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>
53 #include <net/sctp/stream_sched.h>
55 /* Declare internal functions here. */
56 static int sctp_acked(struct sctp_sackhdr
*sack
, __u32 tsn
);
57 static void sctp_check_transmitted(struct sctp_outq
*q
,
58 struct list_head
*transmitted_queue
,
59 struct sctp_transport
*transport
,
60 union sctp_addr
*saddr
,
61 struct sctp_sackhdr
*sack
,
62 __u32
*highest_new_tsn
);
64 static void sctp_mark_missing(struct sctp_outq
*q
,
65 struct list_head
*transmitted_queue
,
66 struct sctp_transport
*transport
,
67 __u32 highest_new_tsn
,
68 int count_of_newacks
);
70 static void sctp_generate_fwdtsn(struct sctp_outq
*q
, __u32 sack_ctsn
);
72 static void sctp_outq_flush(struct sctp_outq
*q
, int rtx_timeout
, gfp_t gfp
);
74 /* Add data to the front of the queue. */
75 static inline void sctp_outq_head_data(struct sctp_outq
*q
,
76 struct sctp_chunk
*ch
)
78 struct sctp_stream_out_ext
*oute
;
81 list_add(&ch
->list
, &q
->out_chunk_list
);
82 q
->out_qlen
+= ch
->skb
->len
;
84 stream
= sctp_chunk_stream_no(ch
);
85 oute
= q
->asoc
->stream
.out
[stream
].ext
;
86 list_add(&ch
->stream_list
, &oute
->outq
);
89 /* Take data from the front of the queue. */
90 static inline struct sctp_chunk
*sctp_outq_dequeue_data(struct sctp_outq
*q
)
92 return q
->sched
->dequeue(q
);
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 struct sctp_stream_out_ext
*oute
;
102 list_add_tail(&ch
->list
, &q
->out_chunk_list
);
103 q
->out_qlen
+= ch
->skb
->len
;
105 stream
= sctp_chunk_stream_no(ch
);
106 oute
= q
->asoc
->stream
.out
[stream
].ext
;
107 list_add_tail(&ch
->stream_list
, &oute
->outq
);
111 * SFR-CACC algorithm:
112 * D) If count_of_newacks is greater than or equal to 2
113 * and t was not sent to the current primary then the
114 * sender MUST NOT increment missing report count for t.
116 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport
*primary
,
117 struct sctp_transport
*transport
,
118 int count_of_newacks
)
120 if (count_of_newacks
>= 2 && transport
!= primary
)
126 * SFR-CACC algorithm:
127 * F) If count_of_newacks is less than 2, let d be the
128 * destination to which t was sent. If cacc_saw_newack
129 * is 0 for destination d, then the sender MUST NOT
130 * increment missing report count for t.
132 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport
*transport
,
133 int count_of_newacks
)
135 if (count_of_newacks
< 2 &&
136 (transport
&& !transport
->cacc
.cacc_saw_newack
))
142 * SFR-CACC algorithm:
143 * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD
144 * execute steps C, D, F.
146 * C has been implemented in sctp_outq_sack
148 static inline int sctp_cacc_skip_3_1(struct sctp_transport
*primary
,
149 struct sctp_transport
*transport
,
150 int count_of_newacks
)
152 if (!primary
->cacc
.cycling_changeover
) {
153 if (sctp_cacc_skip_3_1_d(primary
, transport
, count_of_newacks
))
155 if (sctp_cacc_skip_3_1_f(transport
, count_of_newacks
))
163 * SFR-CACC algorithm:
164 * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less
165 * than next_tsn_at_change of the current primary, then
166 * the sender MUST NOT increment missing report count
169 static inline int sctp_cacc_skip_3_2(struct sctp_transport
*primary
, __u32 tsn
)
171 if (primary
->cacc
.cycling_changeover
&&
172 TSN_lt(tsn
, primary
->cacc
.next_tsn_at_change
))
178 * SFR-CACC algorithm:
179 * 3) If the missing report count for TSN t is to be
180 * incremented according to [RFC2960] and
181 * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set,
182 * then the sender MUST further execute steps 3.1 and
183 * 3.2 to determine if the missing report count for
184 * TSN t SHOULD NOT be incremented.
186 * 3.3) If 3.1 and 3.2 do not dictate that the missing
187 * report count for t should not be incremented, then
188 * the sender SHOULD increment missing report count for
189 * t (according to [RFC2960] and [SCTP_STEWART_2002]).
191 static inline int sctp_cacc_skip(struct sctp_transport
*primary
,
192 struct sctp_transport
*transport
,
193 int count_of_newacks
,
196 if (primary
->cacc
.changeover_active
&&
197 (sctp_cacc_skip_3_1(primary
, transport
, count_of_newacks
) ||
198 sctp_cacc_skip_3_2(primary
, tsn
)))
203 /* Initialize an existing sctp_outq. This does the boring stuff.
204 * You still need to define handlers if you really want to DO
205 * something with this structure...
207 void sctp_outq_init(struct sctp_association
*asoc
, struct sctp_outq
*q
)
209 memset(q
, 0, sizeof(struct sctp_outq
));
212 INIT_LIST_HEAD(&q
->out_chunk_list
);
213 INIT_LIST_HEAD(&q
->control_chunk_list
);
214 INIT_LIST_HEAD(&q
->retransmit
);
215 INIT_LIST_HEAD(&q
->sacked
);
216 INIT_LIST_HEAD(&q
->abandoned
);
217 sctp_sched_set_sched(asoc
, SCTP_SS_FCFS
);
220 /* Free the outqueue structure and any related pending chunks.
222 static void __sctp_outq_teardown(struct sctp_outq
*q
)
224 struct sctp_transport
*transport
;
225 struct list_head
*lchunk
, *temp
;
226 struct sctp_chunk
*chunk
, *tmp
;
228 /* Throw away unacknowledged chunks. */
229 list_for_each_entry(transport
, &q
->asoc
->peer
.transport_addr_list
,
231 while ((lchunk
= sctp_list_dequeue(&transport
->transmitted
)) != NULL
) {
232 chunk
= list_entry(lchunk
, struct sctp_chunk
,
234 /* Mark as part of a failed message. */
235 sctp_chunk_fail(chunk
, q
->error
);
236 sctp_chunk_free(chunk
);
240 /* Throw away chunks that have been gap ACKed. */
241 list_for_each_safe(lchunk
, temp
, &q
->sacked
) {
242 list_del_init(lchunk
);
243 chunk
= list_entry(lchunk
, struct sctp_chunk
,
245 sctp_chunk_fail(chunk
, q
->error
);
246 sctp_chunk_free(chunk
);
249 /* Throw away any chunks in the retransmit queue. */
250 list_for_each_safe(lchunk
, temp
, &q
->retransmit
) {
251 list_del_init(lchunk
);
252 chunk
= list_entry(lchunk
, struct sctp_chunk
,
254 sctp_chunk_fail(chunk
, q
->error
);
255 sctp_chunk_free(chunk
);
258 /* Throw away any chunks that are in the abandoned queue. */
259 list_for_each_safe(lchunk
, temp
, &q
->abandoned
) {
260 list_del_init(lchunk
);
261 chunk
= list_entry(lchunk
, struct sctp_chunk
,
263 sctp_chunk_fail(chunk
, q
->error
);
264 sctp_chunk_free(chunk
);
267 /* Throw away any leftover data chunks. */
268 while ((chunk
= sctp_outq_dequeue_data(q
)) != NULL
) {
269 sctp_sched_dequeue_done(q
, chunk
);
271 /* Mark as send failure. */
272 sctp_chunk_fail(chunk
, q
->error
);
273 sctp_chunk_free(chunk
);
276 /* Throw away any leftover control chunks. */
277 list_for_each_entry_safe(chunk
, tmp
, &q
->control_chunk_list
, list
) {
278 list_del_init(&chunk
->list
);
279 sctp_chunk_free(chunk
);
283 void sctp_outq_teardown(struct sctp_outq
*q
)
285 __sctp_outq_teardown(q
);
286 sctp_outq_init(q
->asoc
, q
);
289 /* Free the outqueue structure and any related pending chunks. */
290 void sctp_outq_free(struct sctp_outq
*q
)
292 /* Throw away leftover chunks. */
293 __sctp_outq_teardown(q
);
296 /* Put a new chunk in an sctp_outq. */
297 void sctp_outq_tail(struct sctp_outq
*q
, struct sctp_chunk
*chunk
, gfp_t gfp
)
299 struct net
*net
= sock_net(q
->asoc
->base
.sk
);
301 pr_debug("%s: outq:%p, chunk:%p[%s]\n", __func__
, q
, chunk
,
302 chunk
&& chunk
->chunk_hdr
?
