Patchwork block migraton: check sectors before shift operation.

login
register
mail settings
Submitter Yoshiaki Tamura
Date July 19, 2010, 4:45 a.m.
Message ID <1279514742-6941-1-git-send-email-tamura.yoshiaki@lab.ntt.co.jp>
Download mbox | patch
Permalink /patch/59179/
State New
Headers show

Comments

Yoshiaki Tamura - July 19, 2010, 4:45 a.m.
Commit d246673dcb9911218ff555bcdf28b250e38fa46c has expanded the types
of block drive that can be initialized for block migration.  Although
bdrv_getlength() may return < 0, current code shifts it without
checking.  This makes block migration initialization invalid and
results in abort() due to calling qemu_malloc() with 0 size at
bdrv_set_dirty_tracking().  This patch checks the return value of
bdrv_getlength() by masking with BDRV_SECTOR_MASK.

Signed-off-by: Yoshiaki Tamura <tamura.yoshiaki@lab.ntt.co.jp>
---
 block-migration.c |    5 +++--
 1 files changed, 3 insertions(+), 2 deletions(-)
Kevin Wolf - July 19, 2010, 9:23 a.m.
Am 19.07.2010 06:45, schrieb Yoshiaki Tamura:
> Commit d246673dcb9911218ff555bcdf28b250e38fa46c has expanded the types
> of block drive that can be initialized for block migration.  Although
> bdrv_getlength() may return < 0, current code shifts it without
> checking.  This makes block migration initialization invalid and
> results in abort() due to calling qemu_malloc() with 0 size at
> bdrv_set_dirty_tracking().  This patch checks the return value of
> bdrv_getlength() by masking with BDRV_SECTOR_MASK.
> 
> Signed-off-by: Yoshiaki Tamura <tamura.yoshiaki@lab.ntt.co.jp>

I applied a similar patch by Shahar Havivi to the block branch a few
days ago.

Kevin
Yoshiaki Tamura - July 19, 2010, 11:22 a.m.
2010/7/19 Kevin Wolf <kwolf@redhat.com>:
> Am 19.07.2010 06:45, schrieb Yoshiaki Tamura:
>> Commit d246673dcb9911218ff555bcdf28b250e38fa46c has expanded the types
>> of block drive that can be initialized for block migration.  Although
>> bdrv_getlength() may return < 0, current code shifts it without
>> checking.  This makes block migration initialization invalid and
>> results in abort() due to calling qemu_malloc() with 0 size at
>> bdrv_set_dirty_tracking().  This patch checks the return value of
>> bdrv_getlength() by masking with BDRV_SECTOR_MASK.
>>
>> Signed-off-by: Yoshiaki Tamura <tamura.yoshiaki@lab.ntt.co.jp>
>
> I applied a similar patch by Shahar Havivi to the block branch a few
> days ago.

Oops.  Missed that discussion.

Yoshi

>
> Kevin
>
>

Patch

diff --git a/block-migration.c b/block-migration.c
index 7db6f02..2e02a4a 100644
--- a/block-migration.c
+++ b/block-migration.c
@@ -237,10 +237,11 @@  static void init_blk_migration_it(void *opaque, BlockDriverState *bs)
     int64_t sectors;
 
     if (!bdrv_is_read_only(bs)) {
-        sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
-        if (sectors == 0) {
+        sectors = bdrv_getlength(bs) & BDRV_SECTOR_MASK;
+        if (sectors <= 0) {
             return;
         }
+        sectors >>= BDRV_SECTOR_BITS;
 
         bmds = qemu_mallocz(sizeof(BlkMigDevState));
         bmds->bs = bs;