req = opaque;
     VuServer *server = req->server;
     VuVirtqElement *elem = &req->elem;
     uint32_t type;
 
-    VuBlockDev *vdev_blk = get_vu_block_device_by_server(server);
-    BlockBackend *backend = vdev_blk->backend;
+    VuBlkExport *vexp = container_of(server, VuBlkExport, vu_server);
+    BlockBackend *blk = vexp->export.blk;
 
     struct iovec *in_iov = elem->in_sg;
     struct iovec *out_iov = elem->out_sg;
@@ -133,16 +132,19 @@ static void coroutine_fn vu_block_virtio_process_req(void *opaque)
         bool is_write = type & VIRTIO_BLK_T_OUT;
         req->sector_num = le64_to_cpu(req->out.sector);
 
-        int64_t offset = req->sector_num * vdev_blk->blk_size;
+        if (is_write && !vexp->writable) {
+            req->in->status = VIRTIO_BLK_S_IOERR;
+            break;
+        }
+
+        int64_t offset = req->sector_num * vexp->blk_size;
         QEMUIOVector qiov;
         if (is_write) {
             qemu_iovec_init_external(&qiov, out_iov, out_num);
-            ret = blk_co_pwritev(backend, offset, qiov.size,
-                                 &qiov, 0);
+            ret = blk_co_pwritev(blk, offset, qiov.size, &qiov, 0);
         } else {
             qemu_iovec_init_external(&qiov, in_iov, in_num);
-            ret = blk_co_preadv(backend, offset, qiov.size,
-                                &qiov, 0);
+            ret = blk_co_preadv(blk, offset, qiov.size, &qiov, 0);
         }
         if (ret >= 0) {
             req->in->status = VIRTIO_BLK_S_OK;
@@ -152,7 +154,7 @@ static void coroutine_fn vu_block_virtio_process_req(void *opaque)
         break;
     }
     case VIRTIO_BLK_T_FLUSH:
-        if (vu_block_flush(req) == 0) {
+        if (blk_co_flush(blk) == 0) {
             req->in->status = VIRTIO_BLK_S_OK;
         } else {
             req->in->status = VIRTIO_BLK_S_IOERR;
@@ -169,8 +171,13 @@ static void coroutine_fn vu_block_virtio_process_req(void *opaque)
     case VIRTIO_BLK_T_DISCARD:
     case VIRTIO_BLK_T_WRITE_ZEROES: {
         int rc;
-        rc = vu_block_discard_write_zeroes(req, &elem->out_sg[1],
-                                           out_num, type);
+
+        if (!vexp->writable) {
+            req->in->status = VIRTIO_BLK_S_IOERR;
+            break;
+        }
+
+        rc = vu_blk_discard_write_zeroes(blk, &elem->out_sg[1], out_num, type);
         if (rc == 0) {
             req->in->status = VIRTIO_BLK_S_OK;
         } else {
@@ -183,22 +190,22 @@ static void coroutine_fn vu_block_virtio_process_req(void *opaque)
         break;
     }
 
-    vu_block_req_complete(req);
+    vu_blk_req_complete(req);
     return;
 
 err:
-    free(elem);
+    free(req);
 }
 
-static void vu_block_process_vq(VuDev *vu_dev, int idx)
+static void vu_blk_process_vq(VuDev *vu_dev, int idx)
 {
     VuServer *server = container_of(vu_dev, VuServer, vu_dev);
     VuVirtq *vq = vu_get_queue(vu_dev, idx);
 
     while (1) {
-        VuBlockReq *req;
+        VuBlkReq *req;
 
-        req = vu_queue_pop(vu_dev, vq, sizeof(VuBlockReq));
+        req = vu_queue_pop(vu_dev, vq, sizeof(VuBlkReq));
         if (!req) {
             break;
         }
@@ -207,26 +214,26 @@ static void vu_block_process_vq(VuDev *vu_dev, int idx)
         req->vq = vq;
 
         Coroutine *co =
-            qemu_coroutine_create(vu_block_virtio_process_req, req);
+            qemu_coroutine_create(vu_blk_virtio_process_req, req);
         qemu_coroutine_enter(co);
     }
 }
 
-static void vu_block_queue_set_started(VuDev *vu_dev, int idx, bool started)
+static void vu_blk_queue_set_started(VuDev *vu_dev, int idx, bool started)
 {
     VuVirtq *vq;
 
     assert(vu_dev);
 
     vq = vu_get_queue(vu_dev, idx);
-    vu_set_queue_handler(vu_dev, vq, started ? vu_block_process_vq : NULL);
+    vu_set_queue_handler(vu_dev, vq, started ? vu_blk_process_vq : NULL);
 }
 
