Re: [PATCH] bcachefs: fix last_seq and last_empty_seq in bch2_fs_journal_start()

[Camila Alvarez Inostroza]

On Sun, 19 May 2024, Kent Overstreet wrote:

> On Wed, May 15, 2024 at 11:19:20PM -0400, Camila Alvarez wrote:
> > Values were left as the next possible sequence number when there were no
> > entries.
> >
> > The fix involves updating the last_seq initial value and
> > setting last_empty_seq to cur_seq - 1.
>
> I think this is correct, but we should try to come up with some better
> assertions or something to make the code clearer; we don't want off by
> ones to lurk so easily.
>
> Could you give it some thought?
You're right. I think the code is written in a confusing way. In 
particular it seems that cur_seq - 1 is used all over the place.
I believe we can abstract cur_seq - 1 in an independent variable (since 
it represents the actual last sequence number), that should make it 
clearer.
I'll share an updated version of the patch.
Thanks for the response!
> > Reported-by: syzbot+10b936c5eaee2819b49b@xxxxxxxxxxxxxxxxxxxxxxxxx
> > Signed-off-by: Camila Alvarez <cam.alvarez.i@xxxxxxxxx>
> > ---
> >  fs/bcachefs/journal.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/fs/bcachefs/journal.c b/fs/bcachefs/journal.c
> > index adec8e1ea73e..3835c458eec9 100644
> > --- a/fs/bcachefs/journal.c
> > +++ b/fs/bcachefs/journal.c
> > @@ -1196,7 +1196,7 @@ int bch2_fs_journal_start(struct journal *j, u64 cur_seq)
> >  	struct journal_replay *i, **_i;
> >  	struct genradix_iter iter;
> >  	bool had_entries = false;
> > -	u64 last_seq = cur_seq, nr, seq;
> > +	u64 last_seq = cur_seq - 1, nr, seq;
> >
> >  	genradix_for_each_reverse(&c->journal_entries, iter, _i) {
> >  		i = *_i;
> > @@ -1256,7 +1256,7 @@ int bch2_fs_journal_start(struct journal *j, u64 cur_seq)
> >  	}
> >
> >  	if (!had_entries)
> > -		j->last_empty_seq = cur_seq;
> > +		j->last_empty_seq = cur_seq - 1;
> >
> >  	spin_lock(&j->lock);
> >
> > --
> > 2.34.1