block: Relax restrictions for blockdev-snapshot
v2 cover block: Relax restrictions for blockdev-snapshot
This series allows libvirt to fix a regression that its switch from drive-mirror to blockdev-mirror caused: It currently requires that the backing chain of the target image is already available when the mirror operation is started.
In reality, the backing chain may only be copied while the operation is in progress, so the backing file of the target image needs to stay disabled until the operation completes and should be attached only at that point. Without this series, we don't have a supported API to attach the backing file at that later point.
v2:
- Added a fix and test case for iothreads [Peter]
- Added a blockdev-snapshot feature flag to indicate that it's usable for attaching a backing chain to an overlay that is already in write-only use (e.g. as a mirror target) [Peter]
Kevin Wolf (6):
block: Make bdrv_get_cumulative_perm() public
block: Relax restrictions for blockdev-snapshot
iotests: Fix run_job() with use_log=False
iotests: Test mirror with temporarily disabled target backing file
block: Fix cross-AioContext blockdev-snapshot
iotests: Add iothread cases to 155
Peter Krempa (1):
qapi: Add '@allow-write-only-overlay' feature for 'blockdev-snapshot'
qapi/block-core.json | 9 +++- include/block/block_int.h | 3 ++ block.c | 7 ++- blockdev.c | 30 ++++-------- tests/qemu-iotests/iotests.py | 5 +- tests/qemu-iotests/085.out | 4 +- tests/qemu-iotests/155 | 88 +++++++++++++++++++++++++++++------ tests/qemu-iotests/155.out | 4 +- 8 files changed, 104 insertions(+), 46 deletions(-)
Signature · 1 line
2.20.1
PATCH v2 1/7 block: Make bdrv_get_cumulative_perm() public
--- include/block/block_int.h | 3 +++ block.c | 6 ++---- 2 files changed, 5 insertions(+), 4 deletions(-)
include/block/block_int.h +3 -0
@@ -1226,6 +1226,9 @@ BdrvChild *bdrv_root_attach_child(BlockDriverState *child_bs, void *opaque, Error **errp);void bdrv_root_unref_child (BdrvChild *child); +void bdrv_get_cumulative_perm (BlockDriverState *bs,uint64_t *perm, +uint64_t *shared_perm); +/** * Sets a BdrvChild's permissions. Avoid if the parent is a BDS; use * bdrv_child_refresh_perms() instead and make the parent's
block.c +2 -4
@@ -1872,8 +1872,6 @@ static int bdrv_child_check_perm(BdrvChild *c, BlockReopenQueue *q, bool *tighten_restrictions, Error **errp);static void bdrv_child_abort_perm_update (BdrvChild *c);static void bdrv_child_set_perm (BdrvChild *c,uint64_t perm,uint64_t shared); -static void bdrv_get_cumulative_perm (BlockDriverState *bs,uint64_t *perm, -uint64_t *shared_perm); typedef struct BlockReopenQueueEntry { bool prepared;
@@ -2097,8 +2095,8 @@ static void bdrv_set_perm(BlockDriverState *bs, uint64_t cumulative_perms, } } -static void bdrv_get_cumulative_perm (BlockDriverState *bs,uint64_t *perm, -uint64_t *shared_perm) +void bdrv_get_cumulative_perm (BlockDriverState *bs,uint64_t *perm, +uint64_t *shared_perm) {BdrvChild *c;uint64_t cumulative_perms =0 ;
Signature · 1 line
2.20.1
Re: [PATCH v2 1/7] block: Make bdrv_get_cumulative_perm() public
Quoted 5 lines · Kevin Wolf
Signed-off-by Kevin Wolf <[email protected]>--- include/block/block_int.h | 3 +++ block.c | 6 ++---- 2 files changed, 5 insertions(+), 4 deletions(-)
PATCH v2 2/7 block: Relax restrictions for blockdev-snapshot
In reply to Kevin Wolfblockdev-snapshot returned an error if the overlay was already in use, which it defined as having any BlockBackend parent. This is in fact both too strict (some parents can tolerate the change of visible data caused by attaching a backing file) and too loose (some non-BlockBackend parents may not be happy with it).
One important use case that is prevented by the too strict check is live storage migration with blockdev-mirror. Here, the target node is usually opened without a backing file so that the active layer is mirrored while its backing chain can be copied in the background.
