xref: /linux/net/sctp/outqueue.c (revision d8327c784b51b57dac2c26cfad87dce0d68dfd98)
1 /* SCTP kernel reference 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 reference Implementation
8  *
9  * These functions implement the sctp_outq class.   The outqueue handles
10  * bundling and queueing of outgoing SCTP chunks.
11  *
12  * The SCTP reference 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  * The SCTP reference 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->outstanding_bytes = 0;
212 	q->empty = 1;
213 	q->cork  = 0;
214 
215 	q->malloced = 0;
216 	q->out_qlen = 0;
217 }
218 
219 /* Free the outqueue structure and any related pending chunks.
220  */
221 void sctp_outq_teardown(struct sctp_outq *q)
222 {
223 	struct sctp_transport *transport;
224 	struct list_head *lchunk, *pos, *temp;
225 	struct sctp_chunk *chunk, *tmp;
226 
227 	/* Throw away unacknowledged chunks. */
228 	list_for_each(pos, &q->asoc->peer.transport_addr_list) {
229 		transport = list_entry(pos, struct sctp_transport, transports);
230 		while ((lchunk = sctp_list_dequeue(&transport->transmitted)) != NULL) {
231 			chunk = list_entry(lchunk, struct sctp_chunk,
232 					   transmitted_list);
233 			/* Mark as part of a failed message. */
234 			sctp_chunk_fail(chunk, q->error);
235 			sctp_chunk_free(chunk);
236 		}
237 	}
238 
239 	/* Throw away chunks that have been gap ACKed.  */
240 	list_for_each_safe(lchunk, temp, &q->sacked) {
241 		list_del_init(lchunk);
242 		chunk = list_entry(lchunk, struct sctp_chunk,
243 				   transmitted_list);
244 		sctp_chunk_fail(chunk, q->error);
245 		sctp_chunk_free(chunk);
246 	}
247 
248 	/* Throw away any chunks in the retransmit queue. */
249 	list_for_each_safe(lchunk, temp, &q->retransmit) {
250 		list_del_init(lchunk);
251 		chunk = list_entry(lchunk, struct sctp_chunk,
252 				   transmitted_list);
253 		sctp_chunk_fail(chunk, q->error);
254 		sctp_chunk_free(chunk);
255 	}
256 
257 	/* Throw away any chunks that are in the abandoned queue. */
258 	list_for_each_safe(lchunk, temp, &q->abandoned) {
259 		list_del_init(lchunk);
260 		chunk = list_entry(lchunk, struct sctp_chunk,
261 				   transmitted_list);
262 		sctp_chunk_fail(chunk, q->error);
263 		sctp_chunk_free(chunk);
264 	}
265 
266 	/* Throw away any leftover data chunks. */
267 	while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
268 
269 		/* Mark as send failure. */
270 		sctp_chunk_fail(chunk, q->error);
271 		sctp_chunk_free(chunk);
272 	}
273 
274 	q->error = 0;
275 
276 	/* Throw away any leftover control chunks. */
277 	list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
278 		list_del_init(&chunk->list);
279 		sctp_chunk_free(chunk);
280 	}
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 	/* If we were kmalloc()'d, free the memory.  */
290 	if (q->malloced)
291 		kfree(q);
292 }
293 
294 /* Put a new chunk in an sctp_outq.  */
295 int sctp_outq_tail(struct sctp_outq *q, struct sctp_chunk *chunk)
296 {
297 	int error = 0;
298 
299 	SCTP_DEBUG_PRINTK("sctp_outq_tail(%p, %p[%s])\n",
300 			  q, chunk, chunk && chunk->chunk_hdr ?
301 			  sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
302 			  : "Illegal Chunk");
303 
304 	/* If it is data, queue it up, otherwise, send it
305 	 * immediately.
306 	 */
307 	if (SCTP_CID_DATA == chunk->chunk_hdr->type) {
308 		/* Is it OK to queue data chunks?  */
309 		/* From 9. Termination of Association
310 		 *
311 		 * When either endpoint performs a shutdown, the
312 		 * association on each peer will stop accepting new
313 		 * data from its user and only deliver data in queue
314 		 * at the time of sending or receiving the SHUTDOWN
315 		 * chunk.
316 		 */
317 		switch (q->asoc->state) {
318 		case SCTP_STATE_EMPTY:
319 		case SCTP_STATE_CLOSED:
320 		case SCTP_STATE_SHUTDOWN_PENDING:
321 		case SCTP_STATE_SHUTDOWN_SENT:
322 		case SCTP_STATE_SHUTDOWN_RECEIVED:
323 		case SCTP_STATE_SHUTDOWN_ACK_SENT:
324 			/* Cannot send after transport endpoint shutdown */
325 			error = -ESHUTDOWN;
326 			break;
327 
328 		default:
329 			SCTP_DEBUG_PRINTK("outqueueing (%p, %p[%s])\n",
330 			  q, chunk, chunk && chunk->chunk_hdr ?
331 			  sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type))
332 			  : "Illegal Chunk");
333 
334 			sctp_outq_tail_data(q, chunk);
335 			if (chunk->chunk_hdr->flags & SCTP_DATA_UNORDERED)
336 				SCTP_INC_STATS(SCTP_MIB_OUTUNORDERCHUNKS);
337 			else
338 				SCTP_INC_STATS(SCTP_MIB_OUTORDERCHUNKS);
339 			q->empty = 0;
340 			break;
341 		};
342 	} else {
343 		list_add_tail(&chunk->list, &q->control_chunk_list);
344 		SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS);
345 	}
346 
347 	if (error < 0)
348 		return error;
349 
350 	if (!q->cork)
351 		error = sctp_outq_flush(q, 0);
352 
353 	return error;
354 }
355 
356 /* Insert a chunk into the sorted list based on the TSNs.  The retransmit list
357  * and the abandoned list are in ascending order.
358  */
359 static void sctp_insert_list(struct list_head *head, struct list_head *new)
360 {
361 	struct list_head *pos;
362 	struct sctp_chunk *nchunk, *lchunk;
363 	__u32 ntsn, ltsn;
364 	int done = 0;
365 
366 	nchunk = list_entry(new, struct sctp_chunk, transmitted_list);
367 	ntsn = ntohl(nchunk->subh.data_hdr->tsn);
368 
369 	list_for_each(pos, head) {
370 		lchunk = list_entry(pos, struct sctp_chunk, transmitted_list);
371 		ltsn = ntohl(lchunk->subh.data_hdr->tsn);
372 		if (TSN_lt(ntsn, ltsn)) {
373 			list_add(new, pos->prev);
374 			done = 1;
375 			break;
376 		}
377 	}
378 	if (!done)
379 		list_add_tail(new, head);
380 }
381 
382 /* Mark all the eligible packets on a transport for retransmission.  */
383 void sctp_retransmit_mark(struct sctp_outq *q,
384 			  struct sctp_transport *transport,
385 			  __u8 fast_retransmit)
386 {
387 	struct list_head *lchunk, *ltemp;
388 	struct sctp_chunk *chunk;
389 
390 	/* Walk through the specified transmitted queue.  */
391 	list_for_each_safe(lchunk, ltemp, &transport->transmitted) {
392 		chunk = list_entry(lchunk, struct sctp_chunk,
393 				   transmitted_list);
394 
395 		/* If the chunk is abandoned, move it to abandoned list. */
396 		if (sctp_chunk_abandoned(chunk)) {
397 			list_del_init(lchunk);
398 			sctp_insert_list(&q->abandoned, lchunk);
399 			continue;
400 		}
401 
402 		/* If we are doing retransmission due to a fast retransmit,
403 		 * only the chunk's that are marked for fast retransmit
404 		 * should be added to the retransmit queue.  If we are doing
405 		 * retransmission due to a timeout or pmtu discovery, only the
406 		 * chunks that are not yet acked should be added to the
407 		 * retransmit queue.