303 sctp_cname(SCTP_ST_CHUNK(chunk
->chunk_hdr
->type
)) :
306 /* If it is data, queue it up, otherwise, send it
309 if (sctp_chunk_is_data(chunk
)) {
310 pr_debug("%s: outqueueing: outq:%p, chunk:%p[%s])\n",
311 __func__
, q
, chunk
, chunk
&& chunk
->chunk_hdr
?
312 sctp_cname(SCTP_ST_CHUNK(chunk
->chunk_hdr
->type
)) :
315 sctp_outq_tail_data(q
, chunk
);
316 if (chunk
->asoc
->peer
.prsctp_capable
&&
317 SCTP_PR_PRIO_ENABLED(chunk
->sinfo
.sinfo_flags
))
318 chunk
->asoc
->sent_cnt_removable
++;
319 if (chunk
->chunk_hdr
->flags
& SCTP_DATA_UNORDERED
)
320 SCTP_INC_STATS(net
, SCTP_MIB_OUTUNORDERCHUNKS
);
322 SCTP_INC_STATS(net
, SCTP_MIB_OUTORDERCHUNKS
);
324 list_add_tail(&chunk
->list
, &q
->control_chunk_list
);
325 SCTP_INC_STATS(net
, SCTP_MIB_OUTCTRLCHUNKS
);
329 sctp_outq_flush(q
, 0, gfp
);
332 /* Insert a chunk into the sorted list based on the TSNs. The retransmit list
333 * and the abandoned list are in ascending order.
335 static void sctp_insert_list(struct list_head
*head
, struct list_head
*new)
337 struct list_head
*pos
;
338 struct sctp_chunk
*nchunk
, *lchunk
;
342 nchunk
= list_entry(new, struct sctp_chunk
, transmitted_list
);
343 ntsn
= ntohl(nchunk
->subh
.data_hdr
->tsn
);
345 list_for_each(pos
, head
) {
346 lchunk
= list_entry(pos
, struct sctp_chunk
, transmitted_list
);
347 ltsn
= ntohl(lchunk
->subh
.data_hdr
->tsn
);
348 if (TSN_lt(ntsn
, ltsn
)) {
349 list_add(new, pos
->prev
);
355 list_add_tail(new, head
);
358 static int sctp_prsctp_prune_sent(struct sctp_association
*asoc
,
359 struct sctp_sndrcvinfo
*sinfo
,
360 struct list_head
*queue
, int msg_len
)
362 struct sctp_chunk
*chk
, *temp
;
364 list_for_each_entry_safe(chk
, temp
, queue
, transmitted_list
) {
365 struct sctp_stream_out
*streamout
;
367 if (!chk
->msg
->abandoned
&&
368 (!SCTP_PR_PRIO_ENABLED(chk
->sinfo
.sinfo_flags
) ||
369 chk
->sinfo
.sinfo_timetolive
<= sinfo
->sinfo_timetolive
))
372 chk
->msg
->abandoned
= 1;
373 list_del_init(&chk
->transmitted_list
);
374 sctp_insert_list(&asoc
->outqueue
.abandoned
,
375 &chk
->transmitted_list
);
377 streamout
= &asoc
->stream
.out
[chk
->sinfo
.sinfo_stream
];
378 asoc
->sent_cnt_removable
--;
379 asoc
->abandoned_sent
[SCTP_PR_INDEX(PRIO
)]++;
380 streamout
->ext
->abandoned_sent
[SCTP_PR_INDEX(PRIO
)]++;
382 if (queue
!= &asoc
->outqueue
.retransmit
&&
383 !chk
->tsn_gap_acked
) {
385 chk
->transport
->flight_size
-=
387 asoc
->outqueue
.outstanding_bytes
-= sctp_data_size(chk
);
390 msg_len
-= SCTP_DATA_SNDSIZE(chk
) +
391 sizeof(struct sk_buff
) +
392 sizeof(struct sctp_chunk
);
400 static int sctp_prsctp_prune_unsent(struct sctp_association
*asoc
,
401 struct sctp_sndrcvinfo
*sinfo
, int msg_len
)
403 struct sctp_outq
*q
= &asoc
->outqueue
;
404 struct sctp_chunk
*chk
, *temp
;
406 q
->sched
->unsched_all(&asoc
->stream
);
408 list_for_each_entry_safe(chk
, temp
, &q
->out_chunk_list
, list
) {
409 if (!chk
->msg
->abandoned
&&
410 (!(chk
->chunk_hdr
->flags
& SCTP_DATA_FIRST_FRAG
) ||
411 !SCTP_PR_PRIO_ENABLED(chk
->sinfo
.sinfo_flags
) ||
412 chk
->sinfo
.sinfo_timetolive
<= sinfo
->sinfo_timetolive
))
415 chk
->msg
->abandoned
= 1;
416 sctp_sched_dequeue_common(q
, chk
);
417 asoc
->sent_cnt_removable
--;
418 asoc
->abandoned_unsent
[SCTP_PR_INDEX(PRIO
)]++;
419 if (chk
->sinfo
.sinfo_stream
< asoc
->stream
.outcnt
) {
420 struct sctp_stream_out
*streamout
=
421 &asoc
->stream
.out
[chk
->sinfo
.sinfo_stream
];
423 streamout
->ext
->abandoned_unsent
[SCTP_PR_INDEX(PRIO
)]++;
426 msg_len
-= SCTP_DATA_SNDSIZE(chk
) +
427 sizeof(struct sk_buff
) +
428 sizeof(struct sctp_chunk
);
429 sctp_chunk_free(chk
);
434 q
->sched
->sched_all(&asoc
->stream
);
439 /* Abandon the chunks according their priorities */
440 void sctp_prsctp_prune(struct sctp_association
*asoc
,
441 struct sctp_sndrcvinfo
*sinfo
, int msg_len
)
443 struct sctp_transport
*transport
;
445 if (!asoc
->peer
.prsctp_capable
|| !asoc
->sent_cnt_removable
)
448 msg_len
= sctp_prsctp_prune_sent(asoc
, sinfo
,
449 &asoc
->outqueue
.retransmit
,
454 list_for_each_entry(transport
, &asoc
->peer
.transport_addr_list
,
456 msg_len
= sctp_prsctp_prune_sent(asoc
, sinfo
,
457 &transport
->transmitted
,
463 sctp_prsctp_prune_unsent(asoc
, sinfo
, msg_len
);
466 /* Mark all the eligible packets on a transport for retransmission. */
467 void sctp_retransmit_mark(struct sctp_outq
*q
,
468 struct sctp_transport
*transport
,
471 struct list_head
*lchunk
, *ltemp
;
472 struct sctp_chunk
*chunk
;
474 /* Walk through the specified transmitted queue. */
475 list_for_each_safe(lchunk
, ltemp
, &transport
->transmitted
) {
476 chunk
= list_entry(lchunk
, struct sctp_chunk
,
479 /* If the chunk is abandoned, move it to abandoned list. */
480 if (sctp_chunk_abandoned(chunk
)) {
481 list_del_init(lchunk
);
482 sctp_insert_list(&q
->abandoned
, lchunk
);
484 /* If this chunk has not been previousely acked,
485 * stop considering it 'outstanding'. Our peer
486 * will most likely never see it since it will
487 * not be retransmitted
489 if (!chunk
->tsn_gap_acked
) {
490 if (chunk
->transport
)
491 chunk
->transport
->flight_size
-=
492 sctp_data_size(chunk
);
493 q
->outstanding_bytes
-= sctp_data_size(chunk
);
494 q
->asoc
->peer
.rwnd
+= sctp_data_size(chunk
);
499 /* If we are doing retransmission due to a timeout or pmtu
500 * discovery, only the chunks that are not yet acked should
501 * be added to the retransmit queue.
503 if ((reason
== SCTP_RTXR_FAST_RTX
&&
504 (chunk
->fast_retransmit
== SCTP_NEED_FRTX
)) ||
505 (reason
!= SCTP_RTXR_FAST_RTX
&& !chunk
->tsn_gap_acked
)) {
506 /* RFC 2960 6.2.1 Processing a Received SACK
508 * C) Any time a DATA chunk is marked for
509 * retransmission (via either T3-rtx timer expiration
510 * (Section 6.3.3) or via fast retransmit
511 * (Section 7.2.4)), add the data size of those
512 * chunks to the rwnd.
514 q
->asoc
->peer
.rwnd
+= sctp_data_size(chunk
);
515 q
->outstanding_bytes
-= sctp_data_size(chunk
);
516 if (chunk
->transport
)
517 transport
->flight_size
-= sctp_data_size(chunk
);
519 /* sctpimpguide-05 Section 2.8.2
520 * M5) If a T3-rtx timer expires, the
521 * 'TSN.Missing.Report' of all affected TSNs is set
524 chunk
->tsn_missing_report
= 0;
526 /* If a chunk that is being used for RTT measurement
527 * has to be retransmitted, we cannot use this chunk
528 * anymore for RTT measurements. Reset rto_pending so
529 * that a new RTT measurement is started when a new
530 * data chunk is sent.
532 if (chunk
->rtt_in_progress
) {
533 chunk
->rtt_in_progress
= 0;
534 transport
->rto_pending
= 0;
537 /* Move the chunk to the retransmit queue. The chunks
538 * on the retransmit queue are always kept in order.
540 list_del_init(lchunk
);
541 sctp_insert_list(&q
->retransmit
, lchunk
);
545 pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d, "
546 "flight_size:%d, pba:%d\n", __func__
, transport
, reason
,
547 transport
->cwnd
, transport
->ssthresh
, transport
->flight_size
,
548 transport
->partial_bytes_acked
);
551 /* Mark all the eligible packets on a transport for retransmission and force
554 void sctp_retransmit(struct sctp_outq
*q
, struct sctp_transport
*transport
,
555 enum sctp_retransmit_reason reason
)
557 struct net
*net
= sock_net(q
->asoc
->base
.sk
);
560 case SCTP_RTXR_T3_RTX
:
561 SCTP_INC_STATS(net
, SCTP_MIB_T3_RETRANSMITS
);
562 sctp_transport_lower_cwnd(transport
, SCTP_LOWER_CWND_T3_RTX
);
563 /* Update the retran path if the T3-rtx timer has expired for
564 * the current retran path.
566 if (transport
== transport
->asoc
->peer
.retran_path
)
567 sctp_assoc_update_retran_path(transport
->asoc
);
568 transport
->asoc
->rtx_data_chunks
+=
569 transport
->asoc
->unack_data
;
571 case SCTP_RTXR_FAST_RTX
:
572 SCTP_INC_STATS(net
, SCTP_MIB_FAST_RETRANSMITS
);
573 sctp_transport_lower_cwnd(transport
, SCTP_LOWER_CWND_FAST_RTX
);
576 case SCTP_RTXR_PMTUD
:
577 SCTP_INC_STATS(net
, SCTP_MIB_PMTUD_RETRANSMITS
);
579 case SCTP_RTXR_T1_RTX
:
580 SCTP_INC_STATS(net
, SCTP_MIB_T1_RETRANSMITS
);
581 transport
->asoc
->init_retries
++;
587 sctp_retransmit_mark(q
, transport
, reason
);
589 /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
590 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
591 * following the procedures outlined in C1 - C5.
593 if (reason
== SCTP_RTXR_T3_RTX
)
594 sctp_generate_fwdtsn(q
, q
->asoc
->ctsn_ack_point
);
596 /* Flush the queues only on timeout, since fast_rtx is only
597 * triggered during sack processing and the queue
598 * will be flushed at the end.
600 if (reason
!= SCTP_RTXR_FAST_RTX
)
601 sctp_outq_flush(q
, /* rtx_timeout */ 1, GFP_ATOMIC
);
605 * Transmit DATA chunks on the retransmit queue. Upon return from
606 * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
607 * need to be transmitted by the caller.
608 * We assume that pkt->transport has already been set.
610 * The return value is a normal kernel error return value.
612 static int sctp_outq_flush_rtx(struct sctp_outq
*q
, struct sctp_packet
*pkt
,
613 int rtx_timeout
, int *start_timer
)
615 struct sctp_transport
*transport
= pkt
->transport
;
616 struct sctp_chunk
*chunk
, *chunk1
;
617 struct list_head
*lqueue
;
618 enum sctp_xmit status
;
624 lqueue
= &q
->retransmit
;
625 fast_rtx
= q
->fast_rtx
;
627 /* This loop handles time-out retransmissions, fast retransmissions,
628 * and retransmissions due to opening of whindow.
630 * RFC 2960 6.3.3 Handle T3-rtx Expiration
632 * E3) Determine how many of the earliest (i.e., lowest TSN)
633 * outstanding DATA chunks for the address for which the
634 * T3-rtx has expired will fit into a single packet, subject
635 * to the MTU constraint for the path corresponding to the
636 * destination transport address to which the retransmission
637 * is being sent (this may be different from the address for
638 * which the timer expires [see Section 6.4]). Call this value
639 * K. Bundle and retransmit those K DATA chunks in a single
640 * packet to the destination endpoint.
642 * [Just to be painfully clear, if we are retransmitting
643 * because a timeout just happened, we should send only ONE
644 * packet of retransmitted data.]
646 * For fast retransmissions we also send only ONE packet. However,
647 * if we are just flushing the queue due to open window, we'll
648 * try to send as much as possible.
650 list_for_each_entry_safe(chunk
, chunk1
, lqueue
, transmitted_list
) {
651 /* If the chunk is abandoned, move it to abandoned list. */
652 if (sctp_chunk_abandoned(chunk
)) {
653 list_del_init(&chunk
->transmitted_list
);
654 sctp_insert_list(&q
->abandoned
,
655 &chunk
->transmitted_list
);
659 /* Make sure that Gap Acked TSNs are not retransmitted. A
660 * simple approach is just to move such TSNs out of the
661 * way and into a 'transmitted' queue and skip to the
664 if (chunk
->tsn_gap_acked
) {
665 list_move_tail(&chunk
->transmitted_list
,
666 &transport
->transmitted
);
670 /* If we are doing fast retransmit, ignore non-fast_rtransmit
673 if (fast_rtx
&& !chunk
->fast_retransmit
)
677 /* Attempt to append this chunk to the packet. */
678 status
= sctp_packet_append_chunk(pkt
, chunk
);
681 case SCTP_XMIT_PMTU_FULL
:
682 if (!pkt
->has_data
&& !pkt
->has_cookie_echo
) {
683 /* If this packet did not contain DATA then
684 * retransmission did not happen, so do it
685 * again. We'll ignore the error here since
686 * control chunks are already freed so there
687 * is nothing we can do.
689 sctp_packet_transmit(pkt
, GFP_ATOMIC
);
693 /* Send this packet. */
694 error
= sctp_packet_transmit(pkt
, GFP_ATOMIC
);
696 /* If we are retransmitting, we should only
697 * send a single packet.
698 * Otherwise, try appending this chunk again.
700 if (rtx_timeout
|| fast_rtx
)
705 /* Bundle next chunk in the next round. */
708 case SCTP_XMIT_RWND_FULL
:
709 /* Send this packet. */
710 error
= sctp_packet_transmit(pkt
, GFP_ATOMIC
);
712 /* Stop sending DATA as there is no more room
718 case SCTP_XMIT_DELAY
:
719 /* Send this packet. */
720 error
= sctp_packet_transmit(pkt
, GFP_ATOMIC
);
722 /* Stop sending DATA because of nagle delay. */
727 /* The append was successful, so add this chunk to
728 * the transmitted list.
