Decrementing drained_end_counter after bdrv_dec_in_flight() (which in
turn invokes bdrv_wakeup() and thus aio_wait_kick()) is not very clever.
We should decrement it beforehand, so that any waiting aio_poll() that
is woken by bdrv_dec_in_flight() sees the decremented
drained_end_counter.
Because the time window between decrementing drained_end_counter and
aio_wait_kick() is very small, I cannot supply a reliable regression
test. However, running e.g. the /bdrv-drain/blockjob/iothread/drain_all
test in test-bdrv-drain has a small chance of hanging without this
patch (about 1/200 or so; it gets to nearly 100 % if you add e.g. an
fputc(' ', stderr); after the bdrv_dec_in_flight()).
Fixes: e037c09c78
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190722133054.21781-2-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
The graph must not change in these loops (or a QLIST_FOREACH_SAFE would
not even be enough). We now ensure this by only polling once in the
root bdrv_drained_end() call, so we can drop the _SAFE suffix. Doing so
makes it clear that the graph must not change.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
We should never poll anywhere in bdrv_do_drained_end() (including its
recursive callees like bdrv_drain_invoke()), because it does not cope
well with graph changes. In fact, it has been written based on the
postulation that no graph changes will happen in it.
Instead, the callers that want to poll must poll, i.e. all currently
globally available wrappers: bdrv_drained_end(),
bdrv_subtree_drained_end(), bdrv_unapply_subtree_drain(), and
bdrv_drain_all_end(). Graph changes there do not matter.
They can poll simply by passing a pointer to a drained_end_counter and
wait until it reaches 0.
This patch also adds a non-polling global wrapper for
bdrv_do_drained_end() that takes a drained_end_counter pointer. We need
such a variant because now no function called anywhere from
bdrv_do_drained_end() must poll. This includes
BdrvChildRole.drained_end(), which already must not poll according to
its interface documentation, but bdrv_child_cb_drained_end() just
violates that by invoking bdrv_drained_end() (which does poll).
Therefore, BdrvChildRole.drained_end() must take a *drained_end_counter
parameter, which bdrv_child_cb_drained_end() can pass on to the new
bdrv_drained_end_no_poll() function.
Note that we now have a pattern of all drained_end-related functions
either polling or receiving a *drained_end_counter to let the caller
poll based on that.
A problem with a single poll loop is that when the drained section in
bdrv_set_aio_context_ignore() ends, some nodes in the subgraph may be in
the old contexts, while others are in the new context already. To let
the collective poll in bdrv_drained_end() work correctly, we must not
hold a lock to the old context, so that the old context can make
progress in case it is different from the current context.
(In the process, remove the comment saying that the current context is
always the old context, because it is wrong.)
In all other places, all nodes in a subtree must be in the same context,
so we can just poll that. The exception of course is
bdrv_drain_all_end(), but that always runs in the main context, so we
can just poll NULL (like bdrv_drain_all_begin() does).
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
These functions are not used outside of block/io.c, there is no reason
why they should be globally available.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Callers can now pass a pointer to an integer that bdrv_drain_invoke()
(and its recursive callees) will increment for every
bdrv_drain_invoke_entry() operation they schedule.
bdrv_drain_invoke_entry() in turn will decrement it once it has invoked
BlockDriver.bdrv_co_drain_end().
We use atomic operations to access the pointee, because the
bdrv_do_drained_end() caller may wish to end drained sections for
multiple nodes in different AioContexts (bdrv_drain_all_end() does, for
example).
This is the first step to moving the polling for BdrvCoDrainData.done to
become true out of bdrv_drain_invoke() and into the root drained_end
function.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Commit 5cb2737e92 laid out why
bdrv_do_drained_end() must decrement the quiesce_counter after
bdrv_drain_invoke(). It did not give a very good reason why it has to
happen after bdrv_parent_drained_end(), instead only claiming symmetry
to bdrv_do_drained_begin().
It turns out that delaying it for so long is wrong.
Situation: We have an active commit job (i.e. a mirror job) from top to
base for the following graph:
filter
|
[file]
|
v
top --[backing]--> base
Now the VM is closed, which results in the job being cancelled and a
bdrv_drain_all() happening pretty much simultaneously.
