xref: /openbmc/linux/net/sctp/outqueue.c (revision 5bd8e16d)
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, write to
26  * the Free Software Foundation, 59 Temple Place - Suite 330,
27  * Boston, MA 02111-1307, USA.
28  *
29  * Please send any bug reports or fixes you make to the
30  * email address(es):
31  *    lksctp developers <linux-sctp@vger.kernel.org>
32  *
33  * Written or modified by:
34  *    La Monte H.P. Yarroll <piggy@acm.org>
35  *    Karl Knutson          <karl@athena.chicago.il.us>
36  *    Perry Melange         <pmelange@null.cc.uic.edu>
37  *    Xingang Guo           <xingang.guo@intel.com>
38  *    Hui Huang 	    <hui.huang@nokia.com>
39  *    Sridhar Samudrala     <sri@us.ibm.com>
40  *    Jon Grimm             <jgrimm@us.ibm.com>
41  */
42 
43 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
44 
45 #include <linux/types.h>
46 #include <linux/list.h>   /* For struct list_head */
47 #include <linux/socket.h>
48 #include <linux/ip.h>
49 #include <linux/slab.h>
50 #include <net/sock.h>	  /* For skb_set_owner_w */
51 
52 #include <net/sctp/sctp.h>
53 #include <net/sctp/sm.h>
54 
55 /* Declare internal functions here.  */
56 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn);
57 static void sctp_check_transmitted(struct sctp_outq *q,
58 				   struct list_head *transmitted_queue,
59 				   struct sctp_transport *transport,
60 				   union sctp_addr *saddr,
61 				   struct sctp_sackhdr *sack,
62 				   __u32 *highest_new_tsn);
63 
64 static void sctp_mark_missing(struct sctp_outq *q,
65 			      struct list_head *transmitted_queue,
66 			      struct sctp_transport *transport,
67 			      __u32 highest_new_tsn,
68 			      int count_of_newacks);
69 
70 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 sack_ctsn);
71 
72 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout);
73 
74 /* Add data to the front of the queue. */
75 static inline void sctp_outq_head_data(struct sctp_outq *q,
76 					struct sctp_chunk *ch)
77 {
78 	list_add(&ch->list, &q->out_chunk_list);
79 	q->out_qlen += ch->skb->len;
80 }
81 
82 /* Take data from the front of the queue. */
83 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q)
84 {
85 	struct sctp_chunk *ch = NULL;
86 
87 	if (!list_empty(&q->out_chunk_list)) {
88 		struct list_head *entry = q->out_chunk_list.next;
89 
90 		ch = list_entry(entry, struct sctp_chunk, list);
91 		list_del_init(entry);
92 		q->out_qlen -= ch->skb->len;
93 	}
94 	return ch;
95 }
96 /* Add data chunk to the end of the queue. */
97 static inline void sctp_outq_tail_data(struct sctp_outq *q,
98 				       struct sctp_chunk *ch)
99 {
100 	list_add_tail(&ch->list, &q->out_chunk_list);
101 	q->out_qlen += ch->skb->len;
102 }
103 
104 /*
105  * SFR-CACC algorithm:
106  * D) If count_of_newacks is greater than or equal to 2
107  * and t was not sent to the current primary then the
108  * sender MUST NOT increment missing report count for t.
109  */
110 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary,
111 				       struct sctp_transport *transport,
112 				       int count_of_newacks)
113 {
114 	if (count_of_newacks >=2 && transport != primary)
115 		return 1;
116 	return 0;
117 }
118 
119 /*
120  * SFR-CACC algorithm:
121  * F) If count_of_newacks is less than 2, let d be the
122  * destination to which t was sent. If cacc_saw_newack
123  * is 0 for destination d, then the sender MUST NOT
124  * increment missing report count for t.
125  */
126 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport,
127 				       int count_of_newacks)
128 {
129 	if (count_of_newacks < 2 &&
130 			(transport && !transport->cacc.cacc_saw_newack))
131 		return 1;
132 	return 0;
133 }
134 
135 /*
136  * SFR-CACC algorithm:
137  * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD
138  * execute steps C, D, F.
139  *
140  * C has been implemented in sctp_outq_sack
141  */
142 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary,
143 				     struct sctp_transport *transport,
144 				     int count_of_newacks)
145 {
146 	if (!primary->cacc.cycling_changeover) {
147 		if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks))
148 			return 1;
149 		if (sctp_cacc_skip_3_1_f(transport, count_of_newacks))
150 			return 1;
151 		return 0;
152 	}
153 	return 0;
154 }
155 
156 /*
157  * SFR-CACC algorithm:
158  * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less
159  * than next_tsn_at_change of the current primary, then
160  * the sender MUST NOT increment missing report count
161  * for t.
162  */
163 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn)
164 {
165 	if (primary->cacc.cycling_changeover &&
166 	    TSN_lt(tsn, primary->cacc.next_tsn_at_change))
167 		return 1;
168 	return 0;
169 }
170 
171 /*
172  * SFR-CACC algorithm:
173  * 3) If the missing report count for TSN t is to be
174  * incremented according to [RFC2960] and
175  * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set,
176  * then the sender MUST further execute steps 3.1 and
177  * 3.2 to determine if the missing report count for
178  * TSN t SHOULD NOT be incremented.
179  *
180  * 3.3) If 3.1 and 3.2 do not dictate that the missing
181  * report count for t should not be incremented, then
182  * the sender SHOULD increment missing report count for
183  * t (according to [RFC2960] and [SCTP_STEWART_2002]).
184  */
185 static inline int sctp_cacc_skip(struct sctp_transport *primary,
186 				 struct sctp_transport *transport,
187 				 int count_of_newacks,
188 				 __u32 tsn)
189 {
190 	if (primary->cacc.changeover_active &&
191 	    (sctp_cacc_skip_3_1(primary, transport, count_of_newacks) ||
192 	     sctp_cacc_skip_3_2(primary, tsn)))
193 		return 1;
194 	return 0;
195 }
196 
197 /* Initialize an existing sctp_outq.  This does the boring stuff.
198  * You still need to define handlers if you really want to DO
199  * something with this structure...
200  */
201 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q)
202 {
203 	memset(q, 0, sizeof(struct sctp_outq));
204 
205 	q->asoc = asoc;
206 	INIT_LIST_HEAD(&q->out_chunk_list);
207 	INIT_LIST_HEAD(&q->control_chunk_list);
208 	INIT_LIST_HEAD(&q->retransmit);
209 	INIT_LIST_HEAD(&q->sacked);
210 	INIT_LIST_HEAD(&q->abandoned);
211 
212 	q->empty = 1;
213 }
214 
215 /* Free the outqueue structure and any related pending chunks.
216  */
217 static void __sctp_outq_teardown(struct sctp_outq *q)
218 {
219 	struct sctp_transport *transport;
220 	struct list_head *lchunk, *temp;
221 	struct sctp_chunk *chunk, *tmp;
222 
223 	/* Throw away unacknowledged chunks. */
224 	list_for_each_entry(transport, &q->asoc->peer.transport_addr_list,
225 			transports) {
226 		while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) {
227 			chunk = list_entry(lchunk, struct sctp_chunk,
228 					   transmitted_list);
229 			/* Mark as part of a failed message. */
230 			sctp_chunk_fail(chunk, q->error);
231 			sctp_chunk_free(chunk);
232 		}
233 	}
234 
235 	/* Throw away chunks that have been gap ACKed.  */
236 	list_for_each_safe(lchunk, temp, &q->sacked) {
237 		list_del_init(lchunk);
238 		chunk = list_entry(lchunk, struct sctp_chunk,
239 				   transmitted_list);
240 		sctp_chunk_fail(chunk, q->error);
241 		sctp_chunk_free(chunk);
242 	}
243 
244 	/* Throw away any chunks in the retransmit queue. */
245 	list_for_each_safe(lchunk, temp, &q->retransmit) {
246 		list_del_init(lchunk);
247 		chunk = list_entry(lchunk, struct sctp_chunk,
248 				   transmitted_list);
249 		sctp_chunk_fail(chunk, q->error);
250 		sctp_chunk_free(chunk);
251 	}
252 
253 	/* Throw away any chunks that are in the abandoned queue. */
254 	list_for_each_safe(lchunk, temp, &q->abandoned) {
255 		list_del_init(lchunk);
256 		chunk = list_entry(lchunk, struct sctp_chunk,
257 				   transmitted_list);
258 		sctp_chunk_fail(chunk, q->error);
259 		sctp_chunk_free(chunk);
260 	}
261 
262 	/* Throw away any leftover data chunks. */
263 	while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
264 
265 		/* Mark as send failure. */
266 		sctp_chunk_fail(chunk, q->error);
267 		sctp_chunk_free(chunk);
268 	}
269 
270 	/* Throw away any leftover control chunks. */
271 	list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
272 		list_del_init(&chunk->list);
273 		sctp_chunk_free(chunk);
274 	}
275 }
276 
277 void sctp_outq_teardown(struct sctp_outq *q)
278 {
279 	__sctp_outq_teardown(q);
280 	sctp_outq_init(q->asoc, q);
281 }
282 
283 /* Free the outqueue structure and any related pending chunks.  */
284 void sctp_outq_free(struct sctp_outq *q)
285 {
286 	/* Throw away leftover chunks. */
287 	__sctp_outq_teardown(q);
288 }
289 
290 /* Put a new chunk in an sctp_outq.  */
291 int sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk)
292 {
293 	struct net *net = sock_net(q->asoc->base.sk);
294 	int error = 0;
295 
296 	pr_debug("%s: outq:%p, chunk:%p[%s]\n", __func__, q, chunk,
297 		 chunk && chunk->chunk_hdr ?
