]> git.proxmox.com Git - mirror_qemu.git/blobdiff - block.c
linux-user: add a function hook to translate sockaddr
[mirror_qemu.git] / block.c
diff --git a/block.c b/block.c
index a2d6238b82de3226dde35c2484e601af88878a62..411edbf9fcc65d1617f5b62f08e4003546a1b548 100644 (file)
--- a/block.c
+++ b/block.c
 #include "qemu/error-report.h"
 #include "qemu/module.h"
 #include "qapi/qmp/qerror.h"
+#include "qapi/qmp/qbool.h"
 #include "qapi/qmp/qjson.h"
 #include "sysemu/block-backend.h"
 #include "sysemu/sysemu.h"
 #include "qemu/notify.h"
-#include "block/coroutine.h"
+#include "qemu/coroutine.h"
 #include "block/qapi.h"
 #include "qmp-commands.h"
 #include "qemu/timer.h"
@@ -73,8 +74,7 @@ struct BdrvDirtyBitmap {
 
 #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */
 
-static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
-    QTAILQ_HEAD_INITIALIZER(bdrv_states);
+struct BdrvStates bdrv_states = QTAILQ_HEAD_INITIALIZER(bdrv_states);
 
 static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states =
     QTAILQ_HEAD_INITIALIZER(graph_bdrv_states);
@@ -257,7 +257,6 @@ BlockDriverState *bdrv_new(void)
     for (i = 0; i < BLOCK_OP_TYPE_MAX; i++) {
         QLIST_INIT(&bs->op_blockers[i]);
     }
-    bdrv_iostatus_disable(bs);
     notifier_list_init(&bs->close_notifiers);
     notifier_with_return_list_init(&bs->before_write_notifiers);
     qemu_co_queue_init(&bs->throttled_reqs[0]);
@@ -625,6 +624,20 @@ static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
     return 0;
 }
 
+/**
+ * Combines a QDict of new block driver @options with any missing options taken
+ * from @old_options, so that leaving out an option defaults to its old value.
+ */
+static void bdrv_join_options(BlockDriverState *bs, QDict *options,
+                              QDict *old_options)
+{
+    if (bs->drv && bs->drv->bdrv_join_options) {
+        bs->drv->bdrv_join_options(options, old_options);
+    } else {
+        qdict_join(options, old_options, false);
+    }
+}
+
 /**
  * Set open flags for a given discard mode
  *
@@ -683,60 +696,81 @@ static int bdrv_temp_snapshot_flags(int flags)
 }
 
 /*
- * Returns the flags that bs->file should get if a protocol driver is expected,
- * based on the given flags for the parent BDS
+ * Returns the options and flags that bs->file should get if a protocol driver
+ * is expected, based on the given options and flags for the parent BDS
  */
-static int bdrv_inherited_flags(int flags)
+static void bdrv_inherited_options(int *child_flags, QDict *child_options,
+                                   int parent_flags, QDict *parent_options)
 {
+    int flags = parent_flags;
+
     /* Enable protocol handling, disable format probing for bs->file */
     flags |= BDRV_O_PROTOCOL;
 
+    /* If the cache mode isn't explicitly set, inherit direct and no-flush from
+     * the parent. */
+    qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
+    qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
+
     /* Our block drivers take care to send flushes and respect unmap policy,
-     * so we can enable both unconditionally on lower layers. */
-    flags |= BDRV_O_CACHE_WB | BDRV_O_UNMAP;
+     * so we can default to enable both on lower layers regardless of the
+     * corresponding parent options. */
+    qdict_set_default_str(child_options, BDRV_OPT_CACHE_WB, "on");
+    flags |= BDRV_O_UNMAP;
 
     /* Clear flags that only apply to the top layer */
     flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING | BDRV_O_COPY_ON_READ);
 
-    return flags;
+    *child_flags = flags;
 }
 
 const BdrvChildRole child_file = {
-    .inherit_flags = bdrv_inherited_flags,
+    .inherit_options = bdrv_inherited_options,
 };
 
 /*
- * Returns the flags that bs->file should get if the use of formats (and not
- * only protocols) is permitted for it, based on the given flags for the parent
- * BDS
+ * Returns the options and flags that bs->file should get if the use of formats
+ * (and not only protocols) is permitted for it, based on the given options and
+ * flags for the parent BDS
  */
-static int bdrv_inherited_fmt_flags(int parent_flags)
+static void bdrv_inherited_fmt_options(int *child_flags, QDict *child_options,
+                                       int parent_flags, QDict *parent_options)
 {
-    int flags = child_file.inherit_flags(parent_flags);
-    return flags & ~BDRV_O_PROTOCOL;
+    child_file.inherit_options(child_flags, child_options,
+                               parent_flags, parent_options);
+
+    *child_flags &= ~BDRV_O_PROTOCOL;
 }
 
 const BdrvChildRole child_format = {
-    .inherit_flags = bdrv_inherited_fmt_flags,
+    .inherit_options = bdrv_inherited_fmt_options,
 };
 
 /*
- * Returns the flags that bs->backing should get, based on the given flags
- * for the parent BDS
+ * Returns the options and flags that bs->backing should get, based on the
+ * given options and flags for the parent BDS
  */
-static int bdrv_backing_flags(int flags)
+static void bdrv_backing_options(int *child_flags, QDict *child_options,
+                                 int parent_flags, QDict *parent_options)
 {
+    int flags = parent_flags;
+
+    /* The cache mode is inherited unmodified for backing files */
+    qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_WB);
+    qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_DIRECT);
+    qdict_copy_default(child_options, parent_options, BDRV_OPT_CACHE_NO_FLUSH);
+
     /* backing files always opened read-only */
     flags &= ~(BDRV_O_RDWR | BDRV_O_COPY_ON_READ);
 
     /* snapshot=on is handled on the top layer */
     flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_TEMPORARY);
 
-    return flags;
+    *child_flags = flags;
 }
 
 static const BdrvChildRole child_backing = {
-    .inherit_flags = bdrv_backing_flags,
+    .inherit_options = bdrv_backing_options,
 };
 
 static int bdrv_open_flags(BlockDriverState *bs, int flags)
@@ -759,16 +793,55 @@ static int bdrv_open_flags(BlockDriverState *bs, int flags)
     return open_flags;
 }
 
+static void update_flags_from_options(int *flags, QemuOpts *opts)
+{
+    *flags &= ~BDRV_O_CACHE_MASK;
+
+    assert(qemu_opt_find(opts, BDRV_OPT_CACHE_WB));
+    if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_WB, false)) {
+        *flags |= BDRV_O_CACHE_WB;
+    }
+
+    assert(qemu_opt_find(opts, BDRV_OPT_CACHE_NO_FLUSH));
+    if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_NO_FLUSH, false)) {
+        *flags |= BDRV_O_NO_FLUSH;
+    }
+
+    assert(qemu_opt_find(opts, BDRV_OPT_CACHE_DIRECT));
+    if (qemu_opt_get_bool(opts, BDRV_OPT_CACHE_DIRECT, false)) {
+        *flags |= BDRV_O_NOCACHE;
+    }
+}
+
+static void update_options_from_flags(QDict *options, int flags)
+{
+    if (!qdict_haskey(options, BDRV_OPT_CACHE_WB)) {
+        qdict_put(options, BDRV_OPT_CACHE_WB,
+                  qbool_from_bool(flags & BDRV_O_CACHE_WB));
+    }
+    if (!qdict_haskey(options, BDRV_OPT_CACHE_DIRECT)) {
+        qdict_put(options, BDRV_OPT_CACHE_DIRECT,
+                  qbool_from_bool(flags & BDRV_O_NOCACHE));
+    }
+    if (!qdict_haskey(options, BDRV_OPT_CACHE_NO_FLUSH)) {
+        qdict_put(options, BDRV_OPT_CACHE_NO_FLUSH,
+                  qbool_from_bool(flags & BDRV_O_NO_FLUSH));
+    }
+}
+
 static void bdrv_assign_node_name(BlockDriverState *bs,
                                   const char *node_name,
                                   Error **errp)
 {
-    if (!node_name) {
-        return;
-    }
+    char *gen_node_name = NULL;
 
