[v6,09/12] block: Add "drained begin/end" for internal snapshot

This ensures the atomicity of the transaction by avoiding processing of
external requests such as those from ioeventfd.

state->bs is assigned right after bdrv_drained_begin. Because it was
used as the flag for deletion or not in abort, now we need a separate
flag - InternalSnapshotState.created.

Signed-off-by: Fam Zheng <famz@redhat.com>
---
 blockdev.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

Am 22.10.2015 um 08:32 hat Fam Zheng geschrieben:
> This ensures the atomicity of the transaction by avoiding processing of
> external requests such as those from ioeventfd.
> 
> state->bs is assigned right after bdrv_drained_begin. Because it was
> used as the flag for deletion or not in abort, now we need a separate
> flag - InternalSnapshotState.created.
> 
> Signed-off-by: Fam Zheng <famz@redhat.com>
> ---
>  blockdev.c | 9 +++++++--
>  1 file changed, 7 insertions(+), 2 deletions(-)
> 
> diff --git a/blockdev.c b/blockdev.c
> index 52f44b2..adc0e69 100644
> --- a/blockdev.c
> +++ b/blockdev.c
> @@ -1370,6 +1370,7 @@ typedef struct InternalSnapshotState {
>      BlockDriverState *bs;
>      AioContext *aio_context;
>      QEMUSnapshotInfo sn;
> +    bool created;
>  } InternalSnapshotState;
>  
>  static void internal_snapshot_prepare(BlkTransactionState *common,
> @@ -1414,6 +1415,9 @@ static void internal_snapshot_prepare(BlkTransactionState *common,
>      }
>      bs = blk_bs(blk);
>  
> +    state->bs = bs;
> +    bdrv_drained_begin(bs);
> +
>      if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
>          return;
>      }

More context:

    /* AioContext is released in .clean() */
    state->aio_context = blk_get_aio_context(blk);
    aio_context_acquire(state->aio_context);

    if (!blk_is_available(blk)) {
        error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
        return;
    }
    bs = blk_bs(blk);

    state->bs = bs;
    bdrv_drained_begin(bs);

If we error out because of !blk_is_available(blk), we will still call
bdrv_drained_end() in .clean even though bdrv_drained_begin() hasn't
been called yet.

Kevin

On Thu, Oct 22, 2015 at 12:18:11PM +0200, Kevin Wolf wrote:
> Am 22.10.2015 um 08:32 hat Fam Zheng geschrieben:
> > This ensures the atomicity of the transaction by avoiding processing of
> > external requests such as those from ioeventfd.
> > 
> > state->bs is assigned right after bdrv_drained_begin. Because it was
> > used as the flag for deletion or not in abort, now we need a separate
> > flag - InternalSnapshotState.created.
> > 
> > Signed-off-by: Fam Zheng <famz@redhat.com>
> > ---
> >  blockdev.c | 9 +++++++--
> >  1 file changed, 7 insertions(+), 2 deletions(-)
> > 
> > diff --git a/blockdev.c b/blockdev.c
> > index 52f44b2..adc0e69 100644
> > --- a/blockdev.c
> > +++ b/blockdev.c
> > @@ -1370,6 +1370,7 @@ typedef struct InternalSnapshotState {
> >      BlockDriverState *bs;
> >      AioContext *aio_context;
> >      QEMUSnapshotInfo sn;
> > +    bool created;
> >  } InternalSnapshotState;
> >  
> >  static void internal_snapshot_prepare(BlkTransactionState *common,
> > @@ -1414,6 +1415,9 @@ static void internal_snapshot_prepare(BlkTransactionState *common,
> >      }
> >      bs = blk_bs(blk);
> >  
> > +    state->bs = bs;
> > +    bdrv_drained_begin(bs);
> > +
> >      if (bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_INTERNAL_SNAPSHOT, errp)) {
> >          return;
> >      }
> 
> More context:
> 
>     /* AioContext is released in .clean() */
>     state->aio_context = blk_get_aio_context(blk);
>     aio_context_acquire(state->aio_context);
> 
>     if (!blk_is_available(blk)) {
>         error_setg(errp, QERR_DEVICE_HAS_NO_MEDIUM, device);
>         return;
>     }
>     bs = blk_bs(blk);
> 
>     state->bs = bs;
>     bdrv_drained_begin(bs);
> 
> If we error out because of !blk_is_available(blk), we will still call
> bdrv_drained_end() in .clean even though bdrv_drained_begin() hasn't
> been called yet.
>

Right, if the bdrv_drained_begin() is moved down as it is in this
patch, then the state->aio_context assignment needs moved down as
well.