298 		 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
299 		 "illegal chunk");
300 
301 	/* If it is data, queue it up, otherwise, send it
302 	 * immediately.
303 	 */
304 	if (sctp_chunk_is_data(chunk)) {
305 		/* Is it OK to queue data chunks?  */
306 		/* From 9. Termination of Association
307 		 *
308 		 * When either endpoint performs a shutdown, the
309 		 * association on each peer will stop accepting new
310 		 * data from its user and only deliver data in queue
311 		 * at the time of sending or receiving the SHUTDOWN
312 		 * chunk.
313 		 */
314 		switch (q->asoc->state) {
315 		case SCTP_STATE_CLOSED:
316 		case SCTP_STATE_SHUTDOWN_PENDING:
317 		case SCTP_STATE_SHUTDOWN_SENT:
318 		case SCTP_STATE_SHUTDOWN_RECEIVED:
319 		case SCTP_STATE_SHUTDOWN_ACK_SENT:
320 			/* Cannot send after transport endpoint shutdown */
321 			error = -ESHUTDOWN;
322 			break;
323 
324 		default:
325 			pr_debug("%s: outqueueing: outq:%p, chunk:%p[%s])\n",
326 				 __func__, q, chunk, chunk && chunk->chunk_hdr ?
327 				 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
328 				 "illegal chunk");
329 
330 			sctp_outq_tail_data(q, chunk);
331 			if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
332 				SCTP_INC_STATS(net, SCTP_MIB_OUTUNORDERCHUNKS);
333 			else
334 				SCTP_INC_STATS(net, SCTP_MIB_OUTORDERCHUNKS);
335 			q->empty = 0;
336 			break;
337 		}
338 	} else {
339 		list_add_tail(&chunk->list, &q->control_chunk_list);
340 		SCTP_INC_STATS(net, SCTP_MIB_OUTCTRLCHUNKS);
341 	}
342 
343 	if (error < 0)
344 		return error;
345 
346 	if (!q->cork)
347 		error = sctp_outq_flush(q, 0);
348 
349 	return error;
350 }
351 
352 /* Insert a chunk into the sorted list based on the TSNs.  The retransmit list
353  * and the abandoned list are in ascending order.
354  */
355 static void sctp_insert_list(struct list_head *head, struct list_head *new)
356 {
357 	struct list_head *pos;
358 	struct sctp_chunk *nchunk, *lchunk;
359 	__u32 ntsn, ltsn;
360 	int done = 0;
361 
362 	nchunk = list_entry(new, struct sctp_chunk, transmitted_list);
363 	ntsn = ntohl(nchunk->subh.data_hdr->tsn);
364 
365 	list_for_each(pos, head) {
366 		lchunk = list_entry(pos, struct sctp_chunk, transmitted_list);
367 		ltsn = ntohl(lchunk->subh.data_hdr->tsn);
368 		if (TSN_lt(ntsn, ltsn)) {
369 			list_add(new, pos->prev);
370 			done = 1;
371 			break;
372 		}
373 	}
374 	if (!done)
375 		list_add_tail(new, head);
376 }
377 
378 /* Mark all the eligible packets on a transport for retransmission.  */
379 void sctp_retransmit_mark(struct sctp_outq *q,
380 			  struct sctp_transport *transport,
381 			  __u8 reason)
382 {
383 	struct list_head *lchunk, *ltemp;
384 	struct sctp_chunk *chunk;
385 
386 	/* Walk through the specified transmitted queue.  */
387 	list_for_each_safe(lchunk, ltemp, &transport->transmitted) {
388 		chunk = list_entry(lchunk, struct sctp_chunk,
389 				   transmitted_list);
390 
391 		/* If the chunk is abandoned, move it to abandoned list. */
392 		if (sctp_chunk_abandoned(chunk)) {
393 			list_del_init(lchunk);
394 			sctp_insert_list(&q->abandoned, lchunk);
395 
396 			/* If this chunk has not been previousely acked,
397 			 * stop considering it 'outstanding'.  Our peer
398 			 * will most likely never see it since it will
399 			 * not be retransmitted
400 			 */
401 			if (!chunk->tsn_gap_acked) {
402 				if (chunk->transport)
403 					chunk->transport->flight_size -=
404 							sctp_data_size(chunk);
405 				q->outstanding_bytes -= sctp_data_size(chunk);
406 				q->asoc->peer.rwnd += sctp_data_size(chunk);
407 			}
408 			continue;
409 		}
410 
411 		/* If we are doing  retransmission due to a timeout or pmtu
412 		 * discovery, only the  chunks that are not yet acked should
413 		 * be added to the retransmit queue.
414 		 */
415 		if ((reason == SCTP_RTXR_FAST_RTX  &&
416 			    (chunk->fast_retransmit == SCTP_NEED_FRTX)) ||
417 		    (reason != SCTP_RTXR_FAST_RTX  && !chunk->tsn_gap_acked)) {
418 			/* RFC 2960 6.2.1 Processing a Received SACK
419 			 *
420 			 * C) Any time a DATA chunk is marked for
421 			 * retransmission (via either T3-rtx timer expiration
422 			 * (Section 6.3.3) or via fast retransmit
423 			 * (Section 7.2.4)), add the data size of those
424 			 * chunks to the rwnd.
425 			 */
426 			q->asoc->peer.rwnd += sctp_data_size(chunk);
427 			q->outstanding_bytes -= sctp_data_size(chunk);
428 			if (chunk->transport)
429 				transport->flight_size -= sctp_data_size(chunk);
430 
431 			/* sctpimpguide-05 Section 2.8.2
432 			 * M5) If a T3-rtx timer expires, the
433 			 * 'TSN.Missing.Report' of all affected TSNs is set
434 			 * to 0.
435 			 */
436 			chunk->tsn_missing_report = 0;
437 
438 			/* If a chunk that is being used for RTT measurement
439 			 * has to be retransmitted, we cannot use this chunk
440 			 * anymore for RTT measurements. Reset rto_pending so
441 			 * that a new RTT measurement is started when a new
442 			 * data chunk is sent.
443 			 */
444 			if (chunk->rtt_in_progress) {
445 				chunk->rtt_in_progress = 0;
446 				transport->rto_pending = 0;
447 			}
448 
449 			/* Move the chunk to the retransmit queue. The chunks
450 			 * on the retransmit queue are always kept in order.
451 			 */
452 			list_del_init(lchunk);
453 			sctp_insert_list(&q->retransmit, lchunk);
454 		}
455 	}
456 
457 	pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d, "
458 		 "flight_size:%d, pba:%d\n", __func__, transport, reason,
459 		 transport->cwnd, transport->ssthresh, transport->flight_size,
460 		 transport->partial_bytes_acked);
461 }
462 
463 /* Mark all the eligible packets on a transport for retransmission and force
464  * one packet out.
465  */
466 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
467 		     sctp_retransmit_reason_t reason)
468 {
469 	struct net *net = sock_net(q->asoc->base.sk);
470 	int error = 0;
471 
472 	switch(reason) {
473 	case SCTP_RTXR_T3_RTX:
474 		SCTP_INC_STATS(net, SCTP_MIB_T3_RETRANSMITS);
475 		sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
476 		/* Update the retran path if the T3-rtx timer has expired for
477 		 * the current retran path.
478 		 */
479 		if (transport == transport->asoc->peer.retran_path)
480 			sctp_assoc_update_retran_path(transport->asoc);
481 		transport->asoc->rtx_data_chunks +=
482 			transport->asoc->unack_data;
483 		break;
484 	case SCTP_RTXR_FAST_RTX:
485 		SCTP_INC_STATS(net, SCTP_MIB_FAST_RETRANSMITS);
486 		sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
487 		q->fast_rtx = 1;
488 		break;
489 	case SCTP_RTXR_PMTUD:
490 		SCTP_INC_STATS(net, SCTP_MIB_PMTUD_RETRANSMITS);
491 		break;
492 	case SCTP_RTXR_T1_RTX:
493 		SCTP_INC_STATS(net, SCTP_MIB_T1_RETRANSMITS);
494 		transport->asoc->init_retries++;
495 		break;
496 	default:
497 		BUG();
498 	}
499 
500 	sctp_retransmit_mark(q, transport, reason);
501 
502 	/* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
503 	 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
504 	 * following the procedures outlined in C1 - C5.
505 	 */
506 	if (reason == SCTP_RTXR_T3_RTX)
507 		sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point);
508 
509 	/* Flush the queues only on timeout, since fast_rtx is only
510 	 * triggered during sack processing and the queue
511 	 * will be flushed at the end.
512 	 */
513 	if (reason != SCTP_RTXR_FAST_RTX)
514 		error = sctp_outq_flush(q, /* rtx_timeout */ 1);
515 
516 	if (error)
517 		q->asoc->base.sk->sk_err = -error;
518 }
519 
520 /*
521  * Transmit DATA chunks on the retransmit queue.  Upon return from
522  * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
523  * need to be transmitted by the caller.
524  * We assume that pkt->transport has already been set.
525  *
526  * The return value is a normal kernel error return value.
527  */
528 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
529 			       int rtx_timeout, int *start_timer)
530 {
531 	struct list_head *lqueue;
532 	struct sctp_transport *transport = pkt->transport;
533 	sctp_xmit_t status;
534 	struct sctp_chunk *chunk, *chunk1;
535 	int fast_rtx;
536 	int error = 0;
537 	int timer = 0;
538 	int done = 0;
539 
540 	lqueue = &q->retransmit;
541 	fast_rtx = q->fast_rtx;
542 
543 	/* This loop handles time-out retransmissions, fast retransmissions,
544 	 * and retransmissions due to opening of whindow.
545 	 *
546 	 * RFC 2960 6.3.3 Handle T3-rtx Expiration
547 	 *
548 	 * E3) Determine how many of the earliest (i.e., lowest TSN)
549 	 * outstanding DATA chunks for the address for which the
550 	 * T3-rtx has expired will fit into a single packet, subject
551 	 * to the MTU constraint for the path corresponding to the
552 	 * destination transport address to which the retransmission
553 	 * is being sent (this may be different from the address for
554 	 * which the timer expires [see Section 6.4]). Call this value
555 	 * K. Bundle and retransmit those K DATA chunks in a single
556 	 * packet to the destination endpoint.
557 	 *
558 	 * [Just to be painfully clear, if we are retransmitting
559 	 * because a timeout just happened, we should send only ONE
560 	 * packet of retransmitted data.]
561 	 *
562 	 * For fast retransmissions we also send only ONE packet.  However,
563 	 * if we are just flushing the queue due to open window, we'll
564 	 * try to send as much as possible.
565 	 */
566 	list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) {
567 		/* If the chunk is abandoned, move it to abandoned list. */
568 		if (sctp_chunk_abandoned(chunk)) {
569 			list_del_init(&chunk->transmitted_list);
570 			sctp_insert_list(&q->abandoned,
571 					 &chunk->transmitted_list);
572 			continue;
573 		}
574 
575 		/* Make sure that Gap Acked TSNs are not retransmitted.  A
576 		 * simple approach is just to move such TSNs out of the
577 		 * way and into a 'transmitted' queue and skip to the
578 		 * next chunk.
579 		 */
580 		if (chunk->tsn_gap_acked) {
581 			list_move_tail(&chunk->transmitted_list,
582 				       &transport->transmitted);
583 			continue;
584 		}
585 
586 		/* If we are doing fast retransmit, ignore non-fast_rtransmit
587 		 * chunks
588 		 */
589 		if (fast_rtx && !chunk->fast_retransmit)
590 			continue;
591 
592 redo:
593 		/* Attempt to append this chunk to the packet. */
594 		status = sctp_packet_append_chunk(pkt, chunk);
595 
596 		switch (status) {
597 		case SCTP_XMIT_PMTU_FULL:
598 			if (!pkt->has_data && !pkt->has_cookie_echo) {
599 				/* If this packet did not contain DATA then
600 				 * retransmission did not happen, so do it
601 				 * again.  We'll ignore the error here since
602 				 * control chunks are already freed so there
603 				 * is nothing we can do.
604 				 */
605 				sctp_packet_transmit(pkt);
606 				goto redo;
607 			}
608 
609 			/* Send this packet.  */
610 			error = sctp_packet_transmit(pkt);
611 
612 			/* If we are retransmitting, we should only
613 			 * send a single packet.
614 			 * Otherwise, try appending this chunk again.
615 			 */
616 			if (rtx_timeout || fast_rtx)
617 				done = 1;
618 			else
619 				goto redo;
620 
621 			/* Bundle next chunk in the next round.  */
622 			break;
623 
624 		case SCTP_XMIT_RWND_FULL:
625 			/* Send this packet. */
626 			error = sctp_packet_transmit(pkt);
627 
628 			/* Stop sending DATA as there is no more room
629 			 * at the receiver.
630 			 */
631 			done = 1;
632 			break;
633 
634 		case SCTP_XMIT_NAGLE_DELAY:
635 			/* Send this packet. */
636 			error = sctp_packet_transmit(pkt);
637 
638 			/* Stop sending DATA because of nagle delay. */
639 			done = 1;
640 			break;
641 
642 		default:
643 			/* The append was successful, so add this chunk to
644 			 * the transmitted list.
645 			 */
646 			list_move_tail(&chunk->transmitted_list,
647 				       &transport->transmitted);
648 
649 			/* Mark the chunk as ineligible for fast retransmit
650 			 * after it is retransmitted.
651 			 */
652 			if (chunk->fast_retransmit == SCTP_NEED_FRTX)
653 				chunk->fast_retransmit = SCTP_DONT_FRTX;
654 
655 			q->empty = 0;
656 			q->asoc->stats.rtxchunks++;
657 			break;
658 		}
659 
660 		/* Set the timer if there were no errors */
661 		if (!error && !timer)
662 			timer = 1;
663 
664 		if (done)
665 			break;
666 	}
667 
668 	/* If we are here due to a retransmit timeout or a fast
669 	 * retransmit and if there are any chunks left in the retransmit
670 	 * queue that could not fit in the PMTU sized packet, they need
671 	 * to be marked as ineligible for a subsequent fast retransmit.
672 	 */
673 	if (rtx_timeout || fast_rtx) {
674 		list_for_each_entry(chunk1, lqueue, transmitted_list) {
675 			if (chunk1->fast_retransmit == SCTP_NEED_FRTX)
676 				chunk1->fast_retransmit = SCTP_DONT_FRTX;
677 		}
678 	}
679 
680 	*start_timer = timer;
681 
682 	/* Clear fast retransmit hint */
683 	if (fast_rtx)
684 		q->fast_rtx = 0;
685 
686 	return error;
687 }
688 
689 /* Cork the outqueue so queued chunks are really queued. */
690 int sctp_outq_uncork(struct sctp_outq *q)
691 {
692 	if (q->cork)
693 		q->cork = 0;
694 
695 	return sctp_outq_flush(q, 0);
696 }
697 
698 
699 /*
700  * Try to flush an outqueue.
701  *
702  * Description: Send everything in q which we legally can, subject to
703  * congestion limitations.
704  * * Note: This function can be called from multiple contexts so appropriate
705  * locking concerns must be made.  Today we use the sock lock to protect
706  * this function.
707  */
708 static int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout)
709 {
710 	struct sctp_packet *packet;
711 	struct sctp_packet singleton;
712 	struct sctp_association *asoc = q->asoc;
713 	__u16 sport = asoc->base.bind_addr.port;
714 	__u16 dport = asoc->peer.port;
715 	__u32 vtag = asoc->peer.i.init_tag;
716 	struct sctp_transport *transport = NULL;
717 	struct sctp_transport *new_transport;
718 	struct sctp_chunk *chunk, *tmp;
719 	sctp_xmit_t status;
720 	int error = 0;
721 	int start_timer = 0;
722 	int one_packet = 0;
723 
724 	/* These transports have chunks to send. */
725 	struct list_head transport_list;
726 	struct list_head *ltransport;
727 
728 	INIT_LIST_HEAD(&transport_list);
729 	packet = NULL;
730 
731 	/*
732 	 * 6.10 Bundling
733 	 *   ...