The backing chain should be attached to the mirror target node when finalising the job, just before switching the users of the source node to the new copy (at which point the mirror job still has a reference to the node). drive-mirror did this automatically, but with blockdev-mirror this is the job of the QMP client, so it needs a way to do this.
blockdev-snapshot is the obvious way, so this patch makes it work in this scenario. The new condition is that no parent uses CONSISTENT_READ permissions. This will ensure that the operation will still be blocked when the node is attached to the guest device, so blockdev-snapshot remains safe.
(For the sake of completeness, x-blockdev-reopen can be used to achieve the same, however it is a big hammer, performs the graph change completely unchecked and is still experimental. So even with the option of using x-blockdev-reopen, there are reasons why blockdev-snapshot should be able to perform this operation.)
--- blockdev.c | 14 ++++++++------ tests/qemu-iotests/085.out | 4 ++-- 2 files changed, 10 insertions(+), 8 deletions(-)
blockdev.c +8 -6
@@ -1501,6 +1501,7 @@ static void external_snapshot_prepare(BlkActionState *common,TransactionAction *action = common->action;AioContext *aio_context;AioContext *old_context; +uint64_t perm, shared;int ret;/* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar
@@ -1616,16 +1617,17 @@ static void external_snapshot_prepare(BlkActionState *common, goto out; } -if (bdrv_has_blk (state->new_bs)) { +/* +* Allow attaching a backing file to an overlay that's already in use only +* if the parents don't assume that they are already seeing a valid image. +* (Specifically, allow it as a mirror target, which is write-only access.) +*/ +bdrv_get_cumulative_perm (state->new_bs, &perm, &shared); +if (perm &BLK_PERM_CONSISTENT_READ ) {error_setg (errp,"The overlay is already in use" ); goto out; } -if (bdrv_op_is_blocked (state->new_bs,BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT , - errp)) { - goto out; - } -if (state->new_bs->backing !=NULL ) {error_setg (errp,"The overlay already has a backing image" ); goto out;
tests/qemu-iotests/085.out +2 -2
@@ -82,7 +82,7 @@ Formatting 'TEST_DIR/12-snapshot-v0.IMGFMT', fmt=IMGFMT size=134217728 backing_f === Invalid command - cannot create a snapshot using a file BDS === { 'execute': 'blockdev-snapshot', 'arguments': { 'node':'virtio0', 'overlay':'file_12' } } -{"error": {"class": "GenericError", "desc": "The overlay does not support backing images"}} +{"error": {"class": "GenericError", "desc": "The overlay is already in use"}} === Invalid command - snapshot node used as active layer ===
@@ -96,7 +96,7 @@ Formatting 'TEST_DIR/12-snapshot-v0.IMGFMT', fmt=IMGFMT size=134217728 backing_f === Invalid command - snapshot node used as backing hd === { 'execute': 'blockdev-snapshot', 'arguments': { 'node': 'virtio0', 'overlay':'snap_11' } } -{"error": {"class": "GenericError", "desc": "Node 'snap_11' is busy: node is used as backing hd of 'snap_12'"}} +{"error": {"class": "GenericError", "desc": "The overlay is already in use"}} === Invalid command - snapshot node has a backing image ===
Signature · 1 line
2.20.1
Quoted 52 lines · Kevin Wolf
blockdev-snapshot returned an error if the overlay was already in use, which it defined as having any BlockBackend parent. This is in fact both too strict (some parents can tolerate the change of visible data caused by attaching a backing file) and too loose (some non-BlockBackend parents may not be happy with it).
One important use case that is prevented by the too strict check is live storage migration with blockdev-mirror. Here, the target node is usually opened without a backing file so that the active layer is mirrored while its backing chain can be copied in the background.
The backing chain should be attached to the mirror target node when finalising the job, just before switching the users of the source node to the new copy (at which point the mirror job still has a reference to the node). drive-mirror did this automatically, but with blockdev-mirror this is the job of the QMP client, so it needs a way to do this.
blockdev-snapshot is the obvious way, so this patch makes it work in this scenario. The new condition is that no parent uses
CONSISTENT_READpermissions. This will ensure that the operation will still be blocked when the node is attached to the guest device, so blockdev-snapshot remains safe.(For the sake of completeness, x-blockdev-reopen can be used to achieve the same, however it is a big hammer, performs the graph change completely unchecked and is still experimental. So even with the option of using x-blockdev-reopen, there are reasons why blockdev-snapshot should be able to perform this operation.)
