On Mon 18-05-15 09:13:59, Mike Snitzer wrote:
> On Mon, May 18 2015 at 3:22am -0400,
> Jan Kara <jack@xxxxxxx> wrote:
>
> > On Thu 14-05-15 17:04:59, Mike Snitzer wrote:
> > > Commit c4cf5261 ("bio: skip atomic inc/dec of ->bi_remaining for
> > > non-chains") regressed all existing callers that followed this pattern:
> > > 1) saving a bio's original bi_end_io
> > > 2) wiring up an intermediate bi_end_io
> > > 3) restoring the original bi_end_io from intermediate bi_end_io
> > > 4) calling bio_endio() to execute the restored original bi_end_io
> > >
> > > The regression was due to BIO_CHAIN only ever getting set if
> > > bio_inc_remaining() is called. For the above pattern it isn't set until
> > > step 3 above (step 2 would've needed to establish BIO_CHAIN). As such
> > > the first bio_endio(), in step 2 above, never decremented __bi_remaining
> > > before calling the intermediate bi_end_io -- leaving __bi_remaining with
> > > the value 1 instead of 0. When bio_inc_remaining() occurred during step
> > > 3 it brought it to a value of 2. When the second bio_endio() was
> > > called, in step 4 above, it should've called the original bi_end_io but
> > > it didn't because there was an extra reference that wasn't dropped (due
> > > to atomic operations being optimized away since BIO_CHAIN wasn't set
> > > upfront).
> > >
> > > Fix this issue by removing the __bi_remaining management complexity for
> > > all callers that use the above pattern -- bio_chain() is the only
> > > interface that _needs_ to be concerned with __bi_remaining. For the
> > > above pattern callers just expect the bi_end_io they set to get called!
> > > Remove bio_endio_nodec() and also remove all bio_inc_remaining() calls
> > > that aren't associated with the bio_chain() interface.
> > >
> > > The bio_inc_remaining() interface has been left exported because it is
> > > still useful for more elaborate uses of bio_chain() -- it will be used
> > > in an upcoming DM commit "dm thin: range discard support".
> > >
> > > Fixes: c4cf5261 ("bio: skip atomic inc/dec of ->bi_remaining for non-chains")
> > > Signed-off-by: Mike Snitzer <snitzer@xxxxxxxxxx>
> > > Cc: Jan Kara <jack@xxxxxxx>
> > > Cc: Chris Mason <clm@xxxxxx>
> > One question: What happens if you stack dm-thin on top of e.g. dm-linear?
> > dm-thin will do it's thing to a bio and passes it to dm-linear. That will
> > split & chain the bio so BIO_CHAIN will be set. And on IO completion you
> > will have troubles in dm-thinp as now bi_remaining gets decremented in
> > bio_endio(). That's the reason why I suggested that we should clear
> > BIO_CHAIN once bi_remaining hits zero...
>
> I think you need to be more precise in explaining the scenario you're
> concerned about. Could be there is an issue but I'm not seeing it yet.
>
> Are you referring to the patch that makes DM thinp use the proposed
> blkdev_issue_discard_async() interface? The bios issued to DM linear
> are generated by blkdev_issue_discard_async(). By using bio_chain()
> they establish ancestory with the parent DM thinp bio (which has
> had BIO_CHAIN set even before calling blkdev_issue_discard_async because
> there is potential for DM thinp to complete the parent bio before all N
> blkdev_issue_discard_async() generated bios complete -- so that is why
> DM thinp itself takes an extra reference on the parent bio using
> bio_inc_remaining() before calling blkdev_issue_discard_async)
No, I'm not referring to your proposed interface. I'm referring to
current kernel + your patch to remove bio_inc_remaining() from all the dm
targets. Ah, after checking again I see where misunderstanding may have
come from - the device below has to be handled by drivers/md/linear.c which
is MD linear driver, not DM one. I confused those two. Anyway here is the
failure I envision (and frankly, I don't understand dm details much so I may
be just completely wrong but I'd like to understand what prevents the following
from happening):
* We have dm-thin stacked on top of drivers/dm/linear.c
* FS issues bio to dm-thin. remap_and_issue_overwrite() sets bi_end_io to
overwrite_endio. dm-thin eventually calls generic_make_request(bio).
* Now linear_make_request() gets called and it ends up calling
bio_chain(split, bio). This sets BIO_CHAIN on bio.
* IO for all chained bios is completed. So bio->bi_remaining is now zero,
bio still has BIO_CHAIN set and overwrite_endio gets called.
* process_prepared_mapping() will eventually try to call original bi_end_io
callback but that never happens because bi_remaining is 0 and BIO_CHAIN
remained set.
Honza
--
Jan Kara <jack@xxxxxxx>
SUSE Labs, CR
--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel
