xref: /linux/net/sctp/outqueue.c (revision f30828a6745281edda735f642b5f814e1123ecd3)
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 <lksctp-developers@lists.sourceforge.net>
32  *
33  * Or submit a bug report through the following website:
34  *    http://www.sf.net/projects/lksctp
35  *
36  * Written or modified by:
37  *    La Monte H.P. Yarroll <piggy@acm.org>
38  *    Karl Knutson          <karl@athena.chicago.il.us>
39  *    Perry Melange         <pmelange@null.cc.uic.edu>
40  *    Xingang Guo           <xingang.guo@intel.com>
41  *    Hui Huang 	    <hui.huang@nokia.com>
42  *    Sridhar Samudrala     <sri@us.ibm.com>
43  *    Jon Grimm             <jgrimm@us.ibm.com>
44  *
45  * Any bugs reported given to us we will try to fix... any fixes shared will
46  * be incorporated into the next SCTP release.
47  */
48 
49 #include <linux/types.h>
50 #include <linux/list.h>   /* For struct list_head */
51 #include <linux/socket.h>
52 #include <linux/ip.h>
53 #include <net/sock.h>	  /* For skb_set_owner_w */
54 
55 #include <net/sctp/sctp.h>
56 #include <net/sctp/sm.h>
57 
58 /* Declare internal functions here.  */
59 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn);
60 static void sctp_check_transmitted(struct sctp_outq *q,
61 				   struct list_head *transmitted_queue,
62 				   struct sctp_transport *transport,
63 				   struct sctp_sackhdr *sack,
64 				   __u32 highest_new_tsn);
65 
66 static void sctp_mark_missing(struct sctp_outq *q,
67 			      struct list_head *transmitted_queue,
68 			      struct sctp_transport *transport,
69 			      __u32 highest_new_tsn,
70 			      int count_of_newacks);
71 
72 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 sack_ctsn);
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 	return;
81 }
82 
83 /* Take data from the front of the queue. */
84 static inline struct sctp_chunk *sctp_outq_dequeue_data(struct sctp_outq *q)
85 {
86 	struct sctp_chunk *ch = NULL;
87 
88 	if (!list_empty(&q->out_chunk_list)) {
89 		struct list_head *entry = q->out_chunk_list.next;
90 
91 		ch = list_entry(entry, struct sctp_chunk, list);
92 		list_del_init(entry);
93 		q->out_qlen -= ch->skb->len;
94 	}
95 	return ch;
96 }
97 /* Add data chunk to the end of the queue. */
98 static inline void sctp_outq_tail_data(struct sctp_outq *q,
99 				       struct sctp_chunk *ch)
100 {
101 	list_add_tail(&ch->list, &q->out_chunk_list);
102 	q->out_qlen += ch->skb->len;
103 	return;
104 }
105 
106 /*
107  * SFR-CACC algorithm:
108  * D) If count_of_newacks is greater than or equal to 2
109  * and t was not sent to the current primary then the
110  * sender MUST NOT increment missing report count for t.
111  */
112 static inline int sctp_cacc_skip_3_1_d(struct sctp_transport *primary,
113 				       struct sctp_transport *transport,
114 				       int count_of_newacks)
115 {
116 	if (count_of_newacks >=2 && transport != primary)
117 		return 1;
118 	return 0;
119 }
120 
121 /*
122  * SFR-CACC algorithm:
123  * F) If count_of_newacks is less than 2, let d be the
124  * destination to which t was sent. If cacc_saw_newack
125  * is 0 for destination d, then the sender MUST NOT
126  * increment missing report count for t.
127  */
128 static inline int sctp_cacc_skip_3_1_f(struct sctp_transport *transport,
129 				       int count_of_newacks)
130 {
131 	if (count_of_newacks < 2 && !transport->cacc.cacc_saw_newack)
132 		return 1;
133 	return 0;
134 }
135 
136 /*
137  * SFR-CACC algorithm:
138  * 3.1) If CYCLING_CHANGEOVER is 0, the sender SHOULD
139  * execute steps C, D, F.
140  *
141  * C has been implemented in sctp_outq_sack
142  */
143 static inline int sctp_cacc_skip_3_1(struct sctp_transport *primary,
144 				     struct sctp_transport *transport,
145 				     int count_of_newacks)
146 {
147 	if (!primary->cacc.cycling_changeover) {
148 		if (sctp_cacc_skip_3_1_d(primary, transport, count_of_newacks))
149 			return 1;
150 		if (sctp_cacc_skip_3_1_f(transport, count_of_newacks))
151 			return 1;
152 		return 0;
153 	}
154 	return 0;
155 }
156 
157 /*
158  * SFR-CACC algorithm:
159  * 3.2) Else if CYCLING_CHANGEOVER is 1, and t is less
160  * than next_tsn_at_change of the current primary, then
161  * the sender MUST NOT increment missing report count
162  * for t.
163  */
164 static inline int sctp_cacc_skip_3_2(struct sctp_transport *primary, __u32 tsn)
165 {
166 	if (primary->cacc.cycling_changeover &&
167 	    TSN_lt(tsn, primary->cacc.next_tsn_at_change))
168 		return 1;
169 	return 0;
170 }
171 
172 /*
173  * SFR-CACC algorithm:
174  * 3) If the missing report count for TSN t is to be
175  * incremented according to [RFC2960] and
176  * [SCTP_STEWART-2002], and CHANGEOVER_ACTIVE is set,
177  * then the sender MUST futher execute steps 3.1 and
178  * 3.2 to determine if the missing report count for
179  * TSN t SHOULD NOT be incremented.
180  *
181  * 3.3) If 3.1 and 3.2 do not dictate that the missing
182  * report count for t should not be incremented, then
183  * the sender SOULD increment missing report count for
184  * t (according to [RFC2960] and [SCTP_STEWART_2002]).
185  */
186 static inline int sctp_cacc_skip(struct sctp_transport *primary,
187 				 struct sctp_transport *transport,
188 				 int count_of_newacks,
189 				 __u32 tsn)
190 {
191 	if (primary->cacc.changeover_active &&
192 	    (sctp_cacc_skip_3_1(primary, transport, count_of_newacks)
193 	     || sctp_cacc_skip_3_2(primary, tsn)))
194 		return 1;
195 	return 0;
196 }
197 
198 /* Initialize an existing sctp_outq.  This does the boring stuff.
199  * You still need to define handlers if you really want to DO
200  * something with this structure...
201  */
202 void sctp_outq_init(struct sctp_association *asoc, struct sctp_outq *q)
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 	q->fast_rtx = 0;
212 	q->outstanding_bytes = 0;
213 	q->empty = 1;
214 	q->cork  = 0;
215 
216 	q->malloced = 0;
217 	q->out_qlen = 0;
218 }
219 
220 /* Free the outqueue structure and any related pending chunks.
221  */
222 void sctp_outq_teardown(struct sctp_outq *q)
223 {
224 	struct sctp_transport *transport;
225 	struct list_head *lchunk, *temp;
226 	struct sctp_chunk *chunk, *tmp;
227 
228 	/* Throw away unacknowledged chunks. */
229 	list_for_each_entry(transport, &q->asoc->peer.transport_addr_list,
230 			transports) {
231 		while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) {
232 			chunk = list_entry(lchunk, struct sctp_chunk,
233 					   transmitted_list);
234 			/* Mark as part of a failed message. */
235 			sctp_chunk_fail(chunk, q->error);
236 			sctp_chunk_free(chunk);
237 		}
238 	}
239 
240 	/* Throw away chunks that have been gap ACKed.  */
241 	list_for_each_safe(lchunk, temp, &q->sacked) {
242 		list_del_init(lchunk);
243 		chunk = list_entry(lchunk, struct sctp_chunk,
244 				   transmitted_list);
245 		sctp_chunk_fail(chunk, q->error);
246 		sctp_chunk_free(chunk);
247 	}
248 
249 	/* Throw away any chunks in the retransmit queue. */
250 	list_for_each_safe(lchunk, temp, &q->retransmit) {
251 		list_del_init(lchunk);
252 		chunk = list_entry(lchunk, struct sctp_chunk,
253 				   transmitted_list);
254 		sctp_chunk_fail(chunk, q->error);
255 		sctp_chunk_free(chunk);
256 	}
257 
258 	/* Throw away any chunks that are in the abandoned queue. */
259 	list_for_each_safe(lchunk, temp, &q->abandoned) {
260 		list_del_init(lchunk);
261 		chunk = list_entry(lchunk, struct sctp_chunk,
262 				   transmitted_list);
263 		sctp_chunk_fail(chunk, q->error);
264 		sctp_chunk_free(chunk);
265 	}
266 
267 	/* Throw away any leftover data chunks. */
268 	while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
269 
270 		/* Mark as send failure. */
271 		sctp_chunk_fail(chunk, q->error);
272 		sctp_chunk_free(chunk);
273 	}
274 
275 	q->error = 0;
276 
277 	/* Throw away any leftover control chunks. */
278 	list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
279 		list_del_init(&chunk->list);
280 		sctp_chunk_free(chunk);
281 	}
282 }
283 
284 /* Free the outqueue structure and any related pending chunks.  */
285 void sctp_outq_free(struct sctp_outq *q)
286 {
287 	/* Throw away leftover chunks. */
288 	sctp_outq_teardown(q);
289 
290 	/* If we were kmalloc()'d, free the memory.  */
291 	if (q->malloced)
292 		kfree(q);
293 }
294 
295 /* Put a new chunk in an sctp_outq.  */
296 int sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk)
297 {
298 	int error = 0;
299 
300 	SCTP_DEBUG_PRINTK("sctp_outq_tail(%p, %p[%s])\n",
301 			  q, chunk, chunk && chunk->chunk_hdr ?
302 			  sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
303 			  : "Illegal Chunk");
304 
305 	/* If it is data, queue it up, otherwise, send it
306 	 * immediately.
307 	 */
308 	if (SCTP_CID_DATA == chunk->chunk_hdr->type) {
309 		/* Is it OK to queue data chunks?  */
310 		/* From 9. Termination of Association
311 		 *
312 		 * When either endpoint performs a shutdown, the
313 		 * association on each peer will stop accepting new
314 		 * data from its user and only deliver data in queue
315 		 * at the time of sending or receiving the SHUTDOWN
316 		 * chunk.
317 		 */
318 		switch (q->asoc->state) {
319 		case SCTP_STATE_EMPTY:
320 		case SCTP_STATE_CLOSED:
321 		case SCTP_STATE_SHUTDOWN_PENDING:
322 		case SCTP_STATE_SHUTDOWN_SENT:
323 		case SCTP_STATE_SHUTDOWN_RECEIVED:
324 		case SCTP_STATE_SHUTDOWN_ACK_SENT:
325 			/* Cannot send after transport endpoint shutdown */
326 			error = -ESHUTDOWN;
327 			break;
328 
329 		default:
330 			SCTP_DEBUG_PRINTK("outqueueing (%p, %p[%s])\n",
331 			  q, chunk, chunk && chunk->chunk_hdr ?
332 			  sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
333 			  : "Illegal Chunk");
334 
335 			sctp_outq_tail_data(q, chunk);
336 			if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
337 				SCTP_INC_STATS(SCTP_MIB_OUTUNORDERCHUNKS);
338 			else
339 				SCTP_INC_STATS(SCTP_MIB_OUTORDERCHUNKS);
340 			q->empty = 0;
341 			break;
342 		}
343 	} else {
344 		list_add_tail(&chunk->list, &q->control_chunk_list);
345 		SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS);
346 	}
347 
348 	if (error < 0)
349 		return error;
350 
351 	if (!q->cork)
352 		error = sctp_outq_flush(q, 0);
353 
354 	return error;
355 }
356 
357 /* Insert a chunk into the sorted list based on the TSNs.  The retransmit list
358  * and the abandoned list are in ascending order.
359  */
360 static void sctp_insert_list(struct list_head *head, struct list_head *new)
361 {
362 	struct list_head *pos;
363 	struct sctp_chunk *nchunk, *lchunk;
364 	__u32 ntsn, ltsn;
365 	int done = 0;
366 
367 	nchunk = list_entry(new, struct sctp_chunk, transmitted_list);
368 	ntsn = ntohl(nchunk->subh.data_hdr->tsn);
369 
370 	list_for_each(pos, head) {
371 		lchunk = list_entry(pos, struct sctp_chunk, transmitted_list);
372 		ltsn = ntohl(lchunk->subh.data_hdr->tsn);
373 		if (TSN_lt(ntsn, ltsn)) {
374 			list_add(new, pos->prev);
375 			done = 1;
376 			break;
377 		}
378 	}
379 	if (!done)
380 		list_add_tail(new, head);
381 }
382 
383 /* Mark all the eligible packets on a transport for retransmission.  */
384 void sctp_retransmit_mark(struct sctp_outq *q,
385 			  struct sctp_transport *transport,
386 			  __u8 reason)
387 {
388 	struct list_head *lchunk, *ltemp;
389 	struct sctp_chunk *chunk;
390 
391 	/* Walk through the specified transmitted queue.  */
392 	list_for_each_safe(lchunk, ltemp, &transport->transmitted) {
393 		chunk = list_entry(lchunk, struct sctp_chunk,
394 				   transmitted_list);
395 
396 		/* If the chunk is abandoned, move it to abandoned list. */
397 		if (sctp_chunk_abandoned(chunk)) {
398 			list_del_init(lchunk);
399 			sctp_insert_list(&q->abandoned, lchunk);
400 
401 			/* If this chunk has not been previousely acked,
402 			 * stop considering it 'outstanding'.  Our peer
403 			 * will most likely never see it since it will
404 			 * not be retransmitted
405 			 */
406 			if (!chunk->tsn_gap_acked) {
407 				chunk->transport->flight_size -=
408 						sctp_data_size(chunk);
409 				q->outstanding_bytes -= sctp_data_size(chunk);
410 				q->asoc->peer.rwnd += (sctp_data_size(chunk) +
411 							sizeof(struct sk_buff));
412 			}
413 			continue;
414 		}
415 
416 		/* If we are doing  retransmission due to a timeout or pmtu
417 		 * discovery, only the  chunks that are not yet acked should
418 		 * be added to the retransmit queue.
419 		 */
420 		if ((reason == SCTP_RTXR_FAST_RTX  &&
421 			    (chunk->fast_retransmit > 0)) ||
422 		    (reason != SCTP_RTXR_FAST_RTX  && !chunk->tsn_gap_acked)) {
423 			/* If this chunk was sent less then 1 rto ago, do not
424 			 * retransmit this chunk, but give the peer time
425 			 * to acknowlege it.  Do this only when
426 			 * retransmitting due to T3 timeout.
427 			 */
428 			if (reason == SCTP_RTXR_T3_RTX &&
429 			    (jiffies - chunk->sent_at) < transport->last_rto)
430 				continue;
431 
432 			/* RFC 2960 6.2.1 Processing a Received SACK
433 			 *
434 			 * C) Any time a DATA chunk is marked for
435 			 * retransmission (via either T3-rtx timer expiration
436 			 * (Section 6.3.3) or via fast retransmit
437 			 * (Section 7.2.4)), add the data size of those
438 			 * chunks to the rwnd.
439 			 */
440 			q->asoc->peer.rwnd += (sctp_data_size(chunk) +
441 						sizeof(struct sk_buff));
442 			q->outstanding_bytes -= sctp_data_size(chunk);
443 			transport->flight_size -= sctp_data_size(chunk);
444 
445 			/* sctpimpguide-05 Section 2.8.2
446 			 * M5) If a T3-rtx timer expires, the
447 			 * 'TSN.Missing.Report' of all affected TSNs is set
448 			 * to 0.
449 			 */
450 			chunk->tsn_missing_report = 0;
451 
452 			/* If a chunk that is being used for RTT measurement
453 			 * has to be retransmitted, we cannot use this chunk
454 			 * anymore for RTT measurements. Reset rto_pending so
455 			 * that a new RTT measurement is started when a new
456 			 * data chunk is sent.
457 			 */
458 			if (chunk->rtt_in_progress) {
459 				chunk->rtt_in_progress = 0;
460 				transport->rto_pending = 0;
461 			}
462 
463 			/* Move the chunk to the retransmit queue. The chunks
464 			 * on the retransmit queue are always kept in order.
465 			 */
466 			list_del_init(lchunk);
467 			sctp_insert_list(&q->retransmit, lchunk);
468 		}
469 	}
470 
471 	SCTP_DEBUG_PRINTK("%s: transport: %p, reason: %d, "
472 			  "cwnd: %d, ssthresh: %d, flight_size: %d, "
473 			  "pba: %d\n", __func__,
474 			  transport, reason,
475 			  transport->cwnd, transport->ssthresh,
476 			  transport->flight_size,
477 			  transport->partial_bytes_acked);
478 
479 }
480 
481 /* Mark all the eligible packets on a transport for retransmission and force
482  * one packet out.
483  */
484 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
485 		     sctp_retransmit_reason_t reason)
486 {
487 	int error = 0;
488 
489 	switch(reason) {
490 	case SCTP_RTXR_T3_RTX:
491 		SCTP_INC_STATS(SCTP_MIB_T3_RETRANSMITS);
492 		sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
493 		/* Update the retran path if the T3-rtx timer has expired for
494 		 * the current retran path.
495 		 */
496 		if (transport == transport->asoc->peer.retran_path)
497 			sctp_assoc_update_retran_path(transport->asoc);
498 		transport->asoc->rtx_data_chunks +=
499 			transport->asoc->unack_data;
500 		break;
501 	case SCTP_RTXR_FAST_RTX:
502 		SCTP_INC_STATS(SCTP_MIB_FAST_RETRANSMITS);
503 		sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
504 		q->fast_rtx = 1;
505 		break;
506 	case SCTP_RTXR_PMTUD:
507 		SCTP_INC_STATS(SCTP_MIB_PMTUD_RETRANSMITS);
508 		break;
509 	case SCTP_RTXR_T1_RTX:
510 		SCTP_INC_STATS(SCTP_MIB_T1_RETRANSMITS);
511 		transport->asoc->init_retries++;
512 		break;
513 	default:
514 		BUG();
515 	}
516 
517 	sctp_retransmit_mark(q, transport, reason);
518 
519 	/* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
520 	 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
521 	 * following the procedures outlined in C1 - C5.
522 	 */
523 	if (reason == SCTP_RTXR_T3_RTX)
524 		sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point);
525 
526 	/* Flush the queues only on timeout, since fast_rtx is only
527 	 * triggered during sack processing and the queue
528 	 * will be flushed at the end.
529 	 */
530 	if (reason != SCTP_RTXR_FAST_RTX)
531 		error = sctp_outq_flush(q, /* rtx_timeout */ 1);
532 
533 	if (error)
534 		q->asoc->base.sk->sk_err = -error;
535 }
536 
537 /*
538  * Transmit DATA chunks on the retransmit queue.  Upon return from
539  * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
540  * need to be transmitted by the caller.
541  * We assume that pkt->transport has already been set.
542  *
543  * The return value is a normal kernel error return value.
544  */
545 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
546 			       int rtx_timeout, int *start_timer)
547 {
548 	struct list_head *lqueue;
549 	struct sctp_transport *transport = pkt->transport;
550 	sctp_xmit_t status;
551 	struct sctp_chunk *chunk, *chunk1;
552 	struct sctp_association *asoc;
553 	int fast_rtx;
554 	int error = 0;
555 	int timer = 0;
556 	int done = 0;
557 
558 	asoc = q->asoc;
559 	lqueue = &q->retransmit;
560 	fast_rtx = q->fast_rtx;
561 
562 	/* This loop handles time-out retransmissions, fast retransmissions,
563 	 * and retransmissions due to opening of whindow.
564 	 *
565 	 * RFC 2960 6.3.3 Handle T3-rtx Expiration
566 	 *
567 	 * E3) Determine how many of the earliest (i.e., lowest TSN)
568 	 * outstanding DATA chunks for the address for which the
569 	 * T3-rtx has expired will fit into a single packet, subject
570 	 * to the MTU constraint for the path corresponding to the
571 	 * destination transport address to which the retransmission
572 	 * is being sent (this may be different from the address for
573 	 * which the timer expires [see Section 6.4]). Call this value
574 	 * K. Bundle and retransmit those K DATA chunks in a single
575 	 * packet to the destination endpoint.
576 	 *
577 	 * [Just to be painfully clear, if we are retransmitting
578 	 * because a timeout just happened, we should send only ONE
579 	 * packet of retransmitted data.]
580 	 *
581 	 * For fast retransmissions we also send only ONE packet.  However,
582 	 * if we are just flushing the queue due to open window, we'll
583 	 * try to send as much as possible.
584 	 */
585 	list_for_each_entry_safe(chunk, chunk1, lqueue, transmitted_list) {
586 
587 		/* Make sure that Gap Acked TSNs are not retransmitted.  A
588 		 * simple approach is just to move such TSNs out of the
589 		 * way and into a 'transmitted' queue and skip to the
590 		 * next chunk.
591 		 */
592 		if (chunk->tsn_gap_acked) {
593 			list_del(&chunk->transmitted_list);
594 			list_add_tail(&chunk->transmitted_list,
595 					&transport->transmitted);
596 			continue;
597 		}
598 
599 		/* If we are doing fast retransmit, ignore non-fast_rtransmit
600 		 * chunks
601 		 */
602 		if (fast_rtx && !chunk->fast_retransmit)
603 			continue;
604 
605 		/* Attempt to append this chunk to the packet. */
606 		status = sctp_packet_append_chunk(pkt, chunk);
607 
608 		switch (status) {
609 		case SCTP_XMIT_PMTU_FULL:
610 			/* Send this packet.  */
611 			error = sctp_packet_transmit(pkt);
612 
613 			/* If we are retransmitting, we should only
614 			 * send a single packet.
615 			 */
616 			if (rtx_timeout || fast_rtx)
617 				done = 1;
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_NAGLE_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_del(&chunk->transmitted_list);
645 			list_add_tail(&chunk->transmitted_list,
646 					&transport->transmitted);
647 
648 			/* Mark the chunk as ineligible for fast retransmit
649 			 * after it is retransmitted.
650 			 */
651 			if (chunk->fast_retransmit > 0)
652 				chunk->fast_retransmit = -1;
653 
654 			/* Force start T3-rtx timer when fast retransmitting
655 			 * the earliest outstanding TSN
656 			 */
657 			if (!timer && fast_rtx &&
658 			    ntohl(chunk->subh.data_hdr->tsn) ==
659 					     asoc->ctsn_ack_point + 1)
660 				timer = 2;
661 
662 			q->empty = 0;
663 			break;
664 		}
665 
666 		/* Set the timer if there were no errors */
667 		if (!error && !timer)
668 			timer = 1;
669 
670 		if (done)
671 			break;
672 	}
673 
674 	/* If we are here due to a retransmit timeout or a fast
675 	 * retransmit and if there are any chunks left in the retransmit
676 	 * queue that could not fit in the PMTU sized packet, they need
677 	 * to be marked as ineligible for a subsequent fast retransmit.
678 	 */
679 	if (rtx_timeout || fast_rtx) {
680 		list_for_each_entry(chunk1, lqueue, transmitted_list) {
681 			if (chunk1->fast_retransmit > 0)
682 				chunk1->fast_retransmit = -1;
683 		}
684 	}
685 
686 	*start_timer = timer;
687 
688 	/* Clear fast retransmit hint */
689 	if (fast_rtx)
690 		q->fast_rtx = 0;
691 
692 	return error;
693 }
694 
695 /* Cork the outqueue so queued chunks are really queued. */
696 int sctp_outq_uncork(struct sctp_outq *q)
697 {
698 	int error = 0;
699 	if (q->cork)
700 		q->cork = 0;
701 	error = sctp_outq_flush(q, 0);
702 	return error;
703 }
704 
705 /*
706  * Try to flush an outqueue.
707  *
708  * Description: Send everything in q which we legally can, subject to
709  * congestion limitations.
710  * * Note: This function can be called from multiple contexts so appropriate
711  * locking concerns must be made.  Today we use the sock lock to protect
712  * this function.
713  */
714 int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout)
715 {
716 	struct sctp_packet *packet;
717 	struct sctp_packet singleton;
718 	struct sctp_association *asoc = q->asoc;
719 	__u16 sport = asoc->base.bind_addr.port;
720 	__u16 dport = asoc->peer.port;
721 	__u32 vtag = asoc->peer.i.init_tag;
722 	struct sctp_transport *transport = NULL;
723 	struct sctp_transport *new_transport;
724 	struct sctp_chunk *chunk, *tmp;
725 	sctp_xmit_t status;
726 	int error = 0;
727 	int start_timer = 0;
728 
729 	/* These transports have chunks to send. */
730 	struct list_head transport_list;
731 	struct list_head *ltransport;
732 
733 	INIT_LIST_HEAD(&transport_list);
734 	packet = NULL;
735 
736 	/*
737 	 * 6.10 Bundling
738 	 *   ...
739 	 *   When bundling control chunks with DATA chunks, an
740 	 *   endpoint MUST place control chunks first in the outbound
741 	 *   SCTP packet.  The transmitter MUST transmit DATA chunks
742 	 *   within a SCTP packet in increasing order of TSN.
743 	 *   ...
744 	 */
745 
746 	list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
747 		list_del_init(&chunk->list);
748 
749 		/* Pick the right transport to use. */
750 		new_transport = chunk->transport;
751 
752 		if (!new_transport) {
753 			/*
754 			 * If we have a prior transport pointer, see if
755 			 * the destination address of the chunk
756 			 * matches the destination address of the
757 			 * current transport.  If not a match, then
758 			 * try to look up the transport with a given
759 			 * destination address.  We do this because
760 			 * after processing ASCONFs, we may have new
761 			 * transports created.
762 			 */
763 			if (transport &&
764 			    sctp_cmp_addr_exact(&chunk->dest,
765 						&transport->ipaddr))
766 					new_transport = transport;
767 			else
768 				new_transport = sctp_assoc_lookup_paddr(asoc,
769 								&chunk->dest);
770 
771 			/* if we still don't have a new transport, then
772 			 * use the current active path.
773 			 */
774 			if (!new_transport)
775 				new_transport = asoc->peer.active_path;
776 		} else if ((new_transport->state == SCTP_INACTIVE) ||
777 			   (new_transport->state == SCTP_UNCONFIRMED)) {
778 			/* If the chunk is Heartbeat or Heartbeat Ack,
779 			 * send it to chunk->transport, even if it's
780 			 * inactive.
781 			 *
782 			 * 3.3.6 Heartbeat Acknowledgement:
783 			 * ...
784 			 * A HEARTBEAT ACK is always sent to the source IP
785 			 * address of the IP datagram containing the
786 			 * HEARTBEAT chunk to which this ack is responding.
787 			 * ...
788 			 *
789 			 * ASCONF_ACKs also must be sent to the source.
790 			 */
791 			if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT &&
792 			    chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK &&
793 			    chunk->chunk_hdr->type != SCTP_CID_ASCONF_ACK)
794 				new_transport = asoc->peer.active_path;
795 		}
796 
797 		/* Are we switching transports?
798 		 * Take care of transport locks.
799 		 */
800 		if (new_transport != transport) {
801 			transport = new_transport;
802 			if (list_empty(&transport->send_ready)) {
803 				list_add_tail(&transport->send_ready,
804 					      &transport_list);
805 			}
806 			packet = &transport->packet;
807 			sctp_packet_config(packet, vtag,
808 					   asoc->peer.ecn_capable);
809 		}
810 
811 		switch (chunk->chunk_hdr->type) {
812 		/*
813 		 * 6.10 Bundling
814 		 *   ...
815 		 *   An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
816 		 *   COMPLETE with any other chunks.  [Send them immediately.]
817 		 */
818 		case SCTP_CID_INIT:
819 		case SCTP_CID_INIT_ACK:
820 		case SCTP_CID_SHUTDOWN_COMPLETE:
821 			sctp_packet_init(&singleton, transport, sport, dport);
822 			sctp_packet_config(&singleton, vtag, 0);
823 			sctp_packet_append_chunk(&singleton, chunk);
824 			error = sctp_packet_transmit(&singleton);
825 			if (error < 0)
826 				return error;
827 			break;
828 
829 		case SCTP_CID_ABORT:
830 			if (sctp_test_T_bit(chunk)) {
831 				packet->vtag = asoc->c.my_vtag;
832 			}
833 		case SCTP_CID_SACK:
834 		case SCTP_CID_HEARTBEAT:
835 		case SCTP_CID_HEARTBEAT_ACK:
836 		case SCTP_CID_SHUTDOWN:
837 		case SCTP_CID_SHUTDOWN_ACK:
838 		case SCTP_CID_ERROR:
839 		case SCTP_CID_COOKIE_ECHO:
840 		case SCTP_CID_COOKIE_ACK:
841 		case SCTP_CID_ECN_ECNE:
842 		case SCTP_CID_ECN_CWR:
843 		case SCTP_CID_ASCONF:
844 		case SCTP_CID_ASCONF_ACK:
845 		case SCTP_CID_FWD_TSN:
846 			sctp_packet_transmit_chunk(packet, chunk);
847 			break;
848 
849 		default:
850 			/* We built a chunk with an illegal type! */
851 			BUG();
852 		}
853 	}
854 
855 	/* Is it OK to send data chunks?  */
856 	switch (asoc->state) {
857 	case SCTP_STATE_COOKIE_ECHOED:
858 		/* Only allow bundling when this packet has a COOKIE-ECHO
859 		 * chunk.
860 		 */
861 		if (!packet || !packet->has_cookie_echo)
862 			break;
863 
864 		/* fallthru */
865 	case SCTP_STATE_ESTABLISHED:
866 	case SCTP_STATE_SHUTDOWN_PENDING:
867 	case SCTP_STATE_SHUTDOWN_RECEIVED:
868 		/*
869 		 * RFC 2960 6.1  Transmission of DATA Chunks
870 		 *
871 		 * C) When the time comes for the sender to transmit,
872 		 * before sending new DATA chunks, the sender MUST
873 		 * first transmit any outstanding DATA chunks which
874 		 * are marked for retransmission (limited by the
875 		 * current cwnd).
876 		 */
877 		if (!list_empty(&q->retransmit)) {
878 			if (transport == asoc->peer.retran_path)
879 				goto retran;
880 
881 			/* Switch transports & prepare the packet.  */
882 
883 			transport = asoc->peer.retran_path;
884 
885 			if (list_empty(&transport->send_ready)) {
886 				list_add_tail(&transport->send_ready,
887 					      &transport_list);
888 			}
889 
890 			packet = &transport->packet;
891 			sctp_packet_config(packet, vtag,
892 					   asoc->peer.ecn_capable);
893 		retran:
894 			error = sctp_outq_flush_rtx(q, packet,
895 						    rtx_timeout, &start_timer);
896 
897 			if (start_timer)
898 				sctp_transport_reset_timers(transport,
899 							    start_timer-1);
900 
901 			/* This can happen on COOKIE-ECHO resend.  Only
902 			 * one chunk can get bundled with a COOKIE-ECHO.
903 			 */
904 			if (packet->has_cookie_echo)
905 				goto sctp_flush_out;
906 
907 			/* Don't send new data if there is still data
908 			 * waiting to retransmit.
909 			 */
910 			if (!list_empty(&q->retransmit))
911 				goto sctp_flush_out;
912 		}
913 
914 		/* Finally, transmit new packets.  */
915 		start_timer = 0;
916 		while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
917 			/* RFC 2960 6.5 Every DATA chunk MUST carry a valid
918 			 * stream identifier.
919 			 */
920 			if (chunk->sinfo.sinfo_stream >=
921 			    asoc->c.sinit_num_ostreams) {
922 
923 				/* Mark as failed send. */
924 				sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM);
925 				sctp_chunk_free(chunk);
926 				continue;
927 			}
928 
929 			/* Has this chunk expired? */
930 			if (sctp_chunk_abandoned(chunk)) {
931 				sctp_chunk_fail(chunk, 0);
932 				sctp_chunk_free(chunk);
933 				continue;
934 			}
935 
936 			/* If there is a specified transport, use it.
937 			 * Otherwise, we want to use the active path.
938 			 */
939 			new_transport = chunk->transport;
940 			if (!new_transport ||
941 			    ((new_transport->state == SCTP_INACTIVE) ||
942 			     (new_transport->state == SCTP_UNCONFIRMED)))
943 				new_transport = asoc->peer.active_path;
944 
945 			/* Change packets if necessary.  */
946 			if (new_transport != transport) {
947 				transport = new_transport;
948 
949 				/* Schedule to have this transport's
950 				 * packet flushed.
951 				 */
952 				if (list_empty(&transport->send_ready)) {
953 					list_add_tail(&transport->send_ready,
954 						      &transport_list);
955 				}
956 
957 				packet = &transport->packet;
958 				sctp_packet_config(packet, vtag,
959 						   asoc->peer.ecn_capable);
960 			}
961 
962 			SCTP_DEBUG_PRINTK("sctp_outq_flush(%p, %p[%s]), ",
963 					  q, chunk,
964 					  chunk && chunk->chunk_hdr ?
965 					  sctp_cname(SCTP_ST_CHUNK(
966 						  chunk->chunk_hdr->type))
967 					  : "Illegal Chunk");
968 
969 			SCTP_DEBUG_PRINTK("TX TSN 0x%x skb->head "
970 					"%p skb->users %d.\n",
971 					ntohl(chunk->subh.data_hdr->tsn),
972 					chunk->skb ?chunk->skb->head : NULL,
973 					chunk->skb ?
974 					atomic_read(&chunk->skb->users) : -1);
975 
976 			/* Add the chunk to the packet.  */
977 			status = sctp_packet_transmit_chunk(packet, chunk);
978 
979 			switch (status) {
980 			case SCTP_XMIT_PMTU_FULL:
981 			case SCTP_XMIT_RWND_FULL:
982 			case SCTP_XMIT_NAGLE_DELAY:
983 				/* We could not append this chunk, so put
984 				 * the chunk back on the output queue.
985 				 */
986 				SCTP_DEBUG_PRINTK("sctp_outq_flush: could "
987 					"not transmit TSN: 0x%x, status: %d\n",
988 					ntohl(chunk->subh.data_hdr->tsn),
989 					status);
990 				sctp_outq_head_data(q, chunk);
991 				goto sctp_flush_out;
992 				break;
993 
994 			case SCTP_XMIT_OK:
995 				break;
996 
997 			default:
998 				BUG();
999 			}
1000 
1001 			/* BUG: We assume that the sctp_packet_transmit()
1002 			 * call below will succeed all the time and add the
1003 			 * chunk to the transmitted list and restart the
1004 			 * timers.
1005 			 * It is possible that the call can fail under OOM
1006 			 * conditions.
1007 			 *
1008 			 * Is this really a problem?  Won't this behave
1009 			 * like a lost TSN?
1010 			 */
1011 			list_add_tail(&chunk->transmitted_list,
1012 				      &transport->transmitted);
1013 
1014 			sctp_transport_reset_timers(transport, start_timer-1);
1015 
1016 			q->empty = 0;
1017 
1018 			/* Only let one DATA chunk get bundled with a
1019 			 * COOKIE-ECHO chunk.
1020 			 */
1021 			if (packet->has_cookie_echo)
1022 				goto sctp_flush_out;
1023 		}
1024 		break;
1025 
1026 	default:
1027 		/* Do nothing.  */
1028 		break;
1029 	}
1030 
1031 sctp_flush_out:
1032 
1033 	/* Before returning, examine all the transports touched in
1034 	 * this call.  Right now, we bluntly force clear all the
1035 	 * transports.  Things might change after we implement Nagle.
1036 	 * But such an examination is still required.
1037 	 *
1038 	 * --xguo
1039 	 */
1040 	while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL ) {
1041 		struct sctp_transport *t = list_entry(ltransport,
1042 						      struct sctp_transport,
1043 						      send_ready);
1044 		packet = &t->packet;
1045 		if (!sctp_packet_empty(packet))
1046 			error = sctp_packet_transmit(packet);
1047 	}
1048 
1049 	return error;
1050 }
1051 
1052 /* Update unack_data based on the incoming SACK chunk */
1053 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
1054 					struct sctp_sackhdr *sack)
1055 {
1056 	sctp_sack_variable_t *frags;
1057 	__u16 unack_data;
1058 	int i;
1059 
1060 	unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
1061 
1062 	frags = sack->variable;
1063 	for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
1064 		unack_data -= ((ntohs(frags[i].gab.end) -
1065 				ntohs(frags[i].gab.start) + 1));
1066 	}
1067 
1068 	assoc->unack_data = unack_data;
1069 }
1070 
1071 /* Return the highest new tsn that is acknowledged by the given SACK chunk. */
1072 static __u32 sctp_highest_new_tsn(struct sctp_sackhdr *sack,
1073 				  struct sctp_association *asoc)
1074 {
1075 	struct sctp_transport *transport;
1076 	struct sctp_chunk *chunk;
1077 	__u32 highest_new_tsn, tsn;
1078 	struct list_head *transport_list = &asoc->peer.transport_addr_list;
1079 
1080 	highest_new_tsn = ntohl(sack->cum_tsn_ack);
1081 
1082 	list_for_each_entry(transport, transport_list, transports) {
1083 		list_for_each_entry(chunk, &transport->transmitted,
1084 				transmitted_list) {
1085 			tsn = ntohl(chunk->subh.data_hdr->tsn);
1086 
1087 			if (!chunk->tsn_gap_acked &&
1088 			    TSN_lt(highest_new_tsn, tsn) &&
1089 			    sctp_acked(sack, tsn))
1090 				highest_new_tsn = tsn;
1091 		}
1092 	}
1093 
1094 	return highest_new_tsn;
1095 }
1096 
1097 /* This is where we REALLY process a SACK.
1098  *
1099  * Process the SACK against the outqueue.  Mostly, this just frees
1100  * things off the transmitted queue.
1101  */
1102 int sctp_outq_sack(struct sctp_outq *q, struct sctp_sackhdr *sack)
1103 {
1104 	struct sctp_association *asoc = q->asoc;
1105 	struct sctp_transport *transport;
1106 	struct sctp_chunk *tchunk = NULL;
1107 	struct list_head *lchunk, *transport_list, *temp;
1108 	sctp_sack_variable_t *frags = sack->variable;
1109 	__u32 sack_ctsn, ctsn, tsn;
1110 	__u32 highest_tsn, highest_new_tsn;
1111 	__u32 sack_a_rwnd;
1112 	unsigned outstanding;
1113 	struct sctp_transport *primary = asoc->peer.primary_path;
1114 	int count_of_newacks = 0;
1115 
1116 	/* Grab the association's destination address list. */
1117 	transport_list = &asoc->peer.transport_addr_list;
1118 
1119 	sack_ctsn = ntohl(sack->cum_tsn_ack);
1120 
1121 	/*
1122 	 * SFR-CACC algorithm:
1123 	 * On receipt of a SACK the sender SHOULD execute the
1124 	 * following statements.
1125 	 *
1126 	 * 1) If the cumulative ack in the SACK passes next tsn_at_change
1127 	 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1128 	 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1129 	 * all destinations.
1130 	 */
1131 	if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1132 		primary->cacc.changeover_active = 0;
1133 		list_for_each_entry(transport, transport_list,
1134 				transports) {
1135 			transport->cacc.cycling_changeover = 0;
1136 		}
1137 	}
1138 
1139 	/*
1140 	 * SFR-CACC algorithm:
1141 	 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1142 	 * is set the receiver of the SACK MUST take the following actions:
1143 	 *
1144 	 * A) Initialize the cacc_saw_newack to 0 for all destination
1145 	 * addresses.
1146 	 */
1147 	if (sack->num_gap_ack_blocks &&
1148 	    primary->cacc.changeover_active) {
1149 		list_for_each_entry(transport, transport_list, transports) {
1150 			transport->cacc.cacc_saw_newack = 0;
1151 		}
1152 	}
1153 
1154 	/* Get the highest TSN in the sack. */
1155 	highest_tsn = sack_ctsn;
1156 	if (sack->num_gap_ack_blocks)
1157 		highest_tsn +=
1158 		    ntohs(frags[ntohs(sack->num_gap_ack_blocks) - 1].gab.end);
1159 
1160 	if (TSN_lt(asoc->highest_sacked, highest_tsn)) {
1161 		highest_new_tsn = highest_tsn;
1162 		asoc->highest_sacked = highest_tsn;
1163 	} else {
1164 		highest_new_tsn = sctp_highest_new_tsn(sack, asoc);
1165 	}
1166 
1167 	/* Run through the retransmit queue.  Credit bytes received
1168 	 * and free those chunks that we can.
1169 	 */
1170 	sctp_check_transmitted(q, &q->retransmit, NULL, sack, highest_new_tsn);
1171 	sctp_mark_missing(q, &q->retransmit, NULL, highest_new_tsn, 0);
1172 
1173 	/* Run through the transmitted queue.
1174 	 * Credit bytes received and free those chunks which we can.
1175 	 *
1176 	 * This is a MASSIVE candidate for optimization.
1177 	 */
1178 	list_for_each_entry(transport, transport_list, transports) {
1179 		sctp_check_transmitted(q, &transport->transmitted,
1180 				       transport, sack, highest_new_tsn);
1181 		/*
1182 		 * SFR-CACC algorithm:
1183 		 * C) Let count_of_newacks be the number of
1184 		 * destinations for which cacc_saw_newack is set.