Signed-off-by Kevin Wolf <[email protected]>--- blockdev.c | 14 ++++++++------ tests/qemu-iotests/085.out | 4 ++-- 2 files changed, 10 insertions(+), 8 deletions(-)blockdev.c +1
-1@@ -1501,6 +1501,7 @@ static void external_snapshot_prepare(BlkActionState *common,TransactionAction *action = common->action;AioContext *aio_context;AioContext *old_context; +uint64_t perm, shared;int ret;/* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar @@ -1616,16 +1617,17 @@ static void external_snapshot_prepare(BlkActionState *common, goto out; } -if (bdrv_has_blk (state->new_bs)) {
blockdev.c static void external_snapshot_prepare(BlkActionState *common, · line 1629
Quoted 10 lines
blockdev.c +7
-0+/* +* Allow attaching a backing file to an overlay that's already in use only +* if the parents don't assume that they are already seeing a valid image. +* (Specifically, allow it as a mirror target, which is write-only access.) +*/ +bdrv_get_cumulative_perm (state->new_bs, &perm, &shared); +if (perm &BLK_PERM_CONSISTENT_READ ) {error_setg (errp,"The overlay is already in use" ); goto out; }
Moving this code a bit further down ...
Quoted 23 lines
blockdev.c +0
-5-if (bdrv_op_is_blocked (state->new_bs,BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT , - errp)) { - goto out; - } -if (state->new_bs->backing !=NULL ) {error_setg (errp,"The overlay already has a backing image" ); goto out;tests/qemu-iotests/085.out +1
-1@@ -82,7 +82,7 @@ Formatting 'TEST_DIR/12-snapshot-v0.IMGFMT', fmt=IMGFMT size=134217728 backing_f === Invalid command - cannot create a snapshot using a file BDS === { 'execute': 'blockdev-snapshot', 'arguments': { 'node':'virtio0', 'overlay':'file_12' } } -{"error": {"class": "GenericError", "desc": "The overlay does not support backing images"}} +{"error": {"class": "GenericError", "desc": "The overlay is already in use"}} === Invalid command - snapshot node used as active layer ===
tests/qemu-iotests/085.out Formatting 'TEST_DIR/12-snapshot-v0.IMGFMT', fmt=IMGFMT size=134217728 backing_f · line 99
Quoted 5 lines
tests/qemu-iotests/085.out +1
-1@@ -96,7 +96,7 @@ Formatting 'TEST_DIR/12-snapshot-v0.IMGFMT', fmt=IMGFMT size=134217728 backing_f === Invalid command - snapshot node used as backing hd === { 'execute': 'blockdev-snapshot', 'arguments': { 'node': 'virtio0', 'overlay':'snap_11' } } -{"error": {"class": "GenericError", "desc": "Node 'snap_11' is busy: node is used as backing hd of 'snap_12'"}} +{"error": {"class": "GenericError", "desc": "The overlay is already in use"}}
Would probably prevent these test changes.
Am 10.03.2020 um 14:15 hat Peter Krempa geschrieben:
Quoted 88 lines · Kevin Wolf
Quoted 52 lines
blockdev-snapshot returned an error if the overlay was already in use, which it defined as having any BlockBackend parent. This is in fact both too strict (some parents can tolerate the change of visible data caused by attaching a backing file) and too loose (some non-BlockBackend parents may not be happy with it).
One important use case that is prevented by the too strict check is live storage migration with blockdev-mirror. Here, the target node is usually opened without a backing file so that the active layer is mirrored while its backing chain can be copied in the background.
The backing chain should be attached to the mirror target node when finalising the job, just before switching the users of the source node to the new copy (at which point the mirror job still has a reference to the node). drive-mirror did this automatically, but with blockdev-mirror this is the job of the QMP client, so it needs a way to do this.
blockdev-snapshot is the obvious way, so this patch makes it work in this scenario. The new condition is that no parent uses
CONSISTENT_READpermissions. This will ensure that the operation will still be blocked when the node is attached to the guest device, so blockdev-snapshot remains safe.(For the sake of completeness, x-blockdev-reopen can be used to achieve the same, however it is a big hammer, performs the graph change completely unchecked and is still experimental. So even with the option of using x-blockdev-reopen, there are reasons why blockdev-snapshot should be able to perform this operation.)