408 		 */
409 		if ((fast_retransmit && (chunk->fast_retransmit > 0)) ||
410 		   (!fast_retransmit && !chunk->tsn_gap_acked)) {
411 			/* RFC 2960 6.2.1 Processing a Received SACK
412 			 *
413 			 * C) Any time a DATA chunk is marked for
414 			 * retransmission (via either T3-rtx timer expiration
415 			 * (Section 6.3.3) or via fast retransmit
416 			 * (Section 7.2.4)), add the data size of those
417 			 * chunks to the rwnd.
418 			 */
419 			q->asoc->peer.rwnd += sctp_data_size(chunk);
420 			q->outstanding_bytes -= sctp_data_size(chunk);
421 			transport->flight_size -= sctp_data_size(chunk);
422 
423 			/* sctpimpguide-05 Section 2.8.2
424 			 * M5) If a T3-rtx timer expires, the
425 			 * 'TSN.Missing.Report' of all affected TSNs is set
426 			 * to 0.
427 			 */
428 			chunk->tsn_missing_report = 0;
429 
430 			/* If a chunk that is being used for RTT measurement
431 			 * has to be retransmitted, we cannot use this chunk
432 			 * anymore for RTT measurements. Reset rto_pending so
433 			 * that a new RTT measurement is started when a new
434 			 * data chunk is sent.
435 			 */
436 			if (chunk->rtt_in_progress) {
437 				chunk->rtt_in_progress = 0;
438 				transport->rto_pending = 0;
439 			}
440 
441 			/* Move the chunk to the retransmit queue. The chunks
442 			 * on the retransmit queue are always kept in order.
443 			 */
444 			list_del_init(lchunk);
445 			sctp_insert_list(&q->retransmit, lchunk);
446 		}
447 	}
448 
449 	SCTP_DEBUG_PRINTK("%s: transport: %p, fast_retransmit: %d, "
450 			  "cwnd: %d, ssthresh: %d, flight_size: %d, "
451 			  "pba: %d\n", __FUNCTION__,
452 			  transport, fast_retransmit,
453 			  transport->cwnd, transport->ssthresh,
454 			  transport->flight_size,
455 			  transport->partial_bytes_acked);
456 
457 }
458 
459 /* Mark all the eligible packets on a transport for retransmission and force
460  * one packet out.
461  */
462 void sctp_retransmit(struct sctp_outq *q, struct sctp_transport *transport,
463 		     sctp_retransmit_reason_t reason)
464 {
465 	int error = 0;
466 	__u8 fast_retransmit = 0;
467 
468 	switch(reason) {
469 	case SCTP_RTXR_T3_RTX:
470 		sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_T3_RTX);
471 		/* Update the retran path if the T3-rtx timer has expired for
472 		 * the current retran path.
473 		 */
474 		if (transport == transport->asoc->peer.retran_path)
475 			sctp_assoc_update_retran_path(transport->asoc);
476 		break;
477 	case SCTP_RTXR_FAST_RTX:
478 		sctp_transport_lower_cwnd(transport, SCTP_LOWER_CWND_FAST_RTX);
479 		fast_retransmit = 1;
480 		break;
481 	case SCTP_RTXR_PMTUD:
482 	default:
483 		break;
484 	}
485 
486 	sctp_retransmit_mark(q, transport, fast_retransmit);
487 
488 	/* PR-SCTP A5) Any time the T3-rtx timer expires, on any destination,
489 	 * the sender SHOULD try to advance the "Advanced.Peer.Ack.Point" by
490 	 * following the procedures outlined in C1 - C5.
491 	 */
492 	sctp_generate_fwdtsn(q, q->asoc->ctsn_ack_point);
493 
494 	error = sctp_outq_flush(q, /* rtx_timeout */ 1);
495 
496 	if (error)
497 		q->asoc->base.sk->sk_err = -error;
498 }
499 
500 /*
501  * Transmit DATA chunks on the retransmit queue.  Upon return from
502  * sctp_outq_flush_rtx() the packet 'pkt' may contain chunks which
503  * need to be transmitted by the caller.
504  * We assume that pkt->transport has already been set.
505  *
506  * The return value is a normal kernel error return value.
507  */
508 static int sctp_outq_flush_rtx(struct sctp_outq *q, struct sctp_packet *pkt,
509 			       int rtx_timeout, int *start_timer)
510 {
511 	struct list_head *lqueue;
512 	struct list_head *lchunk, *lchunk1;
513 	struct sctp_transport *transport = pkt->transport;
514 	sctp_xmit_t status;
515 	struct sctp_chunk *chunk, *chunk1;
516 	struct sctp_association *asoc;
517 	int error = 0;
518 
519 	asoc = q->asoc;
520 	lqueue = &q->retransmit;
521 
522 	/* RFC 2960 6.3.3 Handle T3-rtx Expiration
523 	 *
524 	 * E3) Determine how many of the earliest (i.e., lowest TSN)
525 	 * outstanding DATA chunks for the address for which the
526 	 * T3-rtx has expired will fit into a single packet, subject
527 	 * to the MTU constraint for the path corresponding to the
528 	 * destination transport address to which the retransmission
529 	 * is being sent (this may be different from the address for
530 	 * which the timer expires [see Section 6.4]). Call this value
531 	 * K. Bundle and retransmit those K DATA chunks in a single
532 	 * packet to the destination endpoint.
533 	 *
534 	 * [Just to be painfully clear, if we are retransmitting
535 	 * because a timeout just happened, we should send only ONE
536 	 * packet of retransmitted data.]
537 	 */
538 	lchunk = sctp_list_dequeue(lqueue);
539 
540 	while (lchunk) {
541 		chunk = list_entry(lchunk, struct sctp_chunk,
542 				   transmitted_list);
543 
544 		/* Make sure that Gap Acked TSNs are not retransmitted.  A
545 		 * simple approach is just to move such TSNs out of the
546 		 * way and into a 'transmitted' queue and skip to the
547 		 * next chunk.
548 		 */
549 		if (chunk->tsn_gap_acked) {
550 			list_add_tail(lchunk, &transport->transmitted);
551 			lchunk = sctp_list_dequeue(lqueue);
552 			continue;
553 		}
554 
555 		/* Attempt to append this chunk to the packet. */
556 		status = sctp_packet_append_chunk(pkt, chunk);
557 
558 		switch (status) {
559 		case SCTP_XMIT_PMTU_FULL:
560 			/* Send this packet.  */
561 			if ((error = sctp_packet_transmit(pkt)) == 0)
562 				*start_timer = 1;
563 
564 			/* If we are retransmitting, we should only
565 			 * send a single packet.
566 			 */
567 			if (rtx_timeout) {
568 				list_add(lchunk, lqueue);
569 				lchunk = NULL;
570 			}
571 
572 			/* Bundle lchunk in the next round.  */
573 			break;
574 
575 		case SCTP_XMIT_RWND_FULL:
576 		        /* Send this packet. */
577 			if ((error = sctp_packet_transmit(pkt)) == 0)
578 				*start_timer = 1;
579 
580 			/* Stop sending DATA as there is no more room
581 			 * at the receiver.
