[PATCH v5 0/9] Allow changing bs->file on reopen

[PATCH v5 0/9] Allow changing bs->file on reopen

[Vladimir Sementsov-Ogievskiy]

Hi all!

Alberto, I'm sorry for not consulting with you about sending this.

Actually, I wanted only one patch
"block: Allow changing bs->file on reopen", as I'm going to send another
series based on it. I started to work on updating this one patch it and
it turned into a refactoring series below. So I decided to add you patch
with test cases, for this series to be a complete* v5. I hope you'll
have time to work on the rest of your series on top of this my
suggestion :)

[*], supporting multiple reopen and final publishing x-blockdev-reopen
not included here. Also I didn't address Kevin's comment that we should
have some error-path testing of new feature. Probably I didn't address
some comments on "Allow changing" patch, but it changed so much that
review should start from scratch anyway.

v5:
1-7: new
8: changed a lot
9: unchanged

Alberto Garcia (2):
  block: Allow changing bs->file on reopen
  iotests: Test replacing files with x-blockdev-reopen

Vladimir Sementsov-Ogievskiy (7):
  block: introduce bdrv_remove_file_or_backing_child()
  block: introduce bdrv_set_file_or_backing_noperm()
  block: bdrv_reopen_parse_backing(): don't check aio context
  block: bdrv_reopen_parse_backing(): don't check frozen child
  block: bdrv_reopen_parse_backing(): simplify handling implicit filters
  block: move supports_backing check to
    bdrv_set_file_or_backing_noperm()
  block: BDRVReopenState: drop replace_backing_bs field

 include/block/block.h            |   2 +-
 block.c                          | 287 ++++++++++++++++---------------
 tests/unit/test-bdrv-drain.c     |   1 +
 tests/unit/test-bdrv-graph-mod.c |   1 +
 tests/qemu-iotests/245           | 140 +++++++++++++--
 tests/qemu-iotests/245.out       |  11 +-
 6 files changed, 287 insertions(+), 155 deletions(-)

-- 
2.29.2

[PATCH v5 1/9] block: introduce bdrv_remove_file_or_backing_child()

[Vladimir Sementsov-Ogievskiy]

To be used for reopen in future commit.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
---
 block.c | 25 +++++++++++++++++--------
 1 file changed, 17 insertions(+), 8 deletions(-)

diff --git a/block.c b/block.c
index 9ad725d205..75565ce4d4 100644
--- a/block.c
+++ b/block.c
@@ -4770,16 +4770,14 @@ static TransactionActionDrv bdrv_remove_filter_or_cow_child_drv = {
     .clean = g_free,
 };
 
-/*
- * A function to remove backing-chain child of @bs if exists: cow child for
- * format nodes (always .backing) and filter child for filters (may be .file or
- * .backing)
- */
-static void bdrv_remove_filter_or_cow_child(BlockDriverState *bs,
-                                            Transaction *tran)
+/* A function to remove backing or file child of @bs */
+static void bdrv_remove_file_or_backing_child(BlockDriverState *bs,
+                                              BdrvChild *child,
+                                              Transaction *tran)
 {
     BdrvRemoveFilterOrCowChild *s;
-    BdrvChild *child = bdrv_filter_or_cow_child(bs);
+
+    assert(child == bs->backing || child == bs->file);
 
     if (!child) {
         return;
@@ -4804,6 +4802,17 @@ static void bdrv_remove_filter_or_cow_child(BlockDriverState *bs,
     }
 }
 
+/*
+ * A function to remove backing-chain child of @bs if exists: cow child for
+ * format nodes (always .backing) and filter child for filters (may be .file or
+ * .backing)
+ */
+static void bdrv_remove_filter_or_cow_child(BlockDriverState *bs,
+                                            Transaction *tran)
+{
+    bdrv_remove_file_or_backing_child(bs, bdrv_filter_or_cow_child(bs), tran);
+}
+
 static int bdrv_replace_node_noperm(BlockDriverState *from,
                                     BlockDriverState *to,
                                     bool auto_skip, Transaction *tran,
-- 
2.29.2

[PATCH v5 2/9] block: introduce bdrv_set_file_or_backing_noperm()

[Vladimir Sementsov-Ogievskiy]

To be used for reopen in future commit.

Notes:
 - It seems OK to update inherits_from if new bs is recursively inherits
 from parent bs. Let's just not check for backing_chain_contains, to
 support file child of non-filters.

 - Simply check child->frozen instead of
   bdrv_is_backing_chain_frozen(), as we really interested only in this
   one child.

 - Role determination of new child is a bit more complex: it remains
   the same for backing child, it's obvious for filter driver. But for
   non-filter file child let's for now restrict to only replacing
   existing child (and keeping its role).

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
---
 block.c | 83 +++++++++++++++++++++++++++++++++++++++++++--------------
 1 file changed, 63 insertions(+), 20 deletions(-)

diff --git a/block.c b/block.c
index 75565ce4d4..5141c04815 100644
--- a/block.c
+++ b/block.c
@@ -92,6 +92,9 @@ static int bdrv_attach_child_noperm(BlockDriverState *parent_bs,
                                     BdrvChild **child,
                                     Transaction *tran,
                                     Error **errp);
+static void bdrv_remove_file_or_backing_child(BlockDriverState *bs,
+                                              BdrvChild *child,
+                                              Transaction *tran);
 static void bdrv_remove_filter_or_cow_child(BlockDriverState *bs,
                                             Transaction *tran);
 
@@ -3094,54 +3097,94 @@ static BdrvChildRole bdrv_backing_role(BlockDriverState *bs)
 }
 
 /*
- * Sets the bs->backing link of a BDS. A new reference is created; callers
- * which don't need their own reference any more must call bdrv_unref().
+ * Sets the bs->backing or bs->file link of a BDS. A new reference is created;
+ * callers which don't need their own reference any more must call bdrv_unref().
  */
-static int bdrv_set_backing_noperm(BlockDriverState *bs,
-                                   BlockDriverState *backing_hd,
-                                   Transaction *tran, Error **errp)
+static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs,
+                                           BlockDriverState *child_bs,
+                                           bool is_backing,
+                                           Transaction *tran, Error **errp)
 {
     int ret = 0;
-    bool update_inherits_from = bdrv_chain_contains(bs, backing_hd) &&
-        bdrv_inherits_from_recursive(backing_hd, bs);
+    bool update_inherits_from =
+        bdrv_inherits_from_recursive(child_bs, parent_bs);
+    BdrvChild *child = is_backing ? parent_bs->backing : parent_bs->file;
+    BdrvChildRole role;
 
