[SCTP]: Fix connection hang/slowdown with PR-SCTP
authorVlad Yasevich <vladislav.yasevich@hp.com>
Wed, 21 Feb 2007 10:06:04 +0000 (02:06 -0800)
committerDavid S. Miller <davem@sunset.davemloft.net>
Mon, 26 Feb 2007 19:42:49 +0000 (11:42 -0800)
The problem that this patch corrects happens when all of the following
conditions are satisfisfied:
 1.  PR-SCTP is used and the timeout on the chunks is set below RTO.Max.
 2.  One of the paths on a multihomed associations is brought down.

In this scenario, data will expire within the rto of the initial
transmission and will never be retransmitted.  However this data still
fills the send buffer and is counted against the association as outstanding
data.  This causes any new data not to be sent and retransmission to not
happen.

The fix is to discount the abandoned data from the outstanding count and
peers rwnd estimation.  This allows new data to be sent and a retransmission
timer restarted.  Even though this new data will most likely expire within
the rto, the timer still counts as a strike against the transport and forces
the FORWARD-TSN chunk to be retransmitted as well.

Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: Sridhar Samudrala <sri@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/sctp/outqueue.c

index 5c2ddd10db065629139d0544cc439ffa42e10687..41abfd17627e1a3b0a75029acf1f78031f1c8447 100644 (file)
@@ -396,6 +396,19 @@ void sctp_retransmit_mark(struct sctp_outq *q,
                if (sctp_chunk_abandoned(chunk)) {
                        list_del_init(lchunk);
                        sctp_insert_list(&q->abandoned, lchunk);
+
+                       /* If this chunk has not been previousely acked,
+                        * stop considering it 'outstanding'.  Our peer
+                        * will most likely never see it since it will
+                        * not be retransmitted
+                        */
+                       if (!chunk->tsn_gap_acked) {
+                               chunk->transport->flight_size -=
+                                               sctp_data_size(chunk);
+                               q->outstanding_bytes -= sctp_data_size(chunk);
+                               q->asoc->peer.rwnd += (sctp_data_size(chunk) +
+                                                       sizeof(struct sk_buff));
+                       }
                        continue;
                }
 
@@ -1244,6 +1257,15 @@ static void sctp_check_transmitted(struct sctp_outq *q,
                if (sctp_chunk_abandoned(tchunk)) {
                        /* Move the chunk to abandoned list. */
                        sctp_insert_list(&q->abandoned, lchunk);
+
+                       /* If this chunk has not been acked, stop
+                        * considering it as 'outstanding'.
+                        */
+                       if (!tchunk->tsn_gap_acked) {
+                               tchunk->transport->flight_size -=
+                                               sctp_data_size(tchunk);
+                               q->outstanding_bytes -= sctp_data_size(tchunk);
+                       }
                        continue;
                }
 
@@ -1695,11 +1717,6 @@ static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
                 */
                if (TSN_lte(tsn, ctsn)) {
                        list_del_init(lchunk);
-                       if (!chunk->tsn_gap_acked) {
-                               chunk->transport->flight_size -=
-                                       sctp_data_size(chunk);
-                               q->outstanding_bytes -= sctp_data_size(chunk);
-                       }
                        sctp_chunk_free(chunk);
                } else {
                        if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {