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 void 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 void sctp_outq_tail(struct sctp_outq
*q
, struct sctp_chunk
*chunk
, gfp_t gfp
)
290 struct net
*net
= sock_net(q
->asoc
->base
.sk
);
292 pr_debug("%s: outq:%p, chunk:%p[%s]\n", __func__
, q
, chunk
,
293 chunk
&& chunk
->chunk_hdr
?
294 sctp_cname(SCTP_ST_CHUNK(chunk
->chunk_hdr
->type
)) :
297 /* If it is data, queue it up, otherwise, send it
300 if (sctp_chunk_is_data(chunk
)) {
301 pr_debug("%s: outqueueing: outq:%p, chunk:%p[%s])\n",
302 __func__
, q
, chunk
, chunk
&& chunk
->chunk_hdr
?
303 sctp_cname(SCTP_ST_CHUNK(chunk
->chunk_hdr
->type
)) :
306 sctp_outq_tail_data(q
, chunk
);
307 if (chunk
->asoc
->peer
.prsctp_capable
&&
308 SCTP_PR_PRIO_ENABLED(chunk
->sinfo
.sinfo_flags
))
309 chunk
->asoc
->sent_cnt_removable
++;
310 if (chunk
->chunk_hdr
->flags
& SCTP_DATA_UNORDERED
)
311 SCTP_INC_STATS(net
, SCTP_MIB_OUTUNORDERCHUNKS
);
313 SCTP_INC_STATS(net
, SCTP_MIB_OUTORDERCHUNKS
);
315 list_add_tail(&chunk
->list
, &q
->control_chunk_list
);
316 SCTP_INC_STATS(net
, SCTP_MIB_OUTCTRLCHUNKS
);
320 sctp_outq_flush(q
, 0, gfp
);
323 /* Insert a chunk into the sorted list based on the TSNs. The retransmit list
324 * and the abandoned list are in ascending order.
326 static void sctp_insert_list(struct list_head
*head
, struct list_head
*new)
328 struct list_head
*pos
;
329 struct sctp_chunk
*nchunk
, *lchunk
;
333 nchunk
= list_entry(new, struct sctp_chunk
, transmitted_list
);
334 ntsn
= ntohl(nchunk
->subh
.data_hdr
->tsn
);
336 list_for_each(pos
, head
) {
337 lchunk
= list_entry(pos
, struct sctp_chunk
, transmitted_list
);
338 ltsn
= ntohl(lchunk
->subh
.data_hdr
->tsn
);
339 if (TSN_lt(ntsn
, ltsn
)) {
340 list_add(new, pos
->prev
);
346 list_add_tail(new, head
);
349 static int sctp_prsctp_prune_sent(struct sctp_association
*asoc
,
350 struct sctp_sndrcvinfo
*sinfo
,
351 struct list_head
*queue
, int msg_len
)
353 struct sctp_chunk
*chk
, *temp
;
355 list_for_each_entry_safe(chk
, temp
, queue
, transmitted_list
) {
356 if (!SCTP_PR_PRIO_ENABLED(chk
->sinfo
.sinfo_flags
) ||
357 chk
->sinfo
.sinfo_timetolive
<= sinfo
->sinfo_timetolive
)
360 list_del_init(&chk
->transmitted_list
);
361 sctp_insert_list(&asoc
->outqueue
.abandoned
,
362 &chk
->transmitted_list
);
364 asoc
->sent_cnt_removable
--;
365 asoc
->abandoned_sent
[SCTP_PR_INDEX(PRIO
)]++;
367 if (!chk
->tsn_gap_acked
) {
369 chk
->transport
->flight_size
-=
371 asoc
->outqueue
.outstanding_bytes
-= sctp_data_size(chk
);
374 msg_len
-= SCTP_DATA_SNDSIZE(chk
) +
375 sizeof(struct sk_buff
) +
376 sizeof(struct sctp_chunk
);
384 static int sctp_prsctp_prune_unsent(struct sctp_association
*asoc
,
385 struct sctp_sndrcvinfo
*sinfo
,
386 struct list_head
*queue
, int msg_len
)
388 struct sctp_chunk
*chk
, *temp
;
390 list_for_each_entry_safe(chk
, temp
, queue
, list
) {
391 if (!SCTP_PR_PRIO_ENABLED(chk
->sinfo
.sinfo_flags
) ||
392 chk
->sinfo
.sinfo_timetolive
<= sinfo
->sinfo_timetolive
)
395 list_del_init(&chk
->list
);
396 asoc
->sent_cnt_removable
--;
397 asoc
->abandoned_unsent
[SCTP_PR_INDEX(PRIO
)]++;
399 msg_len
-= SCTP_DATA_SNDSIZE(chk
) +
400 sizeof(struct sk_buff
) +
401 sizeof(struct sctp_chunk
);
402 sctp_chunk_free(chk
);
410 /* Abandon the chunks according their priorities */
411 void sctp_prsctp_prune(struct sctp_association
*asoc
,
412 struct sctp_sndrcvinfo
*sinfo
, int msg_len
)
414 struct sctp_transport
*transport
;
416 if (!asoc
->peer
.prsctp_capable
|| !asoc
->sent_cnt_removable
)
419 msg_len
= sctp_prsctp_prune_sent(asoc
, sinfo
,
420 &asoc
->outqueue
.retransmit
,
425 list_for_each_entry(transport
, &asoc
->peer
.transport_addr_list
,
427 msg_len
= sctp_prsctp_prune_sent(asoc
, sinfo
,
428 &transport
->transmitted
,
434 sctp_prsctp_prune_unsent(asoc
, sinfo
,
435 &asoc
->outqueue
.out_chunk_list
,
439 /* Mark all the eligible packets on a transport for retransmission. */
440 void sctp_retransmit_mark(struct sctp_outq
*q
,
441 struct sctp_transport
*transport
,
444 struct list_head
*lchunk
, *ltemp
;
445 struct sctp_chunk
*chunk
;
447 /* Walk through the specified transmitted queue. */
448 list_for_each_safe(lchunk
, ltemp
, &transport
->transmitted
) {
449 chunk
= list_entry(lchunk
, struct sctp_chunk
,
452 /* If the chunk is abandoned, move it to abandoned list. */
453 if (sctp_chunk_abandoned(chunk
)) {
454 list_del_init(lchunk
);
455 sctp_insert_list(&q
->abandoned
, lchunk
);
457 /* If this chunk has not been previousely acked,
458 * stop considering it 'outstanding'. Our peer
459 * will most likely never see it since it will
460 * not be retransmitted
462 if (!chunk
->tsn_gap_acked
) {
463 if (chunk
->transport
)
464 chunk
->transport
->flight_size
-=
465 sctp_data_size(chunk
);
466 q
->outstanding_bytes
-= sctp_data_size(chunk
);
467 q
->asoc
->peer
.rwnd
+= sctp_data_size(chunk
);
472 /* If we are doing retransmission due to a timeout or pmtu
473 * discovery, only the chunks that are not yet acked should
474 * be added to the retransmit queue.
476 if ((reason
== SCTP_RTXR_FAST_RTX
&&
477 (chunk
->fast_retransmit
== SCTP_NEED_FRTX
)) ||
478 (reason
!= SCTP_RTXR_FAST_RTX
&& !chunk
->tsn_gap_acked
)) {
479 /* RFC 2960 6.2.1 Processing a Received SACK
481 * C) Any time a DATA chunk is marked for
482 * retransmission (via either T3-rtx timer expiration
483 * (Section 6.3.3) or via fast retransmit
484 * (Section 7.2.4)), add the data size of those
485 * chunks to the rwnd.
487 q
->asoc
->peer
.rwnd
+= sctp_data_size(chunk
);
488 q
->outstanding_bytes
-= sctp_data_size(chunk
);
489 if (chunk
->transport
)
490 transport
->flight_size
-= sctp_data_size(chunk
);
492 /* sctpimpguide-05 Section 2.8.2
493 * M5) If a T3-rtx timer expires, the
494 * 'TSN.Missing.Report' of all affected TSNs is set
497 chunk
->tsn_missing_report
= 0;
499 /* If a chunk that is being used for RTT measurement
500 * has to be retransmitted, we cannot use this chunk
501 * anymore for RTT measurements. Reset rto_pending so
502 * that a new RTT measurement is started when a new
503 * data chunk is sent.
505 if (chunk
->rtt_in_progress
) {
506 chunk
->rtt_in_progress
= 0;
507 transport
->rto_pending
= 0;
510 /* Move the chunk to the retransmit queue. The chunks
511 * on the retransmit queue are always kept in order.
513 list_del_init(lchunk
);
514 sctp_insert_list(&q
->retransmit
, lchunk
);
518 pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d, "
519 "flight_size:%d, pba:%d\n", __func__
, transport
, reason
,
520 transport
->cwnd
, transport
->ssthresh
, transport
->flight_size
,
521 transport
->partial_bytes_acked
);
524 /* Mark all the eligible packets on a transport for retransmission and force
527 void sctp_retransmit(struct sctp_outq
*q
, struct sctp_transport
*transport
,
528 sctp_retransmit_reason_t reason
)
530 struct net
*net
= sock_net(q
->asoc
->base
.sk
);
533 case SCTP_RTXR_T3_RTX
:
534 SCTP_INC_STATS(net
, SCTP_MIB_T3_RETRANSMITS
);
535 sctp_transport_lower_cwnd(transport
, SCTP_LOWER_CWND_T3_RTX
);
536 /* Update the retran path if the T3-rtx timer has expired for
537 * the current retran path.
539 if (transport
== transport
->asoc
->peer
.retran_path
)
540 sctp_assoc_update_retran_path(transport
->asoc
);
541 transport
->asoc
->rtx_data_chunks
+=
542 transport
->asoc
->unack_data
;
544 case SCTP_RTXR_FAST_RTX
:
545 SCTP_INC_STATS(net
, SCTP_MIB_FAST_RETRANSMITS
);
546 sctp_transport_lower_cwnd(transport
, SCTP_LOWER_CWND_FAST_RTX
);
549 case SCTP_RTXR_PMTUD
:
550 SCTP_INC_STATS(net
, SCTP_MIB_PMTUD_RETRANSMITS
);
552 case SCTP_RTXR_T1_RTX
:
553 SCTP_INC_STATS(net
, SCTP_MIB_T1_RETRANSMITS
);
554 transport
->asoc
->init_retries
++;
560 sctp_retransmit_mark(q
, transport
, reason
);
562 /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
563 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
564 * following the procedures outlined in C1 - C5.
566 if (reason
== SCTP_RTXR_T3_RTX
)
567 sctp_generate_fwdtsn(q
, q
->asoc
->ctsn_ack_point
);
569 /* Flush the queues only on timeout, since fast_rtx is only
570 * triggered during sack processing and the queue
571 * will be flushed at the end.
573 if (reason
!= SCTP_RTXR_FAST_RTX
)
574 sctp_outq_flush(q
, /* rtx_timeout */ 1, GFP_ATOMIC
);
578 * Transmit DATA chunks on the retransmit queue. Upon return from
579 * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
580 * need to be transmitted by the caller.
581 * We assume that pkt->transport has already been set.
583 * The return value is a normal kernel error return value.
585 static int sctp_outq_flush_rtx(struct sctp_outq
*q
, struct sctp_packet
*pkt
,
586 int rtx_timeout
, int *start_timer
)
588 struct list_head
*lqueue
;
589 struct sctp_transport
*transport
= pkt
->transport
;
591 struct sctp_chunk
*chunk
, *chunk1
;
597 lqueue
= &q
->retransmit
;
598 fast_rtx
= q
->fast_rtx
;
600 /* This loop handles time-out retransmissions, fast retransmissions,
601 * and retransmissions due to opening of whindow.
603 * RFC 2960 6.3.3 Handle T3-rtx Expiration
605 * E3) Determine how many of the earliest (i.e., lowest TSN)
606 * outstanding DATA chunks for the address for which the
607 * T3-rtx has expired will fit into a single packet, subject
608 * to the MTU constraint for the path corresponding to the
609 * destination transport address to which the retransmission
610 * is being sent (this may be different from the address for
611 * which the timer expires [see Section 6.4]). Call this value
612 * K. Bundle and retransmit those K DATA chunks in a single
613 * packet to the destination endpoint.
615 * [Just to be painfully clear, if we are retransmitting
616 * because a timeout just happened, we should send only ONE
617 * packet of retransmitted data.]
619 * For fast retransmissions we also send only ONE packet. However,
620 * if we are just flushing the queue due to open window, we'll
621 * try to send as much as possible.
623 list_for_each_entry_safe(chunk
, chunk1
, lqueue
, transmitted_list
) {
624 /* If the chunk is abandoned, move it to abandoned list. */
625 if (sctp_chunk_abandoned(chunk
)) {
626 list_del_init(&chunk
->transmitted_list
);
627 sctp_insert_list(&q
->abandoned
,
628 &chunk
->transmitted_list
);
632 /* Make sure that Gap Acked TSNs are not retransmitted. A
633 * simple approach is just to move such TSNs out of the
634 * way and into a 'transmitted' queue and skip to the
637 if (chunk
->tsn_gap_acked
) {
638 list_move_tail(&chunk
->transmitted_list
,
639 &transport
->transmitted
);
643 /* If we are doing fast retransmit, ignore non-fast_rtransmit
646 if (fast_rtx
&& !chunk
->fast_retransmit
)
650 /* Attempt to append this chunk to the packet. */
651 status
= sctp_packet_append_chunk(pkt
, chunk
);
654 case SCTP_XMIT_PMTU_FULL
:
655 if (!pkt
->has_data
&& !pkt
->has_cookie_echo
) {
656 /* If this packet did not contain DATA then
657 * retransmission did not happen, so do it
658 * again. We'll ignore the error here since
659 * control chunks are already freed so there
660 * is nothing we can do.
