Calling job_pause_point() while holding the graph reader lock
potentially results in a deadlock: bdrv_graph_wrlock() first drains
everything, including the mirror job, which pauses it. The job is only
unpaused at the end of the drain section, which is when the graph writer
lock has been successfully taken. However, if the job happens to be
paused at a pause point where it still holds the reader lock, the writer
lock can't be taken as long as the job is still paused.
Mark job_pause_point() as GRAPH_UNLOCKED and fix mirror accordingly.
Cc: qemu-stable@nongnu.org
Buglink: https://issues.redhat.com/browse/RHEL-28125
Fixes: 004915a96a ("block: Protect bs->backing with graph_lock")
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20240313153000.33121-1-kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This is the big patch that removes
aio_context_acquire()/aio_context_release() from the block layer and
affected block layer users.
There isn't a clean way to split this patch and the reviewers are likely
the same group of people, so I decided to do it in one patch.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Paul Durrant <paul@xen.org>
Message-ID: <20231205182011.1976568-7-stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Stop acquiring/releasing the AioContext lock in
bdrv_graph_wrlock()/bdrv_graph_unlock() since the lock no longer has any
effect.
The distinction between bdrv_graph_wrunlock() and
bdrv_graph_wrunlock_ctx() becomes meaningless and they can be collapsed
into one function.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20231205182011.1976568-6-stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
bdrv_graph_wrunlock() calls aio_poll(), which may run callbacks that
have a nested event loop. Nested event loops can depend on other
iothreads making progress, so in order to allow them to make progress it
must not hold the AioContext lock of another thread while calling
aio_poll().
This introduces a @bs parameter to bdrv_graph_wrunlock() whose
AioContext is temporarily dropped (which matches bdrv_graph_wrlock()),
and a bdrv_graph_wrunlock_ctx() that can be used if the BlockDriverState
doesn't necessarily exist any more when unlocking.
This also requires a change to bdrv_schedule_unref(), which was relying
on the incorrectly taken lock. It needs to take the lock itself now.
While this is a separate bug, it can't be fixed a separate patch because
otherwise the intermediate state would either deadlock or try to release
a lock that we don't even hold.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20231115172012.112727-3-kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
[kwolf: Fixed up bdrv_schedule_unref()]
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Almost all functions that access bs->backing already take the graph
lock now. Add locking to the remaining users and finally annotate the
struct field itself as protected by the graph lock.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20231027155333.420094-18-kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Instead of taking the writer lock internally, require callers to already
hold it when calling bdrv_replace_node(). Its callers may already want
to hold the graph lock and so wouldn't be able to call functions that
take it internally.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20231027155333.420094-17-kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This adds GRAPH_RDLOCK annotations to declare that callers of
bdrv_(un)freeze_backing_chain() need to hold a reader lock for the
graph because it calls bdrv_filter_or_cow_child(), which accesses
bs->file/backing.
Use the opportunity to make bdrv_is_backing_chain_frozen() static, it
has no external callers.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20231027155333.420094-10-kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This adds GRAPH_RDLOCK annotations to declare that callers of
bdrv_skip_filters() need to hold a reader lock for the graph because it
calls bdrv_filter_child(), which accesses bs->file/backing.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20231027155333.420094-9-kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Instead of taking the writer lock internally, require callers to already
hold it when calling block_job_add_bdrv(). These callers will typically
already hold the graph lock once the locking work is completed, which
means that they can't call functions that take it internally.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20231027155333.420094-6-kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
To start out, only actively-synced is returned.
For example, this is useful for jobs that started out in background
mode and switched to active mode. Once actively-synced is true, it's
clear that the mode switch has been completed. Note that completion of
the switch might happen much earlier, e.g. if the switch happens
before the job is ready, once all background operations have finished.
It's assumed that whether the disks are actively-synced or not is more
interesting than whether the mode switch completed. That information
can still be added if required in the future.
In presence of an iothread, the actively_synced member is now shared
between the iothread and the main thread, so turn accesses to it
atomic.
Requires to adapt the output for iotest 109.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Message-ID: <20231031135431.393137-10-f.ebner@proxmox.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
which allows switching the @copy-mode from 'background' to
'write-blocking'.
This is useful for management applications, so they can start out in
background mode to avoid limiting guest write speed and switch to
active mode when certain criteria are fulfilled.
In presence of an iothread, the copy_mode member is now shared between
the iothread and the main thread, so turn accesses to it atomic.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Message-ID: <20231031135431.393137-6-f.ebner@proxmox.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
In preparation to allow changing the copy_mode via QMP. When running
in an iothread, it could be that copy_mode is changed from the main
thread in between reading copy_mode in bdrv_mirror_top_pwritev() and
reading copy_mode in bdrv_mirror_top_do_write(), so they might end up
disagreeing about whether copy_to_target is true or false. Avoid that
scenario by determining copy_to_target only once and passing it to
bdrv_mirror_top_do_write() as an argument.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Message-ID: <20231031135431.393137-5-f.ebner@proxmox.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
In preparation to allow switching to active mode without draining.
Initialization of the bitmap in mirror_dirty_init() still happens with
the original/backing BlockDriverState, which should be fine, because
the mirror top has the same length.
Suggested-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Message-ID: <20231031135431.393137-4-f.ebner@proxmox.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
When target image is using subclusters, and we align the request during
copy-on-read, it makes sense to align to subcluster_size rather than
cluster_size. Otherwise we end up with unnecessary allocations.
This commit renames bdrv_round_to_clusters() to bdrv_round_to_subclusters()
and utilizes subcluster_size field of BlockDriverInfo to make necessary
alignments. It affects copy-on-read as well as mirror job (which is
using bdrv_round_to_clusters()).
This change also fixes the following bug with failing assert (covered by
the test in the subsequent commit):
qemu-img create -f qcow2 base.qcow2 64K
qemu-img create -f qcow2 -o extended_l2=on,backing_file=base.qcow2,backing_fmt=qcow2 img.qcow2 64K
qemu-io -c "write -P 0xaa 0 2K" img.qcow2
qemu-io -C -c "read -P 0x00 2K 62K" img.qcow2
qemu-io: ../block/io.c:1236: bdrv_co_do_copy_on_readv: Assertion `skip_bytes < pnum' failed.
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Andrey Drobyshev <andrey.drobyshev@virtuozzo.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-ID: <20230711172553.234055-3-andrey.drobyshev@virtuozzo.com>
When jobs are sleeping, for example to enforce a given rate limit, they
can be reentered early, in particular in order to get paused, to update
the rate limit or to get cancelled.
Before this patch, they behave in this case as if they had fully
completed their rate limiting delay. This means that requests are sped
up beyond their limit, violating the constraints that the user gave us.
Change the block jobs to sleep in a loop until the necessary delay is
completed, while still allowing cancelling them immediately as well
pausing (handled by the pause point in job_sleep_ns()) and updating the
rate limit.
This change is also motivated by iotests cases being prone to fail
because drain operations pause and unpause them so often that block jobs
complete earlier than they are supposed to. In particular, the next
commit would fail iotests 030 without this change.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20230510203601.418015-8-kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Kevin Wolf
Stefan HajnocziandKevin Wolf
Fiona EbnerandKevin Wolf
Paolo BonziniandKevin Wolf
Michael Tokarev
Andrey DrobyshevandStefan Hajnoczi
Emanuele Giuseppe EspositoandKevin Wolf