xref: /openbmc/linux/net/sctp/outqueue.c (revision 609e478b)
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.
6  *
7  * This file is part of the SCTP kernel implementation
8  *
9  * These functions implement the sctp_outq class.   The outqueue handles
10  * bundling and queueing of outgoing SCTP chunks.
11  *
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)
16  * any later version.
17  *
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.
23  *
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/>.
27  *
28  * Please send any bug reports or fixes you make to the
29  * email address(es):
30  *    lksctp developers <linux-sctp@vger.kernel.org>
31  *
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>
40  */
41 
42 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
43 
44 #include <linux/types.h>
45 #include <linux/list.h>   /* For struct list_head */
46 #include <linux/socket.h>
47 #include <linux/ip.h>
48 #include <linux/slab.h>
49 #include <net/sock.h>	  /* For skb_set_owner_w */
50 
51 #include <net/sctp/sctp.h>
52 #include <net/sctp/sm.h>
53 
54 /* Declare internal functions here.  */
55 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn);
56 static void sctp_check_transmitted(struct sctp_outq *q,
57 				   struct list_head *transmitted_queue,
58 				   struct sctp_transport *transport,
59 				   union sctp_addr *saddr,
60 				   struct sctp_sackhdr *sack,
61 				   __u32 *highest_new_tsn);
62 
63 static void sctp_mark_missing(struct sctp_outq *q,
64 			      struct list_head *transmitted_queue,
65 			      struct sctp_transport *transport,
66 			      __u32 highest_new_tsn,
67 			      int count_of_newacks);
68 
69 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 sack_ctsn);
70 
71 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout);
72 
73 /* Add data to the front of the queue. */
74 static inline void sctp_outq_head_data(struct sctp_outq *q,
75 					struct sctp_chunk *ch)
76 {
77 	list_add(&ch->list, &q->out_chunk_list);
78 	q->out_qlen += ch->skb->len;
79 }
80 
81 /* Take data from the front of the queue. */
82 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q)
83 {
84 	struct sctp_chunk *ch = NULL;
85 
86 	if (!list_empty(&q->out_chunk_list)) {
87 		struct list_head *entry = q->out_chunk_list.next;
88 
89 		ch = list_entry(entry, struct sctp_chunk, list);
90 		list_del_init(entry);
91 		q->out_qlen -= ch->skb->len;
92 	}
93 	return ch;
94 }
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)
98 {
99 	list_add_tail(&ch->list, &q->out_chunk_list);
100 	q->out_qlen += ch->skb->len;
101 }
102 
103 /*
104  * SFR-CACC algorithm:
105  * D) If count_of_newacks is greater than or equal to 2
106  * and t was not sent to the current primary then the
107  * sender MUST NOT increment missing report count for t.
108  */
109 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary,
110 				       struct sctp_transport *transport,
111 				       int count_of_newacks)
112 {
113 	if (count_of_newacks >= 2 && transport != primary)
114 		return 1;
115 	return 0;
116 }
117 
118 /*
119  * SFR-CACC algorithm:
120  * F) If count_of_newacks is less than 2, let d be the
121  * destination to which t was sent. If cacc_saw_newack
122  * is 0 for destination d, then the sender MUST NOT
123  * increment missing report count for t.
124  */
125 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport,
126 				       int count_of_newacks)
127 {
128 	if (count_of_newacks < 2 &&
129 			(transport && !transport->cacc.cacc_saw_newack))
130 		return 1;
131 	return 0;
132 }
133 
134 /*
135  * SFR-CACC algorithm:
136  * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD
137  * execute steps C, D, F.
138  *
139  * C has been implemented in sctp_outq_sack
140  */
141 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary,
142 				     struct sctp_transport *transport,
143 				     int count_of_newacks)
144 {
145 	if (!primary->cacc.cycling_changeover) {
146 		if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks))
147 			return 1;
148 		if (sctp_cacc_skip_3_1_f(transport, count_of_newacks))
149 			return 1;
150 		return 0;
151 	}
152 	return 0;
153 }
154 
155 /*
156  * SFR-CACC algorithm:
157  * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less
158  * than next_tsn_at_change of the current primary, then
159  * the sender MUST NOT increment missing report count
160  * for t.
161  */
162 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn)
163 {
164 	if (primary->cacc.cycling_changeover &&
165 	    TSN_lt(tsn, primary->cacc.next_tsn_at_change))
166 		return 1;
167 	return 0;
168 }
169 
170 /*
171  * SFR-CACC algorithm:
172  * 3) If the missing report count for TSN t is to be
173  * incremented according to [RFC2960] and
174  * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set,
175  * then the sender MUST further execute steps 3.1 and
176  * 3.2 to determine if the missing report count for
177  * TSN t SHOULD NOT be incremented.
178  *
179  * 3.3) If 3.1 and 3.2 do not dictate that the missing
180  * report count for t should not be incremented, then
181  * the sender SHOULD increment missing report count for
182  * t (according to [RFC2960] and [SCTP_STEWART_2002]).
183  */
184 static inline int sctp_cacc_skip(struct sctp_transport *primary,
185 				 struct sctp_transport *transport,
186 				 int count_of_newacks,
187 				 __u32 tsn)
188 {
189 	if (primary->cacc.changeover_active &&
190 	    (sctp_cacc_skip_3_1(primary, transport, count_of_newacks) ||
191 	     sctp_cacc_skip_3_2(primary, tsn)))
192 		return 1;
193 	return 0;
194 }
195 
196 /* Initialize an existing sctp_outq.  This does the boring stuff.
197  * You still need to define handlers if you really want to DO
198  * something with this structure...
199  */
200 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q)
201 {
202 	memset(q, 0, sizeof(struct sctp_outq));
203 
204 	q->asoc = asoc;
205 	INIT_LIST_HEAD(&q->out_chunk_list);
206 	INIT_LIST_HEAD(&q->control_chunk_list);
207 	INIT_LIST_HEAD(&q->retransmit);
208 	INIT_LIST_HEAD(&q->sacked);
209 	INIT_LIST_HEAD(&q->abandoned);
210 }
211 
212 /* Free the outqueue structure and any related pending chunks.
213  */
214 static void __sctp_outq_teardown(struct sctp_outq *q)
215 {
216 	struct sctp_transport *transport;
217 	struct list_head *lchunk, *temp;
218 	struct sctp_chunk *chunk, *tmp;
219 
220 	/* Throw away unacknowledged chunks. */
221 	list_for_each_entry(transport, &q->asoc->peer.transport_addr_list,
222 			transports) {
223 		while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) {
224 			chunk = list_entry(lchunk, struct sctp_chunk,
225 					   transmitted_list);
226 			/* Mark as part of a failed message. */
227 			sctp_chunk_fail(chunk, q->error);
228 			sctp_chunk_free(chunk);
229 		}
230 	}
231 
232 	/* Throw away chunks that have been gap ACKed.  */
233 	list_for_each_safe(lchunk, temp, &q->sacked) {
234 		list_del_init(lchunk);
235 		chunk = list_entry(lchunk, struct sctp_chunk,
236 				   transmitted_list);
237 		sctp_chunk_fail(chunk, q->error);
238 		sctp_chunk_free(chunk);
239 	}
240 
241 	/* Throw away any chunks in the retransmit queue. */
242 	list_for_each_safe(lchunk, temp, &q->retransmit) {
243 		list_del_init(lchunk);
244 		chunk = list_entry(lchunk, struct sctp_chunk,
245 				   transmitted_list);
246 		sctp_chunk_fail(chunk, q->error);
247 		sctp_chunk_free(chunk);
248 	}
249 
250 	/* Throw away any chunks that are in the abandoned queue. */
251 	list_for_each_safe(lchunk, temp, &q->abandoned) {
252 		list_del_init(lchunk);
253 		chunk = list_entry(lchunk, struct sctp_chunk,
254 				   transmitted_list);
255 		sctp_chunk_fail(chunk, q->error);
256 		sctp_chunk_free(chunk);
257 	}
258 
259 	/* Throw away any leftover data chunks. */
260 	while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
261 
262 		/* Mark as send failure. */
263 		sctp_chunk_fail(chunk, q->error);
264 		sctp_chunk_free(chunk);
265 	}
266 
267 	/* Throw away any leftover control chunks. */
268 	list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
269 		list_del_init(&chunk->list);
270 		sctp_chunk_free(chunk);
271 	}
272 }
273 
274 void sctp_outq_teardown(struct sctp_outq *q)
275 {
276 	__sctp_outq_teardown(q);
277 	sctp_outq_init(q->asoc, q);
278 }
279 
280 /* Free the outqueue structure and any related pending chunks.  */
281 void sctp_outq_free(struct sctp_outq *q)
282 {
283 	/* Throw away leftover chunks. */
284 	__sctp_outq_teardown(q);
285 }
286 
287 /* Put a new chunk in an sctp_outq.  */
288 int sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk)
289 {
290 	struct net *net = sock_net(q->asoc->base.sk);
291 	int error = 0;
292 
293 	pr_debug("%s: outq:%p, chunk:%p[%s]\n", __func__, q, chunk,
294 		 chunk && chunk->chunk_hdr ?
