On Wed, Jan 03, 2018 at 03:31:00PM +0800, Xin Long wrote:
> On Wed, Jan 3, 2018 at 5:44 AM, Marcelo Ricardo Leitner
> <marcelo.leit...@gmail.com> wrote:
> > syzbot reported a hang involving SCTP, on which it kept flooding dmesg
> > with the message:
> > [ 246.742374] sctp: sctp_transport_update_pmtu: Reported pmtu 508 too
> > low, using default minimum of 512
> >
> > That happened because whenever SCTP hits an ICMP Frag Needed, it tries
> > to adjust to the new MTU and triggers an immediate retransmission. But
> > it didn't consider the fact that MTUs smaller than the SCTP minimum MTU
> > allowed (512) would not cause the PMTU to change, and issued the
> > retransmission anyway (thus leading to another ICMP Frag Needed, and so
> > on).
> >
> > The fix is to disable Path MTU discovery for such transport and to skip
> > the retransmission in such cases. By doing this, SCTP will do the
> > backoff retransmissions as needed and will likely switch to another
> > transport if available.
> >
> > See-also: https://lkml.org/lkml/2017/12/22/811
> > Reported-by: syzbot <syzkal...@googlegroups.com>
> > Signed-off-by: Marcelo Ricardo Leitner <marcelo.leit...@gmail.com>
> > ---
> > net/sctp/input.c | 5 ++++-
> > net/sctp/transport.c | 2 ++
> > 2 files changed, 6 insertions(+), 1 deletion(-)
> >
> > diff --git a/net/sctp/input.c b/net/sctp/input.c
> > index
> > 621b5ca3fd1c17c3d7ef7bb1c7677ab98cebbe77..a24658c6f181e03d85f12dbe929c8bb4abaefcbd
> > 100644
> > --- a/net/sctp/input.c
> > +++ b/net/sctp/input.c
> > @@ -412,8 +412,11 @@ void sctp_icmp_frag_needed(struct sock *sk, struct
> > sctp_association *asoc,
> > * Needed will never be sent, but if a message was sent before
> > * PMTU discovery was disabled that was larger than the PMTU, it
> > * would not be fragmented, so it must be re-transmitted fragmented.
> > + * If the new PMTU is invalid, we will keep getting ICMP Frag
> > + * Needed. In this case, simply avoid the retransmit.
> > */
> > - sctp_retransmit(&asoc->outqueue, t, SCTP_RTXR_PMTUD);
> > + if (pmtu >= SCTP_DEFAULT_MINSEGMENT)
> > + sctp_retransmit(&asoc->outqueue, t, SCTP_RTXR_PMTUD);
> > }
> >
> > void sctp_icmp_redirect(struct sock *sk, struct sctp_transport *t,
> > diff --git a/net/sctp/transport.c b/net/sctp/transport.c
> > index
> > 1e5a22430cf56e40a6f323081beb97836b506384..fbd9fe25764d4d98f93c60a48eccefd9cc6b4165
> > 100644
> > --- a/net/sctp/transport.c
> > +++ b/net/sctp/transport.c
> > @@ -259,6 +259,8 @@ void sctp_transport_update_pmtu(struct sctp_transport
> > *t, u32 pmtu)
> > * pmtu discovery on this transport.
> > */
> > t->pathmtu = SCTP_DEFAULT_MINSEGMENT;
> > + t->param_flags = (t->param_flags & ~SPP_PMTUD) |
> > + SPP_PMTUD_DISABLE;
> It seems that once it hits here, this transport will have the minimum pmtu
> forever, even after t->dst has expired. It means this tx path will not come
> back to normal any more even when it gets a needfrag with reasonable
> pmtu. is it too harsh to this transport ?
That was the idea. That is what the comment above these lines is
describing already. Though I missed 06ad391919b2 ("[SCTP] Don't
disable PMTU discovery when mtu is small") and yes, too harsh.
>
> Another thing is on sctp_sendmsg, it also checks pmtu_pending that may
> be set by needfrag, and goes to sctp_assoc_sync_pmtu to trigger this
> warning again.
That is true but that's not an issue, is it? We are not trying to get
ride of the warning, instead we want to not cause a flood of
bogus retransmissions (which led to the flood of warnings).
By not disabling PMTU discovery (as above) we will have such warning
every now and then again for the same transport. We may add
_ratelimited to it, that would help in the case of we have like a
thousand transports suddenly being affected by such small MTU, but
won't omit it completely.
I'll spin a v2, thanks.
>
> > } else {
> > t->pathmtu = pmtu;
> > }
> > --
> > 2.14.3
> >
> --
> To unsubscribe from this list: send the line "unsubscribe linux-sctp" in
> the body of a message to majord...@vger.kernel.org
> More majordomo info at http://vger.kernel.org/majordomo-info.html