-    /* Check for empty string or invalid characters */
-    if (!id_wellformed(node_name)) {
+    if (!node_name) {
+        node_name = gen_node_name = id_generate(ID_BLOCK);
+    } else if (!id_wellformed(node_name)) {
+        /*
+         * Check for empty string or invalid characters, but not if it is
+         * generated (generated names use characters not available to the user)
+         */
         error_setg(errp, "Invalid node name");
         return;
     }
@@ -777,18 +850,20 @@ static void bdrv_assign_node_name(BlockDriverState *bs,
     if (blk_by_name(node_name)) {
         error_setg(errp, "node-name=%s is conflicting with a device id",
                    node_name);
-        return;
+        goto out;
     }
 
     /* takes care of avoiding duplicates node names */
     if (bdrv_find_node(node_name)) {
         error_setg(errp, "Duplicate node name");
-        return;
+        goto out;
     }
 
     /* copy node name into the bs and insert it into the graph list */
     pstrcpy(bs->node_name, sizeof(bs->node_name), node_name);
     QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list);
+out:
+    g_free(gen_node_name);
 }
 
 static QemuOptsList bdrv_runtime_opts = {
@@ -800,6 +875,26 @@ static QemuOptsList bdrv_runtime_opts = {
             .type = QEMU_OPT_STRING,
             .help = "Node name of the block device node",
         },
+        {
+            .name = "driver",
+            .type = QEMU_OPT_STRING,
+            .help = "Block driver to use for the node",
+        },
+        {
+            .name = BDRV_OPT_CACHE_WB,
+            .type = QEMU_OPT_BOOL,
+            .help = "Enable writeback mode",
+        },
+        {
+            .name = BDRV_OPT_CACHE_DIRECT,
+            .type = QEMU_OPT_BOOL,
+            .help = "Bypass software writeback cache on the host",
+        },
+        {
+            .name = BDRV_OPT_CACHE_NO_FLUSH,
+            .type = QEMU_OPT_BOOL,
+            .help = "Ignore flush requests",
+        },
         { /* end of list */ }
     },
 };
@@ -810,18 +905,31 @@ static QemuOptsList bdrv_runtime_opts = {
  * Removes all processed options from *options.
  */
 static int bdrv_open_common(BlockDriverState *bs, BdrvChild *file,
-    QDict *options, int flags, BlockDriver *drv, Error **errp)
+                            QDict *options, int flags, Error **errp)
 {
     int ret, open_flags;
     const char *filename;
+    const char *driver_name = NULL;
     const char *node_name = NULL;
     QemuOpts *opts;
+    BlockDriver *drv;
     Error *local_err = NULL;
 
-    assert(drv != NULL);
     assert(bs->file == NULL);
     assert(options != NULL && bs->options != options);
 
+    opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
+    qemu_opts_absorb_qdict(opts, options, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        ret = -EINVAL;
+        goto fail_opts;
+    }
+
+    driver_name = qemu_opt_get(opts, "driver");
+    drv = bdrv_find_format(driver_name);
+    assert(drv != NULL);
+
     if (file != NULL) {
         filename = file->bs->filename;
     } else {
@@ -831,19 +939,12 @@ static int bdrv_open_common(BlockDriverState *bs, BdrvChild *file,
     if (drv->bdrv_needs_filename && !filename) {
         error_setg(errp, "The '%s' block driver requires a file name",
                    drv->format_name);
-        return -EINVAL;
-    }
-
-    trace_bdrv_open_common(bs, filename ?: "", flags, drv->format_name);
-
-    opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
-    qemu_opts_absorb_qdict(opts, options, &local_err);
-    if (local_err) {
-        error_propagate(errp, local_err);
         ret = -EINVAL;
         goto fail_opts;
     }
 
+    trace_bdrv_open_common(bs, filename ?: "", flags, drv->format_name);
+
     node_name = qemu_opt_get(opts, "node-name");
     bdrv_assign_node_name(bs, node_name, &local_err);
     if (local_err) {
@@ -852,7 +953,6 @@ static int bdrv_open_common(BlockDriverState *bs, BdrvChild *file,
         goto fail_opts;
     }
 
-    bs->guest_block_size = 512;
     bs->request_alignment = 512;
     bs->zero_beyond_eof = true;
     open_flags = bdrv_open_flags(bs, flags);
@@ -889,7 +989,9 @@ static int bdrv_open_common(BlockDriverState *bs, BdrvChild *file,
     bs->drv = drv;
     bs->opaque = g_malloc0(drv->instance_size);
 
-    bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB);
+    /* Apply cache mode options */
+    update_flags_from_options(&bs->open_flags, opts);
+    bdrv_set_enable_write_cache(bs, bs->open_flags & BDRV_O_CACHE_WB);
 
     /* Open the image, either directly or using a protocol */
     if (drv->bdrv_file_open) {
@@ -982,37 +1084,45 @@ static QDict *parse_json_filename(const char *filename, Error **errp)
     return options;
 }
 
+static void parse_json_protocol(QDict *options, const char **pfilename,
+                                Error **errp)
+{
+    QDict *json_options;
+    Error *local_err = NULL;
+
+    /* Parse json: pseudo-protocol */
+    if (!*pfilename || !g_str_has_prefix(*pfilename, "json:")) {
+        return;
+    }
+
+    json_options = parse_json_filename(*pfilename, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return;
+    }
+
+    /* Options given in the filename have lower priority than options
+     * specified directly */
+    qdict_join(options, json_options, false);
+    QDECREF(json_options);
+    *pfilename = NULL;
+}
+
 /*
  * Fills in default options for opening images and converts the legacy
  * filename/flags pair to option QDict entries.
  * The BDRV_O_PROTOCOL flag in *flags will be set or cleared accordingly if a
  * block driver has been specified explicitly.
  */
-static int bdrv_fill_options(QDict **options, const char **pfilename,
+static int bdrv_fill_options(QDict **options, const char *filename,
                              int *flags, Error **errp)
 {
-    const char *filename = *pfilename;
     const char *drvname;
     bool protocol = *flags & BDRV_O_PROTOCOL;
     bool parse_filename = false;
     BlockDriver *drv = NULL;
     Error *local_err = NULL;
 