295 		 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
296 		 "illegal chunk");
297 
298 	/* If it is data, queue it up, otherwise, send it
299 	 * immediately.
300 	 */
301 	if (sctp_chunk_is_data(chunk)) {
302 		/* Is it OK to queue data chunks?  */
303 		/* From 9. Termination of Association
304 		 *
305 		 * When either endpoint performs a shutdown, the
306 		 * association on each peer will stop accepting new
307 		 * data from its user and only deliver data in queue
308 		 * at the time of sending or receiving the SHUTDOWN
309 		 * chunk.
310 		 */
311 		switch (q->asoc->state) {
312 		case SCTP_STATE_CLOSED:
313 		case SCTP_STATE_SHUTDOWN_PENDING:
314 		case SCTP_STATE_SHUTDOWN_SENT:
315 		case SCTP_STATE_SHUTDOWN_RECEIVED:
316 		case SCTP_STATE_SHUTDOWN_ACK_SENT:
317 			/* Cannot send after transport endpoint shutdown */
318 			error = -ESHUTDOWN;
319 			break;
320 
321 		default:
322 			pr_debug("%s: outqueueing: outq:%p, chunk:%p[%s])\n",
323 				 __func__, q, chunk, chunk && chunk->chunk_hdr ?
324 				 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
325 				 "illegal chunk");
326 
327 			sctp_outq_tail_data(q, chunk);
328 			if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
329 				SCTP_INC_STATS(net, SCTP_MIB_OUTUNORDERCHUNKS);
330 			else
331 				SCTP_INC_STATS(net, SCTP_MIB_OUTORDERCHUNKS);
332 			break;
333 		}
334 	} else {
335 		list_add_tail(&chunk->list, &q->control_chunk_list);
336 		SCTP_INC_STATS(net, SCTP_MIB_OUTCTRLCHUNKS);
337 	}
338 
339 	if (error < 0)
340 		return error;
341 
342 	if (!q->cork)
343 		error = sctp_outq_flush(q, 0);
344 
345 	return error;
346 }
347 
348 /* Insert a chunk into the sorted list based on the TSNs.  The retransmit list
349  * and the abandoned list are in ascending order.
350  */
351 static void sctp_insert_list(struct list_head *head, struct list_head *new)
352 {
353 	struct list_head *pos;
354 	struct sctp_chunk *nchunk, *lchunk;
355 	__u32 ntsn, ltsn;
356 	int done = 0;
357 
358 	nchunk = list_entry(new, struct sctp_chunk, transmitted_list);
359 	ntsn = ntohl(nchunk->subh.data_hdr->tsn);
360 
361 	list_for_each(pos, head) {
362 		lchunk = list_entry(pos, struct sctp_chunk, transmitted_list);
363 		ltsn = ntohl(lchunk->subh.data_hdr->tsn);
364 		if (TSN_lt(ntsn, ltsn)) {
365 			list_add(new, pos->prev);
366 			done = 1;
367 			break;
368 		}
369 	}
370 	if (!done)
371 		list_add_tail(new, head);
372 }
373 
374 /* Mark all the eligible packets on a transport for retransmission.  */
375 void sctp_retransmit_mark(struct sctp_outq *q,
376 			  struct sctp_transport *transport,
377 			  __u8 reason)
378 {
379 	struct list_head *lchunk, *ltemp;
380 	struct sctp_chunk *chunk;
381 
382 	/* Walk through the specified transmitted queue.  */
383 	list_for_each_safe(lchunk, ltemp, &transport->transmitted) {
384 		chunk = list_entry(lchunk, struct sctp_chunk,
385 				   transmitted_list);
386 
387 		/* If the chunk is abandoned, move it to abandoned list. */
388 		if (sctp_chunk_abandoned(chunk)) {
389 			list_del_init(lchunk);
390 			sctp_insert_list(&q->abandoned, lchunk);
391 
392 			/* If this chunk has not been previousely acked,
393 			 * stop considering it 'outstanding'.  Our peer
394 			 * will most likely never see it since it will
395 			 * not be retransmitted
396 			 */
397 			if (!chunk->tsn_gap_acked) {
398 				if (chunk->transport)
399 					chunk->transport->flight_size -=
400 							sctp_data_size(chunk);
401 				q->outstanding_bytes -= sctp_data_size(chunk);
402 				q->asoc->peer.rwnd += sctp_data_size(chunk);
403 			}
404 			continue;
405 		}
406 
407 		/* If we are doing  retransmission due to a timeout or pmtu
408 		 * discovery, only the  chunks that are not yet acked should
409 		 * be added to the retransmit queue.
410 		 */
411 		if ((reason == SCTP_RTXR_FAST_RTX  &&
412 			    (chunk->fast_retransmit == SCTP_NEED_FRTX)) ||
413 		    (reason != SCTP_RTXR_FAST_RTX  && !chunk->tsn_gap_acked)) {
414 			/* RFC 2960 6.2.1 Processing a Received SACK
415 			 *
416 			 * C) Any time a DATA chunk is marked for
417 			 * retransmission (via either T3-rtx timer expiration
418 			 * (Section 6.3.3) or via fast retransmit
419 			 * (Section 7.2.4)), add the data size of those
420 			 * chunks to the rwnd.
421 			 */
422 			q->asoc->peer.rwnd += sctp_data_size(chunk);
423 			q->outstanding_bytes -= sctp_data_size(chunk);
424 			if (chunk->transport)
425 				transport->flight_size -= sctp_data_size(chunk);
426 
427 			/* sctpimpguide-05 Section 2.8.2
428 			 * M5) If a T3-rtx timer expires, the
429 			 * 'TSN.Missing.Report' of all affected TSNs is set
430 			 * to 0.
431 			 */
432 			chunk->tsn_missing_report = 0;
433 
434 			/* If a chunk that is being used for RTT measurement
435 			 * has to be retransmitted, we cannot use this chunk
436 			 * anymore for RTT measurements. Reset rto_pending so
437 			 * that a new RTT measurement is started when a new
438 			 * data chunk is sent.
439 			 */
440 			if (chunk->rtt_in_progress) {
441 				chunk->rtt_in_progress = 0;
442 				transport->rto_pending = 0;
443 			}
444 
445 			chunk->resent = 1;
446 
447 			/* Move the chunk to the retransmit queue. The chunks
448 			 * on the retransmit queue are always kept in order.
449 			 */
450 			list_del_init(lchunk);
451 			sctp_insert_list(&q->retransmit, lchunk);
452 		}
453 	}
454 
455 	pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d, "
456 		 "flight_size:%d, pba:%d\n", __func__, transport, reason,
457 		 transport->cwnd, transport->ssthresh, transport->flight_size,
458 		 transport->partial_bytes_acked);
459 }
460 
461 /* Mark all the eligible packets on a transport for retransmission and force
462  * one packet out.
463  */
464 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
465 		     sctp_retransmit_reason_t reason)
466 {
467 	struct net *net = sock_net(q->asoc->base.sk);
468 	int error = 0;
469 
470 	switch (reason) {
471 	case SCTP_RTXR_T3_RTX:
472 		SCTP_INC_STATS(net, SCTP_MIB_T3_RETRANSMITS);
473 		sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
474 		/* Update the retran path if the T3-rtx timer has expired for
475 		 * the current retran path.