662 sctp_packet_transmit(pkt
, GFP_ATOMIC
);
666 /* Send this packet. */
667 error
= sctp_packet_transmit(pkt
, GFP_ATOMIC
);
669 /* If we are retransmitting, we should only
670 * send a single packet.
671 * Otherwise, try appending this chunk again.
673 if (rtx_timeout
|| fast_rtx
)
678 /* Bundle next chunk in the next round. */
681 case SCTP_XMIT_RWND_FULL
:
682 /* Send this packet. */
683 error
= sctp_packet_transmit(pkt
, GFP_ATOMIC
);
685 /* Stop sending DATA as there is no more room
691 case SCTP_XMIT_DELAY
:
692 /* Send this packet. */
693 error
= sctp_packet_transmit(pkt
, GFP_ATOMIC
);
695 /* Stop sending DATA because of nagle delay. */
700 /* The append was successful, so add this chunk to
701 * the transmitted list.
703 list_move_tail(&chunk
->transmitted_list
,
704 &transport
->transmitted
);
706 /* Mark the chunk as ineligible for fast retransmit
707 * after it is retransmitted.
709 if (chunk
->fast_retransmit
== SCTP_NEED_FRTX
)
710 chunk
->fast_retransmit
= SCTP_DONT_FRTX
;
712 q
->asoc
->stats
.rtxchunks
++;
716 /* Set the timer if there were no errors */
717 if (!error
&& !timer
)
724 /* If we are here due to a retransmit timeout or a fast
725 * retransmit and if there are any chunks left in the retransmit
726 * queue that could not fit in the PMTU sized packet, they need
727 * to be marked as ineligible for a subsequent fast retransmit.
729 if (rtx_timeout
|| fast_rtx
) {
730 list_for_each_entry(chunk1
, lqueue
, transmitted_list
) {
731 if (chunk1
->fast_retransmit
== SCTP_NEED_FRTX
)
732 chunk1
->fast_retransmit
= SCTP_DONT_FRTX
;
736 *start_timer
= timer
;
738 /* Clear fast retransmit hint */
745 /* Cork the outqueue so queued chunks are really queued. */
746 void sctp_outq_uncork(struct sctp_outq
*q
, gfp_t gfp
)
751 sctp_outq_flush(q
, 0, gfp
);
756 * Try to flush an outqueue.
758 * Description: Send everything in q which we legally can, subject to
759 * congestion limitations.
760 * * Note: This function can be called from multiple contexts so appropriate
761 * locking concerns must be made. Today we use the sock lock to protect
764 static void sctp_outq_flush(struct sctp_outq
*q
, int rtx_timeout
, gfp_t gfp
)
766 struct sctp_packet
*packet
;
767 struct sctp_packet singleton
;
768 struct sctp_association
*asoc
= q
->asoc
;
769 __u16 sport
= asoc
->base
.bind_addr
.port
;
770 __u16 dport
= asoc
->peer
.port
;
771 __u32 vtag
= asoc
->peer
.i
.init_tag
;
772 struct sctp_transport
*transport
= NULL
;
773 struct sctp_transport
*new_transport
;
774 struct sctp_chunk
*chunk
, *tmp
;
780 /* These transports have chunks to send. */
781 struct list_head transport_list
;
782 struct list_head
*ltransport
;
784 INIT_LIST_HEAD(&transport_list
);
790 * When bundling control chunks with DATA chunks, an
791 * endpoint MUST place control chunks first in the outbound
792 * SCTP packet. The transmitter MUST transmit DATA chunks
793 * within a SCTP packet in increasing order of TSN.
797 list_for_each_entry_safe(chunk
, tmp
, &q
->control_chunk_list
, list
) {
799 * F1) This means that until such time as the ASCONF
800 * containing the add is acknowledged, the sender MUST
801 * NOT use the new IP address as a source for ANY SCTP
802 * packet except on carrying an ASCONF Chunk.
804 if (asoc
->src_out_of_asoc_ok
&&
805 chunk
->chunk_hdr
->type
!= SCTP_CID_ASCONF
)
808 list_del_init(&chunk
->list
);
810 /* Pick the right transport to use. */
811 new_transport
= chunk
->transport
;
813 if (!new_transport
) {
815 * If we have a prior transport pointer, see if
816 * the destination address of the chunk
817 * matches the destination address of the
818 * current transport. If not a match, then
819 * try to look up the transport with a given
820 * destination address. We do this because
821 * after processing ASCONFs, we may have new
822 * transports created.
825 sctp_cmp_addr_exact(&chunk
->dest
,
827 new_transport
= transport
;
829 new_transport
= sctp_assoc_lookup_paddr(asoc
,
832 /* if we still don't have a new transport, then
833 * use the current active path.
836 new_transport
= asoc
->peer
.active_path
;
837 } else if ((new_transport
->state
== SCTP_INACTIVE
) ||
838 (new_transport
->state
== SCTP_UNCONFIRMED
) ||
839 (new_transport
->state
== SCTP_PF
)) {
840 /* If the chunk is Heartbeat or Heartbeat Ack,
841 * send it to chunk->transport, even if it's
844 * 3.3.6 Heartbeat Acknowledgement:
846 * A HEARTBEAT ACK is always sent to the source IP
847 * address of the IP datagram containing the
848 * HEARTBEAT chunk to which this ack is responding.
851 * ASCONF_ACKs also must be sent to the source.
853 if (chunk
->chunk_hdr
->type
!= SCTP_CID_HEARTBEAT
&&
854 chunk
->chunk_hdr
->type
!= SCTP_CID_HEARTBEAT_ACK
&&
855 chunk
->chunk_hdr
->type
!= SCTP_CID_ASCONF_ACK
)
856 new_transport
= asoc
->peer
.active_path
;
859 /* Are we switching transports?
860 * Take care of transport locks.
862 if (new_transport
!= transport
) {
863 transport
= new_transport
;
864 if (list_empty(&transport
->send_ready
)) {
865 list_add_tail(&transport
->send_ready
,
868 packet
= &transport
->packet
;
869 sctp_packet_config(packet
, vtag
,
870 asoc
->peer
.ecn_capable
);
873 switch (chunk
->chunk_hdr
->type
) {
877 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
878 * COMPLETE with any other chunks. [Send them immediately.]
