Re: [Qemu-devel] [PATCH 02/18] block: BDS deletion during bdrv_drain_recurse

[Max Reitz]

On 2017-10-10 10:36, Kevin Wolf wrote:
> Am 13.09.2017 um 20:18 hat Max Reitz geschrieben:
>> Drainined a BDS child may lead to both the original BDS and/or its other
>> children being deleted (e.g. if the original BDS represents a block
>> job).  We should prepare for this in both bdrv_drain_recurse() and
>> bdrv_drained_begin() by monitoring whether the BDS we are about to drain
>> still exists at all.
>>
>> Signed-off-by: Max Reitz <address@hidden>
> 
> How hard would it be to write a test case for this? qemu-iotests
> probably isn't the right tool, but I feel a C unit test would be
> possible.

I can look into it, but I can't promise anything.

>> -    QLIST_FOREACH_SAFE(child, &bs->children, next, tmp) {
>> -        BlockDriverState *bs = child->bs;
>> -        bool in_main_loop =
>> -            qemu_get_current_aio_context() == qemu_get_aio_context();
>> -        assert(bs->refcnt > 0);
> 
> Would it make sense to keep this assertion for the !deleted case?

Sure, why not.

>> -        if (in_main_loop) {
>> -            /* In case the recursive bdrv_drain_recurse processes a
>> -             * block_job_defer_to_main_loop BH and modifies the graph,
>> -             * let's hold a reference to bs until we are done.
>> -             *
>> -             * IOThread doesn't have such a BH, and it is not safe to call
>> -             * bdrv_unref without BQL, so skip doing it there.
>> -             */
>> -            bdrv_ref(bs);
>> -        }
>> -        waited |= bdrv_drain_recurse(bs);
>> -        if (in_main_loop) {
>> -            bdrv_unref(bs);
>> +    /* Draining children may result in other children being removed and 
>> maybe
>> +     * even deleted, so copy the children list first */
> 
> Maybe it's just me, but I failed to understand this correctly at first.
> How about "being removed from their parent" to clarify that it's not the
> BDS that is removed, but just the reference?

Well, it's the BdrvChild that's removed, that's what I meant by
"children".  But then the comment speaks of "children list" and means
creation of a list of BDSs, sooo...  Yes, some change necessary.

Max

signature.asc
Description: OpenPGP digital signature