Beginning the drain means the job is paused once whenever one of its
nodes is quiesced. This is reversed when the drain ends.
With how the code currently is, after base's drain ends (which means
that it will have unpaused the job once), its quiesce_counter remains at
1 while it goes to undrain its parents (bdrv_parent_drained_end()). For
some reason or another, undraining filter causes the job to be kicked
and enter mirror_exit_common(), where it proceeds to invoke
block_job_remove_all_bdrv().
Now base will be detached from the job. Because its quiesce_counter is
still 1, it will unpause the job once more. So in total, undraining
base will unpause the job twice. Eventually, this will lead to the
job's pause_count going negative -- well, it would, were there not an
assertion against this, which crashes qemu.
The general problem is that if in bdrv_parent_drained_end() we undrain
parent A, and then undrain parent B, which then leads to A detaching the
child, bdrv_replace_child_noperm() will undrain A as if we had not done
so yet; that is, one time too many.
It follows that we cannot decrement the quiesce_counter after invoking
bdrv_parent_drained_end().
Unfortunately, decrementing it before bdrv_parent_drained_end() would be
wrong, too. Imagine the above situation in reverse: Undraining A leads
to B detaching the child. If we had already decremented the
quiesce_counter by that point, bdrv_replace_child_noperm() would undrain
B one time too little; because it expects bdrv_parent_drained_end() to
issue this undrain. But bdrv_parent_drained_end() won't do that,
because B is no longer a parent.
Therefore, we have to do something else. This patch opts for
introducing a second quiesce_counter that counts how many times a
child's parent has been quiesced (though c->role->drained_*). With
that, bdrv_replace_child_noperm() just has to undrain the parent exactly
that many times when removing a child, and it will always be right.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This patch is used in the 'block/stream: introduce a bottom node'
that is following. Instead of the base node, the caller may pass
the node that has the base as its backing image to the function
bdrv_is_allocated_above() with a new parameter include_base = true
and get rid of the dependency on the base that may change during
commit/stream parallel jobs. Now, if the specified base is not
found in the backing image chain, the QEMU will abort.
Suggested-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Andrey Shinkevich <andrey.shinkevich@virtuozzo.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 1559152576-281803-2-git-send-email-andrey.shinkevich@virtuozzo.com
[mreitz: Squashed in the following as a rebase on conflicting patches:]
Message-id: e3cf99ae-62e9-8b6e-5a06-d3c8b9363b85@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
This fixes at least one overflow in qcow2_process_discards, which
passes 64bit region length to bdrv_pdiscard where bytes (or sectors in
the past) parameter is int since its introduction in 0b919fae.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
When ending a drained section, bdrv_do_drained_end() currently first
decrements the quiesce_counter, and only then actually ends the drain.
The bdrv_drain_invoke(bs, false) call may cause graph changes. Say the
graph change involves replacing an existing BB's ("blk") BDS
(blk_bs(blk)) by @bs. Let us introducing the following values:
- bs_oqc = old_quiesce_counter
(so bs->quiesce_counter == bs_oqc - 1)
- obs_qc = blk_bs(blk)->quiesce_counter (before bdrv_drain_invoke())
Let us assume there is no blk_pread_unthrottled() involved, so
blk->quiesce_counter == obs_qc (before bdrv_drain_invoke()).
Now replacing blk_bs(blk) by @bs will reduce blk->quiesce_counter by
obs_qc (making it 0) and increase it by bs_oqc-1 (making it bs_oqc-1).
bdrv_drain_invoke() returns and we invoke bdrv_parent_drained_end().
This will decrement blk->quiesce_counter by one, so it would be -1 --
were there not an assertion against that in blk_root_drained_end().
We therefore have to keep the quiesce_counter up at least until
bdrv_drain_invoke() returns, so that bdrv_parent_drained_end() does the
right thing for the parents @bs got during bdrv_drain_invoke().
But let us delay it even further, namely until bdrv_parent_drained_end()
returns, because then it mirrors bdrv_do_drained_begin(): There, we
first increment the quiesce_counter, then begin draining the parents,
and then call bdrv_drain_invoke(). It makes sense to let
bdrv_do_drained_end() unravel this exactly in reverse.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@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>
There are a few places in which we turn a number of bytes into sectors
in order to compare the result against BDRV_REQUEST_MAX_SECTORS
instead of using BDRV_REQUEST_MAX_BYTES directly.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
No one is using these functions anymore, all callers have switched to
the byte-based bdrv_pread() and bdrv_pwrite()
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
On a file system used by the customer, fallocate() returns an error
if the block is not properly aligned. So, bdrv_co_pwrite_zeroes()
fails. We can handle that case the same way as it is done for the
unsupported cases, namely, call to bdrv_driver_pwritev() that writes
zeroes to an image for the unaligned chunk of the block.
Suggested-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Andrey Shinkevich <andrey.shinkevich@virtuozzo.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Message-id: 1554474244-553661-1-git-send-email-andrey.shinkevich@virtuozzo.com
Message-Id: <1554474244-553661-1-git-send-email-andrey.shinkevich@virtuozzo.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
For qemu-img convert, we want an operation that zeroes out the whole
image if this can be done efficiently, but that returns an error
otherwise so we don't write explicit zeroes and immediately overwrite
them with the real data, potentially doubling the amount of data to be
written.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Acked-by: Eric Blake <eblake@redhat.com>
There is only a single caller of bdrv_make_zero(), which is qemu-img
convert. If the function fails, we just fall back to a different method
of zeroing out blocks on the target image. There is no good reason to
print error messages on stderr when the higher level operation will
actually succeed.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Acked-by: Eric Blake <eblake@redhat.com>
Use new qemu_iovec_init_buf() instead of
qemu_iovec_init_external( ... , 1), which simplifies the code.
While being here, use qemu_try_blockalign0 as well.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 20190218140926.333779-3-vsementsov@virtuozzo.com
Message-Id: <20190218140926.333779-3-vsementsov@virtuozzo.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
In the block layer, synchronous APIs are often implemented by creating a
coroutine that calls the asynchronous coroutine-based implementation and
then waiting for completion with BDRV_POLL_WHILE().
For this to work with iothreads (more specifically, when the synchronous
API is called in a thread that is not the home thread of the block
device, so that the coroutine will run in a different thread), we must
make sure to call aio_wait_kick() at the end of the operation. Many
places are missing this, so that BDRV_POLL_WHILE() keeps hanging even if
the condition has long become false.
Note that bdrv_dec_in_flight() involves an aio_wait_kick() call. This
corresponds to the BDRV_POLL_WHILE() in the drain functions, but it is
generally not enough for most other operations because they haven't set
the return value in the coroutine entry stub yet. To avoid race
conditions there, we need to kick after setting the return value.
The race window is small enough that the problem doesn't usually surface
in the common path. However, it does surface and causes easily
reproducible hangs if the operation can return early before even calling
bdrv_inc/dec_in_flight, which many of them do (trivial error or no-op
success paths).
The bug in bdrv_truncate(), bdrv_check() and bdrv_invalidate_cache() is
slightly different: These functions even neglected to schedule the
coroutine in the home thread of the node. This avoids the hang, but is
obviously wrong, too. Fix those to schedule the coroutine in the right
AioContext in addition to adding aio_wait_kick() calls.
Cc: qemu-stable@nongnu.org
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
When draining a block node, we recurse to its parent and for subtree
drains also to its children. A single AIO_WAIT_WHILE() is then used to
wait for bdrv_drain_poll() to become true, which depends on all of the
nodes we recursed to. However, if the respective child or parent becomes
quiescent and calls bdrv_wakeup(), only the AioWait of the child/parent
is checked, while AIO_WAIT_WHILE() depends on the AioWait of the
original node.
Fix this by using a single AioWait for all callers of AIO_WAIT_WHILE().
This may mean that the draining thread gets a few more unnecessary
wakeups because an unrelated operation got completed, but we already
wake it up when something _could_ have changed rather than only if it
has certainly changed.
Apart from that, drain is a slow path anyway. In theory it would be
possible to use wakeups more selectively and still correctly, but the
gains are likely not worth the additional complexity. In fact, this
patch is a nice simplification for some places in the code.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
bdrv_drain_poll_top_level() was buggy because it didn't release the
AioContext lock of the node to be drained before calling aio_poll().
