[SCTP]: Flush fragment queue when exiting partial delivery.
authorVlad Yasevich <vladislav.yasevich@hp.com>
Sun, 16 Dec 2007 22:05:45 +0000 (14:05 -0800)
committerDavid S. Miller <davem@davemloft.net>
Sun, 16 Dec 2007 22:05:45 +0000 (14:05 -0800)
At the end of partial delivery, we may have complete messages
sitting on the fragment queue.  These messages are stuck there
until a new fragment arrives.  This can comletely stall a
given association.  When clearing partial delivery state, flush
any complete messages from the fragment queue and send them on
their way up.

Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/sctp/ulpqueue.c

index 4908041..1733fa2 100644 (file)
@@ -53,6 +53,7 @@ static struct sctp_ulpevent * sctp_ulpq_reasm(struct sctp_ulpq *ulpq,
                                              struct sctp_ulpevent *);
 static struct sctp_ulpevent * sctp_ulpq_order(struct sctp_ulpq *,
                                              struct sctp_ulpevent *);
+static void sctp_ulpq_reasm_drain(struct sctp_ulpq *ulpq);
 
 /* 1st Level Abstractions */
 
@@ -190,6 +191,7 @@ static void sctp_ulpq_set_pd(struct sctp_ulpq *ulpq)
 static int sctp_ulpq_clear_pd(struct sctp_ulpq *ulpq)
 {
        ulpq->pd_mode = 0;
+       sctp_ulpq_reasm_drain(ulpq);
        return sctp_clear_pd(ulpq->asoc->base.sk, ulpq->asoc);
 }
 
@@ -699,6 +701,37 @@ void sctp_ulpq_reasm_flushtsn(struct sctp_ulpq *ulpq, __u32 fwd_tsn)
        }
 }
 
+/*
+ * Drain the reassembly queue.  If we just cleared parted delivery, it
+ * is possible that the reassembly queue will contain already reassembled
+ * messages.  Retrieve any such messages and give them to the user.
+ */
+static void sctp_ulpq_reasm_drain(struct sctp_ulpq *ulpq)
+{
+       struct sctp_ulpevent *event = NULL;
+       struct sk_buff_head temp;
+
+       if (skb_queue_empty(&ulpq->reasm))
+               return;
+
+       while ((event = sctp_ulpq_retrieve_reassembled(ulpq)) != NULL) {
+               /* Do ordering if needed.  */
+               if ((event) && (event->msg_flags & MSG_EOR)){
+                       skb_queue_head_init(&temp);
+                       __skb_queue_tail(&temp, sctp_event2skb(event));
+
+                       event = sctp_ulpq_order(ulpq, event);
+               }
+
+               /* Send event to the ULP.  'event' is the
+                * sctp_ulpevent for  very first SKB on the  temp' list.
+                */
+               if (event)
+                       sctp_ulpq_tail_event(ulpq, event);
+       }
+}
+
+
 /* Helper function to gather skbs that have possibly become
  * ordered by an an incoming chunk.
  */