-    /* Parse json: pseudo-protocol */
-    if (filename && g_str_has_prefix(filename, "json:")) {
-        QDict *json_options = parse_json_filename(filename, &local_err);
-        if (local_err) {
-            error_propagate(errp, local_err);
-            return -EINVAL;
-        }
-
-        /* Options given in the filename have lower priority than options
-         * specified directly */
-        qdict_join(*options, json_options, false);
-        QDECREF(json_options);
-        *pfilename = filename = NULL;
-    }
-
     drvname = qdict_get_try_str(*options, "driver");
     if (drvname) {
         drv = bdrv_find_format(drvname);
@@ -1031,6 +1141,9 @@ static int bdrv_fill_options(QDict **options, const char **pfilename,
         *flags &= ~BDRV_O_PROTOCOL;
     }
 
+    /* Translate cache options from flags into options */
+    update_options_from_flags(*options, *flags);
+
     /* Fetch the file name from the options QDict if necessary */
     if (protocol && filename) {
         if (!qdict_haskey(*options, "filename")) {
@@ -1076,38 +1189,52 @@ static int bdrv_fill_options(QDict **options, const char **pfilename,
         }
     }
 
+    if (runstate_check(RUN_STATE_INMIGRATE)) {
+        *flags |= BDRV_O_INCOMING;
+    }
+
     return 0;
 }
 
 static BdrvChild *bdrv_attach_child(BlockDriverState *parent_bs,
                                     BlockDriverState *child_bs,
+                                    const char *child_name,
                                     const BdrvChildRole *child_role)
 {
     BdrvChild *child = g_new(BdrvChild, 1);
     *child = (BdrvChild) {
         .bs     = child_bs,
+        .name   = g_strdup(child_name),
         .role   = child_role,
     };
 
     QLIST_INSERT_HEAD(&parent_bs->children, child, next);
+    QLIST_INSERT_HEAD(&child_bs->parents, child, next_parent);
 
     return child;
 }
 
-void bdrv_detach_child(BdrvChild *child)
+static void bdrv_detach_child(BdrvChild *child)
 {
     QLIST_REMOVE(child, next);
+    QLIST_REMOVE(child, next_parent);
+    g_free(child->name);
     g_free(child);
 }
 
 void bdrv_unref_child(BlockDriverState *parent, BdrvChild *child)
 {
-    BlockDriverState *child_bs = child->bs;
+    BlockDriverState *child_bs;
+
+    if (child == NULL) {
+        return;
+    }
 
     if (child->bs->inherits_from == parent) {
         child->bs->inherits_from = NULL;
     }
 
+    child_bs = child->bs;
     bdrv_detach_child(child);
     bdrv_unref(child_bs);
 }
@@ -1138,7 +1265,7 @@ void bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd)
         bs->backing = NULL;
         goto out;
     }
-    bs->backing = bdrv_attach_child(bs, backing_hd, &child_backing);
+    bs->backing = bdrv_attach_child(bs, backing_hd, "backing", &child_backing);
     bs->open_flags &= ~BDRV_O_NO_BACKING;
     pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_hd->filename);
     pstrcpy(bs->backing_format, sizeof(bs->backing_format),
@@ -1155,30 +1282,43 @@ out:
 /*
  * Opens the backing file for a BlockDriverState if not yet open
  *
- * options is a QDict of options to pass to the block drivers, or NULL for an
- * empty set of options. The reference to the QDict is transferred to this
- * function (even on failure), so if the caller intends to reuse the dictionary,
- * it needs to use QINCREF() before calling bdrv_file_open.
+ * bdref_key specifies the key for the image's BlockdevRef in the options QDict.
+ * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict
+ * itself, all options starting with "${bdref_key}." are considered part of the
+ * BlockdevRef.
+ *
+ * TODO Can this be unified with bdrv_open_image()?
  */
-int bdrv_open_backing_file(BlockDriverState *bs, QDict *options, Error **errp)
+int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
+                           const char *bdref_key, Error **errp)
 {
     char *backing_filename = g_malloc0(PATH_MAX);
+    char *bdref_key_dot;
+    const char *reference = NULL;
     int ret = 0;
     BlockDriverState *backing_hd;
+    QDict *options;
+    QDict *tmp_parent_options = NULL;
     Error *local_err = NULL;
 
     if (bs->backing != NULL) {
-        QDECREF(options);
         goto free_exit;
     }
 
     /* NULL means an empty set of options */
-    if (options == NULL) {
-        options = qdict_new();
+    if (parent_options == NULL) {
+        tmp_parent_options = qdict_new();
+        parent_options = tmp_parent_options;
     }
 
     bs->open_flags &= ~BDRV_O_NO_BACKING;
-    if (qdict_haskey(options, "file.filename")) {
+
+    bdref_key_dot = g_strdup_printf("%s.", bdref_key);
+    qdict_extract_subqdict(parent_options, &options, bdref_key_dot);
+    g_free(bdref_key_dot);
+
+    reference = qdict_get_try_str(parent_options, bdref_key);
+    if (reference || qdict_haskey(options, "file.filename")) {
         backing_filename[0] = '\0';
     } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) {
         QDECREF(options);
@@ -1201,19 +1341,16 @@ int bdrv_open_backing_file(BlockDriverState *bs, QDict *options, Error **errp)
         goto free_exit;
     }
 
-    backing_hd = bdrv_new();
-
     if (bs->backing_format[0] != '\0' && !qdict_haskey(options, "driver")) {
         qdict_put(options, "driver", qstring_from_str(bs->backing_format));
     }
 
-    assert(bs->backing == NULL);
+    backing_hd = NULL;
     ret = bdrv_open_inherit(&backing_hd,
                             *backing_filename ? backing_filename : NULL,
-                            NULL, options, 0, bs, &child_backing, &local_err);
+                            reference, options, 0, bs, &child_backing,
+                            &local_err);
     if (ret < 0) {
-        bdrv_unref(backing_hd);
-        backing_hd = NULL;
         bs->open_flags |= BDRV_O_NO_BACKING;
         error_setg(errp, "Could not open backing file: %s",
                    error_get_pretty(local_err));
@@ -1226,8 +1363,11 @@ int bdrv_open_backing_file(BlockDriverState *bs, QDict *options, Error **errp)
     bdrv_set_backing_hd(bs, backing_hd);
     bdrv_unref(backing_hd);
 
+    qdict_del(parent_options, bdref_key);
+
 free_exit:
     g_free(backing_filename);
+    QDECREF(tmp_parent_options);
     return ret;
 }
 
@@ -1281,7 +1421,7 @@ BdrvChild *bdrv_open_child(const char *filename,
         goto done;
     }
 
-    c = bdrv_attach_child(parent, bs, child_role);
+    c = bdrv_attach_child(parent, bs, bdref_key, child_role);
 
 done:
     qdict_del(options, bdref_key);
@@ -1380,6 +1520,7 @@ static int bdrv_open_inherit(BlockDriverState **pbs, const char *filename,
     BlockDriverState *bs;
     BlockDriver *drv = NULL;
     const char *drvname;
+    const char *backing;
     Error *local_err = NULL;
     int snapshot_flags = 0;
 
@@ -1423,21 +1564,34 @@ static int bdrv_open_inherit(BlockDriverState **pbs, const char *filename,
         options = qdict_new();
     }
 
+    /* json: syntax counts as explicit options, as if in the QDict */
+    parse_json_protocol(options, &filename, &local_err);
+    if (local_err) {
+        ret = -EINVAL;
+        goto fail;
+    }
+
+    bs->explicit_options = qdict_clone_shallow(options);
+
     if (child_role) {
         bs->inherits_from = parent;
-        flags = child_role->inherit_flags(parent->open_flags);
+        child_role->inherit_options(&flags, options,
+                                    parent->open_flags, parent->options);
     }
 