881 case SCTP_CID_INIT_ACK
:
882 case SCTP_CID_SHUTDOWN_COMPLETE
:
883 sctp_packet_init(&singleton
, transport
, sport
, dport
);
884 sctp_packet_config(&singleton
, vtag
, 0);
885 sctp_packet_append_chunk(&singleton
, chunk
);
886 error
= sctp_packet_transmit(&singleton
, gfp
);
888 asoc
->base
.sk
->sk_err
= -error
;
894 if (sctp_test_T_bit(chunk
)) {
895 packet
->vtag
= asoc
->c
.my_vtag
;
897 /* The following chunks are "response" chunks, i.e.
898 * they are generated in response to something we
899 * received. If we are sending these, then we can
900 * send only 1 packet containing these chunks.
902 case SCTP_CID_HEARTBEAT_ACK
:
903 case SCTP_CID_SHUTDOWN_ACK
:
904 case SCTP_CID_COOKIE_ACK
:
905 case SCTP_CID_COOKIE_ECHO
:
907 case SCTP_CID_ECN_CWR
:
908 case SCTP_CID_ASCONF_ACK
:
913 case SCTP_CID_HEARTBEAT
:
914 case SCTP_CID_SHUTDOWN
:
915 case SCTP_CID_ECN_ECNE
:
916 case SCTP_CID_ASCONF
:
917 case SCTP_CID_FWD_TSN
:
918 status
= sctp_packet_transmit_chunk(packet
, chunk
,
920 if (status
!= SCTP_XMIT_OK
) {
921 /* put the chunk back */
922 list_add(&chunk
->list
, &q
->control_chunk_list
);
924 asoc
->stats
.octrlchunks
++;
925 /* PR-SCTP C5) If a FORWARD TSN is sent, the
926 * sender MUST assure that at least one T3-rtx
929 if (chunk
->chunk_hdr
->type
== SCTP_CID_FWD_TSN
) {
930 sctp_transport_reset_t3_rtx(transport
);
931 transport
->last_time_sent
= jiffies
;
937 /* We built a chunk with an illegal type! */
942 if (q
->asoc
->src_out_of_asoc_ok
)
945 /* Is it OK to send data chunks? */
946 switch (asoc
->state
) {
947 case SCTP_STATE_COOKIE_ECHOED
:
948 /* Only allow bundling when this packet has a COOKIE-ECHO
951 if (!packet
|| !packet
->has_cookie_echo
)
955 case SCTP_STATE_ESTABLISHED
:
956 case SCTP_STATE_SHUTDOWN_PENDING
:
957 case SCTP_STATE_SHUTDOWN_RECEIVED
:
959 * RFC 2960 6.1 Transmission of DATA Chunks
961 * C) When the time comes for the sender to transmit,
962 * before sending new DATA chunks, the sender MUST
963 * first transmit any outstanding DATA chunks which
964 * are marked for retransmission (limited by the
967 if (!list_empty(&q
->retransmit
)) {
968 if (asoc
->peer
.retran_path
->state
== SCTP_UNCONFIRMED
)
970 if (transport
== asoc
->peer
.retran_path
)
973 /* Switch transports & prepare the packet. */
975 transport
= asoc
->peer
.retran_path
;
977 if (list_empty(&transport
->send_ready
)) {
978 list_add_tail(&transport
->send_ready
,
982 packet
= &transport
->packet
;
983 sctp_packet_config(packet
, vtag
,
984 asoc
->peer
.ecn_capable
);
986 error
= sctp_outq_flush_rtx(q
, packet
,
987 rtx_timeout
, &start_timer
);
989 asoc
->base
.sk
->sk_err
= -error
;
992 sctp_transport_reset_t3_rtx(transport
);
993 transport
->last_time_sent
= jiffies
;
996 /* This can happen on COOKIE-ECHO resend. Only
997 * one chunk can get bundled with a COOKIE-ECHO.
999 if (packet
->has_cookie_echo
)
1000 goto sctp_flush_out
;
1002 /* Don't send new data if there is still data
1003 * waiting to retransmit.
1005 if (!list_empty(&q
->retransmit
))
1006 goto sctp_flush_out
;
1009 /* Apply Max.Burst limitation to the current transport in
1010 * case it will be used for new data. We are going to
1011 * rest it before we return, but we want to apply the limit
1012 * to the currently queued data.
1015 sctp_transport_burst_limited(transport
);
1017 /* Finally, transmit new packets. */
1018 while ((chunk
= sctp_outq_dequeue_data(q
)) != NULL
) {
1019 /* RFC 2960 6.5 Every DATA chunk MUST carry a valid
1020 * stream identifier.
1022 if (chunk
->sinfo
.sinfo_stream
>=
1023 asoc
->c
.sinit_num_ostreams
) {
1025 /* Mark as failed send. */
1026 sctp_chunk_fail(chunk
, SCTP_ERROR_INV_STRM
);
1027 if (asoc
->peer
.prsctp_capable
&&
1028 SCTP_PR_PRIO_ENABLED(chunk
->sinfo
.sinfo_flags
))
1029 asoc
->sent_cnt_removable
--;
1030 sctp_chunk_free(chunk
);
1034 /* Has this chunk expired? */
1035 if (sctp_chunk_abandoned(chunk
)) {
1036 sctp_chunk_fail(chunk
, 0);
1037 sctp_chunk_free(chunk
);
1041 /* If there is a specified transport, use it.
1042 * Otherwise, we want to use the active path.
1044 new_transport
= chunk
->transport
;
1045 if (!new_transport
||
1046 ((new_transport
->state
== SCTP_INACTIVE
) ||
1047 (new_transport
->state
== SCTP_UNCONFIRMED
) ||
1048 (new_transport
->state
== SCTP_PF
)))
1049 new_transport
= asoc
->peer
.active_path
;
1050 if (new_transport
->state
== SCTP_UNCONFIRMED
) {
1051 WARN_ONCE(1, "Atempt to send packet on unconfirmed path.");
1052 sctp_chunk_fail(chunk
, 0);
1053 sctp_chunk_free(chunk
);
1057 /* Change packets if necessary. */
1058 if (new_transport
!= transport
) {
1059 transport
= new_transport
;
1061 /* Schedule to have this transport's
1064 if (list_empty(&transport
->send_ready
)) {
1065 list_add_tail(&transport
->send_ready
,
1069 packet
= &transport
->packet
;
1070 sctp_packet_config(packet
, vtag
,
1071 asoc
->peer
.ecn_capable
);
1072 /* We've switched transports, so apply the
1073 * Burst limit to the new transport.
1075 sctp_transport_burst_limited(transport
);
1078 pr_debug("%s: outq:%p, chunk:%p[%s], tx-tsn:0x%x skb->head:%p "
1080 __func__
, q
, chunk
, chunk
&& chunk
->chunk_hdr
?
1081 sctp_cname(SCTP_ST_CHUNK(chunk
->chunk_hdr
->type
)) :
1082 "illegal chunk", ntohl(chunk
->subh
.data_hdr
->tsn
),
1083 chunk
->skb
? chunk
->skb
->head
: NULL
, chunk
->skb
?