Signed-off-by Kevin Wolf <[email protected]>--- blockdev.c | 14 ++++++++------ tests/qemu-iotests/085.out | 4 ++-- 2 files changed, 10 insertions(+), 8 deletions(-)blockdev.c +1
-1@@ -1501,6 +1501,7 @@ static void external_snapshot_prepare(BlkActionState *common,TransactionAction *action = common->action;AioContext *aio_context;AioContext *old_context; +uint64_t perm, shared;int ret;/* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar @@ -1616,16 +1617,17 @@ static void external_snapshot_prepare(BlkActionState *common, goto out; } -if (bdrv_has_blk (state->new_bs)) {blockdev.c static void external_snapshot_prepare(BlkActionState *common, · line 1629
Quoted 10 lines
blockdev.c +7
-0+/* +* Allow attaching a backing file to an overlay that's already in use only +* if the parents don't assume that they are already seeing a valid image. +* (Specifically, allow it as a mirror target, which is write-only access.) +*/ +bdrv_get_cumulative_perm (state->new_bs, &perm, &shared); +if (perm &BLK_PERM_CONSISTENT_READ ) {error_setg (errp,"The overlay is already in use" ); goto out; }Moving this code a bit further down ...
Quoted 22 lines
blockdev.c +0
-5-if (bdrv_op_is_blocked (state->new_bs,BLOCK_OP_TYPE_EXTERNAL_SNAPSHOT , - errp)) { - goto out; - } -if (state->new_bs->backing !=NULL ) {error_setg (errp,"The overlay already has a backing image" ); goto out;tests/qemu-iotests/085.out +1
-1@@ -82,7 +82,7 @@ Formatting 'TEST_DIR/12-snapshot-v0.IMGFMT', fmt=IMGFMT size=134217728 backing_f === Invalid command - cannot create a snapshot using a file BDS === { 'execute': 'blockdev-snapshot', 'arguments': { 'node':'virtio0', 'overlay':'file_12' } } -{"error": {"class": "GenericError", "desc": "The overlay does not support backing images"}} +{"error": {"class": "GenericError", "desc": "The overlay is already in use"}} === Invalid command - snapshot node used as active layer ===
Quoted 8 lines
tests/qemu-iotests/085.out Formatting 'TEST_DIR/12-snapshot-v0.IMGFMT', fmt=IMGFMT size=134217728 backing_f · line 99
Quoted 6 lines
tests/qemu-iotests/085.out +1
-1@@ -96,7 +96,7 @@ Formatting 'TEST_DIR/12-snapshot-v0.IMGFMT', fmt=IMGFMT size=134217728 backing_f === Invalid command - snapshot node used as backing hd === { 'execute': 'blockdev-snapshot', 'arguments': { 'node': 'virtio0', 'overlay':'snap_11' } } -{"error": {"class": "GenericError", "desc": "Node 'snap_11' is busy: node is used as backing hd of 'snap_12'"}} +{"error": {"class": "GenericError", "desc": "The overlay is already in use"}}Would probably prevent these test changes.
It would prevent the first one, though I don't think it's important which of the two checks make it error out.
The second message came from the bdrv_op_is_blocked() that this patch removes, so this change would be there anyway.
Quoted 2 lines
Reviewed-by Peter Krempa <[email protected]>Tested-by Peter Krempa <[email protected]>
Thanks!
Kevin
PATCH v2 3/7 iotests: Fix run_job() with use_log=False
In reply to Kevin WolfThe 'job-complete' QMP command should be run with qmp() rather than qmp_log() if use_log=False is passed.