476 		 */
477 		if (transport == transport->asoc->peer.retran_path)
478 			sctp_assoc_update_retran_path(transport->asoc);
479 		transport->asoc->rtx_data_chunks +=
480 			transport->asoc->unack_data;
481 		break;
482 	case SCTP_RTXR_FAST_RTX:
483 		SCTP_INC_STATS(net, SCTP_MIB_FAST_RETRANSMITS);
484 		sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
485 		q->fast_rtx = 1;
486 		break;
487 	case SCTP_RTXR_PMTUD:
488 		SCTP_INC_STATS(net, SCTP_MIB_PMTUD_RETRANSMITS);
489 		break;
490 	case SCTP_RTXR_T1_RTX:
491 		SCTP_INC_STATS(net, SCTP_MIB_T1_RETRANSMITS);
492 		transport->asoc->init_retries++;
493 		break;
494 	default:
495 		BUG();
496 	}
497 
498 	sctp_retransmit_mark(q, transport, reason);
499 
500 	/* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
501 	 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
502 	 * following the procedures outlined in C1 - C5.
503 	 */
504 	if (reason == SCTP_RTXR_T3_RTX)
505 		sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point);
506 
507 	/* Flush the queues only on timeout, since fast_rtx is only
508 	 * triggered during sack processing and the queue
509 	 * will be flushed at the end.
510 	 */
511 	if (reason != SCTP_RTXR_FAST_RTX)
512 		error = sctp_outq_flush(q, /* rtx_timeout */ 1);
513 
514 	if (error)
515 		q->asoc->base.sk->sk_err = -error;
516 }
517 
518 /*
519  * Transmit DATA chunks on the retransmit queue.  Upon return from
520  * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
521  * need to be transmitted by the caller.
522  * We assume that pkt->transport has already been set.
523  *
524  * The return value is a normal kernel error return value.
525  */
526 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
527 			       int rtx_timeout, int *start_timer)
528 {
529 	struct list_head *lqueue;
530 	struct sctp_transport *transport = pkt->transport;
531 	sctp_xmit_t status;
532 	struct sctp_chunk *chunk, *chunk1;
533 	int fast_rtx;
534 	int error = 0;
535 	int timer = 0;
536 	int done = 0;
537 
538 	lqueue = &q->retransmit;
539 	fast_rtx = q->fast_rtx;
540 
541 	/* This loop handles time-out retransmissions, fast retransmissions,
542 	 * and retransmissions due to opening of whindow.
543 	 *
544 	 * RFC 2960 6.3.3 Handle T3-rtx Expiration
545 	 *
546 	 * E3) Determine how many of the earliest (i.e., lowest TSN)
547 	 * outstanding DATA chunks for the address for which the
548 	 * T3-rtx has expired will fit into a single packet, subject
549 	 * to the MTU constraint for the path corresponding to the
550 	 * destination transport address to which the retransmission
551 	 * is being sent (this may be different from the address for
552 	 * which the timer expires [see Section 6.4]). Call this value
553 	 * K. Bundle and retransmit those K DATA chunks in a single
554 	 * packet to the destination endpoint.
555 	 *
556 	 * [Just to be painfully clear, if we are retransmitting
557 	 * because a timeout just happened, we should send only ONE
558 	 * packet of retransmitted data.]
559 	 *
560 	 * For fast retransmissions we also send only ONE packet.  However,
561 	 * if we are just flushing the queue due to open window, we'll
562 	 * try to send as much as possible.
563 	 */
564 	list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) {
565 		/* If the chunk is abandoned, move it to abandoned list. */
566 		if (sctp_chunk_abandoned(chunk)) {
567 			list_del_init(&chunk->transmitted_list);
568 			sctp_insert_list(&q->abandoned,
569 					 &chunk->transmitted_list);
570 			continue;
571 		}
572 
573 		/* Make sure that Gap Acked TSNs are not retransmitted.  A
574 		 * simple approach is just to move such TSNs out of the
575 		 * way and into a 'transmitted' queue and skip to the
576 		 * next chunk.
577 		 */
578 		if (chunk->tsn_gap_acked) {
579 			list_move_tail(&chunk->transmitted_list,
580 				       &transport->transmitted);
581 			continue;
582 		}
583 
584 		/* If we are doing fast retransmit, ignore non-fast_rtransmit
585 		 * chunks
586 		 */
587 		if (fast_rtx && !chunk->fast_retransmit)
588 			continue;
589 
590 redo:
591 		/* Attempt to append this chunk to the packet. */
592 		status = sctp_packet_append_chunk(pkt, chunk);
593 
594 		switch (status) {
595 		case SCTP_XMIT_PMTU_FULL:
596 			if (!pkt->has_data && !pkt->has_cookie_echo) {
597 				/* If this packet did not contain DATA then
598 				 * retransmission did not happen, so do it
599 				 * again.  We'll ignore the error here since
600 				 * control chunks are already freed so there
601 				 * is nothing we can do.
602 				 */
603 				sctp_packet_transmit(pkt);
604 				goto redo;
605 			}
606 
607 			/* Send this packet.  */
608 			error = sctp_packet_transmit(pkt);
609 
610 			/* If we are retransmitting, we should only
611 			 * send a single packet.
612 			 * Otherwise, try appending this chunk again.
613 			 */
614 			if (rtx_timeout || fast_rtx)
615 				done = 1;
616 			else
617 				goto redo;
618 
619 			/* Bundle next chunk in the next round.  */
620 			break;
621 
622 		case SCTP_XMIT_RWND_FULL:
623 			/* Send this packet. */
624 			error = sctp_packet_transmit(pkt);
625 
626 			/* Stop sending DATA as there is no more room
627 			 * at the receiver.
628 			 */
629 			done = 1;
630 			break;
631 
632 		case SCTP_XMIT_DELAY:
633 			/* Send this packet. */
634 			error = sctp_packet_transmit(pkt);
635 
636 			/* Stop sending DATA because of nagle delay. */
637 			done = 1;
638 			break;
639 
640 		default:
641 			/* The append was successful, so add this chunk to
642 			 * the transmitted list.
643 			 */
644 			list_move_tail(&chunk->transmitted_list,
645 				       &transport->transmitted);
646 
647 			/* Mark the chunk as ineligible for fast retransmit
648 			 * after it is retransmitted.
649 			 */
650 			if (chunk->fast_retransmit == SCTP_NEED_FRTX)
651 				chunk->fast_retransmit = SCTP_DONT_FRTX;
652 
653 			q->asoc->stats.rtxchunks++;
654 			break;
655 		}
656 
657 		/* Set the timer if there were no errors */
658 		if (!error && !timer)
659 			timer = 1;
660 
661 		if (done)
662 			break;
663 	}
664 
665 	/* If we are here due to a retransmit timeout or a fast
666 	 * retransmit and if there are any chunks left in the retransmit
667 	 * queue that could not fit in the PMTU sized packet, they need
668 	 * to be marked as ineligible for a subsequent fast retransmit.
669 	 */
670 	if (rtx_timeout || fast_rtx) {
671 		list_for_each_entry(chunk1, lqueue, transmitted_list) {
672 			if (chunk1->fast_retransmit == SCTP_NEED_FRTX)
673 				chunk1->fast_retransmit = SCTP_DONT_FRTX;
674 		}
675 	}
676 
677 	*start_timer = timer;
678 
679 	/* Clear fast retransmit hint */
680 	if (fast_rtx)
681 		q->fast_rtx = 0;
682 
683 	return error;
684 }
685 
686 /* Cork the outqueue so queued chunks are really queued. */
687 int sctp_outq_uncork(struct sctp_outq *q)
688 {
689 	if (q->cork)
690 		q->cork = 0;
691 
692 	return sctp_outq_flush(q, 0);
693 }
694 
695 
696 /*
697  * Try to flush an outqueue.
698  *
699  * Description: Send everything in q which we legally can, subject to
700  * congestion limitations.
701  * * Note: This function can be called from multiple contexts so appropriate
702  * locking concerns must be made.  Today we use the sock lock to protect
703  * this function.
704  */
705 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout)
706 {
707 	struct sctp_packet *packet;
708 	struct sctp_packet singleton;
709 	struct sctp_association *asoc = q->asoc;
710 	__u16 sport = asoc->base.bind_addr.port;
711 	__u16 dport = asoc->peer.port;
712 	__u32 vtag = asoc->peer.i.init_tag;
713 	struct sctp_transport *transport = NULL;
714 	struct sctp_transport *new_transport;
715 	struct sctp_chunk *chunk, *tmp;
716 	sctp_xmit_t status;
717 	int error = 0;
718 	int start_timer = 0;
719 	int one_packet = 0;
720 
721 	/* These transports have chunks to send. */
722 	struct list_head transport_list;
723 	struct list_head *ltransport;
724 
725 	INIT_LIST_HEAD(&transport_list);
726 	packet = NULL;
727 
728 	/*
729 	 * 6.10 Bundling
730 	 *   ...
