1 /* SCTP kernel implementation
2 * (C) Copyright IBM Corp. 2001, 2004
3 * Copyright (c) 1999-2000 Cisco, Inc.
4 * Copyright (c) 1999-2001 Motorola, Inc.
5 * Copyright (c) 2001-2003 Intel Corp.
7 * This file is part of the SCTP kernel implementation
9 * These functions implement the sctp_outq class. The outqueue handles
10 * bundling and queueing of outgoing SCTP chunks.
12 * This SCTP implementation is free software;
13 * you can redistribute it and/or modify it under the terms of
14 * the GNU General Public License as published by
15 * the Free Software Foundation; either version 2, or (at your option)
18 * This SCTP implementation is distributed in the hope that it
19 * will be useful, but WITHOUT ANY WARRANTY; without even the implied
20 * ************************
21 * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
22 * See the GNU General Public License for more details.
24 * You should have received a copy of the GNU General Public License
25 * along with GNU CC; see the file COPYING. If not, see
26 * <http://www.gnu.org/licenses/>.
28 * Please send any bug reports or fixes you make to the
30 * lksctp developers <linux-sctp@vger.kernel.org>
32 * Written or modified by:
33 * La Monte H.P. Yarroll <piggy@acm.org>
34 * Karl Knutson <karl@athena.chicago.il.us>
35 * Perry Melange <pmelange@null.cc.uic.edu>
36 * Xingang Guo <xingang.guo@intel.com>
37 * Hui Huang <hui.huang@nokia.com>
38 * Sridhar Samudrala <sri@us.ibm.com>
39 * Jon Grimm <jgrimm@us.ibm.com>
42 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
44 #include <linux/types.h>
45 #include <linux/list.h> /* For struct list_head */
46 #include <linux/socket.h>
48 #include <linux/slab.h>
49 #include <net/sock.h> /* For skb_set_owner_w */
51 #include <net/sctp/sctp.h>
52 #include <net/sctp/sm.h>
54 /* Declare internal functions here. */
55 static int sctp_acked(struct sctp_sackhdr
*sack
, __u32 tsn
);
56 static void sctp_check_transmitted(struct sctp_outq
*q
,
57 struct list_head
*transmitted_queue
,
58 struct sctp_transport
*transport
,
59 union sctp_addr
*saddr
,
60 struct sctp_sackhdr
*sack
,
61 __u32
*highest_new_tsn
);
63 static void sctp_mark_missing(struct sctp_outq
*q
,
64 struct list_head
*transmitted_queue
,
65 struct sctp_transport
*transport
,
66 __u32 highest_new_tsn
,
67 int count_of_newacks
);
69 static void sctp_generate_fwdtsn(struct sctp_outq
*q
, __u32 sack_ctsn
);
71 static int sctp_outq_flush(struct sctp_outq
*q
, int rtx_timeout
, gfp_t gfp
);
73 /* Add data to the front of the queue. */
74 static inline void sctp_outq_head_data(struct sctp_outq
*q
,
75 struct sctp_chunk
*ch
)
77 list_add(&ch
->list
, &q
->out_chunk_list
);
78 q
->out_qlen
+= ch
->skb
->len
;
81 /* Take data from the front of the queue. */
82 static inline struct sctp_chunk
*sctp_outq_dequeue_data(struct sctp_outq
*q
)
84 struct sctp_chunk
*ch
= NULL
;
86 if (!list_empty(&q
->out_chunk_list
)) {
87 struct list_head
*entry
= q
->out_chunk_list
.next
;
89 ch
= list_entry(entry
, struct sctp_chunk
, list
);
91 q
->out_qlen
-= ch
->skb
->len
;
95 /* Add data chunk to the end of the queue. */
96 static inline void sctp_outq_tail_data(struct sctp_outq
*q
,
97 struct sctp_chunk
*ch
)
99 list_add_tail(&ch
->list
, &q
->out_chunk_list
);
100 q
->out_qlen
+= ch
->skb
->len
;
104 * SFR-CACC algorithm:
105 * D) If count_of_newacks is greater than or equal to 2
106 * and t was not sent to the current primary then the
107 * sender MUST NOT increment missing report count for t.
109 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport
*primary
,
110 struct sctp_transport
*transport
,
111 int count_of_newacks
)
113 if (count_of_newacks
>= 2 && transport
!= primary
)
119 * SFR-CACC algorithm:
120 * F) If count_of_newacks is less than 2, let d be the
121 * destination to which t was sent. If cacc_saw_newack
122 * is 0 for destination d, then the sender MUST NOT
123 * increment missing report count for t.
125 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport
*transport
,
126 int count_of_newacks
)
128 if (count_of_newacks
< 2 &&
129 (transport
&& !transport
->cacc
.cacc_saw_newack
))
135 * SFR-CACC algorithm:
136 * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD
137 * execute steps C, D, F.
139 * C has been implemented in sctp_outq_sack
141 static inline int sctp_cacc_skip_3_1(struct sctp_transport
*primary
,
142 struct sctp_transport
*transport
,
143 int count_of_newacks
)
145 if (!primary
->cacc
.cycling_changeover
) {
146 if (sctp_cacc_skip_3_1_d(primary
, transport
, count_of_newacks
))
148 if (sctp_cacc_skip_3_1_f(transport
, count_of_newacks
))
156 * SFR-CACC algorithm:
157 * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less
158 * than next_tsn_at_change of the current primary, then
159 * the sender MUST NOT increment missing report count
162 static inline int sctp_cacc_skip_3_2(struct sctp_transport
*primary
, __u32 tsn
)
164 if (primary
->cacc
.cycling_changeover
&&
165 TSN_lt(tsn
, primary
->cacc
.next_tsn_at_change
))
171 * SFR-CACC algorithm:
172 * 3) If the missing report count for TSN t is to be
173 * incremented according to [RFC2960] and
174 * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set,
175 * then the sender MUST further execute steps 3.1 and
176 * 3.2 to determine if the missing report count for
177 * TSN t SHOULD NOT be incremented.
179 * 3.3) If 3.1 and 3.2 do not dictate that the missing
180 * report count for t should not be incremented, then
181 * the sender SHOULD increment missing report count for
182 * t (according to [RFC2960] and [SCTP_STEWART_2002]).
184 static inline int sctp_cacc_skip(struct sctp_transport
*primary
,
185 struct sctp_transport
*transport
,
186 int count_of_newacks
,
189 if (primary
->cacc
.changeover_active
&&
190 (sctp_cacc_skip_3_1(primary
, transport
, count_of_newacks
) ||
191 sctp_cacc_skip_3_2(primary
, tsn
)))
196 /* Initialize an existing sctp_outq. This does the boring stuff.
197 * You still need to define handlers if you really want to DO
198 * something with this structure...
200 void sctp_outq_init(struct sctp_association
*asoc
, struct sctp_outq
*q
)
202 memset(q
, 0, sizeof(struct sctp_outq
));
205 INIT_LIST_HEAD(&q
->out_chunk_list
);
206 INIT_LIST_HEAD(&q
->control_chunk_list
);
207 INIT_LIST_HEAD(&q
->retransmit
);
208 INIT_LIST_HEAD(&q
->sacked
);
209 INIT_LIST_HEAD(&q
->abandoned
);
212 /* Free the outqueue structure and any related pending chunks.