-static uint64_t vu_block_get_features(VuDev *dev)
+static uint64_t vu_blk_get_features(VuDev *dev)
 {
     uint64_t features;
     VuServer *server = container_of(dev, VuServer, vu_dev);
-    VuBlockDev *vdev_blk = get_vu_block_device_by_server(server);
+    VuBlkExport *vexp = container_of(server, VuBlkExport, vu_server);
     features = 1ull << VIRTIO_BLK_F_SIZE_MAX |
                1ull << VIRTIO_BLK_F_SEG_MAX |
                1ull << VIRTIO_BLK_F_TOPOLOGY |
@@ -240,35 +247,35 @@ static uint64_t vu_block_get_features(VuDev *dev)
                1ull << VIRTIO_RING_F_EVENT_IDX |
                1ull << VHOST_USER_F_PROTOCOL_FEATURES;
 
-    if (!vdev_blk->writable) {
+    if (!vexp->writable) {
         features |= 1ull << VIRTIO_BLK_F_RO;
     }
 
     return features;
 }
 
-static uint64_t vu_block_get_protocol_features(VuDev *dev)
+static uint64_t vu_blk_get_protocol_features(VuDev *dev)
 {
     return 1ull << VHOST_USER_PROTOCOL_F_CONFIG |
            1ull << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD;
 }
 
 static int
-vu_block_get_config(VuDev *vu_dev, uint8_t *config, uint32_t len)
+vu_blk_get_config(VuDev *vu_dev, uint8_t *config, uint32_t len)
 {
+    /* TODO blkcfg must be little-endian for VIRTIO 1.0 */
     VuServer *server = container_of(vu_dev, VuServer, vu_dev);
-    VuBlockDev *vdev_blk = get_vu_block_device_by_server(server);
-    memcpy(config, &vdev_blk->blkcfg, len);
-
+    VuBlkExport *vexp = container_of(server, VuBlkExport, vu_server);
+    memcpy(config, &vexp->blkcfg, len);
     return 0;
 }
 
 static int
-vu_block_set_config(VuDev *vu_dev, const uint8_t *data,
+vu_blk_set_config(VuDev *vu_dev, const uint8_t *data,
                     uint32_t offset, uint32_t size, uint32_t flags)
 {
     VuServer *server = container_of(vu_dev, VuServer, vu_dev);
-    VuBlockDev *vdev_blk = get_vu_block_device_by_server(server);
+    VuBlkExport *vexp = container_of(server, VuBlkExport, vu_server);
     uint8_t wce;
 
     /* don't support live migration */
@@ -282,8 +289,8 @@ vu_block_set_config(VuDev *vu_dev, const uint8_t *data,
     }
 
     wce = *data;
-    vdev_blk->blkcfg.wce = wce;
-    blk_set_enable_write_cache(vdev_blk->backend, wce);
+    vexp->blkcfg.wce = wce;
+    blk_set_enable_write_cache(vexp->export.blk, wce);
     return 0;
 }
 
@@ -295,7 +302,7 @@ vu_block_set_config(VuDev *vu_dev, const uint8_t *data,
  * of vu_process_message.
  *
  */
-static int vu_block_process_msg(VuDev *dev, VhostUserMsg *vmsg, int *do_reply)
+static int vu_blk_process_msg(VuDev *dev, VhostUserMsg *vmsg, int *do_reply)
 {
     if (vmsg->request == VHOST_USER_NONE) {
         dev->panic(dev, "disconnect");
@@ -304,29 +311,29 @@ static int vu_block_process_msg(VuDev *dev, VhostUserMsg *vmsg, int *do_reply)
     return false;
 }
 