731 	 *   When bundling control chunks with DATA chunks, an
732 	 *   endpoint MUST place control chunks first in the outbound
733 	 *   SCTP packet.  The transmitter MUST transmit DATA chunks
734 	 *   within a SCTP packet in increasing order of TSN.
735 	 *   ...
736 	 */
737 
738 	list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
739 		/* RFC 5061, 5.3
740 		 * F1) This means that until such time as the ASCONF
741 		 * containing the add is acknowledged, the sender MUST
742 		 * NOT use the new IP address as a source for ANY SCTP
743 		 * packet except on carrying an ASCONF Chunk.
744 		 */
745 		if (asoc->src_out_of_asoc_ok &&
746 		    chunk->chunk_hdr->type != SCTP_CID_ASCONF)
747 			continue;
748 
749 		list_del_init(&chunk->list);
750 
751 		/* Pick the right transport to use. */
752 		new_transport = chunk->transport;
753 
754 		if (!new_transport) {
755 			/*
756 			 * If we have a prior transport pointer, see if
757 			 * the destination address of the chunk
758 			 * matches the destination address of the
759 			 * current transport.  If not a match, then
760 			 * try to look up the transport with a given
761 			 * destination address.  We do this because
762 			 * after processing ASCONFs, we may have new
763 			 * transports created.
764 			 */
765 			if (transport &&
766 			    sctp_cmp_addr_exact(&chunk->dest,
767 						&transport->ipaddr))
768 					new_transport = transport;
769 			else
770 				new_transport = sctp_assoc_lookup_paddr(asoc,
771 								&chunk->dest);
772 
773 			/* if we still don't have a new transport, then
774 			 * use the current active path.
775 			 */
776 			if (!new_transport)
777 				new_transport = asoc->peer.active_path;
778 		} else if ((new_transport->state == SCTP_INACTIVE) ||
779 			   (new_transport->state == SCTP_UNCONFIRMED) ||
780 			   (new_transport->state == SCTP_PF)) {
781 			/* If the chunk is Heartbeat or Heartbeat Ack,
782 			 * send it to chunk->transport, even if it's
783 			 * inactive.
784 			 *
785 			 * 3.3.6 Heartbeat Acknowledgement:
786 			 * ...
787 			 * A HEARTBEAT ACK is always sent to the source IP
788 			 * address of the IP datagram containing the
789 			 * HEARTBEAT chunk to which this ack is responding.
790 			 * ...
791 			 *
792 			 * ASCONF_ACKs also must be sent to the source.
793 			 */
794 			if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT &&
795 			    chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK &&
796 			    chunk->chunk_hdr->type != SCTP_CID_ASCONF_ACK)
797 				new_transport = asoc->peer.active_path;
798 		}
799 
800 		/* Are we switching transports?
801 		 * Take care of transport locks.
802 		 */
803 		if (new_transport != transport) {
804 			transport = new_transport;
805 			if (list_empty(&transport->send_ready)) {
806 				list_add_tail(&transport->send_ready,
807 					      &transport_list);
808 			}
809 			packet = &transport->packet;
810 			sctp_packet_config(packet, vtag,
811 					   asoc->peer.ecn_capable);
812 		}
813 
814 		switch (chunk->chunk_hdr->type) {
815 		/*
816 		 * 6.10 Bundling
817 		 *   ...
818 		 *   An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
819 		 *   COMPLETE with any other chunks.  [Send them immediately.]
820 		 */
821 		case SCTP_CID_INIT:
822 		case SCTP_CID_INIT_ACK:
823 		case SCTP_CID_SHUTDOWN_COMPLETE:
824 			sctp_packet_init(&singleton, transport, sport, dport);
825 			sctp_packet_config(&singleton, vtag, 0);
826 			sctp_packet_append_chunk(&singleton, chunk);
827 			error = sctp_packet_transmit(&singleton);
828 			if (error < 0)
829 				return error;
830 			break;
831 
832 		case SCTP_CID_ABORT:
833 			if (sctp_test_T_bit(chunk)) {
834 				packet->vtag = asoc->c.my_vtag;
835 			}
836 		/* The following chunks are "response" chunks, i.e.
837 		 * they are generated in response to something we
838 		 * received.  If we are sending these, then we can
839 		 * send only 1 packet containing these chunks.
840 		 */
841 		case SCTP_CID_HEARTBEAT_ACK:
842 		case SCTP_CID_SHUTDOWN_ACK:
843 		case SCTP_CID_COOKIE_ACK:
844 		case SCTP_CID_COOKIE_ECHO:
845 		case SCTP_CID_ERROR:
846 		case SCTP_CID_ECN_CWR:
847 		case SCTP_CID_ASCONF_ACK:
848 			one_packet = 1;
849 			/* Fall through */
850 
851 		case SCTP_CID_SACK:
852 		case SCTP_CID_HEARTBEAT:
853 		case SCTP_CID_SHUTDOWN:
854 		case SCTP_CID_ECN_ECNE:
855 		case SCTP_CID_ASCONF:
856 		case SCTP_CID_FWD_TSN:
857 			status = sctp_packet_transmit_chunk(packet, chunk,
858 							    one_packet);
859 			if (status  != SCTP_XMIT_OK) {
860 				/* put the chunk back */
861 				list_add(&chunk->list, &q->control_chunk_list);
862 			} else {
863 				asoc->stats.octrlchunks++;
864 				/* PR-SCTP C5) If a FORWARD TSN is sent, the
865 				 * sender MUST assure that at least one T3-rtx
866 				 * timer is running.
867 				 */
868 				if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN)
869 					sctp_transport_reset_timers(transport);
870 			}
871 			break;
872 
873 		default:
874 			/* We built a chunk with an illegal type! */
875 			BUG();
876 		}
877 	}
878 
879 	if (q->asoc->src_out_of_asoc_ok)
880 		goto sctp_flush_out;
881 
882 	/* Is it OK to send data chunks?  */
883 	switch (asoc->state) {
884 	case SCTP_STATE_COOKIE_ECHOED:
885 		/* Only allow bundling when this packet has a COOKIE-ECHO
886 		 * chunk.
887 		 */
888 		if (!packet || !packet->has_cookie_echo)
889 			break;
890 
891 		/* fallthru */
892 	case SCTP_STATE_ESTABLISHED:
893 	case SCTP_STATE_SHUTDOWN_PENDING:
894 	case SCTP_STATE_SHUTDOWN_RECEIVED:
895 		/*
896 		 * RFC 2960 6.1  Transmission of DATA Chunks
897 		 *
898 		 * C) When the time comes for the sender to transmit,
899 		 * before sending new DATA chunks, the sender MUST
900 		 * first transmit any outstanding DATA chunks which
901 		 * are marked for retransmission (limited by the
902 		 * current cwnd).
903 		 */
904 		if (!list_empty(&q->retransmit)) {
905 			if (asoc->peer.retran_path->state == SCTP_UNCONFIRMED)
906 				goto sctp_flush_out;
907 			if (transport == asoc->peer.retran_path)
908 				goto retran;
909 
910 			/* Switch transports & prepare the packet.  */
911 
912 			transport = asoc->peer.retran_path;
913 
914 			if (list_empty(&transport->send_ready)) {
915 				list_add_tail(&transport->send_ready,
916 					      &transport_list);
917 			}
918 
919 			packet = &transport->packet;
920 			sctp_packet_config(packet, vtag,
921 					   asoc->peer.ecn_capable);
922 		retran:
923 			error = sctp_outq_flush_rtx(q, packet,
924 						    rtx_timeout, &start_timer);
925 
926 			if (start_timer)
927 				sctp_transport_reset_timers(transport);
928 
929 			/* This can happen on COOKIE-ECHO resend.  Only
930 			 * one chunk can get bundled with a COOKIE-ECHO.
931 			 */
932 			if (packet->has_cookie_echo)
933 				goto sctp_flush_out;
934 
935 			/* Don't send new data if there is still data
936 			 * waiting to retransmit.