--- tests/qemu-iotests/iotests.py | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)
tests/qemu-iotests/iotests.py +4 -1
@@ -624,7 +624,10 @@ class VM(qtest.QEMUQtestMachine): if use_log: log('Job failed: %s' % (j['error'])) elif status == 'ready': - self.qmp_log('job-complete', id=job) + if use_log: + self.qmp_log('job-complete', id=job) + else: + self.qmp('job-complete', id=job) elif status == 'pending' and not auto_finalize: if pre_finalize: pre_finalize()
Signature · 1 line
2.20.1
Re: [PATCH v2 3/7] iotests: Fix run_job() with use_log=False
Quoted 7 lines · Kevin Wolf
The 'job-complete' QMP command should be run with
qmp()rather thanqmp_log()if use_log=False is passed.Signed-off-by Kevin Wolf <[email protected]>--- tests/qemu-iotests/iotests.py | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)
PATCH v2 4/7 iotests: Test mirror with temporarily disabled target backing file
In reply to Kevin WolfThe newly tested scenario is a common live storage migration scenario:
The target node is opened without a backing file so that the active layer is mirrored while its backing chain can be copied in the background.
The backing chain should be attached to the mirror target node when finalising the job, just before switching the users of the source node to the new copy (at which point the mirror job still has a reference to the node). drive-mirror did this automatically, but with blockdev-mirror this is the job of the QMP client.
This patch adds test cases for two ways to achieve the desired result, using either x-blockdev-reopen or blockdev-snapshot.
--- tests/qemu-iotests/155 | 56 ++++++++++++++++++++++++++++++++++---- tests/qemu-iotests/155.out | 4 +-- 2 files changed, 53 insertions(+), 7 deletions(-)
tests/qemu-iotests/155 +51 -5
@@ -45,10 +45,15 @@ target_img = os.path.join(iotests.test_dir, 'target.' + iotests.imgfmt) # image during runtime, only makes sense if # target_blockdev_backing is not None # (None: same as target_backing) +# target_open_with_backing: If True, the target image is added with its backing +# chain opened right away. If False, blockdev-add +# opens it without a backing file and job completion +# is supposed to open the backing chain. class BaseClass(iotests.QMPTestCase): target_blockdev_backing = None target_real_backing = None + target_open_with_backing = True def setUp(self): qemu_img('create', '-f', iotests.imgfmt, back0_img, '1440K')
@@ -80,9 +85,13 @@ class BaseClass(iotests.QMPTestCase): options = { 'node-name': 'target', 'driver': iotests.imgfmt, 'file': { 'driver': 'file', + 'node-name': 'target-file', 'filename': target_img } } - if self.target_blockdev_backing: - options['backing'] = self.target_blockdev_backing + + if not self.target_open_with_backing: + options['backing'] = None + elif self.target_blockdev_backing: + options['backing'] = self.target_blockdev_backing result = self.vm.qmp('blockdev-add', **options) self.assert_qmp(result, 'return', {})
@@ -147,10 +156,14 @@ class BaseClass(iotests.QMPTestCase): # cmd: Mirroring command to execute, either drive-mirror or blockdev-mirror class MirrorBaseClass(BaseClass): + def openBacking(self): + pass + def runMirror(self, sync): if self.cmd == 'blockdev-mirror': result = self.vm.qmp(self.cmd, job_id='mirror-job', device='source', - sync=sync, target='target') + sync=sync, target='target', + auto_finalize=False) else: if self.existing: mode = 'existing'
@@ -159,11 +172,12 @@ class MirrorBaseClass(BaseClass): result = self.vm.qmp(self.cmd, job_id='mirror-job', device='source', sync=sync, target=target_img, format=iotests.imgfmt, mode=mode, - node_name='target') + node_name='target', auto_finalize=False) self.assert_qmp(result, 'return', {}) - self.complete_and_wait('mirror-job') + self.vm.run_job('mirror-job', use_log=False, auto_finalize=False, + pre_finalize=self.openBacking, auto_dismiss=True) def testFull(self): self.runMirror('full')