1185 		 */
1186 		if (transport->cacc.cacc_saw_newack)
1187 			count_of_newacks ++;
1188 	}
1189 
1190 	list_for_each_entry(transport, transport_list, transports) {
1191 		sctp_mark_missing(q, &transport->transmitted, transport,
1192 				  highest_new_tsn, count_of_newacks);
1193 	}
1194 
1195 	/* Move the Cumulative TSN Ack Point if appropriate.  */
1196 	if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn))
1197 		asoc->ctsn_ack_point = sack_ctsn;
1198 
1199 	/* Update unack_data field in the assoc. */
1200 	sctp_sack_update_unack_data(asoc, sack);
1201 
1202 	ctsn = asoc->ctsn_ack_point;
1203 
1204 	/* Throw away stuff rotting on the sack queue.  */
1205 	list_for_each_safe(lchunk, temp, &q->sacked) {
1206 		tchunk = list_entry(lchunk, struct sctp_chunk,
1207 				    transmitted_list);
1208 		tsn = ntohl(tchunk->subh.data_hdr->tsn);
1209 		if (TSN_lte(tsn, ctsn)) {
1210 			list_del_init(&tchunk->transmitted_list);
1211 			sctp_chunk_free(tchunk);
1212 		}
1213 	}
1214 
1215 	/* ii) Set rwnd equal to the newly received a_rwnd minus the
1216 	 *     number of bytes still outstanding after processing the
1217 	 *     Cumulative TSN Ack and the Gap Ack Blocks.
1218 	 */
1219 
1220 	sack_a_rwnd = ntohl(sack->a_rwnd);
1221 	outstanding = q->outstanding_bytes;
1222 
1223 	if (outstanding < sack_a_rwnd)
1224 		sack_a_rwnd -= outstanding;
1225 	else
1226 		sack_a_rwnd = 0;
1227 
1228 	asoc->peer.rwnd = sack_a_rwnd;
1229 
1230 	sctp_generate_fwdtsn(q, sack_ctsn);
1231 
1232 	SCTP_DEBUG_PRINTK("%s: sack Cumulative TSN Ack is 0x%x.\n",
1233 			  __func__, sack_ctsn);
1234 	SCTP_DEBUG_PRINTK("%s: Cumulative TSN Ack of association, "
1235 			  "%p is 0x%x. Adv peer ack point: 0x%x\n",
1236 			  __func__, asoc, ctsn, asoc->adv_peer_ack_point);
1237 
1238 	/* See if all chunks are acked.
1239 	 * Make sure the empty queue handler will get run later.
1240 	 */
1241 	q->empty = (list_empty(&q->out_chunk_list) &&
1242 		    list_empty(&q->control_chunk_list) &&
1243 		    list_empty(&q->retransmit));
1244 	if (!q->empty)
1245 		goto finish;
1246 
1247 	list_for_each_entry(transport, transport_list, transports) {
1248 		q->empty = q->empty && list_empty(&transport->transmitted);
1249 		if (!q->empty)
1250 			goto finish;
1251 	}
1252 
1253 	SCTP_DEBUG_PRINTK("sack queue is empty.\n");
1254 finish:
1255 	return q->empty;
1256 }
1257 
1258 /* Is the outqueue empty?  */
1259 int sctp_outq_is_empty(const struct sctp_outq *q)
1260 {
1261 	return q->empty;
1262 }
1263 
1264 /********************************************************************
1265  * 2nd Level Abstractions
1266  ********************************************************************/
1267 
1268 /* Go through a transport's transmitted list or the association's retransmit
1269  * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1270  * The retransmit list will not have an associated transport.
1271  *
1272  * I added coherent debug information output.	--xguo
1273  *
1274  * Instead of printing 'sacked' or 'kept' for each TSN on the
1275  * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1276  * KEPT TSN6-TSN7, etc.
1277  */
1278 static void sctp_check_transmitted(struct sctp_outq *q,
1279 				   struct list_head *transmitted_queue,
1280 				   struct sctp_transport *transport,
1281 				   struct sctp_sackhdr *sack,
1282 				   __u32 highest_new_tsn_in_sack)
1283 {
1284 	struct list_head *lchunk;
1285 	struct sctp_chunk *tchunk;
1286 	struct list_head tlist;
1287 	__u32 tsn;
1288 	__u32 sack_ctsn;
1289 	__u32 rtt;
1290 	__u8 restart_timer = 0;
1291 	int bytes_acked = 0;
1292 
1293 	/* These state variables are for coherent debug output. --xguo */
1294 
1295 #if SCTP_DEBUG
1296 	__u32 dbg_ack_tsn = 0;	/* An ACKed TSN range starts here... */
1297 	__u32 dbg_last_ack_tsn = 0;  /* ...and finishes here.	     */
1298 	__u32 dbg_kept_tsn = 0;	/* An un-ACKed range starts here...  */
1299 	__u32 dbg_last_kept_tsn = 0; /* ...and finishes here.	     */
1300 
1301 	/* 0 : The last TSN was ACKed.
1302 	 * 1 : The last TSN was NOT ACKed (i.e. KEPT).
1303 	 * -1: We need to initialize.
1304 	 */
1305 	int dbg_prt_state = -1;
1306 #endif /* SCTP_DEBUG */
1307 
1308 	sack_ctsn = ntohl(sack->cum_tsn_ack);
1309 
1310 	INIT_LIST_HEAD(&tlist);
1311 
1312 	/* The while loop will skip empty transmitted queues. */
1313 	while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1314 		tchunk = list_entry(lchunk, struct sctp_chunk,
1315 				    transmitted_list);
1316 
1317 		if (sctp_chunk_abandoned(tchunk)) {
1318 			/* Move the chunk to abandoned list. */
1319 			sctp_insert_list(&q->abandoned, lchunk);
1320 
1321 			/* If this chunk has not been acked, stop
1322 			 * considering it as 'outstanding'.
1323 			 */
1324 			if (!tchunk->tsn_gap_acked) {
1325 				tchunk->transport->flight_size -=
1326 						sctp_data_size(tchunk);
1327 				q->outstanding_bytes -= sctp_data_size(tchunk);
1328 			}
1329 			continue;
1330 		}
1331 
1332 		tsn = ntohl(tchunk->subh.data_hdr->tsn);
1333 		if (sctp_acked(sack, tsn)) {
1334 			/* If this queue is the retransmit queue, the
1335 			 * retransmit timer has already reclaimed
1336 			 * the outstanding bytes for this chunk, so only
1337 			 * count bytes associated with a transport.
1338 			 */
1339 			if (transport) {
1340 				/* If this chunk is being used for RTT
1341 				 * measurement, calculate the RTT and update
1342 				 * the RTO using this value.
1343 				 *
1344 				 * 6.3.1 C5) Karn's algorithm: RTT measurements
1345 				 * MUST NOT be made using packets that were
1346 				 * retransmitted (and thus for which it is
1347 				 * ambiguous whether the reply was for the
1348 				 * first instance of the packet or a later
1349 				 * instance).
1350 				 */
1351 				if (!tchunk->tsn_gap_acked &&
1352 				    !tchunk->resent &&
1353 				    tchunk->rtt_in_progress) {
1354 					tchunk->rtt_in_progress = 0;
1355 					rtt = jiffies - tchunk->sent_at;
1356 					sctp_transport_update_rto(transport,
1357 								  rtt);
1358 				}
1359 			}
1360 			if (TSN_lte(tsn, sack_ctsn)) {
1361 				/* RFC 2960  6.3.2 Retransmission Timer Rules
1362 				 *
1363 				 * R3) Whenever a SACK is received
1364 				 * that acknowledges the DATA chunk
1365 				 * with the earliest outstanding TSN
1366 				 * for that address, restart T3-rtx
1367 				 * timer for that address with its
1368 				 * current RTO.
1369 				 */
1370 				restart_timer = 1;
1371 
1372 				if (!tchunk->tsn_gap_acked) {
1373 					tchunk->tsn_gap_acked = 1;
1374 					bytes_acked += sctp_data_size(tchunk);
1375 					/*
1376 					 * SFR-CACC algorithm:
1377 					 * 2) If the SACK contains gap acks
1378 					 * and the flag CHANGEOVER_ACTIVE is
1379 					 * set the receiver of the SACK MUST
1380 					 * take the following action:
1381 					 *
1382 					 * B) For each TSN t being acked that
1383 					 * has not been acked in any SACK so
1384 					 * far, set cacc_saw_newack to 1 for
1385 					 * the destination that the TSN was
1386 					 * sent to.
1387 					 */
1388 					if (transport &&
1389 					    sack->num_gap_ack_blocks &&
1390 					    q->asoc->peer.primary_path->cacc.
