RE: [for-4.2 PATCH 3/6] replay: update docs for record/replay with block devices

[Pavel Dovgalyuk]

> From: Kevin Wolf [mailto:address@hidden]
> Am 19.09.2019 um 14:10 hat Pavel Dovgalyuk geschrieben:
> > > From: Kevin Wolf [mailto:address@hidden]
> > > Am 19.09.2019 um 11:05 hat Pavel Dovgalyuk geschrieben:
> > > > > From: Kevin Wolf [mailto:address@hidden]
> > > > > > >
> > > > > > > However, global -snapshot is just a convenient shortcut for 
> > > > > > > specifying
> > > > > > > snapshot=on for all -drive arguments. So if -snapshot is 
> > > > > > > incompatible
> > > > > > > with replay, shouldn't manually marking all drives as snapshot=on 
> > > > > > > be
> > > > > > > incompatible as well?
> > > > > > >
> > > > > > > Maybe you're really interested in some specific drive not having
> > > > > > > snapshot=on? But then it might be better to check that specific 
> > > > > > > drive
> > > > > > > instad of forbidding just the shortcut for setting it.
> > > > > >
> > > > > > -snapshot adds the flag for top-level drive, making driver 
> > > > > > operations
> > > > > > dependent on temporary file structure.
> > > > > >
> > > > > > Moving this overlay beneath blkreplay driver makes drive operations
> > > > > > deterministic for the top-level device.
> > > > >
> > > > > So the real requirement is that blkreplay is the top-level node of any
> > > > > guest device, right? And only because of this, you can't use -snapshot
> > > > > (or snapshot=on on the blkreplay driver).
> > > > >
> > > > > If we instead check e.g. in blk_insert_bs() or blk_attach_dev() that 
> > > > > in
> > > > > record/replay mode, the root node of the BlockBackend is blkreplay,
> > > > > wouldn't we catch many more incorrect setups?
> > > >
> > > > That sounds interesting.
> > > > Will it help to check that every backend is connected to blkreplay?
> > >
> > > Yes, it would return an error when you try to attach a non-blkreplay
> > > node to a BlockBackend (and every guest device uses a BlockBackend).
> > >
> > > Note that this restriction would currently make block jobs unavailable
> > > on non-blkreplay nodes as they also use BlockBackends internally (though
> > > this is going to change in the long run). I believe this restriction is
> > > harmless and the typical replay use case doesn't involve any block jobs,
> > > but if you do think it's a problem, blk_attach_dev() would be the place
> > > that affects only devices.
> > >
> > > > How then this check has to be done?
> > >
> > > Only compile-tested, but maybe something like below?
> > >
> > > Kevin
> > >
> > > diff --git a/include/block/block_int.h b/include/block/block_int.h
> > > index 0422acdf1c..9fa72bea51 100644
> > > --- a/include/block/block_int.h
> > > +++ b/include/block/block_int.h
> > > @@ -955,6 +955,7 @@ static inline BlockDriverState 
> > > *backing_bs(BlockDriverState *bs)
> > >  extern BlockDriver bdrv_file;
> > >  extern BlockDriver bdrv_raw;
> > >  extern BlockDriver bdrv_qcow2;
> > > +extern BlockDriver bdrv_blkreplay;
> > >
> > >  int coroutine_fn bdrv_co_preadv(BdrvChild *child,
> > >      int64_t offset, unsigned int bytes, QEMUIOVector *qiov,
> > > diff --git a/block/blkreplay.c b/block/blkreplay.c
> > > index 2b7931b940..16a4f1df6a 100644
> > > --- a/block/blkreplay.c
> > > +++ b/block/blkreplay.c
> > > @@ -126,7 +126,7 @@ static int coroutine_fn 
> > > blkreplay_co_flush(BlockDriverState *bs)
> > >      return ret;
> > >  }
> > >
> > > -static BlockDriver bdrv_blkreplay = {
> > > +BlockDriver bdrv_blkreplay = {
> > >      .format_name            = "blkreplay",
> > >      .instance_size          = 0,
> > >
> > > diff --git a/block/block-backend.c b/block/block-backend.c
> > > index 1c605d5444..c57d3d9fdf 100644
> > > --- a/block/block-backend.c
> > > +++ b/block/block-backend.c
> > > @@ -17,6 +17,7 @@
> > >  #include "block/throttle-groups.h"
> > >  #include "hw/qdev-core.h"
> > >  #include "sysemu/blockdev.h"
> > > +#include "sysemu/replay.h"
> > >  #include "sysemu/runstate.h"
> > >  #include "qapi/error.h"
> > >  #include "qapi/qapi-events-block.h"
> > > @@ -808,6 +809,12 @@ void blk_remove_bs(BlockBackend *blk)
> > >  int blk_insert_bs(BlockBackend *blk, BlockDriverState *bs, Error **errp)
> > >  {
> > >      ThrottleGroupMember *tgm = &blk->public.throttle_group_member;
> > > +
> > > +    if (replay_mode != REPLAY_MODE_NONE && bs->drv != &bdrv_blkreplay) {
> > > +        error_setg(errp, "Root node must be blkreplay");
> > > +        return -ENOTSUP;
> > > +    }
> >
> > I guess this is opposite direction - bs->drv is bdrv_file.
> > And we should check its parent.
> 
> If bs->drv is bdrv_file, you want this to fail because only
> bdrv_blkreplay should be able to be attached to devices.

There was a regular rr invocation (as described in docs).
And bs->drv always was a pointer to bdrv_file: for original image,
and for temporary snapshot.

Pavel Dovgalyuk