]> git.proxmox.com Git - qemu.git/commit
qcow2: Bring synchronous read/write back to life stable-0.11 origin/stable-0.11
authorKevin Wolf <kwolf@redhat.com>
Thu, 3 Dec 2009 09:28:44 +0000 (10:28 +0100)
committerAurelien Jarno <aurelien@aurel32.net>
Thu, 3 Dec 2009 10:29:25 +0000 (11:29 +0100)
commit08fd2f30bd3ee5d04596da8293689af4d4f7eb6c
treeeb72c9f023390352184a981fb23df34234e2a6e5
parent35bfc7324e2e6946c4113ada5db30553a1a7c40b
qcow2: Bring synchronous read/write back to life

When the synchronous read and write functions were dropped, they were replaced
by generic emulation functions. Unfortunately, these emulation functions don't
provide the same semantics as the original functions did.

The original bdrv_read would mean that we read some data synchronously and that
we won't be interrupted during this read. The latter assumption is no longer
true with the emulation function which needs to use qemu_aio_poll and therefore
allows the callback of any other concurrent AIO request to be run during the
read. Which in turn means that (meta)data read earlier could have changed and
be invalid now. qcow2 is not prepared to work in this way and it's just scary
how many places there are where other requests could run.

I'm not sure yet where exactly it breaks, but you'll see breakage with virtio
on qcow2 with a backing file. Providing synchronous functions again fixes the
problem for me.

Patchworks-ID: 35437
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>
(cherry picked from commit ef845c3bf421290153154635dc18eaa677cecb43)

Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
block/qcow2-cluster.c
block/qcow2.c
block/qcow2.h