@@ -221,6 +235,38 @@ class TestBlockdevMirrorForcedBacking(MirrorBaseClass): target_blockdev_backing = { 'driver': 'null-co' } target_real_backing = 'null-co://' +# Attach the backing chain only during completion, with blockdev-reopen +class TestBlockdevMirrorReopen(MirrorBaseClass): + cmd = 'blockdev-mirror' + existing = True + target_backing = 'null-co://' + target_open_with_backing = False + + def openBacking(self): + if not self.target_open_with_backing: + result = self.vm.qmp('blockdev-add', node_name="backing", + driver="null-co") + self.assert_qmp(result, 'return', {}) + result = self.vm.qmp('x-blockdev-reopen', node_name="target", + driver=iotests.imgfmt, file="target-file", + backing="backing") + self.assert_qmp(result, 'return', {}) + +# Attach the backing chain only during completion, with blockdev-snapshot +class TestBlockdevMirrorSnapshot(MirrorBaseClass): + cmd = 'blockdev-mirror' + existing = True + target_backing = 'null-co://' + target_open_with_backing = False + + def openBacking(self): + if not self.target_open_with_backing: + result = self.vm.qmp('blockdev-add', node_name="backing", + driver="null-co") + self.assert_qmp(result, 'return', {}) + result = self.vm.qmp('blockdev-snapshot', node="backing", + overlay="target") + self.assert_qmp(result, 'return', {}) class TestCommit(BaseClass): existing = False
tests/qemu-iotests/155.out +2 -2
@@ -1,5 +1,5 @@ -................... +......................... ---------------------------------------------------------------------- -Ran 19 tests +Ran 25 tests OK
Signature · 1 line
2.20.1
Re: [PATCH v2 4/7] iotests: Test mirror with temporarily disabled target backing file
Quoted 11 lines · Kevin Wolf
The newly tested scenario is a common live storage migration scenario:
The target node is opened without a backing file so that the active layer is mirrored while its backing chain can be copied in the background.The backing chain should be attached to the mirror target node when finalising the job, just before switching the users of the source node to the new copy (at which point the mirror job still has a reference to the node). drive-mirror did this automatically, but with blockdev-mirror this is the job of the QMP client.
Quoted 8 lines
This patch adds test cases for two ways to achieve the desired result, using either x-blockdev-reopen or blockdev-snapshot.
Signed-off-by Kevin Wolf <[email protected]>--- tests/qemu-iotests/155 | 56 ++++++++++++++++++++++++++++++++++---- tests/qemu-iotests/155.out | 4 +-- 2 files changed, 53 insertions(+), 7 deletions(-)
PATCH v2 5/7 block: Fix cross-AioContext blockdev-snapshot
In reply to Kevin Wolfexternal_snapshot_prepare() tries to move the overlay to the AioContext of the backing file (the snapshotted node). However, it's possible that this doesn't work, but the backing file can instead be moved to the overlay's AioContext (e.g. opening the backing chain for a mirror target).
bdrv_append() already indirectly uses bdrv_attach_node(), which takes care to move nodes to make sure they use the same AioContext and which tries both directions.
So the problem has a simple fix: Just delete the unnecessary extra bdrv_try_set_aio_context() call in external_snapshot_prepare() and instead assert in bdrv_append() that both nodes were indeed moved to the same AioContext.
--- block.c | 1 + blockdev.c | 16 ---------------- 2 files changed, 1 insertion(+), 16 deletions(-)
block.c +1 -0
@@ -4365,6 +4365,7 @@ void bdrv_replace_node(BlockDriverState *from, BlockDriverState *to,bdrv_ref (from);assert (qemu_get_current_aio_context () ==qemu_get_aio_context ()); +assert (bdrv_get_aio_context (from) ==bdrv_get_aio_context (to));bdrv_drained_begin (from);/* Put all parents into @list and calculate their cumulative permissions */
blockdev.c +0 -16
@@ -1500,9 +1500,7 @@ static void external_snapshot_prepare(BlkActionState *common,DO_UPCAST (ExternalSnapshotState, common, common);TransactionAction *action = common->action;AioContext *aio_context; -AioContext *old_context;uint64_t perm, shared; -int ret;/* 'blockdev-snapshot' and 'blockdev-snapshot-sync' have similar * purpose but a different set of parameters */
@@ -1638,20 +1636,6 @@ static void external_snapshot_prepare(BlkActionState *common, goto out; } -/* Honor bdrv_try_set_aio_context() context acquisition requirements. */ - old_context =bdrv_get_aio_context (state->new_bs); -aio_context_release (aio_context); -aio_context_acquire (old_context); - - ret =bdrv_try_set_aio_context (state->new_bs, aio_context, errp); - -aio_context_release (old_context); -aio_context_acquire (aio_context); - -if (ret <0 ) { - goto out; - } -/* This removes our old bs and adds the new bs. This is an operation that * can fail, so we need to do it in .prepare; undoing it for abort is * always possible. */
Signature · 1 line
2.20.1
Re: [PATCH v2 5/7] block: Fix cross-AioContext blockdev-snapshot
Quoted 15 lines · Kevin Wolf
external_snapshot_prepare()tries to move the overlay to the AioContext of the backing file (the snapshotted node). However, it's possible that this doesn't work, but the backing file can instead be moved to the overlay's AioContext (e.g. opening the backing chain for a mirror target).
