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, write to
26 * the Free Software Foundation, 59 Temple Place - Suite 330,
27 * Boston, MA 02111-1307, USA.
29 * Please send any bug reports or fixes you make to the
31 * lksctp developers <lksctp-developers@lists.sourceforge.net>
33 * Or submit a bug report through the following website:
34 * http://www.sf.net/projects/lksctp
36 * Written or modified by:
37 * La Monte H.P. Yarroll <piggy@acm.org>
38 * Karl Knutson <karl@athena.chicago.il.us>
39 * Perry Melange <pmelange@null.cc.uic.edu>
40 * Xingang Guo <xingang.guo@intel.com>
41 * Hui Huang <hui.huang@nokia.com>
42 * Sridhar Samudrala <sri@us.ibm.com>
43 * Jon Grimm <jgrimm@us.ibm.com>
45 * Any bugs reported given to us we will try to fix... any fixes shared will
46 * be incorporated into the next SCTP release.
49 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
51 #include <linux/types.h>
52 #include <linux/list.h> /* For struct list_head */
53 #include <linux/socket.h>
55 #include <linux/slab.h>
56 #include <net/sock.h> /* For skb_set_owner_w */
58 #include <net/sctp/sctp.h>
59 #include <net/sctp/sm.h>
61 /* Declare internal functions here. */
62 static int sctp_acked(struct sctp_sackhdr
*sack
, __u32 tsn
);
63 static void sctp_check_transmitted(struct sctp_outq
*q
,
64 struct list_head
*transmitted_queue
,
65 struct sctp_transport
*transport
,
66 union sctp_addr
*saddr
,
67 struct sctp_sackhdr
*sack
,
68 __u32
*highest_new_tsn
);
70 static void sctp_mark_missing(struct sctp_outq
*q
,
71 struct list_head
*transmitted_queue
,
72 struct sctp_transport
*transport
,
73 __u32 highest_new_tsn
,
74 int count_of_newacks
);
76 static void sctp_generate_fwdtsn(struct sctp_outq
*q
, __u32 sack_ctsn
);
78 static int sctp_outq_flush(struct sctp_outq
*q
, int rtx_timeout
);
80 /* Add data to the front of the queue. */
81 static inline void sctp_outq_head_data(struct sctp_outq
*q
,
82 struct sctp_chunk
*ch
)
84 list_add(&ch
->list
, &q
->out_chunk_list
);
85 q
->out_qlen
+= ch
->skb
->len
;
88 /* Take data from the front of the queue. */
89 static inline struct sctp_chunk
*sctp_outq_dequeue_data(struct sctp_outq
*q
)
91 struct sctp_chunk
*ch
= NULL
;
93 if (!list_empty(&q
->out_chunk_list
)) {
94 struct list_head
*entry
= q
->out_chunk_list
.next
;
96 ch
= list_entry(entry
, struct sctp_chunk
, list
);
98 q
->out_qlen
-= ch
->skb
->len
;
102 /* Add data chunk to the end of the queue. */
103 static inline void sctp_outq_tail_data(struct sctp_outq
*q
,
104 struct sctp_chunk
*ch
)
106 list_add_tail(&ch
->list
, &q
->out_chunk_list
);
107 q
->out_qlen
+= ch
->skb
->len
;
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
)
210 INIT_LIST_HEAD(&q
->out_chunk_list
);
211 INIT_LIST_HEAD(&q
->control_chunk_list
);
212 INIT_LIST_HEAD(&q
->retransmit
);
213 INIT_LIST_HEAD(&q
->sacked
);
214 INIT_LIST_HEAD(&q
->abandoned
);
217 q
->outstanding_bytes
= 0;
225 /* Free the outqueue structure and any related pending chunks.
227 void sctp_outq_teardown(struct sctp_outq
*q
)
229 struct sctp_transport
*transport
;
230 struct list_head
*lchunk
, *temp
;
231 struct sctp_chunk
*chunk
, *tmp
;
233 /* Throw away unacknowledged chunks. */
234 list_for_each_entry(transport
, &q
->asoc
->peer
.transport_addr_list
,
236 while ((lchunk
= sctp_list_dequeue(&transport
->transmitted
)) != NULL
) {
237 chunk
= list_entry(lchunk
, struct sctp_chunk
,
239 /* Mark as part of a failed message. */
240 sctp_chunk_fail(chunk
, q
->error
);
241 sctp_chunk_free(chunk
);
245 /* Throw away chunks that have been gap ACKed. */
246 list_for_each_safe(lchunk
, temp
, &q
->sacked
) {
247 list_del_init(lchunk
);
248 chunk
= list_entry(lchunk
, struct sctp_chunk
,
250 sctp_chunk_fail(chunk
, q
->error
);
251 sctp_chunk_free(chunk
);
254 /* Throw away any chunks in the retransmit queue. */
255 list_for_each_safe(lchunk
, temp
, &q
->retransmit
) {
256 list_del_init(lchunk
);
257 chunk
= list_entry(lchunk
, struct sctp_chunk
,
259 sctp_chunk_fail(chunk
, q
->error
);
260 sctp_chunk_free(chunk
);
263 /* Throw away any chunks that are in the abandoned queue. */
264 list_for_each_safe(lchunk
, temp
, &q
->abandoned
) {
265 list_del_init(lchunk
);
266 chunk
= list_entry(lchunk
, struct sctp_chunk
,
268 sctp_chunk_fail(chunk
, q
->error
);
269 sctp_chunk_free(chunk
);
272 /* Throw away any leftover data chunks. */
273 while ((chunk
= sctp_outq_dequeue_data(q
)) != NULL
) {
275 /* Mark as send failure. */
276 sctp_chunk_fail(chunk
, q
->error
);
277 sctp_chunk_free(chunk
);
282 /* Throw away any leftover control chunks. */
283 list_for_each_entry_safe(chunk
, tmp
, &q
->control_chunk_list
, list
) {
284 list_del_init(&chunk
->list
);
285 sctp_chunk_free(chunk
);
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
);
295 /* If we were kmalloc()'d, free the memory. */
300 /* Put a new chunk in an sctp_outq. */
301 int sctp_outq_tail(struct sctp_outq
*q
, struct sctp_chunk
*chunk
)
303 struct net
*net
= sock_net(q
->asoc
->base
.sk
);
306 SCTP_DEBUG_PRINTK("sctp_outq_tail(%p, %p[%s])\n",
307 q
, chunk
, chunk
&& chunk
->chunk_hdr
?
308 sctp_cname(SCTP_ST_CHUNK(chunk
->chunk_hdr
->type
))
311 /* If it is data, queue it up, otherwise, send it
314 if (sctp_chunk_is_data(chunk
)) {
315 /* Is it OK to queue data chunks? */
316 /* From 9. Termination of Association
318 * When either endpoint performs a shutdown, the
319 * association on each peer will stop accepting new
320 * data from its user and only deliver data in queue
321 * at the time of sending or receiving the SHUTDOWN
324 switch (q
->asoc
->state
) {
325 case SCTP_STATE_CLOSED
:
326 case SCTP_STATE_SHUTDOWN_PENDING
:
327 case SCTP_STATE_SHUTDOWN_SENT
:
328 case SCTP_STATE_SHUTDOWN_RECEIVED
:
329 case SCTP_STATE_SHUTDOWN_ACK_SENT
:
330 /* Cannot send after transport endpoint shutdown */
335 SCTP_DEBUG_PRINTK("outqueueing (%p, %p[%s])\n",
336 q
, chunk
, chunk
&& chunk
->chunk_hdr
?