582 			 */
583 			list_add(lchunk, lqueue);
584 			lchunk = NULL;
585 			break;
586 
587 		case SCTP_XMIT_NAGLE_DELAY:
588 		        /* Send this packet. */
589 			if ((error = sctp_packet_transmit(pkt)) == 0)
590 				*start_timer = 1;
591 
592 			/* Stop sending DATA because of nagle delay. */
593 			list_add(lchunk, lqueue);
594 			lchunk = NULL;
595 			break;
596 
597 		default:
598 			/* The append was successful, so add this chunk to
599 			 * the transmitted list.
600 			 */
601 			list_add_tail(lchunk, &transport->transmitted);
602 
603 			/* Mark the chunk as ineligible for fast retransmit
604 			 * after it is retransmitted.
605 			 */
606 			if (chunk->fast_retransmit > 0)
607 				chunk->fast_retransmit = -1;
608 
609 			*start_timer = 1;
610 			q->empty = 0;
611 
612 			/* Retrieve a new chunk to bundle. */
613 			lchunk = sctp_list_dequeue(lqueue);
614 			break;
615 		};
616 
617 		/* If we are here due to a retransmit timeout or a fast
618 		 * retransmit and if there are any chunks left in the retransmit
619 		 * queue that could not fit in the PMTU sized packet, they need			 * to be marked as ineligible for a subsequent fast retransmit.
620 		 */
621 		if (rtx_timeout && !lchunk) {
622 			list_for_each(lchunk1, lqueue) {
623 				chunk1 = list_entry(lchunk1, struct sctp_chunk,
624 						    transmitted_list);
625 				if (chunk1->fast_retransmit > 0)
626 					chunk1->fast_retransmit = -1;
627 			}
628 		}
629 	}
630 
631 	return error;
632 }
633 
634 /* Cork the outqueue so queued chunks are really queued. */
635 int sctp_outq_uncork(struct sctp_outq *q)
636 {
637 	int error = 0;
638 	if (q->cork) {
639 		q->cork = 0;
640 		error = sctp_outq_flush(q, 0);
641 	}
642 	return error;
643 }
644 
645 /*
646  * Try to flush an outqueue.
647  *
648  * Description: Send everything in q which we legally can, subject to
649  * congestion limitations.
650  * * Note: This function can be called from multiple contexts so appropriate
651  * locking concerns must be made.  Today we use the sock lock to protect
652  * this function.
653  */
654 int sctp_outq_flush(struct sctp_outq *q, int rtx_timeout)
655 {
656 	struct sctp_packet *packet;
657 	struct sctp_packet singleton;
658 	struct sctp_association *asoc = q->asoc;
659 	__u16 sport = asoc->base.bind_addr.port;
660 	__u16 dport = asoc->peer.port;
661 	__u32 vtag = asoc->peer.i.init_tag;
662 	struct sctp_transport *transport = NULL;
663 	struct sctp_transport *new_transport;
664 	struct sctp_chunk *chunk, *tmp;
665 	sctp_xmit_t status;
666 	int error = 0;
667 	int start_timer = 0;
668 
669 	/* These transports have chunks to send. */
670 	struct list_head transport_list;
671 	struct list_head *ltransport;
672 
673 	INIT_LIST_HEAD(&transport_list);
674 	packet = NULL;
675 
676 	/*
677 	 * 6.10 Bundling
678 	 *   ...
679 	 *   When bundling control chunks with DATA chunks, an
680 	 *   endpoint MUST place control chunks first in the outbound
681 	 *   SCTP packet.  The transmitter MUST transmit DATA chunks
682 	 *   within a SCTP packet in increasing order of TSN.
683 	 *   ...
684 	 */
685 
686 	list_for_each_entry_safe(chunk, tmp, &q->control_chunk_list, list) {
687 		list_del_init(&chunk->list);
688 
689 		/* Pick the right transport to use. */
690 		new_transport = chunk->transport;
691 
692 		if (!new_transport) {
693 			new_transport = asoc->peer.active_path;
694 		} else if (new_transport->state == SCTP_INACTIVE) {
695 			/* If the chunk is Heartbeat or Heartbeat Ack,
696 			 * send it to chunk->transport, even if it's
697 			 * inactive.
698 			 *
699 			 * 3.3.6 Heartbeat Acknowledgement:
700 			 * ...
701 			 * A HEARTBEAT ACK is always sent to the source IP
702 			 * address of the IP datagram containing the
703 			 * HEARTBEAT chunk to which this ack is responding.
704 			 * ...
705 			 */
706 			if (chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT &&
707 			    chunk->chunk_hdr->type != SCTP_CID_HEARTBEAT_ACK)
708 				new_transport = asoc->peer.active_path;
709 		}
710 
711 		/* Are we switching transports?
712 		 * Take care of transport locks.
713 		 */
714 		if (new_transport != transport) {
715 			transport = new_transport;
716 			if (list_empty(&transport->send_ready)) {
717 				list_add_tail(&transport->send_ready,
718 					      &transport_list);
719 			}
720 			packet = &transport->packet;
721 			sctp_packet_config(packet, vtag,
722 					   asoc->peer.ecn_capable);
723 		}
724 
725 		switch (chunk->chunk_hdr->type) {
726 		/*
727 		 * 6.10 Bundling
728 		 *   ...
729 		 *   An endpoint MUST NOT bundle INIT, INIT ACK or SHUTDOWN
730 		 *   COMPLETE with any other chunks.  [Send them immediately.]
731 		 */
732 		case SCTP_CID_INIT:
733 		case SCTP_CID_INIT_ACK:
734 		case SCTP_CID_SHUTDOWN_COMPLETE:
735 			sctp_packet_init(&singleton, transport, sport, dport);
736 			sctp_packet_config(&singleton, vtag, 0);
737 			sctp_packet_append_chunk(&singleton, chunk);
738 			error = sctp_packet_transmit(&singleton);
739 			if (error < 0)
740 				return error;
741 			break;
742 
743 		case SCTP_CID_ABORT:
744 		case SCTP_CID_SACK:
745 		case SCTP_CID_HEARTBEAT:
746 		case SCTP_CID_HEARTBEAT_ACK:
747 		case SCTP_CID_SHUTDOWN:
748 		case SCTP_CID_SHUTDOWN_ACK:
749 		case SCTP_CID_ERROR:
750 		case SCTP_CID_COOKIE_ECHO:
751 		case SCTP_CID_COOKIE_ACK:
752 		case SCTP_CID_ECN_ECNE:
753 		case SCTP_CID_ECN_CWR:
754 		case SCTP_CID_ASCONF:
755 		case SCTP_CID_ASCONF_ACK:
756 		case SCTP_CID_FWD_TSN:
757 			sctp_packet_transmit_chunk(packet, chunk);
758 			break;
759 
760 		default:
761 			/* We built a chunk with an illegal type! */
762 			BUG();
763 		};
764 	}
765 
766 	/* Is it OK to send data chunks?  */
767 	switch (asoc->state) {
768 	case SCTP_STATE_COOKIE_ECHOED:
769 		/* Only allow bundling when this packet has a COOKIE-ECHO
770 		 * chunk.
771 		 */
772 		if (!packet || !packet->has_cookie_echo)
773 			break;
774 
775 		/* fallthru */
776 	case SCTP_STATE_ESTABLISHED:
777 	case SCTP_STATE_SHUTDOWN_PENDING:
778 	case SCTP_STATE_SHUTDOWN_RECEIVED:
779 		/*
780 		 * RFC 2960 6.1  Transmission of DATA Chunks
781 		 *
782 		 * C) When the time comes for the sender to transmit,
783 		 * before sending new DATA chunks, the sender MUST
784 		 * first transmit any outstanding DATA chunks which
785 		 * are marked for retransmission (limited by the
786 		 * current cwnd).