937 			 */
938 			if (!list_empty(&q->retransmit))
939 				goto sctp_flush_out;
940 		}
941 
942 		/* Apply Max.Burst limitation to the current transport in
943 		 * case it will be used for new data.  We are going to
944 		 * rest it before we return, but we want to apply the limit
945 		 * to the currently queued data.
946 		 */
947 		if (transport)
948 			sctp_transport_burst_limited(transport);
949 
950 		/* Finally, transmit new packets.  */
951 		while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
952 			/* RFC 2960 6.5 Every DATA chunk MUST carry a valid
953 			 * stream identifier.
954 			 */
955 			if (chunk->sinfo.sinfo_stream >=
956 			    asoc->c.sinit_num_ostreams) {
957 
958 				/* Mark as failed send. */
959 				sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM);
960 				sctp_chunk_free(chunk);
961 				continue;
962 			}
963 
964 			/* Has this chunk expired? */
965 			if (sctp_chunk_abandoned(chunk)) {
966 				sctp_chunk_fail(chunk, 0);
967 				sctp_chunk_free(chunk);
968 				continue;
969 			}
970 
971 			/* If there is a specified transport, use it.
972 			 * Otherwise, we want to use the active path.
973 			 */
974 			new_transport = chunk->transport;
975 			if (!new_transport ||
976 			    ((new_transport->state == SCTP_INACTIVE) ||
977 			     (new_transport->state == SCTP_UNCONFIRMED) ||
978 			     (new_transport->state == SCTP_PF)))
979 				new_transport = asoc->peer.active_path;
980 			if (new_transport->state == SCTP_UNCONFIRMED)
981 				continue;
982 
983 			/* Change packets if necessary.  */
984 			if (new_transport != transport) {
985 				transport = new_transport;
986 
987 				/* Schedule to have this transport's
988 				 * packet flushed.
989 				 */
990 				if (list_empty(&transport->send_ready)) {
991 					list_add_tail(&transport->send_ready,
992 						      &transport_list);
993 				}
994 
995 				packet = &transport->packet;
996 				sctp_packet_config(packet, vtag,
997 						   asoc->peer.ecn_capable);
998 				/* We've switched transports, so apply the
999 				 * Burst limit to the new transport.
1000 				 */
1001 				sctp_transport_burst_limited(transport);
1002 			}
1003 
1004 			pr_debug("%s: outq:%p, chunk:%p[%s], tx-tsn:0x%x skb->head:%p "
1005 				 "skb->users:%d\n",
1006 				 __func__, q, chunk, chunk && chunk->chunk_hdr ?
1007 				 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
1008 				 "illegal chunk", ntohl(chunk->subh.data_hdr->tsn),
1009 				 chunk->skb ? chunk->skb->head : NULL, chunk->skb ?
1010 				 atomic_read(&chunk->skb->users) : -1);
1011 
1012 			/* Add the chunk to the packet.  */
1013 			status = sctp_packet_transmit_chunk(packet, chunk, 0);
1014 
1015 			switch (status) {
1016 			case SCTP_XMIT_PMTU_FULL:
1017 			case SCTP_XMIT_RWND_FULL:
1018 			case SCTP_XMIT_DELAY:
1019 				/* We could not append this chunk, so put
1020 				 * the chunk back on the output queue.
1021 				 */
1022 				pr_debug("%s: could not transmit tsn:0x%x, status:%d\n",
1023 					 __func__, ntohl(chunk->subh.data_hdr->tsn),
1024 					 status);
1025 
1026 				sctp_outq_head_data(q, chunk);
1027 				goto sctp_flush_out;
1028 
1029 			case SCTP_XMIT_OK:
1030 				/* The sender is in the SHUTDOWN-PENDING state,
1031 				 * The sender MAY set the I-bit in the DATA
1032 				 * chunk header.
1033 				 */
1034 				if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING)
1035 					chunk->chunk_hdr->flags |= SCTP_DATA_SACK_IMM;
1036 				if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
1037 					asoc->stats.ouodchunks++;
1038 				else
1039 					asoc->stats.oodchunks++;
1040 
1041 				break;
1042 
1043 			default:
1044 				BUG();
1045 			}
1046 
1047 			/* BUG: We assume that the sctp_packet_transmit()
1048 			 * call below will succeed all the time and add the
1049 			 * chunk to the transmitted list and restart the
1050 			 * timers.
1051 			 * It is possible that the call can fail under OOM
1052 			 * conditions.
1053 			 *
1054 			 * Is this really a problem?  Won't this behave
1055 			 * like a lost TSN?
1056 			 */
1057 			list_add_tail(&chunk->transmitted_list,
1058 				      &transport->transmitted);
1059 
1060 			sctp_transport_reset_timers(transport);
1061 
1062 			/* Only let one DATA chunk get bundled with a
1063 			 * COOKIE-ECHO chunk.
1064 			 */
1065 			if (packet->has_cookie_echo)
1066 				goto sctp_flush_out;
1067 		}
1068 		break;
1069 
1070 	default:
1071 		/* Do nothing.  */
1072 		break;
1073 	}
1074 
1075 sctp_flush_out:
1076 
1077 	/* Before returning, examine all the transports touched in
1078 	 * this call.  Right now, we bluntly force clear all the
1079 	 * transports.  Things might change after we implement Nagle.
1080 	 * But such an examination is still required.
1081 	 *
1082 	 * --xguo
1083 	 */
1084 	while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL) {
1085 		struct sctp_transport *t = list_entry(ltransport,
1086 						      struct sctp_transport,
1087 						      send_ready);
1088 		packet = &t->packet;
1089 		if (!sctp_packet_empty(packet))
1090 			error = sctp_packet_transmit(packet);
1091 
1092 		/* Clear the burst limited state, if any */
1093 		sctp_transport_burst_reset(t);
1094 	}
1095 
1096 	return error;
1097 }
1098 
1099 /* Update unack_data based on the incoming SACK chunk */
1100 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
1101 					struct sctp_sackhdr *sack)
1102 {
1103 	sctp_sack_variable_t *frags;
1104 	__u16 unack_data;
1105 	int i;
1106 
1107 	unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
1108 
1109 	frags = sack->variable;
1110 	for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
1111 		unack_data -= ((ntohs(frags[i].gab.end) -
1112 				ntohs(frags[i].gab.start) + 1));
1113 	}
1114 
1115 	assoc->unack_data = unack_data;
1116 }
1117 
1118 /* This is where we REALLY process a SACK.
1119  *
1120  * Process the SACK against the outqueue.  Mostly, this just frees
1121  * things off the transmitted queue.
1122  */
1123 int sctp_outq_sack(struct sctp_outq *q, struct sctp_chunk *chunk)
1124 {
1125 	struct sctp_association *asoc = q->asoc;
1126 	struct sctp_sackhdr *sack = chunk->subh.sack_hdr;
1127 	struct sctp_transport *transport;
1128 	struct sctp_chunk *tchunk = NULL;
1129 	struct list_head *lchunk, *transport_list, *temp;
1130 	sctp_sack_variable_t *frags = sack->variable;
1131 	__u32 sack_ctsn, ctsn, tsn;
1132 	__u32 highest_tsn, highest_new_tsn;
1133 	__u32 sack_a_rwnd;
1134 	unsigned int outstanding;
1135 	struct sctp_transport *primary = asoc->peer.primary_path;
1136 	int count_of_newacks = 0;
1137 	int gap_ack_blocks;
1138 	u8 accum_moved = 0;
1139 
1140 	/* Grab the association's destination address list. */
1141 	transport_list = &asoc->peer.transport_addr_list;
1142 
1143 	sack_ctsn = ntohl(sack->cum_tsn_ack);
1144 	gap_ack_blocks = ntohs(sack->num_gap_ack_blocks);
1145 	asoc->stats.gapcnt += gap_ack_blocks;
1146 	/*
1147 	 * SFR-CACC algorithm:
1148 	 * On receipt of a SACK the sender SHOULD execute the
1149 	 * following statements.
1150 	 *
1151 	 * 1) If the cumulative ack in the SACK passes next tsn_at_change
1152 	 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1153 	 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1154 	 * all destinations.
1155 	 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1156 	 * is set the receiver of the SACK MUST take the following actions:
1157 	 *
1158 	 * A) Initialize the cacc_saw_newack to 0 for all destination
1159 	 * addresses.
1160 	 *
1161 	 * Only bother if changeover_active is set. Otherwise, this is
1162 	 * totally suboptimal to do on every SACK.
