Re: [PATCH v2 03/36] block: bdrv_append(): don't consume reference

[Vladimir Sementsov-Ogievskiy]

18.01.2021 17:18, Kevin Wolf wrote:
> Am 27.11.2020 um 15:44 hat Vladimir Sementsov-Ogievskiy geschrieben:
> > We have too much comments for this feature. It seems better just don't
> > do it. Most of real users (tests don't count) have to create additional
> > reference.
> >
> > Drop also comment in external_snapshot_prepare:
> >   - bdrv_append doesn't "remove" old bs in common sense, it sounds
> >     strange
> >   - the fact that bdrv_append can fail is obvious from the context
> >   - the fact that we must rollback all changes in transaction abort is
> >     known (it's the direct role of abort)
> >
> > Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
> > ---
> >   block.c                     | 19 +++----------------
> >   block/backup-top.c          |  1 -
> >   block/commit.c              |  1 +
> >   block/mirror.c              |  3 ---
> >   blockdev.c                  |  4 ----
> >   tests/test-bdrv-drain.c     |  2 +-
> >   tests/test-bdrv-graph-mod.c |  2 ++
> >   7 files changed, 7 insertions(+), 25 deletions(-)
> >
> > diff --git a/block.c b/block.c
> > index 0dd28f0902..55efef3c9d 100644
> > --- a/block.c
> > +++ b/block.c
> > @@ -3145,11 +3145,6 @@ static BlockDriverState 
> > *bdrv_append_temp_snapshot(BlockDriverState *bs,
> >           goto out;
> >       }
> >   
> > -    /* bdrv_append() consumes a strong reference to bs_snapshot
> > -     * (i.e. it will call bdrv_unref() on it) even on error, so in
> > -     * order to be able to return one, we have to increase
> > -     * bs_snapshot's refcount here */
> > -    bdrv_ref(bs_snapshot);
> >       bdrv_append(bs_snapshot, bs, &local_err);
> >       if (local_err) {
> >           error_propagate(errp, local_err);
> > @@ -4608,10 +4603,8 @@ void bdrv_replace_node(BlockDriverState *from, 
> > BlockDriverState *to,
> >    *
> >    * This function does not create any image files.
> >    *
> > - * bdrv_append() takes ownership of a bs_new reference and unrefs it because
> > - * that's what the callers commonly need. bs_new will be referenced by the old
> > - * parents of bs_top after bdrv_append() returns. If the caller needs to keep a
> > - * reference of its own, it must call bdrv_ref().
> > + * Recent update: bdrv_append does NOT eat bs_new reference for now. Drop this
> > + * comment several moths later.
>
> A comment like this is unusual. Do you think there is a high risk of
> somebody introducing a new bdrv_append() in parallel and that they would
> read this comment when rebasing their existing patches?

Or even later, someone may remember that bdrv_append() eat the reference, and 
then face some strange behavior with a new call of bdrv_append(), finally go to 
check the function code and see the new comment.. I don't insist, we can drop 
the comment

> If we do keep the comment: s/for now/now/ (it has recently changed,
> we're not intending to change it later) and s/moths/months/.
>
> >    */
> >   void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top,
> >                    Error **errp)
> > @@ -4621,20 +4614,14 @@ void bdrv_append(BlockDriverState *bs_new, 
> > BlockDriverState *bs_top,
> >       bdrv_set_backing_hd(bs_new, bs_top, &local_err);
> >       if (local_err) {
> >           error_propagate(errp, local_err);
> > -        goto out;
> > +        return;
> >       }
> >   
> >       bdrv_replace_node(bs_top, bs_new, &local_err);
> >       if (local_err) {
> >           error_propagate(errp, local_err);
> >           bdrv_set_backing_hd(bs_new, NULL, &error_abort);
> > -        goto out;
>
> Can we leave a return here just in case that new code will be added at
> the end of the function?

sure

> >       }
> > -
> > -    /* bs_new is now referenced by its new parents, we don't need the
> > -     * additional reference any more. */
> > -out:
> > -    bdrv_unref(bs_new);
> >   }
> >   
> >   static void bdrv_delete(BlockDriverState *bs)
> > diff --git a/block/backup-top.c b/block/backup-top.c
> > index fe6883cc97..650ed6195c 100644
> > --- a/block/backup-top.c
> > +++ b/block/backup-top.c
> > @@ -222,7 +222,6 @@ BlockDriverState *bdrv_backup_top_append(BlockDriverState 
> > *source,
> >   
> >       bdrv_drained_begin(source);
> >   
> > -    bdrv_ref(top);
> >       bdrv_append(top, source, &local_err);
> >       if (local_err) {
> >           error_prepend(&local_err, "Cannot append backup-top filter: ");
> > diff --git a/block/commit.c b/block/commit.c
> > index 71db7ba747..61924bcf66 100644
> > --- a/block/commit.c
> > +++ b/block/commit.c
> > @@ -313,6 +313,7 @@ void commit_start(const char *job_id, BlockDriverState *bs,
> >       commit_top_bs->total_sectors = top->total_sectors;
> >   
> >       bdrv_append(commit_top_bs, top, &local_err);
> > +    bdrv_unref(commit_top_bs); /* referenced by new parents or failed */
> >       if (local_err) {
> >           commit_top_bs = NULL;
> >           error_propagate(errp, local_err);
> > diff --git a/block/mirror.c b/block/mirror.c
> > index 8e1ad6eceb..13f7ecc998 100644
> > --- a/block/mirror.c
> > +++ b/block/mirror.c
> > @@ -1605,9 +1605,6 @@ static BlockJob *mirror_start_job(
> >       bs_opaque = g_new0(MirrorBDSOpaque, 1);
> >       mirror_top_bs->opaque = bs_opaque;
> >   
> > -    /* bdrv_append takes ownership of the mirror_top_bs reference, need to keep
> > -     * it alive until block_job_create() succeeds even if bs has no parent. */
> > -    bdrv_ref(mirror_top_bs);
> >       bdrv_drained_begin(bs);
> >       bdrv_append(mirror_top_bs, bs, &local_err);
> >       bdrv_drained_end(bs);
> > diff --git a/blockdev.c b/blockdev.c
> > index b5f11c524b..96c96f8ba6 100644
> > --- a/blockdev.c
> > +++ b/blockdev.c
> > @@ -1587,10 +1587,6 @@ static void external_snapshot_prepare(BlkActionState 
> > *common,
> >           goto out;
> >       }
> >   
> > -    /* This removes our old bs and adds the new bs. This is an operation that
> > -     * can fail, so we need to do it in .prepare; undoing it for abort is
> > -     * always possible. */
>
> This comment is still relevant, it's unrelated to the bdrv_ref().

I described in commit message, why I dislike this comment :) I can keep it of 
course, it's not critical

> > -    bdrv_ref(state->new_bs);
> >       bdrv_append(state->new_bs, state->old_bs, &local_err);
> >       if (local_err) {
> >           error_propagate(errp, local_err);
>
> Kevin


--
Best regards,
Vladimir