Re: [Qemu-devel] [PATCH 08/27] block/parallels: _co_writev callback for Parallels format

["Denis V. Lunev" <den@openvz.org>]

On 23/04/15 12:20, Stefan Hajnoczi wrote:
> On Wed, Apr 22, 2015 at 04:16:38PM +0300, Denis V. Lunev wrote:
>> On 22/04/15 16:08, Stefan Hajnoczi wrote:
>>> On Wed, Mar 11, 2015 at 01:28:02PM +0300, Denis V. Lunev wrote:
>>>> +static int64_t allocate_cluster(BlockDriverState *bs, int64_t sector_num)
>>>> +{
>>>> +    BDRVParallelsState *s = bs->opaque;
>>>> +    uint32_t idx, offset, tmp;
>>>> +    int64_t pos;
>>>> +    int ret;
>>>> +
>>>> +    idx = sector_num / s->tracks;
>>>> +    offset = sector_num % s->tracks;
>>>> +
>>>> +    if (idx >= s->catalog_size) {
>>>> +        return -EINVAL;
>>>> +    }
>>>> +    if (s->catalog_bitmap[idx] != 0) {
>>>> +        return (uint64_t)s->catalog_bitmap[idx] * s->off_multiplier + offset;
>>>> +    }
>>>> +
>>>> +    pos = bdrv_getlength(bs->file) >> BDRV_SECTOR_BITS;
>>>> +    bdrv_truncate(bs->file, (pos + s->tracks) << BDRV_SECTOR_BITS);
>>>> +    s->catalog_bitmap[idx] = pos / s->off_multiplier;
>>>> +
>>>> +    tmp = cpu_to_le32(s->catalog_bitmap[idx]);
>>>> +
>>>> +    ret = bdrv_pwrite_sync(bs->file,
>>>> +            sizeof(ParallelsHeader) + idx * sizeof(tmp), &tmp, sizeof(tmp));
>>> What is the purpose of the sync?
>> This is necessary to preserve image consistency on crash from
>> my point of view. There is no check consistency at the moment.
>> The sync will be removed later when proper crash detection
>> code will be added (patches 19, 20, 21)
> Let's look at possible orderings in case of failure:
>
> A. BAT update
> B. Data write
>
> This sync enforces A, B ordering.  If we can see B, then A must also
> have happened thanks to the sync.
>
> But A, B ordering is too conservative.  Imagine B, A ordering and the
> failure where we crash before A.  It means we wrote the data but never
> linked it into the BAT.
>
> What happens in that case?  We've leaked a cluster in the underlying
> image file but it doesn't corrupt the visible disk from the guest
> point-of-view.
>
> Because your implementation uses truncate to extend the file size before
> A, even the A, B failure case results in a leaked cluster.  So the B, A
> case is not worse in any way.
>
> Why do other image formats sync cluster allocation updates?  Because
> they support backing files and in that case an A, B ordering results in
> data corruption so they enforce B, A ordering (the opposite of what
> you're trying to do!).
>
> The reason why A, B ordering results in data corruption when backing
> files are in use is because the guest's write request might touch only a
> subset of the cluster (a couple of sectors out of the whole cluster).
> So the guest needs to copy the remaining sectors from the backing file.
> If there is a dangling BAT entry like in the A, B failure case, then the
> guest will see a zeroed cluster instead of the contents of the backing
> file.  This is a data corruption, but only if a backing file is being
> used!
>
> So the sync is not necessary, both A, B and B, A ordering work for
> block/parallels.c.
>
> Stefan
this is very interesting speculation. I have never considered the
problem from this point of view. Thank you very much for this!