-static const VuDevIface vu_block_iface = {
-    .get_features          = vu_block_get_features,
-    .queue_set_started     = vu_block_queue_set_started,
-    .get_protocol_features = vu_block_get_protocol_features,
-    .get_config            = vu_block_get_config,
-    .set_config            = vu_block_set_config,
-    .process_msg           = vu_block_process_msg,
+static const VuDevIface vu_blk_iface = {
+    .get_features          = vu_blk_get_features,
+    .queue_set_started     = vu_blk_queue_set_started,
+    .get_protocol_features = vu_blk_get_protocol_features,
+    .get_config            = vu_blk_get_config,
+    .set_config            = vu_blk_set_config,
+    .process_msg           = vu_blk_process_msg,
 };
 
 static void blk_aio_attached(AioContext *ctx, void *opaque)
 {
-    VuBlockDev *vub_dev = opaque;
-    vhost_user_server_attach_aio_context(&vub_dev->vu_server, ctx);
+    VuBlkExport *vexp = opaque;
+    vhost_user_server_attach_aio_context(&vexp->vu_server, ctx);
 }
 
 static void blk_aio_detach(void *opaque)
 {
-    VuBlockDev *vub_dev = opaque;
-    vhost_user_server_detach_aio_context(&vub_dev->vu_server);
+    VuBlkExport *vexp = opaque;
+    vhost_user_server_detach_aio_context(&vexp->vu_server);
 }
 
 static void
-vu_block_initialize_config(BlockDriverState *bs,
+vu_blk_initialize_config(BlockDriverState *bs,
                            struct virtio_blk_config *config, uint32_t blk_size)
 {
     config->capacity = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
@@ -343,290 +350,67 @@ vu_block_initialize_config(BlockDriverState *bs,
     config->max_write_zeroes_seg = 1;
 }
 
-static VuBlockDev *vu_block_init(VuBlockDev *vu_block_device, Error **errp)
+static void vu_blk_exp_request_shutdown(BlockExport *exp)
 {
+    VuBlkExport *vexp = container_of(exp, VuBlkExport, export);
 
-    BlockBackend *blk;
-    Error *local_error = NULL;
-    const char *node_name = vu_block_device->node_name;
-    bool writable = vu_block_device->writable;
-    uint64_t perm = BLK_PERM_CONSISTENT_READ;
-    int ret;
-
-    AioContext *ctx;
-
-    BlockDriverState *bs = bdrv_lookup_bs(node_name, node_name, &local_error);
-
-    if (!bs) {
-        error_propagate(errp, local_error);
-        return NULL;
-    }
-
-    if (bdrv_is_read_only(bs)) {
-        writable = false;
-    }
-
-    if (writable) {
-        perm |= BLK_PERM_WRITE;
-    }
-
-    ctx = bdrv_get_aio_context(bs);
-    aio_context_acquire(ctx);
-    bdrv_invalidate_cache(bs, NULL);
-    aio_context_release(ctx);
-
-    /*
-     * Don't allow resize while the vhost user server is running,
-     * otherwise we don't care what happens with the node.
-     */
-    blk = blk_new(bdrv_get_aio_context(bs), perm,
-                  BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
-                  BLK_PERM_WRITE | BLK_PERM_GRAPH_MOD);
-    ret = blk_insert_bs(blk, bs, errp);
-
-    if (ret < 0) {
-        goto fail;
-    }
-
-    blk_set_enable_write_cache(blk, false);
-
-    blk_set_allow_aio_context_change(blk, true);
-
-    vu_block_device->blkcfg.wce = 0;
-    vu_block_device->backend = blk;
-    if (!vu_block_device->blk_size) {
-        vu_block_device->blk_size = BDRV_SECTOR_SIZE;
-    }
-    vu_block_device->blkcfg.blk_size = vu_block_device->blk_size;
-    blk_set_guest_block_size(blk, vu_block_device->blk_size);
-    vu_block_initialize_config(bs, &vu_block_device->blkcfg,
-                                   vu_block_device->blk_size);
-    return vu_block_device;
-
-fail:
-    blk_unref(blk);
-    return NULL;
-}
-
-static void vu_block_deinit(VuBlockDev *vu_block_device)
-{
-    if (vu_block_device->backend) {
-        blk_remove_aio_context_notifier(vu_block_device->backend, blk_aio_attached,
-                                        blk_aio_detach, vu_block_device);
-    }
-
-    blk_unref(vu_block_device->backend);
+    vhost_user_server_stop(&vexp->vu_server);
 }
 
