Re: [PATCH 3/4] qcow2: Fix corruption on discard during write with COW

[Kevin Wolf <kwolf@redhat.com>]

Am 29.04.2026 um 17:28 hat Fiona Ebner geschrieben:
> Am 27.04.26 um 7:04 PM schrieb Kevin Wolf:
> > @@ -2000,6 +2028,15 @@ int qcow2_cluster_discard(BlockDriverState *bs, uint64_t offset,
> >      int64_t cleared;
> >      int ret;
> >  
> > +    /*
> > +     * If we're touching a cluster for which allocating writes are in flight,
> > +     * wait for them to complete to avoid conflicting metadata updates.
> > +     *
> > +     * We don't need to allocate a QCowL2Meta for the discard operation because
> > +     * s->lock is held for the duration of the whole operation.
> 
> The caller in qcow2_make_empty() does not acquire s->lock AFAICS. Can
> that be a problem?

This is the case the comment inside wait_for_dependencies() is about:

    /*
     * Discard has some non-coroutine callers (creating internal snapshots and
     * make empty). They are calling from qemu-img or in a drained section, so
     * we know that no writes can be in progress.
     */

So it should be fine in practice because we know a bigger hammer is
already in use for synchronisation.

Kevin

> > +     */
> > +    wait_for_dependencies(bs, offset, bytes);
> > +
> >      /* Caller must pass aligned values, except at image end */
> >      assert(QEMU_IS_ALIGNED(offset, s->cluster_size));
> >      assert(QEMU_IS_ALIGNED(end_offset, s->cluster_size) ||