Merge tag 'for-upstream' of https://repo.or.cz/qemu/kevin into staging

Block layer patches - Managing inactive nodes (enables QSD migration with shared storage) - Fix swapped values for BLOCK_IO_ERROR 'device' and 'qom-path' - vpc: Read images exported from Azure correctly - scripts/qemu-gdb: Support coroutine dumps in coredumps - Minor cleanups # -----BEGIN PGP SIGNATURE----- # # iQJFBAABCAAvFiEE3D3rFZqa+V09dFb+fwmycsiPL9YFAmek34IRHGt3b2xmQHJl # ZGhhdC5jb20ACgkQfwmycsiPL9bDpxAAnTvwmdazAXG0g9GzqvrEB/+6rStjAsqE # 9MTWV4WxyN41d0RXxN8CYKb8CXSiTRyw6r3CSGNYEI2eShe9e934PriSkZm41HyX # n9Yh5YxqGZqitzvPtx62Ii/1KG+PcjQbfHuK1p4+rlKa0yQ2eGlio1JIIrZrCkBZ # ikZcQUrhIyD0XV8hTQ2+Ysa+ZN6itjnlTQIG3gS3m8f8WR7kyUXD8YFMQFJFyjVx # NrAIpLnc/ln9+5PZR9tje8U7XEn2KCgI5pgGaQnrd0h0G1H4ig8ogzYYnKTLhjU/ # AmQpS8np8Tyg6S1UZTiekEq0VuAhThEQc5b3sGbmHWH/R2ABMStyf18oCBAkPzZ7 # s6h+3XzTKKY2Q5Q3ZG/ANkUJjTNBhdj1fcaARvbSWsqsuk5CWX/I3jzvgihFtCSs # eGu+b/bLeW6P7hu4qPHBcgLHuB1Fc7Rd2t4BoIGM1wcO2CeC9DzUKOiIMZOEJIh0 # GGqCkEWDHgckDTakD4/vSqm0UDKt6FSlQC9ga/ILBY3IB5HpHoArY58selymy28i # X7MgAvbjdsmNuUuXDZZOiObcFt3j8jlmwPJpPyzXPQIiPX1RXeBPRhVAEeZCKn6Z # tfHr72SJdMeVOGXVTvOrJ2iW+4g03rPdmkDFCUhpOwo62RODq7ahvCIXsNf3nEFR # rSB3T1M/8EM= # =iQLP # -----END PGP SIGNATURE----- # gpg: Signature made Thu 06 Feb 2025 11:12:50 EST # gpg: using RSA key DC3DEB159A9AF95D3D7456FE7F09B272C88F2FD6 # gpg: issuer "kwolf@redhat.com" # gpg: Good signature from "Kevin Wolf <kwolf@redhat.com>" [full] # Primary key fingerprint: DC3D EB15 9A9A F95D 3D74 56FE 7F09 B272 C88F 2FD6 * tag 'for-upstream' of https://repo.or.cz/qemu/kevin: (25 commits) block: remove unused BLOCK_OP_TYPE_DATAPLANE iotests: Add (NBD-based) tests for inactive nodes iotests: Add qsd-migrate case iotests: Add filter_qtest() nbd/server: Support inactive nodes block/export: Add option to allow export of inactive nodes block: Drain nodes before inactivating them block/export: Don't ignore image activation error in blk_exp_add() block: Support inactive nodes in blk_insert_bs() block: Add blockdev-set-active QMP command block: Add option to create inactive nodes block: Fix crash on block_resize on inactive node block: Don't attach inactive child to active node migration/block-active: Remove global active flag block: Inactivate external snapshot overlays when necessary block: Allow inactivating already inactive nodes block: Add 'active' field to BlockDeviceInfo block-backend: Fix argument order when calling 'qapi_event_send_block_io_error()' scripts/qemu-gdb: Support coroutine dumps in coredumps scripts/qemu-gdb: Simplify fs_base fetching for coroutines ... Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
2026-07-06 00:20:22 -07:00 · 2025-02-10 13:25:36 -05:00
parent 4f1d018512 fc4e394b28
commit f2ec48fefd
35 changed files with 1133 additions and 166 deletions
@@ -1573,6 +1573,10 @@ static void update_flags_from_options(int *flags, QemuOpts *opts)
    if (qemu_opt_get_bool_del(opts, BDRV_OPT_AUTO_READ_ONLY, false)) {
        *flags |= BDRV_O_AUTO_RDONLY;
    }
+
+    if (!qemu_opt_get_bool_del(opts, BDRV_OPT_ACTIVE, true)) {
+        *flags |= BDRV_O_INACTIVE;
+    }
 }