214 static void __sctp_outq_teardown(struct sctp_outq
*q
)
216 struct sctp_transport
*transport
;
217 struct list_head
*lchunk
, *temp
;
218 struct sctp_chunk
*chunk
, *tmp
;
220 /* Throw away unacknowledged chunks. */
221 list_for_each_entry(transport
, &q
->asoc
->peer
.transport_addr_list
,
223 while ((lchunk
= sctp_list_dequeue(&transport
->transmitted
)) != NULL
) {
224 chunk
= list_entry(lchunk
, struct sctp_chunk
,
226 /* Mark as part of a failed message. */
227 sctp_chunk_fail(chunk
, q
->error
);
228 sctp_chunk_free(chunk
);
232 /* Throw away chunks that have been gap ACKed. */
233 list_for_each_safe(lchunk
, temp
, &q
->sacked
) {
234 list_del_init(lchunk
);
235 chunk
= list_entry(lchunk
, struct sctp_chunk
,
237 sctp_chunk_fail(chunk
, q
->error
);
238 sctp_chunk_free(chunk
);
241 /* Throw away any chunks in the retransmit queue. */
242 list_for_each_safe(lchunk
, temp
, &q
->retransmit
) {
243 list_del_init(lchunk
);
244 chunk
= list_entry(lchunk
, struct sctp_chunk
,
246 sctp_chunk_fail(chunk
, q
->error
);
247 sctp_chunk_free(chunk
);
250 /* Throw away any chunks that are in the abandoned queue. */
251 list_for_each_safe(lchunk
, temp
, &q
->abandoned
) {
252 list_del_init(lchunk
);
253 chunk
= list_entry(lchunk
, struct sctp_chunk
,
255 sctp_chunk_fail(chunk
, q
->error
);
256 sctp_chunk_free(chunk
);
259 /* Throw away any leftover data chunks. */
260 while ((chunk
= sctp_outq_dequeue_data(q
)) != NULL
) {
262 /* Mark as send failure. */
263 sctp_chunk_fail(chunk
, q
->error
);
264 sctp_chunk_free(chunk
);
267 /* Throw away any leftover control chunks. */
268 list_for_each_entry_safe(chunk
, tmp
, &q
->control_chunk_list
, list
) {
269 list_del_init(&chunk
->list
);
270 sctp_chunk_free(chunk
);
274 void sctp_outq_teardown(struct sctp_outq
*q
)
276 __sctp_outq_teardown(q
);
277 sctp_outq_init(q
->asoc
, q
);
280 /* Free the outqueue structure and any related pending chunks. */
281 void sctp_outq_free(struct sctp_outq
*q
)
283 /* Throw away leftover chunks. */
284 __sctp_outq_teardown(q
);
287 /* Put a new chunk in an sctp_outq. */
288 int sctp_outq_tail(struct sctp_outq
*q
, struct sctp_chunk
*chunk
, gfp_t gfp
)
290 struct net
*net
= sock_net(q
->asoc
->base
.sk
);
293 pr_debug("%s: outq:%p, chunk:%p[%s]\n", __func__
, q
, chunk
,
294 chunk
&& chunk
->chunk_hdr
?
295 sctp_cname(SCTP_ST_CHUNK(chunk
->chunk_hdr
->type
)) :
298 /* If it is data, queue it up, otherwise, send it
301 if (sctp_chunk_is_data(chunk
)) {
302 /* Is it OK to queue data chunks? */
303 /* From 9. Termination of Association
305 * When either endpoint performs a shutdown, the
306 * association on each peer will stop accepting new
307 * data from its user and only deliver data in queue
308 * at the time of sending or receiving the SHUTDOWN
311 switch (q
->asoc
->state
) {
312 case SCTP_STATE_CLOSED
:
313 case SCTP_STATE_SHUTDOWN_PENDING
:
314 case SCTP_STATE_SHUTDOWN_SENT
:
315 case SCTP_STATE_SHUTDOWN_RECEIVED
:
316 case SCTP_STATE_SHUTDOWN_ACK_SENT
:
317 /* Cannot send after transport endpoint shutdown */
322 pr_debug("%s: outqueueing: outq:%p, chunk:%p[%s])\n",
323 __func__
, q
, chunk
, chunk
&& chunk
->chunk_hdr
?
324 sctp_cname(SCTP_ST_CHUNK(chunk
->chunk_hdr
->type
)) :
327 sctp_chunk_hold(chunk
);
328 sctp_outq_tail_data(q
, chunk
);
329 if (chunk
->chunk_hdr
->flags
& SCTP_DATA_UNORDERED
)
330 SCTP_INC_STATS(net
, SCTP_MIB_OUTUNORDERCHUNKS
);
332 SCTP_INC_STATS(net
, SCTP_MIB_OUTORDERCHUNKS
);
336 list_add_tail(&chunk
->list
, &q
->control_chunk_list
);
337 SCTP_INC_STATS(net
, SCTP_MIB_OUTCTRLCHUNKS
);
344 error
= sctp_outq_flush(q
, 0, gfp
);
349 /* Insert a chunk into the sorted list based on the TSNs. The retransmit list
350 * and the abandoned list are in ascending order.
352 static void sctp_insert_list(struct list_head
*head
, struct list_head
*new)
354 struct list_head
*pos
;
355 struct sctp_chunk
*nchunk
, *lchunk
;
359 nchunk
= list_entry(new, struct sctp_chunk
, transmitted_list
);
360 ntsn
= ntohl(nchunk
->subh
.data_hdr
->tsn
);
362 list_for_each(pos
, head
) {
363 lchunk
= list_entry(pos
, struct sctp_chunk
, transmitted_list
);
364 ltsn
= ntohl(lchunk
->subh
.data_hdr
->tsn
);
365 if (TSN_lt(ntsn
, ltsn
)) {
366 list_add(new, pos
->prev
);
372 list_add_tail(new, head
);
375 /* Mark all the eligible packets on a transport for retransmission. */
376 void sctp_retransmit_mark(struct sctp_outq
*q
,
377 struct sctp_transport
*transport
,
380 struct list_head
*lchunk
, *ltemp
;
381 struct sctp_chunk
*chunk
;
383 /* Walk through the specified transmitted queue. */
384 list_for_each_safe(lchunk
, ltemp
, &transport
->transmitted
) {
385 chunk
= list_entry(lchunk
, struct sctp_chunk
,
388 /* If the chunk is abandoned, move it to abandoned list. */
389 if (sctp_chunk_abandoned(chunk
)) {
390 list_del_init(lchunk
);
391 sctp_insert_list(&q
->abandoned
, lchunk
);
393 /* If this chunk has not been previousely acked,
394 * stop considering it 'outstanding'. Our peer
395 * will most likely never see it since it will
396 * not be retransmitted
398 if (!chunk
->tsn_gap_acked
) {
399 if (chunk
->transport
)
400 chunk
->transport
->flight_size
-=
401 sctp_data_size(chunk
);
402 q
->outstanding_bytes
-= sctp_data_size(chunk
);
403 q
->asoc
->peer
.rwnd
+= sctp_data_size(chunk
);
408 /* If we are doing retransmission due to a timeout or pmtu
409 * discovery, only the chunks that are not yet acked should
410 * be added to the retransmit queue.
412 if ((reason
== SCTP_RTXR_FAST_RTX
&&
413 (chunk
->fast_retransmit
== SCTP_NEED_FRTX
)) ||
414 (reason
!= SCTP_RTXR_FAST_RTX
&& !chunk
->tsn_gap_acked
)) {
415 /* RFC 2960 6.2.1 Processing a Received SACK
417 * C) Any time a DATA chunk is marked for
418 * retransmission (via either T3-rtx timer expiration
419 * (Section 6.3.3) or via fast retransmit
420 * (Section 7.2.4)), add the data size of those
421 * chunks to the rwnd.