bdrv_append()already indirectly usesbdrv_attach_node(), which takes care to move nodes to make sure they use the same AioContext and which tries both directions.So the problem has a simple fix: Just delete the unnecessary extra
bdrv_try_set_aio_context()call inexternal_snapshot_prepare()and instead assert inbdrv_append()that both nodes were indeed moved to the same AioContext.
Quoted 5 lines
Signed-off-by Kevin Wolf <[email protected]>--- block.c | 1 + blockdev.c | 16 ---------------- 2 files changed, 1 insertion(+), 16 deletions(-)
Unfortunately I don't feel confident enough in the intricacies of the aio contexts. Nevertheless Fixing bugs by deleting code is awesome!
PATCH v2 6/7 iotests: Add iothread cases to 155
In reply to Kevin WolfThis patch adds test cases for attaching the backing chain to a mirror job target right before finalising the job, where the image is in a non-mainloop AioContext (i.e. the backing chain needs to be moved to the AioContext of the mirror target).
This requires switching the test case from virtio-blk to virtio-scsi because virtio-blk only actually starts using the iothreads when the guest driver initialises the device (which never happens in a test case without a guest OS). virtio-scsi always keeps its block nodes in the AioContext of the the requested iothread without guest interaction.
--- tests/qemu-iotests/155 | 32 +++++++++++++++++++++++--------- tests/qemu-iotests/155.out | 4 ++-- 2 files changed, 25 insertions(+), 11 deletions(-)
tests/qemu-iotests/155 +23 -9
@@ -49,11 +49,14 @@ target_img = os.path.join(iotests.test_dir, 'target.' + iotests.imgfmt) # chain opened right away. If False, blockdev-add # opens it without a backing file and job completion # is supposed to open the backing chain. +# use_iothread: If True, an iothread is configured for the virtio-blk device +# that uses the image being mirrored class BaseClass(iotests.QMPTestCase): target_blockdev_backing = None target_real_backing = None target_open_with_backing = True + use_iothread = False def setUp(self): qemu_img('create', '-f', iotests.imgfmt, back0_img, '1440K')
@@ -69,7 +72,16 @@ class BaseClass(iotests.QMPTestCase): 'file': {'driver': 'file', 'filename': source_img}} self.vm.add_blockdev(self.vm.qmp_to_opts(blockdev)) - self.vm.add_device('virtio-blk,id=qdev0,drive=source') + + if self.use_iothread: + self.vm.add_object('iothread,id=iothread0') + iothread = ",iothread=iothread0" + else: + iothread = "" + + self.vm.add_device('virtio-scsi%s' % iothread) + self.vm.add_device('scsi-hd,id=qdev0,drive=source') + self.vm.launch() self.assertIntactSourceBackingChain()
@@ -182,24 +194,21 @@ class MirrorBaseClass(BaseClass): def testFull(self): self.runMirror('full') - node = self.findBlockNode('target', - '/machine/peripheral/qdev0/virtio-backend') + node = self.findBlockNode('target', 'qdev0') self.assertCorrectBackingImage(node, None) self.assertIntactSourceBackingChain() def testTop(self): self.runMirror('top') - node = self.findBlockNode('target', - '/machine/peripheral/qdev0/virtio-backend') + node = self.findBlockNode('target', 'qdev0') self.assertCorrectBackingImage(node, back2_img) self.assertIntactSourceBackingChain() def testNone(self): self.runMirror('none') - node = self.findBlockNode('target', - '/machine/peripheral/qdev0/virtio-backend') + node = self.findBlockNode('target', 'qdev0') self.assertCorrectBackingImage(node, source_img) self.assertIntactSourceBackingChain()
@@ -252,6 +261,9 @@ class TestBlockdevMirrorReopen(MirrorBaseClass): backing="backing") self.assert_qmp(result, 'return', {}) +class TestBlockdevMirrorReopenIothread(TestBlockdevMirrorReopen): + use_iothread = True + # Attach the backing chain only during completion, with blockdev-snapshot class TestBlockdevMirrorSnapshot(MirrorBaseClass): cmd = 'blockdev-mirror'