734 	 *   When bundling control chunks with DATA chunks, an
735 	 *   endpoint MUST place control chunks first in the outbound
736 	 *   SCTP packet.  The transmitter MUST transmit DATA chunks
737 	 *   within a SCTP packet in increasing order of TSN.
738 	 *   ...
739 	 */
740 
741 	list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
742 		/* RFC 5061, 5.3
743 		 * F1) This means that until such time as the ASCONF
744 		 * containing the add is acknowledged, the sender MUST
745 		 * NOT use the new IP address as a source for ANY SCTP
746 		 * packet except on carrying an ASCONF Chunk.
747 		 */
748 		if (asoc->src_out_of_asoc_ok &&
749 		    chunk->chunk_hdr->type != SCTP_CID_ASCONF)
750 			continue;
751 
752 		list_del_init(&chunk->list);
753 
754 		/* Pick the right transport to use. */
755 		new_transport = chunk->transport;
756 
757 		if (!new_transport) {
758 			/*
759 			 * If we have a prior transport pointer, see if
760 			 * the destination address of the chunk
761 			 * matches the destination address of the
762 			 * current transport.  If not a match, then
763 			 * try to look up the transport with a given
764 			 * destination address.  We do this because
765 			 * after processing ASCONFs, we may have new
766 			 * transports created.
767 			 */
768 			if (transport &&
769 			    sctp_cmp_addr_exact(&chunk->dest,
770 						&transport->ipaddr))
771 					new_transport = transport;
772 			else
773 				new_transport = sctp_assoc_lookup_paddr(asoc,
774 								&chunk->dest);
775 
776 			/* if we still don't have a new transport, then
777 			 * use the current active path.
778 			 */
779 			if (!new_transport)
780 				new_transport = asoc->peer.active_path;
781 		} else if ((new_transport->state == SCTP_INACTIVE) ||
782 			   (new_transport->state == SCTP_UNCONFIRMED) ||
783 			   (new_transport->state == SCTP_PF)) {
784 			/* If the chunk is Heartbeat or Heartbeat Ack,
785 			 * send it to chunk->transport, even if it's
786 			 * inactive.
787 			 *
788 			 * 3.3.6 Heartbeat Acknowledgement:
789 			 * ...
790 			 * A HEARTBEAT ACK is always sent to the source IP
791 			 * address of the IP datagram containing the
792 			 * HEARTBEAT chunk to which this ack is responding.
793 			 * ...
794 			 *
795 			 * ASCONF_ACKs also must be sent to the source.
796 			 */
797 			if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT &&
798 			    chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK &&
799 			    chunk->chunk_hdr->type != SCTP_CID_ASCONF_ACK)
800 				new_transport = asoc->peer.active_path;
801 		}
802 
803 		/* Are we switching transports?
804 		 * Take care of transport locks.
805 		 */
806 		if (new_transport != transport) {
807 			transport = new_transport;
808 			if (list_empty(&transport->send_ready)) {
809 				list_add_tail(&transport->send_ready,
810 					      &transport_list);
811 			}
812 			packet = &transport->packet;
813 			sctp_packet_config(packet, vtag,
814 					   asoc->peer.ecn_capable);
815 		}
816 
817 		switch (chunk->chunk_hdr->type) {
818 		/*
819 		 * 6.10 Bundling
820 		 *   ...
821 		 *   An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
822 		 *   COMPLETE with any other chunks.  [Send them immediately.]
823 		 */
824 		case SCTP_CID_INIT:
825 		case SCTP_CID_INIT_ACK:
826 		case SCTP_CID_SHUTDOWN_COMPLETE:
827 			sctp_packet_init(&singleton, transport, sport, dport);
828 			sctp_packet_config(&singleton, vtag, 0);
829 			sctp_packet_append_chunk(&singleton, chunk);
830 			error = sctp_packet_transmit(&singleton);
831 			if (error < 0)
832 				return error;
833 			break;
834 
835 		case SCTP_CID_ABORT:
836 			if (sctp_test_T_bit(chunk)) {
837 				packet->vtag = asoc->c.my_vtag;
838 			}
839 		/* The following chunks are "response" chunks, i.e.
840 		 * they are generated in response to something we
841 		 * received.  If we are sending these, then we can
842 		 * send only 1 packet containing these chunks.
843 		 */
844 		case SCTP_CID_HEARTBEAT_ACK:
845 		case SCTP_CID_SHUTDOWN_ACK:
846 		case SCTP_CID_COOKIE_ACK:
847 		case SCTP_CID_COOKIE_ECHO:
848 		case SCTP_CID_ERROR:
849 		case SCTP_CID_ECN_CWR:
850 		case SCTP_CID_ASCONF_ACK:
851 			one_packet = 1;
852 			/* Fall through */
853 
854 		case SCTP_CID_SACK:
855 		case SCTP_CID_HEARTBEAT:
856 		case SCTP_CID_SHUTDOWN:
857 		case SCTP_CID_ECN_ECNE:
858 		case SCTP_CID_ASCONF:
859 		case SCTP_CID_FWD_TSN:
860 			status = sctp_packet_transmit_chunk(packet, chunk,
861 							    one_packet);
862 			if (status  != SCTP_XMIT_OK) {
863 				/* put the chunk back */
864 				list_add(&chunk->list, &q->control_chunk_list);
865 			} else {
866 				asoc->stats.octrlchunks++;
867 				/* PR-SCTP C5) If a FORWARD TSN is sent, the
868 				 * sender MUST assure that at least one T3-rtx
869 				 * timer is running.
870 				 */
871 				if (chunk->chunk_hdr->type == SCTP_CID_FWD_TSN)
872 					sctp_transport_reset_timers(transport);
873 			}
874 			break;
875 
876 		default:
877 			/* We built a chunk with an illegal type! */
878 			BUG();
879 		}
880 	}
881 
882 	if (q->asoc->src_out_of_asoc_ok)
883 		goto sctp_flush_out;
884 
885 	/* Is it OK to send data chunks?  */
886 	switch (asoc->state) {
887 	case SCTP_STATE_COOKIE_ECHOED:
888 		/* Only allow bundling when this packet has a COOKIE-ECHO
889 		 * chunk.
890 		 */
891 		if (!packet || !packet->has_cookie_echo)
892 			break;
893 
894 		/* fallthru */
895 	case SCTP_STATE_ESTABLISHED:
896 	case SCTP_STATE_SHUTDOWN_PENDING:
897 	case SCTP_STATE_SHUTDOWN_RECEIVED:
898 		/*
899 		 * RFC 2960 6.1  Transmission of DATA Chunks
900 		 *
901 		 * C) When the time comes for the sender to transmit,
902 		 * before sending new DATA chunks, the sender MUST
903 		 * first transmit any outstanding DATA chunks which
904 		 * are marked for retransmission (limited by the
905 		 * current cwnd).
906 		 */
907 		if (!list_empty(&q->retransmit)) {
908 			if (asoc->peer.retran_path->state == SCTP_UNCONFIRMED)
909 				goto sctp_flush_out;
910 			if (transport == asoc->peer.retran_path)
911 				goto retran;
912 
913 			/* Switch transports & prepare the packet.  */
914 
915 			transport = asoc->peer.retran_path;
916 
917 			if (list_empty(&transport->send_ready)) {
918 				list_add_tail(&transport->send_ready,
919 					      &transport_list);
920 			}
921 
922 			packet = &transport->packet;
923 			sctp_packet_config(packet, vtag,
924 					   asoc->peer.ecn_capable);
925 		retran:
926 			error = sctp_outq_flush_rtx(q, packet,
927 						    rtx_timeout, &start_timer);
928 
929 			if (start_timer)
930 				sctp_transport_reset_timers(transport);
931 
932 			/* This can happen on COOKIE-ECHO resend.  Only
933 			 * one chunk can get bundled with a COOKIE-ECHO.
934 			 */
935 			if (packet->has_cookie_echo)
936 				goto sctp_flush_out;
937 
938 			/* Don't send new data if there is still data
939 			 * waiting to retransmit.