730 list_move_tail(&chunk
->transmitted_list
,
731 &transport
->transmitted
);
733 /* Mark the chunk as ineligible for fast retransmit
734 * after it is retransmitted.
736 if (chunk
->fast_retransmit
== SCTP_NEED_FRTX
)
737 chunk
->fast_retransmit
= SCTP_DONT_FRTX
;
739 q
->asoc
->stats
.rtxchunks
++;
743 /* Set the timer if there were no errors */
744 if (!error
&& !timer
)
751 /* If we are here due to a retransmit timeout or a fast
752 * retransmit and if there are any chunks left in the retransmit
753 * queue that could not fit in the PMTU sized packet, they need
754 * to be marked as ineligible for a subsequent fast retransmit.
756 if (rtx_timeout
|| fast_rtx
) {
757 list_for_each_entry(chunk1
, lqueue
, transmitted_list
) {
758 if (chunk1
->fast_retransmit
== SCTP_NEED_FRTX
)
759 chunk1
->fast_retransmit
= SCTP_DONT_FRTX
;
763 *start_timer
= timer
;
765 /* Clear fast retransmit hint */
772 /* Cork the outqueue so queued chunks are really queued. */
773 void sctp_outq_uncork(struct sctp_outq
*q
, gfp_t gfp
)
778 sctp_outq_flush(q
, 0, gfp
);
783 * Try to flush an outqueue.
785 * Description: Send everything in q which we legally can, subject to
786 * congestion limitations.
787 * * Note: This function can be called from multiple contexts so appropriate
788 * locking concerns must be made. Today we use the sock lock to protect
791 static void sctp_outq_flush(struct sctp_outq
*q
, int rtx_timeout
, gfp_t gfp
)
793 struct sctp_packet
*packet
;
794 struct sctp_packet singleton
;
795 struct sctp_association
*asoc
= q
->asoc
;
796 __u16 sport
= asoc
->base
.bind_addr
.port
;
797 __u16 dport
= asoc
->peer
.port
;
798 __u32 vtag
= asoc
->peer
.i
.init_tag
;
799 struct sctp_transport
*transport
= NULL
;
800 struct sctp_transport
*new_transport
;
801 struct sctp_chunk
*chunk
, *tmp
;
802 enum sctp_xmit status
;
807 /* These transports have chunks to send. */
808 struct list_head transport_list
;
809 struct list_head
*ltransport
;
811 INIT_LIST_HEAD(&transport_list
);
817 * When bundling control chunks with DATA chunks, an
818 * endpoint MUST place control chunks first in the outbound
819 * SCTP packet. The transmitter MUST transmit DATA chunks
820 * within a SCTP packet in increasing order of TSN.
824 list_for_each_entry_safe(chunk
, tmp
, &q
->control_chunk_list
, list
) {
826 * F1) This means that until such time as the ASCONF
827 * containing the add is acknowledged, the sender MUST
828 * NOT use the new IP address as a source for ANY SCTP
829 * packet except on carrying an ASCONF Chunk.
831 if (asoc
->src_out_of_asoc_ok
&&
832 chunk
->chunk_hdr
->type
!= SCTP_CID_ASCONF
)
835 list_del_init(&chunk
->list
);
837 /* Pick the right transport to use. */
838 new_transport
= chunk
->transport
;
840 if (!new_transport
) {
842 * If we have a prior transport pointer, see if
843 * the destination address of the chunk
844 * matches the destination address of the
845 * current transport. If not a match, then
846 * try to look up the transport with a given
847 * destination address. We do this because
848 * after processing ASCONFs, we may have new
849 * transports created.
852 sctp_cmp_addr_exact(&chunk
->dest
,
854 new_transport
= transport
;
856 new_transport
= sctp_assoc_lookup_paddr(asoc
,
859 /* if we still don't have a new transport, then
860 * use the current active path.
863 new_transport
= asoc
->peer
.active_path
;
864 } else if ((new_transport
->state
== SCTP_INACTIVE
) ||
865 (new_transport
->state
== SCTP_UNCONFIRMED
) ||
866 (new_transport
->state
== SCTP_PF
)) {
867 /* If the chunk is Heartbeat or Heartbeat Ack,
868 * send it to chunk->transport, even if it's
871 * 3.3.6 Heartbeat Acknowledgement:
873 * A HEARTBEAT ACK is always sent to the source IP
874 * address of the IP datagram containing the
875 * HEARTBEAT chunk to which this ack is responding.
878 * ASCONF_ACKs also must be sent to the source.
880 if (chunk
->chunk_hdr
->type
!= SCTP_CID_HEARTBEAT
&&
881 chunk
->chunk_hdr
->type
!= SCTP_CID_HEARTBEAT_ACK
&&
882 chunk
->chunk_hdr
->type
!= SCTP_CID_ASCONF_ACK
)
883 new_transport
= asoc
->peer
.active_path
;
886 /* Are we switching transports?
887 * Take care of transport locks.
889 if (new_transport
!= transport
) {
890 transport
= new_transport
;
891 if (list_empty(&transport
->send_ready
)) {
892 list_add_tail(&transport
->send_ready
,
895 packet
= &transport
->packet
;
896 sctp_packet_config(packet
, vtag
,
897 asoc
->peer
.ecn_capable
);
900 switch (chunk
->chunk_hdr
->type
) {
904 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
905 * COMPLETE with any other chunks. [Send them immediately.]
908 case SCTP_CID_INIT_ACK
:
909 case SCTP_CID_SHUTDOWN_COMPLETE
:
910 sctp_packet_init(&singleton
, transport
, sport
, dport
);
911 sctp_packet_config(&singleton
, vtag
, 0);
912 sctp_packet_append_chunk(&singleton
, chunk
);
913 error
= sctp_packet_transmit(&singleton
, gfp
);
915 asoc
->base
.sk
->sk_err
= -error
;
921 if (sctp_test_T_bit(chunk
))
922 packet
->vtag
= asoc
->c
.my_vtag
;
924 /* The following chunks are "response" chunks, i.e.
925 * they are generated in response to something we
926 * received. If we are sending these, then we can
927 * send only 1 packet containing these chunks.
929 case SCTP_CID_HEARTBEAT_ACK
:
930 case SCTP_CID_SHUTDOWN_ACK
:
931 case SCTP_CID_COOKIE_ACK
:
932 case SCTP_CID_COOKIE_ECHO
:
934 case SCTP_CID_ECN_CWR
:
935 case SCTP_CID_ASCONF_ACK
:
940 case SCTP_CID_HEARTBEAT
:
941 case SCTP_CID_SHUTDOWN
:
942 case SCTP_CID_ECN_ECNE
:
943 case SCTP_CID_ASCONF
:
944 case SCTP_CID_FWD_TSN
:
945 case SCTP_CID_RECONF
:
946 status
= sctp_packet_transmit_chunk(packet
, chunk
,
948 if (status
!= SCTP_XMIT_OK
) {
949 /* put the chunk back */
950 list_add(&chunk
->list
, &q
->control_chunk_list
);
954 asoc
->stats
.octrlchunks
++;
955 /* PR-SCTP C5) If a FORWARD TSN is sent, the
956 * sender MUST assure that at least one T3-rtx
959 if (chunk
->chunk_hdr
->type
== SCTP_CID_FWD_TSN
) {
960 sctp_transport_reset_t3_rtx(transport
);
961 transport
->last_time_sent
= jiffies
;
964 if (chunk
== asoc
->strreset_chunk
)
965 sctp_transport_reset_reconf_timer(transport
);
970 /* We built a chunk with an illegal type! */
975 if (q
->asoc
->src_out_of_asoc_ok
)
978 /* Is it OK to send data chunks? */
979 switch (asoc
->state
) {
980 case SCTP_STATE_COOKIE_ECHOED
:
981 /* Only allow bundling when this packet has a COOKIE-ECHO
984 if (!packet
|| !packet
->has_cookie_echo
)
988 case SCTP_STATE_ESTABLISHED
:
989 case SCTP_STATE_SHUTDOWN_PENDING
:
990 case SCTP_STATE_SHUTDOWN_RECEIVED
:
992 * RFC 2960 6.1 Transmission of DATA Chunks
994 * C) When the time comes for the sender to transmit,
995 * before sending new DATA chunks, the sender MUST
996 * first transmit any outstanding DATA chunks which
997 * are marked for retransmission (limited by the
1000 if (!list_empty(&q
->retransmit
)) {
1001 if (asoc
->peer
.retran_path
->state
== SCTP_UNCONFIRMED
)
1002 goto sctp_flush_out
;
1003 if (transport
== asoc
->peer
.retran_path
)
1006 /* Switch transports & prepare the packet. */
1008 transport
= asoc
->peer
.retran_path
;
1010 if (list_empty(&transport
->send_ready
)) {
1011 list_add_tail(&transport
->send_ready
,
1015 packet
= &transport
->packet
;
1016 sctp_packet_config(packet
, vtag
,
1017 asoc
->peer
.ecn_capable
);
1019 error
= sctp_outq_flush_rtx(q
, packet
,
1020 rtx_timeout
, &start_timer
);
1022 asoc
->base
.sk
->sk_err
= -error
;
1025 sctp_transport_reset_t3_rtx(transport
);
1026 transport
->last_time_sent
= jiffies
;
1029 /* This can happen on COOKIE-ECHO resend. Only
1030 * one chunk can get bundled with a COOKIE-ECHO.
