This patch completes the series with the COR-filter applied to
block-stream operations.
Adding the filter makes it possible in future implement discarding
copied regions in backing files during the block-stream job, to reduce
the disk overuse (we need control on permissions).
Also, the filter now is smart enough to do copy-on-read with specified
base, so we have benefit on guest reads even when doing block-stream of
the part of the backing chain.
Several iotests are slightly modified due to filter insertion.
Signed-off-by: Andrey Shinkevich <andrey.shinkevich@virtuozzo.com>
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20201216061703.70908-14-vsementsov@virtuozzo.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
With bash 5.1, the output of the following script changes:
a=("double space")
a=${a[@]:0:1}
echo "$a"
from "double space" to "double space", i.e. all white space is
preserved as-is. This is probably what we actually want here (judging
from the "...to accommodate pathnames with spaces" comment), but before
5.1, we would have to quote the ${} slice to get the same behavior.
In any case, without quoting, the reference output of many iotests is
different between bash 5.1 and pre-5.1, which is not very good. The
output of 5.1 is what we want, so whatever we do to get pre-5.1 to the
same result, it means we have to fix the reference output of basically
all tests that invoke _send_qemu_cmd (except the ones that only use
single spaces in the commands they invoke).
Instead of quoting the ${} slice (cmd="${$@: 1:...}"), we can also just
not use array slicing and replace the whole thing with a simple "cmd=$1;
shift", which works because all callers quote the whole $cmd argument
anyway.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20201217153803.101231-3-mreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
There are many existing qcow2 images that specify a backing file but
no format. This has been the source of CVEs in the past, but has
become more prominent of a problem now that libvirt has switched to
-blockdev. With older -drive, at least the probing was always done by
qemu (so the only risk of a changed format between successive boots of
a guest was if qemu was upgraded and probed differently). But with
newer -blockdev, libvirt must specify a format; if libvirt guesses raw
where the image was formatted, this results in data corruption visible
to the guest; conversely, if libvirt guesses qcow2 where qemu was
using raw, this can result in potential security holes, so modern
libvirt instead refuses to use images without explicit backing format.
The change in libvirt to reject images without explicit backing format
has pointed out that a number of tools have been far too reliant on
probing in the past. It's time to set a better example in our own
iotests of properly setting this parameter.
iotest calls to create, rebase, and convert are all impacted to some
degree. It's a bit annoying that we are inconsistent on command line
- while all of those accept -o backing_file=...,backing_fmt=..., the
shortcuts are different: create and rebase have -b and -F, while
convert has -B but no -F. (amend has no shortcuts, but the previous
patch just deprecated the use of amend to change backing chains).
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200706203954.341758-9-eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Issuing a drive-backup from qmp_drive_backup takes a slightly
different path than when it's issued from a transaction. In the code,
this is manifested as some redundancy between do_drive_backup() and
drive_backup_prepare().
This change unifies both paths, merging do_drive_backup() and
drive_backup_prepare(), and changing qmp_drive_backup() to create a
transaction instead of calling do_backup_common() direcly.
As a side-effect, now qmp_drive_backup() is executed inside a drained
section, as it happens when creating a drive-backup transaction. This
change is visible from the user's perspective, as the job gets paused
and immediately resumed before starting the actual work.
Also fix tests 141, 185 and 219 to cope with the extra
JOB_STATUS_CHANGE lines.
Signed-off-by: Sergio Lopez <slp@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
We generally include relevant HMP input in .out files, by virtue of
the fact that HMP echoes its input. But QMP does not, so we have to
explicitly inject it in the output stream (appropriately filtered to
keep the tests passing), in order to make it easier to read .out files
to see what behavior is being tested (especially true where the output
file is a sequence of {'return': {}}).
Suggested-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20191114213415.23499-4-eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Drop write notifiers and use filter node instead.
= Changes =
1. Add filter-node-name argument for backup qmp api. We have to do it
in this commit, as 257 needs to be fixed.
2. There are no more write notifiers here, so is_write_notifier
parameter is dropped from block-copy paths.
3. To sync with in-flight requests at job finish we now have drained
removing of the filter, we don't need rw-lock.
4. Block-copy is now using BdrvChildren instead of BlockBackends
5. As backup-top owns these children, we also move block-copy state
into backup-top's ownership.
= Iotest changes =
56: op-blocker doesn't shoot now, as we set it on source, but then
check on filter, when trying to start second backup.
To keep the test we instead can catch another collision: both jobs will
get 'drive0' job-id, as job-id parameter is unspecified. To prevent
interleaving with file-posix locks (as they are dependent on config)
let's use another target for second backup.
Also, it's obvious now that we'd like to drop this op-blocker at all
and add a test-case for two backups from one node (to different
destinations) actually works. But not in these series.
141: Output changed: prepatch, "Node is in use" comes from bdrv_has_blk
check inside qmp_blockdev_del. But we've dropped block-copy blk
objects, so no more blk objects on source bs (job blk is on backup-top
filter bs). New message is from op-blocker, which is the next check in
qmp_blockdev_add.
257: The test wants to emulate guest write during backup. They should
go to filter node, not to original source node, of course. Therefore we
need to specify filter node name and use it.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-id: 20191001131409.14202-6-vsementsov@virtuozzo.com
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
69f47505ee has changed qcow2 in such a way that the commit job run in
test 141 (and 144[1]) returns before it emits the READY event. However,
141 also runs with qed, where the order is still the other way around.
Just filter out the {"return": {}} so the test passes for qed again.
[1] 144 only runs with qcow2, so it is fine as it is.
Suggested-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Fixes: 69f47505ee
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190809185253.17535-1-mreitz@redhat.com
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
drv_co_block_status digs bs->file for additional, more accurate search
for hole inside region, reported as DATA by bs since 5daa74a6eb.
This accuracy is not free: assume we have qcow2 disk. Actually, qcow2
knows, where are holes and where is data. But every block_status
request calls lseek additionally. Assume a big disk, full of
data, in any iterative copying block job (or img convert) we'll call
lseek(HOLE) on every iteration, and each of these lseeks will have to
iterate through all metadata up to the end of file. It's obviously
ineffective behavior. And for many scenarios we don't need this lseek
at all.
However, lseek is needed when we have metadata-preallocated image.
So, let's detect metadata-preallocation case and don't dig qcow2's
protocol file in other cases.
The idea is to compare allocation size in POV of filesystem with
allocations size in POV of Qcow2 (by refcounts). If allocation in fs is
significantly lower, consider it as metadata-preallocation case.
102 iotest changed, as our detector can't detect shrinked file as
metadata-preallocation, which don't seem to be wrong, as with metadata
preallocation we always have valid file length.
Two other iotests have a slight change in their QMP output sequence:
Active 'block-commit' returns earlier because the job coroutine yields
earlier on a blocking operation. This operation is loading the refcount
blocks in qcow2_detect_metadata_preallocation().
Suggested-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The mirror_start_job() function used for the commit-active job blocks
the source, target and all intermediate nodes for the duration of the
job.
target <- intermediate <- source
Since 4ef85a9c23 this function creates a dummy mirror_top_bs that
goes on top of the source node, and it is this dummy node that gets
blocked instead. The source node is never blocked or added to the
job's list of nodes.
target <- intermediate <- source <- mirror_top
At the moment I don't think it is possible to exploit this problem
because any additional job on 'source' would either be forbidden for
other reasons or it would need to involve an additional node that is
blocked, causing an error.
This can be seen in the error messages, however, because they never
refer to the source node being blocked:
$ qemu-img create -f qcow2 hd0.qcow2 1M
$ qemu-img create -f qcow2 -b hd0.qcow2 hd1.qcow2
$ qemu-io -c 'write 0 1M' hd0.qcow2
$ $QEMU -drive if=none,file=hd1.qcow2,node-name=hd1
{ "execute": "qmp_capabilities" }
{ "execute": "block-commit", "arguments": {"device": "hd1", "speed": 256}}
{ "execute": "block-stream", "arguments": {"device": "hd1"}}
{ "error": {"class": "GenericError",
"desc": "Node 'hd0' is busy: block device is in use by block job: commit"}}
After this patch the error message refers to 'hd1', as it should.
The expected output of iotest 141 also needs to be updated for the
same reason.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This adds a QMP event that is emitted whenever a job transitions from
one status to another.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
The mirror block job is mainly used for two different scenarios:
Mirroring to an otherwise unused, independent target node, or for active
commit where the target node is part of the backing chain of the source.
Similarly to the commit block job patch, we need to insert a new filter
node to keep the permissions correct during active commit.
Note that one change this implies is that job->blk points to
mirror_top_bs as its root now, and mirror_top_bs (rather than the actual
source node) contains the bs->job pointer. This requires qemu-img commit
to get the job by name now rather than just taking bs->job.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Acked-by: Fam Zheng <famz@redhat.com>
Acked-by: Max Reitz <mreitz@redhat.com>
We want to remove the 'id' option for blockdev-add. This removes one
user of the option and makes it use only node names.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Some guests (win2008 server for example) do a lot of unnecessary
flushing when underlying media has not changed. This adds additional
overhead on host when calling fsync/fdatasync.
This change introduces a write generation scheme in BlockDriverState.
Current write generation is checked against last flushed generation to
avoid unnessesary flushes.
The problem with excessive flushing was found by a performance test
which does parallel directory tree creation (from 2 processes).
Results improved from 0.424 loops/sec to 0.432 loops/sec.
Each loop creates 10^3 directories with 10 files in each.
This affected some blkdebug testcases that were expecting error logs from
failure-injected flushes which are now skipped entirely
(tests 026 071 089).
This also affects the performance of block jobs and thus BLOCK_JOB_READY
events for driver-mirror and active block-commit commands now arrives
faster, before QMP send successfully returns to caller (tests 141 144).
Signed-off-by: Evgeny Yakovlev <eyakovlev@virtuozzo.com>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-id: 1468870792-7411-5-git-send-email-den@openvz.org
CC: Kevin Wolf <kwolf@redhat.com>
CC: Max Reitz <mreitz@redhat.com>
CC: Stefan Hajnoczi <stefanha@redhat.com>
CC: Fam Zheng <famz@redhat.com>
CC: John Snow <jsnow@redhat.com>
Signed-off-by: John Snow <jsnow@redhat.com>
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
