Mirror jobs for drives with iothreads make QEMU to abort with "block.c:1895: bdrv_attach_child: Assertion `bdrv_get_aio_context(parent_bs) == bdrv_get_aio_context(child_bs)' failed."
Description of problem:
* Preparing the images
# qemu-img create -f raw base.img 1g
# qemu-img create -f qcow2 -o backing_file=base.img top1.qcow2
# qemu-img create -f qcow2 -o backing_file=base.img top2.qcow2
* Running with iothreads
# /usr/libexec/qemu-kvm -m 1g -drive if=none,file=top1.qcow2,format=qcow2,id=drive0 -object iothread,id=iothread0 -device virtio-blk-pci,iothread=iothread0,drive=drive0 -monitor stdio
QEMU 2.9.0 monitor - type 'help' for more information
(qemu) drive_mirror -n drive0 top2.qcow2
(qemu) info block-jobs
Type mirror, device drive0: Completed 0 of 0 bytes, speed limit 0 bytes/s
(qemu) block_job_complete drive0
qemu-kvm: block.c:1895: bdrv_attach_child: Assertion `bdrv_get_aio_context(parent_bs) == bdrv_get_aio_context(child_bs)' failed.
Aborted (core dumped)
* Running without iothreads
# /usr/libexec/qemu-kvm -m 1g -drive if=none,file=top1.qcow2,format=qcow2,id=drive0 -device virtio-blk-pci,drive=drive0 -monitor stdio
QEMU 2.9.0 monitor - type 'help' for more information
(qemu) drive_mirror -n drive0 top2.qcow2
(qemu) info block-jobs
Type mirror, device drive0: Completed 0 of 0 bytes, speed limit 0 bytes/s
(qemu) block_job_complete drive0
(qemu) info block-jobs
No active jobs
Version-Release number of selected component (if applicable):
qemu-kvm-rhev-2.9.0-16.el7_4.13.x86_64
How reproducible:
Always
Additional info:
This has been fixed in upstream by this oneliner:
<snip>
[slopezpa@dritchie qemu]$ git show 5ce6bfe255091d532ec4555d0ede816294cb1703
commit 5ce6bfe255091d532ec4555d0ede816294cb1703
Author: sochin.jiang <sochin.jiang>
Date: Mon Jun 26 19:04:24 2017 +0800
mirror: Fix inconsistent backing AioContext for after mirroring
mirror_complete opens the backing chain, which should have the same
AioContext as the top when using iothreads. Make the code guarantee
this, which fixes a failed assertion in bdrv_attach_child.
Signed-off-by: sochin.jiang <sochin.jiang>
Message-id: 1498475064-39816-1-git-send-email-sochin.jiang
[mreitz: Reworded commit message]
Signed-off-by: Max Reitz <mreitz>
diff --git a/block.c b/block.c
index edfa6b7a2c..b9e5114a27 100644
--- a/block.c
+++ b/block.c
@@ -2185,6 +2185,7 @@ int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options,
ret = -EINVAL;
goto free_exit;
}
+ bdrv_set_aio_context(backing_hd, bdrv_get_aio_context(bs));
/* Hook up the backing file link; drop our reference, bs owns the
* backing_hd reference now */
</snip>
Please consider backporting this. Thanks.
Reproduced on qemu-kvm-rhev-10:2.9.0-14.el7.x86_64 with steps in comment 0.
If mirror source file is base, wont trigger the core dump, the source file must be a snapshot for this issue.
Same steps work well on qemu-kvm-rhev-2.10.0-16.el7.x86_64, so the latest version qemu should have already fixed this issue.
(In reply to Qianqian Zhu from comment #4)
> Reproduced on qemu-kvm-rhev-10:2.9.0-14.el7.x86_64 with steps in comment 0.
>
> If mirror source file is base, wont trigger the core dump, the source file
> must be a snapshot for this issue.
>
> Same steps work well on qemu-kvm-rhev-2.10.0-16.el7.x86_64, so the latest
> version qemu should have already fixed this issue.
Indeed, I confirmed the upstream patch pointed in Comment #0 is included in qemu-2.10, changing BZ to MODIFIED.
Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.
For information on the advisory, and where to find the updated
files, follow the link below.
If the solution does not work for you, open a new bug report.
https://access.redhat.com/errata/RHSA-2018:1104
Description of problem: * Preparing the images # qemu-img create -f raw base.img 1g # qemu-img create -f qcow2 -o backing_file=base.img top1.qcow2 # qemu-img create -f qcow2 -o backing_file=base.img top2.qcow2 * Running with iothreads # /usr/libexec/qemu-kvm -m 1g -drive if=none,file=top1.qcow2,format=qcow2,id=drive0 -object iothread,id=iothread0 -device virtio-blk-pci,iothread=iothread0,drive=drive0 -monitor stdio QEMU 2.9.0 monitor - type 'help' for more information (qemu) drive_mirror -n drive0 top2.qcow2 (qemu) info block-jobs Type mirror, device drive0: Completed 0 of 0 bytes, speed limit 0 bytes/s (qemu) block_job_complete drive0 qemu-kvm: block.c:1895: bdrv_attach_child: Assertion `bdrv_get_aio_context(parent_bs) == bdrv_get_aio_context(child_bs)' failed. Aborted (core dumped) * Running without iothreads # /usr/libexec/qemu-kvm -m 1g -drive if=none,file=top1.qcow2,format=qcow2,id=drive0 -device virtio-blk-pci,drive=drive0 -monitor stdio QEMU 2.9.0 monitor - type 'help' for more information (qemu) drive_mirror -n drive0 top2.qcow2 (qemu) info block-jobs Type mirror, device drive0: Completed 0 of 0 bytes, speed limit 0 bytes/s (qemu) block_job_complete drive0 (qemu) info block-jobs No active jobs Version-Release number of selected component (if applicable): qemu-kvm-rhev-2.9.0-16.el7_4.13.x86_64 How reproducible: Always Additional info: This has been fixed in upstream by this oneliner: <snip> [slopezpa@dritchie qemu]$ git show 5ce6bfe255091d532ec4555d0ede816294cb1703 commit 5ce6bfe255091d532ec4555d0ede816294cb1703 Author: sochin.jiang <sochin.jiang> Date: Mon Jun 26 19:04:24 2017 +0800 mirror: Fix inconsistent backing AioContext for after mirroring mirror_complete opens the backing chain, which should have the same AioContext as the top when using iothreads. Make the code guarantee this, which fixes a failed assertion in bdrv_attach_child. Signed-off-by: sochin.jiang <sochin.jiang> Message-id: 1498475064-39816-1-git-send-email-sochin.jiang [mreitz: Reworded commit message] Signed-off-by: Max Reitz <mreitz> diff --git a/block.c b/block.c index edfa6b7a2c..b9e5114a27 100644 --- a/block.c +++ b/block.c @@ -2185,6 +2185,7 @@ int bdrv_open_backing_file(BlockDriverState *bs, QDict *parent_options, ret = -EINVAL; goto free_exit; } + bdrv_set_aio_context(backing_hd, bdrv_get_aio_context(bs)); /* Hook up the backing file link; drop our reference, bs owns the * backing_hd reference now */ </snip> Please consider backporting this. Thanks.