Re: [Qemu-devel] [PATCH] migration: Appease coverity, skip empty block trees

[John Snow <jsnow@redhat.com>]

On 06/29/2018 01:07 PM, Stefan Hajnoczi wrote:
> On Fri, Jun 22, 2018 at 04:11:22PM -0400, John Snow wrote:
>> If a tree consists exclusively of implicit filter nodes, we might crash
>> QEMU. This configuration should not exist in practice, but if it did,
>> skipping it would be fine.
>>
>> For the purposes of debug builds, throw an assert to remind us that
>> this configuration is truly unexpected, but if it's compiled out we
>> will cope just fine.
>>
>> Signed-off-by: John Snow <jsnow@redhat.com>
>> ---
>>  migration/block-dirty-bitmap.c | 4 ++++
>>  1 file changed, 4 insertions(+)
>>
>> diff --git a/migration/block-dirty-bitmap.c b/migration/block-dirty-bitmap.c
>> index 3bafbbdc4c..02725293dd 100644
>> --- a/migration/block-dirty-bitmap.c
>> +++ b/migration/block-dirty-bitmap.c
>> @@ -287,6 +287,10 @@ static int init_dirty_bitmap_migration(void)
>>          while (bs && bs->drv && bs->implicit) {
>>              bs = backing_bs(bs);
>>          }
>> +        if (!bs) {
>> +            g_assert_not_reached();
>> +            continue;
>> +        }
> 
> If bs can never be NULL, why test that it is non-NULL in the while loop
> condition?
> 
> Try:
> 
>   /* Precondition: bs != NULL thanks to the for loop */
>   while (bs->drv && bs->implicit) {
>       bs = backing_bs(bs);
>   }
>   /* Postcondition: bs != NULL due to implicit node layout assumption */
> 
> Does this silence Coverity?  ISTR it looks for cues like the bs check in
> the while loop condition to decide whether it's likely that a variable
> could be NULL.
> 

I'll give this a go, but mechanically it looks suspect without an
assert(bs) in the loop body, but that would definitely silence Coverity.

--js