[Qemu-devel] [PATCH for-3.1 v2 0/2] block: Fix crash on migration with explicit child nodes

[Qemu-devel] [PATCH for-3.1 v2 0/2] block: Fix crash on migration with explicit child nodes

[Kevin Wolf]

v2:
- Cover children that were created after their parents [Max]

Kevin Wolf (2):
  block: Don't inactivate children before parents
  iotests: Test migration with -blockdev

 block.c                    |  32 ++++++++++-
 tests/qemu-iotests/234     | 115 +++++++++++++++++++++++++++++++++++++
 tests/qemu-iotests/234.out |  27 +++++++++
 tests/qemu-iotests/group   |   1 +
 4 files changed, 174 insertions(+), 1 deletion(-)
 create mode 100755 tests/qemu-iotests/234
 create mode 100644 tests/qemu-iotests/234.out

-- 
2.19.1

[Qemu-devel] [PATCH for-3.1 v2 1/2] block: Don't inactivate children before parents

[Kevin Wolf]

bdrv_child_cb_inactivate() asserts that parents are already inactive
when children get inactivated. This precondition is necessary because
parents could still issue requests in their inactivation code.

When block nodes are created individually with -blockdev, all of them
are monitor owned and will be returned by bdrv_next() in an undefined
order (in practice, in the order of their creation, which is usually
children before parents), which obviously fails the assertion:

qemu: block.c:899: bdrv_child_cb_inactivate: Assertion `bs->open_flags & BDRV_O_INACTIVE' failed.

This patch fixes the ordering by skipping nodes with still active
parents in bdrv_inactivate_recurse() because we know that they will be
covered by recursion when the last active parent becomes inactive.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
 block.c | 32 +++++++++++++++++++++++++++++++-
 1 file changed, 31 insertions(+), 1 deletion(-)

diff --git a/block.c b/block.c
index 5ba3435f8f..e9181c3be7 100644
--- a/block.c
+++ b/block.c
@@ -4612,6 +4612,22 @@ void bdrv_invalidate_cache_all(Error **errp)
     }
 }
 
+static bool bdrv_has_bds_parent(BlockDriverState *bs, bool only_active)
+{
+    BdrvChild *parent;
+
+    QLIST_FOREACH(parent, &bs->parents, next_parent) {
+        if (parent->role->parent_is_bds) {
+            BlockDriverState *parent_bs = parent->opaque;
+            if (!only_active || !(parent_bs->open_flags & BDRV_O_INACTIVE)) {
+                return true;
+            }
+        }
+    }
+
+    return false;
+}
+
 static int bdrv_inactivate_recurse(BlockDriverState *bs,
                                    bool setting_flag)
 {
@@ -4622,6 +4638,14 @@ static int bdrv_inactivate_recurse(BlockDriverState *bs,
         return -ENOMEDIUM;
     }
 
+    /* Make sure that we don't inactivate a child before its parent.
+     * It will be covered by recursion from the yet active parent. */
+    if (bdrv_has_bds_parent(bs, true)) {
+        return 0;
+    }
+
+    assert(!(bs->open_flags & BDRV_O_INACTIVE));
+
     if (!setting_flag && bs->drv->bdrv_inactivate) {
         ret = bs->drv->bdrv_inactivate(bs);
         if (ret < 0) {
@@ -4629,7 +4653,7 @@ static int bdrv_inactivate_recurse(BlockDriverState *bs,
         }
     }
 
-    if (setting_flag && !(bs->open_flags & BDRV_O_INACTIVE)) {
+    if (setting_flag) {
         uint64_t perm, shared_perm;
 
         QLIST_FOREACH(parent, &bs->parents, next_parent) {
@@ -4682,6 +4706,12 @@ int bdrv_inactivate_all(void)
      * is allowed. */
     for (pass = 0; pass < 2; pass++) {
         for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
+            /* Nodes with BDS parents are covered by recursion from the last
+             * parent that gets inactivated. Don't inactivate them a second
+             * time if that has already happened. */
+            if (bdrv_has_bds_parent(bs, false)) {
+                continue;
+            }
             ret = bdrv_inactivate_recurse(bs, pass);
             if (ret < 0) {
                 bdrv_next_cleanup(&it);
-- 
2.19.1

[Qemu-devel] [PATCH for-3.1 v2 2/2] iotests: Test migration with -blockdev

[Kevin Wolf]

Check that block node activation and inactivation works with a block
graph that is built with individually created nodes.

