mirror: Freeze backing chain for sync=top
cover mirror: Freeze backing chain for sync=top
Hi,
On review of the “Deal with filters” series, Kevin noted that mirror doesn’t freeze its backing chain, even though we keep references to the base (for sync=top). That seems like a recipe for disaster, and as the test added in patch 3 shows, indeed it is.
Patch 1 could also be squashed into “mirror: Deal with filters”.
As for patch 2; maybe it would be sufficient to freeze the backing chain only down to base_overlay and until mirror_dirty_init() is done. Then we bdrv_ref(s->base) so can always be attached to the target when mirror completes. But I think the conservative approach I implemented here is fine until someone complains.
Max Reitz (3):
mirror: Set s->base_overlay only if s->base is set
mirror: Freeze backing chain for sync=top iotests/041: x-blockdev-reopen during mirror
block/mirror.c | 28 +++++++++--- tests/qemu-iotests/041 | 92 ++++++++++++++++++++++++++++++++++++++ tests/qemu-iotests/041.out | 4 +- 3 files changed, 116 insertions(+), 8 deletions(-)
Signature · 1 line
2.26.2
PATCH 1/3 mirror: Set s->base_overlay only if s->base is set
This way, sync=full will not need a reference to any node other than the source and the target.
--- block/mirror.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-)
block/mirror.c +5 -3
@@ -42,6 +42,7 @@ typedef struct MirrorBlockJob {BlockBackend *target;BlockDriverState *mirror_top_bs;BlockDriverState *base; +/* Overlay of @base if @base is non-NULL; NULL otherwise */ BlockDriverState *base_overlay;/* The name of the graph node to replace */
@@ -839,8 +840,9 @@ static int coroutine_fn mirror_dirty_init(MirrorBlockJob *s)return 0 ; } - ret =bdrv_is_allocated_above (bs, s->base_overlay, true, offset, bytes, - &count); + ret =bdrv_is_allocated_above (bs, s->base_overlay, + s->base_overlay !=NULL , + offset, bytes, &count);if (ret <0 ) {return ret; }
@@ -1710,7 +1712,7 @@ static BlockJob *mirror_start_job( s->zero_target = zero_target; s->copy_mode = copy_mode; s->base = base; - s->base_overlay =bdrv_find_overlay (bs, base); + s->base_overlay = base ?bdrv_find_overlay (bs, base) :NULL ; s->granularity = granularity; s->buf_size =ROUND_UP (buf_size, granularity); s->unmap = unmap;
Signature · 1 line
2.26.2
PATCH 2/3 mirror: Freeze backing chain for sync=top
In reply to Max Reitz--- block/mirror.c | 20 +++++++++++++++++--- 1 file changed, 17 insertions(+), 3 deletions(-)
block/mirror.c +17 -3
@@ -649,8 +649,8 @@ static int mirror_exit_common(Job *job) src = mirror_top_bs->backing->bs; target_bs =blk_bs (s->target); -if (bdrv_chain_contains (src, target_bs)) { -bdrv_unfreeze_backing_chain (mirror_top_bs, target_bs); +if (s->base) { +bdrv_unfreeze_backing_chain (mirror_top_bs, s->base); }bdrv_release_dirty_bitmap (s->dirty_bitmap);
@@ -1780,8 +1780,22 @@ static BlockJob *mirror_start_job( goto fail; } } + } + +if (s->base) { +/* +* For active commit or mirror with sync=top, we need to +* freeze the backing chain down to the base, because we keep +* pointers to it and its overlay. For other cases (mirror +* with sync=full or sync=none), we do not, so there is no +* need to freeze any part of the chain. +* (s->base is non-NULL only in these cases.) +*/ +if (target_is_backing) { +assert (s->base == target); + } -if (bdrv_freeze_backing_chain (mirror_top_bs, target, errp) <0 ) { +if (bdrv_freeze_backing_chain (mirror_top_bs, s->base, errp) <0 ) { goto fail; } }
Signature · 1 line
2.26.2
Quoted 29 lines · Max Reitz
Reported-by Kevin Wolf <[email protected]>Signed-off-by Max Reitz <[email protected]>--- block/mirror.c | 20 +++++++++++++++++--- 1 file changed, 17 insertions(+), 3 deletions(-)block/mirror.c +6
-2@@ -649,8 +649,8 @@ static int mirror_exit_common(Job *job) src = mirror_top_bs->backing->bs; target_bs =blk_bs (s->target); -if (bdrv_chain_contains (src, target_bs)) { -bdrv_unfreeze_backing_chain (mirror_top_bs, target_bs); +if (s->base) { +bdrv_unfreeze_backing_chain (mirror_top_bs, s->base); }bdrv_release_dirty_bitmap (s->dirty_bitmap);@@ -1780,8 +1780,22 @@ static BlockJob *mirror_start_job( goto fail; } } + } + +if (s->base) { +/*
block/mirror.c static BlockJob *mirror_start_job( · line 1796
Quoted 10 lines
block/mirror.c +10
-0+* For active commit or mirror with sync=top, we need to +* freeze the backing chain down to the base, because we keep +* pointers to it and its overlay. For other cases (mirror +* with sync=full or sync=none), we do not, so there is no +* need to freeze any part of the chain. +* (s->base is non-NULL only in these cases.) +*/ +if (target_is_backing) { +assert (s->base == target); + }
On second thought, this assertion doesn’t hold true when mirroring to an image in the backing chain (i.e., specifically not using commit). Now, this mostly doesn’t work thanks to op blockers, but after “Deal with filters”, it’s possible. I don’t know whether it produces any sensible results, though.