-static void vhost_user_blk_server_stop(VuBlockDev *vu_block_device)
+static int vu_blk_exp_create(BlockExport *exp, BlockExportOptions *opts,
+                             Error **errp)
 {
-    vhost_user_server_stop(&vu_block_device->vu_server);
-    vu_block_deinit(vu_block_device);
-}
-
-static void vhost_user_blk_server_start(VuBlockDev *vu_block_device,
-                                        Error **errp)
-{
-    AioContext *ctx;
-    SocketAddress *addr = vu_block_device->addr;
-
-    if (!vu_block_init(vu_block_device, errp)) {
-        return;
-    }
-
-    ctx = bdrv_get_aio_context(blk_bs(vu_block_device->backend));
-
-    if (!vhost_user_server_start(&vu_block_device->vu_server, addr, ctx,
-                                 VHOST_USER_BLK_MAX_QUEUES, &vu_block_iface,
-                                 errp)) {
-        goto error;
-    }
-
-    blk_add_aio_context_notifier(vu_block_device->backend, blk_aio_attached,
-                                 blk_aio_detach, vu_block_device);
-    vu_block_device->running = true;
-    return;
-
- error:
-    vu_block_deinit(vu_block_device);
-}
-
-static bool vu_prop_modifiable(VuBlockDev *vus, Error **errp)
-{
-    if (vus->running) {
-            error_setg(errp, "The property can't be modified "
-                       "while the server is running");
-            return false;
-    }
-    return true;
-}
-
-static void vu_set_node_name(Object *obj, const char *value, Error **errp)
-{
-    VuBlockDev *vus = VHOST_USER_BLK_SERVER(obj);
-
-    if (!vu_prop_modifiable(vus, errp)) {
-        return;
-    }
-
-    if (vus->node_name) {
-        g_free(vus->node_name);
-    }
-
-    vus->node_name = g_strdup(value);
-}
-
-static char *vu_get_node_name(Object *obj, Error **errp)
-{
-    VuBlockDev *vus = VHOST_USER_BLK_SERVER(obj);
-    return g_strdup(vus->node_name);
-}
-
-static void free_socket_addr(SocketAddress *addr)
-{
-        g_free(addr->u.q_unix.path);
-        g_free(addr);
-}
-
-static void vu_set_unix_socket(Object *obj, const char *value,
-                               Error **errp)
-{
-    VuBlockDev *vus = VHOST_USER_BLK_SERVER(obj);
-
-    if (!vu_prop_modifiable(vus, errp)) {
-        return;
-    }
-
-    if (vus->addr) {
-        free_socket_addr(vus->addr);
-    }
-
-    SocketAddress *addr = g_new0(SocketAddress, 1);
-    addr->type = SOCKET_ADDRESS_TYPE_UNIX;
-    addr->u.q_unix.path = g_strdup(value);
-    vus->addr = addr;
-}
-
-static char *vu_get_unix_socket(Object *obj, Error **errp)
-{
-    VuBlockDev *vus = VHOST_USER_BLK_SERVER(obj);
-    return g_strdup(vus->addr->u.q_unix.path);
-}
-
-static bool vu_get_block_writable(Object *obj, Error **errp)
-{
-    VuBlockDev *vus = VHOST_USER_BLK_SERVER(obj);
-    return vus->writable;
-}
-
-static void vu_set_block_writable(Object *obj, bool value, Error **errp)
-{
-    VuBlockDev *vus = VHOST_USER_BLK_SERVER(obj);
-
-    if (!vu_prop_modifiable(vus, errp)) {
-            return;
-    }
-
-    vus->writable = value;
-}
-
-static void vu_get_blk_size(Object *obj, Visitor *v, const char *name,
-                            void *opaque, Error **errp)
-{
-    VuBlockDev *vus = VHOST_USER_BLK_SERVER(obj);
-    uint32_t value = vus->blk_size;
-
-    visit_type_uint32(v, name, &value, errp);
-}
-
-static void vu_set_blk_size(Object *obj, Visitor *v, const char *name,
-                            void *opaque, Error **errp)
-{
-    VuBlockDev *vus = VHOST_USER_BLK_SERVER(obj);
-
+    VuBlkExport *vexp = container_of(exp, VuBlkExport, export);
+    BlockExportOptionsVhostUserBlk *vu_opts = &opts->u.vhost_user_blk;
     Error *local_err = NULL;
-    uint32_t value;
+    uint64_t logical_block_size;
 
-    if (!vu_prop_modifiable(vus, errp)) {
-            return;
-    }
+    vexp->writable = opts->writable;
+    vexp->blkcfg.wce = 0;
 
-    visit_type_uint32(v, name, &value, &local_err);
-    if (local_err) {
-        goto out;
+    if (vu_opts->has_logical_block_size) {
+        logical_block_size = vu_opts->logical_block_size;
+    } else {
+        logical_block_size = BDRV_SECTOR_SIZE;
     }
-
-    check_block_size(object_get_typename(obj), name, value, &local_err);
+    check_block_size(exp->id, "logical-block-size", logical_block_size,
+                     &local_err);
     if (local_err) {
-        goto out;
+        error_propagate(errp, local_err);
+        return -EINVAL;
     }
+    vexp->blk_size = logical_block_size;
+    blk_set_guest_block_size(exp->blk, logical_block_size);
+    vu_blk_initialize_config(blk_bs(exp->blk), &vexp->blkcfg,
+                               logical_block_size);
 