1032 if (packet
->has_cookie_echo
)
1033 goto sctp_flush_out
;
1035 /* Don't send new data if there is still data
1036 * waiting to retransmit.
1038 if (!list_empty(&q
->retransmit
))
1039 goto sctp_flush_out
;
1042 /* Apply Max.Burst limitation to the current transport in
1043 * case it will be used for new data. We are going to
1044 * rest it before we return, but we want to apply the limit
1045 * to the currently queued data.
1048 sctp_transport_burst_limited(transport
);
1050 /* Finally, transmit new packets. */
1051 while ((chunk
= sctp_outq_dequeue_data(q
)) != NULL
) {
1052 __u32 sid
= ntohs(chunk
->subh
.data_hdr
->stream
);
1054 /* Has this chunk expired? */
1055 if (sctp_chunk_abandoned(chunk
)) {
1056 sctp_sched_dequeue_done(q
, chunk
);
1057 sctp_chunk_fail(chunk
, 0);
1058 sctp_chunk_free(chunk
);
1062 if (asoc
->stream
.out
[sid
].state
== SCTP_STREAM_CLOSED
) {
1063 sctp_outq_head_data(q
, chunk
);
1064 goto sctp_flush_out
;
1067 /* If there is a specified transport, use it.
1068 * Otherwise, we want to use the active path.
1070 new_transport
= chunk
->transport
;
1071 if (!new_transport
||
1072 ((new_transport
->state
== SCTP_INACTIVE
) ||
1073 (new_transport
->state
== SCTP_UNCONFIRMED
) ||
1074 (new_transport
->state
== SCTP_PF
)))
1075 new_transport
= asoc
->peer
.active_path
;
1076 if (new_transport
->state
== SCTP_UNCONFIRMED
) {
1077 WARN_ONCE(1, "Attempt to send packet on unconfirmed path.");
1078 sctp_sched_dequeue_done(q
, chunk
);
1079 sctp_chunk_fail(chunk
, 0);
1080 sctp_chunk_free(chunk
);
1084 /* Change packets if necessary. */
1085 if (new_transport
!= transport
) {
1086 transport
= new_transport
;
1088 /* Schedule to have this transport's
1091 if (list_empty(&transport
->send_ready
)) {
1092 list_add_tail(&transport
->send_ready
,
1096 packet
= &transport
->packet
;
1097 sctp_packet_config(packet
, vtag
,
1098 asoc
->peer
.ecn_capable
);
1099 /* We've switched transports, so apply the
1100 * Burst limit to the new transport.
1102 sctp_transport_burst_limited(transport
);
1105 pr_debug("%s: outq:%p, chunk:%p[%s], tx-tsn:0x%x skb->head:%p "
1107 __func__
, q
, chunk
, chunk
&& chunk
->chunk_hdr
?
1108 sctp_cname(SCTP_ST_CHUNK(chunk
->chunk_hdr
->type
)) :
1109 "illegal chunk", ntohl(chunk
->subh
.data_hdr
->tsn
),
1110 chunk
->skb
? chunk
->skb
->head
: NULL
, chunk
->skb
?
1111 refcount_read(&chunk
->skb
->users
) : -1);
1113 /* Add the chunk to the packet. */
1114 status
= sctp_packet_transmit_chunk(packet
, chunk
, 0, gfp
);
1117 case SCTP_XMIT_PMTU_FULL
:
1118 case SCTP_XMIT_RWND_FULL
:
1119 case SCTP_XMIT_DELAY
:
1120 /* We could not append this chunk, so put
1121 * the chunk back on the output queue.
1123 pr_debug("%s: could not transmit tsn:0x%x, status:%d\n",
1124 __func__
, ntohl(chunk
->subh
.data_hdr
->tsn
),
1127 sctp_outq_head_data(q
, chunk
);
1128 goto sctp_flush_out
;
1131 /* The sender is in the SHUTDOWN-PENDING state,
1132 * The sender MAY set the I-bit in the DATA
1135 if (asoc
->state
== SCTP_STATE_SHUTDOWN_PENDING
)
1136 chunk
->chunk_hdr
->flags
|= SCTP_DATA_SACK_IMM
;
1137 if (chunk
->chunk_hdr
->flags
& SCTP_DATA_UNORDERED
)
1138 asoc
->stats
.ouodchunks
++;
1140 asoc
->stats
.oodchunks
++;
1142 /* Only now it's safe to consider this
1143 * chunk as sent, sched-wise.
1145 sctp_sched_dequeue_done(q
, chunk
);
1153 /* BUG: We assume that the sctp_packet_transmit()
1154 * call below will succeed all the time and add the
1155 * chunk to the transmitted list and restart the
1157 * It is possible that the call can fail under OOM
1160 * Is this really a problem? Won't this behave
1163 list_add_tail(&chunk
->transmitted_list
,
1164 &transport
->transmitted
);
1166 sctp_transport_reset_t3_rtx(transport
);
1167 transport
->last_time_sent
= jiffies
;
1169 /* Only let one DATA chunk get bundled with a
1170 * COOKIE-ECHO chunk.
1172 if (packet
->has_cookie_echo
)
1173 goto sctp_flush_out
;
1184 /* Before returning, examine all the transports touched in
1185 * this call. Right now, we bluntly force clear all the
1186 * transports. Things might change after we implement Nagle.
1187 * But such an examination is still required.
1191 while ((ltransport
= sctp_list_dequeue(&transport_list
)) != NULL
) {
1192 struct sctp_transport
*t
= list_entry(ltransport
,
1193 struct sctp_transport
,
1195 packet
= &t
->packet
;
1196 if (!sctp_packet_empty(packet
)) {
1197 error
= sctp_packet_transmit(packet
, gfp
);
1199 asoc
->base
.sk
->sk_err
= -error
;
1202 /* Clear the burst limited state, if any */
1203 sctp_transport_burst_reset(t
);
1207 /* Update unack_data based on the incoming SACK chunk */
1208 static void sctp_sack_update_unack_data(struct sctp_association
*assoc
,
1209 struct sctp_sackhdr
*sack
)
1211 union sctp_sack_variable
*frags
;
1215 unack_data
= assoc
->next_tsn
- assoc
->ctsn_ack_point
- 1;
1217 frags
= sack
->variable
;
1218 for (i
= 0; i
< ntohs(sack
->num_gap_ack_blocks
); i
++) {
1219 unack_data
-= ((ntohs(frags
[i
].gab
.end
) -
1220 ntohs(frags
[i
].gab
.start
) + 1));
1223 assoc
->unack_data
= unack_data
;
1226 /* This is where we REALLY process a SACK.
1228 * Process the SACK against the outqueue. Mostly, this just frees
1229 * things off the transmitted queue.
1231 int sctp_outq_sack(struct sctp_outq
*q
, struct sctp_chunk
*chunk
)
1233 struct sctp_association
*asoc
= q
->asoc
;
1234 struct sctp_sackhdr
*sack
= chunk
->subh
.sack_hdr
;
1235 struct sctp_transport
*transport
;
1236 struct sctp_chunk
*tchunk
= NULL
;
1237 struct list_head
*lchunk
, *transport_list
, *temp
;
1238 union sctp_sack_variable
*frags
= sack
->variable
;
1239 __u32 sack_ctsn
, ctsn
, tsn
;
1240 __u32 highest_tsn
, highest_new_tsn
;
1242 unsigned int outstanding
;
1243 struct sctp_transport
*primary
= asoc
->peer
.primary_path
;
1244 int count_of_newacks
= 0;
1248 /* Grab the association's destination address list. */
1249 transport_list
= &asoc
->peer
.transport_addr_list
;
1251 sack_ctsn
= ntohl(sack
->cum_tsn_ack
);
1252 gap_ack_blocks
= ntohs(sack
->num_gap_ack_blocks
);
1253 asoc
->stats
.gapcnt
+= gap_ack_blocks
;
1255 * SFR-CACC algorithm:
1256 * On receipt of a SACK the sender SHOULD execute the
1257 * following statements.
1259 * 1) If the cumulative ack in the SACK passes next tsn_at_change
1260 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1261 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1263 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1264 * is set the receiver of the SACK MUST take the following actions:
1266 * A) Initialize the cacc_saw_newack to 0 for all destination
1269 * Only bother if changeover_active is set. Otherwise, this is
1270 * totally suboptimal to do on every SACK.
1272 if (primary
->cacc
.changeover_active
) {
1273 u8 clear_cycling
= 0;
1275 if (TSN_lte(primary
->cacc
.next_tsn_at_change
, sack_ctsn
)) {
1276 primary
->cacc
.changeover_active
= 0;
1280 if (clear_cycling
|| gap_ack_blocks
) {
1281 list_for_each_entry(transport
, transport_list
,
1284 transport
->cacc
.cycling_changeover
= 0;
1286 transport
->cacc
.cacc_saw_newack
= 0;
1291 /* Get the highest TSN in the sack. */
1292 highest_tsn
= sack_ctsn
;
1294 highest_tsn
+= ntohs(frags
[gap_ack_blocks
- 1].gab
.end
);
1296 if (TSN_lt(asoc
->highest_sacked
, highest_tsn
))
1297 asoc
->highest_sacked
= highest_tsn
;
1299 highest_new_tsn
= sack_ctsn
;
1301 /* Run through the retransmit queue. Credit bytes received
1302 * and free those chunks that we can.
1304 sctp_check_transmitted(q
, &q
->retransmit
, NULL
, NULL
, sack
, &highest_new_tsn
);
1306 /* Run through the transmitted queue.
1307 * Credit bytes received and free those chunks which we can.
1309 * This is a MASSIVE candidate for optimization.
1311 list_for_each_entry(transport
, transport_list
, transports
) {
1312 sctp_check_transmitted(q
, &transport
->transmitted
,
1313 transport
, &chunk
->source
, sack
,
1316 * SFR-CACC algorithm:
1317 * C) Let count_of_newacks be the number of
1318 * destinations for which cacc_saw_newack is set.
1320 if (transport
->cacc
.cacc_saw_newack
)
1324 /* Move the Cumulative TSN Ack Point if appropriate. */
1325 if (TSN_lt(asoc
->ctsn_ack_point
, sack_ctsn
)) {
1326 asoc
->ctsn_ack_point
= sack_ctsn
;
1330 if (gap_ack_blocks
) {
1332 if (asoc
->fast_recovery
&& accum_moved
)
1333 highest_new_tsn
= highest_tsn
;
1335 list_for_each_entry(transport
, transport_list
, transports
)
1336 sctp_mark_missing(q
, &transport
->transmitted
, transport
,
1337 highest_new_tsn
, count_of_newacks
);
1340 /* Update unack_data field in the assoc. */
1341 sctp_sack_update_unack_data(asoc
, sack
);
1343 ctsn
= asoc
->ctsn_ack_point
;
1345 /* Throw away stuff rotting on the sack queue. */
1346 list_for_each_safe(lchunk
, temp
, &q
->sacked
) {
1347 tchunk
= list_entry(lchunk
, struct sctp_chunk
,
1349 tsn
= ntohl(tchunk
->subh
.data_hdr
->tsn
);
1350 if (TSN_lte(tsn
, ctsn
)) {
1351 list_del_init(&tchunk
->transmitted_list
);
1352 if (asoc
->peer
.prsctp_capable
&&
1353 SCTP_PR_PRIO_ENABLED(chunk
->sinfo
.sinfo_flags
))
1354 asoc
->sent_cnt_removable
--;
1355 sctp_chunk_free(tchunk
);
1359 /* ii) Set rwnd equal to the newly received a_rwnd minus the
1360 * number of bytes still outstanding after processing the
1361 * Cumulative TSN Ack and the Gap Ack Blocks.
1364 sack_a_rwnd
= ntohl(sack
->a_rwnd
);
1365 asoc
->peer
.zero_window_announced
= !sack_a_rwnd
;
1366 outstanding
= q
->outstanding_bytes
;
1368 if (outstanding
< sack_a_rwnd
)
1369 sack_a_rwnd
-= outstanding
;
1373 asoc
->peer
.rwnd
= sack_a_rwnd
;
1375 sctp_generate_fwdtsn(q
, sack_ctsn
);
1377 pr_debug("%s: sack cumulative tsn ack:0x%x\n", __func__
, sack_ctsn
);
1378 pr_debug("%s: cumulative tsn ack of assoc:%p is 0x%x, "
1379 "advertised peer ack point:0x%x\n", __func__
, asoc
, ctsn
,
1380 asoc
->adv_peer_ack_point
);
1382 return sctp_outq_is_empty(q
);
1385 /* Is the outqueue empty?
1386 * The queue is empty when we have not pending data, no in-flight data
1387 * and nothing pending retransmissions.
1389 int sctp_outq_is_empty(const struct sctp_outq
*q
)
1391 return q
->out_qlen
== 0 && q
->outstanding_bytes
== 0 &&
1392 list_empty(&q
->retransmit
);
1395 /********************************************************************
1396 * 2nd Level Abstractions
1397 ********************************************************************/
1399 /* Go through a transport's transmitted list or the association's retransmit
1400 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1401 * The retransmit list will not have an associated transport.
1403 * I added coherent debug information output. --xguo
1405 * Instead of printing 'sacked' or 'kept' for each TSN on the
1406 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1407 * KEPT TSN6-TSN7, etc.
1409 static void sctp_check_transmitted(struct sctp_outq
*q
,
1410 struct list_head
*transmitted_queue
,
1411 struct sctp_transport
*transport
,
1412 union sctp_addr
*saddr
,
1413 struct sctp_sackhdr
*sack
,
1414 __u32
*highest_new_tsn_in_sack
)
1416 struct list_head
*lchunk
;
1417 struct sctp_chunk
*tchunk
;
1418 struct list_head tlist
;
1422 __u8 restart_timer
= 0;
1423 int bytes_acked
= 0;
1424 int migrate_bytes
= 0;
1425 bool forward_progress
= false;
1427 sack_ctsn
= ntohl(sack
->cum_tsn_ack
);
1429 INIT_LIST_HEAD(&tlist
);
1431 /* The while loop will skip empty transmitted queues. */
1432 while (NULL
!= (lchunk
= sctp_list_dequeue(transmitted_queue
))) {
1433 tchunk
= list_entry(lchunk
, struct sctp_chunk
,
1436 if (sctp_chunk_abandoned(tchunk
)) {
1437 /* Move the chunk to abandoned list. */
1438 sctp_insert_list(&q
->abandoned
, lchunk
);
1440 /* If this chunk has not been acked, stop
1441 * considering it as 'outstanding'.