1084 atomic_read(&chunk
->skb
->users
) : -1);
1086 /* Add the chunk to the packet. */
1087 status
= sctp_packet_transmit_chunk(packet
, chunk
, 0, gfp
);
1090 case SCTP_XMIT_PMTU_FULL
:
1091 case SCTP_XMIT_RWND_FULL
:
1092 case SCTP_XMIT_DELAY
:
1093 /* We could not append this chunk, so put
1094 * the chunk back on the output queue.
1096 pr_debug("%s: could not transmit tsn:0x%x, status:%d\n",
1097 __func__
, ntohl(chunk
->subh
.data_hdr
->tsn
),
1100 sctp_outq_head_data(q
, chunk
);
1101 goto sctp_flush_out
;
1104 /* The sender is in the SHUTDOWN-PENDING state,
1105 * The sender MAY set the I-bit in the DATA
1108 if (asoc
->state
== SCTP_STATE_SHUTDOWN_PENDING
)
1109 chunk
->chunk_hdr
->flags
|= SCTP_DATA_SACK_IMM
;
1110 if (chunk
->chunk_hdr
->flags
& SCTP_DATA_UNORDERED
)
1111 asoc
->stats
.ouodchunks
++;
1113 asoc
->stats
.oodchunks
++;
1121 /* BUG: We assume that the sctp_packet_transmit()
1122 * call below will succeed all the time and add the
1123 * chunk to the transmitted list and restart the
1125 * It is possible that the call can fail under OOM
1128 * Is this really a problem? Won't this behave
1131 list_add_tail(&chunk
->transmitted_list
,
1132 &transport
->transmitted
);
1134 sctp_transport_reset_t3_rtx(transport
);
1135 transport
->last_time_sent
= jiffies
;
1137 /* Only let one DATA chunk get bundled with a
1138 * COOKIE-ECHO chunk.
1140 if (packet
->has_cookie_echo
)
1141 goto sctp_flush_out
;
1152 /* Before returning, examine all the transports touched in
1153 * this call. Right now, we bluntly force clear all the
1154 * transports. Things might change after we implement Nagle.
1155 * But such an examination is still required.
1159 while ((ltransport
= sctp_list_dequeue(&transport_list
)) != NULL
) {
1160 struct sctp_transport
*t
= list_entry(ltransport
,
1161 struct sctp_transport
,
1163 packet
= &t
->packet
;
1164 if (!sctp_packet_empty(packet
)) {
1165 error
= sctp_packet_transmit(packet
, gfp
);
1167 asoc
->base
.sk
->sk_err
= -error
;
1170 /* Clear the burst limited state, if any */
1171 sctp_transport_burst_reset(t
);
1175 /* Update unack_data based on the incoming SACK chunk */
1176 static void sctp_sack_update_unack_data(struct sctp_association
*assoc
,
1177 struct sctp_sackhdr
*sack
)
1179 sctp_sack_variable_t
*frags
;
1183 unack_data
= assoc
->next_tsn
- assoc
->ctsn_ack_point
- 1;
1185 frags
= sack
->variable
;
1186 for (i
= 0; i
< ntohs(sack
->num_gap_ack_blocks
); i
++) {
1187 unack_data
-= ((ntohs(frags
[i
].gab
.end
) -
1188 ntohs(frags
[i
].gab
.start
) + 1));
1191 assoc
->unack_data
= unack_data
;
1194 /* This is where we REALLY process a SACK.
1196 * Process the SACK against the outqueue. Mostly, this just frees
1197 * things off the transmitted queue.
1199 int sctp_outq_sack(struct sctp_outq
*q
, struct sctp_chunk
*chunk
)
1201 struct sctp_association
*asoc
= q
->asoc
;
1202 struct sctp_sackhdr
*sack
= chunk
->subh
.sack_hdr
;
1203 struct sctp_transport
*transport
;
1204 struct sctp_chunk
*tchunk
= NULL
;
1205 struct list_head
*lchunk
, *transport_list
, *temp
;
1206 sctp_sack_variable_t
*frags
= sack
->variable
;
1207 __u32 sack_ctsn
, ctsn
, tsn
;
1208 __u32 highest_tsn
, highest_new_tsn
;
1210 unsigned int outstanding
;
1211 struct sctp_transport
*primary
= asoc
->peer
.primary_path
;
1212 int count_of_newacks
= 0;
1216 /* Grab the association's destination address list. */
1217 transport_list
= &asoc
->peer
.transport_addr_list
;
1219 sack_ctsn
= ntohl(sack
->cum_tsn_ack
);
1220 gap_ack_blocks
= ntohs(sack
->num_gap_ack_blocks
);
1221 asoc
->stats
.gapcnt
+= gap_ack_blocks
;
1223 * SFR-CACC algorithm:
1224 * On receipt of a SACK the sender SHOULD execute the
1225 * following statements.
1227 * 1) If the cumulative ack in the SACK passes next tsn_at_change
1228 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1229 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1231 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1232 * is set the receiver of the SACK MUST take the following actions:
1234 * A) Initialize the cacc_saw_newack to 0 for all destination
1237 * Only bother if changeover_active is set. Otherwise, this is
1238 * totally suboptimal to do on every SACK.
1240 if (primary
->cacc
.changeover_active
) {
1241 u8 clear_cycling
= 0;
1243 if (TSN_lte(primary
->cacc
.next_tsn_at_change
, sack_ctsn
)) {
1244 primary
->cacc
.changeover_active
= 0;
1248 if (clear_cycling
|| gap_ack_blocks
) {
1249 list_for_each_entry(transport
, transport_list
,
1252 transport
->cacc
.cycling_changeover
= 0;
1254 transport
->cacc
.cacc_saw_newack
= 0;
1259 /* Get the highest TSN in the sack. */
1260 highest_tsn
= sack_ctsn
;
1262 highest_tsn
+= ntohs(frags
[gap_ack_blocks
- 1].gab
.end
);
1264 if (TSN_lt(asoc
->highest_sacked
, highest_tsn
))
1265 asoc
->highest_sacked
= highest_tsn
;
1267 highest_new_tsn
= sack_ctsn
;
1269 /* Run through the retransmit queue. Credit bytes received
1270 * and free those chunks that we can.
1272 sctp_check_transmitted(q
, &q
->retransmit
, NULL
, NULL
, sack
, &highest_new_tsn
);
1274 /* Run through the transmitted queue.
1275 * Credit bytes received and free those chunks which we can.
1277 * This is a MASSIVE candidate for optimization.
1279 list_for_each_entry(transport
, transport_list
, transports
) {
1280 sctp_check_transmitted(q
, &transport
->transmitted
,
1281 transport
, &chunk
->source
, sack
,
1284 * SFR-CACC algorithm:
1285 * C) Let count_of_newacks be the number of
1286 * destinations for which cacc_saw_newack is set.