423 q
->asoc
->peer
.rwnd
+= sctp_data_size(chunk
);
424 q
->outstanding_bytes
-= sctp_data_size(chunk
);
425 if (chunk
->transport
)
426 transport
->flight_size
-= sctp_data_size(chunk
);
428 /* sctpimpguide-05 Section 2.8.2
429 * M5) If a T3-rtx timer expires, the
430 * 'TSN.Missing.Report' of all affected TSNs is set
433 chunk
->tsn_missing_report
= 0;
435 /* If a chunk that is being used for RTT measurement
436 * has to be retransmitted, we cannot use this chunk
437 * anymore for RTT measurements. Reset rto_pending so
438 * that a new RTT measurement is started when a new
439 * data chunk is sent.
441 if (chunk
->rtt_in_progress
) {
442 chunk
->rtt_in_progress
= 0;
443 transport
->rto_pending
= 0;
448 /* Move the chunk to the retransmit queue. The chunks
449 * on the retransmit queue are always kept in order.
451 list_del_init(lchunk
);
452 sctp_insert_list(&q
->retransmit
, lchunk
);
456 pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d, "
457 "flight_size:%d, pba:%d\n", __func__
, transport
, reason
,
458 transport
->cwnd
, transport
->ssthresh
, transport
->flight_size
,
459 transport
->partial_bytes_acked
);
462 /* Mark all the eligible packets on a transport for retransmission and force
465 void sctp_retransmit(struct sctp_outq
*q
, struct sctp_transport
*transport
,
466 sctp_retransmit_reason_t reason
)
468 struct net
*net
= sock_net(q
->asoc
->base
.sk
);
472 case SCTP_RTXR_T3_RTX
:
473 SCTP_INC_STATS(net
, SCTP_MIB_T3_RETRANSMITS
);
474 sctp_transport_lower_cwnd(transport
, SCTP_LOWER_CWND_T3_RTX
);
475 /* Update the retran path if the T3-rtx timer has expired for
476 * the current retran path.
478 if (transport
== transport
->asoc
->peer
.retran_path
)
479 sctp_assoc_update_retran_path(transport
->asoc
);
480 transport
->asoc
->rtx_data_chunks
+=
481 transport
->asoc
->unack_data
;
483 case SCTP_RTXR_FAST_RTX
:
484 SCTP_INC_STATS(net
, SCTP_MIB_FAST_RETRANSMITS
);
485 sctp_transport_lower_cwnd(transport
, SCTP_LOWER_CWND_FAST_RTX
);
488 case SCTP_RTXR_PMTUD
:
489 SCTP_INC_STATS(net
, SCTP_MIB_PMTUD_RETRANSMITS
);
491 case SCTP_RTXR_T1_RTX
:
492 SCTP_INC_STATS(net
, SCTP_MIB_T1_RETRANSMITS
);
493 transport
->asoc
->init_retries
++;
499 sctp_retransmit_mark(q
, transport
, reason
);
501 /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
502 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
503 * following the procedures outlined in C1 - C5.
505 if (reason
== SCTP_RTXR_T3_RTX
)
506 sctp_generate_fwdtsn(q
, q
->asoc
->ctsn_ack_point
);
508 /* Flush the queues only on timeout, since fast_rtx is only
509 * triggered during sack processing and the queue
510 * will be flushed at the end.
512 if (reason
!= SCTP_RTXR_FAST_RTX
)
513 error
= sctp_outq_flush(q
, /* rtx_timeout */ 1, GFP_ATOMIC
);
516 q
->asoc
->base
.sk
->sk_err
= -error
;
520 * Transmit DATA chunks on the retransmit queue. Upon return from
521 * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
522 * need to be transmitted by the caller.
523 * We assume that pkt->transport has already been set.
525 * The return value is a normal kernel error return value.
527 static int sctp_outq_flush_rtx(struct sctp_outq
*q
, struct sctp_packet
*pkt
,
528 int rtx_timeout
, int *start_timer
)
530 struct list_head
*lqueue
;
531 struct sctp_transport
*transport
= pkt
->transport
;
533 struct sctp_chunk
*chunk
, *chunk1
;
539 lqueue
= &q
->retransmit
;
540 fast_rtx
= q
->fast_rtx
;
542 /* This loop handles time-out retransmissions, fast retransmissions,
543 * and retransmissions due to opening of whindow.
545 * RFC 2960 6.3.3 Handle T3-rtx Expiration
547 * E3) Determine how many of the earliest (i.e., lowest TSN)
548 * outstanding DATA chunks for the address for which the
549 * T3-rtx has expired will fit into a single packet, subject
550 * to the MTU constraint for the path corresponding to the
551 * destination transport address to which the retransmission
552 * is being sent (this may be different from the address for
553 * which the timer expires [see Section 6.4]). Call this value
554 * K. Bundle and retransmit those K DATA chunks in a single
555 * packet to the destination endpoint.
557 * [Just to be painfully clear, if we are retransmitting
558 * because a timeout just happened, we should send only ONE
559 * packet of retransmitted data.]
561 * For fast retransmissions we also send only ONE packet. However,
562 * if we are just flushing the queue due to open window, we'll
563 * try to send as much as possible.
565 list_for_each_entry_safe(chunk
, chunk1
, lqueue
, transmitted_list
) {
566 /* If the chunk is abandoned, move it to abandoned list. */
567 if (sctp_chunk_abandoned(chunk
)) {
568 list_del_init(&chunk
->transmitted_list
);
569 sctp_insert_list(&q
->abandoned
,
570 &chunk
->transmitted_list
);
574 /* Make sure that Gap Acked TSNs are not retransmitted. A
575 * simple approach is just to move such TSNs out of the
576 * way and into a 'transmitted' queue and skip to the
579 if (chunk
->tsn_gap_acked
) {
580 list_move_tail(&chunk
->transmitted_list
,
581 &transport
->transmitted
);
585 /* If we are doing fast retransmit, ignore non-fast_rtransmit
588 if (fast_rtx
&& !chunk
->fast_retransmit
)
592 /* Attempt to append this chunk to the packet. */
593 status
= sctp_packet_append_chunk(pkt
, chunk
);
596 case SCTP_XMIT_PMTU_FULL
:
597 if (!pkt
->has_data
&& !pkt
->has_cookie_echo
) {
598 /* If this packet did not contain DATA then
599 * retransmission did not happen, so do it
600 * again. We'll ignore the error here since
601 * control chunks are already freed so there
602 * is nothing we can do.
604 sctp_packet_transmit(pkt
, GFP_ATOMIC
);
608 /* Send this packet. */
609 error
= sctp_packet_transmit(pkt
, GFP_ATOMIC
);
611 /* If we are retransmitting, we should only
612 * send a single packet.
613 * Otherwise, try appending this chunk again.
615 if (rtx_timeout
|| fast_rtx
)
620 /* Bundle next chunk in the next round. */
623 case SCTP_XMIT_RWND_FULL
:
624 /* Send this packet. */
625 error
= sctp_packet_transmit(pkt
, GFP_ATOMIC
);
627 /* Stop sending DATA as there is no more room
633 case SCTP_XMIT_DELAY
:
634 /* Send this packet. */
635 error
= sctp_packet_transmit(pkt
, GFP_ATOMIC
);
637 /* Stop sending DATA because of nagle delay. */
642 /* The append was successful, so add this chunk to
643 * the transmitted list.
645 list_move_tail(&chunk
->transmitted_list
,
646 &transport
->transmitted
);
648 /* Mark the chunk as ineligible for fast retransmit
649 * after it is retransmitted.