1163 	 */
1164 	if (primary->cacc.changeover_active) {
1165 		u8 clear_cycling = 0;
1166 
1167 		if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1168 			primary->cacc.changeover_active = 0;
1169 			clear_cycling = 1;
1170 		}
1171 
1172 		if (clear_cycling || gap_ack_blocks) {
1173 			list_for_each_entry(transport, transport_list,
1174 					transports) {
1175 				if (clear_cycling)
1176 					transport->cacc.cycling_changeover = 0;
1177 				if (gap_ack_blocks)
1178 					transport->cacc.cacc_saw_newack = 0;
1179 			}
1180 		}
1181 	}
1182 
1183 	/* Get the highest TSN in the sack. */
1184 	highest_tsn = sack_ctsn;
1185 	if (gap_ack_blocks)
1186 		highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end);
1187 
1188 	if (TSN_lt(asoc->highest_sacked, highest_tsn))
1189 		asoc->highest_sacked = highest_tsn;
1190 
1191 	highest_new_tsn = sack_ctsn;
1192 
1193 	/* Run through the retransmit queue.  Credit bytes received
1194 	 * and free those chunks that we can.
1195 	 */
1196 	sctp_check_transmitted(q, &q->retransmit, NULL, NULL, sack, &highest_new_tsn);
1197 
1198 	/* Run through the transmitted queue.
1199 	 * Credit bytes received and free those chunks which we can.
1200 	 *
1201 	 * This is a MASSIVE candidate for optimization.
1202 	 */
1203 	list_for_each_entry(transport, transport_list, transports) {
1204 		sctp_check_transmitted(q, &transport->transmitted,
1205 				       transport, &chunk->source, sack,
1206 				       &highest_new_tsn);
1207 		/*
1208 		 * SFR-CACC algorithm:
1209 		 * C) Let count_of_newacks be the number of
1210 		 * destinations for which cacc_saw_newack is set.
1211 		 */
1212 		if (transport->cacc.cacc_saw_newack)
1213 			count_of_newacks++;
1214 	}
1215 
1216 	/* Move the Cumulative TSN Ack Point if appropriate.  */
1217 	if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) {
1218 		asoc->ctsn_ack_point = sack_ctsn;
1219 		accum_moved = 1;
1220 	}
1221 
1222 	if (gap_ack_blocks) {
1223 
1224 		if (asoc->fast_recovery && accum_moved)
1225 			highest_new_tsn = highest_tsn;
1226 
1227 		list_for_each_entry(transport, transport_list, transports)
1228 			sctp_mark_missing(q, &transport->transmitted, transport,
1229 					  highest_new_tsn, count_of_newacks);
1230 	}
1231 
1232 	/* Update unack_data field in the assoc. */
1233 	sctp_sack_update_unack_data(asoc, sack);
1234 
1235 	ctsn = asoc->ctsn_ack_point;
1236 
1237 	/* Throw away stuff rotting on the sack queue.  */
1238 	list_for_each_safe(lchunk, temp, &q->sacked) {
1239 		tchunk = list_entry(lchunk, struct sctp_chunk,
1240 				    transmitted_list);
1241 		tsn = ntohl(tchunk->subh.data_hdr->tsn);
1242 		if (TSN_lte(tsn, ctsn)) {
1243 			list_del_init(&tchunk->transmitted_list);
1244 			sctp_chunk_free(tchunk);
1245 		}
1246 	}
1247 
1248 	/* ii) Set rwnd equal to the newly received a_rwnd minus the
1249 	 *     number of bytes still outstanding after processing the
1250 	 *     Cumulative TSN Ack and the Gap Ack Blocks.
1251 	 */
1252 
1253 	sack_a_rwnd = ntohl(sack->a_rwnd);
1254 	outstanding = q->outstanding_bytes;
1255 
1256 	if (outstanding < sack_a_rwnd)
1257 		sack_a_rwnd -= outstanding;
1258 	else
1259 		sack_a_rwnd = 0;
1260 
1261 	asoc->peer.rwnd = sack_a_rwnd;
1262 
1263 	sctp_generate_fwdtsn(q, sack_ctsn);
1264 
1265 	pr_debug("%s: sack cumulative tsn ack:0x%x\n", __func__, sack_ctsn);
1266 	pr_debug("%s: cumulative tsn ack of assoc:%p is 0x%x, "
1267 		 "advertised peer ack point:0x%x\n", __func__, asoc, ctsn,
1268 		 asoc->adv_peer_ack_point);
1269 
1270 	return sctp_outq_is_empty(q);
1271 }
1272 
1273 /* Is the outqueue empty?
1274  * The queue is empty when we have not pending data, no in-flight data
1275  * and nothing pending retransmissions.
1276  */
1277 int sctp_outq_is_empty(const struct sctp_outq *q)
1278 {
1279 	return q->out_qlen == 0 && q->outstanding_bytes == 0 &&
1280 	       list_empty(&q->retransmit);
1281 }
1282 
1283 /********************************************************************
1284  * 2nd Level Abstractions
1285  ********************************************************************/
1286 
1287 /* Go through a transport's transmitted list or the association's retransmit
1288  * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1289  * The retransmit list will not have an associated transport.
1290  *
1291  * I added coherent debug information output.	--xguo
1292  *
1293  * Instead of printing 'sacked' or 'kept' for each TSN on the
1294  * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1295  * KEPT TSN6-TSN7, etc.
1296  */
1297 static void sctp_check_transmitted(struct sctp_outq *q,
1298 				   struct list_head *transmitted_queue,
1299 				   struct sctp_transport *transport,
1300 				   union sctp_addr *saddr,
1301 				   struct sctp_sackhdr *sack,
1302 				   __u32 *highest_new_tsn_in_sack)
1303 {
1304 	struct list_head *lchunk;
1305 	struct sctp_chunk *tchunk;
1306 	struct list_head tlist;
1307 	__u32 tsn;
1308 	__u32 sack_ctsn;
1309 	__u32 rtt;
1310 	__u8 restart_timer = 0;
1311 	int bytes_acked = 0;
1312 	int migrate_bytes = 0;
1313 	bool forward_progress = false;
1314 
1315 	sack_ctsn = ntohl(sack->cum_tsn_ack);
1316 
1317 	INIT_LIST_HEAD(&tlist);
1318 
1319 	/* The while loop will skip empty transmitted queues. */
1320 	while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1321 		tchunk = list_entry(lchunk, struct sctp_chunk,
1322 				    transmitted_list);
1323 
1324 		if (sctp_chunk_abandoned(tchunk)) {
1325 			/* Move the chunk to abandoned list. */
1326 			sctp_insert_list(&q->abandoned, lchunk);
1327 
1328 			/* If this chunk has not been acked, stop
1329 			 * considering it as 'outstanding'.
1330 			 */
1331 			if (!tchunk->tsn_gap_acked) {
1332 				if (tchunk->transport)
1333 					tchunk->transport->flight_size -=
1334 							sctp_data_size(tchunk);
1335 				q->outstanding_bytes -= sctp_data_size(tchunk);
1336 			}
1337 			continue;
1338 		}
1339 
1340 		tsn = ntohl(tchunk->subh.data_hdr->tsn);
1341 		if (sctp_acked(sack, tsn)) {
1342 			/* If this queue is the retransmit queue, the
1343 			 * retransmit timer has already reclaimed
1344 			 * the outstanding bytes for this chunk, so only
1345 			 * count bytes associated with a transport.
1346 			 */
1347 			if (transport) {
1348 				/* If this chunk is being used for RTT
1349 				 * measurement, calculate the RTT and update
1350 				 * the RTO using this value.
1351 				 *
1352 				 * 6.3.1 C5) Karn's algorithm: RTT measurements
1353 				 * MUST NOT be made using packets that were
1354 				 * retransmitted (and thus for which it is
1355 				 * ambiguous whether the reply was for the
1356 				 * first instance of the packet or a later
1357 				 * instance).
1358 				 */
1359 				if (!tchunk->tsn_gap_acked &&
1360 				    !tchunk->resent &&
1361 				    tchunk->rtt_in_progress) {
1362 					tchunk->rtt_in_progress = 0;
1363 					rtt = jiffies - tchunk->sent_at;
1364 					sctp_transport_update_rto(transport,
1365 								  rtt);
1366 				}
1367 			}
1368 
1369 			/* If the chunk hasn't been marked as ACKED,
1370 			 * mark it and account bytes_acked if the
1371 			 * chunk had a valid transport (it will not
1372 			 * have a transport if ASCONF had deleted it
1373 			 * while DATA was outstanding).