1391 					    changeover_active)
1392 						transport->cacc.cacc_saw_newack
1393 							= 1;
1394 				}
1395 
1396 				list_add_tail(&tchunk->transmitted_list,
1397 					      &q->sacked);
1398 			} else {
1399 				/* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1400 				 * M2) Each time a SACK arrives reporting
1401 				 * 'Stray DATA chunk(s)' record the highest TSN
1402 				 * reported as newly acknowledged, call this
1403 				 * value 'HighestTSNinSack'. A newly
1404 				 * acknowledged DATA chunk is one not
1405 				 * previously acknowledged in a SACK.
1406 				 *
1407 				 * When the SCTP sender of data receives a SACK
1408 				 * chunk that acknowledges, for the first time,
1409 				 * the receipt of a DATA chunk, all the still
1410 				 * unacknowledged DATA chunks whose TSN is
1411 				 * older than that newly acknowledged DATA
1412 				 * chunk, are qualified as 'Stray DATA chunks'.
1413 				 */
1414 				if (!tchunk->tsn_gap_acked) {
1415 					tchunk->tsn_gap_acked = 1;
1416 					bytes_acked += sctp_data_size(tchunk);
1417 				}
1418 				list_add_tail(lchunk, &tlist);
1419 			}
1420 
1421 #if SCTP_DEBUG
1422 			switch (dbg_prt_state) {
1423 			case 0:	/* last TSN was ACKed */
1424 				if (dbg_last_ack_tsn + 1 == tsn) {
1425 					/* This TSN belongs to the
1426 					 * current ACK range.
1427 					 */
1428 					break;
1429 				}
1430 
1431 				if (dbg_last_ack_tsn != dbg_ack_tsn) {
1432 					/* Display the end of the
1433 					 * current range.
1434 					 */
1435 					SCTP_DEBUG_PRINTK("-%08x",
1436 							  dbg_last_ack_tsn);
1437 				}
1438 
1439 				/* Start a new range.  */
1440 				SCTP_DEBUG_PRINTK(",%08x", tsn);
1441 				dbg_ack_tsn = tsn;
1442 				break;
1443 
1444 			case 1:	/* The last TSN was NOT ACKed. */
1445 				if (dbg_last_kept_tsn != dbg_kept_tsn) {
1446 					/* Display the end of current range. */
1447 					SCTP_DEBUG_PRINTK("-%08x",
1448 							  dbg_last_kept_tsn);
1449 				}
1450 
1451 				SCTP_DEBUG_PRINTK("\n");
1452 
1453 				/* FALL THROUGH... */
1454 			default:
1455 				/* This is the first-ever TSN we examined.  */
1456 				/* Start a new range of ACK-ed TSNs.  */
1457 				SCTP_DEBUG_PRINTK("ACKed: %08x", tsn);
1458 				dbg_prt_state = 0;
1459 				dbg_ack_tsn = tsn;
1460 			}
1461 
1462 			dbg_last_ack_tsn = tsn;
1463 #endif /* SCTP_DEBUG */
1464 
1465 		} else {
1466 			if (tchunk->tsn_gap_acked) {
1467 				SCTP_DEBUG_PRINTK("%s: Receiver reneged on "
1468 						  "data TSN: 0x%x\n",
1469 						  __func__,
1470 						  tsn);
1471 				tchunk->tsn_gap_acked = 0;
1472 
1473 				bytes_acked -= sctp_data_size(tchunk);
1474 
1475 				/* RFC 2960 6.3.2 Retransmission Timer Rules
1476 				 *
1477 				 * R4) Whenever a SACK is received missing a
1478 				 * TSN that was previously acknowledged via a
1479 				 * Gap Ack Block, start T3-rtx for the
1480 				 * destination address to which the DATA
1481 				 * chunk was originally
1482 				 * transmitted if it is not already running.
1483 				 */
1484 				restart_timer = 1;
1485 			}
1486 
1487 			list_add_tail(lchunk, &tlist);
1488 
1489 #if SCTP_DEBUG
1490 			/* See the above comments on ACK-ed TSNs. */
1491 			switch (dbg_prt_state) {
1492 			case 1:
1493 				if (dbg_last_kept_tsn + 1 == tsn)
1494 					break;
1495 
1496 				if (dbg_last_kept_tsn != dbg_kept_tsn)
1497 					SCTP_DEBUG_PRINTK("-%08x",
1498 							  dbg_last_kept_tsn);
1499 
1500 				SCTP_DEBUG_PRINTK(",%08x", tsn);
1501 				dbg_kept_tsn = tsn;
1502 				break;
1503 
1504 			case 0:
1505 				if (dbg_last_ack_tsn != dbg_ack_tsn)
1506 					SCTP_DEBUG_PRINTK("-%08x",
1507 							  dbg_last_ack_tsn);
1508 				SCTP_DEBUG_PRINTK("\n");
1509 
1510 				/* FALL THROUGH... */
1511 			default:
1512 				SCTP_DEBUG_PRINTK("KEPT: %08x",tsn);
1513 				dbg_prt_state = 1;
1514 				dbg_kept_tsn = tsn;
1515 			}
1516 
1517 			dbg_last_kept_tsn = tsn;
1518 #endif /* SCTP_DEBUG */
1519 		}
1520 	}
1521 
1522 #if SCTP_DEBUG
1523 	/* Finish off the last range, displaying its ending TSN.  */
1524 	switch (dbg_prt_state) {
1525 	case 0:
1526 		if (dbg_last_ack_tsn != dbg_ack_tsn) {
1527 			SCTP_DEBUG_PRINTK("-%08x\n", dbg_last_ack_tsn);
1528 		} else {
1529 			SCTP_DEBUG_PRINTK("\n");
1530 		}
1531 	break;
1532 
1533 	case 1:
1534 		if (dbg_last_kept_tsn != dbg_kept_tsn) {
1535 			SCTP_DEBUG_PRINTK("-%08x\n", dbg_last_kept_tsn);
1536 		} else {
1537 			SCTP_DEBUG_PRINTK("\n");
1538 		}
1539 	}
1540 #endif /* SCTP_DEBUG */
1541 	if (transport) {
1542 		if (bytes_acked) {
1543 			/* 8.2. When an outstanding TSN is acknowledged,
1544 			 * the endpoint shall clear the error counter of
1545 			 * the destination transport address to which the
1546 			 * DATA chunk was last sent.
1547 			 * The association's overall error counter is
1548 			 * also cleared.
1549 			 */
1550 			transport->error_count = 0;
1551 			transport->asoc->overall_error_count = 0;
1552 
1553 			/* Mark the destination transport address as
1554 			 * active if it is not so marked.
1555 			 */
1556 			if ((transport->state == SCTP_INACTIVE) ||
1557 			    (transport->state == SCTP_UNCONFIRMED)) {
1558 				sctp_assoc_control_transport(
1559 					transport->asoc,
1560 					transport,
1561 					SCTP_TRANSPORT_UP,
1562 					SCTP_RECEIVED_SACK);
1563 			}
1564 
1565 			sctp_transport_raise_cwnd(transport, sack_ctsn,
1566 						  bytes_acked);
1567 
1568 			transport->flight_size -= bytes_acked;
1569 			if (transport->flight_size == 0)
1570 				transport->partial_bytes_acked = 0;
1571 			q->outstanding_bytes -= bytes_acked;
1572 		} else {
1573 			/* RFC 2960 6.1, sctpimpguide-06 2.15.2
1574 			 * When a sender is doing zero window probing, it
1575 			 * should not timeout the association if it continues
1576 			 * to receive new packets from the receiver. The
1577 			 * reason is that the receiver MAY keep its window
1578 			 * closed for an indefinite time.
1579 			 * A sender is doing zero window probing when the
1580 			 * receiver's advertised window is zero, and there is
1581 			 * only one data chunk in flight to the receiver.
1582 			 */
1583 			if (!q->asoc->peer.rwnd &&
1584 			    !list_empty(&tlist) &&
1585 			    (sack_ctsn+2 == q->asoc->next_tsn)) {
1586 				SCTP_DEBUG_PRINTK("%s: SACK received for zero "
1587 						  "window probe: %u\n",
1588 						  __func__, sack_ctsn);
1589 				q->asoc->overall_error_count = 0;
1590 				transport->error_count = 0;
1591 			}
1592 		}
1593 
1594 		/* RFC 2960 6.3.2 Retransmission Timer Rules
1595 		 *
1596 		 * R2) Whenever all outstanding data sent to an address have
1597 		 * been acknowledged, turn off the T3-rtx timer of that
1598 		 * address.