-    if (bdrv_is_backing_chain_frozen(bs, child_bs(bs->backing), errp)) {
+    if (!parent_bs->drv) {
+        /*
+         * Node without drv is an object without a class :/. TODO: finally fix
+         * qcow2 driver to never clear bs->drv and implement format corruption
+         * handling in other way.
+         */
+        error_setg(errp, "Node corrupted");
+        return -EINVAL;
+    }
+
+    if (child && child->frozen) {
+        error_setg(errp, "Cannot change frozen '%s' link from '%s' to '%s'",
+                   child->name, parent_bs->node_name, child->bs->node_name);
         return -EPERM;
     }
 
-    if (bs->backing) {
-        /* Cannot be frozen, we checked that above */
-        bdrv_unset_inherits_from(bs, bs->backing, tran);
-        bdrv_remove_filter_or_cow_child(bs, tran);
+    if (parent_bs->drv->is_filter) {
+        role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
+    } else if (is_backing) {
+        role = BDRV_CHILD_COW;
+    } else {
+        /*
+         * We only can use same role as it is in existing child. We don't have
+         * infrastructure to determine role of file child in generic way
+         */
+        if (!child) {
+            error_setg(errp, "Cannot set file child to format node without "
+                       "file child");
+            return -EINVAL;
+        }
+        role = child->role;
     }
 
-    if (!backing_hd) {
+    if (child) {
+        bdrv_unset_inherits_from(parent_bs, child, tran);
+        bdrv_remove_file_or_backing_child(parent_bs, child, tran);
+    }
+
+    if (!child_bs) {
         goto out;
     }
 
-    ret = bdrv_attach_child_noperm(bs, backing_hd, "backing",
-                                   &child_of_bds, bdrv_backing_role(bs),
-                                   &bs->backing, tran, errp);
+    ret = bdrv_attach_child_noperm(parent_bs, child_bs,
+                                   is_backing ? "backing" : "file",
+                                   &child_of_bds, role,
+                                   is_backing ? &parent_bs->backing :
+                                                &parent_bs->file,
+                                   tran, errp);
     if (ret < 0) {
         return ret;
     }
 
 
     /*
-     * If backing_hd was already part of bs's backing chain, and
-     * inherits_from pointed recursively to bs then let's update it to
+     * If inherits_from pointed recursively to bs then let's update it to
      * point directly to bs (else it will become NULL).
      */
     if (update_inherits_from) {
-        bdrv_set_inherits_from(backing_hd, bs, tran);
+        bdrv_set_inherits_from(child_bs, parent_bs, tran);
     }
 
 out:
-    bdrv_refresh_limits(bs, tran, NULL);
+    bdrv_refresh_limits(parent_bs, tran, NULL);
 
     return 0;
 }
 
+static int bdrv_set_backing_noperm(BlockDriverState *bs,
+                                   BlockDriverState *backing_hd,
+                                   Transaction *tran, Error **errp)
+{
+    return bdrv_set_file_or_backing_noperm(bs, backing_hd, true, tran, errp);
+}
+
 int bdrv_set_backing_hd(BlockDriverState *bs, BlockDriverState *backing_hd,
                         Error **errp)
 {
-- 
2.29.2

[PATCH v5 3/9] block: bdrv_reopen_parse_backing(): don't check aio context

[Vladimir Sementsov-Ogievskiy]

We don't need this check: bdrv_set_backing_noperm() will do it anyway
(actually in bdrv_attach_child_common()).

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
---
 block.c | 33 ---------------------------------
 1 file changed, 33 deletions(-)

diff --git a/block.c b/block.c
index 5141c04815..4e18bc3177 100644
--- a/block.c
+++ b/block.c
@@ -4190,29 +4190,6 @@ int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
     return ret;
 }
 
-static bool bdrv_reopen_can_attach(BlockDriverState *parent,
-                                   BdrvChild *child,
-                                   BlockDriverState *new_child,
-                                   Error **errp)
-{
-    AioContext *parent_ctx = bdrv_get_aio_context(parent);
-    AioContext *child_ctx = bdrv_get_aio_context(new_child);
-    GSList *ignore;
-    bool ret;
-
-    ignore = g_slist_prepend(NULL, child);
-    ret = bdrv_can_set_aio_context(new_child, parent_ctx, &ignore, NULL);
-    g_slist_free(ignore);
-    if (ret) {
-        return ret;
-    }
-
-    ignore = g_slist_prepend(NULL, child);
-    ret = bdrv_can_set_aio_context(parent, child_ctx, &ignore, errp);
-    g_slist_free(ignore);
-    return ret;
-}
-
 /*
  * Take a BDRVReopenState and check if the value of 'backing' in the
  * reopen_state->options QDict is valid or not.
@@ -4264,16 +4241,6 @@ static int bdrv_reopen_parse_backing(BDRVReopenState *reopen_state,
         g_assert_not_reached();
     }
 
-    /*
-     * Check AioContext compatibility so that the bdrv_set_backing_hd() call in
-     * bdrv_reopen_commit() won't fail.
-     */
-    if (new_backing_bs) {
-        if (!bdrv_reopen_can_attach(bs, bs->backing, new_backing_bs, errp)) {
-            return -EINVAL;
-        }
-    }
-
     /*
      * Ensure that @bs can really handle backing files, because we are
      * about to give it one (or swap the existing one)
-- 
2.29.2

[PATCH v5 4/9] block: bdrv_reopen_parse_backing(): don't check frozen child

[Vladimir Sementsov-Ogievskiy]

bdrv_set_backing_noperm() takes care of it (actual check is in
bdrv_set_file_or_backing_noperm()), so we don't need to check it here.

