Drivers can have internal request sources that generate IO, like the
need_check_timer in QED. Since we want quiesced periods that contain
nested event loops in block layer, we need to have a way to disable such
event sources.
Block drivers must implement the "bdrv_drain" callback if it has any
internal sources that can generate I/O activity, like a timer or a
worker thread (even in a library) that can schedule QEMUBH in an
asynchronous callback.
Update the comments of bdrv_drain and bdrv_drained_begin accordingly.
Like bdrv_requests_pending(), we should consider all the children of bs.
Before, the while loop just works, as bdrv_requests_pending() already
tracks its children; now we mustn't miss the callback, so recurse down
explicitly.
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Message-id: 1447064214-29930-9-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Currently all drivers that support .bdrv_aio_ioctl also implement
.bdrv_ioctl redundantly. To track ioctl requests in block layer it is
easier if we unify the two paths, because we'll need to run it in a
coroutine, as required by tracked_request_begin. While we're at it, use
.bdrv_aio_ioctl plus aio_poll() to emulate bdrv_ioctl().
Signed-off-by: Fam Zheng <famz@redhat.com>
Message-id: 1447064214-29930-7-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Both bdrv_discard and bdrv_aio_discard will call into bdrv_co_discard,
so add tracked_request_begin/end calls around the loop.
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1447064214-29930-4-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Both bdrv_flush and bdrv_aio_flush eventually call bdrv_co_flush, add
tracked_request_begin and tracked_request_end pair in that function so
that all flush requests are now tracked.
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1447064214-29930-3-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
We'll track more request types besides read and write, change the
boolean field to an enum.
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1447064214-29930-2-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The function manually recursed into bs->file and bs->backing to check
whether there were any requests pending, but it ignored other children.
There's no need to special case file and backing here, so just replace
these two explicit recursions by a loop recursing for all child nodes.
Reported-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Message-id: 1446029211-27148-1-git-send-email-kwolf@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The semantics is that after bdrv_drained_begin(bs), bs will not get new external
requests until the matching bdrv_drained_end(bs).
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
As the comment above bdrv_get_stats() says, BlockAcctStats is something
which belongs to the device instead of each BlockDriverState. This patch
therefore moves it into the BlockBackend.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
BlockAcctStats contains statistics about the data transferred from and
to the device; wr_highest_sector does not fit in with the rest.
Furthermore, those statistics are supposed to be specific for a certain
device and not necessarily for a BDS (see the comment above
bdrv_get_stats()); on the other hand, wr_highest_sector may be a rather
important information to know for each BDS. When BlockAcctStats is
finally removed from the BDS, we will want to keep wr_highest_sector in
the BDS.
Finally, wr_highest_sector is renamed to wr_highest_offset and given the
appropriate meaning. Externally, it is represented as an offset so there
is no point in doing something different internally. Its definition is
changed to match that in qapi/block-core.json which is "the offset after
the greatest byte written to". Doing so should not cause any harm since
if external programs tried to calculate the volume usage by
(wr_highest_offset + 512) / volume_size, after this patch they will just
assume the volume to be full slightly earlier than before.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
This is the final step in converting all of the BlockDriverState
pointers that block drivers use to BdrvChild.
After this patch, bs->children contains the full list of child nodes
that are referenced by a given BDS, and these children are only
referenced through BdrvChild, so that updating the pointer in there is
enough for changing edges in the graph.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
This patch removes the temporary duplication between bs->file and
bs->file_child by converting everything to BdrvChild.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Simplify memory allocation by sticking with a single API. GSlice
is not that fast anyway (tcmalloc/jemalloc are better).
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
In some cases, we need to disable copy-on-read, and just
read the data.
Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Message-id: 1441682913-14320-2-git-send-email-wency@cn.fujitsu.com
Signed-off-by: Jeff Cody <jcody@redhat.com>
The doc comments for bdrv_drain_all() and bdrv_drain() are outdated:
* The bdrv_drain() comment is a poor man's bdrv_lock()/bdrv_unlock()
which Fam Zheng is currently developing. Unfortunately this warning
was never really enough because devices keep submitting I/O and op
blockers don't prevent that.
* The bdrv_drain_all() comment is still partially correct but reflects
the nature of the implementation rather than API documentation.
Do make it clear that bdrv_drain() is only appropriate within an
AioContext. For anything spanning AioContexts you need
bdrv_drain_all().
Cc: Markus Armbruster <armbru@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Message-id: 1435854281-6078-1-git-send-email-stefanha@redhat.com
An empty GSList is represented by a NULL pointer, therefore it's a
perfectly valid argument for g_slist_find() and there's no need to
make any additional check.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Message-id: 1435583533-5758-1-git-send-email-berto@igalia.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Unsetting dirty globally with discard is not very correct. The discard may zero
out sectors (depending on can_write_zeroes_with_unmap), we should replicate
this change to destination side to make sure that the guest sees the same data.
Calling bdrv_reset_dirty also troubles mirror job because the hbitmap iterator
doesn't expect unsetting of bits after current position.
So let's do it the opposite way which fixes both problems: set the dirty bits
if we are to discard it.
Reported-by: wangxiaolong@ucloud.cn
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Like bdrv_is_allocated_above, this function follows the backing chain until seeing
BDRV_BLOCK_ALLOCATED. Base is not included.
Reimplement bdrv_is_allocated on top.
[Initialized bdrv_co_get_block_status_above() ret to 0 to silence
mingw64 compiler warning about the unitialized variable. assert(bs !=
base) prevents that case but I suppose the program could be compiled
with -DNDEBUG.
--Stefan]
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
During migration, QEMU uses fsync()/fdatasync() on the open file
descriptor for read-write block devices to flush data just before
stopping the VM.
However, fsync() on a scsi-generic device returns -EINVAL which
causes the migration to fail. This patch skips flushing data in case
of an SG device, since submitting SCSI commands directly via an SG
character device (e.g. /dev/sg0) bypasses the page cache completely,
anyway.
Note that fsync() not only flushes the page cache but also the disk
cache. The scsi-generic device never sends flushes, and for
migration it assumes that the same SCSI device is used by the
destination host, so it does not issue any SCSI SYNCHRONIZE CACHE
(10) command.
Finally, remove the bdrv_is_sg() test from iscsi_co_flush() since
this is now redundant (we flush the underlying protocol at the end
of bdrv_co_flush() which, with this patch, we never reach).
Signed-off-by: Dimitris Aragiorgis <dimara@arrikto.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1435056300-14924-3-git-send-email-dimara@arrikto.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
After the commit 9b536adc ("block: acquire AioContext in
bdrv_drain_all()") the aio_poll() function got called for every
BlockDriverState, in assumption that every device may have its own
AioContext. If we have thousands of disks attached, there are a lot of
BlockDriverStates but only a few AioContexts, leading to tons of
unnecessary aio_poll() calls.
This patch changes the bdrv_drain_all() function allowing it find shared
AioContexts and to call aio_poll() only for unique ones.
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Cornelia Huck <cornelia.huck@de.ibm.com>
Cc: Kevin Wolf <kwolf@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Alexander Yarygin <yarygin@linux.vnet.ibm.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Tested-by: Christian Borntraeger <borntraeger@de.ibm.com>
Message-id: 1433936297-7098-4-git-send-email-yarygin@linux.vnet.ibm.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The throttle group support use a cooperative round robin scheduling
algorithm.
The principles of the algorithm are simple:
- Each BDS of the group is used as a token in a circular way.
- The active BDS computes if a wait must be done and arms the right
timer.
- If a wait must be done the token timer will be armed so the token
will become the next active BDS.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: f0082a86f3ac01c46170f7eafe2101a92e8fde39.1433779731.git.berto@igalia.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Group throttling will share ThrottleState between multiple bs.
As a consequence the ThrottleState will be accessed by multiple aio
context.
Timers are tied to their aio context so they must go out of the
ThrottleState structure.
This commit paves the way for each bs of a common ThrottleState to
have its own timer.
Signed-off-by: Benoit Canet <benoit.canet@nodalink.com>
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 6cf9ea96d8b32ae2f8769cead38f68a6a0c8c909.1433779731.git.berto@igalia.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
A bit of Boolean algebra (and common sense) tells us that the
second "if" here is looking for blocks that are not allocated.
This is the opposite of the "if" that sets BDRV_BLOCK_ALLOCATED,
and thus it can use an "else".
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Message-id: 1431599702-10431-1-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
For zero write, callers pass in NULL qiov (qemu-io "write -z" or
scsi-disk "write same").