1599 		 */
1600 		if (!transport->flight_size) {
1601 			if (timer_pending(&transport->T3_rtx_timer) &&
1602 			    del_timer(&transport->T3_rtx_timer)) {
1603 				sctp_transport_put(transport);
1604 			}
1605 		} else if (restart_timer) {
1606 			if (!mod_timer(&transport->T3_rtx_timer,
1607 				       jiffies + transport->rto))
1608 				sctp_transport_hold(transport);
1609 		}
1610 	}
1611 
1612 	list_splice(&tlist, transmitted_queue);
1613 }
1614 
1615 /* Mark chunks as missing and consequently may get retransmitted. */
1616 static void sctp_mark_missing(struct sctp_outq *q,
1617 			      struct list_head *transmitted_queue,
1618 			      struct sctp_transport *transport,
1619 			      __u32 highest_new_tsn_in_sack,
1620 			      int count_of_newacks)
1621 {
1622 	struct sctp_chunk *chunk;
1623 	__u32 tsn;
1624 	char do_fast_retransmit = 0;
1625 	struct sctp_transport *primary = q->asoc->peer.primary_path;
1626 
1627 	list_for_each_entry(chunk, transmitted_queue, transmitted_list) {
1628 
1629 		tsn = ntohl(chunk->subh.data_hdr->tsn);
1630 
1631 		/* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1632 		 * 'Unacknowledged TSN's', if the TSN number of an
1633 		 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1634 		 * value, increment the 'TSN.Missing.Report' count on that
1635 		 * chunk if it has NOT been fast retransmitted or marked for
1636 		 * fast retransmit already.
1637 		 */
1638 		if (!chunk->fast_retransmit &&
1639 		    !chunk->tsn_gap_acked &&
1640 		    TSN_lt(tsn, highest_new_tsn_in_sack)) {
1641 
1642 			/* SFR-CACC may require us to skip marking
1643 			 * this chunk as missing.
1644 			 */
1645 			if (!transport || !sctp_cacc_skip(primary, transport,
1646 					    count_of_newacks, tsn)) {
1647 				chunk->tsn_missing_report++;
1648 
1649 				SCTP_DEBUG_PRINTK(
1650 					"%s: TSN 0x%x missing counter: %d\n",
1651 					__func__, tsn,
1652 					chunk->tsn_missing_report);
1653 			}
1654 		}
1655 		/*
1656 		 * M4) If any DATA chunk is found to have a
1657 		 * 'TSN.Missing.Report'
1658 		 * value larger than or equal to 3, mark that chunk for
1659 		 * retransmission and start the fast retransmit procedure.
1660 		 */
1661 
1662 		if (chunk->tsn_missing_report >= 3) {
1663 			chunk->fast_retransmit = 1;
1664 			do_fast_retransmit = 1;
1665 		}
1666 	}
1667 
1668 	if (transport) {
1669 		if (do_fast_retransmit)
1670 			sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1671 
1672 		SCTP_DEBUG_PRINTK("%s: transport: %p, cwnd: %d, "
1673 				  "ssthresh: %d, flight_size: %d, pba: %d\n",
1674 				  __func__, transport, transport->cwnd,
1675 				  transport->ssthresh, transport->flight_size,
1676 				  transport->partial_bytes_acked);
1677 	}
1678 }
1679 
1680 /* Is the given TSN acked by this packet?  */
1681 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1682 {
1683 	int i;
1684 	sctp_sack_variable_t *frags;
1685 	__u16 gap;
1686 	__u32 ctsn = ntohl(sack->cum_tsn_ack);
1687 
1688 	if (TSN_lte(tsn, ctsn))
1689 		goto pass;
1690 
1691 	/* 3.3.4 Selective Acknowledgement (SACK) (3):
1692 	 *
1693 	 * Gap Ack Blocks:
1694 	 *  These fields contain the Gap Ack Blocks. They are repeated
1695 	 *  for each Gap Ack Block up to the number of Gap Ack Blocks
1696 	 *  defined in the Number of Gap Ack Blocks field. All DATA
1697 	 *  chunks with TSNs greater than or equal to (Cumulative TSN
1698 	 *  Ack + Gap Ack Block Start) and less than or equal to
1699 	 *  (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1700 	 *  Block are assumed to have been received correctly.
1701 	 */
1702 
1703 	frags = sack->variable;
1704 	gap = tsn - ctsn;
1705 	for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) {
1706 		if (TSN_lte(ntohs(frags[i].gab.start), gap) &&
1707 		    TSN_lte(gap, ntohs(frags[i].gab.end)))
1708 			goto pass;
1709 	}
1710 
1711 	return 0;
1712 pass:
1713 	return 1;
1714 }
1715 
1716 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1717 				    int nskips, __be16 stream)
1718 {
1719 	int i;
1720 
1721 	for (i = 0; i < nskips; i++) {
1722 		if (skiplist[i].stream == stream)
1723 			return i;
1724 	}
1725 	return i;
1726 }
1727 
1728 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1729 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1730 {
1731 	struct sctp_association *asoc = q->asoc;
1732 	struct sctp_chunk *ftsn_chunk = NULL;
1733 	struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1734 	int nskips = 0;
1735 	int skip_pos = 0;
1736 	__u32 tsn;
1737 	struct sctp_chunk *chunk;
1738 	struct list_head *lchunk, *temp;
1739 
1740 	/* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1741 	 * received SACK.
1742 	 *
1743 	 * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1744 	 * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1745 	 */
1746 	if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1747 		asoc->adv_peer_ack_point = ctsn;
1748 
1749 	/* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1750 	 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1751 	 * the chunk next in the out-queue space is marked as "abandoned" as
1752 	 * shown in the following example:
1753 	 *
1754 	 * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1755 	 * and the Advanced.Peer.Ack.Point is updated to this value:
1756 	 *
1757 	 *   out-queue at the end of  ==>   out-queue after Adv.Ack.Point
1758 	 *   normal SACK processing           local advancement
1759 	 *                ...                           ...
1760 	 *   Adv.Ack.Pt-> 102 acked                     102 acked
1761 	 *                103 abandoned                 103 abandoned
1762 	 *                104 abandoned     Adv.Ack.P-> 104 abandoned
1763 	 *                105                           105
1764 	 *                106 acked                     106 acked
1765 	 *                ...                           ...
1766 	 *
1767 	 * In this example, the data sender successfully advanced the
1768 	 * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1769 	 */
1770 	list_for_each_safe(lchunk, temp, &q->abandoned) {
1771 		chunk = list_entry(lchunk, struct sctp_chunk,
1772 					transmitted_list);
1773 		tsn = ntohl(chunk->subh.data_hdr->tsn);
1774 
1775 		/* Remove any chunks in the abandoned queue that are acked by
1776 		 * the ctsn.
1777 		 */
1778 		if (TSN_lte(tsn, ctsn)) {
1779 			list_del_init(lchunk);
1780 			sctp_chunk_free(chunk);
1781 		} else {
1782 			if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1783 				asoc->adv_peer_ack_point = tsn;
1784 				if (chunk->chunk_hdr->flags &
1785 					 SCTP_DATA_UNORDERED)
1786 					continue;
1787 				skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1788 						nskips,
1789 						chunk->subh.data_hdr->stream);
1790 				ftsn_skip_arr[skip_pos].stream =
1791 					chunk->subh.data_hdr->stream;
1792 				ftsn_skip_arr[skip_pos].ssn =
1793 					 chunk->subh.data_hdr->ssn;
1794 				if (skip_pos == nskips)
1795 					nskips++;
1796 				if (nskips == 10)
1797 					break;
1798 			} else
1799 				break;
1800 		}
1801 	}
1802 
1803 	/* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1804 	 * is greater than the Cumulative TSN ACK carried in the received
1805 	 * SACK, the data sender MUST send the data receiver a FORWARD TSN
1806 	 * chunk containing the latest value of the
1807 	 * "Advanced.Peer.Ack.Point".
1808 	 *
1809 	 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1810 	 * list each stream and sequence number in the forwarded TSN. This
1811 	 * information will enable the receiver to easily find any
1812 	 * stranded TSN's waiting on stream reorder queues. Each stream
1813 	 * SHOULD only be reported once; this means that if multiple
1814 	 * abandoned messages occur in the same stream then only the
1815 	 * highest abandoned stream sequence number is reported. If the
1816 	 * total size of the FORWARD TSN does NOT fit in a single MTU then
1817 	 * the sender of the FORWARD TSN SHOULD lower the
1818 	 * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1819 	 * single MTU.
1820 	 */
1821 	if (asoc->adv_peer_ack_point > ctsn)
1822 		ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1823 					      nskips, &ftsn_skip_arr[0]);
1824 
1825 	if (ftsn_chunk) {
1826 		list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1827 		SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS);
1828 	}
1829 }
1830