This way, callbacks called by aio_poll() would possibly take the lock a
second time and run into a deadlock with a nested AIO_WAIT_WHILE() call.
However, it turns out that the aio_poll() call isn't actually needed any
more. It was introduced in commit 91af091f92, which is effectively
reverted by this patch. The cases it was supposed to fix are now covered
by bdrv_drain_poll(), which waits for block jobs to reach a quiescent
state.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
bdrv_do_drained_begin/end() assume that they are called with the
AioContext lock of bs held. If we call drain functions from a coroutine
with the AioContext lock held, we yield and schedule a BH to move out of
coroutine context. This means that the lock for the home context of the
coroutine is released and must be re-acquired in the bottom half.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Truncation is the last to convert from open coded req handling to
reusing helpers. This time the permission check in prepare has to adapt
to the new caller: it checks a different permission bit, and doesn't
trigger the before write notifier.
Also, truncation should always trigger a bs->total_sectors update and in
turn call parent resize_cb. Update the condition in finish helper, too.
It's intended to do a duplicated bs->read_only check before calling
bdrv_co_write_req_prepare() so that we can be more informative with the
error message, as bdrv_co_write_req_prepare() doesn't have Error
parameter.
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
If we are growing the image and potentially using preallocation for the
new area, we need to make sure that no write requests are made to the
"preallocated" area which is [@old_size, @offset), not
[@offset, offset * 2 - @old_size).
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This brings the request handling logic inline with write and discard,
fixing write_gen, resize_cb, dirty bitmaps and image size refreshing.
The last of these issues broke iotest case 222, which is now fixed.
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reuse the new bdrv_co_write_req_prepare/finish helpers. The variation
here is that discard requests don't affect bs->wr_highest_offset, and it
cannot extend the image.
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Two problems exist when a write request that enlarges the image (i.e.
write beyond EOF) finishes:
1) parent is not notified about size change;
2) dirty bitmap is not resized although we try to set the dirty bits;
Fix them just like how bdrv_co_truncate works.
Reported-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
As a mechanical refactoring patch, this is the first step towards
unified and more correct write code paths. This is helpful because
multiple BlockDriverState fields need to be updated after modifying
image data, and it's hard to maintain in multiple places such as copy
offload, discard and truncate.
Suggested-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This matches the types used for bytes in the rest parts of block layer.
In the case of bdrv_co_truncate, new_bytes can be the image size which
probably doesn't fit in a 32 bit int.
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Other I/O functions are already using a BdrvChild pointer in the API, so
make discard do the same. It makes it possible to initiate the same
permission checks before doing I/O, and much easier to share the
helper functions for this, which will be added and used by write,
truncate and copy range paths.
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
A few trace points that can help reveal what is happening in a copy
offloading I/O path.
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Serialized writes should be used in copy-on-write of backup(sync=none)
for image fleecing scheme.
We need to change an assert in bdrv_aligned_pwritev, added in
28de2dcd88. The assert may fail now, because call to
wait_serialising_requests here may become first call to it for this
request with serializing flag set. It occurs if the request is aligned
(otherwise, we should already set serializing flag before calling
bdrv_aligned_pwritev and correspondingly waited for all intersecting
requests). However, for aligned requests, we should not care about
outdating of previously read data, as there no such data. Therefore,
let's just update an assert to not care about aligned requests.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Pass read flags and write flags separately. This is needed to handle
coming BDRV_REQ_NO_SERIALISING clearly in following patches.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Here two things are fixed:
1. Architecture
On each recursion step, we go to the child of src or dst, only for one
of them. So, it's wrong to create tracked requests for both on each
step. It leads to tracked requests duplication.
2. Wait for serializing requests on write path independently of
BDRV_REQ_NO_SERIALISING
Before commit 9ded4a0114 "backup: Use copy offloading",
BDRV_REQ_NO_SERIALISING was used for only one case: read in
copy-on-write operation during backup. Also, the flag was handled only
on read path (in bdrv_co_preadv and bdrv_aligned_preadv).