940 			 */
941 			if (!list_empty(&q->retransmit))
942 				goto sctp_flush_out;
943 		}
944 
945 		/* Apply Max.Burst limitation to the current transport in
946 		 * case it will be used for new data.  We are going to
947 		 * rest it before we return, but we want to apply the limit
948 		 * to the currently queued data.
949 		 */
950 		if (transport)
951 			sctp_transport_burst_limited(transport);
952 
953 		/* Finally, transmit new packets.  */
954 		while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
955 			/* RFC 2960 6.5 Every DATA chunk MUST carry a valid
956 			 * stream identifier.
957 			 */
958 			if (chunk->sinfo.sinfo_stream >=
959 			    asoc->c.sinit_num_ostreams) {
960 
961 				/* Mark as failed send. */
962 				sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM);
963 				sctp_chunk_free(chunk);
964 				continue;
965 			}
966 
967 			/* Has this chunk expired? */
968 			if (sctp_chunk_abandoned(chunk)) {
969 				sctp_chunk_fail(chunk, 0);
970 				sctp_chunk_free(chunk);
971 				continue;
972 			}
973 
974 			/* If there is a specified transport, use it.
975 			 * Otherwise, we want to use the active path.
976 			 */
977 			new_transport = chunk->transport;
978 			if (!new_transport ||
979 			    ((new_transport->state == SCTP_INACTIVE) ||
980 			     (new_transport->state == SCTP_UNCONFIRMED) ||
981 			     (new_transport->state == SCTP_PF)))
982 				new_transport = asoc->peer.active_path;
983 			if (new_transport->state == SCTP_UNCONFIRMED)
984 				continue;
985 
986 			/* Change packets if necessary.  */
987 			if (new_transport != transport) {
988 				transport = new_transport;
989 
990 				/* Schedule to have this transport's
991 				 * packet flushed.
992 				 */
993 				if (list_empty(&transport->send_ready)) {
994 					list_add_tail(&transport->send_ready,
995 						      &transport_list);
996 				}
997 
998 				packet = &transport->packet;
999 				sctp_packet_config(packet, vtag,
1000 						   asoc->peer.ecn_capable);
1001 				/* We've switched transports, so apply the
1002 				 * Burst limit to the new transport.
1003 				 */
1004 				sctp_transport_burst_limited(transport);
1005 			}
1006 
1007 			pr_debug("%s: outq:%p, chunk:%p[%s], tx-tsn:0x%x skb->head:%p "
1008 				 "skb->users:%d\n",
1009 				 __func__, q, chunk, chunk && chunk->chunk_hdr ?
1010 				 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)) :
1011 				 "illegal chunk", ntohl(chunk->subh.data_hdr->tsn),
1012 				 chunk->skb ? chunk->skb->head : NULL, chunk->skb ?
1013 				 atomic_read(&chunk->skb->users) : -1);
1014 
1015 			/* Add the chunk to the packet.  */
1016 			status = sctp_packet_transmit_chunk(packet, chunk, 0);
1017 
1018 			switch (status) {
1019 			case SCTP_XMIT_PMTU_FULL:
1020 			case SCTP_XMIT_RWND_FULL:
1021 			case SCTP_XMIT_NAGLE_DELAY:
1022 				/* We could not append this chunk, so put
1023 				 * the chunk back on the output queue.
1024 				 */
1025 				pr_debug("%s: could not transmit tsn:0x%x, status:%d\n",
1026 					 __func__, ntohl(chunk->subh.data_hdr->tsn),
1027 					 status);
1028 
1029 				sctp_outq_head_data(q, chunk);
1030 				goto sctp_flush_out;
1031 				break;
1032 
1033 			case SCTP_XMIT_OK:
1034 				/* The sender is in the SHUTDOWN-PENDING state,
1035 				 * The sender MAY set the I-bit in the DATA
1036 				 * chunk header.
1037 				 */
1038 				if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING)
1039 					chunk->chunk_hdr->flags |= SCTP_DATA_SACK_IMM;
1040 				if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
1041 					asoc->stats.ouodchunks++;
1042 				else
1043 					asoc->stats.oodchunks++;
1044 
1045 				break;
1046 
1047 			default:
1048 				BUG();
1049 			}
1050 
1051 			/* BUG: We assume that the sctp_packet_transmit()
1052 			 * call below will succeed all the time and add the
1053 			 * chunk to the transmitted list and restart the
1054 			 * timers.
1055 			 * It is possible that the call can fail under OOM
1056 			 * conditions.
1057 			 *
1058 			 * Is this really a problem?  Won't this behave
1059 			 * like a lost TSN?
1060 			 */
1061 			list_add_tail(&chunk->transmitted_list,
1062 				      &transport->transmitted);
1063 
1064 			sctp_transport_reset_timers(transport);
1065 
1066 			q->empty = 0;
1067 
1068 			/* Only let one DATA chunk get bundled with a
1069 			 * COOKIE-ECHO chunk.
1070 			 */
1071 			if (packet->has_cookie_echo)
1072 				goto sctp_flush_out;
1073 		}
1074 		break;
1075 
1076 	default:
1077 		/* Do nothing.  */
1078 		break;
1079 	}
1080 
1081 sctp_flush_out:
1082 
1083 	/* Before returning, examine all the transports touched in
1084 	 * this call.  Right now, we bluntly force clear all the
1085 	 * transports.  Things might change after we implement Nagle.
1086 	 * But such an examination is still required.
1087 	 *
1088 	 * --xguo
1089 	 */
1090 	while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL ) {
1091 		struct sctp_transport *t = list_entry(ltransport,
1092 						      struct sctp_transport,
1093 						      send_ready);
1094 		packet = &t->packet;
1095 		if (!sctp_packet_empty(packet))
1096 			error = sctp_packet_transmit(packet);
1097 
1098 		/* Clear the burst limited state, if any */
1099 		sctp_transport_burst_reset(t);
1100 	}
1101 
1102 	return error;
1103 }
1104 
1105 /* Update unack_data based on the incoming SACK chunk */
1106 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
1107 					struct sctp_sackhdr *sack)
1108 {
1109 	sctp_sack_variable_t *frags;
1110 	__u16 unack_data;
1111 	int i;
1112 
1113 	unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
1114 
1115 	frags = sack->variable;
1116 	for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
1117 		unack_data -= ((ntohs(frags[i].gab.end) -
1118 				ntohs(frags[i].gab.start) + 1));
1119 	}
1120 
1121 	assoc->unack_data = unack_data;
1122 }
1123 
1124 /* This is where we REALLY process a SACK.
1125  *
1126  * Process the SACK against the outqueue.  Mostly, this just frees
1127  * things off the transmitted queue.
1128  */
1129 int sctp_outq_sack(struct sctp_outq *q, struct sctp_chunk *chunk)
1130 {
1131 	struct sctp_association *asoc = q->asoc;
1132 	struct sctp_sackhdr *sack = chunk->subh.sack_hdr;
1133 	struct sctp_transport *transport;
1134 	struct sctp_chunk *tchunk = NULL;
1135 	struct list_head *lchunk, *transport_list, *temp;
1136 	sctp_sack_variable_t *frags = sack->variable;
1137 	__u32 sack_ctsn, ctsn, tsn;
1138 	__u32 highest_tsn, highest_new_tsn;
1139 	__u32 sack_a_rwnd;
1140 	unsigned int outstanding;
1141 	struct sctp_transport *primary = asoc->peer.primary_path;
1142 	int count_of_newacks = 0;
1143 	int gap_ack_blocks;
1144 	u8 accum_moved = 0;
1145 
1146 	/* Grab the association's destination address list. */
1147 	transport_list = &asoc->peer.transport_addr_list;
1148 
1149 	sack_ctsn = ntohl(sack->cum_tsn_ack);
1150 	gap_ack_blocks = ntohs(sack->num_gap_ack_blocks);
1151 	asoc->stats.gapcnt += gap_ack_blocks;
1152 	/*
1153 	 * SFR-CACC algorithm:
1154 	 * On receipt of a SACK the sender SHOULD execute the
1155 	 * following statements.
1156 	 *
1157 	 * 1) If the cumulative ack in the SACK passes next tsn_at_change
1158 	 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1159 	 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1160 	 * all destinations.
1161 	 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1162 	 * is set the receiver of the SACK MUST take the following actions:
1163 	 *
1164 	 * A) Initialize the cacc_saw_newack to 0 for all destination
1165 	 * addresses.
1166 	 *
1167 	 * Only bother if changeover_active is set. Otherwise, this is
1168 	 * totally suboptimal to do on every SACK.