-    ret = bdrv_fill_options(&options, &filename, &flags, &local_err);
+    ret = bdrv_fill_options(&options, filename, &flags, &local_err);
     if (local_err) {
         goto fail;
     }
 
+    bs->open_flags = flags;
+    bs->options = options;
+    options = qdict_clone_shallow(options);
+
     /* Find the right image format driver */
     drvname = qdict_get_try_str(options, "driver");
     if (drvname) {
         drv = bdrv_find_format(drvname);
-        qdict_del(options, "driver");
         if (!drv) {
             error_setg(errp, "Unknown driver: '%s'", drvname);
             ret = -EINVAL;
@@ -1447,9 +1601,11 @@ static int bdrv_open_inherit(BlockDriverState **pbs, const char *filename,
 
     assert(drvname || !(flags & BDRV_O_PROTOCOL));
 
-    bs->open_flags = flags;
-    bs->options = options;
-    options = qdict_clone_shallow(options);
+    backing = qdict_get_try_str(options, "backing");
+    if (backing && *backing == '\0') {
+        flags |= BDRV_O_NO_BACKING;
+        qdict_del(options, "backing");
+    }
 
     /* Open image file without format layer */
     if ((flags & BDRV_O_PROTOCOL) == 0) {
@@ -1458,7 +1614,7 @@ static int bdrv_open_inherit(BlockDriverState **pbs, const char *filename,
         }
         if (flags & BDRV_O_SNAPSHOT) {
             snapshot_flags = bdrv_temp_snapshot_flags(flags);
-            flags = bdrv_backing_flags(flags);
+            bdrv_backing_options(&flags, options, flags, options);
         }
 
         bs->open_flags = flags;
@@ -1478,6 +1634,19 @@ static int bdrv_open_inherit(BlockDriverState **pbs, const char *filename,
         if (ret < 0) {
             goto fail;
         }
+        /*
+         * This option update would logically belong in bdrv_fill_options(),
+         * but we first need to open bs->file for the probing to work, while
+         * opening bs->file already requires the (mostly) final set of options
+         * so that cache mode etc. can be inherited.
+         *
+         * Adding the driver later is somewhat ugly, but it's not an option
+         * that would ever be inherited, so it's correct. We just need to make
+         * sure to update both bs->options (which has the full effective
+         * options for bs) and options (which has file.* already removed).
+         */
+        qdict_put(bs->options, "driver", qstring_from_str(drv->format_name));
+        qdict_put(options, "driver", qstring_from_str(drv->format_name));
     } else if (!drv) {
         error_setg(errp, "Must specify either driver or file");
         ret = -EINVAL;
@@ -1491,7 +1660,7 @@ static int bdrv_open_inherit(BlockDriverState **pbs, const char *filename,
     assert(!(flags & BDRV_O_PROTOCOL) || !file);
 
     /* Open the image */
-    ret = bdrv_open_common(bs, file, options, flags, drv, &local_err);
+    ret = bdrv_open_common(bs, file, options, flags, &local_err);
     if (ret < 0) {
         goto fail;
     }
@@ -1503,10 +1672,7 @@ static int bdrv_open_inherit(BlockDriverState **pbs, const char *filename,
 
     /* If there is a backing file, use it */
     if ((flags & BDRV_O_NO_BACKING) == 0) {
-        QDict *backing_options;
-
-        qdict_extract_subqdict(options, &backing_options, "backing.");
-        ret = bdrv_open_backing_file(bs, backing_options, &local_err);
+        ret = bdrv_open_backing_file(bs, options, "backing", &local_err);
         if (ret < 0) {
             goto close_and_fail;
         }
@@ -1514,15 +1680,6 @@ static int bdrv_open_inherit(BlockDriverState **pbs, const char *filename,
 
     bdrv_refresh_filename(bs);
 
-    /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
-     * temporary snapshot afterwards. */
-    if (snapshot_flags) {
-        ret = bdrv_append_temp_snapshot(bs, snapshot_flags, &local_err);
-        if (local_err) {
-            goto close_and_fail;
-        }
-    }
-
     /* Check if any unknown options were used */
     if (options && (qdict_size(options) != 0)) {
         const QDictEntry *entry = qdict_first(options);
@@ -1554,12 +1711,23 @@ static int bdrv_open_inherit(BlockDriverState **pbs, const char *filename,
 
     QDECREF(options);
     *pbs = bs;
+
+    /* For snapshot=on, create a temporary qcow2 overlay. bs points to the
+     * temporary snapshot afterwards. */
+    if (snapshot_flags) {
+        ret = bdrv_append_temp_snapshot(bs, snapshot_flags, &local_err);
+        if (local_err) {
+            goto close_and_fail;
+        }
+    }
+
     return 0;
 
 fail:
     if (file != NULL) {
         bdrv_unref_child(bs, file);
     }
+    QDECREF(bs->explicit_options);
     QDECREF(bs->options);
     QDECREF(options);
     bs->options = NULL;
@@ -1622,15 +1790,19 @@ typedef struct BlockReopenQueueEntry {
  * bs_queue, or the existing bs_queue being used.
  *
  */
-BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
-                                    BlockDriverState *bs,
-                                    QDict *options, int flags)
+static BlockReopenQueue *bdrv_reopen_queue_child(BlockReopenQueue *bs_queue,
+                                                 BlockDriverState *bs,
+                                                 QDict *options,
+                                                 int flags,
+                                                 const BdrvChildRole *role,
+                                                 QDict *parent_options,
+                                                 int parent_flags)
 {
     assert(bs != NULL);
 
     BlockReopenQueueEntry *bs_entry;
     BdrvChild *child;
-    QDict *old_options;
+    QDict *old_options, *explicit_options;
 
     if (bs_queue == NULL) {
         bs_queue = g_new0(BlockReopenQueue, 1);
@@ -1641,23 +1813,63 @@ BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
         options = qdict_new();
     }
 
+    /*
+     * Precedence of options:
+     * 1. Explicitly passed in options (highest)
+     * 2. Set in flags (only for top level)
+     * 3. Retained from explicitly set options of bs
+     * 4. Inherited from parent node
+     * 5. Retained from effective options of bs
+     */
+
+    if (!parent_options) {
+        /*
+         * Any setting represented by flags is always updated. If the
+         * corresponding QDict option is set, it takes precedence. Otherwise
+         * the flag is translated into a QDict option. The old setting of bs is
+         * not considered.
+         */
+        update_options_from_flags(options, flags);
+    }
+
+    /* Old explicitly set values (don't overwrite by inherited value) */
+    old_options = qdict_clone_shallow(bs->explicit_options);
+    bdrv_join_options(bs, options, old_options);
+    QDECREF(old_options);
+
+    explicit_options = qdict_clone_shallow(options);
+
+    /* Inherit from parent node */
+    if (parent_options) {
+        assert(!flags);
+        role->inherit_options(&flags, options, parent_flags, parent_options);
+    }
+
+    /* Old values are used for options that aren't set yet */
     old_options = qdict_clone_shallow(bs->options);
-    qdict_join(options, old_options, false);
+    bdrv_join_options(bs, options, old_options);
     QDECREF(old_options);
 
     /* bdrv_open() masks this flag out */
     flags &= ~BDRV_O_PROTOCOL;
 