 static void update_options_from_flags(QDict *options, int flags)
@@ -1799,6 +1803,11 @@ QemuOptsList bdrv_runtime_opts = {
            .type = QEMU_OPT_BOOL,
            .help = "Ignore flush requests",
        },
+        {
+            .name = BDRV_OPT_ACTIVE,
+            .type = QEMU_OPT_BOOL,
+            .help = "Node is activated",
+        },
        {
            .name = BDRV_OPT_READ_ONLY,
            .type = QEMU_OPT_BOOL,
@@ -3077,6 +3086,13 @@ bdrv_attach_child_common(BlockDriverState *child_bs,
    assert(child_class->get_parent_desc);
    GLOBAL_STATE_CODE();

+    if (bdrv_is_inactive(child_bs) && (perm & ~BLK_PERM_CONSISTENT_READ)) {
+        g_autofree char *perm_names = bdrv_perm_names(perm);
+        error_setg(errp, "Permission '%s' unavailable on inactive node",
+                   perm_names);
+        return NULL;
+    }
+
    new_child = g_new(BdrvChild, 1);
    *new_child = (BdrvChild) {
        .bs             = NULL,
@@ -3183,6 +3199,11 @@ bdrv_attach_child_noperm(BlockDriverState *parent_bs,
                   child_bs->node_name, child_name, parent_bs->node_name);
        return NULL;
    }
+    if (bdrv_is_inactive(child_bs) && !bdrv_is_inactive(parent_bs)) {
+        error_setg(errp, "Inactive '%s' can't be a %s child of active '%s'",
+                   child_bs->node_name, child_name, parent_bs->node_name);
+        return NULL;
+    }

    bdrv_get_cumulative_perm(parent_bs, &perm, &shared_perm);
    bdrv_child_perm(parent_bs, child_bs, NULL, child_role, NULL,
@@ -6824,6 +6845,10 @@ void bdrv_init_with_whitelist(void)
    bdrv_init();
 }

+bool bdrv_is_inactive(BlockDriverState *bs) {
+    return bs->open_flags & BDRV_O_INACTIVE;
+}
+
 int bdrv_activate(BlockDriverState *bs, Error **errp)
 {
    BdrvChild *child, *parent;
@@ -6955,7 +6980,8 @@ bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
    return false;
 }

-static int GRAPH_RDLOCK bdrv_inactivate_recurse(BlockDriverState *bs)
+static int GRAPH_RDLOCK
+bdrv_inactivate_recurse(BlockDriverState *bs, bool top_level)
 {
    BdrvChild *child, *parent;
    int ret;
@@ -6973,7 +6999,14 @@ static int GRAPH_RDLOCK bdrv_inactivate_recurse(BlockDriverState *bs)
        return 0;
    }

-    assert(!(bs->open_flags & BDRV_O_INACTIVE));
+    /*
+     * Inactivating an already inactive node on user request is harmless, but if
+     * a child is already inactive before its parent, that's bad.
+     */
+    if (bs->open_flags & BDRV_O_INACTIVE) {
+        assert(top_level);
+        return 0;
+    }

    /* Inactivate this node */
    if (bs->drv->bdrv_inactivate) {
@@ -6999,7 +7032,9 @@ static int GRAPH_RDLOCK bdrv_inactivate_recurse(BlockDriverState *bs)
        return -EPERM;
    }

+    bdrv_drained_begin(bs);
    bs->open_flags |= BDRV_O_INACTIVE;
+    bdrv_drained_end(bs);

    /*
     * Update permissions, they may differ for inactive nodes.
@@ -7010,7 +7045,7 @@ static int GRAPH_RDLOCK bdrv_inactivate_recurse(BlockDriverState *bs)

    /* Recursively inactivate children */
    QLIST_FOREACH(child, &bs->children, next) {
-        ret = bdrv_inactivate_recurse(child->bs);
+        ret = bdrv_inactivate_recurse(child->bs, false);
        if (ret < 0) {
            return ret;
        }
@@ -7019,6 +7054,27 @@ static int GRAPH_RDLOCK bdrv_inactivate_recurse(BlockDriverState *bs)
    return 0;
 }

+int bdrv_inactivate(BlockDriverState *bs, Error **errp)
+{
+    int ret;
+
+    GLOBAL_STATE_CODE();
+    GRAPH_RDLOCK_GUARD_MAINLOOP();
+
+    if (bdrv_has_bds_parent(bs, true)) {
+        error_setg(errp, "Node has active parent node");
+        return -EPERM;
+    }
+
+    ret = bdrv_inactivate_recurse(bs, true);
+    if (ret < 0) {
+        error_setg_errno(errp, -ret, "Failed to inactivate node");
+        return ret;
+    }
+
+    return 0;
+}
+
 int bdrv_inactivate_all(void)
 {
    BlockDriverState *bs = NULL;
@@ -7035,7 +7091,7 @@ int bdrv_inactivate_all(void)
        if (bdrv_has_bds_parent(bs, false)) {
            continue;
        }
-        ret = bdrv_inactivate_recurse(bs);
+        ret = bdrv_inactivate_recurse(bs, true);
        if (ret < 0) {
            bdrv_next_cleanup(&it);
            break;
@@ -253,7 +253,7 @@ static bool blk_can_inactivate(BlockBackend *blk)
     * guest.  For block job BBs that satisfy this, we can just allow
     * it.  This is the case for mirror job source, which is required
     * by libvirt non-shared block migration. */
-    if (!(blk->perm & (BLK_PERM_WRITE | BLK_PERM_WRITE_UNCHANGED))) {
+    if (!(blk->perm & ~BLK_PERM_CONSISTENT_READ)) {
        return true;
    }

@@ -900,14 +900,24 @@ void blk_remove_bs(BlockBackend *blk)
 int blk_insert_bs(BlockBackend *blk, BlockDriverState *bs, Error **errp)
 {
    ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
+    uint64_t perm, shared_perm;

    GLOBAL_STATE_CODE();
    bdrv_ref(bs);
    bdrv_graph_wrlock();
+
+    if ((bs->open_flags & BDRV_O_INACTIVE) && blk_can_inactivate(blk)) {
+        blk->disable_perm = true;
+        perm = 0;
+        shared_perm = BLK_PERM_ALL;
+    } else {
+        perm = blk->perm;
+        shared_perm = blk->shared_perm;
+    }
+
    blk->root = bdrv_root_attach_child(bs, "root", &child_root,
                                       BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY,
-                                       blk->perm, blk->shared_perm,
-                                       blk, errp);
+                                       perm, shared_perm, blk, errp);
    bdrv_graph_wrunlock();
    if (blk->root == NULL) {
        return -EPERM;
@@ -1019,6 +1029,10 @@ DeviceState *blk_get_attached_dev(BlockBackend *blk)
    return blk->dev;
 }

+/*
+ * The caller is responsible for releasing the value returned
+ * with g_free() after use.
+ */
 static char *blk_get_attached_dev_id_or_path(BlockBackend *blk, bool want_id)
 {
    DeviceState *dev = blk->dev;
@@ -1033,15 +1047,15 @@ static char *blk_get_attached_dev_id_or_path(BlockBackend *blk, bool want_id)
    return object_get_canonical_path(OBJECT(dev)) ?: g_strdup("");
 }

-/*
- * Return the qdev ID, or if no ID is assigned the QOM path, of the block
- * device attached to the BlockBackend.
- */
 char *blk_get_attached_dev_id(BlockBackend *blk)
 {
    return blk_get_attached_dev_id_or_path(blk, true);
 }

+/*
+ * The caller is responsible for releasing the value returned
+ * with g_free() after use.
+ */
 static char *blk_get_attached_dev_path(BlockBackend *blk)
 {
    return blk_get_attached_dev_id_or_path(blk, false);
@@ -2134,10 +2148,10 @@ static void send_qmp_error_event(BlockBackend *blk,
 {
    IoOperationType optype;
    BlockDriverState *bs = blk_bs(blk);
+    g_autofree char *path = blk_get_attached_dev_path(blk);

    optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
-    qapi_event_send_block_io_error(blk_name(blk),
-                                   blk_get_attached_dev_path(blk),
+    qapi_event_send_block_io_error(path, blk_name(blk),
                                   bs ? bdrv_get_node_name(bs) : NULL, optype,
                                   action, blk_iostatus_is_enabled(blk),
                                   error == ENOSPC, strerror(error));
@@ -75,6 +75,7 @@ static const BlockExportDriver *blk_exp_find_driver(BlockExportType type)
 BlockExport *blk_exp_add(BlockExportOptions *export, Error **errp)
 {
    bool fixed_iothread = export->has_fixed_iothread && export->fixed_iothread;
+    bool allow_inactive = export->has_allow_inactive && export->allow_inactive;
    const BlockExportDriver *drv;
    BlockExport *exp = NULL;
    BlockDriverState *bs;
@@ -138,14 +139,25 @@ BlockExport *blk_exp_add(BlockExportOptions *export, Error **errp)
        }
    }

-    /*
-     * Block exports are used for non-shared storage migration. Make sure
-     * that BDRV_O_INACTIVE is cleared and the image is ready for write
-     * access since the export could be available before migration handover.
-     * ctx was acquired in the caller.
-     */
    bdrv_graph_rdlock_main_loop();
-    bdrv_activate(bs, NULL);
+    if (allow_inactive) {
+        if (!drv->supports_inactive) {
+            error_setg(errp, "Export type does not support inactive exports");
+            bdrv_graph_rdunlock_main_loop();
+            goto fail;
+        }
+    } else {
+        /*
+         * Block exports are used for non-shared storage migration. Make sure
+         * that BDRV_O_INACTIVE is cleared and the image is ready for write
+         * access since the export could be available before migration handover.
+         */
+        ret = bdrv_activate(bs, errp);
+        if (ret < 0) {
+            bdrv_graph_rdunlock_main_loop();
+            goto fail;
+        }
+    }
    bdrv_graph_rdunlock_main_loop();

    perm = BLK_PERM_CONSISTENT_READ;
@@ -158,6 +170,9 @@ BlockExport *blk_exp_add(BlockExportOptions *export, Error **errp)
    if (!fixed_iothread) {
        blk_set_allow_aio_context_change(blk, true);
    }
+    if (allow_inactive) {
+        blk_set_force_allow_inactivate(blk);
+    }

    ret = blk_insert_bs(blk, bs, errp);
    if (ret < 0) {
@@ -630,11 +630,12 @@ static void print_block_info(Monitor *mon, BlockInfo *info,
    }

    if (inserted) {
-        monitor_printf(mon, ": %s (%s%s%s)\n",
+        monitor_printf(mon, ": %s (%s%s%s%s)\n",
                       inserted->file,
                       inserted->drv,
                       inserted->ro ? ", read-only" : "",
-                       inserted->encrypted ? ", encrypted" : "");
+                       inserted->encrypted ? ", encrypted" : "",
+                       inserted->active ? "" : ", inactive");
    } else {
        monitor_printf(mon, ": [not inserted]\n");
    }
@@ -63,6 +63,7 @@ BlockDeviceInfo *bdrv_block_device_info(BlockBackend *blk,
    info->file                   = g_strdup(bs->filename);
    info->ro                     = bdrv_is_read_only(bs);
    info->drv                    = g_strdup(bs->drv->format_name);
+    info->active                 = !bdrv_is_inactive(bs);
    info->encrypted              = bs->encrypted;

    info->cache = g_new(BlockdevCacheInfo, 1);
@@ -576,7 +576,6 @@ static void replication_start(ReplicationState *rs, ReplicationMode mode,
            return;
        }
        bdrv_op_block_all(top_bs, s->blocker);
-        bdrv_op_unblock(top_bs, BLOCK_OP_TYPE_DATAPLANE, s->blocker);

        bdrv_graph_wrunlock();

@@ -216,6 +216,39 @@ static void vpc_parse_options(BlockDriverState *bs, QemuOpts *opts,
    }
 }

+/*
+ * Microsoft Virtual PC and Microsoft Hyper-V produce and read
+ * VHD image sizes differently.  VPC will rely on CHS geometry,
+ * while Hyper-V and disk2vhd use the size specified in the footer.
+ *
+ * We use a couple of approaches to try and determine the correct method:
+ * look at the Creator App field, and look for images that have CHS
+ * geometry that is the maximum value.
+ *
+ * If the CHS geometry is the maximum CHS geometry, then we assume that
+ * the size is the footer->current_size to avoid truncation.  Otherwise,
+ * we follow the table based on footer->creator_app:
+ *
+ *  Known creator apps:
+ *      'vpc '  :  CHS              Virtual PC (uses disk geometry)
+ *      'qemu'  :  CHS              QEMU (uses disk geometry)
+ *      'qem2'  :  current_size     QEMU (uses current_size)
+ *      'win '  :  current_size     Hyper-V
+ *      'd2v '  :  current_size     Disk2vhd
+ *      'tap\0' :  current_size     XenServer
+ *      'CTXS'  :  current_size     XenConverter
+ *      'wa\0\0':  current_size     Azure
+ *
+ *  The user can override the table values via drive options, however
+ *  even with an override we will still use current_size for images
+ *  that have CHS geometry of the maximum size.
+ */
+static bool vpc_ignore_current_size(VHDFooter *footer)
+{
+    return !strncmp(footer->creator_app, "vpc ", 4) ||
+           !strncmp(footer->creator_app, "qemu", 4);
+}
+
 static int vpc_open(BlockDriverState *bs, QDict *options, int flags,
                    Error **errp)
 {
@@ -304,36 +337,8 @@ static int vpc_open(BlockDriverState *bs, QDict *options, int flags,
    bs->total_sectors = (int64_t)
        be16_to_cpu(footer->cyls) * footer->heads * footer->secs_per_cyl;

-    /* Microsoft Virtual PC and Microsoft Hyper-V produce and read
-     * VHD image sizes differently.  VPC will rely on CHS geometry,
-     * while Hyper-V and disk2vhd use the size specified in the footer.
-     *
-     * We use a couple of approaches to try and determine the correct method:
-     * look at the Creator App field, and look for images that have CHS
-     * geometry that is the maximum value.
-     *
-     * If the CHS geometry is the maximum CHS geometry, then we assume that
-     * the size is the footer->current_size to avoid truncation.  Otherwise,
-     * we follow the table based on footer->creator_app:
-     *
-     *  Known creator apps:
-     *      'vpc '  :  CHS              Virtual PC (uses disk geometry)
-     *      'qemu'  :  CHS              QEMU (uses disk geometry)
-     *      'qem2'  :  current_size     QEMU (uses current_size)
-     *      'win '  :  current_size     Hyper-V
-     *      'd2v '  :  current_size     Disk2vhd
-     *      'tap\0' :  current_size     XenServer
-     *      'CTXS'  :  current_size     XenConverter
-     *
-     *  The user can override the table values via drive options, however
-     *  even with an override we will still use current_size for images
-     *  that have CHS geometry of the maximum size.
-     */
-    use_chs = (!!strncmp(footer->creator_app, "win ", 4) &&
-               !!strncmp(footer->creator_app, "qem2", 4) &&
-               !!strncmp(footer->creator_app, "d2v ", 4) &&
-               !!strncmp(footer->creator_app, "CTXS", 4) &&
-               !!memcmp(footer->creator_app, "tap", 4)) || s->force_use_chs;
+    /* Use CHS or current_size to determine the image size. */
+    use_chs = vpc_ignore_current_size(footer) || s->force_use_chs;

    if (!use_chs || bs->total_sectors == VHD_MAX_GEOMETRY || s->force_use_sz) {
        bs->total_sectors = be64_to_cpu(footer->current_size) /
@@ -1497,6 +1497,22 @@ static void external_snapshot_action(TransactionAction *action,
        return;
    }

+    /*
+     * Older QEMU versions have allowed adding an active parent node to an
+     * inactive child node. This is unsafe in the general case, but there is an
+     * important use case, which is taking a VM snapshot with migration to file
+     * and then adding an external snapshot while the VM is still stopped and
+     * images are inactive. Requiring the user to explicitly create the overlay
+     * as inactive would break compatibility, so just do it automatically here
+     * to keep this working.
+     */
+    if (bdrv_is_inactive(state->old_bs) && !bdrv_is_inactive(state->new_bs)) {
+        ret = bdrv_inactivate(state->new_bs, errp);
+        if (ret < 0) {
+            return;
+        }
+    }
+
    ret = bdrv_append(state->new_bs, state->old_bs, errp);
    if (ret < 0) {
        return;
@@ -3455,6 +3471,38 @@ void qmp_blockdev_del(const char *node_name, Error **errp)
    bdrv_unref(bs);
 }

