| Message ID | 20231026145018.GA19598@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | nfsd_copy_write_verifier: use read_seqbegin() rather than read_seqbegin_or_lock() | expand |
On 10/27, Chuck Lever wrote: > > On Thu, Oct 26, 2023 at 04:50:18PM +0200, Oleg Nesterov wrote: > > The usage of read_seqbegin_or_lock() in nfsd_copy_write_verifier() > > is wrong. "seq" is always even and thus "or_lock" has no effect, > > this code can never take ->writeverf_lock for writing. > > > > I guess this is fine, nfsd_copy_write_verifier() just copies 8 bytes > > and nfsd_reset_write_verifier() is supposed to be very rare operation > > so we do not need the adaptive locking in this case. > > > > Yet the code looks wrong and sub-optimal, it can use read_seqbegin() > > without changing the behaviour. > > I was debating whether to add Fixes/Cc-stable, but if the behavior > doesn't change, this doesn't need a backport. Yes, yes, sorry for confusion. This code is not buggy. Just a) it looks confusing because read_seqbegin_or_lock() doesn't do what it is supposed to do, and b) I am going to change the semantics of done_seqretry() to enforce the locking on the 2nd pass. Chuck, I can reword the changelog to make it more clear and send V2 if you think this makes sense. Thanks, Oleg.
> On Oct 27, 2023, at 12:34 PM, Oleg Nesterov <oleg@redhat.com> wrote: > > On 10/27, Chuck Lever wrote: >> >> On Thu, Oct 26, 2023 at 04:50:18PM +0200, Oleg Nesterov wrote: >>> The usage of read_seqbegin_or_lock() in nfsd_copy_write_verifier() >>> is wrong. "seq" is always even and thus "or_lock" has no effect, >>> this code can never take ->writeverf_lock for writing. >>> >>> I guess this is fine, nfsd_copy_write_verifier() just copies 8 bytes >>> and nfsd_reset_write_verifier() is supposed to be very rare operation >>> so we do not need the adaptive locking in this case. >>> >>> Yet the code looks wrong and sub-optimal, it can use read_seqbegin() >>> without changing the behaviour. >> >> I was debating whether to add Fixes/Cc-stable, but if the behavior >> doesn't change, this doesn't need a backport. > > Yes, yes, sorry for confusion. This code is not buggy. Just a) it looks > confusing because read_seqbegin_or_lock() doesn't do what it is supposed > to do, and b) I am going to change the semantics of done_seqretry() to > enforce the locking on the 2nd pass. > > Chuck, I can reword the changelog to make it more clear and send V2 if > you think this makes sense. No confusion, the changelog is clear to me. I'm simply stating my intention for other reviewers and the lore archive that I will leave off Fixes/Cc-stable when I commit your patch. So far there has been no review comment that suggests we need a v2. -- Chuck Lever
On Thu, 2023-10-26 at 16:50 +0200, Oleg Nesterov wrote: > The usage of read_seqbegin_or_lock() in nfsd_copy_write_verifier() > is wrong. "seq" is always even and thus "or_lock" has no effect, > this code can never take ->writeverf_lock for writing. > > I guess this is fine, nfsd_copy_write_verifier() just copies 8 bytes > and nfsd_reset_write_verifier() is supposed to be very rare operation > so we do not need the adaptive locking in this case. > > Yet the code looks wrong and sub-optimal, it can use read_seqbegin() > without changing the behaviour. > > Signed-off-by: Oleg Nesterov <oleg@redhat.com> > --- > fs/nfsd/nfssvc.c | 7 +++---- > 1 file changed, 3 insertions(+), 4 deletions(-) > > diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c > index c7af1095f6b5..094b765c5397 100644 > --- a/fs/nfsd/nfssvc.c > +++ b/fs/nfsd/nfssvc.c > @@ -359,13 +359,12 @@ static bool nfsd_needs_lockd(struct nfsd_net *nn) > */ > void nfsd_copy_write_verifier(__be32 verf[2], struct nfsd_net *nn) > { > - int seq = 0; > + unsigned seq; > > do { > - read_seqbegin_or_lock(&nn->writeverf_lock, &seq); > + seq = read_seqbegin(&nn->writeverf_lock); > memcpy(verf, nn->writeverf, sizeof(nn->writeverf)); > - } while (need_seqretry(&nn->writeverf_lock, seq)); > - done_seqretry(&nn->writeverf_lock, seq); > + } while (read_seqretry(&nn->writeverf_lock, seq)); > } > > static void nfsd_reset_write_verifier_locked(struct nfsd_net *nn) Reviewed-by: Jeff Layton <jlayton@kernel.org>
On Fri, 27 Oct 2023, Oleg Nesterov wrote: > The usage of read_seqbegin_or_lock() in nfsd_copy_write_verifier() > is wrong. "seq" is always even and thus "or_lock" has no effect, > this code can never take ->writeverf_lock for writing. > > I guess this is fine, nfsd_copy_write_verifier() just copies 8 bytes > and nfsd_reset_write_verifier() is supposed to be very rare operation > so we do not need the adaptive locking in this case. > > Yet the code looks wrong and sub-optimal, it can use read_seqbegin() > without changing the behaviour. Wow! read_seqbegin_or_lock() has never locked since Commit: 88a411c07b6f ("seqlock: livelock fix") in Linux v2.6.26 (2008). That's rather embarrassing. I agree we don't need the lock on the read-side for nfsd. Reviewed-by: NeilBrown <neilb@suse.de> NeilBrown > > Signed-off-by: Oleg Nesterov <oleg@redhat.com> > --- > fs/nfsd/nfssvc.c | 7 +++---- > 1 file changed, 3 insertions(+), 4 deletions(-) > > diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c > index c7af1095f6b5..094b765c5397 100644 > --- a/fs/nfsd/nfssvc.c > +++ b/fs/nfsd/nfssvc.c > @@ -359,13 +359,12 @@ static bool nfsd_needs_lockd(struct nfsd_net *nn) > */ > void nfsd_copy_write_verifier(__be32 verf[2], struct nfsd_net *nn) > { > - int seq = 0; > + unsigned seq; > > do { > - read_seqbegin_or_lock(&nn->writeverf_lock, &seq); > + seq = read_seqbegin(&nn->writeverf_lock); > memcpy(verf, nn->writeverf, sizeof(nn->writeverf)); > - } while (need_seqretry(&nn->writeverf_lock, seq)); > - done_seqretry(&nn->writeverf_lock, seq); > + } while (read_seqretry(&nn->writeverf_lock, seq)); > } > > static void nfsd_reset_write_verifier_locked(struct nfsd_net *nn) > -- > 2.25.1.362.g51ebf55 > > >
diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c index c7af1095f6b5..094b765c5397 100644 --- a/fs/nfsd/nfssvc.c +++ b/fs/nfsd/nfssvc.c @@ -359,13 +359,12 @@ static bool nfsd_needs_lockd(struct nfsd_net *nn) */ void nfsd_copy_write_verifier(__be32 verf[2], struct nfsd_net *nn) { - int seq = 0; + unsigned seq; do { - read_seqbegin_or_lock(&nn->writeverf_lock, &seq); + seq = read_seqbegin(&nn->writeverf_lock); memcpy(verf, nn->writeverf, sizeof(nn->writeverf)); - } while (need_seqretry(&nn->writeverf_lock, seq)); - done_seqretry(&nn->writeverf_lock, seq); + } while (read_seqretry(&nn->writeverf_lock, seq)); } static void nfsd_reset_write_verifier_locked(struct nfsd_net *nn)
The usage of read_seqbegin_or_lock() in nfsd_copy_write_verifier() is wrong. "seq" is always even and thus "or_lock" has no effect, this code can never take ->writeverf_lock for writing. I guess this is fine, nfsd_copy_write_verifier() just copies 8 bytes and nfsd_reset_write_verifier() is supposed to be very rare operation so we do not need the adaptive locking in this case. Yet the code looks wrong and sub-optimal, it can use read_seqbegin() without changing the behaviour. Signed-off-by: Oleg Nesterov <oleg@redhat.com> --- fs/nfsd/nfssvc.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)