651 if (chunk
->fast_retransmit
== SCTP_NEED_FRTX
)
652 chunk
->fast_retransmit
= SCTP_DONT_FRTX
;
654 q
->asoc
->stats
.rtxchunks
++;
658 /* Set the timer if there were no errors */
659 if (!error
&& !timer
)
666 /* If we are here due to a retransmit timeout or a fast
667 * retransmit and if there are any chunks left in the retransmit
668 * queue that could not fit in the PMTU sized packet, they need
669 * to be marked as ineligible for a subsequent fast retransmit.
671 if (rtx_timeout
|| fast_rtx
) {
672 list_for_each_entry(chunk1
, lqueue
, transmitted_list
) {
673 if (chunk1
->fast_retransmit
== SCTP_NEED_FRTX
)
674 chunk1
->fast_retransmit
= SCTP_DONT_FRTX
;
678 *start_timer
= timer
;
680 /* Clear fast retransmit hint */
687 /* Cork the outqueue so queued chunks are really queued. */
688 int sctp_outq_uncork(struct sctp_outq
*q
, gfp_t gfp
)
693 return sctp_outq_flush(q
, 0, gfp
);
698 * Try to flush an outqueue.
700 * Description: Send everything in q which we legally can, subject to
701 * congestion limitations.
702 * * Note: This function can be called from multiple contexts so appropriate
703 * locking concerns must be made. Today we use the sock lock to protect
706 static int sctp_outq_flush(struct sctp_outq
*q
, int rtx_timeout
, gfp_t gfp
)
708 struct sctp_packet
*packet
;
709 struct sctp_packet singleton
;
710 struct sctp_association
*asoc
= q
->asoc
;
711 __u16 sport
= asoc
->base
.bind_addr
.port
;
712 __u16 dport
= asoc
->peer
.port
;
713 __u32 vtag
= asoc
->peer
.i
.init_tag
;
714 struct sctp_transport
*transport
= NULL
;
715 struct sctp_transport
*new_transport
;
716 struct sctp_chunk
*chunk
, *tmp
;
722 /* These transports have chunks to send. */
723 struct list_head transport_list
;
724 struct list_head
*ltransport
;
726 INIT_LIST_HEAD(&transport_list
);
732 * When bundling control chunks with DATA chunks, an
733 * endpoint MUST place control chunks first in the outbound
734 * SCTP packet. The transmitter MUST transmit DATA chunks
735 * within a SCTP packet in increasing order of TSN.
739 list_for_each_entry_safe(chunk
, tmp
, &q
->control_chunk_list
, list
) {
741 * F1) This means that until such time as the ASCONF
742 * containing the add is acknowledged, the sender MUST
743 * NOT use the new IP address as a source for ANY SCTP
744 * packet except on carrying an ASCONF Chunk.
746 if (asoc
->src_out_of_asoc_ok
&&
747 chunk
->chunk_hdr
->type
!= SCTP_CID_ASCONF
)
750 list_del_init(&chunk
->list
);
752 /* Pick the right transport to use. */
753 new_transport
= chunk
->transport
;
755 if (!new_transport
) {
757 * If we have a prior transport pointer, see if
758 * the destination address of the chunk
759 * matches the destination address of the
760 * current transport. If not a match, then
761 * try to look up the transport with a given
762 * destination address. We do this because
763 * after processing ASCONFs, we may have new
764 * transports created.
767 sctp_cmp_addr_exact(&chunk
->dest
,
769 new_transport
= transport
;
771 new_transport
= sctp_assoc_lookup_paddr(asoc
,
774 /* if we still don't have a new transport, then
775 * use the current active path.
778 new_transport
= asoc
->peer
.active_path
;
779 } else if ((new_transport
->state
== SCTP_INACTIVE
) ||
780 (new_transport
->state
== SCTP_UNCONFIRMED
) ||
781 (new_transport
->state
== SCTP_PF
)) {
782 /* If the chunk is Heartbeat or Heartbeat Ack,
783 * send it to chunk->transport, even if it's
786 * 3.3.6 Heartbeat Acknowledgement:
788 * A HEARTBEAT ACK is always sent to the source IP
789 * address of the IP datagram containing the
790 * HEARTBEAT chunk to which this ack is responding.
793 * ASCONF_ACKs also must be sent to the source.
795 if (chunk
->chunk_hdr
->type
!= SCTP_CID_HEARTBEAT
&&
796 chunk
->chunk_hdr
->type
!= SCTP_CID_HEARTBEAT_ACK
&&
797 chunk
->chunk_hdr
->type
!= SCTP_CID_ASCONF_ACK
)
798 new_transport
= asoc
->peer
.active_path
;
801 /* Are we switching transports?
802 * Take care of transport locks.
804 if (new_transport
!= transport
) {
805 transport
= new_transport
;
806 if (list_empty(&transport
->send_ready
)) {
807 list_add_tail(&transport
->send_ready
,
810 packet
= &transport
->packet
;
811 sctp_packet_config(packet
, vtag
,
812 asoc
->peer
.ecn_capable
);
815 switch (chunk
->chunk_hdr
->type
) {
819 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
820 * COMPLETE with any other chunks. [Send them immediately.]
823 case SCTP_CID_INIT_ACK
:
824 case SCTP_CID_SHUTDOWN_COMPLETE
:
825 sctp_packet_init(&singleton
, transport
, sport
, dport
);
826 sctp_packet_config(&singleton
, vtag
, 0);
827 sctp_packet_append_chunk(&singleton
, chunk
);
828 error
= sctp_packet_transmit(&singleton
, gfp
);
834 if (sctp_test_T_bit(chunk
)) {
835 packet
->vtag
= asoc
->c
.my_vtag
;
837 /* The following chunks are "response" chunks, i.e.
838 * they are generated in response to something we
839 * received. If we are sending these, then we can
840 * send only 1 packet containing these chunks.
842 case SCTP_CID_HEARTBEAT_ACK
:
843 case SCTP_CID_SHUTDOWN_ACK
:
844 case SCTP_CID_COOKIE_ACK
:
845 case SCTP_CID_COOKIE_ECHO
:
847 case SCTP_CID_ECN_CWR
:
848 case SCTP_CID_ASCONF_ACK
:
853 case SCTP_CID_HEARTBEAT
:
854 case SCTP_CID_SHUTDOWN
:
855 case SCTP_CID_ECN_ECNE
:
856 case SCTP_CID_ASCONF
:
857 case SCTP_CID_FWD_TSN
:
858 status
= sctp_packet_transmit_chunk(packet
, chunk
,
860 if (status
!= SCTP_XMIT_OK
) {
861 /* put the chunk back */
862 list_add(&chunk
->list
, &q
->control_chunk_list
);
864 asoc
->stats
.octrlchunks
++;
865 /* PR-SCTP C5) If a FORWARD TSN is sent, the
866 * sender MUST assure that at least one T3-rtx
869 if (chunk
->chunk_hdr
->type
== SCTP_CID_FWD_TSN
)
870 sctp_transport_reset_timers(transport
);
875 /* We built a chunk with an illegal type! */
880 if (q
->asoc
->src_out_of_asoc_ok
)
883 /* Is it OK to send data chunks? */
884 switch (asoc
->state
) {
885 case SCTP_STATE_COOKIE_ECHOED
:
886 /* Only allow bundling when this packet has a COOKIE-ECHO
889 if (!packet
|| !packet
->has_cookie_echo
)
893 case SCTP_STATE_ESTABLISHED
:
894 case SCTP_STATE_SHUTDOWN_PENDING
:
895 case SCTP_STATE_SHUTDOWN_RECEIVED
:
897 * RFC 2960 6.1 Transmission of DATA Chunks
899 * C) When the time comes for the sender to transmit,
900 * before sending new DATA chunks, the sender MUST
901 * first transmit any outstanding DATA chunks which
902 * are marked for retransmission (limited by the
905 if (!list_empty(&q
->retransmit
)) {
906 if (asoc
->peer
.retran_path
->state
== SCTP_UNCONFIRMED
)
908 if (transport
== asoc
->peer
.retran_path
)
911 /* Switch transports & prepare the packet. */
913 transport
= asoc
->peer
.retran_path
;
915 if (list_empty(&transport
->send_ready
)) {
916 list_add_tail(&transport
->send_ready
,
920 packet
= &transport
->packet
;
921 sctp_packet_config(packet
, vtag
,
922 asoc
->peer
.ecn_capable
);
924 error
= sctp_outq_flush_rtx(q
, packet
,
925 rtx_timeout
, &start_timer
);
928 sctp_transport_reset_timers(transport
);
930 /* This can happen on COOKIE-ECHO resend. Only
931 * one chunk can get bundled with a COOKIE-ECHO.