+void qmp_blockdev_set_active(const char *node_name, bool active, Error **errp)
+{
+    int ret;
+
+    GLOBAL_STATE_CODE();
+    GRAPH_RDLOCK_GUARD_MAINLOOP();
+
+    if (!node_name) {
+        if (active) {
+            bdrv_activate_all(errp);
+        } else {
+            ret = bdrv_inactivate_all();
+            if (ret < 0) {
+                error_setg_errno(errp, -ret, "Failed to inactivate all nodes");
+            }
+        }
+    } else {
+        BlockDriverState *bs = bdrv_find_node(node_name);
+        if (!bs) {
+            error_setg(errp, "Failed to find node with node-name='%s'",
+                       node_name);
+            return;
+        }
+
+        if (active) {
+            bdrv_activate(bs, errp);
+        } else {
+            bdrv_inactivate(bs, errp);
+        }
+    }
+}
+
 static BdrvChild * GRAPH_RDLOCK
 bdrv_find_child(BlockDriverState *parent_bs, const char *child_name)
 {
@@ -539,8 +539,6 @@ void *block_job_create(const char *job_id, const BlockJobDriver *driver,
        goto fail;
    }

-    bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker);
-
    if (!block_job_set_speed(job, speed, errp)) {
        goto fail;
    }
@@ -1562,15 +1562,6 @@ static bool virtio_blk_vq_aio_context_init(VirtIOBlock *s, Error **errp)
            error_setg(errp, "ioeventfd is required for iothread");
            return false;
        }
-
-        /*
-         * If ioeventfd is (re-)enabled while the guest is running there could
-         * be block jobs that can conflict.
-         */
-        if (blk_op_is_blocked(conf->conf.blk, BLOCK_OP_TYPE_DATAPLANE, errp)) {
-            error_prepend(errp, "cannot start virtio-blk ioeventfd: ");
-            return false;
-        }
    }

    s->vq_aio_context = g_new(AioContext *, conf->num_queues);
@@ -1065,9 +1065,6 @@ static void virtio_scsi_hotplug(HotplugHandler *hotplug_dev, DeviceState *dev,
    int ret;

    if (s->ctx && !s->dataplane_fenced) {
-        if (blk_op_is_blocked(sd->conf.blk, BLOCK_OP_TYPE_DATAPLANE, errp)) {
-            return;
-        }
        ret = blk_set_aio_context(sd->conf.blk, s->ctx, errp);
        if (ret < 0) {
            return;
@@ -257,6 +257,7 @@ typedef enum {
 #define BDRV_OPT_AUTO_READ_ONLY "auto-read-only"
 #define BDRV_OPT_DISCARD        "discard"
 #define BDRV_OPT_FORCE_SHARE    "force-share"
+#define BDRV_OPT_ACTIVE         "active"


 #define BDRV_SECTOR_BITS   9
@@ -355,7 +356,6 @@ typedef enum BlockOpType {
    BLOCK_OP_TYPE_CHANGE,
    BLOCK_OP_TYPE_COMMIT_SOURCE,
    BLOCK_OP_TYPE_COMMIT_TARGET,
-    BLOCK_OP_TYPE_DATAPLANE,
    BLOCK_OP_TYPE_DRIVE_DEL,
    BLOCK_OP_TYPE_EJECT,
    BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT,
@@ -175,12 +175,18 @@ BlockDriverState * GRAPH_RDLOCK
 check_to_replace_node(BlockDriverState *parent_bs, const char *node_name,
                      Error **errp);

+
+bool GRAPH_RDLOCK bdrv_is_inactive(BlockDriverState *bs);
+
 int no_coroutine_fn GRAPH_RDLOCK
 bdrv_activate(BlockDriverState *bs, Error **errp);

 int coroutine_fn no_co_wrapper_bdrv_rdlock
 bdrv_co_activate(BlockDriverState *bs, Error **errp);

+int no_coroutine_fn
+bdrv_inactivate(BlockDriverState *bs, Error **errp);
+
 void bdrv_activate_all(Error **errp);
 int bdrv_inactivate_all(void);