1443 if (transmitted_queue
!= &q
->retransmit
&&
1444 !tchunk
->tsn_gap_acked
) {
1445 if (tchunk
->transport
)
1446 tchunk
->transport
->flight_size
-=
1447 sctp_data_size(tchunk
);
1448 q
->outstanding_bytes
-= sctp_data_size(tchunk
);
1453 tsn
= ntohl(tchunk
->subh
.data_hdr
->tsn
);
1454 if (sctp_acked(sack
, tsn
)) {
1455 /* If this queue is the retransmit queue, the
1456 * retransmit timer has already reclaimed
1457 * the outstanding bytes for this chunk, so only
1458 * count bytes associated with a transport.
1461 /* If this chunk is being used for RTT
1462 * measurement, calculate the RTT and update
1463 * the RTO using this value.
1465 * 6.3.1 C5) Karn's algorithm: RTT measurements
1466 * MUST NOT be made using packets that were
1467 * retransmitted (and thus for which it is
1468 * ambiguous whether the reply was for the
1469 * first instance of the packet or a later
1472 if (!tchunk
->tsn_gap_acked
&&
1473 !sctp_chunk_retransmitted(tchunk
) &&
1474 tchunk
->rtt_in_progress
) {
1475 tchunk
->rtt_in_progress
= 0;
1476 rtt
= jiffies
- tchunk
->sent_at
;
1477 sctp_transport_update_rto(transport
,
1482 /* If the chunk hasn't been marked as ACKED,
1483 * mark it and account bytes_acked if the
1484 * chunk had a valid transport (it will not
1485 * have a transport if ASCONF had deleted it
1486 * while DATA was outstanding).
1488 if (!tchunk
->tsn_gap_acked
) {
1489 tchunk
->tsn_gap_acked
= 1;
1490 if (TSN_lt(*highest_new_tsn_in_sack
, tsn
))
1491 *highest_new_tsn_in_sack
= tsn
;
1492 bytes_acked
+= sctp_data_size(tchunk
);
1493 if (!tchunk
->transport
)
1494 migrate_bytes
+= sctp_data_size(tchunk
);
1495 forward_progress
= true;
1498 if (TSN_lte(tsn
, sack_ctsn
)) {
1499 /* RFC 2960 6.3.2 Retransmission Timer Rules
1501 * R3) Whenever a SACK is received
1502 * that acknowledges the DATA chunk
1503 * with the earliest outstanding TSN
1504 * for that address, restart T3-rtx
1505 * timer for that address with its
1509 forward_progress
= true;
1511 if (!tchunk
->tsn_gap_acked
) {
1513 * SFR-CACC algorithm:
1514 * 2) If the SACK contains gap acks
1515 * and the flag CHANGEOVER_ACTIVE is
1516 * set the receiver of the SACK MUST
1517 * take the following action:
1519 * B) For each TSN t being acked that
1520 * has not been acked in any SACK so
1521 * far, set cacc_saw_newack to 1 for
1522 * the destination that the TSN was
1526 sack
->num_gap_ack_blocks
&&
1527 q
->asoc
->peer
.primary_path
->cacc
.
1529 transport
->cacc
.cacc_saw_newack
1533 list_add_tail(&tchunk
->transmitted_list
,
1536 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1537 * M2) Each time a SACK arrives reporting
1538 * 'Stray DATA chunk(s)' record the highest TSN
1539 * reported as newly acknowledged, call this
1540 * value 'HighestTSNinSack'. A newly
1541 * acknowledged DATA chunk is one not
1542 * previously acknowledged in a SACK.
1544 * When the SCTP sender of data receives a SACK
1545 * chunk that acknowledges, for the first time,
1546 * the receipt of a DATA chunk, all the still
1547 * unacknowledged DATA chunks whose TSN is
1548 * older than that newly acknowledged DATA
1549 * chunk, are qualified as 'Stray DATA chunks'.
1551 list_add_tail(lchunk
, &tlist
);
1554 if (tchunk
->tsn_gap_acked
) {
1555 pr_debug("%s: receiver reneged on data TSN:0x%x\n",
1558 tchunk
->tsn_gap_acked
= 0;
1560 if (tchunk
->transport
)
1561 bytes_acked
-= sctp_data_size(tchunk
);
1563 /* RFC 2960 6.3.2 Retransmission Timer Rules
1565 * R4) Whenever a SACK is received missing a
1566 * TSN that was previously acknowledged via a
1567 * Gap Ack Block, start T3-rtx for the
1568 * destination address to which the DATA
1569 * chunk was originally
1570 * transmitted if it is not already running.
1575 list_add_tail(lchunk
, &tlist
);
1581 struct sctp_association
*asoc
= transport
->asoc
;
1583 /* We may have counted DATA that was migrated
1584 * to this transport due to DEL-IP operation.
1585 * Subtract those bytes, since the were never
1586 * send on this transport and shouldn't be
1587 * credited to this transport.
1589 bytes_acked
-= migrate_bytes
;
1591 /* 8.2. When an outstanding TSN is acknowledged,
1592 * the endpoint shall clear the error counter of
1593 * the destination transport address to which the
1594 * DATA chunk was last sent.
1595 * The association's overall error counter is
1598 transport
->error_count
= 0;
1599 transport
->asoc
->overall_error_count
= 0;
1600 forward_progress
= true;
1603 * While in SHUTDOWN PENDING, we may have started
1604 * the T5 shutdown guard timer after reaching the
1605 * retransmission limit. Stop that timer as soon
1606 * as the receiver acknowledged any data.
1608 if (asoc
->state
== SCTP_STATE_SHUTDOWN_PENDING
&&
1609 del_timer(&asoc
->timers
1610 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD
]))
1611 sctp_association_put(asoc
);
1613 /* Mark the destination transport address as
1614 * active if it is not so marked.
1616 if ((transport
->state
== SCTP_INACTIVE
||
1617 transport
->state
== SCTP_UNCONFIRMED
) &&
1618 sctp_cmp_addr_exact(&transport
->ipaddr
, saddr
)) {
1619 sctp_assoc_control_transport(
1623 SCTP_RECEIVED_SACK
);
1626 sctp_transport_raise_cwnd(transport
, sack_ctsn
,
1629 transport
->flight_size
-= bytes_acked
;
1630 if (transport
->flight_size
== 0)
1631 transport
->partial_bytes_acked
= 0;
1632 q
->outstanding_bytes
-= bytes_acked
+ migrate_bytes
;
1634 /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1635 * When a sender is doing zero window probing, it
1636 * should not timeout the association if it continues
1637 * to receive new packets from the receiver. The
1638 * reason is that the receiver MAY keep its window
1639 * closed for an indefinite time.
1640 * A sender is doing zero window probing when the
1641 * receiver's advertised window is zero, and there is
1642 * only one data chunk in flight to the receiver.
1644 * Allow the association to timeout while in SHUTDOWN
1645 * PENDING or SHUTDOWN RECEIVED in case the receiver
1646 * stays in zero window mode forever.