Commit fc3959e466 fixed bdrv_co_write_zeroes which is the common case
for this bug, but it still exists in bdrv_aio_write_zeroes. A simpler
fix would be in bdrv_co_do_pwritev which is the NULL dereference point
and covers both cases.
So don't access it in bdrv_co_do_pwritev in this case, use three aligned
writes.
[Initialize ret to 0 in bdrv_co_do_zero_pwritev() to avoid uninitialized
variable warning with gcc 4.9.2.
--Stefan]
Signed-off-by: Fam Zheng <famz@redhat.com>
Message-id: 1431522721-3266-3-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This reverts commit fc3959e466.
The core write code already handles the case, so remove this
duplication.
Because commit 61007b316 moved the touched code from block.c to
block/io.c, the change is manually reverted.
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1431522721-3266-2-git-send-email-famz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The following sequence
int fd = open(argv[1], O_RDWR | O_CREAT | O_DIRECT, 0644);
for (i = 0; i < 100000; i++)
write(fd, buf, 4096);
performs 5% better if buf is aligned to 4096 bytes.
The difference is quite reliable.
On the other hand we do not want at the moment to enforce bounce
buffering if guest request is aligned to 512 bytes.
The patch changes default bounce buffer optimal alignment to
MAX(page size, 4k). 4k is chosen as maximal known sector size on real
HDD.
The justification of the performance improve is quite interesting.
From the kernel point of view each request to the disk was split
by two. This could be seen by blktrace like this:
9,0 11 1 0.000000000 11151 Q WS 312737792 + 1023 [qemu-img]
9,0 11 2 0.000007938 11151 Q WS 312738815 + 8 [qemu-img]
9,0 11 3 0.000030735 11151 Q WS 312738823 + 1016 [qemu-img]
9,0 11 4 0.000032482 11151 Q WS 312739839 + 8 [qemu-img]
9,0 11 5 0.000041379 11151 Q WS 312739847 + 1016 [qemu-img]
9,0 11 6 0.000042818 11151 Q WS 312740863 + 8 [qemu-img]
9,0 11 7 0.000051236 11151 Q WS 312740871 + 1017 [qemu-img]
9,0 5 1 0.169071519 11151 Q WS 312741888 + 1023 [qemu-img]
After the patch the pattern becomes normal:
9,0 6 1 0.000000000 12422 Q WS 314834944 + 1024 [qemu-img]
9,0 6 2 0.000038527 12422 Q WS 314835968 + 1024 [qemu-img]
9,0 6 3 0.000072849 12422 Q WS 314836992 + 1024 [qemu-img]
9,0 6 4 0.000106276 12422 Q WS 314838016 + 1024 [qemu-img]
and the amount of requests sent to disk (could be calculated counting
number of lines in the output of blktrace) is reduced about 2 times.
Both qemu-img and qemu-io are affected while qemu-kvm is not. The guest
does his job well and real requests comes properly aligned (to page).
Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1431441056-26198-3-git-send-email-den@openvz.org
CC: Paolo Bonzini <pbonzini@redhat.com>
CC: Kevin Wolf <kwolf@redhat.com>
CC: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The patch introduces new concept: minimal memory alignment for bounce
buffers. Original so called "optimal" value is actually minimal required
value for aligment. It should be used for validation that the IOVec
is properly aligned and bounce buffer is not required.
Though, from the performance point of view, it would be better if
bounce buffer or IOVec allocated by QEMU will be aligned stricter.
The patch does not change any alignment value yet.
Signed-off-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1431441056-26198-2-git-send-email-den@openvz.org
CC: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
CC: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This is the behavior in the operating system, for example Linux's
blkdev_write_iter has the following:
if (bdev_read_only(I_BDEV(bd_inode)))
return -EPERM;
This does not apply to opening a device for read/write, when the
device only supports read-only operation. In this case any of
EACCES, EPERM or EROFS is acceptable depending on why writing is
not possible.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-id: 1431013548-22492-1-git-send-email-pbonzini@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The block.c file has grown to over 6000 lines. It is time to split this
file so there are fewer conflicts and the code is easier to maintain.
Extract I/O request processing code:
* Read
* Write
* Zero writes and making the image empty
* Flush
* Discard
* ioctl
* Tracked requests and queuing
* Throttling and copy-on-read
* Block status and allocated functions
* Refreshing block limits
* Reading/writing vmstate
* qemu_blockalign() and friends
The patch simply moves code from block.c into block/io.c.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