     QLIST_FOREACH(child, &bs->children, next) {
-        int child_flags;
+        QDict *new_child_options;
+        char *child_key_dot;
 
+        /* reopen can only change the options of block devices that were
+         * implicitly created and inherited options. For other (referenced)
+         * block devices, a syntax like "backing.foo" results in an error. */
         if (child->bs->inherits_from != bs) {
             continue;
         }
 
-        child_flags = child->role->inherit_flags(flags);
-        /* TODO Pass down child flags (backing.*, extents.*, ...) */
-        bdrv_reopen_queue(bs_queue, child->bs, NULL, child_flags);
+        child_key_dot = g_strdup_printf("%s.", child->name);
+        qdict_extract_subqdict(options, &new_child_options, child_key_dot);
+        g_free(child_key_dot);
+
+        bdrv_reopen_queue_child(bs_queue, child->bs, new_child_options, 0,
+                                child->role, options, flags);
     }
 
     bs_entry = g_new0(BlockReopenQueueEntry, 1);
@@ -1665,11 +1877,20 @@ BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
 
     bs_entry->state.bs = bs;
     bs_entry->state.options = options;
+    bs_entry->state.explicit_options = explicit_options;
     bs_entry->state.flags = flags;
 
     return bs_queue;
 }
 
+BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue,
+                                    BlockDriverState *bs,
+                                    QDict *options, int flags)
+{
+    return bdrv_reopen_queue_child(bs_queue, bs, options, flags,
+                                   NULL, NULL, 0);
+}
+
 /*
  * Reopen multiple BlockDriverStates atomically & transactionally.
  *
@@ -1716,6 +1937,8 @@ cleanup:
     QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
         if (ret && bs_entry->prepared) {
             bdrv_reopen_abort(&bs_entry->state);
+        } else if (ret) {
+            QDECREF(bs_entry->state.explicit_options);
         }
         QDECREF(bs_entry->state.options);
         g_free(bs_entry);
@@ -1763,11 +1986,47 @@ int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
     int ret = -1;
     Error *local_err = NULL;
     BlockDriver *drv;
+    QemuOpts *opts;
+    const char *value;
 
     assert(reopen_state != NULL);
     assert(reopen_state->bs->drv != NULL);
     drv = reopen_state->bs->drv;
 
+    /* Process generic block layer options */
+    opts = qemu_opts_create(&bdrv_runtime_opts, NULL, 0, &error_abort);
+    qemu_opts_absorb_qdict(opts, reopen_state->options, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        ret = -EINVAL;
+        goto error;
+    }
+
+    update_flags_from_options(&reopen_state->flags, opts);
+
+    /* If a guest device is attached, it owns WCE */
+    if (reopen_state->bs->blk && blk_get_attached_dev(reopen_state->bs->blk)) {
+        bool old_wce = bdrv_enable_write_cache(reopen_state->bs);
+        bool new_wce = (reopen_state->flags & BDRV_O_CACHE_WB);
+        if (old_wce != new_wce) {
+            error_setg(errp, "Cannot change cache.writeback: Device attached");
+            ret = -EINVAL;
+            goto error;
+        }
+    }
+
+    /* node-name and driver must be unchanged. Put them back into the QDict, so
+     * that they are checked at the end of this function. */
+    value = qemu_opt_get(opts, "node-name");
+    if (value) {
+        qdict_put(reopen_state->options, "node-name", qstring_from_str(value));
+    }
+
+    value = qemu_opt_get(opts, "driver");
+    if (value) {
+        qdict_put(reopen_state->options, "driver", qstring_from_str(value));
+    }
+
     /* if we are to stay read-only, do not allow permission change
      * to r/w */
     if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) &&
@@ -1780,8 +2039,7 @@ int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
 
     ret = bdrv_flush(reopen_state->bs);
     if (ret) {
-        error_set(errp, ERROR_CLASS_GENERIC_ERROR, "Error (%s) flushing drive",
-                  strerror(-ret));
+        error_setg_errno(errp, -ret, "Error flushing drive");
         goto error;
     }
 
@@ -1829,6 +2087,7 @@ int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue,
     ret = 0;
 
 error:
+    qemu_opts_del(opts);
     return ret;
 }
 
@@ -1851,6 +2110,9 @@ void bdrv_reopen_commit(BDRVReopenState *reopen_state)
     }
 
     /* set BDS specific flags now */
+    QDECREF(reopen_state->bs->explicit_options);
+
+    reopen_state->bs->explicit_options   = reopen_state->explicit_options;
     reopen_state->bs->open_flags         = reopen_state->flags;
     reopen_state->bs->enable_write_cache = !!(reopen_state->flags &
                                               BDRV_O_CACHE_WB);
@@ -1874,6 +2136,8 @@ void bdrv_reopen_abort(BDRVReopenState *reopen_state)
     if (drv->bdrv_reopen_abort) {
         drv->bdrv_reopen_abort(reopen_state);
     }
+
+    QDECREF(reopen_state->explicit_options);
 }
 
 
@@ -1886,7 +2150,7 @@ void bdrv_close(BlockDriverState *bs)
     }
 
     /* Disable I/O limits and drain all pending throttled requests */
-    if (bs->io_limits_enabled) {
+    if (bs->throttle_state) {
         bdrv_io_limits_disable(bs);
     }
 
@@ -1895,6 +2159,10 @@ void bdrv_close(BlockDriverState *bs)
     bdrv_drain(bs); /* in case flush left pending I/O */
     notifier_list_notify(&bs->close_notifiers, bs);
 
+    if (bs->blk) {
+        blk_dev_change_media_cb(bs->blk, false);
+    }
+
     if (bs->drv) {
         BdrvChild *child, *next;
 
@@ -1928,15 +2196,12 @@ void bdrv_close(BlockDriverState *bs)
         bs->sg = 0;
         bs->zero_beyond_eof = false;
         QDECREF(bs->options);
+        QDECREF(bs->explicit_options);
         bs->options = NULL;
         QDECREF(bs->full_open_options);
         bs->full_open_options = NULL;
     }
 
-    if (bs->blk) {
-        blk_dev_change_media_cb(bs->blk, false);
-    }
-
     QLIST_FOREACH_SAFE(ban, &bs->aio_notifiers, list, ban_next) {
         g_free(ban);
     }
@@ -1978,179 +2243,58 @@ void bdrv_make_anon(BlockDriverState *bs)
     bs->node_name[0] = '\0';
 }
 
-static void bdrv_rebind(BlockDriverState *bs)
-{
-    if (bs->drv && bs->drv->bdrv_rebind) {
-        bs->drv->bdrv_rebind(bs);
-    }
-}
-
-/* Fields that need to stay with the top-level BDS, no matter whether the
- * address of the top-level BDS stays the same or not. */
+/* Fields that need to stay with the top-level BDS */
 static void bdrv_move_feature_fields(BlockDriverState *bs_dest,
                                      BlockDriverState *bs_src)
 {
     /* move some fields that need to stay attached to the device */
 
     /* dev info */
-    bs_dest->guest_block_size   = bs_src->guest_block_size;
     bs_dest->copy_on_read       = bs_src->copy_on_read;
 
     bs_dest->enable_write_cache = bs_src->enable_write_cache;
 
-    /* i/o throttled req */
-    bs_dest->throttle_state     = bs_src->throttle_state,
-    bs_dest->io_limits_enabled  = bs_src->io_limits_enabled;
-    bs_dest->pending_reqs[0]    = bs_src->pending_reqs[0];
-    bs_dest->pending_reqs[1]    = bs_src->pending_reqs[1];
-    bs_dest->throttled_reqs[0]  = bs_src->throttled_reqs[0];
-    bs_dest->throttled_reqs[1]  = bs_src->throttled_reqs[1];
-    memcpy(&bs_dest->round_robin,
-           &bs_src->round_robin,
-           sizeof(bs_dest->round_robin));
-    memcpy(&bs_dest->throttle_timers,
-           &bs_src->throttle_timers,
-           sizeof(ThrottleTimers));
-
-    /* r/w error */
-    bs_dest->on_read_error      = bs_src->on_read_error;
-    bs_dest->on_write_error     = bs_src->on_write_error;
-
-    /* i/o status */
-    bs_dest->iostatus_enabled   = bs_src->iostatus_enabled;
-    bs_dest->iostatus           = bs_src->iostatus;
-
     /* dirty bitmap */
     bs_dest->dirty_bitmaps      = bs_src->dirty_bitmaps;
 }
 
-/* Fields that only need to be swapped if the contents of BDSes is swapped
- * rather than pointers being changed in the parents. */
-static void bdrv_move_reference_fields(BlockDriverState *bs_dest,
-                                       BlockDriverState *bs_src)
+static void change_parent_backing_link(BlockDriverState *from,
+                                       BlockDriverState *to)
 {
-    /* reference count */
-    bs_dest->refcnt             = bs_src->refcnt;
-
-    /* job */
-    bs_dest->job                = bs_src->job;
-
-    /* keep the same entry in bdrv_states */
-    bs_dest->device_list = bs_src->device_list;
-    bs_dest->blk = bs_src->blk;
+    BdrvChild *c, *next;
 
-    memcpy(bs_dest->op_blockers, bs_src->op_blockers,
-           sizeof(bs_dest->op_blockers));
+    QLIST_FOREACH_SAFE(c, &from->parents, next_parent, next) {
+        assert(c->role != &child_backing);
+        c->bs = to;
+        QLIST_REMOVE(c, next_parent);
+        QLIST_INSERT_HEAD(&to->parents, c, next_parent);
+        bdrv_ref(to);
+        bdrv_unref(from);
+    }
+    if (from->blk) {
+        blk_set_bs(from->blk, to);
+        if (!to->device_list.tqe_prev) {
+            QTAILQ_INSERT_BEFORE(from, to, device_list);
+        }
+        QTAILQ_REMOVE(&bdrv_states, from, device_list);
+    }
 }
 
-/*
- * Swap bs contents for two image chains while they are live,
- * while keeping required fields on the BlockDriverState that is
- * actually attached to a device.
- *
- * This will modify the BlockDriverState fields, and swap contents
- * between bs_new and bs_old. Both bs_new and bs_old are modified.
- *
- * bs_new must not be attached to a BlockBackend.
- *
- * This function does not create any image files.
- */
-void bdrv_swap(BlockDriverState *bs_new, BlockDriverState *bs_old)
+static void swap_feature_fields(BlockDriverState *bs_top,
+                                BlockDriverState *bs_new)
 {
     BlockDriverState tmp;
-    BdrvChild *child;
-
-    bdrv_drain(bs_new);
-    bdrv_drain(bs_old);
-
-    /* The code needs to swap the node_name but simply swapping node_list won't
-     * work so first remove the nodes from the graph list, do the swap then
-     * insert them back if needed.
-     */
-    if (bs_new->node_name[0] != '\0') {
-        QTAILQ_REMOVE(&graph_bdrv_states, bs_new, node_list);
-    }
-    if (bs_old->node_name[0] != '\0') {
-        QTAILQ_REMOVE(&graph_bdrv_states, bs_old, node_list);
-    }
-
-    /* If the BlockDriverState is part of a throttling group acquire
-     * its lock since we're going to mess with the protected fields.
-     * Otherwise there's no need to worry since no one else can touch
-     * them. */
-    if (bs_old->throttle_state) {
-        throttle_group_lock(bs_old);
-    }
 
-    /* bs_new must be unattached and shouldn't have anything fancy enabled */
-    assert(!bs_new->blk);
-    assert(QLIST_EMPTY(&bs_new->dirty_bitmaps));
-    assert(bs_new->job == NULL);
-    assert(bs_new->io_limits_enabled == false);
-    assert(bs_new->throttle_state == NULL);
-    assert(!throttle_timers_are_initialized(&bs_new->throttle_timers));
-
-    tmp = *bs_new;
-    *bs_new = *bs_old;
-    *bs_old = tmp;
-
-    /* there are some fields that should not be swapped, move them back */
-    bdrv_move_feature_fields(&tmp, bs_old);
-    bdrv_move_feature_fields(bs_old, bs_new);
+    bdrv_move_feature_fields(&tmp, bs_top);
+    bdrv_move_feature_fields(bs_top, bs_new);
     bdrv_move_feature_fields(bs_new, &tmp);
 
-    bdrv_move_reference_fields(&tmp, bs_old);
-    bdrv_move_reference_fields(bs_old, bs_new);
-    bdrv_move_reference_fields(bs_new, &tmp);
-
-    /* bs_new must remain unattached */
-    assert(!bs_new->blk);
-
-    /* Check a few fields that should remain attached to the device */
-    assert(bs_new->job == NULL);
-    assert(bs_new->io_limits_enabled == false);
-    assert(bs_new->throttle_state == NULL);
-    assert(!throttle_timers_are_initialized(&bs_new->throttle_timers));
-
-    /* Release the ThrottleGroup lock */
-    if (bs_old->throttle_state) {
-        throttle_group_unlock(bs_old);
+    assert(!bs_new->throttle_state);
+    if (bs_top->throttle_state) {
+        assert(bs_top->io_limits_enabled);
+        bdrv_io_limits_enable(bs_new, throttle_group_get_name(bs_top));
+        bdrv_io_limits_disable(bs_top);
     }
-
-    /* insert the nodes back into the graph node list if needed */
-    if (bs_new->node_name[0] != '\0') {
-        QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs_new, node_list);
-    }
-    if (bs_old->node_name[0] != '\0') {
-        QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs_old, node_list);
-    }
-
-    /*
-     * Update lh_first.le_prev for non-empty lists.
-     *
-     * The head of the op blocker list doesn't change because it is moved back
-     * in bdrv_move_feature_fields().
-     */
-    assert(QLIST_EMPTY(&bs_old->tracked_requests));
-    assert(QLIST_EMPTY(&bs_new->tracked_requests));
-
-    QLIST_FIX_HEAD_PTR(&bs_new->children, next);
-    QLIST_FIX_HEAD_PTR(&bs_old->children, next);
-
-    /* Update references in bs->opaque and children */
-    QLIST_FOREACH(child, &bs_old->children, next) {
-        if (child->bs->inherits_from == bs_new) {
-            child->bs->inherits_from = bs_old;
-        }
-    }
-    QLIST_FOREACH(child, &bs_new->children, next) {
-        if (child->bs->inherits_from == bs_old) {
-            child->bs->inherits_from = bs_new;
-        }
-    }
-
-    bdrv_rebind(bs_new);
-    bdrv_rebind(bs_old);
 }
 