-    vus->blk_size = value;
-
-out:
-    error_propagate(errp, local_err);
-}
-
-static void vhost_user_blk_server_instance_finalize(Object *obj)
-{
-    VuBlockDev *vub = VHOST_USER_BLK_SERVER(obj);
+    blk_set_allow_aio_context_change(exp->blk, true);
+    blk_add_aio_context_notifier(exp->blk, blk_aio_attached, blk_aio_detach,
+                                 vexp);
 
-    vhost_user_blk_server_stop(vub);
-
-    /*
-     * Unlike object_property_add_str, object_class_property_add_str
-     * doesn't have a release method. Thus manual memory freeing is
-     * needed.
-     */
-    free_socket_addr(vub->addr);
-    g_free(vub->node_name);
-}
-
-static void vhost_user_blk_server_complete(UserCreatable *obj, Error **errp)
-{
-    VuBlockDev *vub = VHOST_USER_BLK_SERVER(obj);
+    if (!vhost_user_server_start(&vexp->vu_server, vu_opts->addr, exp->ctx,
+                                 VHOST_USER_BLK_MAX_QUEUES, &vu_blk_iface,
+                                 errp)) {
+        blk_remove_aio_context_notifier(exp->blk, blk_aio_attached,
+                                        blk_aio_detach, vexp);
+        return -EADDRNOTAVAIL;
+    }
 
-    vhost_user_blk_server_start(vub, errp);
+    return 0;
 }
 
-static void vhost_user_blk_server_class_init(ObjectClass *klass,
-                                             void *class_data)
+static void vu_blk_exp_delete(BlockExport *exp)
 {
-    UserCreatableClass *ucc = USER_CREATABLE_CLASS(klass);
-    ucc->complete = vhost_user_blk_server_complete;
+    VuBlkExport *vexp = container_of(exp, VuBlkExport, export);
 
-    object_class_property_add_bool(klass, "writable",
-                                   vu_get_block_writable,
-                                   vu_set_block_writable);
-
-    object_class_property_add_str(klass, "node-name",
-                                  vu_get_node_name,
-                                  vu_set_node_name);
-
-    object_class_property_add_str(klass, "unix-socket",
-                                  vu_get_unix_socket,
-                                  vu_set_unix_socket);
-
-    object_class_property_add(klass, "logical-block-size", "uint32",
-                              vu_get_blk_size, vu_set_blk_size,
-                              NULL, NULL);
+    blk_remove_aio_context_notifier(exp->blk, blk_aio_attached, blk_aio_detach,
+                                    vexp);
 }
 
-static const TypeInfo vhost_user_blk_server_info = {
-    .name = TYPE_VHOST_USER_BLK_SERVER,
-    .parent = TYPE_OBJECT,
-    .instance_size = sizeof(VuBlockDev),
-    .instance_finalize = vhost_user_blk_server_instance_finalize,
-    .class_init = vhost_user_blk_server_class_init,
-    .interfaces = (InterfaceInfo[]) {
-        {TYPE_USER_CREATABLE},
-        {}
-    },
+const BlockExportDriver blk_exp_vhost_user_blk = {
+    .type               = BLOCK_EXPORT_TYPE_VHOST_USER_BLK,
+    .instance_size      = sizeof(VuBlkExport),
+    .create             = vu_blk_exp_create,
+    .delete             = vu_blk_exp_delete,
+    .request_shutdown   = vu_blk_exp_request_shutdown,
 };
-
-static void vhost_user_blk_server_register_types(void)
-{
-    type_register_static(&vhost_user_blk_server_info);
-}
-
-type_init(vhost_user_blk_server_register_types)
index f06f37c4c83b813af848377be4f1820fe37ebf35..fcf46fc8a5fce9b362a35a8a6837ec3a6bb80cb2 100644 (file)
 
 #ifndef VHOST_USER_BLK_SERVER_H
 #define VHOST_USER_BLK_SERVER_H
-#include "util/vhost-user-server.h"
 