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
---
 tests/qemu-iotests/234     | 115 +++++++++++++++++++++++++++++++++++++
 tests/qemu-iotests/234.out |  27 +++++++++
 tests/qemu-iotests/group   |   1 +
 3 files changed, 143 insertions(+)
 create mode 100755 tests/qemu-iotests/234
 create mode 100644 tests/qemu-iotests/234.out

diff --git a/tests/qemu-iotests/234 b/tests/qemu-iotests/234
new file mode 100755
index 0000000000..58886b8dea
--- /dev/null
+++ b/tests/qemu-iotests/234
@@ -0,0 +1,115 @@
+#!/usr/bin/env python
+#
+# Copyright (C) 2018 Red Hat, Inc.
+#
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program.  If not, see <http://www.gnu.org/licenses/>.
+#
+# Creator/Owner: Kevin Wolf <kwolf@redhat.com>
+#
+# Check that block node activation and inactivation works with a block graph
+# that is built with individually created nodes
+
+import iotests
+import os
+
+iotests.verify_image_format(supported_fmts=['raw', 'qcow2'])
+iotests.verify_platform(['linux'])
+
+with iotests.FilePath('img') as img_path, \
+     iotests.FilePath('backing') as backing_path, \
+     iotests.FilePath('mig_fifo_a') as fifo_a, \
+     iotests.FilePath('mig_fifo_b') as fifo_b, \
+     iotests.VM(path_suffix='a') as vm_a, \
+     iotests.VM(path_suffix='b') as vm_b:
+
+    iotests.qemu_img_pipe('create', '-f', iotests.imgfmt, backing_path, '64M')
+    iotests.qemu_img_pipe('create', '-f', iotests.imgfmt, img_path, '64M')
+
+    os.mkfifo(fifo_a)
+    os.mkfifo(fifo_b)
+
+    iotests.log('Launching source VM...')
+    (vm_a.add_blockdev('file,filename=%s,node-name=drive0-file' % (img_path))
+         .add_blockdev('%s,file=drive0-file,node-name=drive0' % (iotests.imgfmt))
+         .add_blockdev('file,filename=%s,node-name=drive0-backing-file' % (backing_path))
+         .add_blockdev('%s,file=drive0-backing-file,node-name=drive0-backing' % (iotests.imgfmt))
+         .launch())
+
+    iotests.log('Launching destination VM...')
+    (vm_b.add_blockdev('file,filename=%s,node-name=drive0-file' % (img_path))
+         .add_blockdev('%s,file=drive0-file,node-name=drive0' % (iotests.imgfmt))
+         .add_blockdev('file,filename=%s,node-name=drive0-backing-file' % (backing_path))
+         .add_blockdev('%s,file=drive0-backing-file,node-name=drive0-backing' % (iotests.imgfmt))
+         .add_incoming("exec: cat '%s'" % (fifo_a))
+         .launch())
+
+    # Add a child node that was created after the parent node. The reverse case
+    # is covered by the -blockdev options above.
+    vm_a.qmp('blockdev-snapshot', node='drive0-backing', overlay='drive0')
+
+    iotests.log('Enabling migration QMP events on A...')
+    iotests.log(vm_a.qmp('migrate-set-capabilities', capabilities=[
+        {
+            'capability': 'events',
+            'state': True
+        }
+    ]))
+
+    iotests.log('Starting migration to B...')
+    iotests.log(vm_a.qmp('migrate', uri='exec:cat >%s' % (fifo_a)))
+    with iotests.Timeout(3, 'Migration does not complete'):
+        while True:
+            event = vm_a.event_wait('MIGRATION')
+            iotests.log(event, filters=[iotests.filter_qmp_event])
+            if event['data']['status'] == 'completed':
+                break
+
+    iotests.log(vm_a.qmp('query-migrate')['return']['status'])
+    iotests.log(vm_b.qmp('query-migrate')['return']['status'])
+
+    iotests.log(vm_a.qmp('query-status'))
+    iotests.log(vm_b.qmp('query-status'))
+
+    iotests.log('Add a second parent to drive0-file...')
+    iotests.log(vm_b.qmp('blockdev-add', driver='raw', file='drive0-file',
+                         node_name='drive0-raw'))
+
+    iotests.log('Restart A with -incoming and second parent...')
+    vm_a.shutdown()
+    (vm_a.add_blockdev('raw,file=drive0-file,node-name=drive0-raw')
+         .add_incoming("exec: cat '%s'" % (fifo_b))
+         .launch())
+
+    iotests.log('Enabling migration QMP events on B...')
+    iotests.log(vm_b.qmp('migrate-set-capabilities', capabilities=[
+        {
+            'capability': 'events',
+            'state': True
+        }
+    ]))
+
+    iotests.log('Starting migration back to A...')
+    iotests.log(vm_b.qmp('migrate', uri='exec:cat >%s' % (fifo_b)))
+    with iotests.Timeout(3, 'Migration does not complete'):
+        while True:
+            event = vm_b.event_wait('MIGRATION')
+            iotests.log(event, filters=[iotests.filter_qmp_event])
+            if event['data']['status'] == 'completed':
+                break
+
+    iotests.log(vm_a.qmp('query-migrate')['return']['status'])
+    iotests.log(vm_b.qmp('query-migrate')['return']['status'])
+
+    iotests.log(vm_a.qmp('query-status'))
+    iotests.log(vm_b.qmp('query-status'))
diff --git a/tests/qemu-iotests/234.out b/tests/qemu-iotests/234.out
new file mode 100644
index 0000000000..115537cd5c
--- /dev/null
+++ b/tests/qemu-iotests/234.out
@@ -0,0 +1,27 @@
+Launching source VM...
+Launching destination VM...
+Enabling migration QMP events on A...
+{"return": {}}
+Starting migration to B...
+{"return": {}}
+{"data": {"status": "setup"}, "event": "MIGRATION", "timestamp": {"microseconds": "USECS", "seconds": "SECS"}}
+{"data": {"status": "active"}, "event": "MIGRATION", "timestamp": {"microseconds": "USECS", "seconds": "SECS"}}
+{"data": {"status": "completed"}, "event": "MIGRATION", "timestamp": {"microseconds": "USECS", "seconds": "SECS"}}
+completed
+completed
+{"return": {"running": false, "singlestep": false, "status": "postmigrate"}}
+{"return": {"running": true, "singlestep": false, "status": "running"}}
+Add a second parent to drive0-file...
+{"return": {}}
+Restart A with -incoming and second parent...
+Enabling migration QMP events on B...
+{"return": {}}
+Starting migration back to A...
+{"return": {}}
+{"data": {"status": "setup"}, "event": "MIGRATION", "timestamp": {"microseconds": "USECS", "seconds": "SECS"}}
+{"data": {"status": "active"}, "event": "MIGRATION", "timestamp": {"microseconds": "USECS", "seconds": "SECS"}}
+{"data": {"status": "completed"}, "event": "MIGRATION", "timestamp": {"microseconds": "USECS", "seconds": "SECS"}}
+completed
+completed
+{"return": {"running": true, "singlestep": false, "status": "running"}}
+{"return": {"running": false, "singlestep": false, "status": "postmigrate"}}
diff --git a/tests/qemu-iotests/group b/tests/qemu-iotests/group
index ddf1a5b549..8c56a0ad11 100644
--- a/tests/qemu-iotests/group
+++ b/tests/qemu-iotests/group
@@ -231,3 +231,4 @@
 231 auto quick
 232 auto quick
 233 auto quick
+234 auto quick migration
-- 
2.19.1