1288 if (transport
->cacc
.cacc_saw_newack
)
1292 /* Move the Cumulative TSN Ack Point if appropriate. */
1293 if (TSN_lt(asoc
->ctsn_ack_point
, sack_ctsn
)) {
1294 asoc
->ctsn_ack_point
= sack_ctsn
;
1298 if (gap_ack_blocks
) {
1300 if (asoc
->fast_recovery
&& accum_moved
)
1301 highest_new_tsn
= highest_tsn
;
1303 list_for_each_entry(transport
, transport_list
, transports
)
1304 sctp_mark_missing(q
, &transport
->transmitted
, transport
,
1305 highest_new_tsn
, count_of_newacks
);
1308 /* Update unack_data field in the assoc. */
1309 sctp_sack_update_unack_data(asoc
, sack
);
1311 ctsn
= asoc
->ctsn_ack_point
;
1313 /* Throw away stuff rotting on the sack queue. */
1314 list_for_each_safe(lchunk
, temp
, &q
->sacked
) {
1315 tchunk
= list_entry(lchunk
, struct sctp_chunk
,
1317 tsn
= ntohl(tchunk
->subh
.data_hdr
->tsn
);
1318 if (TSN_lte(tsn
, ctsn
)) {
1319 list_del_init(&tchunk
->transmitted_list
);
1320 if (asoc
->peer
.prsctp_capable
&&
1321 SCTP_PR_PRIO_ENABLED(chunk
->sinfo
.sinfo_flags
))
1322 asoc
->sent_cnt_removable
--;
1323 sctp_chunk_free(tchunk
);
1327 /* ii) Set rwnd equal to the newly received a_rwnd minus the
1328 * number of bytes still outstanding after processing the
1329 * Cumulative TSN Ack and the Gap Ack Blocks.
1332 sack_a_rwnd
= ntohl(sack
->a_rwnd
);
1333 asoc
->peer
.zero_window_announced
= !sack_a_rwnd
;
1334 outstanding
= q
->outstanding_bytes
;
1336 if (outstanding
< sack_a_rwnd
)
1337 sack_a_rwnd
-= outstanding
;
1341 asoc
->peer
.rwnd
= sack_a_rwnd
;
1343 sctp_generate_fwdtsn(q
, sack_ctsn
);
1345 pr_debug("%s: sack cumulative tsn ack:0x%x\n", __func__
, sack_ctsn
);
1346 pr_debug("%s: cumulative tsn ack of assoc:%p is 0x%x, "
1347 "advertised peer ack point:0x%x\n", __func__
, asoc
, ctsn
,
1348 asoc
->adv_peer_ack_point
);
1350 return sctp_outq_is_empty(q
);
1353 /* Is the outqueue empty?
1354 * The queue is empty when we have not pending data, no in-flight data
1355 * and nothing pending retransmissions.
1357 int sctp_outq_is_empty(const struct sctp_outq
*q
)
1359 return q
->out_qlen
== 0 && q
->outstanding_bytes
== 0 &&
1360 list_empty(&q
->retransmit
);
1363 /********************************************************************
1364 * 2nd Level Abstractions
1365 ********************************************************************/
1367 /* Go through a transport's transmitted list or the association's retransmit
1368 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1369 * The retransmit list will not have an associated transport.
1371 * I added coherent debug information output. --xguo
1373 * Instead of printing 'sacked' or 'kept' for each TSN on the
1374 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1375 * KEPT TSN6-TSN7, etc.
1377 static void sctp_check_transmitted(struct sctp_outq
*q
,
1378 struct list_head
*transmitted_queue
,
1379 struct sctp_transport
*transport
,
1380 union sctp_addr
*saddr
,
1381 struct sctp_sackhdr
*sack
,
1382 __u32
*highest_new_tsn_in_sack
)
1384 struct list_head
*lchunk
;
1385 struct sctp_chunk
*tchunk
;
1386 struct list_head tlist
;
1390 __u8 restart_timer
= 0;
1391 int bytes_acked
= 0;
1392 int migrate_bytes
= 0;
1393 bool forward_progress
= false;
1395 sack_ctsn
= ntohl(sack
->cum_tsn_ack
);
1397 INIT_LIST_HEAD(&tlist
);
1399 /* The while loop will skip empty transmitted queues. */
1400 while (NULL
!= (lchunk
= sctp_list_dequeue(transmitted_queue
))) {
1401 tchunk
= list_entry(lchunk
, struct sctp_chunk
,
1404 if (sctp_chunk_abandoned(tchunk
)) {
1405 /* Move the chunk to abandoned list. */
1406 sctp_insert_list(&q
->abandoned
, lchunk
);
1408 /* If this chunk has not been acked, stop
1409 * considering it as 'outstanding'.
1411 if (!tchunk
->tsn_gap_acked
) {
1412 if (tchunk
->transport
)
1413 tchunk
->transport
->flight_size
-=
1414 sctp_data_size(tchunk
);
1415 q
->outstanding_bytes
-= sctp_data_size(tchunk
);
1420 tsn
= ntohl(tchunk
->subh
.data_hdr
->tsn
);
1421 if (sctp_acked(sack
, tsn
)) {
1422 /* If this queue is the retransmit queue, the
1423 * retransmit timer has already reclaimed
1424 * the outstanding bytes for this chunk, so only
1425 * count bytes associated with a transport.
1428 /* If this chunk is being used for RTT
1429 * measurement, calculate the RTT and update
1430 * the RTO using this value.
1432 * 6.3.1 C5) Karn's algorithm: RTT measurements
1433 * MUST NOT be made using packets that were
1434 * retransmitted (and thus for which it is
1435 * ambiguous whether the reply was for the
1436 * first instance of the packet or a later
1439 if (!tchunk
->tsn_gap_acked
&&
1440 !sctp_chunk_retransmitted(tchunk
) &&
1441 tchunk
->rtt_in_progress
) {
1442 tchunk
->rtt_in_progress
= 0;
1443 rtt
= jiffies
- tchunk
->sent_at
;
1444 sctp_transport_update_rto(transport
,
1449 /* If the chunk hasn't been marked as ACKED,
1450 * mark it and account bytes_acked if the
1451 * chunk had a valid transport (it will not
1452 * have a transport if ASCONF had deleted it
1453 * while DATA was outstanding).