787 		 */
788 		if (!list_empty(&q->retransmit)) {
789 			if (transport == asoc->peer.retran_path)
790 				goto retran;
791 
792 			/* Switch transports & prepare the packet.  */
793 
794 			transport = asoc->peer.retran_path;
795 
796 			if (list_empty(&transport->send_ready)) {
797 				list_add_tail(&transport->send_ready,
798 					      &transport_list);
799 			}
800 
801 			packet = &transport->packet;
802 			sctp_packet_config(packet, vtag,
803 					   asoc->peer.ecn_capable);
804 		retran:
805 			error = sctp_outq_flush_rtx(q, packet,
806 						    rtx_timeout, &start_timer);
807 
808 			if (start_timer)
809 				sctp_transport_reset_timers(transport);
810 
811 			/* This can happen on COOKIE-ECHO resend.  Only
812 			 * one chunk can get bundled with a COOKIE-ECHO.
813 			 */
814 			if (packet->has_cookie_echo)
815 				goto sctp_flush_out;
816 
817 			/* Don't send new data if there is still data
818 			 * waiting to retransmit.
819 			 */
820 			if (!list_empty(&q->retransmit))
821 				goto sctp_flush_out;
822 		}
823 
824 		/* Finally, transmit new packets.  */
825 		start_timer = 0;
826 		while ((chunk = sctp_outq_dequeue_data(q)) != NULL) {
827 			/* RFC 2960 6.5 Every DATA chunk MUST carry a valid
828 			 * stream identifier.
829 			 */
830 			if (chunk->sinfo.sinfo_stream >=
831 			    asoc->c.sinit_num_ostreams) {
832 
833 				/* Mark as failed send. */
834 				sctp_chunk_fail(chunk, SCTP_ERROR_INV_STRM);
835 				sctp_chunk_free(chunk);
836 				continue;
837 			}
838 
839 			/* Has this chunk expired? */
840 			if (sctp_chunk_abandoned(chunk)) {
841 				sctp_chunk_fail(chunk, 0);
842 				sctp_chunk_free(chunk);
843 				continue;
844 			}
845 
846 			/* If there is a specified transport, use it.
847 			 * Otherwise, we want to use the active path.
848 			 */
849 			new_transport = chunk->transport;
850 			if (!new_transport ||
851 			    new_transport->state == SCTP_INACTIVE)
852 				new_transport = asoc->peer.active_path;
853 
854 			/* Change packets if necessary.  */
855 			if (new_transport != transport) {
856 				transport = new_transport;
857 
858 				/* Schedule to have this transport's
859 				 * packet flushed.
860 				 */
861 				if (list_empty(&transport->send_ready)) {
862 					list_add_tail(&transport->send_ready,
863 						      &transport_list);
864 				}
865 
866 				packet = &transport->packet;
867 				sctp_packet_config(packet, vtag,
868 						   asoc->peer.ecn_capable);
869 			}
870 
871 			SCTP_DEBUG_PRINTK("sctp_outq_flush(%p, %p[%s]), ",
872 					  q, chunk,
873 					  chunk && chunk->chunk_hdr ?
874 					  sctp_cname(SCTP_ST_CHUNK(
875 						  chunk->chunk_hdr->type))
876 					  : "Illegal Chunk");
877 
878 			SCTP_DEBUG_PRINTK("TX TSN 0x%x skb->head "
879 					"%p skb->users %d.\n",
880 					ntohl(chunk->subh.data_hdr->tsn),
881 					chunk->skb ?chunk->skb->head : NULL,
882 					chunk->skb ?
883 					atomic_read(&chunk->skb->users) : -1);
884 
885 			/* Add the chunk to the packet.  */
886 			status = sctp_packet_transmit_chunk(packet, chunk);
887 
888 			switch (status) {
889 			case SCTP_XMIT_PMTU_FULL:
890 			case SCTP_XMIT_RWND_FULL:
891 			case SCTP_XMIT_NAGLE_DELAY:
892 				/* We could not append this chunk, so put
893 				 * the chunk back on the output queue.
894 				 */
895 				SCTP_DEBUG_PRINTK("sctp_outq_flush: could "
896 					"not transmit TSN: 0x%x, status: %d\n",
897 					ntohl(chunk->subh.data_hdr->tsn),
898 					status);
899 				sctp_outq_head_data(q, chunk);
900 				goto sctp_flush_out;
901 				break;
902 
903 			case SCTP_XMIT_OK:
904 				break;
905 
906 			default:
907 				BUG();
908 			}
909 
910 			/* BUG: We assume that the sctp_packet_transmit()
911 			 * call below will succeed all the time and add the
912 			 * chunk to the transmitted list and restart the
913 			 * timers.
914 			 * It is possible that the call can fail under OOM
915 			 * conditions.
916 			 *
917 			 * Is this really a problem?  Won't this behave
918 			 * like a lost TSN?
919 			 */
920 			list_add_tail(&chunk->transmitted_list,
921 				      &transport->transmitted);
922 
923 			sctp_transport_reset_timers(transport);
924 
925 			q->empty = 0;
926 
927 			/* Only let one DATA chunk get bundled with a
928 			 * COOKIE-ECHO chunk.
929 			 */
930 			if (packet->has_cookie_echo)
931 				goto sctp_flush_out;
932 		}
933 		break;
934 
935 	default:
936 		/* Do nothing.  */
937 		break;
938 	}
939 
940 sctp_flush_out:
941 
942 	/* Before returning, examine all the transports touched in
943 	 * this call.  Right now, we bluntly force clear all the
944 	 * transports.  Things might change after we implement Nagle.
945 	 * But such an examination is still required.
946 	 *
947 	 * --xguo
948 	 */
949 	while ((ltransport = sctp_list_dequeue(&transport_list)) != NULL ) {
950 		struct sctp_transport *t = list_entry(ltransport,
951 						      struct sctp_transport,
952 						      send_ready);
953 		packet = &t->packet;
954 		if (!sctp_packet_empty(packet))
955 			error = sctp_packet_transmit(packet);
956 	}
957 
958 	return error;
959 }
960 
961 /* Update unack_data based on the incoming SACK chunk */
962 static void sctp_sack_update_unack_data(struct sctp_association *assoc,
963 					struct sctp_sackhdr *sack)
964 {
965 	sctp_sack_variable_t *frags;
966 	__u16 unack_data;
967 	int i;
968 
969 	unack_data = assoc->next_tsn - assoc->ctsn_ack_point - 1;
970 
971 	frags = sack->variable;
972 	for (i = 0; i < ntohs(sack->num_gap_ack_blocks); i++) {
973 		unack_data -= ((ntohs(frags[i].gab.end) -
974 				ntohs(frags[i].gab.start) + 1));
975 	}
976 
977 	assoc->unack_data = unack_data;
978 }
979 
980 /* Return the highest new tsn that is acknowledged by the given SACK chunk. */
981 static __u32 sctp_highest_new_tsn(struct sctp_sackhdr *sack,
982 				  struct sctp_association *asoc)
983 {
984 	struct list_head *ltransport, *lchunk;
985 	struct sctp_transport *transport;
986 	struct sctp_chunk *chunk;
987 	__u32 highest_new_tsn, tsn;
988 	struct list_head *transport_list = &asoc->peer.transport_addr_list;
989 
990 	highest_new_tsn = ntohl(sack->cum_tsn_ack);
991 
992 	list_for_each(ltransport, transport_list) {
993 		transport = list_entry(ltransport, struct sctp_transport,
994 				       transports);
995 		list_for_each(lchunk, &transport->transmitted) {
996 			chunk = list_entry(lchunk, struct sctp_chunk,
997 					   transmitted_list);
998 			tsn = ntohl(chunk->subh.data_hdr->tsn);
999 
1000 			if (!chunk->tsn_gap_acked &&
1001 			    TSN_lt(highest_new_tsn, tsn) &&
1002 			    sctp_acked(sack, tsn))
1003 				highest_new_tsn = tsn;
1004 		}
1005 	}
1006 
1007 	return highest_new_tsn;
1008 }
1009 
1010 /* This is where we REALLY process a SACK.