While being here, improve error message a bit.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
---
 block.c                | 14 +-------------
 tests/qemu-iotests/245 |  8 ++++----
 2 files changed, 5 insertions(+), 17 deletions(-)

diff --git a/block.c b/block.c
index 4e18bc3177..bfa497813b 100644
--- a/block.c
+++ b/block.c
@@ -4283,19 +4283,7 @@ static int bdrv_reopen_parse_backing(BDRVReopenState *reopen_state,
                        "an implicit backing file", bs->node_name);
             return -EPERM;
         }
-        /*
-         * Check if the backing link that we want to replace is frozen.
-         * Note that
-         * bdrv_filter_or_cow_child(overlay_bs) == overlay_bs->backing,
-         * because we know that overlay_bs == bs, and that @bs
-         * either is a filter that uses ->backing or a COW format BDS
-         * with bs->drv->supports_backing == true.
-         */
-        if (bdrv_is_backing_chain_frozen(overlay_bs,
-                                         child_bs(overlay_bs->backing), errp))
-        {
-            return -EPERM;
-        }
+
         reopen_state->replace_backing_bs = true;
         reopen_state->old_backing_bs = bs->backing ? bs->backing->bs : NULL;
         ret = bdrv_set_backing_noperm(bs, new_backing_bs, set_backings_tran,
diff --git a/tests/qemu-iotests/245 b/tests/qemu-iotests/245
index fc5297e268..c7d671366a 100755
--- a/tests/qemu-iotests/245
+++ b/tests/qemu-iotests/245
@@ -878,7 +878,7 @@ class TestBlockdevReopen(iotests.QMPTestCase):
 
         # We can't remove hd1 while the stream job is ongoing
         opts['backing'] = None
-        self.reopen(opts, {}, "Cannot change 'backing' link from 'hd0' to 'hd1'")
+        self.reopen(opts, {}, "Cannot change frozen 'backing' link from 'hd0' to 'hd1'")
 
         self.vm.run_job('stream0', auto_finalize = False, auto_dismiss = True)
 
@@ -910,7 +910,7 @@ class TestBlockdevReopen(iotests.QMPTestCase):
         # We can't remove hd2 while the stream job is ongoing
         opts['backing']['backing'] = None
         self.reopen(opts['backing'], {'read-only': False},
-                    "Cannot change 'backing' link from 'hd1' to 'hd2'")
+                    "Cannot change frozen 'backing' link from 'hd1' to 'hd2'")
 
         # We can detach hd1 from hd0 because it doesn't affect the stream job
         opts['backing'] = None
@@ -933,11 +933,11 @@ class TestBlockdevReopen(iotests.QMPTestCase):
 
         # We can't remove hd2 while the commit job is ongoing
         opts['backing']['backing'] = None
-        self.reopen(opts, {}, "Cannot change 'backing' link from 'hd1' to 'hd2'")
+        self.reopen(opts, {}, "Cannot change frozen 'backing' link from 'hd1' to 'hd2'")
 
         # We can't remove hd1 while the commit job is ongoing
         opts['backing'] = None
-        self.reopen(opts, {}, "Cannot change 'backing' link from 'hd0' to 'hd1'")
+        self.reopen(opts, {}, "Cannot change frozen 'backing' link from 'hd0' to 'hd1'")
 
         event = self.vm.event_wait(name='BLOCK_JOB_READY')
         self.assert_qmp(event, 'data/device', 'commit0')
-- 
2.29.2

[PATCH v5 5/9] block: bdrv_reopen_parse_backing(): simplify handling implicit filters

[Vladimir Sementsov-Ogievskiy]

The logic around finding overlay here is not obvious. Actually it does
two simple things:

1. If new bs is already in backing chain, split from parent bs by
   several implicit filters we are done, do nothing.

2. Otherwise, don't try to replace implicit filter.

Let's rewrite this in more obvious way.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
---
 block.c | 53 ++++++++++++++++-------------------------------------
 1 file changed, 16 insertions(+), 37 deletions(-)

diff --git a/block.c b/block.c
index bfa497813b..7d208c708d 100644
--- a/block.c
+++ b/block.c
@@ -4212,7 +4212,7 @@ static int bdrv_reopen_parse_backing(BDRVReopenState *reopen_state,
                                      Error **errp)
 {
     BlockDriverState *bs = reopen_state->bs;
-    BlockDriverState *overlay_bs, *below_bs, *new_backing_bs;
+    BlockDriverState *new_backing_bs;
     QObject *value;
     const char *str;
 
@@ -4241,6 +4241,18 @@ static int bdrv_reopen_parse_backing(BDRVReopenState *reopen_state,
         g_assert_not_reached();
     }
 
+    if (bs->backing) {
+        if (bdrv_skip_implicit_filters(bs->backing->bs) == new_backing_bs) {
+            return 0;
+        }
+
+        if (bs->backing->bs->implicit) {
+            error_setg(errp, "Cannot change backing link if '%s' has "
+                       "an implicit backing file", bs->node_name);
+            return -EPERM;
+        }
+    }
+
     /*
      * Ensure that @bs can really handle backing files, because we are
      * about to give it one (or swap the existing one)
@@ -4258,42 +4270,9 @@ static int bdrv_reopen_parse_backing(BDRVReopenState *reopen_state,
         return -EINVAL;
     }
 
-    /*
-     * Find the "actual" backing file by skipping all links that point
-     * to an implicit node, if any (e.g. a commit filter node).
-     * We cannot use any of the bdrv_skip_*() functions here because
-     * those return the first explicit node, while we are looking for
-     * its overlay here.
-     */
-    overlay_bs = bs;
-    for (below_bs = bdrv_filter_or_cow_bs(overlay_bs);
-         below_bs && below_bs->implicit;
-         below_bs = bdrv_filter_or_cow_bs(overlay_bs))
-    {
-        overlay_bs = below_bs;
-    }
-
-    /* If we want to replace the backing file we need some extra checks */
-    if (new_backing_bs != bdrv_filter_or_cow_bs(overlay_bs)) {
-        int ret;
-
-        /* Check for implicit nodes between bs and its backing file */
-        if (bs != overlay_bs) {
-            error_setg(errp, "Cannot change backing link if '%s' has "
-                       "an implicit backing file", bs->node_name);
-            return -EPERM;
-        }
-
-        reopen_state->replace_backing_bs = true;
-        reopen_state->old_backing_bs = bs->backing ? bs->backing->bs : NULL;
-        ret = bdrv_set_backing_noperm(bs, new_backing_bs, set_backings_tran,
-                                      errp);
-        if (ret < 0) {
-            return ret;
-        }
-    }
-
-    return 0;
+    reopen_state->replace_backing_bs = true;
+    reopen_state->old_backing_bs = bs->backing ? bs->backing->bs : NULL;
+    return bdrv_set_backing_noperm(bs, new_backing_bs, set_backings_tran, errp);
 }
 