933 if (packet
->has_cookie_echo
)
936 /* Don't send new data if there is still data
937 * waiting to retransmit.
939 if (!list_empty(&q
->retransmit
))
943 /* Apply Max.Burst limitation to the current transport in
944 * case it will be used for new data. We are going to
945 * rest it before we return, but we want to apply the limit
946 * to the currently queued data.
949 sctp_transport_burst_limited(transport
);
951 /* Finally, transmit new packets. */
952 while ((chunk
= sctp_outq_dequeue_data(q
)) != NULL
) {
953 /* RFC 2960 6.5 Every DATA chunk MUST carry a valid
956 if (chunk
->sinfo
.sinfo_stream
>=
957 asoc
->c
.sinit_num_ostreams
) {
959 /* Mark as failed send. */
960 sctp_chunk_fail(chunk
, SCTP_ERROR_INV_STRM
);
961 sctp_chunk_free(chunk
);
965 /* Has this chunk expired? */
966 if (sctp_chunk_abandoned(chunk
)) {
967 sctp_chunk_fail(chunk
, 0);
968 sctp_chunk_free(chunk
);
972 /* If there is a specified transport, use it.
973 * Otherwise, we want to use the active path.
975 new_transport
= chunk
->transport
;
976 if (!new_transport
||
977 ((new_transport
->state
== SCTP_INACTIVE
) ||
978 (new_transport
->state
== SCTP_UNCONFIRMED
) ||
979 (new_transport
->state
== SCTP_PF
)))
980 new_transport
= asoc
->peer
.active_path
;
981 if (new_transport
->state
== SCTP_UNCONFIRMED
) {
982 WARN_ONCE(1, "Atempt to send packet on unconfirmed path.");
983 sctp_chunk_fail(chunk
, 0);
984 sctp_chunk_free(chunk
);
988 /* Change packets if necessary. */
989 if (new_transport
!= transport
) {
990 transport
= new_transport
;
992 /* Schedule to have this transport's
995 if (list_empty(&transport
->send_ready
)) {
996 list_add_tail(&transport
->send_ready
,
1000 packet
= &transport
->packet
;
1001 sctp_packet_config(packet
, vtag
,
1002 asoc
->peer
.ecn_capable
);
1003 /* We've switched transports, so apply the
1004 * Burst limit to the new transport.
1006 sctp_transport_burst_limited(transport
);
1009 pr_debug("%s: outq:%p, chunk:%p[%s], tx-tsn:0x%x skb->head:%p "
1011 __func__
, q
, chunk
, chunk
&& chunk
->chunk_hdr
?
1012 sctp_cname(SCTP_ST_CHUNK(chunk
->chunk_hdr
->type
)) :
1013 "illegal chunk", ntohl(chunk
->subh
.data_hdr
->tsn
),
1014 chunk
->skb
? chunk
->skb
->head
: NULL
, chunk
->skb
?
1015 atomic_read(&chunk
->skb
->users
) : -1);
1017 /* Add the chunk to the packet. */
1018 status
= sctp_packet_transmit_chunk(packet
, chunk
, 0, gfp
);
1021 case SCTP_XMIT_PMTU_FULL
:
1022 case SCTP_XMIT_RWND_FULL
:
1023 case SCTP_XMIT_DELAY
:
1024 /* We could not append this chunk, so put
1025 * the chunk back on the output queue.
1027 pr_debug("%s: could not transmit tsn:0x%x, status:%d\n",
1028 __func__
, ntohl(chunk
->subh
.data_hdr
->tsn
),
1031 sctp_outq_head_data(q
, chunk
);
1032 goto sctp_flush_out
;
1035 /* The sender is in the SHUTDOWN-PENDING state,
1036 * The sender MAY set the I-bit in the DATA
1039 if (asoc
->state
== SCTP_STATE_SHUTDOWN_PENDING
)
1040 chunk
->chunk_hdr
->flags
|= SCTP_DATA_SACK_IMM
;
1041 if (chunk
->chunk_hdr
->flags
& SCTP_DATA_UNORDERED
)
1042 asoc
->stats
.ouodchunks
++;
1044 asoc
->stats
.oodchunks
++;
1052 /* BUG: We assume that the sctp_packet_transmit()
1053 * call below will succeed all the time and add the
1054 * chunk to the transmitted list and restart the
1056 * It is possible that the call can fail under OOM
1059 * Is this really a problem? Won't this behave
1062 list_add_tail(&chunk
->transmitted_list
,
1063 &transport
->transmitted
);
1065 sctp_transport_reset_timers(transport
);
1067 /* Only let one DATA chunk get bundled with a
1068 * COOKIE-ECHO chunk.
1070 if (packet
->has_cookie_echo
)
1071 goto sctp_flush_out
;
1082 /* Before returning, examine all the transports touched in
1083 * this call. Right now, we bluntly force clear all the
1084 * transports. Things might change after we implement Nagle.
1085 * But such an examination is still required.
1089 while ((ltransport
= sctp_list_dequeue(&transport_list
)) != NULL
) {
1090 struct sctp_transport
*t
= list_entry(ltransport
,
1091 struct sctp_transport
,
1093 packet
= &t
->packet
;
1094 if (!sctp_packet_empty(packet
))
1095 error
= sctp_packet_transmit(packet
, gfp
);
1097 /* Clear the burst limited state, if any */
1098 sctp_transport_burst_reset(t
);
1104 /* Update unack_data based on the incoming SACK chunk */
1105 static void sctp_sack_update_unack_data(struct sctp_association
*assoc
,
1106 struct sctp_sackhdr
*sack
)
1108 sctp_sack_variable_t
*frags
;
1112 unack_data
= assoc
->next_tsn
- assoc
->ctsn_ack_point
- 1;
1114 frags
= sack
->variable
;
1115 for (i
= 0; i
< ntohs(sack
->num_gap_ack_blocks
); i
++) {
1116 unack_data
-= ((ntohs(frags
[i
].gab
.end
) -
1117 ntohs(frags
[i
].gab
.start
) + 1));
1120 assoc
->unack_data
= unack_data
;
1123 /* This is where we REALLY process a SACK.
1125 * Process the SACK against the outqueue. Mostly, this just frees
1126 * things off the transmitted queue.