337 sctp_cname(SCTP_ST_CHUNK(chunk
->chunk_hdr
->type
))
340 sctp_outq_tail_data(q
, chunk
);
341 if (chunk
->chunk_hdr
->flags
& SCTP_DATA_UNORDERED
)
342 SCTP_INC_STATS(net
, SCTP_MIB_OUTUNORDERCHUNKS
);
344 SCTP_INC_STATS(net
, SCTP_MIB_OUTORDERCHUNKS
);
349 list_add_tail(&chunk
->list
, &q
->control_chunk_list
);
350 SCTP_INC_STATS(net
, SCTP_MIB_OUTCTRLCHUNKS
);
357 error
= sctp_outq_flush(q
, 0);
362 /* Insert a chunk into the sorted list based on the TSNs. The retransmit list
363 * and the abandoned list are in ascending order.
365 static void sctp_insert_list(struct list_head
*head
, struct list_head
*new)
367 struct list_head
*pos
;
368 struct sctp_chunk
*nchunk
, *lchunk
;
372 nchunk
= list_entry(new, struct sctp_chunk
, transmitted_list
);
373 ntsn
= ntohl(nchunk
->subh
.data_hdr
->tsn
);
375 list_for_each(pos
, head
) {
376 lchunk
= list_entry(pos
, struct sctp_chunk
, transmitted_list
);
377 ltsn
= ntohl(lchunk
->subh
.data_hdr
->tsn
);
378 if (TSN_lt(ntsn
, ltsn
)) {
379 list_add(new, pos
->prev
);
385 list_add_tail(new, head
);
388 /* Mark all the eligible packets on a transport for retransmission. */
389 void sctp_retransmit_mark(struct sctp_outq
*q
,
390 struct sctp_transport
*transport
,
393 struct list_head
*lchunk
, *ltemp
;
394 struct sctp_chunk
*chunk
;
396 /* Walk through the specified transmitted queue. */
397 list_for_each_safe(lchunk
, ltemp
, &transport
->transmitted
) {
398 chunk
= list_entry(lchunk
, struct sctp_chunk
,
401 /* If the chunk is abandoned, move it to abandoned list. */
402 if (sctp_chunk_abandoned(chunk
)) {
403 list_del_init(lchunk
);
404 sctp_insert_list(&q
->abandoned
, lchunk
);
406 /* If this chunk has not been previousely acked,
407 * stop considering it 'outstanding'. Our peer
408 * will most likely never see it since it will
409 * not be retransmitted
411 if (!chunk
->tsn_gap_acked
) {
412 if (chunk
->transport
)
413 chunk
->transport
->flight_size
-=
414 sctp_data_size(chunk
);
415 q
->outstanding_bytes
-= sctp_data_size(chunk
);
416 q
->asoc
->peer
.rwnd
+= sctp_data_size(chunk
);
421 /* If we are doing retransmission due to a timeout or pmtu
422 * discovery, only the chunks that are not yet acked should
423 * be added to the retransmit queue.
425 if ((reason
== SCTP_RTXR_FAST_RTX
&&
426 (chunk
->fast_retransmit
== SCTP_NEED_FRTX
)) ||
427 (reason
!= SCTP_RTXR_FAST_RTX
&& !chunk
->tsn_gap_acked
)) {
428 /* RFC 2960 6.2.1 Processing a Received SACK
430 * C) Any time a DATA chunk is marked for
431 * retransmission (via either T3-rtx timer expiration
432 * (Section 6.3.3) or via fast retransmit
433 * (Section 7.2.4)), add the data size of those
434 * chunks to the rwnd.
436 q
->asoc
->peer
.rwnd
+= sctp_data_size(chunk
);
437 q
->outstanding_bytes
-= sctp_data_size(chunk
);
438 if (chunk
->transport
)
439 transport
->flight_size
-= sctp_data_size(chunk
);
441 /* sctpimpguide-05 Section 2.8.2
442 * M5) If a T3-rtx timer expires, the
443 * 'TSN.Missing.Report' of all affected TSNs is set
446 chunk
->tsn_missing_report
= 0;
448 /* If a chunk that is being used for RTT measurement
449 * has to be retransmitted, we cannot use this chunk
450 * anymore for RTT measurements. Reset rto_pending so
451 * that a new RTT measurement is started when a new
452 * data chunk is sent.
454 if (chunk
->rtt_in_progress
) {
455 chunk
->rtt_in_progress
= 0;
456 transport
->rto_pending
= 0;
459 /* Move the chunk to the retransmit queue. The chunks
460 * on the retransmit queue are always kept in order.
462 list_del_init(lchunk
);
463 sctp_insert_list(&q
->retransmit
, lchunk
);
467 SCTP_DEBUG_PRINTK("%s: transport: %p, reason: %d, "
468 "cwnd: %d, ssthresh: %d, flight_size: %d, "
469 "pba: %d\n", __func__
,
471 transport
->cwnd
, transport
->ssthresh
,
472 transport
->flight_size
,
473 transport
->partial_bytes_acked
);
477 /* Mark all the eligible packets on a transport for retransmission and force
480 void sctp_retransmit(struct sctp_outq
*q
, struct sctp_transport
*transport
,
481 sctp_retransmit_reason_t reason
)
483 struct net
*net
= sock_net(q
->asoc
->base
.sk
);
487 case SCTP_RTXR_T3_RTX
:
488 SCTP_INC_STATS(net
, SCTP_MIB_T3_RETRANSMITS
);
489 sctp_transport_lower_cwnd(transport
, SCTP_LOWER_CWND_T3_RTX
);
490 /* Update the retran path if the T3-rtx timer has expired for
491 * the current retran path.
493 if (transport
== transport
->asoc
->peer
.retran_path
)
494 sctp_assoc_update_retran_path(transport
->asoc
);
495 transport
->asoc
->rtx_data_chunks
+=
496 transport
->asoc
->unack_data
;
498 case SCTP_RTXR_FAST_RTX
:
499 SCTP_INC_STATS(net
, SCTP_MIB_FAST_RETRANSMITS
);
500 sctp_transport_lower_cwnd(transport
, SCTP_LOWER_CWND_FAST_RTX
);
503 case SCTP_RTXR_PMTUD
:
504 SCTP_INC_STATS(net
, SCTP_MIB_PMTUD_RETRANSMITS
);
506 case SCTP_RTXR_T1_RTX
:
507 SCTP_INC_STATS(net
, SCTP_MIB_T1_RETRANSMITS
);
508 transport
->asoc
->init_retries
++;
514 sctp_retransmit_mark(q
, transport
, reason
);
516 /* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
517 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
518 * following the procedures outlined in C1 - C5.
520 if (reason
== SCTP_RTXR_T3_RTX
)
521 sctp_generate_fwdtsn(q
, q
->asoc
->ctsn_ack_point
);
523 /* Flush the queues only on timeout, since fast_rtx is only
524 * triggered during sack processing and the queue
525 * will be flushed at the end.
527 if (reason
!= SCTP_RTXR_FAST_RTX
)
528 error
= sctp_outq_flush(q
, /* rtx_timeout */ 1);
531 q
->asoc
->base
.sk
->sk_err
= -error
;
535 * Transmit DATA chunks on the retransmit queue. Upon return from
536 * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
537 * need to be transmitted by the caller.
538 * We assume that pkt->transport has already been set.
540 * The return value is a normal kernel error return value.
542 static int sctp_outq_flush_rtx(struct sctp_outq
*q
, struct sctp_packet
*pkt
,
543 int rtx_timeout
, int *start_timer
)
545 struct list_head
*lqueue
;
546 struct sctp_transport
*transport
= pkt
->transport
;
548 struct sctp_chunk
*chunk
, *chunk1
;
554 lqueue
= &q
->retransmit
;
555 fast_rtx
= q
->fast_rtx
;
557 /* This loop handles time-out retransmissions, fast retransmissions,
558 * and retransmissions due to opening of whindow.
560 * RFC 2960 6.3.3 Handle T3-rtx Expiration
562 * E3) Determine how many of the earliest (i.e., lowest TSN)
563 * outstanding DATA chunks for the address for which the
564 * T3-rtx has expired will fit into a single packet, subject
565 * to the MTU constraint for the path corresponding to the
566 * destination transport address to which the retransmission
567 * is being sent (this may be different from the address for
568 * which the timer expires [see Section 6.4]). Call this value
569 * K. Bundle and retransmit those K DATA chunks in a single
570 * packet to the destination endpoint.
572 * [Just to be painfully clear, if we are retransmitting
573 * because a timeout just happened, we should send only ONE
574 * packet of retransmitted data.]
576 * For fast retransmissions we also send only ONE packet. However,
577 * if we are just flushing the queue due to open window, we'll
578 * try to send as much as possible.
580 list_for_each_entry_safe(chunk
, chunk1
, lqueue
, transmitted_list
) {
581 /* If the chunk is abandoned, move it to abandoned list. */
582 if (sctp_chunk_abandoned(chunk
)) {
583 list_del_init(&chunk
->transmitted_list
);
584 sctp_insert_list(&q
->abandoned
,
585 &chunk
->transmitted_list
);
589 /* Make sure that Gap Acked TSNs are not retransmitted. A
590 * simple approach is just to move such TSNs out of the
591 * way and into a 'transmitted' queue and skip to the
594 if (chunk
->tsn_gap_acked
) {
595 list_move_tail(&chunk
->transmitted_list
,
596 &transport
->transmitted
);
600 /* If we are doing fast retransmit, ignore non-fast_rtransmit
603 if (fast_rtx
&& !chunk
->fast_retransmit
)
607 /* Attempt to append this chunk to the packet. */
608 status
= sctp_packet_append_chunk(pkt
, chunk
);
611 case SCTP_XMIT_PMTU_FULL
:
612 if (!pkt
->has_data
&& !pkt
->has_cookie_echo
) {
613 /* If this packet did not contain DATA then
614 * retransmission did not happen, so do it
615 * again. We'll ignore the error here since
616 * control chunks are already freed so there
617 * is nothing we can do.
619 sctp_packet_transmit(pkt
);
623 /* Send this packet. */
624 error
= sctp_packet_transmit(pkt
);
626 /* If we are retransmitting, we should only
627 * send a single packet.
628 * Otherwise, try appending this chunk again.
630 if (rtx_timeout
|| fast_rtx
)
635 /* Bundle next chunk in the next round. */
638 case SCTP_XMIT_RWND_FULL
:
639 /* Send this packet. */
640 error
= sctp_packet_transmit(pkt
);
642 /* Stop sending DATA as there is no more room
648 case SCTP_XMIT_NAGLE_DELAY
:
649 /* Send this packet. */
650 error
= sctp_packet_transmit(pkt
);
652 /* Stop sending DATA because of nagle delay. */
657 /* The append was successful, so add this chunk to
658 * the transmitted list.
660 list_move_tail(&chunk
->transmitted_list
,
661 &transport
->transmitted
);
663 /* Mark the chunk as ineligible for fast retransmit
664 * after it is retransmitted.
666 if (chunk
->fast_retransmit
== SCTP_NEED_FRTX
)
667 chunk
->fast_retransmit
= SCTP_DONT_FRTX
;
673 /* Set the timer if there were no errors */
674 if (!error
&& !timer
)
681 /* If we are here due to a retransmit timeout or a fast
682 * retransmit and if there are any chunks left in the retransmit
683 * queue that could not fit in the PMTU sized packet, they need
684 * to be marked as ineligible for a subsequent fast retransmit.
686 if (rtx_timeout
|| fast_rtx
) {
687 list_for_each_entry(chunk1
, lqueue
, transmitted_list
) {
688 if (chunk1
->fast_retransmit
== SCTP_NEED_FRTX
)
689 chunk1
->fast_retransmit
= SCTP_DONT_FRTX
;
693 *start_timer
= timer
;
695 /* Clear fast retransmit hint */
702 /* Cork the outqueue so queued chunks are really queued. */
703 int sctp_outq_uncork(struct sctp_outq
*q
)
708 error
= sctp_outq_flush(q
, 0);
714 * Try to flush an outqueue.
716 * Description: Send everything in q which we legally can, subject to
717 * congestion limitations.
718 * * Note: This function can be called from multiple contexts so appropriate
719 * locking concerns must be made. Today we use the sock lock to protect
722 static int sctp_outq_flush(struct sctp_outq
*q
, int rtx_timeout
)
724 struct sctp_packet
*packet
;
725 struct sctp_packet singleton
;
726 struct sctp_association
*asoc
= q
->asoc
;
727 __u16 sport
= asoc
->base
.bind_addr
.port
;
728 __u16 dport
= asoc
->peer
.port
;
729 __u32 vtag
= asoc
->peer
.i
.init_tag
;
730 struct sctp_transport
*transport
= NULL
;
731 struct sctp_transport
*new_transport
;
732 struct sctp_chunk
*chunk
, *tmp
;
738 /* These transports have chunks to send. */
739 struct list_head transport_list
;
740 struct list_head
*ltransport
;
742 INIT_LIST_HEAD(&transport_list
);
748 * When bundling control chunks with DATA chunks, an
749 * endpoint MUST place control chunks first in the outbound
750 * SCTP packet. The transmitter MUST transmit DATA chunks
751 * within a SCTP packet in increasing order of TSN.
755 list_for_each_entry_safe(chunk
, tmp
, &q
->control_chunk_list
, list
) {
757 * F1) This means that until such time as the ASCONF
758 * containing the add is acknowledged, the sender MUST
759 * NOT use the new IP address as a source for ANY SCTP
760 * packet except on carrying an ASCONF Chunk.
762 if (asoc
->src_out_of_asoc_ok
&&
763 chunk
->chunk_hdr
->type
!= SCTP_CID_ASCONF
)
766 list_del_init(&chunk
->list
);
768 /* Pick the right transport to use. */
769 new_transport
= chunk
->transport
;
771 if (!new_transport
) {
773 * If we have a prior transport pointer, see if
774 * the destination address of the chunk
775 * matches the destination address of the
776 * current transport. If not a match, then
777 * try to look up the transport with a given
778 * destination address. We do this because
779 * after processing ASCONFs, we may have new
780 * transports created.
783 sctp_cmp_addr_exact(&chunk
->dest
,
785 new_transport
= transport
;
787 new_transport
= sctp_assoc_lookup_paddr(asoc
,
790 /* if we still don't have a new transport, then
791 * use the current active path.
794 new_transport
= asoc
->peer
.active_path
;
795 } else if ((new_transport
->state
== SCTP_INACTIVE
) ||
796 (new_transport
->state
== SCTP_UNCONFIRMED
) ||
797 (new_transport
->state
== SCTP_PF
)) {
798 /* If the chunk is Heartbeat or Heartbeat Ack,
799 * send it to chunk->transport, even if it's
802 * 3.3.6 Heartbeat Acknowledgement:
804 * A HEARTBEAT ACK is always sent to the source IP
805 * address of the IP datagram containing the
806 * HEARTBEAT chunk to which this ack is responding.