 /*
-- 
2.29.2

[PATCH v5 6/9] block: move supports_backing check to bdrv_set_file_or_backing_noperm()

[Vladimir Sementsov-Ogievskiy]

Move supports_backing check of bdrv_reopen_parse_backing to called
(through bdrv_set_backing_noperm()) bdrv_set_file_or_backing_noperm()
function. The check applies to general case, so it's appropriate for
bdrv_set_file_or_backing_noperm().

We have to declare backing support for two test drivers, otherwise new
check fails.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
---
 block.c                          | 29 +++++++++++++++--------------
 tests/unit/test-bdrv-drain.c     |  1 +
 tests/unit/test-bdrv-graph-mod.c |  1 +
 3 files changed, 17 insertions(+), 14 deletions(-)

diff --git a/block.c b/block.c
index 7d208c708d..d90b35f8f1 100644
--- a/block.c
+++ b/block.c
@@ -3127,6 +3127,14 @@ static int bdrv_set_file_or_backing_noperm(BlockDriverState *parent_bs,
         return -EPERM;
     }
 
+    if (is_backing && !parent_bs->drv->is_filter &&
+        !parent_bs->drv->supports_backing)
+    {
+        error_setg(errp, "Driver '%s' of node '%s' does not support backing "
+                   "files", parent_bs->drv->format_name, parent_bs->node_name);
+        return -EINVAL;
+    }
+
     if (parent_bs->drv->is_filter) {
         role = BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY;
     } else if (is_backing) {
@@ -4253,20 +4261,13 @@ static int bdrv_reopen_parse_backing(BDRVReopenState *reopen_state,
         }
     }
 
-    /*
-     * Ensure that @bs can really handle backing files, because we are
-     * about to give it one (or swap the existing one)
-     */
-    if (bs->drv->is_filter) {
-        /* Filters always have a file or a backing child */
-        if (!bs->backing) {
-            error_setg(errp, "'%s' is a %s filter node that does not support a "
-                       "backing child", bs->node_name, bs->drv->format_name);
-            return -EINVAL;
-        }
-    } else if (!bs->drv->supports_backing) {
-        error_setg(errp, "Driver '%s' of node '%s' does not support backing "
-                   "files", bs->drv->format_name, bs->node_name);
+    if (bs->drv->is_filter && !bs->backing) {
+        /*
+         * Filters always have a file or a backing child, so we are trying to
+         * change wrong child
+         */
+        error_setg(errp, "'%s' is a %s filter node that does not support a "
+                   "backing child", bs->node_name, bs->drv->format_name);
         return -EINVAL;
     }
 
diff --git a/tests/unit/test-bdrv-drain.c b/tests/unit/test-bdrv-drain.c
index 892f7f47d8..ce071b5fc5 100644
--- a/tests/unit/test-bdrv-drain.c
+++ b/tests/unit/test-bdrv-drain.c
@@ -95,6 +95,7 @@ static int bdrv_test_change_backing_file(BlockDriverState *bs,
 static BlockDriver bdrv_test = {
     .format_name            = "test",
     .instance_size          = sizeof(BDRVTestState),
+    .supports_backing       = true,
 
     .bdrv_close             = bdrv_test_close,
     .bdrv_co_preadv         = bdrv_test_co_preadv,
diff --git a/tests/unit/test-bdrv-graph-mod.c b/tests/unit/test-bdrv-graph-mod.c
index 88f25c0cdb..a6e3bb79be 100644
--- a/tests/unit/test-bdrv-graph-mod.c
+++ b/tests/unit/test-bdrv-graph-mod.c
@@ -41,6 +41,7 @@ static void no_perm_default_perms(BlockDriverState *bs, BdrvChild *c,
 
 static BlockDriver bdrv_no_perm = {
     .format_name = "no-perm",
+    .supports_backing = true,
     .bdrv_child_perm = no_perm_default_perms,
 };
 
-- 
2.29.2

[PATCH v5 7/9] block: BDRVReopenState: drop replace_backing_bs field

[Vladimir Sementsov-Ogievskiy]

It's used only in bdrv_reopen_commit(). "backing" is covered by the
loop through all children except for case when we removed backing child
during reopen.

Make it more obvious and drop extra boolean field: qdict_del will not
fail if there is no such entry.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
---
 include/block/block.h |  1 -
 block.c               | 10 ++++------
 2 files changed, 4 insertions(+), 7 deletions(-)

diff --git a/include/block/block.h b/include/block/block.h
index 82185965ff..0796f75b49 100644
--- a/include/block/block.h
+++ b/include/block/block.h
@@ -208,7 +208,6 @@ typedef struct BDRVReopenState {
     int flags;
     BlockdevDetectZeroesOptions detect_zeroes;
     bool backing_missing;
-    bool replace_backing_bs;  /* new_backing_bs is ignored if this is false */
     BlockDriverState *old_backing_bs; /* keep pointer for permissions update */
     QDict *options;
     QDict *explicit_options;
diff --git a/block.c b/block.c
index d90b35f8f1..819e133dfa 100644
--- a/block.c
+++ b/block.c
@@ -4271,7 +4271,6 @@ static int bdrv_reopen_parse_backing(BDRVReopenState *reopen_state,
         return -EINVAL;
     }
 
-    reopen_state->replace_backing_bs = true;
     reopen_state->old_backing_bs = bs->backing ? bs->backing->bs : NULL;
     return bdrv_set_backing_noperm(bs, new_backing_bs, set_backings_tran, errp);
 }
@@ -4526,17 +4525,16 @@ static void bdrv_reopen_commit(BDRVReopenState *reopen_state)
     bs->read_only = !(reopen_state->flags & BDRV_O_RDWR);
     bs->detect_zeroes      = reopen_state->detect_zeroes;
 