1128 int sctp_outq_sack(struct sctp_outq
*q
, struct sctp_chunk
*chunk
)
1130 struct sctp_association
*asoc
= q
->asoc
;
1131 struct sctp_sackhdr
*sack
= chunk
->subh
.sack_hdr
;
1132 struct sctp_transport
*transport
;
1133 struct sctp_chunk
*tchunk
= NULL
;
1134 struct list_head
*lchunk
, *transport_list
, *temp
;
1135 sctp_sack_variable_t
*frags
= sack
->variable
;
1136 __u32 sack_ctsn
, ctsn
, tsn
;
1137 __u32 highest_tsn
, highest_new_tsn
;
1139 unsigned int outstanding
;
1140 struct sctp_transport
*primary
= asoc
->peer
.primary_path
;
1141 int count_of_newacks
= 0;
1145 /* Grab the association's destination address list. */
1146 transport_list
= &asoc
->peer
.transport_addr_list
;
1148 sack_ctsn
= ntohl(sack
->cum_tsn_ack
);
1149 gap_ack_blocks
= ntohs(sack
->num_gap_ack_blocks
);
1150 asoc
->stats
.gapcnt
+= gap_ack_blocks
;
1152 * SFR-CACC algorithm:
1153 * On receipt of a SACK the sender SHOULD execute the
1154 * following statements.
1156 * 1) If the cumulative ack in the SACK passes next tsn_at_change
1157 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1158 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1160 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1161 * is set the receiver of the SACK MUST take the following actions:
1163 * A) Initialize the cacc_saw_newack to 0 for all destination
1166 * Only bother if changeover_active is set. Otherwise, this is
1167 * totally suboptimal to do on every SACK.
1169 if (primary
->cacc
.changeover_active
) {
1170 u8 clear_cycling
= 0;
1172 if (TSN_lte(primary
->cacc
.next_tsn_at_change
, sack_ctsn
)) {
1173 primary
->cacc
.changeover_active
= 0;
1177 if (clear_cycling
|| gap_ack_blocks
) {
1178 list_for_each_entry(transport
, transport_list
,
1181 transport
->cacc
.cycling_changeover
= 0;
1183 transport
->cacc
.cacc_saw_newack
= 0;
1188 /* Get the highest TSN in the sack. */
1189 highest_tsn
= sack_ctsn
;
1191 highest_tsn
+= ntohs(frags
[gap_ack_blocks
- 1].gab
.end
);
1193 if (TSN_lt(asoc
->highest_sacked
, highest_tsn
))
1194 asoc
->highest_sacked
= highest_tsn
;
1196 highest_new_tsn
= sack_ctsn
;
1198 /* Run through the retransmit queue. Credit bytes received
1199 * and free those chunks that we can.
1201 sctp_check_transmitted(q
, &q
->retransmit
, NULL
, NULL
, sack
, &highest_new_tsn
);
1203 /* Run through the transmitted queue.
1204 * Credit bytes received and free those chunks which we can.
1206 * This is a MASSIVE candidate for optimization.
1208 list_for_each_entry(transport
, transport_list
, transports
) {
1209 sctp_check_transmitted(q
, &transport
->transmitted
,
1210 transport
, &chunk
->source
, sack
,
1213 * SFR-CACC algorithm:
1214 * C) Let count_of_newacks be the number of
1215 * destinations for which cacc_saw_newack is set.
1217 if (transport
->cacc
.cacc_saw_newack
)
1221 /* Move the Cumulative TSN Ack Point if appropriate. */
1222 if (TSN_lt(asoc
->ctsn_ack_point
, sack_ctsn
)) {
1223 asoc
->ctsn_ack_point
= sack_ctsn
;
1227 if (gap_ack_blocks
) {
1229 if (asoc
->fast_recovery
&& accum_moved
)
1230 highest_new_tsn
= highest_tsn
;
1232 list_for_each_entry(transport
, transport_list
, transports
)
1233 sctp_mark_missing(q
, &transport
->transmitted
, transport
,
1234 highest_new_tsn
, count_of_newacks
);
1237 /* Update unack_data field in the assoc. */
1238 sctp_sack_update_unack_data(asoc
, sack
);
1240 ctsn
= asoc
->ctsn_ack_point
;
1242 /* Throw away stuff rotting on the sack queue. */
1243 list_for_each_safe(lchunk
, temp
, &q
->sacked
) {
1244 tchunk
= list_entry(lchunk
, struct sctp_chunk
,
1246 tsn
= ntohl(tchunk
->subh
.data_hdr
->tsn
);
1247 if (TSN_lte(tsn
, ctsn
)) {
1248 list_del_init(&tchunk
->transmitted_list
);
1249 sctp_chunk_free(tchunk
);
1253 /* ii) Set rwnd equal to the newly received a_rwnd minus the
1254 * number of bytes still outstanding after processing the
1255 * Cumulative TSN Ack and the Gap Ack Blocks.
1258 sack_a_rwnd
= ntohl(sack
->a_rwnd
);
1259 asoc
->peer
.zero_window_announced
= !sack_a_rwnd
;
1260 outstanding
= q
->outstanding_bytes
;
1262 if (outstanding
< sack_a_rwnd
)
1263 sack_a_rwnd
-= outstanding
;
1267 asoc
->peer
.rwnd
= sack_a_rwnd
;
1269 sctp_generate_fwdtsn(q
, sack_ctsn
);
1271 pr_debug("%s: sack cumulative tsn ack:0x%x\n", __func__
, sack_ctsn
);
1272 pr_debug("%s: cumulative tsn ack of assoc:%p is 0x%x, "
1273 "advertised peer ack point:0x%x\n", __func__
, asoc
, ctsn
,
1274 asoc
->adv_peer_ack_point
);
1276 return sctp_outq_is_empty(q
);
1279 /* Is the outqueue empty?
1280 * The queue is empty when we have not pending data, no in-flight data
1281 * and nothing pending retransmissions.
1283 int sctp_outq_is_empty(const struct sctp_outq
*q
)
1285 return q
->out_qlen
== 0 && q
->outstanding_bytes
== 0 &&
1286 list_empty(&q
->retransmit
);
1289 /********************************************************************
1290 * 2nd Level Abstractions
1291 ********************************************************************/
1293 /* Go through a transport's transmitted list or the association's retransmit
1294 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1295 * The retransmit list will not have an associated transport.
1297 * I added coherent debug information output. --xguo
1299 * Instead of printing 'sacked' or 'kept' for each TSN on the
1300 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1301 * KEPT TSN6-TSN7, etc.
1303 static void sctp_check_transmitted(struct sctp_outq
*q
,
1304 struct list_head
*transmitted_queue
,
1305 struct sctp_transport
*transport
,
1306 union sctp_addr
*saddr
,
1307 struct sctp_sackhdr
*sack
,
1308 __u32
*highest_new_tsn_in_sack
)
1310 struct list_head
*lchunk
;
1311 struct sctp_chunk
*tchunk
;
1312 struct list_head tlist
;
1316 __u8 restart_timer
= 0;
1317 int bytes_acked
= 0;
1318 int migrate_bytes
= 0;
1319 bool forward_progress
= false;
1321 sack_ctsn
= ntohl(sack
->cum_tsn_ack
);
1323 INIT_LIST_HEAD(&tlist
);
1325 /* The while loop will skip empty transmitted queues. */
1326 while (NULL
!= (lchunk
= sctp_list_dequeue(transmitted_queue
))) {
1327 tchunk
= list_entry(lchunk
, struct sctp_chunk
,
1330 if (sctp_chunk_abandoned(tchunk
)) {
1331 /* Move the chunk to abandoned list. */
1332 sctp_insert_list(&q
->abandoned
, lchunk
);
1334 /* If this chunk has not been acked, stop
1335 * considering it as 'outstanding'.
1337 if (!tchunk
->tsn_gap_acked
) {
1338 if (tchunk
->transport
)
1339 tchunk
->transport
->flight_size
-=
1340 sctp_data_size(tchunk
);
1341 q
->outstanding_bytes
-= sctp_data_size(tchunk
);
1346 tsn
= ntohl(tchunk
->subh
.data_hdr
->tsn
);
1347 if (sctp_acked(sack
, tsn
)) {
1348 /* If this queue is the retransmit queue, the
1349 * retransmit timer has already reclaimed
1350 * the outstanding bytes for this chunk, so only
1351 * count bytes associated with a transport.