809 * ASCONF_ACKs also must be sent to the source.
811 if (chunk
->chunk_hdr
->type
!= SCTP_CID_HEARTBEAT
&&
812 chunk
->chunk_hdr
->type
!= SCTP_CID_HEARTBEAT_ACK
&&
813 chunk
->chunk_hdr
->type
!= SCTP_CID_ASCONF_ACK
)
814 new_transport
= asoc
->peer
.active_path
;
817 /* Are we switching transports?
818 * Take care of transport locks.
820 if (new_transport
!= transport
) {
821 transport
= new_transport
;
822 if (list_empty(&transport
->send_ready
)) {
823 list_add_tail(&transport
->send_ready
,
826 packet
= &transport
->packet
;
827 sctp_packet_config(packet
, vtag
,
828 asoc
->peer
.ecn_capable
);
831 switch (chunk
->chunk_hdr
->type
) {
835 * An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
836 * COMPLETE with any other chunks. [Send them immediately.]
839 case SCTP_CID_INIT_ACK
:
840 case SCTP_CID_SHUTDOWN_COMPLETE
:
841 sctp_packet_init(&singleton
, transport
, sport
, dport
);
842 sctp_packet_config(&singleton
, vtag
, 0);
843 sctp_packet_append_chunk(&singleton
, chunk
);
844 error
= sctp_packet_transmit(&singleton
);
850 if (sctp_test_T_bit(chunk
)) {
851 packet
->vtag
= asoc
->c
.my_vtag
;
853 /* The following chunks are "response" chunks, i.e.
854 * they are generated in response to something we
855 * received. If we are sending these, then we can
856 * send only 1 packet containing these chunks.
858 case SCTP_CID_HEARTBEAT_ACK
:
859 case SCTP_CID_SHUTDOWN_ACK
:
860 case SCTP_CID_COOKIE_ACK
:
861 case SCTP_CID_COOKIE_ECHO
:
863 case SCTP_CID_ECN_CWR
:
864 case SCTP_CID_ASCONF_ACK
:
869 case SCTP_CID_HEARTBEAT
:
870 case SCTP_CID_SHUTDOWN
:
871 case SCTP_CID_ECN_ECNE
:
872 case SCTP_CID_ASCONF
:
873 case SCTP_CID_FWD_TSN
:
874 status
= sctp_packet_transmit_chunk(packet
, chunk
,
876 if (status
!= SCTP_XMIT_OK
) {
877 /* put the chunk back */
878 list_add(&chunk
->list
, &q
->control_chunk_list
);
879 } else if (chunk
->chunk_hdr
->type
== SCTP_CID_FWD_TSN
) {
880 /* PR-SCTP C5) If a FORWARD TSN is sent, the
881 * sender MUST assure that at least one T3-rtx
884 sctp_transport_reset_timers(transport
);
889 /* We built a chunk with an illegal type! */
894 if (q
->asoc
->src_out_of_asoc_ok
)
897 /* Is it OK to send data chunks? */
898 switch (asoc
->state
) {
899 case SCTP_STATE_COOKIE_ECHOED
:
900 /* Only allow bundling when this packet has a COOKIE-ECHO
903 if (!packet
|| !packet
->has_cookie_echo
)
907 case SCTP_STATE_ESTABLISHED
:
908 case SCTP_STATE_SHUTDOWN_PENDING
:
909 case SCTP_STATE_SHUTDOWN_RECEIVED
:
911 * RFC 2960 6.1 Transmission of DATA Chunks
913 * C) When the time comes for the sender to transmit,
914 * before sending new DATA chunks, the sender MUST
915 * first transmit any outstanding DATA chunks which
916 * are marked for retransmission (limited by the
919 if (!list_empty(&q
->retransmit
)) {
920 if (asoc
->peer
.retran_path
->state
== SCTP_UNCONFIRMED
)
922 if (transport
== asoc
->peer
.retran_path
)
925 /* Switch transports & prepare the packet. */
927 transport
= asoc
->peer
.retran_path
;
929 if (list_empty(&transport
->send_ready
)) {
930 list_add_tail(&transport
->send_ready
,
934 packet
= &transport
->packet
;
935 sctp_packet_config(packet
, vtag
,
936 asoc
->peer
.ecn_capable
);
938 error
= sctp_outq_flush_rtx(q
, packet
,
939 rtx_timeout
, &start_timer
);
942 sctp_transport_reset_timers(transport
);
944 /* This can happen on COOKIE-ECHO resend. Only
945 * one chunk can get bundled with a COOKIE-ECHO.
947 if (packet
->has_cookie_echo
)
950 /* Don't send new data if there is still data
951 * waiting to retransmit.
953 if (!list_empty(&q
->retransmit
))
957 /* Apply Max.Burst limitation to the current transport in
958 * case it will be used for new data. We are going to
959 * rest it before we return, but we want to apply the limit
960 * to the currently queued data.
963 sctp_transport_burst_limited(transport
);
965 /* Finally, transmit new packets. */
966 while ((chunk
= sctp_outq_dequeue_data(q
)) != NULL
) {
967 /* RFC 2960 6.5 Every DATA chunk MUST carry a valid
970 if (chunk
->sinfo
.sinfo_stream
>=
971 asoc
->c
.sinit_num_ostreams
) {
973 /* Mark as failed send. */
974 sctp_chunk_fail(chunk
, SCTP_ERROR_INV_STRM
);
975 sctp_chunk_free(chunk
);
979 /* Has this chunk expired? */
980 if (sctp_chunk_abandoned(chunk
)) {
981 sctp_chunk_fail(chunk
, 0);
982 sctp_chunk_free(chunk
);
986 /* If there is a specified transport, use it.
987 * Otherwise, we want to use the active path.
989 new_transport
= chunk
->transport
;
990 if (!new_transport
||
991 ((new_transport
->state
== SCTP_INACTIVE
) ||
992 (new_transport
->state
== SCTP_UNCONFIRMED
) ||
993 (new_transport
->state
== SCTP_PF
)))
994 new_transport
= asoc
->peer
.active_path
;
995 if (new_transport
->state
== SCTP_UNCONFIRMED
)
998 /* Change packets if necessary. */
999 if (new_transport
!= transport
) {
1000 transport
= new_transport
;
1002 /* Schedule to have this transport's
1005 if (list_empty(&transport
->send_ready
)) {
1006 list_add_tail(&transport
->send_ready
,
1010 packet
= &transport
->packet
;
1011 sctp_packet_config(packet
, vtag
,
1012 asoc
->peer
.ecn_capable
);
1013 /* We've switched transports, so apply the
1014 * Burst limit to the new transport.
1016 sctp_transport_burst_limited(transport
);
1019 SCTP_DEBUG_PRINTK("sctp_outq_flush(%p, %p[%s]), ",
1021 chunk
&& chunk
->chunk_hdr
?
1022 sctp_cname(SCTP_ST_CHUNK(
1023 chunk
->chunk_hdr
->type
))
1026 SCTP_DEBUG_PRINTK("TX TSN 0x%x skb->head "
1027 "%p skb->users %d.\n",
1028 ntohl(chunk
->subh
.data_hdr
->tsn
),
1029 chunk
->skb
?chunk
->skb
->head
: NULL
,
1031 atomic_read(&chunk
->skb
->users
) : -1);
1033 /* Add the chunk to the packet. */
1034 status
= sctp_packet_transmit_chunk(packet
, chunk
, 0);
1037 case SCTP_XMIT_PMTU_FULL
:
1038 case SCTP_XMIT_RWND_FULL
:
1039 case SCTP_XMIT_NAGLE_DELAY
:
1040 /* We could not append this chunk, so put
1041 * the chunk back on the output queue.