After 9ded4a0114, flag is used for not waiting serializing operations
on backup target (in same case of copy-on-write operation). This
behavior change is unsubstantiated and potentially dangerous, let's
drop it and add additional asserts and documentation.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
If the virtual disk size isn't aligned to full clusters,
bdrv_co_do_copy_on_readv() may get pnum == 0 before having the full
cluster completed, which will let it run into an assertion failure:
qemu-io: block/io.c:1203: bdrv_co_do_copy_on_readv: Assertion `skip_bytes < pnum' failed.
Check for EOF, assert that we read at least as much as the read request
originally wanted to have (which is true at EOF because otherwise
bdrv_check_byte_request() would already have returned an error) and
return success early even though we couldn't copy the full cluster.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Commit dcf94a23b1 ('block: Don't poll in parent drain callbacks')
removed polling in bdrv_child_cb_drained_begin() on the grounds that the
original bdrv_drain() already will poll and BdrvChildRole.drained_begin
calls must not cause graph changes (and therefore must not call
aio_poll() or the recursion through the graph will break.
This reasoning is correct for calls through bdrv_do_drained_begin().
However, BdrvChildRole.drained_begin is also called when a node that is
already in a drained section (i.e. bdrv_do_drained_begin() has already
returned and therefore can't poll any more) is attached to a new parent.
In this case, we must explicitly poll to have all requests completed
before the drained new child can be attached to the parent.
In bdrv_replace_child_noperm(), we know that we're not inside the
recursion of bdrv_do_drained_begin() because graph changes are not
allowed there, and bdrv_replace_child_noperm() is a graph change. The
call of BdrvChildRole.drained_begin() must therefore be followed by a
BDRV_POLL_WHILE() that waits for the completion of requests.
Reported-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This semantics is needed by drive-backup so implement it before using
this API there.
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
Message-id: 20180703023758.14422-3-famz@redhat.com
Signed-off-by: Jeff Cody <jcody@redhat.com>
src may be NULL if BDRV_REQ_ZERO_WRITE flag is set, in this case only
check dst and dst->bs. This bug was introduced when moving in the
request tracking code from bdrv_co_copy_range, in 37aec7d75e.
This especially fixes the possible segfault when initializing src_bs
with a NULL src.
Signed-off-by: Fam Zheng <famz@redhat.com>
Message-id: 20180703023758.14422-2-famz@redhat.com
Reviewed-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Jeff Cody <jcody@redhat.com>
We are gradually moving away from sector-based interfaces, towards
byte-based. Now that all callers of vectored I/O have been converted
to use our preferred byte-based bdrv_co_p{read,write}v(), we can
delete the unused bdrv_co_{read,write}v().
Furthermore, this gets rid of the signature difference between the
public bdrv_co_writev() and the callback .bdrv_co_writev (the
latter still exists, because some drivers still need more work
before they are fully byte-based).
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
in_flight and tracked requests need to be tracked in every layer during
recursion. For now the only user is qemu-img convert where overlapping
requests and IOThreads don't exist, therefore this change doesn't make
much difference form user point of view, but it is incorrect as part of
the API. Fix it.
Reported-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
When growing an image, block drivers (especially protocol drivers) may
initialise the newly added area. I/O requests to the same area need to
wait for this initialisation to be completed so that data writes don't
get overwritten and reads don't read uninitialised data.
To avoid overhead in the fast I/O path by adding new locking in the
protocol drivers and to restrict the impact to requests that actually
touch the new area, reuse the existing tracked request infrastructure in
block/io.c and mark all discard requests as serialising.
With this change, it is safe for protocol drivers to make
.bdrv_co_truncate actually asynchronous.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
This moves the bdrv_truncate() implementation from block.c to block/io.c
so it can have access to the tracked requests infrastructure.
This involves making refresh_total_sectors() public (in block_int.h).
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
bdrv_drain_all_*() used bdrv_next() to iterate over all root nodes and
did a subtree drain for each of them. This works fine as long as the
graph is static, but sadly, reality looks different.
If the graph changes so that root nodes are added or removed, we would
have to compensate for this. bdrv_next() returns each root node only
once even if it's the root node for multiple BlockBackends or for a
monitor-owned block driver tree, which would only complicate things.
The much easier and more obviously correct way is to fundamentally
change the way the functions work: Iterate over all BlockDriverStates,
no matter who owns them, and drain them individually. Compensation is
only necessary when a new BDS is created inside a drain_all section.