@@ -29,6 +29,9 @@ typedef struct BlockExportDriver {
     */
    size_t instance_size;

+    /* True if the export type supports running on an inactive node */
+    bool supports_inactive;
+
    /* Creates and starts a new block export */
    int (*create)(BlockExport *, BlockExportOptions *, Error **);

@@ -32,6 +32,13 @@ void blk_set_allow_aio_context_change(BlockBackend *blk, bool allow);
 void blk_set_disable_request_queuing(BlockBackend *blk, bool disable);
 bool blk_iostatus_is_enabled(const BlockBackend *blk);

+/*
+ * Return the qdev ID, or if no ID is assigned the QOM path,
+ * of the block device attached to the BlockBackend.
+ *
+ * The caller is responsible for releasing the value returned
+ * with g_free() after use.
+ */
 char *blk_get_attached_dev_id(BlockBackend *blk);

 BlockAIOCB *blk_aio_pwrite_zeroes(BlockBackend *blk, int64_t offset,
@@ -12,51 +12,12 @@
 #include "qemu/error-report.h"
 #include "trace.h"

-/*
- * Migration-only cache to remember the block layer activation status.
- * Protected by BQL.
- *
- * We need this because..
- *
- * - Migration can fail after block devices are invalidated (during
- *   switchover phase).  When that happens, we need to be able to recover
- *   the block drive status by re-activating them.
- *
- * - Currently bdrv_inactivate_all() is not safe to be invoked on top of
- *   invalidated drives (even if bdrv_activate_all() is actually safe to be
- *   called any time!).  It means remembering this could help migration to
- *   make sure it won't invalidate twice in a row, crashing QEMU.  It can
- *   happen when we migrate a PAUSED VM from host1 to host2, then migrate
- *   again to host3 without starting it.  TODO: a cleaner solution is to
- *   allow safe invoke of bdrv_inactivate_all() at anytime, like
- *   bdrv_activate_all().
- *
- * For freshly started QEMU, the flag is initialized to TRUE reflecting the
- * scenario where QEMU owns block device ownerships.
- *
- * For incoming QEMU taking a migration stream, the flag is initialized to
- * FALSE reflecting that the incoming side doesn't own the block devices,
- * not until switchover happens.
- */
-static bool migration_block_active;
-
-/* Setup the disk activation status */
-void migration_block_active_setup(bool active)
-{
-    migration_block_active = active;
-}
-
 bool migration_block_activate(Error **errp)
 {
    ERRP_GUARD();

    assert(bql_locked());

-    if (migration_block_active) {
-        trace_migration_block_activation("active-skipped");
-        return true;
-    }
-
    trace_migration_block_activation("active");

    bdrv_activate_all(errp);
@@ -65,7 +26,6 @@ bool migration_block_activate(Error **errp)
        return false;
    }

-    migration_block_active = true;
    return true;
 }

@@ -75,11 +35,6 @@ bool migration_block_inactivate(void)

    assert(bql_locked());

-    if (!migration_block_active) {
-        trace_migration_block_activation("inactive-skipped");
-        return true;
-    }
-
    trace_migration_block_activation("inactive");

    ret = bdrv_inactivate_all();
@@ -89,6 +44,5 @@ bool migration_block_inactivate(void)
        return false;
    }

-    migration_block_active = false;
    return true;
 }
@@ -1895,12 +1895,6 @@ void qmp_migrate_incoming(const char *uri, bool has_channels,
        return;
    }

-    /*
-     * Newly setup incoming QEMU.  Mark the block active state to reflect
-     * that the src currently owns the disks.
-     */
-    migration_block_active_setup(false);
-
    once = false;
 }

@@ -3992,8 +3986,6 @@ static void migration_instance_init(Object *obj)
    ms->state = MIGRATION_STATUS_NONE;
    ms->mbps = -1;
    ms->pages_per_second = -1;
-    /* Freshly started QEMU owns all the block devices */
-    migration_block_active_setup(true);
    qemu_sem_init(&ms->pause_sem, 0);
    qemu_mutex_init(&ms->error_mutex);

@@ -554,7 +554,4 @@ void migration_bitmap_sync_precopy(bool last_stage);
 void dirty_bitmap_mig_init(void);
 bool should_send_vmdesc(void);

