Especially the combination of iothreads, block jobs and drain tends to
lead to hangs currently. This series fixes a few of these bugs, although
there are more of them, to be addressed in separate patches.
The primary goal of this series is to fix the scenario from:
https://bugzilla.redhat.com/show_bug.cgi?id=1601212
A simplified reproducer of the reported problem looks like this (two concurrent
commit block jobs for disks in an iothread):
$qemu -qmp stdio \
-object iothread,id=iothread1 \
-device virtio-scsi-pci,id=virtio_scsi_pci0,bus=pci.0,addr=0x6,iothread=iothread1 \
-drive id=drive_image1,if=none,snapshot=off,aio=threads,cache=none,format=qcow2,file=hd0 \
-device scsi-hd,drive=drive_image1,id=image1,bootindex=1 \
-drive id=drive_image2,if=none,snapshot=off,aio=threads,cache=none,format=qcow2,file=hd1 \
-device scsi-hd,drive=drive_image2,id=image2,bootindex=2
{"execute":"qmp_capabilities"}
{"execute":"blockdev-snapshot-sync","arguments":{"device":"drive_image1","snapshot-file":"sn1"}}
{"execute":"blockdev-snapshot-sync","arguments":{"device":"drive_image1","snapshot-file":"sn11"}}
{"execute":"blockdev-snapshot-sync","arguments":{"device":"drive_image1","snapshot-file":"sn111"}}
{"execute":"blockdev-snapshot-sync","arguments":{"device":"drive_image2","snapshot-file":"sn2"}}
{"execute":"blockdev-snapshot-sync","arguments":{"device":"drive_image2","snapshot-file":"sn22"}}
{"execute":"blockdev-snapshot-sync","arguments":{"device":"drive_image2","snapshot-file":"sn222"}}
{ "execute": "block-commit", "arguments": { "device": "drive_image2","base":"sn2","backing-file":"sn2","top":"sn22"}}
{ "execute": "block-commit", "arguments": { "device": "drive_image1","base":"sn1","backing-file":"sn1","top":"sn11"}}
{"execute":"quit"}
Kevin Wolf (14):
blockjob: Wake up BDS when job becomes idle
test-bdrv-drain: Drain with block jobs in an I/O thread
test-blockjob: Acquire AioContext around job_finish_sync()
job: Use AIO_WAIT_WHILE() in job_finish_sync()
test-bdrv-drain: Test AIO_WAIT_WHILE() in completion callback
block: Add missing locking in bdrv_co_drain_bh_cb()
aio-wait: Increase num_waiters even in home thread
block-backend: Add .drained_poll callback
block-backend: Fix potential double blk_delete()
block-backend: Decrease in_flight only after callback
mirror: Fix potential use-after-free in active commit
blockjob: Lie better in child_job_drained_poll()
block: Remove aio_poll() in bdrv_drain_poll variants
test-bdrv-drain: Test nested poll in bdrv_drain_poll_top_level()
include/block/aio-wait.h | 2 +
include/block/blockjob.h | 13 ++++++
include/qemu/coroutine.h | 5 +++
include/qemu/job.h | 12 ++++++
block/block-backend.c | 26 ++++++++++-
block/io.c | 23 ++++++----
block/mirror.c | 9 ++++
blockjob.c | 20 ++++++++-
job.c | 29 +++++++++----
tests/test-bdrv-drain.c | 110 ++++++++++++++++++++++++++++++++++++++++++++---
tests/test-blockjob.c | 6 +++
util/qemu-coroutine.c | 5 +++
12 files changed, 235 insertions(+), 25 deletions(-)
--
2.13.6