 /*
@@ -2163,14 +2307,59 @@ void bdrv_swap(BlockDriverState *bs_new, BlockDriverState *bs_old)
  * bs_new must not be attached to a BlockBackend.
  *
  * This function does not create any image files.
+ *
+ * bdrv_append() takes ownership of a bs_new reference and unrefs it because
+ * that's what the callers commonly need. bs_new will be referenced by the old
+ * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
+ * reference of its own, it must call bdrv_ref().
  */
 void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top)
 {
-    bdrv_swap(bs_new, bs_top);
+    assert(!bdrv_requests_pending(bs_top));
+    assert(!bdrv_requests_pending(bs_new));
+
+    bdrv_ref(bs_top);
+    change_parent_backing_link(bs_top, bs_new);
+
+    /* Some fields always stay on top of the backing file chain */
+    swap_feature_fields(bs_top, bs_new);
+
+    bdrv_set_backing_hd(bs_new, bs_top);
+    bdrv_unref(bs_top);
 
-    /* The contents of 'tmp' will become bs_top, as we are
-     * swapping bs_new and bs_top contents. */
-    bdrv_set_backing_hd(bs_top, bs_new);
+    /* bs_new is now referenced by its new parents, we don't need the
+     * additional reference any more. */
+    bdrv_unref(bs_new);
+}
+
+void bdrv_replace_in_backing_chain(BlockDriverState *old, BlockDriverState *new)
+{
+    assert(!bdrv_requests_pending(old));
+    assert(!bdrv_requests_pending(new));
+
+    bdrv_ref(old);
+
+    if (old->blk) {
+        /* As long as these fields aren't in BlockBackend, but in the top-level
+         * BlockDriverState, it's not possible for a BDS to have two BBs.
+         *
+         * We really want to copy the fields from old to new, but we go for a
+         * swap instead so that pointers aren't duplicated and cause trouble.
+         * (Also, bdrv_swap() used to do the same.) */
+        assert(!new->blk);
+        swap_feature_fields(old, new);
+    }
+    change_parent_backing_link(old, new);
+
+    /* Change backing files if a previously independent node is added to the
+     * chain. For active commit, we replace top by its own (indirect) backing
+     * file and don't do anything here so we don't build a loop. */
+    if (new->backing == NULL && !bdrv_chain_contains(backing_bs(old), new)) {
+        bdrv_set_backing_hd(new, backing_bs(old));
+        bdrv_set_backing_hd(old, NULL);
+    }
+
+    bdrv_unref(old);
 }
 
 static void bdrv_delete(BlockDriverState *bs)
@@ -2551,82 +2740,6 @@ void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
     *nb_sectors_ptr = nb_sectors < 0 ? 0 : nb_sectors;
 }
 
-void bdrv_set_on_error(BlockDriverState *bs, BlockdevOnError on_read_error,
-                       BlockdevOnError on_write_error)
-{
-    bs->on_read_error = on_read_error;
-    bs->on_write_error = on_write_error;
-}
-
-BlockdevOnError bdrv_get_on_error(BlockDriverState *bs, bool is_read)
-{
-    return is_read ? bs->on_read_error : bs->on_write_error;
-}
-
-BlockErrorAction bdrv_get_error_action(BlockDriverState *bs, bool is_read, int error)
-{
-    BlockdevOnError on_err = is_read ? bs->on_read_error : bs->on_write_error;
-
-    switch (on_err) {
-    case BLOCKDEV_ON_ERROR_ENOSPC:
-        return (error == ENOSPC) ?
-               BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT;
-    case BLOCKDEV_ON_ERROR_STOP:
-        return BLOCK_ERROR_ACTION_STOP;
-    case BLOCKDEV_ON_ERROR_REPORT:
-        return BLOCK_ERROR_ACTION_REPORT;
-    case BLOCKDEV_ON_ERROR_IGNORE:
-        return BLOCK_ERROR_ACTION_IGNORE;
-    default:
-        abort();
-    }
-}
-
-static void send_qmp_error_event(BlockDriverState *bs,
-                                 BlockErrorAction action,
-                                 bool is_read, int error)
-{
-    IoOperationType optype;
-
-    optype = is_read ? IO_OPERATION_TYPE_READ : IO_OPERATION_TYPE_WRITE;
-    qapi_event_send_block_io_error(bdrv_get_device_name(bs), optype, action,
-                                   bdrv_iostatus_is_enabled(bs),
-                                   error == ENOSPC, strerror(error),
-                                   &error_abort);
-}
-
-/* This is done by device models because, while the block layer knows
- * about the error, it does not know whether an operation comes from
- * the device or the block layer (from a job, for example).
- */
-void bdrv_error_action(BlockDriverState *bs, BlockErrorAction action,
-                       bool is_read, int error)
-{
-    assert(error >= 0);
-
-    if (action == BLOCK_ERROR_ACTION_STOP) {
-        /* First set the iostatus, so that "info block" returns an iostatus
-         * that matches the events raised so far (an additional error iostatus
-         * is fine, but not a lost one).
-         */
-        bdrv_iostatus_set_err(bs, error);
-
-        /* Then raise the request to stop the VM and the event.
-         * qemu_system_vmstop_request_prepare has two effects.  First,
-         * it ensures that the STOP event always comes after the
-         * BLOCK_IO_ERROR event.  Second, it ensures that even if management
-         * can observe the STOP event and do a "cont" before the STOP
-         * event is issued, the VM will not stop.  In this case, vm_start()
-         * also ensures that the STOP/RESUME pair of events is emitted.
-         */
-        qemu_system_vmstop_request_prepare();
-        send_qmp_error_event(bs, action, is_read, error);
-        qemu_system_vmstop_request(RUN_STATE_IO_ERROR);
-    } else {
-        send_qmp_error_event(bs, action, is_read, error);
-    }
-}
-
 int bdrv_is_read_only(BlockDriverState *bs)
 {
     return bs->read_only;
@@ -2820,7 +2933,12 @@ BlockDriverState *bdrv_lookup_bs(const char *device,
         blk = blk_by_name(device);
 
         if (blk) {
-            return blk_bs(blk);
+            bs = blk_bs(blk);
+            if (!bs) {
+                error_setg(errp, "Device '%s' has no medium", device);
+            }
+
+            return bs;
         }
     }
 
@@ -2978,7 +3096,7 @@ ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs)
     return NULL;
 }
 
-void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
+void bdrv_debug_event(BlockDriverState *bs, BlkdebugEvent event)
 {
     if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) {
         return;
@@ -3190,15 +3308,23 @@ void bdrv_invalidate_cache_all(Error **errp)
 /**
  * Return TRUE if the media is present
  */
-int bdrv_is_inserted(BlockDriverState *bs)
+bool bdrv_is_inserted(BlockDriverState *bs)
 {
     BlockDriver *drv = bs->drv;
+    BdrvChild *child;
 
-    if (!drv)
-        return 0;
-    if (!drv->bdrv_is_inserted)
-        return 1;
-    return drv->bdrv_is_inserted(bs);
+    if (!drv) {
+        return false;
+    }
+    if (drv->bdrv_is_inserted) {
+        return drv->bdrv_is_inserted(bs);
+    }
+    QLIST_FOREACH(child, &bs->children, next) {
+        if (!bdrv_is_inserted(child->bs)) {
+            return false;
+        }
+    }
+    return true;
 }
 