1043 SCTP_DEBUG_PRINTK("sctp_outq_flush: could "
1044 "not transmit TSN: 0x%x, status: %d\n",
1045 ntohl(chunk
->subh
.data_hdr
->tsn
),
1047 sctp_outq_head_data(q
, chunk
);
1048 goto sctp_flush_out
;
1052 /* The sender is in the SHUTDOWN-PENDING state,
1053 * The sender MAY set the I-bit in the DATA
1056 if (asoc
->state
== SCTP_STATE_SHUTDOWN_PENDING
)
1057 chunk
->chunk_hdr
->flags
|= SCTP_DATA_SACK_IMM
;
1065 /* BUG: We assume that the sctp_packet_transmit()
1066 * call below will succeed all the time and add the
1067 * chunk to the transmitted list and restart the
1069 * It is possible that the call can fail under OOM
1072 * Is this really a problem? Won't this behave
1075 list_add_tail(&chunk
->transmitted_list
,
1076 &transport
->transmitted
);
1078 sctp_transport_reset_timers(transport
);
1082 /* Only let one DATA chunk get bundled with a
1083 * COOKIE-ECHO chunk.
1085 if (packet
->has_cookie_echo
)
1086 goto sctp_flush_out
;
1097 /* Before returning, examine all the transports touched in
1098 * this call. Right now, we bluntly force clear all the
1099 * transports. Things might change after we implement Nagle.
1100 * But such an examination is still required.
1104 while ((ltransport
= sctp_list_dequeue(&transport_list
)) != NULL
) {
1105 struct sctp_transport
*t
= list_entry(ltransport
,
1106 struct sctp_transport
,
1108 packet
= &t
->packet
;
1109 if (!sctp_packet_empty(packet
))
1110 error
= sctp_packet_transmit(packet
);
1112 /* Clear the burst limited state, if any */
1113 sctp_transport_burst_reset(t
);
1119 /* Update unack_data based on the incoming SACK chunk */
1120 static void sctp_sack_update_unack_data(struct sctp_association
*assoc
,
1121 struct sctp_sackhdr
*sack
)
1123 sctp_sack_variable_t
*frags
;
1127 unack_data
= assoc
->next_tsn
- assoc
->ctsn_ack_point
- 1;
1129 frags
= sack
->variable
;
1130 for (i
= 0; i
< ntohs(sack
->num_gap_ack_blocks
); i
++) {
1131 unack_data
-= ((ntohs(frags
[i
].gab
.end
) -
1132 ntohs(frags
[i
].gab
.start
) + 1));
1135 assoc
->unack_data
= unack_data
;
1138 /* This is where we REALLY process a SACK.
1140 * Process the SACK against the outqueue. Mostly, this just frees
1141 * things off the transmitted queue.
1143 int sctp_outq_sack(struct sctp_outq
*q
, struct sctp_chunk
*chunk
)
1145 struct sctp_association
*asoc
= q
->asoc
;
1146 struct sctp_sackhdr
*sack
= chunk
->subh
.sack_hdr
;
1147 struct sctp_transport
*transport
;
1148 struct sctp_chunk
*tchunk
= NULL
;
1149 struct list_head
*lchunk
, *transport_list
, *temp
;
1150 sctp_sack_variable_t
*frags
= sack
->variable
;
1151 __u32 sack_ctsn
, ctsn
, tsn
;
1152 __u32 highest_tsn
, highest_new_tsn
;
1154 unsigned int outstanding
;
1155 struct sctp_transport
*primary
= asoc
->peer
.primary_path
;
1156 int count_of_newacks
= 0;
1160 /* Grab the association's destination address list. */
1161 transport_list
= &asoc
->peer
.transport_addr_list
;
1163 sack_ctsn
= ntohl(sack
->cum_tsn_ack
);
1164 gap_ack_blocks
= ntohs(sack
->num_gap_ack_blocks
);
1166 * SFR-CACC algorithm:
1167 * On receipt of a SACK the sender SHOULD execute the
1168 * following statements.
1170 * 1) If the cumulative ack in the SACK passes next tsn_at_change
1171 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1172 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1174 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1175 * is set the receiver of the SACK MUST take the following actions:
1177 * A) Initialize the cacc_saw_newack to 0 for all destination
1180 * Only bother if changeover_active is set. Otherwise, this is
1181 * totally suboptimal to do on every SACK.
1183 if (primary
->cacc
.changeover_active
) {
1184 u8 clear_cycling
= 0;
1186 if (TSN_lte(primary
->cacc
.next_tsn_at_change
, sack_ctsn
)) {
1187 primary
->cacc
.changeover_active
= 0;
1191 if (clear_cycling
|| gap_ack_blocks
) {
1192 list_for_each_entry(transport
, transport_list
,
1195 transport
->cacc
.cycling_changeover
= 0;
1197 transport
->cacc
.cacc_saw_newack
= 0;
1202 /* Get the highest TSN in the sack. */
1203 highest_tsn
= sack_ctsn
;
1205 highest_tsn
+= ntohs(frags
[gap_ack_blocks
- 1].gab
.end
);
1207 if (TSN_lt(asoc
->highest_sacked
, highest_tsn
))
1208 asoc
->highest_sacked
= highest_tsn
;
1210 highest_new_tsn
= sack_ctsn
;
1212 /* Run through the retransmit queue. Credit bytes received
1213 * and free those chunks that we can.
1215 sctp_check_transmitted(q
, &q
->retransmit
, NULL
, NULL
, sack
, &highest_new_tsn
);
1217 /* Run through the transmitted queue.
1218 * Credit bytes received and free those chunks which we can.
1220 * This is a MASSIVE candidate for optimization.
1222 list_for_each_entry(transport
, transport_list
, transports
) {
1223 sctp_check_transmitted(q
, &transport
->transmitted
,
1224 transport
, &chunk
->source
, sack
,
1227 * SFR-CACC algorithm:
1228 * C) Let count_of_newacks be the number of
1229 * destinations for which cacc_saw_newack is set.
1231 if (transport
->cacc
.cacc_saw_newack
)
1232 count_of_newacks
++;
1235 /* Move the Cumulative TSN Ack Point if appropriate. */
1236 if (TSN_lt(asoc
->ctsn_ack_point
, sack_ctsn
)) {
1237 asoc
->ctsn_ack_point
= sack_ctsn
;
1241 if (gap_ack_blocks
) {
1243 if (asoc
->fast_recovery
&& accum_moved
)
1244 highest_new_tsn
= highest_tsn
;
1246 list_for_each_entry(transport
, transport_list
, transports
)
1247 sctp_mark_missing(q
, &transport
->transmitted
, transport
,
1248 highest_new_tsn
, count_of_newacks
);
1251 /* Update unack_data field in the assoc. */
1252 sctp_sack_update_unack_data(asoc
, sack
);
1254 ctsn
= asoc
->ctsn_ack_point
;
1256 /* Throw away stuff rotting on the sack queue. */
1257 list_for_each_safe(lchunk
, temp
, &q
->sacked
) {
1258 tchunk
= list_entry(lchunk
, struct sctp_chunk
,
1260 tsn
= ntohl(tchunk
->subh
.data_hdr
->tsn
);
1261 if (TSN_lte(tsn
, ctsn
)) {
1262 list_del_init(&tchunk
->transmitted_list
);
1263 sctp_chunk_free(tchunk
);
1267 /* ii) Set rwnd equal to the newly received a_rwnd minus the
1268 * number of bytes still outstanding after processing the
1269 * Cumulative TSN Ack and the Gap Ack Blocks.