Re: [Qemu-devel] [PATCH for-3.1 v2 1/2] block: Don't inactivate children before parents

[Max Reitz]

[-- Attachment #1: Type: text/plain, Size: 2031 bytes --]

On 26.11.18 15:16, Kevin Wolf wrote:
> bdrv_child_cb_inactivate() asserts that parents are already inactive
> when children get inactivated. This precondition is necessary because
> parents could still issue requests in their inactivation code.
> 
> When block nodes are created individually with -blockdev, all of them
> are monitor owned and will be returned by bdrv_next() in an undefined
> order (in practice, in the order of their creation, which is usually
> children before parents), which obviously fails the assertion:
> 
> qemu: block.c:899: bdrv_child_cb_inactivate: Assertion `bs->open_flags & BDRV_O_INACTIVE' failed.
> 
> This patch fixes the ordering by skipping nodes with still active
> parents in bdrv_inactivate_recurse() because we know that they will be
> covered by recursion when the last active parent becomes inactive.
> 
> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
> ---
>  block.c | 32 +++++++++++++++++++++++++++++++-
>  1 file changed, 31 insertions(+), 1 deletion(-)
> 
> diff --git a/block.c b/block.c
> index 5ba3435f8f..e9181c3be7 100644
> --- a/block.c
> +++ b/block.c

[...]

> @@ -4622,6 +4638,14 @@ static int bdrv_inactivate_recurse(BlockDriverState *bs,
>          return -ENOMEDIUM;
>      }
>  
> +    /* Make sure that we don't inactivate a child before its parent.
> +     * It will be covered by recursion from the yet active parent. */
> +    if (bdrv_has_bds_parent(bs, true)) {

Overall the patch looks good to me, apart from the fact that I still
think this should only be checked if setting_flag is true.