-    if (reopen_state->replace_backing_bs) {
-        qdict_del(bs->explicit_options, "backing");
-        qdict_del(bs->options, "backing");
-    }
-
     /* Remove child references from bs->options and bs->explicit_options.
      * Child options were already removed in bdrv_reopen_queue_child() */
     QLIST_FOREACH(child, &bs->children, next) {
         qdict_del(bs->explicit_options, child->name);
         qdict_del(bs->options, child->name);
     }
+    /* backing is probably removed, so it's not handled by previous loop */
+    qdict_del(bs->explicit_options, "backing");
+    qdict_del(bs->options, "backing");
+
     bdrv_refresh_limits(bs, NULL, NULL);
 }
 
-- 
2.29.2

[PATCH v5 8/9] block: Allow changing bs->file on reopen

[Vladimir Sementsov-Ogievskiy]

From: Alberto Garcia <berto@igalia.com>

When the x-blockdev-reopen was added it allowed reconfiguring the
graph by replacing backing files, but changing the 'file' option was
forbidden. Because of this restriction some operations are not
possible, notably inserting and removing block filters.

This patch adds support for replacing the 'file' option. This is
similar to replacing the backing file and the user is likewise
responsible for the correctness of the resulting graph, otherwise this
can lead to data corruption.

Signed-off-by: Alberto Garcia <berto@igalia.com>
 [vsementsov: bdrv_reopen_parse_file_or_backing() is modified a lot]
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
---
 include/block/block.h  |  1 +
 block.c                | 78 +++++++++++++++++++++++++++++-------------
 tests/qemu-iotests/245 | 23 +++++++------
 3 files changed, 67 insertions(+), 35 deletions(-)

diff --git a/include/block/block.h b/include/block/block.h
index 0796f75b49..daae24073d 100644
--- a/include/block/block.h
+++ b/include/block/block.h
@@ -209,6 +209,7 @@ typedef struct BDRVReopenState {
     BlockdevDetectZeroesOptions detect_zeroes;
     bool backing_missing;
     BlockDriverState *old_backing_bs; /* keep pointer for permissions update */
+    BlockDriverState *old_file_bs; /* keep pointer for permissions update */
     QDict *options;
     QDict *explicit_options;
     void *opaque;
diff --git a/block.c b/block.c
index 819e133dfa..564443ac32 100644
--- a/block.c
+++ b/block.c
@@ -100,7 +100,7 @@ static void bdrv_remove_filter_or_cow_child(BlockDriverState *bs,
 
 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
                                BlockReopenQueue *queue,
-                               Transaction *set_backings_tran, Error **errp);
+                               Transaction *change_child_tran, Error **errp);
 static void bdrv_reopen_commit(BDRVReopenState *reopen_state);
 static void bdrv_reopen_abort(BDRVReopenState *reopen_state);
 
@@ -4123,6 +4123,10 @@ int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp)
             refresh_list = bdrv_topological_dfs(refresh_list, found,
                                                 state->old_backing_bs);
         }
+        if (state->old_file_bs) {
+            refresh_list = bdrv_topological_dfs(refresh_list, found,
+                                                state->old_file_bs);
+        }
     }
 
     /*
@@ -4215,64 +4219,81 @@ int bdrv_reopen_set_read_only(BlockDriverState *bs, bool read_only,
  *
  * Return 0 on success, otherwise return < 0 and set @errp.
  */
-static int bdrv_reopen_parse_backing(BDRVReopenState *reopen_state,
-                                     Transaction *set_backings_tran,
-                                     Error **errp)
+static int bdrv_reopen_parse_file_or_backing(BDRVReopenState *reopen_state,
+                                             bool is_backing, Transaction *tran,
+                                             Error **errp)
 {
     BlockDriverState *bs = reopen_state->bs;
-    BlockDriverState *new_backing_bs;
+    BlockDriverState *new_child_bs;
+    BlockDriverState *old_child_bs = is_backing ? child_bs(bs->backing) :
+                                                  child_bs(bs->file);
+    const char *child_name = is_backing ? "backing" : "file";
     QObject *value;
     const char *str;
 
-    value = qdict_get(reopen_state->options, "backing");
+    value = qdict_get(reopen_state->options, child_name);
     if (value == NULL) {
         return 0;
     }
 
     switch (qobject_type(value)) {
     case QTYPE_QNULL:
-        new_backing_bs = NULL;
+        assert(is_backing); /* The 'file' option does not allow a null value */
+        new_child_bs = NULL;
         break;
     case QTYPE_QSTRING:
         str = qstring_get_str(qobject_to(QString, value));
-        new_backing_bs = bdrv_lookup_bs(NULL, str, errp);
-        if (new_backing_bs == NULL) {
+        new_child_bs = bdrv_lookup_bs(NULL, str, errp);
+        if (new_child_bs == NULL) {
             return -EINVAL;
-        } else if (bdrv_recurse_has_child(new_backing_bs, bs)) {
-            error_setg(errp, "Making '%s' a backing file of '%s' "
-                       "would create a cycle", str, bs->node_name);
+        } else if (bdrv_recurse_has_child(new_child_bs, bs)) {
+            error_setg(errp, "Making '%s' a %s child of '%s' would create a "
+                       "cycle", str, child_name, bs->node_name);
             return -EINVAL;
         }
         break;
     default:
-        /* 'backing' does not allow any other data type */
+        /*
+         * The options QDict has been flattened, so 'backing' and 'file'
+         * do not allow any other data type here.
+         */
         g_assert_not_reached();
     }
 
-    if (bs->backing) {
-        if (bdrv_skip_implicit_filters(bs->backing->bs) == new_backing_bs) {
+    if (old_child_bs == new_child_bs) {
+        return 0;
+    }
+
+    if (old_child_bs) {
+        if (bdrv_skip_implicit_filters(old_child_bs) == new_child_bs) {
             return 0;
         }
 
-        if (bs->backing->bs->implicit) {
-            error_setg(errp, "Cannot change backing link if '%s' has "
-                       "an implicit backing file", bs->node_name);
+        if (old_child_bs->implicit) {
+            error_setg(errp, "Cannot replace implicit %s child of %s",
+                       child_name, bs->node_name);
             return -EPERM;
         }
     }
 
-    if (bs->drv->is_filter && !bs->backing) {
+    if (bs->drv->is_filter && !old_child_bs) {
         /*
          * Filters always have a file or a backing child, so we are trying to
          * change wrong child
          */
         error_setg(errp, "'%s' is a %s filter node that does not support a "
-                   "backing child", bs->node_name, bs->drv->format_name);
+                   "%s child", bs->node_name, bs->drv->format_name, child_name);
         return -EINVAL;
     }
 
-    reopen_state->old_backing_bs = bs->backing ? bs->backing->bs : NULL;
-    return bdrv_set_backing_noperm(bs, new_backing_bs, set_backings_tran, errp);
+    if (is_backing) {
+        reopen_state->old_backing_bs = old_child_bs;
+    } else {
+        reopen_state->old_file_bs = old_child_bs;
+    }
+
+    return bdrv_set_file_or_backing_noperm(bs, new_child_bs, is_backing,
+                                           tran, errp);
 }
 
 /*
@@ -4294,7 +4315,7 @@ static int bdrv_reopen_parse_backing(BDRVReopenState *reopen_state,
  */
 static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
                                BlockReopenQueue *queue,
-                               Transaction *set_backings_tran, Error **errp)
+                               Transaction *change_child_tran, Error **errp)
 {
     int ret = -1;
     int old_flags;
@@ -4414,12 +4435,21 @@ static int bdrv_reopen_prepare(BDRVReopenState *reopen_state,
      * either a reference to an existing node (using its node name)
      * or NULL to simply detach the current backing file.
      */
-    ret = bdrv_reopen_parse_backing(reopen_state, set_backings_tran, errp);
+    ret = bdrv_reopen_parse_file_or_backing(reopen_state, true,
+                                            change_child_tran, errp);
     if (ret < 0) {
         goto error;
     }
     qdict_del(reopen_state->options, "backing");
 
+    /* Allow changing the 'file' option. In this case NULL is not allowed */
+    ret = bdrv_reopen_parse_file_or_backing(reopen_state, false,
+                                            change_child_tran, errp);
+    if (ret < 0) {
+        goto error;
+    }
+    qdict_del(reopen_state->options, "file");
+
     /* Options that are not handled are only okay if they are unchanged
      * compared to the old state. It is expected that some options are only
      * used for the initial open, but not reopen (e.g. filename) */
diff --git a/tests/qemu-iotests/245 b/tests/qemu-iotests/245
index c7d671366a..87b59666f5 100755
--- a/tests/qemu-iotests/245
+++ b/tests/qemu-iotests/245
@@ -146,8 +146,8 @@ class TestBlockdevReopen(iotests.QMPTestCase):
         self.reopen(opts, {'driver': 'raw'}, "Cannot change the option 'driver'")
         self.reopen(opts, {'driver': ''}, "Invalid parameter ''")
         self.reopen(opts, {'driver': None}, "Invalid parameter type for 'driver', expected: string")
-        self.reopen(opts, {'file': 'not-found'}, "Cannot change the option 'file'")
-        self.reopen(opts, {'file': ''}, "Cannot change the option 'file'")
+        self.reopen(opts, {'file': 'not-found'}, "Cannot find device='' nor node-name='not-found'")
+        self.reopen(opts, {'file': ''}, "Cannot find device='' nor node-name=''")
         self.reopen(opts, {'file': None}, "Invalid parameter type for 'file', expected: BlockdevRef")
         self.reopen(opts, {'file.node-name': 'newname'}, "Cannot change the option 'node-name'")
         self.reopen(opts, {'file.driver': 'host_device'}, "Cannot change the option 'driver'")
@@ -443,7 +443,7 @@ class TestBlockdevReopen(iotests.QMPTestCase):
 
         # Illegal operation: hd2 is a child of hd1
         self.reopen(opts[2], {'backing': 'hd1'},
-                    "Making 'hd1' a backing file of 'hd2' would create a cycle")
+                    "Making 'hd1' a backing child of 'hd2' would create a cycle")
 
         # hd2 <- hd0, hd2 <- hd1
         self.reopen(opts[0], {'backing': 'hd2'})
@@ -454,8 +454,9 @@ class TestBlockdevReopen(iotests.QMPTestCase):
 
         # More illegal operations
         self.reopen(opts[2], {'backing': 'hd1'},
-                    "Making 'hd1' a backing file of 'hd2' would create a cycle")
-        self.reopen(opts[2], {'file': 'hd0-file'}, "Cannot change the option 'file'")
+                    "Making 'hd1' a backing child of 'hd2' would create a cycle")
+        self.reopen(opts[2], {'file': 'hd0-file'},
+                    "Conflicts with use by hd2 as 'file', which does not allow 'write, resize' on hd0-file")
 
         result = self.vm.qmp('blockdev-del', conv_keys = True, node_name = 'hd2')
         self.assert_qmp(result, 'error/class', 'GenericError')
@@ -497,18 +498,18 @@ class TestBlockdevReopen(iotests.QMPTestCase):
 
         # Illegal: hd2 is backed by hd1
         self.reopen(opts[1], {'backing': 'hd2'},
-                    "Making 'hd2' a backing file of 'hd1' would create a cycle")
+                    "Making 'hd2' a backing child of 'hd1' would create a cycle")
 
         # hd1 <- hd0 <- hd2
         self.reopen(opts[2], {'backing': 'hd0'})
 
         # Illegal: hd2 is backed by hd0, which is backed by hd1
         self.reopen(opts[1], {'backing': 'hd2'},
-                    "Making 'hd2' a backing file of 'hd1' would create a cycle")
+                    "Making 'hd2' a backing child of 'hd1' would create a cycle")
 
         # Illegal: hd1 cannot point to itself
         self.reopen(opts[1], {'backing': 'hd1'},
-                    "Making 'hd1' a backing file of 'hd1' would create a cycle")
+                    "Making 'hd1' a backing child of 'hd1' would create a cycle")
 