1011  *
1012  * Process the SACK against the outqueue.  Mostly, this just frees
1013  * things off the transmitted queue.
1014  */
1015 int sctp_outq_sack(struct sctp_outq *q, struct sctp_sackhdr *sack)
1016 {
1017 	struct sctp_association *asoc = q->asoc;
1018 	struct sctp_transport *transport;
1019 	struct sctp_chunk *tchunk = NULL;
1020 	struct list_head *lchunk, *transport_list, *pos, *temp;
1021 	sctp_sack_variable_t *frags = sack->variable;
1022 	__u32 sack_ctsn, ctsn, tsn;
1023 	__u32 highest_tsn, highest_new_tsn;
1024 	__u32 sack_a_rwnd;
1025 	unsigned outstanding;
1026 	struct sctp_transport *primary = asoc->peer.primary_path;
1027 	int count_of_newacks = 0;
1028 
1029 	/* Grab the association's destination address list. */
1030 	transport_list = &asoc->peer.transport_addr_list;
1031 
1032 	sack_ctsn = ntohl(sack->cum_tsn_ack);
1033 
1034 	/*
1035 	 * SFR-CACC algorithm:
1036 	 * On receipt of a SACK the sender SHOULD execute the
1037 	 * following statements.
1038 	 *
1039 	 * 1) If the cumulative ack in the SACK passes next tsn_at_change
1040 	 * on the current primary, the CHANGEOVER_ACTIVE flag SHOULD be
1041 	 * cleared. The CYCLING_CHANGEOVER flag SHOULD also be cleared for
1042 	 * all destinations.
1043 	 */
1044 	if (TSN_lte(primary->cacc.next_tsn_at_change, sack_ctsn)) {
1045 		primary->cacc.changeover_active = 0;
1046 		list_for_each(pos, transport_list) {
1047 			transport = list_entry(pos, struct sctp_transport,
1048 					transports);
1049 			transport->cacc.cycling_changeover = 0;
1050 		}
1051 	}
1052 
1053 	/*
1054 	 * SFR-CACC algorithm:
1055 	 * 2) If the SACK contains gap acks and the flag CHANGEOVER_ACTIVE
1056 	 * is set the receiver of the SACK MUST take the following actions:
1057 	 *
1058 	 * A) Initialize the cacc_saw_newack to 0 for all destination
1059 	 * addresses.
1060 	 */
1061 	if (sack->num_gap_ack_blocks > 0 &&
1062 	    primary->cacc.changeover_active) {
1063 		list_for_each(pos, transport_list) {
1064 			transport = list_entry(pos, struct sctp_transport,
1065 					transports);
1066 			transport->cacc.cacc_saw_newack = 0;
1067 		}
1068 	}
1069 
1070 	/* Get the highest TSN in the sack. */
1071 	highest_tsn = sack_ctsn;
1072 	if (sack->num_gap_ack_blocks)
1073 		highest_tsn +=
1074 		    ntohs(frags[ntohs(sack->num_gap_ack_blocks) - 1].gab.end);
1075 
1076 	if (TSN_lt(asoc->highest_sacked, highest_tsn)) {
1077 		highest_new_tsn = highest_tsn;
1078 		asoc->highest_sacked = highest_tsn;
1079 	} else {
1080 		highest_new_tsn = sctp_highest_new_tsn(sack, asoc);
1081 	}
1082 
1083 	/* Run through the retransmit queue.  Credit bytes received
1084 	 * and free those chunks that we can.
1085 	 */
1086 	sctp_check_transmitted(q, &q->retransmit, NULL, sack, highest_new_tsn);
1087 	sctp_mark_missing(q, &q->retransmit, NULL, highest_new_tsn, 0);
1088 
1089 	/* Run through the transmitted queue.
1090 	 * Credit bytes received and free those chunks which we can.
1091 	 *
1092 	 * This is a MASSIVE candidate for optimization.
1093 	 */
1094 	list_for_each(pos, transport_list) {
1095 		transport  = list_entry(pos, struct sctp_transport,
1096 					transports);
1097 		sctp_check_transmitted(q, &transport->transmitted,
1098 				       transport, sack, highest_new_tsn);
1099 		/*
1100 		 * SFR-CACC algorithm:
1101 		 * C) Let count_of_newacks be the number of
1102 		 * destinations for which cacc_saw_newack is set.
1103 		 */
1104 		if (transport->cacc.cacc_saw_newack)
1105 			count_of_newacks ++;
1106 	}
1107 
1108 	list_for_each(pos, transport_list) {
1109 		transport  = list_entry(pos, struct sctp_transport,
1110 					transports);
1111 		sctp_mark_missing(q, &transport->transmitted, transport,
1112 				  highest_new_tsn, count_of_newacks);
1113 	}
1114 
1115 	/* Move the Cumulative TSN Ack Point if appropriate.  */
1116 	if (TSN_lt(asoc->ctsn_ack_point, sack_ctsn))
1117 		asoc->ctsn_ack_point = sack_ctsn;
1118 
1119 	/* Update unack_data field in the assoc. */
1120 	sctp_sack_update_unack_data(asoc, sack);
1121 
1122 	ctsn = asoc->ctsn_ack_point;
1123 
1124 	/* Throw away stuff rotting on the sack queue.  */
1125 	list_for_each_safe(lchunk, temp, &q->sacked) {
1126 		tchunk = list_entry(lchunk, struct sctp_chunk,
1127 				    transmitted_list);
1128 		tsn = ntohl(tchunk->subh.data_hdr->tsn);
1129 		if (TSN_lte(tsn, ctsn))
1130 			sctp_chunk_free(tchunk);
1131 	}
1132 
1133 	/* ii) Set rwnd equal to the newly received a_rwnd minus the
1134 	 *     number of bytes still outstanding after processing the
1135 	 *     Cumulative TSN Ack and the Gap Ack Blocks.
1136 	 */
1137 
1138 	sack_a_rwnd = ntohl(sack->a_rwnd);
1139 	outstanding = q->outstanding_bytes;
1140 
1141 	if (outstanding < sack_a_rwnd)
1142 		sack_a_rwnd -= outstanding;
1143 	else
1144 		sack_a_rwnd = 0;
1145 
1146 	asoc->peer.rwnd = sack_a_rwnd;
1147 
1148 	sctp_generate_fwdtsn(q, sack_ctsn);
1149 
1150 	SCTP_DEBUG_PRINTK("%s: sack Cumulative TSN Ack is 0x%x.\n",
1151 			  __FUNCTION__, sack_ctsn);
1152 	SCTP_DEBUG_PRINTK("%s: Cumulative TSN Ack of association, "
1153 			  "%p is 0x%x. Adv peer ack point: 0x%x\n",
1154 			  __FUNCTION__, asoc, ctsn, asoc->adv_peer_ack_point);
1155 
1156 	/* See if all chunks are acked.