1455 if (!tchunk
->tsn_gap_acked
) {
1456 tchunk
->tsn_gap_acked
= 1;
1457 if (TSN_lt(*highest_new_tsn_in_sack
, tsn
))
1458 *highest_new_tsn_in_sack
= tsn
;
1459 bytes_acked
+= sctp_data_size(tchunk
);
1460 if (!tchunk
->transport
)
1461 migrate_bytes
+= sctp_data_size(tchunk
);
1462 forward_progress
= true;
1465 if (TSN_lte(tsn
, sack_ctsn
)) {
1466 /* RFC 2960 6.3.2 Retransmission Timer Rules
1468 * R3) Whenever a SACK is received
1469 * that acknowledges the DATA chunk
1470 * with the earliest outstanding TSN
1471 * for that address, restart T3-rtx
1472 * timer for that address with its
1476 forward_progress
= true;
1478 if (!tchunk
->tsn_gap_acked
) {
1480 * SFR-CACC algorithm:
1481 * 2) If the SACK contains gap acks
1482 * and the flag CHANGEOVER_ACTIVE is
1483 * set the receiver of the SACK MUST
1484 * take the following action:
1486 * B) For each TSN t being acked that
1487 * has not been acked in any SACK so
1488 * far, set cacc_saw_newack to 1 for
1489 * the destination that the TSN was
1493 sack
->num_gap_ack_blocks
&&
1494 q
->asoc
->peer
.primary_path
->cacc
.
1496 transport
->cacc
.cacc_saw_newack
1500 list_add_tail(&tchunk
->transmitted_list
,
1503 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1504 * M2) Each time a SACK arrives reporting
1505 * 'Stray DATA chunk(s)' record the highest TSN
1506 * reported as newly acknowledged, call this
1507 * value 'HighestTSNinSack'. A newly
1508 * acknowledged DATA chunk is one not
1509 * previously acknowledged in a SACK.
1511 * When the SCTP sender of data receives a SACK
1512 * chunk that acknowledges, for the first time,
1513 * the receipt of a DATA chunk, all the still
1514 * unacknowledged DATA chunks whose TSN is
1515 * older than that newly acknowledged DATA
1516 * chunk, are qualified as 'Stray DATA chunks'.
1518 list_add_tail(lchunk
, &tlist
);
1521 if (tchunk
->tsn_gap_acked
) {
1522 pr_debug("%s: receiver reneged on data TSN:0x%x\n",
1525 tchunk
->tsn_gap_acked
= 0;
1527 if (tchunk
->transport
)
1528 bytes_acked
-= sctp_data_size(tchunk
);
1530 /* RFC 2960 6.3.2 Retransmission Timer Rules
1532 * R4) Whenever a SACK is received missing a
1533 * TSN that was previously acknowledged via a
1534 * Gap Ack Block, start T3-rtx for the
1535 * destination address to which the DATA
1536 * chunk was originally
1537 * transmitted if it is not already running.
1542 list_add_tail(lchunk
, &tlist
);
1548 struct sctp_association
*asoc
= transport
->asoc
;
1550 /* We may have counted DATA that was migrated
1551 * to this transport due to DEL-IP operation.
1552 * Subtract those bytes, since the were never
1553 * send on this transport and shouldn't be
1554 * credited to this transport.
1556 bytes_acked
-= migrate_bytes
;
1558 /* 8.2. When an outstanding TSN is acknowledged,
1559 * the endpoint shall clear the error counter of
1560 * the destination transport address to which the
1561 * DATA chunk was last sent.
1562 * The association's overall error counter is
1565 transport
->error_count
= 0;
1566 transport
->asoc
->overall_error_count
= 0;
1567 forward_progress
= true;
1570 * While in SHUTDOWN PENDING, we may have started
1571 * the T5 shutdown guard timer after reaching the
1572 * retransmission limit. Stop that timer as soon
1573 * as the receiver acknowledged any data.
1575 if (asoc
->state
== SCTP_STATE_SHUTDOWN_PENDING
&&
1576 del_timer(&asoc
->timers
1577 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD
]))
1578 sctp_association_put(asoc
);
1580 /* Mark the destination transport address as
1581 * active if it is not so marked.
1583 if ((transport
->state
== SCTP_INACTIVE
||
1584 transport
->state
== SCTP_UNCONFIRMED
) &&
1585 sctp_cmp_addr_exact(&transport
->ipaddr
, saddr
)) {
1586 sctp_assoc_control_transport(
1590 SCTP_RECEIVED_SACK
);
1593 sctp_transport_raise_cwnd(transport
, sack_ctsn
,
1596 transport
->flight_size
-= bytes_acked
;
1597 if (transport
->flight_size
== 0)
1598 transport
->partial_bytes_acked
= 0;
1599 q
->outstanding_bytes
-= bytes_acked
+ migrate_bytes
;
1601 /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1602 * When a sender is doing zero window probing, it
1603 * should not timeout the association if it continues
1604 * to receive new packets from the receiver. The
1605 * reason is that the receiver MAY keep its window
1606 * closed for an indefinite time.
1607 * A sender is doing zero window probing when the
1608 * receiver's advertised window is zero, and there is
1609 * only one data chunk in flight to the receiver.
1611 * Allow the association to timeout while in SHUTDOWN
1612 * PENDING or SHUTDOWN RECEIVED in case the receiver
1613 * stays in zero window mode forever.
1615 if (!q
->asoc
->peer
.rwnd
&&
1616 !list_empty(&tlist
) &&
1617 (sack_ctsn
+2 == q
->asoc
->next_tsn
) &&
1618 q
->asoc
->state
< SCTP_STATE_SHUTDOWN_PENDING
) {
1619 pr_debug("%s: sack received for zero window "
1620 "probe:%u\n", __func__
, sack_ctsn
);
1622 q
->asoc
->overall_error_count
= 0;
1623 transport
->error_count
= 0;
1627 /* RFC 2960 6.3.2 Retransmission Timer Rules
1629 * R2) Whenever all outstanding data sent to an address have
1630 * been acknowledged, turn off the T3-rtx timer of that
1633 if (!transport
->flight_size
) {
1634 if (del_timer(&transport
->T3_rtx_timer
))
1635 sctp_transport_put(transport
);
1636 } else if (restart_timer
) {
1637 if (!mod_timer(&transport
->T3_rtx_timer
,
1638 jiffies
+ transport
->rto
))
1639 sctp_transport_hold(transport
);
1642 if (forward_progress
) {
1644 dst_confirm(transport
->dst
);
1648 list_splice(&tlist
, transmitted_queue
);
1651 /* Mark chunks as missing and consequently may get retransmitted. */
1652 static void sctp_mark_missing(struct sctp_outq
*q
,
1653 struct list_head
*transmitted_queue
,
1654 struct sctp_transport
*transport
,
1655 __u32 highest_new_tsn_in_sack
,
1656 int count_of_newacks
)
1658 struct sctp_chunk
*chunk
;
1660 char do_fast_retransmit
= 0;
1661 struct sctp_association
*asoc
= q
->asoc
;
1662 struct sctp_transport
*primary
= asoc
->peer
.primary_path
;
1664 list_for_each_entry(chunk
, transmitted_queue
, transmitted_list
) {
1666 tsn
= ntohl(chunk
->subh
.data_hdr
->tsn
);
1668 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1669 * 'Unacknowledged TSN's', if the TSN number of an
1670 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1671 * value, increment the 'TSN.Missing.Report' count on that
1672 * chunk if it has NOT been fast retransmitted or marked for
1673 * fast retransmit already.