Should we leave it possible, or should we make mirroring to a node in the backing chain (i.e. target_is_backing && s->base != target) an error?
Max
Quoted 7 lines
block/mirror.c +1
-1-if (bdrv_freeze_backing_chain (mirror_top_bs, target, errp) <0 ) { +if (bdrv_freeze_backing_chain (mirror_top_bs, s->base, errp) <0 ) { goto fail; } }
PATCH 3/3 iotests/041: x-blockdev-reopen during mirror
In reply to Max ReitzTest what happens when you remove the backing file during a mirror with sync=top.
--- tests/qemu-iotests/041 | 92 ++++++++++++++++++++++++++++++++++++++ tests/qemu-iotests/041.out | 4 +- 2 files changed, 94 insertions(+), 2 deletions(-)
tests/qemu-iotests/041 +92 -0
@@ -1432,6 +1432,98 @@ class TestFilters(iotests.QMPTestCase): self.complete_and_wait('mirror') +class TestReconfDuringMirror(iotests.QMPTestCase): + def setUp(self): + qemu_img('create', '-f', iotests.imgfmt, backing_img, '1M') + qemu_img('create', '-f', iotests.imgfmt, '-b', backing_img, + '-F', iotests.imgfmt, test_img) + + qemu_io('-c', 'write -P 1 0 1M', test_img) + + self.vm = iotests.VM() + self.vm.launch() + + blockdevs = ( + { 'node-name': 'backing-file', + 'driver': 'file', + 'filename': backing_img }, + + { 'node-name': 'backing', + 'driver': iotests.imgfmt, + 'file': 'backing-file', + 'read-only': True }, + + { 'node-name': 'source-file', + 'driver': 'file', + 'filename': test_img }, + + { 'node-name': 'source', + 'driver': iotests.imgfmt, + 'file': 'source-file', + 'backing': 'backing' } + ) + + for blockdev in blockdevs: + result = self.vm.qmp('blockdev-add', **blockdev) + self.assert_qmp(result, 'return', {}) + + def tearDown(self): + self.vm.shutdown() + + os.remove(test_img) + os.remove(backing_img) + + try: + os.remove(target_img) + except OSError: + pass + + def test_reconf_during_top_mirror(self): + result = self.vm.qmp('drive-mirror', + job_id='mirror', + device='source', + sync='top', + target=target_img, + speed=65536) + self.assert_qmp(result, 'return', {}) + + result = self.vm.qmp('x-blockdev-reopen', + **{ 'node-name': 'source', + 'driver': iotests.imgfmt, + 'file': 'source-file', + 'backing': None }) + if 'error' in result: + self.assert_qmp(result, 'error/desc', + "Cannot change 'backing' link " + "from 'source' to 'backing'") + + result = self.vm.qmp('block-job-set-speed', + device='mirror', speed=0) + self.assert_qmp(result, 'return', {}) + + self.complete_and_wait('mirror') + + # Pass + return + + # This is not expected. Let's see what happens when we drop + # the backing file altogether. + for node in ('backing', 'backing-file'): + result = self.vm.qmp('blockdev-del', node_name=node) + self.assert_qmp(result, 'return', {}) + + result = self.vm.qmp('block-job-set-speed', + device='mirror', speed=0) + self.assert_qmp(result, 'return', {}) + + # Completing will now probably result in a segfault, because + # it tries to assign 'backing' as the target's backing node + self.complete_and_wait('mirror') + + # Either way, this is a failure. + self.fail('x-blockdev-reopen should not have succeeded') + + if __name__ == '__main__': iotests.main(supported_fmts=['qcow2', 'qed'], supported_protocols=['file'],
tests/qemu-iotests/041.out +2 -2
@@ -1,5 +1,5 @@ -........................................................................................................... +............................................................................................................ ---------------------------------------------------------------------- -Ran 107 tests +Ran 108 tests OK
Signature · 1 line
2.26.2