1157 	 * Make sure the empty queue handler will get run later.
1158 	 */
1159 	q->empty = (list_empty(&q->out_chunk_list) &&
1160 		    list_empty(&q->control_chunk_list) &&
1161 		    list_empty(&q->retransmit));
1162 	if (!q->empty)
1163 		goto finish;
1164 
1165 	list_for_each(pos, transport_list) {
1166 		transport  = list_entry(pos, struct sctp_transport,
1167 					transports);
1168 		q->empty = q->empty && list_empty(&transport->transmitted);
1169 		if (!q->empty)
1170 			goto finish;
1171 	}
1172 
1173 	SCTP_DEBUG_PRINTK("sack queue is empty.\n");
1174 finish:
1175 	return q->empty;
1176 }
1177 
1178 /* Is the outqueue empty?  */
1179 int sctp_outq_is_empty(const struct sctp_outq *q)
1180 {
1181 	return q->empty;
1182 }
1183 
1184 /********************************************************************
1185  * 2nd Level Abstractions
1186  ********************************************************************/
1187 
1188 /* Go through a transport's transmitted list or the association's retransmit
1189  * list and move chunks that are acked by the Cumulative TSN Ack to q->sacked.
1190  * The retransmit list will not have an associated transport.
1191  *
1192  * I added coherent debug information output.	--xguo
1193  *
1194  * Instead of printing 'sacked' or 'kept' for each TSN on the
1195  * transmitted_queue, we print a range: SACKED: TSN1-TSN2, TSN3, TSN4-TSN5.
1196  * KEPT TSN6-TSN7, etc.
1197  */
1198 static void sctp_check_transmitted(struct sctp_outq *q,
1199 				   struct list_head *transmitted_queue,
1200 				   struct sctp_transport *transport,
1201 				   struct sctp_sackhdr *sack,
1202 				   __u32 highest_new_tsn_in_sack)
1203 {
1204 	struct list_head *lchunk;
1205 	struct sctp_chunk *tchunk;
1206 	struct list_head tlist;
1207 	__u32 tsn;
1208 	__u32 sack_ctsn;
1209 	__u32 rtt;
1210 	__u8 restart_timer = 0;
1211 	int bytes_acked = 0;
1212 
1213 	/* These state variables are for coherent debug output. --xguo */
1214 
1215 #if SCTP_DEBUG
1216 	__u32 dbg_ack_tsn = 0;	/* An ACKed TSN range starts here... */
1217 	__u32 dbg_last_ack_tsn = 0;  /* ...and finishes here.	     */
1218 	__u32 dbg_kept_tsn = 0;	/* An un-ACKed range starts here...  */
1219 	__u32 dbg_last_kept_tsn = 0; /* ...and finishes here.	     */
1220 
1221 	/* 0 : The last TSN was ACKed.
1222 	 * 1 : The last TSN was NOT ACKed (i.e. KEPT).
1223 	 * -1: We need to initialize.
1224 	 */
1225 	int dbg_prt_state = -1;
1226 #endif /* SCTP_DEBUG */
1227 
1228 	sack_ctsn = ntohl(sack->cum_tsn_ack);
1229 
1230 	INIT_LIST_HEAD(&tlist);
1231 
1232 	/* The while loop will skip empty transmitted queues. */
1233 	while (NULL != (lchunk = sctp_list_dequeue(transmitted_queue))) {
1234 		tchunk = list_entry(lchunk, struct sctp_chunk,
1235 				    transmitted_list);
1236 
1237 		if (sctp_chunk_abandoned(tchunk)) {
1238 			/* Move the chunk to abandoned list. */
1239 			sctp_insert_list(&q->abandoned, lchunk);
1240 			continue;
1241 		}
1242 
1243 		tsn = ntohl(tchunk->subh.data_hdr->tsn);
1244 		if (sctp_acked(sack, tsn)) {
1245 			/* If this queue is the retransmit queue, the
1246 			 * retransmit timer has already reclaimed
1247 			 * the outstanding bytes for this chunk, so only
1248 			 * count bytes associated with a transport.
1249 			 */
1250 			if (transport) {
1251 				/* If this chunk is being used for RTT
1252 				 * measurement, calculate the RTT and update
1253 				 * the RTO using this value.
1254 				 *
1255 				 * 6.3.1 C5) Karn's algorithm: RTT measurements
1256 				 * MUST NOT be made using packets that were
1257 				 * retransmitted (and thus for which it is
1258 				 * ambiguous whether the reply was for the
1259 				 * first instance of the packet or a later
1260 				 * instance).
1261 				 */
1262 			   	if (!tchunk->tsn_gap_acked &&
1263 				    !tchunk->resent &&
1264 				    tchunk->rtt_in_progress) {
1265 					rtt = jiffies - tchunk->sent_at;
1266 					sctp_transport_update_rto(transport,
1267 								  rtt);
1268 				}
1269 			}
1270                         if (TSN_lte(tsn, sack_ctsn)) {
1271 				/* RFC 2960  6.3.2 Retransmission Timer Rules
1272 				 *
1273 				 * R3) Whenever a SACK is received
1274 				 * that acknowledges the DATA chunk
1275 				 * with the earliest outstanding TSN
1276 				 * for that address, restart T3-rtx
1277 				 * timer for that address with its
1278 				 * current RTO.
1279 				 */
1280 				restart_timer = 1;
1281 
1282 				if (!tchunk->tsn_gap_acked) {
1283 					tchunk->tsn_gap_acked = 1;
1284 					bytes_acked += sctp_data_size(tchunk);
1285 					/*
1286 					 * SFR-CACC algorithm:
1287 					 * 2) If the SACK contains gap acks
1288 					 * and the flag CHANGEOVER_ACTIVE is
1289 					 * set the receiver of the SACK MUST
1290 					 * take the following action:
1291 					 *
1292 					 * B) For each TSN t being acked that
1293 					 * has not been acked in any SACK so
1294 					 * far, set cacc_saw_newack to 1 for
1295 					 * the destination that the TSN was
1296 					 * sent to.
1297 					 */
1298 					if (transport &&
1299 					    sack->num_gap_ack_blocks &&
1300 					    q->asoc->peer.primary_path->cacc.
1301 					    changeover_active)
1302 						transport->cacc.cacc_saw_newack
1303 							= 1;
1304 				}
1305 
1306 				list_add_tail(&tchunk->transmitted_list,
1307 					      &q->sacked);
1308 			} else {
1309 				/* RFC2960 7.2.4, sctpimpguide-05 2.8.2
1310 				 * M2) Each time a SACK arrives reporting
1311 				 * 'Stray DATA chunk(s)' record the highest TSN
1312 				 * reported as newly acknowledged, call this
1313 				 * value 'HighestTSNinSack'. A newly
1314 				 * acknowledged DATA chunk is one not
1315 				 * previously acknowledged in a SACK.
1316 				 *
1317 				 * When the SCTP sender of data receives a SACK
1318 				 * chunk that acknowledges, for the first time,
1319 				 * the receipt of a DATA chunk, all the still
1320 				 * unacknowledged DATA chunks whose TSN is
1321 				 * older than that newly acknowledged DATA
1322 				 * chunk, are qualified as 'Stray DATA chunks'.