@@ -268,6 +280,9 @@ class TestBlockdevMirrorSnapshot(MirrorBaseClass): overlay="target") self.assert_qmp(result, 'return', {}) +class TestBlockdevMirrorSnapshotIothread(TestBlockdevMirrorSnapshot): + use_iothread = True + class TestCommit(BaseClass): existing = False
@@ -283,8 +298,7 @@ class TestCommit(BaseClass): self.vm.event_wait('BLOCK_JOB_COMPLETED') - node = self.findBlockNode(None, - '/machine/peripheral/qdev0/virtio-backend') + node = self.findBlockNode(None, 'qdev0') self.assert_qmp(node, 'image' + '/backing-image' * 0 + '/filename', back1_img) self.assert_qmp(node, 'image' + '/backing-image' * 1 + '/filename',
tests/qemu-iotests/155.out +2 -2
@@ -1,5 +1,5 @@ -......................... +............................... ---------------------------------------------------------------------- -Ran 25 tests +Ran 31 tests OK
Signature · 1 line
2.20.1
Re: [PATCH v2 6/7] iotests: Add iothread cases to 155
Quoted 11 lines · Kevin Wolf
This patch adds test cases for attaching the backing chain to a mirror job target right before finalising the job, where the image is in a non-mainloop AioContext (i.e. the backing chain needs to be moved to the AioContext of the mirror target).
This requires switching the test case from virtio-blk to virtio-scsi because virtio-blk only actually starts using the iothreads when the guest driver initialises the device (which never happens in a test case without a guest OS). virtio-scsi always keeps its block nodes in the AioContext of the the requested iothread without guest interaction.
Quoted 5 lines
Signed-off-by Kevin Wolf <[email protected]>--- tests/qemu-iotests/155 | 32 +++++++++++++++++++++++--------- tests/qemu-iotests/155.out | 4 ++-- 2 files changed, 25 insertions(+), 11 deletions(-)
PATCH v2 7/7 qapi: Add '@allow-write-only-overlay' feature for 'blockdev-snapshot'
In reply to Kevin WolfFrom: Peter Krempa <[email protected]>
Anounce that 'blockdev-snapshot' command's permissions allow changing of the backing file if the 'consistent_read' permission is not required.
This is useful for libvirt to allow late opening of the backing chain during a blockdev-mirror.
--- qapi/block-core.json | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
qapi/block-core.json +8 -1
@@ -1472,6 +1472,12 @@ # # For the arguments, see the documentation of BlockdevSnapshot. # +# Features: +# @allow-write-only-overlay: If present, the check whether this operation is safe +# was relaxed so that it can be used to change +# backing file of a destination of a blockdev-mirror. +# (since 5.0) +# # Since: 2.5 # # Example:
@@ -1492,7 +1498,8 @@ # ## { 'command': 'blockdev-snapshot', - 'data': 'BlockdevSnapshot' } + 'data': 'BlockdevSnapshot', + 'features': [ 'allow-write-only-overlay' ] } ## # @change-backing-file:
Signature · 1 line
2.20.1
Am 10.03.2020 um 12:38 hat Kevin Wolf geschrieben:
Quoted 11 lines
This series allows libvirt to fix a regression that its switch from drive-mirror to blockdev-mirror caused: It currently requires that the backing chain of the target image is already available when the mirror operation is started.
In reality, the backing chain may only be copied while the operation is in progress, so the backing file of the target image needs to stay disabled until the operation completes and should be attached only at that point. Without this series, we don't have a supported API to attach the backing file at that later point.
Quoted 5 lines
v2:
- Added a fix and test case for iothreads [Peter]
- Added a blockdev-snapshot feature flag to indicate that it's usable for attaching a backing chain to an overlay that is already in write-only use (e.g. as a mirror target) [Peter]
Thanks for review and testing, applied to the block branch.
Kevin