1648 if (!q
->asoc
->peer
.rwnd
&&
1649 !list_empty(&tlist
) &&
1650 (sack_ctsn
+2 == q
->asoc
->next_tsn
) &&
1651 q
->asoc
->state
< SCTP_STATE_SHUTDOWN_PENDING
) {
1652 pr_debug("%s: sack received for zero window "
1653 "probe:%u\n", __func__
, sack_ctsn
);
1655 q
->asoc
->overall_error_count
= 0;
1656 transport
->error_count
= 0;
1660 /* RFC 2960 6.3.2 Retransmission Timer Rules
1662 * R2) Whenever all outstanding data sent to an address have
1663 * been acknowledged, turn off the T3-rtx timer of that
1666 if (!transport
->flight_size
) {
1667 if (del_timer(&transport
->T3_rtx_timer
))
1668 sctp_transport_put(transport
);
1669 } else if (restart_timer
) {
1670 if (!mod_timer(&transport
->T3_rtx_timer
,
1671 jiffies
+ transport
->rto
))
1672 sctp_transport_hold(transport
);
1675 if (forward_progress
) {
1677 sctp_transport_dst_confirm(transport
);
1681 list_splice(&tlist
, transmitted_queue
);
1684 /* Mark chunks as missing and consequently may get retransmitted. */
1685 static void sctp_mark_missing(struct sctp_outq
*q
,
1686 struct list_head
*transmitted_queue
,
1687 struct sctp_transport
*transport
,
1688 __u32 highest_new_tsn_in_sack
,
1689 int count_of_newacks
)
1691 struct sctp_chunk
*chunk
;
1693 char do_fast_retransmit
= 0;
1694 struct sctp_association
*asoc
= q
->asoc
;
1695 struct sctp_transport
*primary
= asoc
->peer
.primary_path
;
1697 list_for_each_entry(chunk
, transmitted_queue
, transmitted_list
) {
1699 tsn
= ntohl(chunk
->subh
.data_hdr
->tsn
);
1701 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1702 * 'Unacknowledged TSN's', if the TSN number of an
1703 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1704 * value, increment the 'TSN.Missing.Report' count on that
1705 * chunk if it has NOT been fast retransmitted or marked for
1706 * fast retransmit already.
1708 if (chunk
->fast_retransmit
== SCTP_CAN_FRTX
&&
1709 !chunk
->tsn_gap_acked
&&
1710 TSN_lt(tsn
, highest_new_tsn_in_sack
)) {
1712 /* SFR-CACC may require us to skip marking
1713 * this chunk as missing.
1715 if (!transport
|| !sctp_cacc_skip(primary
,
1717 count_of_newacks
, tsn
)) {
1718 chunk
->tsn_missing_report
++;
1720 pr_debug("%s: tsn:0x%x missing counter:%d\n",
1721 __func__
, tsn
, chunk
->tsn_missing_report
);
1725 * M4) If any DATA chunk is found to have a
1726 * 'TSN.Missing.Report'
1727 * value larger than or equal to 3, mark that chunk for
1728 * retransmission and start the fast retransmit procedure.
1731 if (chunk
->tsn_missing_report
>= 3) {
1732 chunk
->fast_retransmit
= SCTP_NEED_FRTX
;
1733 do_fast_retransmit
= 1;
1738 if (do_fast_retransmit
)
1739 sctp_retransmit(q
, transport
, SCTP_RTXR_FAST_RTX
);
1741 pr_debug("%s: transport:%p, cwnd:%d, ssthresh:%d, "
1742 "flight_size:%d, pba:%d\n", __func__
, transport
,
1743 transport
->cwnd
, transport
->ssthresh
,
1744 transport
->flight_size
, transport
->partial_bytes_acked
);
1748 /* Is the given TSN acked by this packet? */
1749 static int sctp_acked(struct sctp_sackhdr
*sack
, __u32 tsn
)
1751 __u32 ctsn
= ntohl(sack
->cum_tsn_ack
);
1752 union sctp_sack_variable
*frags
;
1753 __u16 tsn_offset
, blocks
;
1756 if (TSN_lte(tsn
, ctsn
))
1759 /* 3.3.4 Selective Acknowledgement (SACK) (3):
1762 * These fields contain the Gap Ack Blocks. They are repeated
1763 * for each Gap Ack Block up to the number of Gap Ack Blocks
1764 * defined in the Number of Gap Ack Blocks field. All DATA
1765 * chunks with TSNs greater than or equal to (Cumulative TSN
1766 * Ack + Gap Ack Block Start) and less than or equal to
1767 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1768 * Block are assumed to have been received correctly.
1771 frags
= sack
->variable
;
1772 blocks
= ntohs(sack
->num_gap_ack_blocks
);
1773 tsn_offset
= tsn
- ctsn
;
1774 for (i
= 0; i
< blocks
; ++i
) {
1775 if (tsn_offset
>= ntohs(frags
[i
].gab
.start
) &&
1776 tsn_offset
<= ntohs(frags
[i
].gab
.end
))
1785 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip
*skiplist
,
1786 int nskips
, __be16 stream
)
1790 for (i
= 0; i
< nskips
; i
++) {
1791 if (skiplist
[i
].stream
== stream
)
1797 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1798 static void sctp_generate_fwdtsn(struct sctp_outq
*q
, __u32 ctsn
)
1800 struct sctp_association
*asoc
= q
->asoc
;
1801 struct sctp_chunk
*ftsn_chunk
= NULL
;
1802 struct sctp_fwdtsn_skip ftsn_skip_arr
[10];
1806 struct sctp_chunk
*chunk
;
1807 struct list_head
*lchunk
, *temp
;
1809 if (!asoc
->peer
.prsctp_capable
)
1812 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1815 * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1816 * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1818 if (TSN_lt(asoc
->adv_peer_ack_point
, ctsn
))
1819 asoc
->adv_peer_ack_point
= ctsn
;
1821 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1822 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1823 * the chunk next in the out-queue space is marked as "abandoned" as
1824 * shown in the following example:
1826 * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1827 * and the Advanced.Peer.Ack.Point is updated to this value:
1829 * out-queue at the end of ==> out-queue after Adv.Ack.Point
1830 * normal SACK processing local advancement
1832 * Adv.Ack.Pt-> 102 acked 102 acked
1833 * 103 abandoned 103 abandoned
1834 * 104 abandoned Adv.Ack.P-> 104 abandoned
1836 * 106 acked 106 acked
1839 * In this example, the data sender successfully advanced the
1840 * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1842 list_for_each_safe(lchunk
, temp
, &q
->abandoned
) {
1843 chunk
= list_entry(lchunk
, struct sctp_chunk
,
1845 tsn
= ntohl(chunk
->subh
.data_hdr
->tsn
);
1847 /* Remove any chunks in the abandoned queue that are acked by
1850 if (TSN_lte(tsn
, ctsn
)) {
1851 list_del_init(lchunk
);
1852 sctp_chunk_free(chunk
);
1854 if (TSN_lte(tsn
, asoc
->adv_peer_ack_point
+1)) {
1855 asoc
->adv_peer_ack_point
= tsn
;
1856 if (chunk
->chunk_hdr
->flags
&
1857 SCTP_DATA_UNORDERED
)
1859 skip_pos
= sctp_get_skip_pos(&ftsn_skip_arr
[0],
1861 chunk
->subh
.data_hdr
->stream
);
1862 ftsn_skip_arr
[skip_pos
].stream
=
1863 chunk
->subh
.data_hdr
->stream
;
1864 ftsn_skip_arr
[skip_pos
].ssn
=
1865 chunk
->subh
.data_hdr
->ssn
;
1866 if (skip_pos
== nskips
)
1875 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1876 * is greater than the Cumulative TSN ACK carried in the received
1877 * SACK, the data sender MUST send the data receiver a FORWARD TSN
1878 * chunk containing the latest value of the
1879 * "Advanced.Peer.Ack.Point".
1881 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1882 * list each stream and sequence number in the forwarded TSN. This
1883 * information will enable the receiver to easily find any
1884 * stranded TSN's waiting on stream reorder queues. Each stream
1885 * SHOULD only be reported once; this means that if multiple
1886 * abandoned messages occur in the same stream then only the
1887 * highest abandoned stream sequence number is reported. If the
1888 * total size of the FORWARD TSN does NOT fit in a single MTU then
1889 * the sender of the FORWARD TSN SHOULD lower the
1890 * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1893 if (asoc
->adv_peer_ack_point
> ctsn
)
1894 ftsn_chunk
= sctp_make_fwdtsn(asoc
, asoc
->adv_peer_ack_point
,
1895 nskips
, &ftsn_skip_arr
[0]);
1898 list_add_tail(&ftsn_chunk
->list
, &q
->control_chunk_list
);
1899 SCTP_INC_STATS(sock_net(asoc
->base
.sk
), SCTP_MIB_OUTCTRLCHUNKS
);