Removal of a BDS doesn't require any action because it's gone afterwards
anyway.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
In the future, bdrv_drained_all_begin/end() will drain all invidiual
nodes separately rather than whole subtrees. This means that we don't
want to propagate the drain to all parents any more: If the parent is a
BDS, it will already be drained separately. Recursing to all parents is
unnecessary work and would make it an O(n²) operation.
Prepare the drain function for the changed drain_all by adding an
ignore_bds_parents parameter to the internal implementation that
prevents the propagation of the drain to BDS parents. We still (have to)
propagate it to non-BDS parents like BlockBackends or Jobs because those
are not drained separately.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Before we can introduce a single polling loop for all nodes in
bdrv_drain_all_begin(), we must make sure to run it outside of coroutine
context like we already do for bdrv_do_drained_begin().
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
We cannot allow aio_poll() in bdrv_drain_invoke(begin=true) until we're
done with propagating the drain through the graph and are doing the
single final BDRV_POLL_WHILE().
Just schedule the coroutine with the callback and increase bs->in_flight
to make sure that the polling phase will wait for it.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
bdrv_do_drained_begin() is only safe if we have a single
BDRV_POLL_WHILE() after quiescing all affected nodes. We cannot allow
that parent callbacks introduce a nested polling loop that could cause
graph changes while we're traversing the graph.
Split off bdrv_do_drained_begin_quiesce(), which only quiesces a single
node without waiting for its requests to complete. These requests will
be waited for in the BDRV_POLL_WHILE() call down the call chain.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Anything can happen inside BDRV_POLL_WHILE(), including graph
changes that may interfere with its callers (e.g. child list iteration
in recursive callers of bdrv_do_drained_begin).
Switch to a single BDRV_POLL_WHILE() call for the whole subtree at the
end of bdrv_do_drained_begin() to avoid such effects. The recursion
happens now inside the loop condition. As the graph can only change
between bdrv_drain_poll() calls, but not inside of it, doing the
recursion here is safe.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
For bdrv_drain(), recursively waiting for child node requests is
pointless because we didn't quiesce their parents, so new requests could
come in anyway. Letting the function work only on a single node makes it
more consistent.
For subtree drains and drain_all, we already have the recursion in
bdrv_do_drained_begin(), so the extra recursion doesn't add anything
either.
Remove the useless code.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
We already requested that block jobs be paused in .bdrv_drained_begin,
but no guarantee was made that the job was actually inactive at the
point where bdrv_drained_begin() returned.
This introduces a new callback BdrvChildRole.bdrv_drained_poll() and
uses it to make bdrv_drain_poll() consider block jobs using the node to
be drained.
For the test case to work as expected, we have to switch from
block_job_sleep_ns() to qemu_co_sleep_ns() so that the test job is even
considered active and must be waited for when draining the node.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Commit 91af091f92 added an additional aio_poll() to BDRV_POLL_WHILE()
in order to make sure that all pending BHs are executed on drain. This
was the wrong place to make the fix, as it is useless overhead for all
other users of the macro and unnecessarily complicates the mechanism.
This patch effectively reverts said commit (the context has changed a
bit and the code has moved to AIO_WAIT_WHILE()) and instead polls in the
loop condition for drain.
The effect is probably hard to measure in any real-world use case
because actual I/O will dominate, but if I run only the initialisation
part of 'qemu-img convert' where it calls bdrv_block_status() for the
whole image to find out how much data there is copy, this phase actually
needs only roughly half the time after this patch.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
All involved nodes are already idle, we called bdrv_do_drain_begin() on
them.
The comment in the code suggested that this was not correct because the
completion of a request on one node could spawn a new request on a
different node (which might have been drained before, so we wouldn't
drain the new request). In reality, new requests to different nodes
aren't spawned out of nothing, but only in the context of a parent
request, and they aren't submitted to random nodes, but only to child
nodes. As long as we still poll for the completion of the parent request
(which we do), draining each root node separately is good enough.
Remove the additional polling code from bdrv_drain_all_begin() and
replace it with an assertion that all nodes are already idle after we
drained them separately.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Max Reitz
Andrey Shinkevich
Vladimir Sementsov-Ogievskiy
Alberto Garcia
Kevin Wolf
Fam Zheng
Eric Blake