 /**
@@ -3249,11 +3375,6 @@ void bdrv_lock_medium(BlockDriverState *bs, bool locked)
     }
 }
 
-void bdrv_set_guest_block_size(BlockDriverState *bs, int align)
-{
-    bs->guest_block_size = align;
-}
-
 BdrvDirtyBitmap *bdrv_find_dirty_bitmap(BlockDriverState *bs, const char *name)
 {
     BdrvDirtyBitmap *bm;
@@ -3528,10 +3649,25 @@ void bdrv_reset_dirty_bitmap(BdrvDirtyBitmap *bitmap,
     hbitmap_reset(bitmap->bitmap, cur_sector, nr_sectors);
 }
 
-void bdrv_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap)
+void bdrv_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap **out)
+{
+    assert(bdrv_dirty_bitmap_enabled(bitmap));
+    if (!out) {
+        hbitmap_reset_all(bitmap->bitmap);
+    } else {
+        HBitmap *backup = bitmap->bitmap;
+        bitmap->bitmap = hbitmap_alloc(bitmap->size,
+                                       hbitmap_granularity(backup));
+        *out = backup;
+    }
+}
+
+void bdrv_undo_clear_dirty_bitmap(BdrvDirtyBitmap *bitmap, HBitmap *in)
 {
+    HBitmap *tmp = bitmap->bitmap;
     assert(bdrv_dirty_bitmap_enabled(bitmap));
-    hbitmap_reset_all(bitmap->bitmap);
+    bitmap->bitmap = in;
+    hbitmap_free(tmp);
 }
 
 void bdrv_set_dirty(BlockDriverState *bs, int64_t cur_sector,
@@ -3651,46 +3787,6 @@ bool bdrv_op_blocker_is_empty(BlockDriverState *bs)
     return true;
 }
 
-void bdrv_iostatus_enable(BlockDriverState *bs)
-{
-    bs->iostatus_enabled = true;
-    bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
-}
-
-/* The I/O status is only enabled if the drive explicitly
- * enables it _and_ the VM is configured to stop on errors */
-bool bdrv_iostatus_is_enabled(const BlockDriverState *bs)
-{
-    return (bs->iostatus_enabled &&
-           (bs->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC ||
-            bs->on_write_error == BLOCKDEV_ON_ERROR_STOP   ||
-            bs->on_read_error == BLOCKDEV_ON_ERROR_STOP));
-}
-
-void bdrv_iostatus_disable(BlockDriverState *bs)
-{
-    bs->iostatus_enabled = false;
-}
-
-void bdrv_iostatus_reset(BlockDriverState *bs)
-{
-    if (bdrv_iostatus_is_enabled(bs)) {
-        bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK;
-        if (bs->job) {
-            block_job_iostatus_reset(bs->job);
-        }
-    }
-}
-
-void bdrv_iostatus_set_err(BlockDriverState *bs, int error)
-{
-    assert(bdrv_iostatus_is_enabled(bs));
-    if (bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
-        bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE :
-                                         BLOCK_DEVICE_IO_STATUS_FAILED;
-    }
-}
-
 void bdrv_img_create(const char *filename, const char *fmt,
                      const char *base_filename, const char *base_fmt,
                      char *options, uint64_t img_size, int flags,
@@ -3875,7 +3971,7 @@ void bdrv_detach_aio_context(BlockDriverState *bs)
         baf->detach_aio_context(baf->opaque);
     }
 
-    if (bs->io_limits_enabled) {
+    if (bs->throttle_state) {
         throttle_timers_detach_aio_context(&bs->throttle_timers);
     }
     if (bs->drv->bdrv_detach_aio_context) {
@@ -3911,7 +4007,7 @@ void bdrv_attach_aio_context(BlockDriverState *bs,
     if (bs->drv->bdrv_attach_aio_context) {
         bs->drv->bdrv_attach_aio_context(bs, new_context);
     }
-    if (bs->io_limits_enabled) {
+    if (bs->throttle_state) {
         throttle_timers_attach_aio_context(&bs->throttle_timers, new_context);
     }
 
@@ -3972,12 +4068,12 @@ void bdrv_remove_aio_context_notifier(BlockDriverState *bs,
 }
 
 int bdrv_amend_options(BlockDriverState *bs, QemuOpts *opts,
-                       BlockDriverAmendStatusCB *status_cb)
+                       BlockDriverAmendStatusCB *status_cb, void *cb_opaque)
 {
     if (!bs->drv->bdrv_amend_options) {
         return -ENOTSUP;
     }
-    return bs->drv->bdrv_amend_options(bs, opts, status_cb);
+    return bs->drv->bdrv_amend_options(bs, opts, status_cb, cb_opaque);
 }
 
 /* This function will be called by the bdrv_recurse_is_first_non_filter method
@@ -4075,20 +4171,39 @@ out:
 static bool append_open_options(QDict *d, BlockDriverState *bs)
 {
     const QDictEntry *entry;
+    QemuOptDesc *desc;
+    BdrvChild *child;
     bool found_any = false;
+    const char *p;
 
     for (entry = qdict_first(bs->options); entry;
          entry = qdict_next(bs->options, entry))
     {
-        /* Only take options for this level and exclude all non-driver-specific
-         * options */
-        if (!strchr(qdict_entry_key(entry), '.') &&
-            strcmp(qdict_entry_key(entry), "node-name"))
-        {
-            qobject_incref(qdict_entry_value(entry));
-            qdict_put_obj(d, qdict_entry_key(entry), qdict_entry_value(entry));
-            found_any = true;
+        /* Exclude options for children */
+        QLIST_FOREACH(child, &bs->children, next) {
+            if (strstart(qdict_entry_key(entry), child->name, &p)
+                && (!*p || *p == '.'))
+            {
+                break;
+            }
         }
+        if (child) {
+            continue;
+        }
+
+        /* And exclude all non-driver-specific options */
+        for (desc = bdrv_runtime_opts.desc; desc->name; desc++) {
+            if (!strcmp(qdict_entry_key(entry), desc->name)) {
+                break;
+            }
+        }
+        if (desc->name) {
+            continue;
+        }
+
+        qobject_incref(qdict_entry_value(entry));
+        qdict_put_obj(d, qdict_entry_key(entry), qdict_entry_value(entry));
+        found_any = true;
     }
 
     return found_any;
@@ -4130,7 +4245,10 @@ void bdrv_refresh_filename(BlockDriverState *bs)
             bs->full_open_options = NULL;
         }
 
-        drv->bdrv_refresh_filename(bs);
+        opts = qdict_new();
+        append_open_options(opts, bs);
+        drv->bdrv_refresh_filename(bs, opts);
+        QDECREF(opts);
     } else if (bs->file) {
         /* Try to reconstruct valid information from the underlying file */
         bool has_open_options;
@@ -4202,14 +4320,3 @@ void bdrv_refresh_filename(BlockDriverState *bs)
         QDECREF(json);
     }
 }
-
-/* This accessor function purpose is to allow the device models to access the
- * BlockAcctStats structure embedded inside a BlockDriverState without being
- * aware of the BlockDriverState structure layout.
- * It will go away when the BlockAcctStats structure will be moved inside
- * the device models.
- */
-BlockAcctStats *bdrv_get_stats(BlockDriverState *bs)
-{
-    return &bs->stats;
-}