[Qemu-devel] [PATCH] sheepdog: fix dynamic grow for running qcow2 format

[Qemu-devel] [PATCH] sheepdog: fix dynamic grow for running qcow2 format

[Liu Yuan]

When running qcow2 over sheepdog, we might meet following problem

  qemu-system-x86_64: shrinking is not supported

And cause IO errors to Guest. This is because we abuse bs->total_sectors, which
is manipulated by generic block layer and race with sheepdog code.

We should directly check if offset > vdi_size to dynamically enlarge the volume
instead of 'offset > bs->total_sectors', which will cause problem when following
case happens:

   vdi_size > offset > bs->total_sectors

   # then trigger sd_truncate() to shrink the volume wrongly.

Cc: qemu-devel@nongnu.org
Cc: Kevin Wolf <kwolf@redhat.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Reported-by: Hadrien KOHL <hadrien.kohl@gmail.com>
Signed-off-by: Liu Yuan <namei.unix@gmail.com>
---
 block/sheepdog.c |    7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/block/sheepdog.c b/block/sheepdog.c
index d1c812d..ba451a9 100644
--- a/block/sheepdog.c
+++ b/block/sheepdog.c
@@ -2048,13 +2048,14 @@ static coroutine_fn int sd_co_writev(BlockDriverState *bs, int64_t sector_num,
 {
     SheepdogAIOCB *acb;
     int ret;
+    int64_t offset = (sector_num + nb_sectors) * BDRV_SECTOR_SIZE;
+    BDRVSheepdogState *s = bs->opaque;
 
-    if (bs->growable && sector_num + nb_sectors > bs->total_sectors) {
-        ret = sd_truncate(bs, (sector_num + nb_sectors) * BDRV_SECTOR_SIZE);
+    if (bs->growable && offset > s->inode.vdi_size) {
+        ret = sd_truncate(bs, offset);
         if (ret < 0) {
             return ret;
         }
-        bs->total_sectors = sector_num + nb_sectors;
     }
 
     acb = sd_aio_setup(bs, qiov, sector_num, nb_sectors);
-- 
1.7.9.5

Re: [Qemu-devel] [PATCH] sheepdog: fix dynamic grow for running qcow2 format

[Stefan Hajnoczi]

On Sat, Dec 14, 2013 at 01:29:28AM +0800, Liu Yuan wrote:
> When running qcow2 over sheepdog, we might meet following problem
> 
>   qemu-system-x86_64: shrinking is not supported
> 
> And cause IO errors to Guest. This is because we abuse bs->total_sectors, which
> is manipulated by generic block layer and race with sheepdog code.
> 
> We should directly check if offset > vdi_size to dynamically enlarge the volume
> instead of 'offset > bs->total_sectors', which will cause problem when following
> case happens:
> 
>    vdi_size > offset > bs->total_sectors
> 
>    # then trigger sd_truncate() to shrink the volume wrongly.
> 
> Cc: qemu-devel@nongnu.org
> Cc: Kevin Wolf <kwolf@redhat.com>
> Cc: Stefan Hajnoczi <stefanha@redhat.com>
> Reported-by: Hadrien KOHL <hadrien.kohl@gmail.com>
> Signed-off-by: Liu Yuan <namei.unix@gmail.com>
> ---
>  block/sheepdog.c |    7 ++++---
>  1 file changed, 4 insertions(+), 3 deletions(-)

block.c already updates bs->total_sectors if bs->growable so this patch
is safe.

Thanks, applied to my block tree:
https://github.com/stefanha/qemu/commits/block

Stefan