1169 	 */
1170 	if (primary->cacc.changeover_active) {
1171 		u8 clear_cycling = 0;
1172 
1173 		if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1174 			primary->cacc.changeover_active = 0;
1175 			clear_cycling = 1;
1176 		}
1177 
1178 		if (clear_cycling || gap_ack_blocks) {
1179 			list_for_each_entry(transport, transport_list,
1180 					transports) {
1181 				if (clear_cycling)
1182 					transport->cacc.cycling_changeover = 0;
1183 				if (gap_ack_blocks)
1184 					transport->cacc.cacc_saw_newack = 0;
1185 			}
1186 		}
1187 	}
1188 
1189 	/* Get the highest TSN in the sack. */
1190 	highest_tsn = sack_ctsn;
1191 	if (gap_ack_blocks)
1192 		highest_tsn += ntohs(frags[gap_ack_blocks - 1].gab.end);
1193 
1194 	if (TSN_lt(asoc->highest_sacked, highest_tsn))
1195 		asoc->highest_sacked = highest_tsn;
1196 
1197 	highest_new_tsn = sack_ctsn;
1198 
1199 	/* Run through the retransmit queue.  Credit bytes received
1200 	 * and free those chunks that we can.
1201 	 */
1202 	sctp_check_transmitted(q, &q->retransmit, NULL, NULL, sack, &highest_new_tsn);
1203 
1204 	/* Run through the transmitted queue.
1205 	 * Credit bytes received and free those chunks which we can.
1206 	 *
1207 	 * This is a MASSIVE candidate for optimization.
1208 	 */
1209 	list_for_each_entry(transport, transport_list, transports) {
1210 		sctp_check_transmitted(q, &transport->transmitted,
1211 				       transport, &chunk->source, sack,
1212 				       &highest_new_tsn);
1213 		/*
1214 		 * SFR-CACC algorithm:
1215 		 * C) Let count_of_newacks be the number of
1216 		 * destinations for which cacc_saw_newack is set.
1217 		 */
1218 		if (transport->cacc.cacc_saw_newack)
1219 			count_of_newacks ++;
1220 	}
1221 
1222 	/* Move the Cumulative TSN Ack Point if appropriate.  */
1223 	if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn)) {
1224 		asoc->ctsn_ack_point = sack_ctsn;
1225 		accum_moved = 1;
1226 	}
1227 
1228 	if (gap_ack_blocks) {
1229 
1230 		if (asoc->fast_recovery && accum_moved)
1231 			highest_new_tsn = highest_tsn;
1232 
1233 		list_for_each_entry(transport, transport_list, transports)
1234 			sctp_mark_missing(q, &transport->transmitted, transport,
1235 					  highest_new_tsn, count_of_newacks);
1236 	}
1237 
1238 	/* Update unack_data field in the assoc. */
1239 	sctp_sack_update_unack_data(asoc, sack);
1240 
1241 	ctsn = asoc->ctsn_ack_point;
1242 
1243 	/* Throw away stuff rotting on the sack queue.  */
1244 	list_for_each_safe(lchunk, temp, &q->sacked) {
1245 		tchunk = list_entry(lchunk, struct sctp_chunk,
1246 				    transmitted_list);
1247 		tsn = ntohl(tchunk->subh.data_hdr->tsn);
1248 		if (TSN_lte(tsn, ctsn)) {
1249 			list_del_init(&tchunk->transmitted_list);
1250 			sctp_chunk_free(tchunk);
1251 		}
1252 	}
1253 
1254 	/* ii) Set rwnd equal to the newly received a_rwnd minus the
1255 	 *     number of bytes still outstanding after processing the
1256 	 *     Cumulative TSN Ack and the Gap Ack Blocks.
1257 	 */
1258 
1259 	sack_a_rwnd = ntohl(sack->a_rwnd);
1260 	outstanding = q->outstanding_bytes;
1261 
1262 	if (outstanding < sack_a_rwnd)
1263 		sack_a_rwnd -= outstanding;
1264 	else
1265 		sack_a_rwnd = 0;
1266 
1267 	asoc->peer.rwnd = sack_a_rwnd;
1268 
1269 	sctp_generate_fwdtsn(q, sack_ctsn);
1270 
1271 	pr_debug("%s: sack cumulative tsn ack:0x%x\n", __func__, sack_ctsn);
1272 	pr_debug("%s: cumulative tsn ack of assoc:%p is 0x%x, "
1273 		 "advertised peer ack point:0x%x\n", __func__, asoc, ctsn,
1274 		 asoc->adv_peer_ack_point);
1275 
1276 	/* See if all chunks are acked.
1277 	 * Make sure the empty queue handler will get run later.
1278 	 */
1279 	q->empty = (list_empty(&q->out_chunk_list) &&
1280 		    list_empty(&q->retransmit));
1281 	if (!q->empty)
1282 		goto finish;
1283 
1284 	list_for_each_entry(transport, transport_list, transports) {
1285 		q->empty = q->empty && list_empty(&transport->transmitted);
1286 		if (!q->empty)
1287 			goto finish;
1288 	}
1289 
1290 	pr_debug("%s: sack queue is empty\n", __func__);
1291 finish:
1292 	return q->empty;
1293 }
1294 
1295 /* Is the outqueue empty?  */
1296 int sctp_outq_is_empty(const struct sctp_outq *q)
1297 {
1298 	return q->empty;
1299 }
1300 
1301 /********************************************************************
1302  * 2nd Level Abstractions
1303  ********************************************************************/
1304 
1305 /* Go through a transport's transmitted list or the association's retransmit
1306  * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1307  * The retransmit list will not have an associated transport.
1308  *
1309  * I added coherent debug information output.	--xguo
1310  *
1311  * Instead of printing 'sacked' or 'kept' for each TSN on the
1312  * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1313  * KEPT TSN6-TSN7, etc.
1314  */
1315 static void sctp_check_transmitted(struct sctp_outq *q,
1316 				   struct list_head *transmitted_queue,
1317 				   struct sctp_transport *transport,
1318 				   union sctp_addr *saddr,
1319 				   struct sctp_sackhdr *sack,
1320 				   __u32 *highest_new_tsn_in_sack)
1321 {
1322 	struct list_head *lchunk;
1323 	struct sctp_chunk *tchunk;
1324 	struct list_head tlist;
1325 	__u32 tsn;
1326 	__u32 sack_ctsn;
1327 	__u32 rtt;
1328 	__u8 restart_timer = 0;
1329 	int bytes_acked = 0;
1330 	int migrate_bytes = 0;
1331 	bool forward_progress = false;
1332 
1333 	sack_ctsn = ntohl(sack->cum_tsn_ack);
1334 
1335 	INIT_LIST_HEAD(&tlist);
1336 
1337 	/* The while loop will skip empty transmitted queues. */
1338 	while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1339 		tchunk = list_entry(lchunk, struct sctp_chunk,
1340 				    transmitted_list);
1341 
1342 		if (sctp_chunk_abandoned(tchunk)) {
1343 			/* Move the chunk to abandoned list. */
1344 			sctp_insert_list(&q->abandoned, lchunk);
1345 
1346 			/* If this chunk has not been acked, stop
1347 			 * considering it as 'outstanding'.
1348 			 */
1349 			if (!tchunk->tsn_gap_acked) {
1350 				if (tchunk->transport)
1351 					tchunk->transport->flight_size -=
1352 							sctp_data_size(tchunk);
1353 				q->outstanding_bytes -= sctp_data_size(tchunk);
1354 			}
1355 			continue;
1356 		}
1357 
1358 		tsn = ntohl(tchunk->subh.data_hdr->tsn);
1359 		if (sctp_acked(sack, tsn)) {
1360 			/* If this queue is the retransmit queue, the
1361 			 * retransmit timer has already reclaimed
1362 			 * the outstanding bytes for this chunk, so only
1363 			 * count bytes associated with a transport.
1364 			 */
1365 			if (transport) {
1366 				/* If this chunk is being used for RTT
1367 				 * measurement, calculate the RTT and update
1368 				 * the RTO using this value.
1369 				 *
1370 				 * 6.3.1 C5) Karn's algorithm: RTT measurements
1371 				 * MUST NOT be made using packets that were
1372 				 * retransmitted (and thus for which it is
1373 				 * ambiguous whether the reply was for the
1374 				 * first instance of the packet or a later
1375 				 * instance).
1376 				 */
1377 				if (!tchunk->tsn_gap_acked &&
1378 				    tchunk->rtt_in_progress) {
1379 					tchunk->rtt_in_progress = 0;
1380 					rtt = jiffies - tchunk->sent_at;
1381 					sctp_transport_update_rto(transport,
1382 								  rtt);
1383 				}
1384 			}
1385 
1386 			/* If the chunk hasn't been marked as ACKED,
1387 			 * mark it and account bytes_acked if the
1388 			 * chunk had a valid transport (it will not
1389 			 * have a transport if ASCONF had deleted it
1390 			 * while DATA was outstanding).
1391 			 */
1392 			if (!tchunk->tsn_gap_acked) {
1393 				tchunk->tsn_gap_acked = 1;
1394 				*highest_new_tsn_in_sack = tsn;
1395 				bytes_acked += sctp_data_size(tchunk);
1396 				if (!tchunk->transport)
1397 					migrate_bytes += sctp_data_size(tchunk);
1398 				forward_progress = true;
1399 			}
1400 
1401 			if (TSN_lte(tsn, sack_ctsn)) {
1402 				/* RFC 2960  6.3.2 Retransmission Timer Rules
1403 				 *
1404 				 * R3) Whenever a SACK is received
1405 				 * that acknowledges the DATA chunk
1406 				 * with the earliest outstanding TSN
1407 				 * for that address, restart T3-rtx
1408 				 * timer for that address with its
1409 				 * current RTO.
1410 				 */
1411 				restart_timer = 1;
1412 				forward_progress = true;
1413 
1414 				if (!tchunk->tsn_gap_acked) {
1415 					/*
1416 					 * SFR-CACC algorithm:
1417 					 * 2) If the SACK contains gap acks
1418 					 * and the flag CHANGEOVER_ACTIVE is
1419 					 * set the receiver of the SACK MUST
1420 					 * take the following action:
1421 					 *
1422 					 * B) For each TSN t being acked that
1423 					 * has not been acked in any SACK so
1424 					 * far, set cacc_saw_newack to 1 for
1425 					 * the destination that the TSN was
1426 					 * sent to.
1427 					 */
1428 					if (transport &&
1429 					    sack->num_gap_ack_blocks &&
1430 					    q->asoc->peer.primary_path->cacc.
1431 					    changeover_active)
1432 						transport->cacc.cacc_saw_newack
1433 							= 1;
1434 				}
1435 
1436 				list_add_tail(&tchunk->transmitted_list,
1437 					      &q->sacked);
1438 			} else {
1439 				/* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1440 				 * M2) Each time a SACK arrives reporting
1441 				 * 'Stray DATA chunk(s)' record the highest TSN
1442 				 * reported as newly acknowledged, call this
1443 				 * value 'HighestTSNinSack'. A newly
1444 				 * acknowledged DATA chunk is one not
1445 				 * previously acknowledged in a SACK.
1446 				 *
1447 				 * When the SCTP sender of data receives a SACK
1448 				 * chunk that acknowledges, for the first time,
1449 				 * the receipt of a DATA chunk, all the still
1450 				 * unacknowledged DATA chunks whose TSN is
1451 				 * older than that newly acknowledged DATA
1452 				 * chunk, are qualified as 'Stray DATA chunks'.
1453 				 */
1454 				list_add_tail(lchunk, &tlist);
1455 			}
1456 		} else {
1457 			if (tchunk->tsn_gap_acked) {
1458 				pr_debug("%s: receiver reneged on data TSN:0x%x\n",
1459 					 __func__, tsn);
1460 
1461 				tchunk->tsn_gap_acked = 0;
1462 
1463 				if (tchunk->transport)
1464 					bytes_acked -= sctp_data_size(tchunk);
1465 
1466 				/* RFC 2960 6.3.2 Retransmission Timer Rules
1467 				 *
1468 				 * R4) Whenever a SACK is received missing a
1469 				 * TSN that was previously acknowledged via a
1470 				 * Gap Ack Block, start T3-rtx for the
1471 				 * destination address to which the DATA
1472 				 * chunk was originally
1473 				 * transmitted if it is not already running.
1474 				 */
1475 				restart_timer = 1;
1476 			}
1477 
1478 			list_add_tail(lchunk, &tlist);
1479 		}
1480 	}
1481 
1482 	if (transport) {
1483 		if (bytes_acked) {
1484 			struct sctp_association *asoc = transport->asoc;
1485 
1486 			/* We may have counted DATA that was migrated
1487 			 * to this transport due to DEL-IP operation.
1488 			 * Subtract those bytes, since the were never
1489 			 * send on this transport and shouldn't be
1490 			 * credited to this transport.
1491 			 */
1492 			bytes_acked -= migrate_bytes;
1493 
1494 			/* 8.2. When an outstanding TSN is acknowledged,
1495 			 * the endpoint shall clear the error counter of
1496 			 * the destination transport address to which the
1497 			 * DATA chunk was last sent.
1498 			 * The association's overall error counter is
1499 			 * also cleared.
1500 			 */
1501 			transport->error_count = 0;
1502 			transport->asoc->overall_error_count = 0;
1503 			forward_progress = true;
1504 
1505 			/*
1506 			 * While in SHUTDOWN PENDING, we may have started
1507 			 * the T5 shutdown guard timer after reaching the
1508 			 * retransmission limit. Stop that timer as soon
1509 			 * as the receiver acknowledged any data.
1510 			 */
1511 			if (asoc->state == SCTP_STATE_SHUTDOWN_PENDING &&
1512 			    del_timer(&asoc->timers
1513 				[SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]))
1514 					sctp_association_put(asoc);
1515 
1516 			/* Mark the destination transport address as
1517 			 * active if it is not so marked.
1518 			 */
1519 			if ((transport->state == SCTP_INACTIVE ||
1520 			     transport->state == SCTP_UNCONFIRMED) &&
1521 			    sctp_cmp_addr_exact(&transport->ipaddr, saddr)) {
1522 				sctp_assoc_control_transport(
1523 					transport->asoc,
1524 					transport,
1525 					SCTP_TRANSPORT_UP,
1526 					SCTP_RECEIVED_SACK);
1527 			}
1528 
1529 			sctp_transport_raise_cwnd(transport, sack_ctsn,
1530 						  bytes_acked);
1531 
1532 			transport->flight_size -= bytes_acked;
1533 			if (transport->flight_size == 0)
1534 				transport->partial_bytes_acked = 0;
1535 			q->outstanding_bytes -= bytes_acked + migrate_bytes;
1536 		} else {
1537 			/* RFC 2960 6.1, sctpimpguide-06 2.15.2
1538 			 * When a sender is doing zero window probing, it
1539 			 * should not timeout the association if it continues
1540 			 * to receive new packets from the receiver. The
1541 			 * reason is that the receiver MAY keep its window
1542 			 * closed for an indefinite time.
1543 			 * A sender is doing zero window probing when the
1544 			 * receiver's advertised window is zero, and there is
1545 			 * only one data chunk in flight to the receiver.
1546 			 *
1547 			 * Allow the association to timeout while in SHUTDOWN
1548 			 * PENDING or SHUTDOWN RECEIVED in case the receiver
1549 			 * stays in zero window mode forever.
1550 			 */
1551 			if (!q->asoc->peer.rwnd &&
1552 			    !list_empty(&tlist) &&
1553 			    (sack_ctsn+2 == q->asoc->next_tsn) &&
1554 			    q->asoc->state < SCTP_STATE_SHUTDOWN_PENDING) {
1555 				pr_debug("%s: sack received for zero window "
1556 					 "probe:%u\n", __func__, sack_ctsn);
1557 
1558 				q->asoc->overall_error_count = 0;
1559 				transport->error_count = 0;
1560 			}
1561 		}
1562 
1563 		/* RFC 2960 6.3.2 Retransmission Timer Rules
1564 		 *
1565 		 * R2) Whenever all outstanding data sent to an address have
1566 		 * been acknowledged, turn off the T3-rtx timer of that
1567 		 * address.