1272 sack_a_rwnd
= ntohl(sack
->a_rwnd
);
1273 outstanding
= q
->outstanding_bytes
;
1275 if (outstanding
< sack_a_rwnd
)
1276 sack_a_rwnd
-= outstanding
;
1280 asoc
->peer
.rwnd
= sack_a_rwnd
;
1282 sctp_generate_fwdtsn(q
, sack_ctsn
);
1284 SCTP_DEBUG_PRINTK("%s: sack Cumulative TSN Ack is 0x%x.\n",
1285 __func__
, sack_ctsn
);
1286 SCTP_DEBUG_PRINTK("%s: Cumulative TSN Ack of association, "
1287 "%p is 0x%x. Adv peer ack point: 0x%x\n",
1288 __func__
, asoc
, ctsn
, asoc
->adv_peer_ack_point
);
1290 /* See if all chunks are acked.
1291 * Make sure the empty queue handler will get run later.
1293 q
->empty
= (list_empty(&q
->out_chunk_list
) &&
1294 list_empty(&q
->retransmit
));
1298 list_for_each_entry(transport
, transport_list
, transports
) {
1299 q
->empty
= q
->empty
&& list_empty(&transport
->transmitted
);
1304 SCTP_DEBUG_PRINTK("sack queue is empty.\n");
1309 /* Is the outqueue empty? */
1310 int sctp_outq_is_empty(const struct sctp_outq
*q
)
1315 /********************************************************************
1316 * 2nd Level Abstractions
1317 ********************************************************************/
1319 /* Go through a transport's transmitted list or the association's retransmit
1320 * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1321 * The retransmit list will not have an associated transport.
1323 * I added coherent debug information output. --xguo
1325 * Instead of printing 'sacked' or 'kept' for each TSN on the
1326 * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1327 * KEPT TSN6-TSN7, etc.
1329 static void sctp_check_transmitted(struct sctp_outq
*q
,
1330 struct list_head
*transmitted_queue
,
1331 struct sctp_transport
*transport
,
1332 union sctp_addr
*saddr
,
1333 struct sctp_sackhdr
*sack
,
1334 __u32
*highest_new_tsn_in_sack
)
1336 struct list_head
*lchunk
;
1337 struct sctp_chunk
*tchunk
;
1338 struct list_head tlist
;
1342 __u8 restart_timer
= 0;
1343 int bytes_acked
= 0;
1344 int migrate_bytes
= 0;
1346 /* These state variables are for coherent debug output. --xguo */
1349 __u32 dbg_ack_tsn
= 0; /* An ACKed TSN range starts here... */
1350 __u32 dbg_last_ack_tsn
= 0; /* ...and finishes here. */
1351 __u32 dbg_kept_tsn
= 0; /* An un-ACKed range starts here... */
1352 __u32 dbg_last_kept_tsn
= 0; /* ...and finishes here. */
1354 /* 0 : The last TSN was ACKed.
1355 * 1 : The last TSN was NOT ACKed (i.e. KEPT).
1356 * -1: We need to initialize.
1358 int dbg_prt_state
= -1;
1359 #endif /* SCTP_DEBUG */
1361 sack_ctsn
= ntohl(sack
->cum_tsn_ack
);
1363 INIT_LIST_HEAD(&tlist
);
1365 /* The while loop will skip empty transmitted queues. */
1366 while (NULL
!= (lchunk
= sctp_list_dequeue(transmitted_queue
))) {
1367 tchunk
= list_entry(lchunk
, struct sctp_chunk
,
1370 if (sctp_chunk_abandoned(tchunk
)) {
1371 /* Move the chunk to abandoned list. */
1372 sctp_insert_list(&q
->abandoned
, lchunk
);
1374 /* If this chunk has not been acked, stop
1375 * considering it as 'outstanding'.
1377 if (!tchunk
->tsn_gap_acked
) {
1378 if (tchunk
->transport
)
1379 tchunk
->transport
->flight_size
-=
1380 sctp_data_size(tchunk
);
1381 q
->outstanding_bytes
-= sctp_data_size(tchunk
);
1386 tsn
= ntohl(tchunk
->subh
.data_hdr
->tsn
);
1387 if (sctp_acked(sack
, tsn
)) {
1388 /* If this queue is the retransmit queue, the
1389 * retransmit timer has already reclaimed
1390 * the outstanding bytes for this chunk, so only
1391 * count bytes associated with a transport.
1394 /* If this chunk is being used for RTT
1395 * measurement, calculate the RTT and update
1396 * the RTO using this value.
1398 * 6.3.1 C5) Karn's algorithm: RTT measurements
1399 * MUST NOT be made using packets that were
1400 * retransmitted (and thus for which it is
1401 * ambiguous whether the reply was for the
1402 * first instance of the packet or a later
1405 if (!tchunk
->tsn_gap_acked
&&
1406 tchunk
->rtt_in_progress
) {
1407 tchunk
->rtt_in_progress
= 0;
1408 rtt
= jiffies
- tchunk
->sent_at
;
1409 sctp_transport_update_rto(transport
,
1414 /* If the chunk hasn't been marked as ACKED,
1415 * mark it and account bytes_acked if the
1416 * chunk had a valid transport (it will not
1417 * have a transport if ASCONF had deleted it
1418 * while DATA was outstanding).
1420 if (!tchunk
->tsn_gap_acked
) {
1421 tchunk
->tsn_gap_acked
= 1;
1422 *highest_new_tsn_in_sack
= tsn
;
1423 bytes_acked
+= sctp_data_size(tchunk
);
1424 if (!tchunk
->transport
)
1425 migrate_bytes
+= sctp_data_size(tchunk
);
1428 if (TSN_lte(tsn
, sack_ctsn
)) {
1429 /* RFC 2960 6.3.2 Retransmission Timer Rules
1431 * R3) Whenever a SACK is received
1432 * that acknowledges the DATA chunk
1433 * with the earliest outstanding TSN
1434 * for that address, restart T3-rtx
1435 * timer for that address with its
1440 if (!tchunk
->tsn_gap_acked
) {
1442 * SFR-CACC algorithm:
1443 * 2) If the SACK contains gap acks
1444 * and the flag CHANGEOVER_ACTIVE is
1445 * set the receiver of the SACK MUST
1446 * take the following action:
1448 * B) For each TSN t being acked that
1449 * has not been acked in any SACK so
1450 * far, set cacc_saw_newack to 1 for
1451 * the destination that the TSN was
1455 sack
->num_gap_ack_blocks
&&
1456 q
->asoc
->peer
.primary_path
->cacc
.
1458 transport
->cacc
.cacc_saw_newack
1462 list_add_tail(&tchunk
->transmitted_list
,
1465 /* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1466 * M2) Each time a SACK arrives reporting
1467 * 'Stray DATA chunk(s)' record the highest TSN
1468 * reported as newly acknowledged, call this
1469 * value 'HighestTSNinSack'. A newly
1470 * acknowledged DATA chunk is one not
1471 * previously acknowledged in a SACK.