1354 /* If this chunk is being used for RTT
1355 * measurement, calculate the RTT and update
1356 * the RTO using this value.
1358 * 6.3.1 C5) Karn's algorithm: RTT measurements
1359 * MUST NOT be made using packets that were
1360 * retransmitted (and thus for which it is
1361 * ambiguous whether the reply was for the
1362 * first instance of the packet or a later
1365 if (!tchunk
->tsn_gap_acked
&&
1367 tchunk
->rtt_in_progress
) {
1368 tchunk
->rtt_in_progress
= 0;
1369 rtt
= jiffies
- tchunk
->sent_at
;
1370 sctp_transport_update_rto(transport
,
1375 /* If the chunk hasn't been marked as ACKED,
1376 * mark it and account bytes_acked if the
1377 * chunk had a valid transport (it will not
1378 * have a transport if ASCONF had deleted it
1379 * while DATA was outstanding).
1381 if (!tchunk
->tsn_gap_acked
) {
1382 tchunk
->tsn_gap_acked
= 1;
1383 if (TSN_lt(*highest_new_tsn_in_sack
, tsn
))
1384 *highest_new_tsn_in_sack
= tsn
;
1385 bytes_acked
+= sctp_data_size(tchunk
);
1386 if (!tchunk
->transport
)
1387 migrate_bytes
+= sctp_data_size(tchunk
);
1388 forward_progress
= true;
1391 if (TSN_lte(tsn
, sack_ctsn
)) {
1392 /* RFC 2960 6.3.2 Retransmission Timer Rules
1394 * R3) Whenever a SACK is received
1395 * that acknowledges the DATA chunk
1396 * with the earliest outstanding TSN
1397 * for that address, restart T3-rtx
1398 * timer for that address with its
1402 forward_progress
= true;
1404 if (!tchunk
->tsn_gap_acked
) {
1406 * SFR-CACC algorithm:
1407 * 2) If the SACK contains gap acks
1408 * and the flag CHANGEOVER_ACTIVE is
1409 * set the receiver of the SACK MUST
1410 * take the following action:
1412 * B) For each TSN t being acked that
1413 * has not been acked in any SACK so
1414 * far, set cacc_saw_newack to 1 for
1415 * the destination that the TSN was
1419 sack
->num_gap_ack_blocks
&&
1420 q
->asoc
->peer
.primary_path
->cacc
.
1422 transport
->cacc
.cacc_saw_newack
1426 list_add_tail(&tchunk
->transmitted_list
,
1429 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1430 * M2) Each time a SACK arrives reporting
1431 * 'Stray DATA chunk(s)' record the highest TSN
1432 * reported as newly acknowledged, call this
1433 * value 'HighestTSNinSack'. A newly
1434 * acknowledged DATA chunk is one not
1435 * previously acknowledged in a SACK.
1437 * When the SCTP sender of data receives a SACK
1438 * chunk that acknowledges, for the first time,
1439 * the receipt of a DATA chunk, all the still
1440 * unacknowledged DATA chunks whose TSN is
1441 * older than that newly acknowledged DATA
1442 * chunk, are qualified as 'Stray DATA chunks'.
1444 list_add_tail(lchunk
, &tlist
);
1447 if (tchunk
->tsn_gap_acked
) {
1448 pr_debug("%s: receiver reneged on data TSN:0x%x\n",
1451 tchunk
->tsn_gap_acked
= 0;
1453 if (tchunk
->transport
)
1454 bytes_acked
-= sctp_data_size(tchunk
);
1456 /* RFC 2960 6.3.2 Retransmission Timer Rules
1458 * R4) Whenever a SACK is received missing a
1459 * TSN that was previously acknowledged via a
1460 * Gap Ack Block, start T3-rtx for the
1461 * destination address to which the DATA
1462 * chunk was originally
1463 * transmitted if it is not already running.
1468 list_add_tail(lchunk
, &tlist
);
1474 struct sctp_association
*asoc
= transport
->asoc
;
1476 /* We may have counted DATA that was migrated
1477 * to this transport due to DEL-IP operation.
1478 * Subtract those bytes, since the were never
1479 * send on this transport and shouldn't be
1480 * credited to this transport.
1482 bytes_acked
-= migrate_bytes
;
1484 /* 8.2. When an outstanding TSN is acknowledged,
1485 * the endpoint shall clear the error counter of
1486 * the destination transport address to which the
1487 * DATA chunk was last sent.
1488 * The association's overall error counter is
1491 transport
->error_count
= 0;
1492 transport
->asoc
->overall_error_count
= 0;
1493 forward_progress
= true;
1496 * While in SHUTDOWN PENDING, we may have started
1497 * the T5 shutdown guard timer after reaching the
1498 * retransmission limit. Stop that timer as soon
1499 * as the receiver acknowledged any data.
1501 if (asoc
->state
== SCTP_STATE_SHUTDOWN_PENDING
&&
1502 del_timer(&asoc
->timers
1503 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD
]))
1504 sctp_association_put(asoc
);
1506 /* Mark the destination transport address as
1507 * active if it is not so marked.
1509 if ((transport
->state
== SCTP_INACTIVE
||
1510 transport
->state
== SCTP_UNCONFIRMED
) &&
1511 sctp_cmp_addr_exact(&transport
->ipaddr
, saddr
)) {
1512 sctp_assoc_control_transport(
1516 SCTP_RECEIVED_SACK
);
1519 sctp_transport_raise_cwnd(transport
, sack_ctsn
,
1522 transport
->flight_size
-= bytes_acked
;
1523 if (transport
->flight_size
== 0)
1524 transport
->partial_bytes_acked
= 0;
1525 q
->outstanding_bytes
-= bytes_acked
+ migrate_bytes
;
1527 /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1528 * When a sender is doing zero window probing, it
1529 * should not timeout the association if it continues
1530 * to receive new packets from the receiver. The
1531 * reason is that the receiver MAY keep its window
1532 * closed for an indefinite time.
1533 * A sender is doing zero window probing when the
1534 * receiver's advertised window is zero, and there is
1535 * only one data chunk in flight to the receiver.
1537 * Allow the association to timeout while in SHUTDOWN
1538 * PENDING or SHUTDOWN RECEIVED in case the receiver
1539 * stays in zero window mode forever.
1541 if (!q
->asoc
->peer
.rwnd
&&
1542 !list_empty(&tlist
) &&
1543 (sack_ctsn
+2 == q
->asoc
->next_tsn
) &&
1544 q
->asoc
->state
< SCTP_STATE_SHUTDOWN_PENDING
) {
1545 pr_debug("%s: sack received for zero window "
1546 "probe:%u\n", __func__
, sack_ctsn
);
1548 q
->asoc
->overall_error_count
= 0;
1549 transport
->error_count
= 0;
1553 /* RFC 2960 6.3.2 Retransmission Timer Rules
1555 * R2) Whenever all outstanding data sent to an address have
1556 * been acknowledged, turn off the T3-rtx timer of that
1559 if (!transport
->flight_size
) {
1560 if (del_timer(&transport
->T3_rtx_timer
))
1561 sctp_transport_put(transport
);
1562 } else if (restart_timer
) {
1563 if (!mod_timer(&transport
->T3_rtx_timer
,
1564 jiffies
+ transport
->rto
))
1565 sctp_transport_hold(transport
);
1568 if (forward_progress
) {
1570 dst_confirm(transport
->dst
);
1574 list_splice(&tlist
, transmitted_queue
);
1577 /* Mark chunks as missing and consequently may get retransmitted. */
1578 static void sctp_mark_missing(struct sctp_outq
*q
,
1579 struct list_head
*transmitted_queue
,
1580 struct sctp_transport
*transport
,
1581 __u32 highest_new_tsn_in_sack
,
1582 int count_of_newacks
)
1584 struct sctp_chunk
*chunk
;
1586 char do_fast_retransmit
= 0;
1587 struct sctp_association
*asoc
= q
->asoc
;
1588 struct sctp_transport
*primary
= asoc
->peer
.primary_path
;
1590 list_for_each_entry(chunk
, transmitted_queue
, transmitted_list
) {
1592 tsn
= ntohl(chunk
->subh
.data_hdr
->tsn
);
1594 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1595 * 'Unacknowledged TSN's', if the TSN number of an
1596 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1597 * value, increment the 'TSN.Missing.Report' count on that
1598 * chunk if it has NOT been fast retransmitted or marked for
1599 * fast retransmit already.