1323 				 */
1324 				if (!tchunk->tsn_gap_acked) {
1325 					tchunk->tsn_gap_acked = 1;
1326 					bytes_acked += sctp_data_size(tchunk);
1327 				}
1328 				list_add_tail(lchunk, &tlist);
1329 			}
1330 
1331 #if SCTP_DEBUG
1332 			switch (dbg_prt_state) {
1333 			case 0:	/* last TSN was ACKed */
1334 				if (dbg_last_ack_tsn + 1 == tsn) {
1335 					/* This TSN belongs to the
1336 					 * current ACK range.
1337 					 */
1338 					break;
1339 				}
1340 
1341 				if (dbg_last_ack_tsn != dbg_ack_tsn) {
1342 					/* Display the end of the
1343 					 * current range.
1344 					 */
1345 					SCTP_DEBUG_PRINTK("-%08x",
1346 							  dbg_last_ack_tsn);
1347 				}
1348 
1349 				/* Start a new range.  */
1350 				SCTP_DEBUG_PRINTK(",%08x", tsn);
1351 				dbg_ack_tsn = tsn;
1352 				break;
1353 
1354 			case 1:	/* The last TSN was NOT ACKed. */
1355 				if (dbg_last_kept_tsn != dbg_kept_tsn) {
1356 					/* Display the end of current range. */
1357 					SCTP_DEBUG_PRINTK("-%08x",
1358 							  dbg_last_kept_tsn);
1359 				}
1360 
1361 				SCTP_DEBUG_PRINTK("\n");
1362 
1363 				/* FALL THROUGH... */
1364 			default:
1365 				/* This is the first-ever TSN we examined.  */
1366 				/* Start a new range of ACK-ed TSNs.  */
1367 				SCTP_DEBUG_PRINTK("ACKed: %08x", tsn);
1368 				dbg_prt_state = 0;
1369 				dbg_ack_tsn = tsn;
1370 			};
1371 
1372 			dbg_last_ack_tsn = tsn;
1373 #endif /* SCTP_DEBUG */
1374 
1375 		} else {
1376 			if (tchunk->tsn_gap_acked) {
1377 				SCTP_DEBUG_PRINTK("%s: Receiver reneged on "
1378 						  "data TSN: 0x%x\n",
1379 						  __FUNCTION__,
1380 						  tsn);
1381 				tchunk->tsn_gap_acked = 0;
1382 
1383 				bytes_acked -= sctp_data_size(tchunk);
1384 
1385 				/* RFC 2960 6.3.2 Retransmission Timer Rules
1386 				 *
1387 				 * R4) Whenever a SACK is received missing a
1388 				 * TSN that was previously acknowledged via a
1389 				 * Gap Ack Block, start T3-rtx for the
1390 				 * destination address to which the DATA
1391 				 * chunk was originally
1392 				 * transmitted if it is not already running.
1393 				 */
1394 				restart_timer = 1;
1395 			}
1396 
1397 			list_add_tail(lchunk, &tlist);
1398 
1399 #if SCTP_DEBUG
1400 			/* See the above comments on ACK-ed TSNs. */
1401 			switch (dbg_prt_state) {
1402 			case 1:
1403 				if (dbg_last_kept_tsn + 1 == tsn)
1404 					break;
1405 
1406 				if (dbg_last_kept_tsn != dbg_kept_tsn)
1407 					SCTP_DEBUG_PRINTK("-%08x",
1408 							  dbg_last_kept_tsn);
1409 
1410 				SCTP_DEBUG_PRINTK(",%08x", tsn);
1411 				dbg_kept_tsn = tsn;
1412 				break;
1413 
1414 			case 0:
1415 				if (dbg_last_ack_tsn != dbg_ack_tsn)
1416 					SCTP_DEBUG_PRINTK("-%08x",
1417 							  dbg_last_ack_tsn);
1418 				SCTP_DEBUG_PRINTK("\n");
1419 
1420 				/* FALL THROUGH... */
1421 			default:
1422 				SCTP_DEBUG_PRINTK("KEPT: %08x",tsn);
1423 				dbg_prt_state = 1;
1424 				dbg_kept_tsn = tsn;
1425 			};
1426 
1427 			dbg_last_kept_tsn = tsn;
1428 #endif /* SCTP_DEBUG */
1429 		}
1430 	}
1431 
1432 #if SCTP_DEBUG
1433 	/* Finish off the last range, displaying its ending TSN.  */
1434 	switch (dbg_prt_state) {
1435 	case 0:
1436 		if (dbg_last_ack_tsn != dbg_ack_tsn) {
1437 			SCTP_DEBUG_PRINTK("-%08x\n", dbg_last_ack_tsn);
1438 		} else {
1439 			SCTP_DEBUG_PRINTK("\n");
1440 		}
1441 	break;
1442 
1443 	case 1:
1444 		if (dbg_last_kept_tsn != dbg_kept_tsn) {
1445 			SCTP_DEBUG_PRINTK("-%08x\n", dbg_last_kept_tsn);
1446 		} else {
1447 			SCTP_DEBUG_PRINTK("\n");
1448 		}
1449 	};
1450 #endif /* SCTP_DEBUG */
1451 	if (transport) {
1452 		if (bytes_acked) {
1453 			/* 8.2. When an outstanding TSN is acknowledged,
1454 			 * the endpoint shall clear the error counter of
1455 			 * the destination transport address to which the
1456 			 * DATA chunk was last sent.
1457 			 * The association's overall error counter is
1458 			 * also cleared.
1459 			 */
1460 			transport->error_count = 0;
1461 			transport->asoc->overall_error_count = 0;
1462 
1463 			/* Mark the destination transport address as
1464 			 * active if it is not so marked.
1465 			 */
1466 			if (transport->state == SCTP_INACTIVE) {
1467 				sctp_assoc_control_transport(
1468 					transport->asoc,
1469 					transport,
1470 					SCTP_TRANSPORT_UP,
1471 					SCTP_RECEIVED_SACK);
1472 			}
1473 
1474 			sctp_transport_raise_cwnd(transport, sack_ctsn,
1475 						  bytes_acked);
1476 
1477 			transport->flight_size -= bytes_acked;
1478 			q->outstanding_bytes -= bytes_acked;
1479 		} else {
1480 			/* RFC 2960 6.1, sctpimpguide-06 2.15.2
1481 			 * When a sender is doing zero window probing, it
1482 			 * should not timeout the association if it continues
1483 			 * to receive new packets from the receiver. The
1484 			 * reason is that the receiver MAY keep its window
1485 			 * closed for an indefinite time.
1486 			 * A sender is doing zero window probing when the
1487 			 * receiver's advertised window is zero, and there is
1488 			 * only one data chunk in flight to the receiver.
1489 			 */
1490 			if (!q->asoc->peer.rwnd &&
1491 			    !list_empty(&tlist) &&
1492 			    (sack_ctsn+2 == q->asoc->next_tsn)) {
1493 				SCTP_DEBUG_PRINTK("%s: SACK received for zero "
1494 						  "window probe: %u\n",
1495 						  __FUNCTION__, sack_ctsn);
1496 				q->asoc->overall_error_count = 0;
1497 				transport->error_count = 0;
1498 			}
1499 		}
1500 
1501 		/* RFC 2960 6.3.2 Retransmission Timer Rules
1502 		 *
1503 		 * R2) Whenever all outstanding data sent to an address have
1504 		 * been acknowledged, turn off the T3-rtx timer of that
1505 		 * address.