(Because BDRV_O_INACTIVE is only set during the second pass, so this
check will lead to the first pass not doing anything (and thus not
calling bs->drv->bdrv_inactivate()) on any non-root bs.)

Max

> +        return 0;
> +    }
> +
> +    assert(!(bs->open_flags & BDRV_O_INACTIVE));
> +
>      if (!setting_flag && bs->drv->bdrv_inactivate) {
>          ret = bs->drv->bdrv_inactivate(bs);
>          if (ret < 0) {


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: [Qemu-devel] [PATCH for-3.1 v2 2/2] iotests: Test migration with -blockdev

[Max Reitz]

[-- Attachment #1: Type: text/plain, Size: 8025 bytes --]

On 26.11.18 15:16, Kevin Wolf wrote:
> Check that block node activation and inactivation works with a block
> graph that is built with individually created nodes.
> 
> Signed-off-by: Kevin Wolf <kwolf@redhat.com>
> ---
>  tests/qemu-iotests/234     | 115 +++++++++++++++++++++++++++++++++++++
>  tests/qemu-iotests/234.out |  27 +++++++++
>  tests/qemu-iotests/group   |   1 +
>  3 files changed, 143 insertions(+)
>  create mode 100755 tests/qemu-iotests/234
>  create mode 100644 tests/qemu-iotests/234.out
> 
> diff --git a/tests/qemu-iotests/234 b/tests/qemu-iotests/234
> new file mode 100755
> index 0000000000..58886b8dea
> --- /dev/null
> +++ b/tests/qemu-iotests/234
> @@ -0,0 +1,115 @@
> +#!/usr/bin/env python
> +#
> +# Copyright (C) 2018 Red Hat, Inc.
> +#
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License as published by
> +# the Free Software Foundation; either version 2 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
> +#
> +# Creator/Owner: Kevin Wolf <kwolf@redhat.com>
> +#
> +# Check that block node activation and inactivation works with a block graph
> +# that is built with individually created nodes
> +
> +import iotests
> +import os
> +
> +iotests.verify_image_format(supported_fmts=['raw', 'qcow2'])
> +iotests.verify_platform(['linux'])
> +
> +with iotests.FilePath('img') as img_path, \
> +     iotests.FilePath('backing') as backing_path, \
> +     iotests.FilePath('mig_fifo_a') as fifo_a, \
> +     iotests.FilePath('mig_fifo_b') as fifo_b, \
> +     iotests.VM(path_suffix='a') as vm_a, \
> +     iotests.VM(path_suffix='b') as vm_b:
> +
> +    iotests.qemu_img_pipe('create', '-f', iotests.imgfmt, backing_path, '64M')
> +    iotests.qemu_img_pipe('create', '-f', iotests.imgfmt, img_path, '64M')
> +
> +    os.mkfifo(fifo_a)
> +    os.mkfifo(fifo_b)
> +
> +    iotests.log('Launching source VM...')
> +    (vm_a.add_blockdev('file,filename=%s,node-name=drive0-file' % (img_path))
> +         .add_blockdev('%s,file=drive0-file,node-name=drive0' % (iotests.imgfmt))
> +         .add_blockdev('file,filename=%s,node-name=drive0-backing-file' % (backing_path))
> +         .add_blockdev('%s,file=drive0-backing-file,node-name=drive0-backing' % (iotests.imgfmt))
> +         .launch())
> +
> +    iotests.log('Launching destination VM...')
> +    (vm_b.add_blockdev('file,filename=%s,node-name=drive0-file' % (img_path))
> +         .add_blockdev('%s,file=drive0-file,node-name=drive0' % (iotests.imgfmt))
> +         .add_blockdev('file,filename=%s,node-name=drive0-backing-file' % (backing_path))
> +         .add_blockdev('%s,file=drive0-backing-file,node-name=drive0-backing' % (iotests.imgfmt))
> +         .add_incoming("exec: cat '%s'" % (fifo_a))
> +         .launch())
> +
> +    # Add a child node that was created after the parent node. The reverse case
> +    # is covered by the -blockdev options above.
> +    vm_a.qmp('blockdev-snapshot', node='drive0-backing', overlay='drive0')

Sorry if I made it sound like my diff was ready for inclusion as-is. O:-)

I think this should be vm_a.qmp_log() at least (and the reference output
needs to be adjusted accordingly).

Also, we probably have to execute the same command on vm_b...