         # Remove all backing files
         self.reopen(opts[0])
@@ -530,7 +531,7 @@ class TestBlockdevReopen(iotests.QMPTestCase):
 
         # Illegal: hd0 is a child of the blkverify node
         self.reopen(opts[0], {'backing': 'bv'},
-                    "Making 'bv' a backing file of 'hd0' would create a cycle")
+                    "Making 'bv' a backing child of 'hd0' would create a cycle")
 
         # Delete the blkverify node
         result = self.vm.qmp('blockdev-del', conv_keys = True, node_name = 'bv')
@@ -563,7 +564,7 @@ class TestBlockdevReopen(iotests.QMPTestCase):
         # You can't make quorum0 a backing file of hd0:
         # hd0 is already a child of quorum0.
         self.reopen(hd_opts(0), {'backing': 'quorum0'},
-                    "Making 'quorum0' a backing file of 'hd0' would create a cycle")
+                    "Making 'quorum0' a backing child of 'hd0' would create a cycle")
 
         # Delete quorum0
         result = self.vm.qmp('blockdev-del', conv_keys = True, node_name = 'quorum0')
@@ -969,7 +970,7 @@ class TestBlockdevReopen(iotests.QMPTestCase):
 
         # We can't remove hd1 while the commit job is ongoing
         opts['backing'] = None
-        self.reopen(opts, {}, "Cannot change backing link if 'hd0' has an implicit backing file")
+        self.reopen(opts, {}, "Cannot replace implicit backing child of hd0")
 
         # hd2 <- hd0
         self.vm.run_job('commit0', auto_finalize = False, auto_dismiss = True)
-- 
2.29.2

[PATCH v5 9/9] iotests: Test replacing files with x-blockdev-reopen

[Vladimir Sementsov-Ogievskiy]

From: Alberto Garcia <berto@igalia.com>

This patch adds new tests in which we use x-blockdev-reopen to change
bs->file

Signed-off-by: Alberto Garcia <berto@igalia.com>
---
 tests/qemu-iotests/245     | 109 ++++++++++++++++++++++++++++++++++++-
 tests/qemu-iotests/245.out |  11 +++-
 2 files changed, 117 insertions(+), 3 deletions(-)

diff --git a/tests/qemu-iotests/245 b/tests/qemu-iotests/245
index 87b59666f5..1623544a6a 100755
--- a/tests/qemu-iotests/245
+++ b/tests/qemu-iotests/245
@@ -79,7 +79,7 @@ class TestBlockdevReopen(iotests.QMPTestCase):
         for line in log.split("\n"):
             if line.startswith("Pattern verification failed"):
                 raise Exception("%s (command #%d)" % (line, found))
-            if re.match("read .*/.* bytes at offset", line):
+            if re.match("(read|wrote) .*/.* bytes at offset", line):
                 found += 1
         self.assertEqual(found, self.total_io_cmds,
                          "Expected output of %d qemu-io commands, found %d" %
@@ -537,6 +537,113 @@ class TestBlockdevReopen(iotests.QMPTestCase):
         result = self.vm.qmp('blockdev-del', conv_keys = True, node_name = 'bv')
         self.assert_qmp(result, 'return', {})
 