1374 			 */
1375 			if (!tchunk->tsn_gap_acked) {
1376 				tchunk->tsn_gap_acked = 1;
1377 				if (TSN_lt(*highest_new_tsn_in_sack, tsn))
1378 					*highest_new_tsn_in_sack = tsn;
1379 				bytes_acked += sctp_data_size(tchunk);
1380 				if (!tchunk->transport)
1381 					migrate_bytes += sctp_data_size(tchunk);
1382 				forward_progress = true;
1383 			}
1384 
1385 			if (TSN_lte(tsn, sack_ctsn)) {
1386 				/* RFC 2960  6.3.2 Retransmission Timer Rules
1387 				 *
1388 				 * R3) Whenever a SACK is received
1389 				 * that acknowledges the DATA chunk
1390 				 * with the earliest outstanding TSN
1391 				 * for that address, restart T3-rtx
1392 				 * timer for that address with its
1393 				 * current RTO.
1394 				 */
1395 				restart_timer = 1;
1396 				forward_progress = true;
1397 
1398 				if (!tchunk->tsn_gap_acked) {
1399 					/*
1400 					 * SFR-CACC algorithm:
1401 					 * 2) If the SACK contains gap acks
1402 					 * and the flag CHANGEOVER_ACTIVE is
1403 					 * set the receiver of the SACK MUST
1404 					 * take the following action:
1405 					 *
1406 					 * B) For each TSN t being acked that
1407 					 * has not been acked in any SACK so
1408 					 * far, set cacc_saw_newack to 1 for
1409 					 * the destination that the TSN was
1410 					 * sent to.
1411 					 */
1412 					if (transport &&
1413 					    sack->num_gap_ack_blocks &&
1414 					    q->asoc->peer.primary_path->cacc.
1415 					    changeover_active)
1416 						transport->cacc.cacc_saw_newack
1417 							= 1;
1418 				}
1419 
1420 				list_add_tail(&tchunk->transmitted_list,
1421 					      &q->sacked);
1422 			} else {
1423 				/* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1424 				 * M2) Each time a SACK arrives reporting
1425 				 * 'Stray DATA chunk(s)' record the highest TSN
1426 				 * reported as newly acknowledged, call this
1427 				 * value 'HighestTSNinSack'. A newly
1428 				 * acknowledged DATA chunk is one not
1429 				 * previously acknowledged in a SACK.
1430 				 *
1431 				 * When the SCTP sender of data receives a SACK
1432 				 * chunk that acknowledges, for the first time,
1433 				 * the receipt of a DATA chunk, all the still
1434 				 * unacknowledged DATA chunks whose TSN is
1435 				 * older than that newly acknowledged DATA
1436 				 * chunk, are qualified as 'Stray DATA chunks'.
1437 				 */
1438 				list_add_tail(lchunk, &tlist);
1439 			}
1440 		} else {
1441 			if (tchunk->tsn_gap_acked) {
1442 				pr_debug("%s: receiver reneged on data TSN:0x%x\n",
1443 					 __func__, tsn);
1444 
1445 				tchunk->tsn_gap_acked = 0;
1446 
1447 				if (tchunk->transport)
1448 					bytes_acked -= sctp_data_size(tchunk);
1449 
1450 				/* RFC 2960 6.3.2 Retransmission Timer Rules
1451 				 *
1452 				 * R4) Whenever a SACK is received missing a
1453 				 * TSN that was previously acknowledged via a
1454 				 * Gap Ack Block, start T3-rtx for the
1455 				 * destination address to which the DATA
1456 				 * chunk was originally
1457 				 * transmitted if it is not already running.
1458 				 */
1459 				restart_timer = 1;
1460 			}
1461 
1462 			list_add_tail(lchunk, &tlist);
1463 		}
1464 	}
1465 
1466 	if (transport) {
1467 		if (bytes_acked) {
1468 			struct sctp_association *asoc = transport->asoc;
1469 
1470 			/* We may have counted DATA that was migrated
1471 			 * to this transport due to DEL-IP operation.
1472 			 * Subtract those bytes, since the were never
1473 			 * send on this transport and shouldn't be
1474 			 * credited to this transport.
1475 			 */
1476 			bytes_acked -= migrate_bytes;
1477 
1478 			/* 8.2. When an outstanding TSN is acknowledged,
1479 			 * the endpoint shall clear the error counter of
1480 			 * the destination transport address to which the
1481 			 * DATA chunk was last sent.
1482 			 * The association's overall error counter is
1483 			 * also cleared.
1484 			 */
1485 			transport->error_count = 0;
1486 			transport->asoc->overall_error_count = 0;
1487 			forward_progress = true;
1488 
1489 			/*
1490 			 * While in SHUTDOWN PENDING, we may have started
1491 			 * the T5 shutdown guard timer after reaching the
1492 			 * retransmission limit. Stop that timer as soon
1493 			 * as the receiver acknowledged any data.
1494 			 */
1495 			if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING &&
1496 			    del_timer(&asoc->timers
1497 				[SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]))
1498 					sctp_association_put(asoc);
1499 
1500 			/* Mark the destination transport address as
1501 			 * active if it is not so marked.
1502 			 */
1503 			if ((transport->state == SCTP_INACTIVE ||
1504 			     transport->state == SCTP_UNCONFIRMED) &&
1505 			    sctp_cmp_addr_exact(&transport->ipaddr, saddr)) {
1506 				sctp_assoc_control_transport(
1507 					transport->asoc,
1508 					transport,
1509 					SCTP_TRANSPORT_UP,
1510 					SCTP_RECEIVED_SACK);
1511 			}
1512 
1513 			sctp_transport_raise_cwnd(transport, sack_ctsn,
1514 						  bytes_acked);
1515 
1516 			transport->flight_size -= bytes_acked;
1517 			if (transport->flight_size == 0)
1518 				transport->partial_bytes_acked = 0;
1519 			q->outstanding_bytes -= bytes_acked + migrate_bytes;
1520 		} else {
1521 			/* RFC 2960 6.1, sctpimpguide-06 2.15.2
1522 			 * When a sender is doing zero window probing, it
1523 			 * should not timeout the association if it continues
1524 			 * to receive new packets from the receiver. The
1525 			 * reason is that the receiver MAY keep its window
1526 			 * closed for an indefinite time.
1527 			 * A sender is doing zero window probing when the
1528 			 * receiver's advertised window is zero, and there is
1529 			 * only one data chunk in flight to the receiver.
1530 			 *
1531 			 * Allow the association to timeout while in SHUTDOWN
1532 			 * PENDING or SHUTDOWN RECEIVED in case the receiver
1533 			 * stays in zero window mode forever.
1534 			 */
1535 			if (!q->asoc->peer.rwnd &&
1536 			    !list_empty(&tlist) &&
1537 			    (sack_ctsn+2 == q->asoc->next_tsn) &&
1538 			    q->asoc->state < SCTP_STATE_SHUTDOWN_PENDING) {
1539 				pr_debug("%s: sack received for zero window "
1540 					 "probe:%u\n", __func__, sack_ctsn);
1541 
1542 				q->asoc->overall_error_count = 0;
1543 				transport->error_count = 0;
1544 			}
1545 		}
1546 
1547 		/* RFC 2960 6.3.2 Retransmission Timer Rules
1548 		 *
1549 		 * R2) Whenever all outstanding data sent to an address have
1550 		 * been acknowledged, turn off the T3-rtx timer of that
1551 		 * address.