> +    iotests.log('Enabling migration QMP events on A...')
> +    iotests.log(vm_a.qmp('migrate-set-capabilities', capabilities=[
> +        {
> +            'capability': 'events',
> +            'state': True
> +        }
> +    ]))
> +
> +    iotests.log('Starting migration to B...')
> +    iotests.log(vm_a.qmp('migrate', uri='exec:cat >%s' % (fifo_a)))
> +    with iotests.Timeout(3, 'Migration does not complete'):
> +        while True:
> +            event = vm_a.event_wait('MIGRATION')
> +            iotests.log(event, filters=[iotests.filter_qmp_event])
> +            if event['data']['status'] == 'completed':
> +                break
> +
> +    iotests.log(vm_a.qmp('query-migrate')['return']['status'])
> +    iotests.log(vm_b.qmp('query-migrate')['return']['status'])
> +
> +    iotests.log(vm_a.qmp('query-status'))
> +    iotests.log(vm_b.qmp('query-status'))
> +
> +    iotests.log('Add a second parent to drive0-file...')
> +    iotests.log(vm_b.qmp('blockdev-add', driver='raw', file='drive0-file',
> +                         node_name='drive0-raw'))
> +
> +    iotests.log('Restart A with -incoming and second parent...')
> +    vm_a.shutdown()
> +    (vm_a.add_blockdev('raw,file=drive0-file,node-name=drive0-raw')
> +         .add_incoming("exec: cat '%s'" % (fifo_b))
> +         .launch())

...and here on vm_a again?

Max

> +    iotests.log('Enabling migration QMP events on B...')
> +    iotests.log(vm_b.qmp('migrate-set-capabilities', capabilities=[
> +        {
> +            'capability': 'events',
> +            'state': True
> +        }
> +    ]))
> +
> +    iotests.log('Starting migration back to A...')
> +    iotests.log(vm_b.qmp('migrate', uri='exec:cat >%s' % (fifo_b)))
> +    with iotests.Timeout(3, 'Migration does not complete'):
> +        while True:
> +            event = vm_b.event_wait('MIGRATION')
> +            iotests.log(event, filters=[iotests.filter_qmp_event])
> +            if event['data']['status'] == 'completed':
> +                break
> +
> +    iotests.log(vm_a.qmp('query-migrate')['return']['status'])
> +    iotests.log(vm_b.qmp('query-migrate')['return']['status'])
> +
> +    iotests.log(vm_a.qmp('query-status'))
> +    iotests.log(vm_b.qmp('query-status'))
> diff --git a/tests/qemu-iotests/234.out b/tests/qemu-iotests/234.out
> new file mode 100644
> index 0000000000..115537cd5c
> --- /dev/null
> +++ b/tests/qemu-iotests/234.out
> @@ -0,0 +1,27 @@
> +Launching source VM...
> +Launching destination VM...
> +Enabling migration QMP events on A...
> +{"return": {}}
> +Starting migration to B...
> +{"return": {}}
> +{"data": {"status": "setup"}, "event": "MIGRATION", "timestamp": {"microseconds": "USECS", "seconds": "SECS"}}
> +{"data": {"status": "active"}, "event": "MIGRATION", "timestamp": {"microseconds": "USECS", "seconds": "SECS"}}
> +{"data": {"status": "completed"}, "event": "MIGRATION", "timestamp": {"microseconds": "USECS", "seconds": "SECS"}}
> +completed
> +completed
> +{"return": {"running": false, "singlestep": false, "status": "postmigrate"}}
> +{"return": {"running": true, "singlestep": false, "status": "running"}}
> +Add a second parent to drive0-file...
> +{"return": {}}
> +Restart A with -incoming and second parent...
> +Enabling migration QMP events on B...
> +{"return": {}}
> +Starting migration back to A...
> +{"return": {}}
> +{"data": {"status": "setup"}, "event": "MIGRATION", "timestamp": {"microseconds": "USECS", "seconds": "SECS"}}
> +{"data": {"status": "active"}, "event": "MIGRATION", "timestamp": {"microseconds": "USECS", "seconds": "SECS"}}
> +{"data": {"status": "completed"}, "event": "MIGRATION", "timestamp": {"microseconds": "USECS", "seconds": "SECS"}}
> +completed
> +completed
> +{"return": {"running": true, "singlestep": false, "status": "running"}}
> +{"return": {"running": false, "singlestep": false, "status": "postmigrate"}}
> diff --git a/tests/qemu-iotests/group b/tests/qemu-iotests/group
> index ddf1a5b549..8c56a0ad11 100644
> --- a/tests/qemu-iotests/group
> +++ b/tests/qemu-iotests/group
> @@ -231,3 +231,4 @@
>  231 auto quick
>  232 auto quick
>  233 auto quick
> +234 auto quick migration
> 



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 bytes --]