+    # Replace the protocol layer ('file' parameter) of a disk image
+    def test_replace_file(self):
+        # Create two small raw images and add them to a running VM
+        qemu_img('create', '-f', 'raw', hd_path[0], '10k')
+        qemu_img('create', '-f', 'raw', hd_path[1], '10k')
+
+        hd0_opts = {'driver': 'file', 'node-name': 'hd0-file', 'filename':  hd_path[0] }
+        hd1_opts = {'driver': 'file', 'node-name': 'hd1-file', 'filename':  hd_path[1] }
+
+        result = self.vm.qmp('blockdev-add', conv_keys = False, **hd0_opts)
+        self.assert_qmp(result, 'return', {})
+        result = self.vm.qmp('blockdev-add', conv_keys = False, **hd1_opts)
+        self.assert_qmp(result, 'return', {})
+
+        # Add a raw format layer that uses hd0-file as its protocol layer
+        opts = {'driver': 'raw', 'node-name': 'hd', 'file': 'hd0-file'}
+
+        result = self.vm.qmp('blockdev-add', conv_keys = False, **opts)
+        self.assert_qmp(result, 'return', {})
+
+        # Fill the image with data
+        self.run_qemu_io("hd", "read  -P 0 0 10k")
+        self.run_qemu_io("hd", "write -P 0xa0 0 10k")
+
+        # Replace hd0-file with hd1-file and fill it with (different) data
+        self.reopen(opts, {'file': 'hd1-file'})
+        self.run_qemu_io("hd", "read  -P 0 0 10k")
+        self.run_qemu_io("hd", "write -P 0xa1 0 10k")
+
+        # Use hd0-file again and check that it contains the expected data
+        self.reopen(opts, {'file': 'hd0-file'})
+        self.run_qemu_io("hd", "read  -P 0xa0 0 10k")
+
+        # And finally do the same with hd1-file
+        self.reopen(opts, {'file': 'hd1-file'})
+        self.run_qemu_io("hd", "read  -P 0xa1 0 10k")
+
+    # Insert (and remove) a throttle filter
+    def test_insert_throttle_filter(self):
+        # Add an image to the VM
+        hd0_opts = hd_opts(0)
+        result = self.vm.qmp('blockdev-add', conv_keys = False, **hd0_opts)
+        self.assert_qmp(result, 'return', {})
+
+        # Create a throttle-group object
+        opts = { 'qom-type': 'throttle-group', 'id': 'group0',
+                 'limits': { 'iops-total': 1000 } }
+        result = self.vm.qmp('object-add', conv_keys = False, **opts)
+        self.assert_qmp(result, 'return', {})
+
+        # Add a throttle filter with the group that we just created.
+        # The filter is not used by anyone yet
+        opts = { 'driver': 'throttle', 'node-name': 'throttle0',
+                 'throttle-group': 'group0', 'file': 'hd0-file' }
+        result = self.vm.qmp('blockdev-add', conv_keys = False, **opts)
+        self.assert_qmp(result, 'return', {})
+
+        # Insert the throttle filter between hd0 and hd0-file
+        self.reopen(hd0_opts, {'file': 'throttle0'})
+
+        # Remove the throttle filter from hd0
+        self.reopen(hd0_opts, {'file': 'hd0-file'})
+
+    # Insert (and remove) a compress filter
+    def test_insert_compress_filter(self):
+        # Add an image to the VM: hd (raw) -> hd0 (qcow2) -> hd0-file (file)
+        opts = {'driver': 'raw', 'node-name': 'hd', 'file': hd_opts(0)}
+        result = self.vm.qmp('blockdev-add', conv_keys = False, **opts)
+        self.assert_qmp(result, 'return', {})
+
+        # Add a 'compress' filter
+        filter_opts = {'driver': 'compress',
+                       'node-name': 'compress0',
+                       'file': 'hd0'}
+        result = self.vm.qmp('blockdev-add', conv_keys = False, **filter_opts)
+        self.assert_qmp(result, 'return', {})
+
+        # Unmap the beginning of the image (we cannot write compressed
+        # data to an allocated cluster)
+        self.run_qemu_io("hd", "write -z -u 0 128k")
+
+        # Write data to the first cluster
+        self.run_qemu_io("hd", "write -P 0xa0 0 64k")
+
+        # Insert the filter then write to the second cluster
+        # hd -> compress0 -> hd0 -> hd0-file
+        self.reopen(opts, {'file': 'compress0'})
+        self.run_qemu_io("hd", "write -P 0xa1 64k 64k")
+
+        # Remove the filter then write to the third cluster
+        # hd -> hd0 -> hd0-file
+        self.reopen(opts, {'file': 'hd0'})
+        self.run_qemu_io("hd", "write -P 0xa2 128k 64k")
+
+        # Verify the data that we just wrote
+        self.run_qemu_io("hd", "read -P 0xa0    0 64k")
+        self.run_qemu_io("hd", "read -P 0xa1  64k 64k")
+        self.run_qemu_io("hd", "read -P 0xa2 128k 64k")
+
+        self.vm.shutdown()
+
+        # Check the first byte of the first three L2 entries and verify that
+        # the second one is compressed (0x40) while the others are not (0x80)
+        iotests.qemu_io_log('-f', 'raw', '-c', 'read -P 0x80 0x40000 1',
+                                         '-c', 'read -P 0x40 0x40008 1',
+                                         '-c', 'read -P 0x80 0x40010 1', hd_path[0])
+
     # Misc reopen tests with different block drivers
     @iotests.skip_if_unsupported(['quorum', 'throttle'])
     def test_misc_drivers(self):
diff --git a/tests/qemu-iotests/245.out b/tests/qemu-iotests/245.out
index 4b33dcaf5c..6ea1b2798f 100644
--- a/tests/qemu-iotests/245.out
+++ b/tests/qemu-iotests/245.out
@@ -10,8 +10,15 @@
 {"return": {}}
 {"data": {"id": "stream0", "type": "stream"}, "event": "BLOCK_JOB_PENDING", "timestamp": {"microseconds": "USECS", "seconds": "SECS"}}
 {"data": {"device": "stream0", "len": 3145728, "offset": 3145728, "speed": 0, "type": "stream"}, "event": "BLOCK_JOB_COMPLETED", "timestamp": {"microseconds": "USECS", "seconds": "SECS"}}
-.....................
+read 1/1 bytes at offset 262144
+1 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
+read 1/1 bytes at offset 262152
+1 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
+read 1/1 bytes at offset 262160
+1 bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
+
+........................
 ----------------------------------------------------------------------
-Ran 21 tests
+Ran 24 tests
 
 OK
-- 
2.29.2

Re: [PATCH v5 0/9] Allow changing bs->file on reopen

[Vladimir Sementsov-Ogievskiy]

15.05.2021 16:46, Vladimir Sementsov-Ogievskiy wrote:
> Hi all!
> 
> Alberto, I'm sorry for not consulting with you about sending this.
> 
> Actually, I wanted only one patch
> "block: Allow changing bs->file on reopen", as I'm going to send another
> series based on it.

Haha. Actually I need instead a way to insert filter between root device and top block node, which is impossible with reopen, so I probably need qom-set instead, as Kevin said.. Or something new. Anyway, I've done some work, hope it's helpful)

> I started to work on updating this one patch it and
> it turned into a refactoring series below. So I decided to add you patch
> with test cases, for this series to be a complete* v5. I hope you'll
> have time to work on the rest of your series on top of this my
> suggestion :)
> 
> [*], supporting multiple reopen and final publishing x-blockdev-reopen
> not included here. Also I didn't address Kevin's comment that we should
> have some error-path testing of new feature. Probably I didn't address
> some comments on "Allow changing" patch, but it changed so much that
> review should start from scratch anyway.
> 
> v5:
> 1-7: new
> 8: changed a lot
> 9: unchanged
> 
> Alberto Garcia (2):
>    block: Allow changing bs->file on reopen
>    iotests: Test replacing files with x-blockdev-reopen
> 
> Vladimir Sementsov-Ogievskiy (7):
>    block: introduce bdrv_remove_file_or_backing_child()
>    block: introduce bdrv_set_file_or_backing_noperm()
>    block: bdrv_reopen_parse_backing(): don't check aio context
>    block: bdrv_reopen_parse_backing(): don't check frozen child
>    block: bdrv_reopen_parse_backing(): simplify handling implicit filters
>    block: move supports_backing check to
>      bdrv_set_file_or_backing_noperm()
>    block: BDRVReopenState: drop replace_backing_bs field
> 
>   include/block/block.h            |   2 +-
>   block.c                          | 287 ++++++++++++++++---------------
>   tests/unit/test-bdrv-drain.c     |   1 +
>   tests/unit/test-bdrv-graph-mod.c |   1 +
>   tests/qemu-iotests/245           | 140 +++++++++++++--
>   tests/qemu-iotests/245.out       |  11 +-
>   6 files changed, 287 insertions(+), 155 deletions(-)
> 


-- 
Best regards,
Vladimir