1675 if (chunk
->fast_retransmit
== SCTP_CAN_FRTX
&&
1676 !chunk
->tsn_gap_acked
&&
1677 TSN_lt(tsn
, highest_new_tsn_in_sack
)) {
1679 /* SFR-CACC may require us to skip marking
1680 * this chunk as missing.
1682 if (!transport
|| !sctp_cacc_skip(primary
,
1684 count_of_newacks
, tsn
)) {
1685 chunk
->tsn_missing_report
++;
1687 pr_debug("%s: tsn:0x%x missing counter:%d\n",
1688 __func__
, tsn
, chunk
->tsn_missing_report
);
1692 * M4) If any DATA chunk is found to have a
1693 * 'TSN.Missing.Report'
1694 * value larger than or equal to 3, mark that chunk for
1695 * retransmission and start the fast retransmit procedure.
1698 if (chunk
->tsn_missing_report
>= 3) {
1699 chunk
->fast_retransmit
= SCTP_NEED_FRTX
;
1700 do_fast_retransmit
= 1;
1705 if (do_fast_retransmit
)
1706 sctp_retransmit(q
, transport
, SCTP_RTXR_FAST_RTX
);
1708 pr_debug("%s: transport:%p, cwnd:%d, ssthresh:%d, "
1709 "flight_size:%d, pba:%d\n", __func__
, transport
,
1710 transport
->cwnd
, transport
->ssthresh
,
1711 transport
->flight_size
, transport
->partial_bytes_acked
);
1715 /* Is the given TSN acked by this packet? */
1716 static int sctp_acked(struct sctp_sackhdr
*sack
, __u32 tsn
)
1719 sctp_sack_variable_t
*frags
;
1720 __u16 tsn_offset
, blocks
;
1721 __u32 ctsn
= ntohl(sack
->cum_tsn_ack
);
1723 if (TSN_lte(tsn
, ctsn
))
1726 /* 3.3.4 Selective Acknowledgement (SACK) (3):
1729 * These fields contain the Gap Ack Blocks. They are repeated
1730 * for each Gap Ack Block up to the number of Gap Ack Blocks
1731 * defined in the Number of Gap Ack Blocks field. All DATA
1732 * chunks with TSNs greater than or equal to (Cumulative TSN
1733 * Ack + Gap Ack Block Start) and less than or equal to
1734 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1735 * Block are assumed to have been received correctly.
1738 frags
= sack
->variable
;
1739 blocks
= ntohs(sack
->num_gap_ack_blocks
);
1740 tsn_offset
= tsn
- ctsn
;
1741 for (i
= 0; i
< blocks
; ++i
) {
1742 if (tsn_offset
>= ntohs(frags
[i
].gab
.start
) &&
1743 tsn_offset
<= ntohs(frags
[i
].gab
.end
))
1752 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip
*skiplist
,
1753 int nskips
, __be16 stream
)
1757 for (i
= 0; i
< nskips
; i
++) {
1758 if (skiplist
[i
].stream
== stream
)
1764 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1765 static void sctp_generate_fwdtsn(struct sctp_outq
*q
, __u32 ctsn
)
1767 struct sctp_association
*asoc
= q
->asoc
;
1768 struct sctp_chunk
*ftsn_chunk
= NULL
;
1769 struct sctp_fwdtsn_skip ftsn_skip_arr
[10];
1773 struct sctp_chunk
*chunk
;
1774 struct list_head
*lchunk
, *temp
;
1776 if (!asoc
->peer
.prsctp_capable
)
1779 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1782 * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1783 * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1785 if (TSN_lt(asoc
->adv_peer_ack_point
, ctsn
))
1786 asoc
->adv_peer_ack_point
= ctsn
;
1788 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1789 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1790 * the chunk next in the out-queue space is marked as "abandoned" as
1791 * shown in the following example:
1793 * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1794 * and the Advanced.Peer.Ack.Point is updated to this value:
1796 * out-queue at the end of ==> out-queue after Adv.Ack.Point
1797 * normal SACK processing local advancement
1799 * Adv.Ack.Pt-> 102 acked 102 acked
1800 * 103 abandoned 103 abandoned
1801 * 104 abandoned Adv.Ack.P-> 104 abandoned
1803 * 106 acked 106 acked
1806 * In this example, the data sender successfully advanced the
1807 * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1809 list_for_each_safe(lchunk
, temp
, &q
->abandoned
) {
1810 chunk
= list_entry(lchunk
, struct sctp_chunk
,
1812 tsn
= ntohl(chunk
->subh
.data_hdr
->tsn
);
1814 /* Remove any chunks in the abandoned queue that are acked by
1817 if (TSN_lte(tsn
, ctsn
)) {
1818 list_del_init(lchunk
);
1819 sctp_chunk_free(chunk
);
1821 if (TSN_lte(tsn
, asoc
->adv_peer_ack_point
+1)) {
1822 asoc
->adv_peer_ack_point
= tsn
;
1823 if (chunk
->chunk_hdr
->flags
&
1824 SCTP_DATA_UNORDERED
)
1826 skip_pos
= sctp_get_skip_pos(&ftsn_skip_arr
[0],
1828 chunk
->subh
.data_hdr
->stream
);
1829 ftsn_skip_arr
[skip_pos
].stream
=
1830 chunk
->subh
.data_hdr
->stream
;
1831 ftsn_skip_arr
[skip_pos
].ssn
=
1832 chunk
->subh
.data_hdr
->ssn
;
1833 if (skip_pos
== nskips
)
1842 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1843 * is greater than the Cumulative TSN ACK carried in the received
1844 * SACK, the data sender MUST send the data receiver a FORWARD TSN
1845 * chunk containing the latest value of the
1846 * "Advanced.Peer.Ack.Point".
1848 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1849 * list each stream and sequence number in the forwarded TSN. This
1850 * information will enable the receiver to easily find any
1851 * stranded TSN's waiting on stream reorder queues. Each stream
1852 * SHOULD only be reported once; this means that if multiple
1853 * abandoned messages occur in the same stream then only the
1854 * highest abandoned stream sequence number is reported. If the
1855 * total size of the FORWARD TSN does NOT fit in a single MTU then
1856 * the sender of the FORWARD TSN SHOULD lower the
1857 * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1860 if (asoc
->adv_peer_ack_point
> ctsn
)
1861 ftsn_chunk
= sctp_make_fwdtsn(asoc
, asoc
->adv_peer_ack_point
,
1862 nskips
, &ftsn_skip_arr
[0]);
1865 list_add_tail(&ftsn_chunk
->list
, &q
->control_chunk_list
);
1866 SCTP_INC_STATS(sock_net(asoc
->base
.sk
), SCTP_MIB_OUTCTRLCHUNKS
);