-typedef struct VuBlockDev VuBlockDev;
-#define TYPE_VHOST_USER_BLK_SERVER "vhost-user-blk-server"
-#define VHOST_USER_BLK_SERVER(obj) \
-   OBJECT_CHECK(VuBlockDev, obj, TYPE_VHOST_USER_BLK_SERVER)
+#include "block/export.h"
 
-/* vhost user block device */
-struct VuBlockDev {
-    Object parent_obj;
-    char *node_name;
-    SocketAddress *addr;
-    AioContext *ctx;
-    VuServer vu_server;
-    bool running;
-    uint32_t blk_size;
-    BlockBackend *backend;
-    QIOChannelSocket *sioc;
-    QTAILQ_ENTRY(VuBlockDev) next;
-    struct virtio_blk_config blkcfg;
-    bool writable;
-};
+/* For block/export/export.c */
+extern const BlockExportDriver blk_exp_vhost_user_blk;
 
 #endif /* VHOST_USER_BLK_SERVER_H */
index 6e6c1dc479b88e11b3d66df54b4b4c1cb8110d1e..78e8b252327a6bd1f9b23eacc92e66eefb33e667 100644 (file)
@@ -60,7 +60,6 @@ block_ss.add(when: 'CONFIG_WIN32', if_true: files('file-win32.c', 'win32-aio.c')
 block_ss.add(when: 'CONFIG_POSIX', if_true: [files('file-posix.c'), coref, iokit])
 block_ss.add(when: 'CONFIG_LIBISCSI', if_true: files('iscsi-opts.c'))
 block_ss.add(when: 'CONFIG_LINUX', if_true: files('nvme.c'))
-block_ss.add(when: 'CONFIG_LINUX', if_true: files('export/vhost-user-blk-server.c', '../contrib/libvhost-user/libvhost-user.c'))
 block_ss.add(when: 'CONFIG_REPLICATION', if_true: files('replication.c'))
 block_ss.add(when: 'CONFIG_SHEEPDOG', if_true: files('sheepdog.c'))
 block_ss.add(when: ['CONFIG_LINUX_AIO', libaio], if_true: files('linux-aio.c'))
index 65804834d905eca88bb661afdce7f2dd91e0704c..a793e34af97d167d9f09486ff8466fe5c6317f70 100644 (file)
   'data': { '*name': 'str', '*description': 'str',
             '*bitmap': 'str' } }
 
+##
+# @BlockExportOptionsVhostUserBlk:
+#
+# A vhost-user-blk block export.
+#
+# @addr: The vhost-user socket on which to listen. Both 'unix' and 'fd'
+#        SocketAddress types are supported. Passed fds must be UNIX domain
+#        sockets.
+# @logical-block-size: Logical block size in bytes. Defaults to 512 bytes.
+#
+# Since: 5.2
+##
+{ 'struct': 'BlockExportOptionsVhostUserBlk',
+  'data': { 'addr': 'SocketAddress', '*logical-block-size': 'size' } }
+
 ##
 # @NbdServerAddOptions:
 #
 # An enumeration of block export types
 #
 # @nbd: NBD export
+# @vhost-user-blk: vhost-user-blk export (since 5.2)
 #
 # Since: 4.2
 ##
 { 'enum': 'BlockExportType',
-  'data': [ 'nbd' ] }
+  'data': [ 'nbd', 'vhost-user-blk' ] }
 
 ##
 # @BlockExportOptions:
             '*writethrough': 'bool' },
   'discriminator': 'type',
   'data': {
-      'nbd': 'BlockExportOptionsNbd'
+      'nbd': 'BlockExportOptionsNbd',
+      'vhost-user-blk': 'BlockExportOptionsVhostUserBlk'
    } }
 
 ##
index c448800e587e26606ce669d168f38e6b8eb6b4c8..516999b38ac6abad1bd37f1e2aea396c85b8caeb 100644 (file)
@@ -408,7 +408,15 @@ bool vhost_user_server_start(VuServer *server,
                              Error **errp)
 {
     QEMUBH *bh;
-    QIONetListener *listener = qio_net_listener_new();
+    QIONetListener *listener;
+
+    if (socket_addr->type != SOCKET_ADDRESS_TYPE_UNIX &&
+        socket_addr->type != SOCKET_ADDRESS_TYPE_FD) {
+        error_setg(errp, "Only socket address types 'unix' and 'fd' are supported");
+        return false;
+    }
+
+    listener = qio_net_listener_new();
     if (qio_net_listener_open_sync(listener, socket_addr, 1,
                                    errp) < 0) {
         object_unref(OBJECT(listener));