1568 		 */
1569 		if (!transport->flight_size) {
1570 			if (del_timer(&transport->T3_rtx_timer))
1571 				sctp_transport_put(transport);
1572 		} else if (restart_timer) {
1573 			if (!mod_timer(&transport->T3_rtx_timer,
1574 				       jiffies + transport->rto))
1575 				sctp_transport_hold(transport);
1576 		}
1577 
1578 		if (forward_progress) {
1579 			if (transport->dst)
1580 				dst_confirm(transport->dst);
1581 		}
1582 	}
1583 
1584 	list_splice(&tlist, transmitted_queue);
1585 }
1586 
1587 /* Mark chunks as missing and consequently may get retransmitted. */
1588 static void sctp_mark_missing(struct sctp_outq *q,
1589 			      struct list_head *transmitted_queue,
1590 			      struct sctp_transport *transport,
1591 			      __u32 highest_new_tsn_in_sack,
1592 			      int count_of_newacks)
1593 {
1594 	struct sctp_chunk *chunk;
1595 	__u32 tsn;
1596 	char do_fast_retransmit = 0;
1597 	struct sctp_association *asoc = q->asoc;
1598 	struct sctp_transport *primary = asoc->peer.primary_path;
1599 
1600 	list_for_each_entry(chunk, transmitted_queue, transmitted_list) {
1601 
1602 		tsn = ntohl(chunk->subh.data_hdr->tsn);
1603 
1604 		/* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1605 		 * 'Unacknowledged TSN's', if the TSN number of an
1606 		 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1607 		 * value, increment the 'TSN.Missing.Report' count on that
1608 		 * chunk if it has NOT been fast retransmitted or marked for
1609 		 * fast retransmit already.
1610 		 */
1611 		if (chunk->fast_retransmit == SCTP_CAN_FRTX &&
1612 		    !chunk->tsn_gap_acked &&
1613 		    TSN_lt(tsn, highest_new_tsn_in_sack)) {
1614 
1615 			/* SFR-CACC may require us to skip marking
1616 			 * this chunk as missing.
1617 			 */
1618 			if (!transport || !sctp_cacc_skip(primary,
1619 						chunk->transport,
1620 						count_of_newacks, tsn)) {
1621 				chunk->tsn_missing_report++;
1622 
1623 				pr_debug("%s: tsn:0x%x missing counter:%d\n",
1624 					 __func__, tsn, chunk->tsn_missing_report);
1625 			}
1626 		}
1627 		/*
1628 		 * M4) If any DATA chunk is found to have a
1629 		 * 'TSN.Missing.Report'
1630 		 * value larger than or equal to 3, mark that chunk for
1631 		 * retransmission and start the fast retransmit procedure.
1632 		 */
1633 
1634 		if (chunk->tsn_missing_report >= 3) {
1635 			chunk->fast_retransmit = SCTP_NEED_FRTX;
1636 			do_fast_retransmit = 1;
1637 		}
1638 	}
1639 
1640 	if (transport) {
1641 		if (do_fast_retransmit)
1642 			sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1643 
1644 		pr_debug("%s: transport:%p, cwnd:%d, ssthresh:%d, "
1645 			 "flight_size:%d, pba:%d\n",  __func__, transport,
1646 			 transport->cwnd, transport->ssthresh,
1647 			 transport->flight_size, transport->partial_bytes_acked);
1648 	}
1649 }
1650 
1651 /* Is the given TSN acked by this packet?  */
1652 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1653 {
1654 	int i;
1655 	sctp_sack_variable_t *frags;
1656 	__u16 gap;
1657 	__u32 ctsn = ntohl(sack->cum_tsn_ack);
1658 
1659 	if (TSN_lte(tsn, ctsn))
1660 		goto pass;
1661 
1662 	/* 3.3.4 Selective Acknowledgement (SACK) (3):
1663 	 *
1664 	 * Gap Ack Blocks:
1665 	 *  These fields contain the Gap Ack Blocks. They are repeated
1666 	 *  for each Gap Ack Block up to the number of Gap Ack Blocks
1667 	 *  defined in the Number of Gap Ack Blocks field. All DATA
1668 	 *  chunks with TSNs greater than or equal to (Cumulative TSN
1669 	 *  Ack + Gap Ack Block Start) and less than or equal to
1670 	 *  (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1671 	 *  Block are assumed to have been received correctly.
1672 	 */
1673 
1674 	frags = sack->variable;
1675 	gap = tsn - ctsn;
1676 	for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) {
1677 		if (TSN_lte(ntohs(frags[i].gab.start), gap) &&
1678 		    TSN_lte(gap, ntohs(frags[i].gab.end)))
1679 			goto pass;
1680 	}
1681 
1682 	return 0;
1683 pass:
1684 	return 1;
1685 }
1686 
1687 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1688 				    int nskips, __be16 stream)
1689 {
1690 	int i;
1691 
1692 	for (i = 0; i < nskips; i++) {
1693 		if (skiplist[i].stream == stream)
1694 			return i;
1695 	}
1696 	return i;
1697 }
1698 
1699 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1700 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1701 {
1702 	struct sctp_association *asoc = q->asoc;
1703 	struct sctp_chunk *ftsn_chunk = NULL;
1704 	struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1705 	int nskips = 0;
1706 	int skip_pos = 0;
1707 	__u32 tsn;
1708 	struct sctp_chunk *chunk;
1709 	struct list_head *lchunk, *temp;
1710 
1711 	if (!asoc->peer.prsctp_capable)
1712 		return;
1713 
1714 	/* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1715 	 * received SACK.
1716 	 *
1717 	 * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1718 	 * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1719 	 */
1720 	if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1721 		asoc->adv_peer_ack_point = ctsn;
1722 
1723 	/* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1724 	 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1725 	 * the chunk next in the out-queue space is marked as "abandoned" as
1726 	 * shown in the following example:
1727 	 *
1728 	 * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1729 	 * and the Advanced.Peer.Ack.Point is updated to this value:
1730 	 *
1731 	 *   out-queue at the end of  ==>   out-queue after Adv.Ack.Point
1732 	 *   normal SACK processing           local advancement
1733 	 *                ...                           ...
1734 	 *   Adv.Ack.Pt-> 102 acked                     102 acked
1735 	 *                103 abandoned                 103 abandoned
1736 	 *                104 abandoned     Adv.Ack.P-> 104 abandoned
1737 	 *                105                           105
1738 	 *                106 acked                     106 acked
1739 	 *                ...                           ...
1740 	 *
1741 	 * In this example, the data sender successfully advanced the
1742 	 * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1743 	 */
1744 	list_for_each_safe(lchunk, temp, &q->abandoned) {
1745 		chunk = list_entry(lchunk, struct sctp_chunk,
1746 					transmitted_list);
1747 		tsn = ntohl(chunk->subh.data_hdr->tsn);
1748 
1749 		/* Remove any chunks in the abandoned queue that are acked by
1750 		 * the ctsn.
1751 		 */
1752 		if (TSN_lte(tsn, ctsn)) {
1753 			list_del_init(lchunk);
1754 			sctp_chunk_free(chunk);
1755 		} else {
1756 			if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1757 				asoc->adv_peer_ack_point = tsn;
1758 				if (chunk->chunk_hdr->flags &
1759 					 SCTP_DATA_UNORDERED)
1760 					continue;
1761 				skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1762 						nskips,
1763 						chunk->subh.data_hdr->stream);
1764 				ftsn_skip_arr[skip_pos].stream =
1765 					chunk->subh.data_hdr->stream;
1766 				ftsn_skip_arr[skip_pos].ssn =
1767 					 chunk->subh.data_hdr->ssn;
1768 				if (skip_pos == nskips)
1769 					nskips++;
1770 				if (nskips == 10)
1771 					break;
1772 			} else
1773 				break;
1774 		}
1775 	}
1776 
1777 	/* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1778 	 * is greater than the Cumulative TSN ACK carried in the received
1779 	 * SACK, the data sender MUST send the data receiver a FORWARD TSN
1780 	 * chunk containing the latest value of the
1781 	 * "Advanced.Peer.Ack.Point".
1782 	 *
1783 	 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1784 	 * list each stream and sequence number in the forwarded TSN. This
1785 	 * information will enable the receiver to easily find any
1786 	 * stranded TSN's waiting on stream reorder queues. Each stream
1787 	 * SHOULD only be reported once; this means that if multiple
1788 	 * abandoned messages occur in the same stream then only the
1789 	 * highest abandoned stream sequence number is reported. If the
1790 	 * total size of the FORWARD TSN does NOT fit in a single MTU then
1791 	 * the sender of the FORWARD TSN SHOULD lower the
1792 	 * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1793 	 * single MTU.
1794 	 */
1795 	if (asoc->adv_peer_ack_point > ctsn)
1796 		ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1797 					      nskips, &ftsn_skip_arr[0]);
1798 
1799 	if (ftsn_chunk) {
1800 		list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1801 		SCTP_INC_STATS(sock_net(asoc->base.sk), SCTP_MIB_OUTCTRLCHUNKS);
1802 	}
1803 }
1804