1473 * When the SCTP sender of data receives a SACK
1474 * chunk that acknowledges, for the first time,
1475 * the receipt of a DATA chunk, all the still
1476 * unacknowledged DATA chunks whose TSN is
1477 * older than that newly acknowledged DATA
1478 * chunk, are qualified as 'Stray DATA chunks'.
1480 list_add_tail(lchunk
, &tlist
);
1484 switch (dbg_prt_state
) {
1485 case 0: /* last TSN was ACKed */
1486 if (dbg_last_ack_tsn
+ 1 == tsn
) {
1487 /* This TSN belongs to the
1488 * current ACK range.
1493 if (dbg_last_ack_tsn
!= dbg_ack_tsn
) {
1494 /* Display the end of the
1497 SCTP_DEBUG_PRINTK_CONT("-%08x",
1501 /* Start a new range. */
1502 SCTP_DEBUG_PRINTK_CONT(",%08x", tsn
);
1506 case 1: /* The last TSN was NOT ACKed. */
1507 if (dbg_last_kept_tsn
!= dbg_kept_tsn
) {
1508 /* Display the end of current range. */
1509 SCTP_DEBUG_PRINTK_CONT("-%08x",
1513 SCTP_DEBUG_PRINTK_CONT("\n");
1515 /* FALL THROUGH... */
1517 /* This is the first-ever TSN we examined. */
1518 /* Start a new range of ACK-ed TSNs. */
1519 SCTP_DEBUG_PRINTK("ACKed: %08x", tsn
);
1524 dbg_last_ack_tsn
= tsn
;
1525 #endif /* SCTP_DEBUG */
1528 if (tchunk
->tsn_gap_acked
) {
1529 SCTP_DEBUG_PRINTK("%s: Receiver reneged on "
1533 tchunk
->tsn_gap_acked
= 0;
1535 if (tchunk
->transport
)
1536 bytes_acked
-= sctp_data_size(tchunk
);
1538 /* RFC 2960 6.3.2 Retransmission Timer Rules
1540 * R4) Whenever a SACK is received missing a
1541 * TSN that was previously acknowledged via a
1542 * Gap Ack Block, start T3-rtx for the
1543 * destination address to which the DATA
1544 * chunk was originally
1545 * transmitted if it is not already running.
1550 list_add_tail(lchunk
, &tlist
);
1553 /* See the above comments on ACK-ed TSNs. */
1554 switch (dbg_prt_state
) {
1556 if (dbg_last_kept_tsn
+ 1 == tsn
)
1559 if (dbg_last_kept_tsn
!= dbg_kept_tsn
)
1560 SCTP_DEBUG_PRINTK_CONT("-%08x",
1563 SCTP_DEBUG_PRINTK_CONT(",%08x", tsn
);
1568 if (dbg_last_ack_tsn
!= dbg_ack_tsn
)
1569 SCTP_DEBUG_PRINTK_CONT("-%08x",
1571 SCTP_DEBUG_PRINTK_CONT("\n");
1573 /* FALL THROUGH... */
1575 SCTP_DEBUG_PRINTK("KEPT: %08x",tsn
);
1580 dbg_last_kept_tsn
= tsn
;
1581 #endif /* SCTP_DEBUG */
1586 /* Finish off the last range, displaying its ending TSN. */
1587 switch (dbg_prt_state
) {
1589 if (dbg_last_ack_tsn
!= dbg_ack_tsn
) {
1590 SCTP_DEBUG_PRINTK_CONT("-%08x\n", dbg_last_ack_tsn
);
1592 SCTP_DEBUG_PRINTK_CONT("\n");
1597 if (dbg_last_kept_tsn
!= dbg_kept_tsn
) {
1598 SCTP_DEBUG_PRINTK_CONT("-%08x\n", dbg_last_kept_tsn
);
1600 SCTP_DEBUG_PRINTK_CONT("\n");
1603 #endif /* SCTP_DEBUG */
1606 struct sctp_association
*asoc
= transport
->asoc
;
1608 /* We may have counted DATA that was migrated
1609 * to this transport due to DEL-IP operation.
1610 * Subtract those bytes, since the were never
1611 * send on this transport and shouldn't be
1612 * credited to this transport.
1614 bytes_acked
-= migrate_bytes
;
1616 /* 8.2. When an outstanding TSN is acknowledged,
1617 * the endpoint shall clear the error counter of
1618 * the destination transport address to which the
1619 * DATA chunk was last sent.
1620 * The association's overall error counter is
1623 transport
->error_count
= 0;
1624 transport
->asoc
->overall_error_count
= 0;
1627 * While in SHUTDOWN PENDING, we may have started
1628 * the T5 shutdown guard timer after reaching the
1629 * retransmission limit. Stop that timer as soon
1630 * as the receiver acknowledged any data.
1632 if (asoc
->state
== SCTP_STATE_SHUTDOWN_PENDING
&&
1633 del_timer(&asoc
->timers
1634 [SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD
]))
1635 sctp_association_put(asoc
);
1637 /* Mark the destination transport address as
1638 * active if it is not so marked.
1640 if ((transport
->state
== SCTP_INACTIVE
||
1641 transport
->state
== SCTP_UNCONFIRMED
) &&
1642 sctp_cmp_addr_exact(&transport
->ipaddr
, saddr
)) {
1643 sctp_assoc_control_transport(
1647 SCTP_RECEIVED_SACK
);
1650 sctp_transport_raise_cwnd(transport
, sack_ctsn
,
1653 transport
->flight_size
-= bytes_acked
;
1654 if (transport
->flight_size
== 0)
1655 transport
->partial_bytes_acked
= 0;
1656 q
->outstanding_bytes
-= bytes_acked
+ migrate_bytes
;
1658 /* RFC 2960 6.1, sctpimpguide-06 2.15.2
1659 * When a sender is doing zero window probing, it
1660 * should not timeout the association if it continues
1661 * to receive new packets from the receiver. The
1662 * reason is that the receiver MAY keep its window
1663 * closed for an indefinite time.
1664 * A sender is doing zero window probing when the
1665 * receiver's advertised window is zero, and there is
1666 * only one data chunk in flight to the receiver.
1668 * Allow the association to timeout while in SHUTDOWN
1669 * PENDING or SHUTDOWN RECEIVED in case the receiver
1670 * stays in zero window mode forever.
1672 if (!q
->asoc
->peer
.rwnd
&&
1673 !list_empty(&tlist
) &&
1674 (sack_ctsn
+2 == q
->asoc
->next_tsn
) &&
1675 q
->asoc
->state
< SCTP_STATE_SHUTDOWN_PENDING
) {
1676 SCTP_DEBUG_PRINTK("%s: SACK received for zero "
1677 "window probe: %u\n",
1678 __func__
, sack_ctsn
);
1679 q
->asoc
->overall_error_count
= 0;
1680 transport
->error_count
= 0;
1684 /* RFC 2960 6.3.2 Retransmission Timer Rules
1686 * R2) Whenever all outstanding data sent to an address have
1687 * been acknowledged, turn off the T3-rtx timer of that
1690 if (!transport
->flight_size
) {
1691 if (timer_pending(&transport
->T3_rtx_timer
) &&
1692 del_timer(&transport
->T3_rtx_timer
)) {
1693 sctp_transport_put(transport
);
1695 } else if (restart_timer
) {
1696 if (!mod_timer(&transport
->T3_rtx_timer
,
1697 jiffies
+ transport
->rto
))
1698 sctp_transport_hold(transport
);
1702 list_splice(&tlist
, transmitted_queue
);
1705 /* Mark chunks as missing and consequently may get retransmitted. */
1706 static void sctp_mark_missing(struct sctp_outq
*q
,
1707 struct list_head
*transmitted_queue
,
1708 struct sctp_transport
*transport
,
1709 __u32 highest_new_tsn_in_sack
,
1710 int count_of_newacks
)
1712 struct sctp_chunk
*chunk
;
1714 char do_fast_retransmit
= 0;
1715 struct sctp_association
*asoc
= q
->asoc
;
1716 struct sctp_transport
*primary
= asoc
->peer
.primary_path
;
1718 list_for_each_entry(chunk
, transmitted_queue
, transmitted_list
) {
1720 tsn
= ntohl(chunk
->subh
.data_hdr
->tsn
);
1722 /* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1723 * 'Unacknowledged TSN's', if the TSN number of an
1724 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1725 * value, increment the 'TSN.Missing.Report' count on that
1726 * chunk if it has NOT been fast retransmitted or marked for
1727 * fast retransmit already.
1729 if (chunk
->fast_retransmit
== SCTP_CAN_FRTX
&&
1730 !chunk
->tsn_gap_acked
&&
1731 TSN_lt(tsn
, highest_new_tsn_in_sack
)) {
1733 /* SFR-CACC may require us to skip marking
1734 * this chunk as missing.
1736 if (!transport
|| !sctp_cacc_skip(primary
,
1738 count_of_newacks
, tsn
)) {
1739 chunk
->tsn_missing_report
++;
1742 "%s: TSN 0x%x missing counter: %d\n",
1744 chunk
->tsn_missing_report
);
1748 * M4) If any DATA chunk is found to have a
1749 * 'TSN.Missing.Report'
1750 * value larger than or equal to 3, mark that chunk for
1751 * retransmission and start the fast retransmit procedure.
1754 if (chunk
->tsn_missing_report
>= 3) {
1755 chunk
->fast_retransmit
= SCTP_NEED_FRTX
;
1756 do_fast_retransmit
= 1;
1761 if (do_fast_retransmit
)
1762 sctp_retransmit(q
, transport
, SCTP_RTXR_FAST_RTX
);
1764 SCTP_DEBUG_PRINTK("%s: transport: %p, cwnd: %d, "
1765 "ssthresh: %d, flight_size: %d, pba: %d\n",
1766 __func__
, transport
, transport
->cwnd
,
1767 transport
->ssthresh
, transport
->flight_size
,
1768 transport
->partial_bytes_acked
);
1772 /* Is the given TSN acked by this packet? */
1773 static int sctp_acked(struct sctp_sackhdr
*sack
, __u32 tsn
)
1776 sctp_sack_variable_t
*frags
;
1778 __u32 ctsn
= ntohl(sack
->cum_tsn_ack
);
1780 if (TSN_lte(tsn
, ctsn
))
1783 /* 3.3.4 Selective Acknowledgement (SACK) (3):
1786 * These fields contain the Gap Ack Blocks. They are repeated
1787 * for each Gap Ack Block up to the number of Gap Ack Blocks
1788 * defined in the Number of Gap Ack Blocks field. All DATA
1789 * chunks with TSNs greater than or equal to (Cumulative TSN
1790 * Ack + Gap Ack Block Start) and less than or equal to
1791 * (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1792 * Block are assumed to have been received correctly.
1795 frags
= sack
->variable
;
1797 for (i
= 0; i
< ntohs(sack
->num_gap_ack_blocks
); ++i
) {
1798 if (TSN_lte(ntohs(frags
[i
].gab
.start
), gap
) &&
1799 TSN_lte(gap
, ntohs(frags
[i
].gab
.end
)))
1808 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip
*skiplist
,
1809 int nskips
, __be16 stream
)
1813 for (i
= 0; i
< nskips
; i
++) {
1814 if (skiplist
[i
].stream
== stream
)
1820 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1821 static void sctp_generate_fwdtsn(struct sctp_outq
*q
, __u32 ctsn
)
1823 struct sctp_association
*asoc
= q
->asoc
;
1824 struct sctp_chunk
*ftsn_chunk
= NULL
;
1825 struct sctp_fwdtsn_skip ftsn_skip_arr
[10];
1829 struct sctp_chunk
*chunk
;
1830 struct list_head
*lchunk
, *temp
;
1832 if (!asoc
->peer
.prsctp_capable
)
1835 /* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1838 * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1839 * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1841 if (TSN_lt(asoc
->adv_peer_ack_point
, ctsn
))
1842 asoc
->adv_peer_ack_point
= ctsn
;
1844 /* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1845 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1846 * the chunk next in the out-queue space is marked as "abandoned" as
1847 * shown in the following example:
1849 * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1850 * and the Advanced.Peer.Ack.Point is updated to this value:
1852 * out-queue at the end of ==> out-queue after Adv.Ack.Point
1853 * normal SACK processing local advancement
1855 * Adv.Ack.Pt-> 102 acked 102 acked
1856 * 103 abandoned 103 abandoned
1857 * 104 abandoned Adv.Ack.P-> 104 abandoned
1859 * 106 acked 106 acked
1862 * In this example, the data sender successfully advanced the
1863 * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1865 list_for_each_safe(lchunk
, temp
, &q
->abandoned
) {
1866 chunk
= list_entry(lchunk
, struct sctp_chunk
,
1868 tsn
= ntohl(chunk
->subh
.data_hdr
->tsn
);
1870 /* Remove any chunks in the abandoned queue that are acked by
1873 if (TSN_lte(tsn
, ctsn
)) {
1874 list_del_init(lchunk
);
1875 sctp_chunk_free(chunk
);
1877 if (TSN_lte(tsn
, asoc
->adv_peer_ack_point
+1)) {
1878 asoc
->adv_peer_ack_point
= tsn
;
1879 if (chunk
->chunk_hdr
->flags
&
1880 SCTP_DATA_UNORDERED
)
1882 skip_pos
= sctp_get_skip_pos(&ftsn_skip_arr
[0],
1884 chunk
->subh
.data_hdr
->stream
);
1885 ftsn_skip_arr
[skip_pos
].stream
=
1886 chunk
->subh
.data_hdr
->stream
;
1887 ftsn_skip_arr
[skip_pos
].ssn
=
1888 chunk
->subh
.data_hdr
->ssn
;
1889 if (skip_pos
== nskips
)
1898 /* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1899 * is greater than the Cumulative TSN ACK carried in the received
1900 * SACK, the data sender MUST send the data receiver a FORWARD TSN
1901 * chunk containing the latest value of the
1902 * "Advanced.Peer.Ack.Point".
1904 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1905 * list each stream and sequence number in the forwarded TSN. This
1906 * information will enable the receiver to easily find any
1907 * stranded TSN's waiting on stream reorder queues. Each stream
1908 * SHOULD only be reported once; this means that if multiple
1909 * abandoned messages occur in the same stream then only the
1910 * highest abandoned stream sequence number is reported. If the
1911 * total size of the FORWARD TSN does NOT fit in a single MTU then
1912 * the sender of the FORWARD TSN SHOULD lower the
1913 * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1916 if (asoc
->adv_peer_ack_point
> ctsn
)
1917 ftsn_chunk
= sctp_make_fwdtsn(asoc
, asoc
->adv_peer_ack_point
,
1918 nskips
, &ftsn_skip_arr
[0]);
1921 list_add_tail(&ftsn_chunk
->list
, &q
->control_chunk_list
);
1922 SCTP_INC_STATS(sock_net(asoc
->base
.sk
), SCTP_MIB_OUTCTRLCHUNKS
);