1552 		 */
1553 		if (!transport->flight_size) {
1554 			if (del_timer(&transport->T3_rtx_timer))
1555 				sctp_transport_put(transport);
1556 		} else if (restart_timer) {
1557 			if (!mod_timer(&transport->T3_rtx_timer,
1558 				       jiffies + transport->rto))
1559 				sctp_transport_hold(transport);
1560 		}
1561 
1562 		if (forward_progress) {
1563 			if (transport->dst)
1564 				dst_confirm(transport->dst);
1565 		}
1566 	}
1567 
1568 	list_splice(&tlist, transmitted_queue);
1569 }
1570 
1571 /* Mark chunks as missing and consequently may get retransmitted. */
1572 static void sctp_mark_missing(struct sctp_outq *q,
1573 			      struct list_head *transmitted_queue,
1574 			      struct sctp_transport *transport,
1575 			      __u32 highest_new_tsn_in_sack,
1576 			      int count_of_newacks)
1577 {
1578 	struct sctp_chunk *chunk;
1579 	__u32 tsn;
1580 	char do_fast_retransmit = 0;
1581 	struct sctp_association *asoc = q->asoc;
1582 	struct sctp_transport *primary = asoc->peer.primary_path;
1583 
1584 	list_for_each_entry(chunk, transmitted_queue, transmitted_list) {
1585 
1586 		tsn = ntohl(chunk->subh.data_hdr->tsn);
1587 
1588 		/* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1589 		 * 'Unacknowledged TSN's', if the TSN number of an
1590 		 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1591 		 * value, increment the 'TSN.Missing.Report' count on that
1592 		 * chunk if it has NOT been fast retransmitted or marked for
1593 		 * fast retransmit already.
1594 		 */
1595 		if (chunk->fast_retransmit == SCTP_CAN_FRTX &&
1596 		    !chunk->tsn_gap_acked &&
1597 		    TSN_lt(tsn, highest_new_tsn_in_sack)) {
1598 
1599 			/* SFR-CACC may require us to skip marking
1600 			 * this chunk as missing.
1601 			 */
1602 			if (!transport || !sctp_cacc_skip(primary,
1603 						chunk->transport,
1604 						count_of_newacks, tsn)) {
1605 				chunk->tsn_missing_report++;
1606 
1607 				pr_debug("%s: tsn:0x%x missing counter:%d\n",
1608 					 __func__, tsn, chunk->tsn_missing_report);
1609 			}
1610 		}
1611 		/*
1612 		 * M4) If any DATA chunk is found to have a
1613 		 * 'TSN.Missing.Report'
1614 		 * value larger than or equal to 3, mark that chunk for
1615 		 * retransmission and start the fast retransmit procedure.
1616 		 */
1617 
1618 		if (chunk->tsn_missing_report >= 3) {
1619 			chunk->fast_retransmit = SCTP_NEED_FRTX;
1620 			do_fast_retransmit = 1;
1621 		}
1622 	}
1623 
1624 	if (transport) {
1625 		if (do_fast_retransmit)
1626 			sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1627 
1628 		pr_debug("%s: transport:%p, cwnd:%d, ssthresh:%d, "
1629 			 "flight_size:%d, pba:%d\n",  __func__, transport,
1630 			 transport->cwnd, transport->ssthresh,
1631 			 transport->flight_size, transport->partial_bytes_acked);
1632 	}
1633 }
1634 
1635 /* Is the given TSN acked by this packet?  */
1636 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1637 {
1638 	int i;
1639 	sctp_sack_variable_t *frags;
1640 	__u16 gap;
1641 	__u32 ctsn = ntohl(sack->cum_tsn_ack);
1642 
1643 	if (TSN_lte(tsn, ctsn))
1644 		goto pass;
1645 
1646 	/* 3.3.4 Selective Acknowledgement (SACK) (3):
1647 	 *
1648 	 * Gap Ack Blocks:
1649 	 *  These fields contain the Gap Ack Blocks. They are repeated
1650 	 *  for each Gap Ack Block up to the number of Gap Ack Blocks
1651 	 *  defined in the Number of Gap Ack Blocks field. All DATA
1652 	 *  chunks with TSNs greater than or equal to (Cumulative TSN
1653 	 *  Ack + Gap Ack Block Start) and less than or equal to
1654 	 *  (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1655 	 *  Block are assumed to have been received correctly.
1656 	 */
1657 
1658 	frags = sack->variable;
1659 	gap = tsn - ctsn;
1660 	for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) {
1661 		if (TSN_lte(ntohs(frags[i].gab.start), gap) &&
1662 		    TSN_lte(gap, ntohs(frags[i].gab.end)))
1663 			goto pass;
1664 	}
1665 
1666 	return 0;
1667 pass:
1668 	return 1;
1669 }
1670 
1671 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1672 				    int nskips, __be16 stream)
1673 {
1674 	int i;
1675 
1676 	for (i = 0; i < nskips; i++) {
1677 		if (skiplist[i].stream == stream)
1678 			return i;
1679 	}
1680 	return i;
1681 }
1682 
1683 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1684 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1685 {
1686 	struct sctp_association *asoc = q->asoc;
1687 	struct sctp_chunk *ftsn_chunk = NULL;
1688 	struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1689 	int nskips = 0;
1690 	int skip_pos = 0;
1691 	__u32 tsn;
1692 	struct sctp_chunk *chunk;
1693 	struct list_head *lchunk, *temp;
1694 
1695 	if (!asoc->peer.prsctp_capable)
1696 		return;
1697 
1698 	/* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1699 	 * received SACK.
1700 	 *
1701 	 * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1702 	 * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1703 	 */
1704 	if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1705 		asoc->adv_peer_ack_point = ctsn;
1706 
1707 	/* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1708 	 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1709 	 * the chunk next in the out-queue space is marked as "abandoned" as
1710 	 * shown in the following example:
1711 	 *
1712 	 * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1713 	 * and the Advanced.Peer.Ack.Point is updated to this value:
1714 	 *
1715 	 *   out-queue at the end of  ==>   out-queue after Adv.Ack.Point
1716 	 *   normal SACK processing           local advancement
1717 	 *                ...                           ...
1718 	 *   Adv.Ack.Pt-> 102 acked                     102 acked
1719 	 *                103 abandoned                 103 abandoned
1720 	 *                104 abandoned     Adv.Ack.P-> 104 abandoned
1721 	 *                105                           105
1722 	 *                106 acked                     106 acked
1723 	 *                ...                           ...
1724 	 *
1725 	 * In this example, the data sender successfully advanced the
1726 	 * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1727 	 */
1728 	list_for_each_safe(lchunk, temp, &q->abandoned) {
1729 		chunk = list_entry(lchunk, struct sctp_chunk,
1730 					transmitted_list);
1731 		tsn = ntohl(chunk->subh.data_hdr->tsn);
1732 
1733 		/* Remove any chunks in the abandoned queue that are acked by
1734 		 * the ctsn.
1735 		 */
1736 		if (TSN_lte(tsn, ctsn)) {
1737 			list_del_init(lchunk);
1738 			sctp_chunk_free(chunk);
1739 		} else {
1740 			if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1741 				asoc->adv_peer_ack_point = tsn;
1742 				if (chunk->chunk_hdr->flags &
1743 					 SCTP_DATA_UNORDERED)
1744 					continue;
1745 				skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1746 						nskips,
1747 						chunk->subh.data_hdr->stream);
1748 				ftsn_skip_arr[skip_pos].stream =
1749 					chunk->subh.data_hdr->stream;
1750 				ftsn_skip_arr[skip_pos].ssn =
1751 					 chunk->subh.data_hdr->ssn;
1752 				if (skip_pos == nskips)
1753 					nskips++;
1754 				if (nskips == 10)
1755 					break;
1756 			} else
1757 				break;
1758 		}
1759 	}
1760 
1761 	/* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1762 	 * is greater than the Cumulative TSN ACK carried in the received
1763 	 * SACK, the data sender MUST send the data receiver a FORWARD TSN
1764 	 * chunk containing the latest value of the
1765 	 * "Advanced.Peer.Ack.Point".
1766 	 *
1767 	 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1768 	 * list each stream and sequence number in the forwarded TSN. This
1769 	 * information will enable the receiver to easily find any
1770 	 * stranded TSN's waiting on stream reorder queues. Each stream
1771 	 * SHOULD only be reported once; this means that if multiple
1772 	 * abandoned messages occur in the same stream then only the
1773 	 * highest abandoned stream sequence number is reported. If the
1774 	 * total size of the FORWARD TSN does NOT fit in a single MTU then
1775 	 * the sender of the FORWARD TSN SHOULD lower the
1776 	 * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1777 	 * single MTU.
1778 	 */
1779 	if (asoc->adv_peer_ack_point > ctsn)
1780 		ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1781 					      nskips, &ftsn_skip_arr[0]);
1782 
1783 	if (ftsn_chunk) {
1784 		list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1785 		SCTP_INC_STATS(sock_net(asoc->base.sk), SCTP_MIB_OUTCTRLCHUNKS);
1786 	}
1787 }
1788