1601 if (chunk
->fast_retransmit
== SCTP_CAN_FRTX
&&
1602 !chunk
->tsn_gap_acked
&&
1603 TSN_lt(tsn
, highest_new_tsn_in_sack
)) {
1605 /* SFR-CACC may require us to skip marking
1606 * this chunk as missing.
1608 if (!transport
|| !sctp_cacc_skip(primary
,
1610 count_of_newacks
, tsn
)) {
1611 chunk
->tsn_missing_report
++;
1613 pr_debug("%s: tsn:0x%x missing counter:%d\n",
1614 __func__
, tsn
, chunk
->tsn_missing_report
);
1618 * M4) If any DATA chunk is found to have a
1619 * 'TSN.Missing.Report'
1620 * value larger than or equal to 3, mark that chunk for
1621 * retransmission and start the fast retransmit procedure.
1624 if (chunk
->tsn_missing_report
>= 3) {
1625 chunk
->fast_retransmit
= SCTP_NEED_FRTX
;
1626 do_fast_retransmit
= 1;
1631 if (do_fast_retransmit
)
1632 sctp_retransmit(q
, transport
, SCTP_RTXR_FAST_RTX
);
1634 pr_debug("%s: transport:%p, cwnd:%d, ssthresh:%d, "
1635 "flight_size:%d, pba:%d\n", __func__
, transport
,
1636 transport
->cwnd
, transport
->ssthresh
,
1637 transport
->flight_size
, transport
->partial_bytes_acked
);
1641 /* Is the given TSN acked by this packet? */
1642 static int sctp_acked(struct sctp_sackhdr
*sack
, __u32 tsn
)
1645 sctp_sack_variable_t
*frags
;
1647 __u32 ctsn
= ntohl(sack
->cum_tsn_ack
);
1649 if (TSN_lte(tsn
, ctsn
))
1652 /* 3.3.4 Selective Acknowledgement (SACK) (3):
1655 * These fields contain the Gap Ack Blocks. They are repeated
1656 * for each Gap Ack Block up to the number of Gap Ack Blocks
1657 * defined in the Number of Gap Ack Blocks field. All DATA
1658 * chunks with TSNs greater than or equal to (Cumulative TSN
1659 * Ack + Gap Ack Block Start) and less than or equal to
1660 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1661 * Block are assumed to have been received correctly.
1664 frags
= sack
->variable
;
1666 for (i
= 0; i
< ntohs(sack
->num_gap_ack_blocks
); ++i
) {
1667 if (TSN_lte(ntohs(frags
[i
].gab
.start
), gap
) &&
1668 TSN_lte(gap
, ntohs(frags
[i
].gab
.end
)))
1677 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip
*skiplist
,
1678 int nskips
, __be16 stream
)
1682 for (i
= 0; i
< nskips
; i
++) {
1683 if (skiplist
[i
].stream
== stream
)
1689 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1690 static void sctp_generate_fwdtsn(struct sctp_outq
*q
, __u32 ctsn
)
1692 struct sctp_association
*asoc
= q
->asoc
;
1693 struct sctp_chunk
*ftsn_chunk
= NULL
;
1694 struct sctp_fwdtsn_skip ftsn_skip_arr
[10];
1698 struct sctp_chunk
*chunk
;
1699 struct list_head
*lchunk
, *temp
;
1701 if (!asoc
->peer
.prsctp_capable
)
1704 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1707 * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1708 * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1710 if (TSN_lt(asoc
->adv_peer_ack_point
, ctsn
))
1711 asoc
->adv_peer_ack_point
= ctsn
;
1713 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1714 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1715 * the chunk next in the out-queue space is marked as "abandoned" as
1716 * shown in the following example:
1718 * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1719 * and the Advanced.Peer.Ack.Point is updated to this value:
1721 * out-queue at the end of ==> out-queue after Adv.Ack.Point
1722 * normal SACK processing local advancement
1724 * Adv.Ack.Pt-> 102 acked 102 acked
1725 * 103 abandoned 103 abandoned
1726 * 104 abandoned Adv.Ack.P-> 104 abandoned
1728 * 106 acked 106 acked
1731 * In this example, the data sender successfully advanced the
1732 * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1734 list_for_each_safe(lchunk
, temp
, &q
->abandoned
) {
1735 chunk
= list_entry(lchunk
, struct sctp_chunk
,
1737 tsn
= ntohl(chunk
->subh
.data_hdr
->tsn
);
1739 /* Remove any chunks in the abandoned queue that are acked by
1742 if (TSN_lte(tsn
, ctsn
)) {
1743 list_del_init(lchunk
);
1744 sctp_chunk_free(chunk
);
1746 if (TSN_lte(tsn
, asoc
->adv_peer_ack_point
+1)) {
1747 asoc
->adv_peer_ack_point
= tsn
;
1748 if (chunk
->chunk_hdr
->flags
&
1749 SCTP_DATA_UNORDERED
)
1751 skip_pos
= sctp_get_skip_pos(&ftsn_skip_arr
[0],
1753 chunk
->subh
.data_hdr
->stream
);
1754 ftsn_skip_arr
[skip_pos
].stream
=
1755 chunk
->subh
.data_hdr
->stream
;
1756 ftsn_skip_arr
[skip_pos
].ssn
=
1757 chunk
->subh
.data_hdr
->ssn
;
1758 if (skip_pos
== nskips
)
1767 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1768 * is greater than the Cumulative TSN ACK carried in the received
1769 * SACK, the data sender MUST send the data receiver a FORWARD TSN
1770 * chunk containing the latest value of the
1771 * "Advanced.Peer.Ack.Point".
1773 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1774 * list each stream and sequence number in the forwarded TSN. This
1775 * information will enable the receiver to easily find any
1776 * stranded TSN's waiting on stream reorder queues. Each stream
1777 * SHOULD only be reported once; this means that if multiple
1778 * abandoned messages occur in the same stream then only the
1779 * highest abandoned stream sequence number is reported. If the
1780 * total size of the FORWARD TSN does NOT fit in a single MTU then
1781 * the sender of the FORWARD TSN SHOULD lower the
1782 * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1785 if (asoc
->adv_peer_ack_point
> ctsn
)
1786 ftsn_chunk
= sctp_make_fwdtsn(asoc
, asoc
->adv_peer_ack_point
,
1787 nskips
, &ftsn_skip_arr
[0]);
1790 list_add_tail(&ftsn_chunk
->list
, &q
->control_chunk_list
);
1791 SCTP_INC_STATS(sock_net(asoc
->base
.sk
), SCTP_MIB_OUTCTRLCHUNKS
);