-/* migration/block-active.c */
-void migration_block_active_setup(bool active);
-
 #endif
@@ -2026,6 +2026,7 @@ static void nbd_export_delete(BlockExport *blk_exp)
 const BlockExportDriver blk_exp_nbd = {
    .type               = BLOCK_EXPORT_TYPE_NBD,
    .instance_size      = sizeof(NBDExport),
+    .supports_inactive  = true,
    .create             = nbd_export_create,
    .delete             = nbd_export_delete,
    .request_shutdown   = nbd_export_request_shutdown,
@@ -2920,6 +2921,22 @@ static coroutine_fn int nbd_handle_request(NBDClient *client,
    NBDExport *exp = client->exp;
    char *msg;
    size_t i;
+    bool inactive;
+
+    WITH_GRAPH_RDLOCK_GUARD() {
+        inactive = bdrv_is_inactive(blk_bs(exp->common.blk));
+        if (inactive) {
+            switch (request->type) {
+            case NBD_CMD_READ:
+                /* These commands are allowed on inactive nodes */
+                break;
+            default:
+                /* Return an error for the rest */
+                return nbd_send_generic_reply(client, request, -EPERM,
+                                              "export is inactive", errp);
+            }
+        }
+    }

    switch (request->type) {
    case NBD_CMD_CACHE:
@@ -486,6 +486,10 @@
 # @backing_file_depth: number of files in the backing file chain
 #     (since: 1.2)
 #
+# @active: true if the backend is active; typical cases for inactive backends
+#     are on the migration source instance after migration completes and on the
+#     destination before it completes. (since: 10.0)
+#
 # @encrypted: true if the backing device is encrypted
 #
 # @detect_zeroes: detect and optimize zero writes (Since 2.1)
@@ -556,7 +560,7 @@
 { 'struct': 'BlockDeviceInfo',
  'data': { 'file': 'str', '*node-name': 'str', 'ro': 'bool', 'drv': 'str',
            '*backing_file': 'str', 'backing_file_depth': 'int',
-            'encrypted': 'bool',
+            'active': 'bool', 'encrypted': 'bool',
            'detect_zeroes': 'BlockdevDetectZeroesOptions',
            'bps': 'int', 'bps_rd': 'int', 'bps_wr': 'int',
            'iops': 'int', 'iops_rd': 'int', 'iops_wr': 'int',
@@ -4679,6 +4683,11 @@
 #
 # @cache: cache-related options
 #
+# @active: whether the block node should be activated (default: true).
+#     Having inactive block nodes is useful primarily for migration because it
+#     allows opening an image on the destination while the source is still
+#     holding locks for it. (Since 10.0)
+#
 # @read-only: whether the block device should be read-only (default:
 #     false).  Note that some block drivers support only read-only
 #     access, either generally or in certain configurations.  In this
@@ -4705,6 +4714,7 @@
            '*node-name': 'str',
            '*discard': 'BlockdevDiscardOptions',
            '*cache': 'BlockdevCacheOptions',
+            '*active': 'bool',
            '*read-only': 'bool',
            '*auto-read-only': 'bool',
            '*force-share': 'bool',
@@ -4935,6 +4945,38 @@
 { 'command': 'blockdev-del', 'data': { 'node-name': 'str' },
  'allow-preconfig': true }

+##
+# @blockdev-set-active:
+#
+# Activate or inactivate a block device. Use this to manage the handover of
+# block devices on migration with qemu-storage-daemon.
+#
+# Activating a node automatically activates all of its child nodes first.
+# Inactivating a node automatically inactivates any of its child nodes that are
+# not in use by a still active node.
+#
+# @node-name: Name of the graph node to activate or inactivate. By default, all
+#     nodes are affected by the operation.
+#
+# @active: true if the nodes should be active when the command returns success,
+#     false if they should be inactive.
+#
+# Since: 10.0
+#
+# .. qmp-example::
+#
+#     -> { "execute": "blockdev-set-active",
+#          "arguments": {
+#               "node-name": "node0",
+#               "active": false
+#          }
+#        }
+#     <- { "return": {} }
+##
+{ 'command': 'blockdev-set-active',
+  'data': { '*node-name': 'str', 'active': 'bool' },
+  'allow-preconfig': true }
+
 ##
 # @BlockdevCreateOptionsFile:
 #
--- a/Show More
+++ b/Show More