1506 		 */
1507 		if (!transport->flight_size) {
1508 			if (timer_pending(&transport->T3_rtx_timer) &&
1509 			    del_timer(&transport->T3_rtx_timer)) {
1510 				sctp_transport_put(transport);
1511 			}
1512 		} else if (restart_timer) {
1513 			if (!mod_timer(&transport->T3_rtx_timer,
1514 				       jiffies + transport->rto))
1515 				sctp_transport_hold(transport);
1516 		}
1517 	}
1518 
1519 	list_splice(&tlist, transmitted_queue);
1520 }
1521 
1522 /* Mark chunks as missing and consequently may get retransmitted. */
1523 static void sctp_mark_missing(struct sctp_outq *q,
1524 			      struct list_head *transmitted_queue,
1525 			      struct sctp_transport *transport,
1526 			      __u32 highest_new_tsn_in_sack,
1527 			      int count_of_newacks)
1528 {
1529 	struct sctp_chunk *chunk;
1530 	struct list_head *pos;
1531 	__u32 tsn;
1532 	char do_fast_retransmit = 0;
1533 	struct sctp_transport *primary = q->asoc->peer.primary_path;
1534 
1535 	list_for_each(pos, transmitted_queue) {
1536 
1537 		chunk = list_entry(pos, struct sctp_chunk, transmitted_list);
1538 		tsn = ntohl(chunk->subh.data_hdr->tsn);
1539 
1540 		/* RFC 2960 7.2.4, sctpimpguide-05 2.8.2 M3) Examine all
1541 		 * 'Unacknowledged TSN's', if the TSN number of an
1542 		 * 'Unacknowledged TSN' is smaller than the 'HighestTSNinSack'
1543 		 * value, increment the 'TSN.Missing.Report' count on that
1544 		 * chunk if it has NOT been fast retransmitted or marked for
1545 		 * fast retransmit already.
1546 		 */
1547 		if (!chunk->fast_retransmit &&
1548 		    !chunk->tsn_gap_acked &&
1549 		    TSN_lt(tsn, highest_new_tsn_in_sack)) {
1550 
1551 			/* SFR-CACC may require us to skip marking
1552 			 * this chunk as missing.
1553 			 */
1554 			if (!transport || !sctp_cacc_skip(primary, transport,
1555 					    count_of_newacks, tsn)) {
1556 				chunk->tsn_missing_report++;
1557 
1558 				SCTP_DEBUG_PRINTK(
1559 					"%s: TSN 0x%x missing counter: %d\n",
1560 					__FUNCTION__, tsn,
1561 					chunk->tsn_missing_report);
1562 			}
1563 		}
1564 		/*
1565 		 * M4) If any DATA chunk is found to have a
1566 		 * 'TSN.Missing.Report'
1567 		 * value larger than or equal to 3, mark that chunk for
1568 		 * retransmission and start the fast retransmit procedure.
1569 		 */
1570 
1571 		if (chunk->tsn_missing_report >= 3) {
1572 			chunk->fast_retransmit = 1;
1573 			do_fast_retransmit = 1;
1574 		}
1575 	}
1576 
1577 	if (transport) {
1578 		if (do_fast_retransmit)
1579 			sctp_retransmit(q, transport, SCTP_RTXR_FAST_RTX);
1580 
1581 		SCTP_DEBUG_PRINTK("%s: transport: %p, cwnd: %d, "
1582 				  "ssthresh: %d, flight_size: %d, pba: %d\n",
1583 				  __FUNCTION__, transport, transport->cwnd,
1584 			  	  transport->ssthresh, transport->flight_size,
1585 				  transport->partial_bytes_acked);
1586 	}
1587 }
1588 
1589 /* Is the given TSN acked by this packet?  */
1590 static int sctp_acked(struct sctp_sackhdr *sack, __u32 tsn)
1591 {
1592 	int i;
1593 	sctp_sack_variable_t *frags;
1594 	__u16 gap;
1595 	__u32 ctsn = ntohl(sack->cum_tsn_ack);
1596 
1597         if (TSN_lte(tsn, ctsn))
1598 		goto pass;
1599 
1600 	/* 3.3.4 Selective Acknowledgement (SACK) (3):
1601 	 *
1602 	 * Gap Ack Blocks:
1603 	 *  These fields contain the Gap Ack Blocks. They are repeated
1604 	 *  for each Gap Ack Block up to the number of Gap Ack Blocks
1605 	 *  defined in the Number of Gap Ack Blocks field. All DATA
1606 	 *  chunks with TSNs greater than or equal to (Cumulative TSN
1607 	 *  Ack + Gap Ack Block Start) and less than or equal to
1608 	 *  (Cumulative TSN Ack + Gap Ack Block End) of each Gap Ack
1609 	 *  Block are assumed to have been received correctly.
1610 	 */
1611 
1612 	frags = sack->variable;
1613 	gap = tsn - ctsn;
1614 	for (i = 0; i < ntohs(sack->num_gap_ack_blocks); ++i) {
1615 		if (TSN_lte(ntohs(frags[i].gab.start), gap) &&
1616 		    TSN_lte(gap, ntohs(frags[i].gab.end)))
1617 			goto pass;
1618 	}
1619 
1620 	return 0;
1621 pass:
1622 	return 1;
1623 }
1624 
1625 static inline int sctp_get_skip_pos(struct sctp_fwdtsn_skip *skiplist,
1626 				    int nskips, __u16 stream)
1627 {
1628 	int i;
1629 
1630 	for (i = 0; i < nskips; i++) {
1631 		if (skiplist[i].stream == stream)
1632 			return i;
1633 	}
1634 	return i;
1635 }
1636 
1637 /* Create and add a fwdtsn chunk to the outq's control queue if needed. */
1638 static void sctp_generate_fwdtsn(struct sctp_outq *q, __u32 ctsn)
1639 {
1640 	struct sctp_association *asoc = q->asoc;
1641 	struct sctp_chunk *ftsn_chunk = NULL;
1642 	struct sctp_fwdtsn_skip ftsn_skip_arr[10];
1643 	int nskips = 0;
1644 	int skip_pos = 0;
1645 	__u32 tsn;
1646 	struct sctp_chunk *chunk;
1647 	struct list_head *lchunk, *temp;
1648 
1649 	/* PR-SCTP C1) Let SackCumAck be the Cumulative TSN ACK carried in the
1650 	 * received SACK.
1651 	 *
1652 	 * If (Advanced.Peer.Ack.Point < SackCumAck), then update
1653 	 * Advanced.Peer.Ack.Point to be equal to SackCumAck.
1654 	 */
1655 	if (TSN_lt(asoc->adv_peer_ack_point, ctsn))
1656 		asoc->adv_peer_ack_point = ctsn;
1657 
1658 	/* PR-SCTP C2) Try to further advance the "Advanced.Peer.Ack.Point"
1659 	 * locally, that is, to move "Advanced.Peer.Ack.Point" up as long as
1660 	 * the chunk next in the out-queue space is marked as "abandoned" as
1661 	 * shown in the following example:
1662 	 *
1663 	 * Assuming that a SACK arrived with the Cumulative TSN ACK 102
1664 	 * and the Advanced.Peer.Ack.Point is updated to this value:
1665 	 *
1666 	 *   out-queue at the end of  ==>   out-queue after Adv.Ack.Point
1667 	 *   normal SACK processing           local advancement
1668 	 *                ...                           ...
1669 	 *   Adv.Ack.Pt-> 102 acked                     102 acked
1670 	 *                103 abandoned                 103 abandoned
1671 	 *                104 abandoned     Adv.Ack.P-> 104 abandoned
1672 	 *                105                           105
1673 	 *                106 acked                     106 acked
1674 	 *                ...                           ...
1675 	 *
1676 	 * In this example, the data sender successfully advanced the
1677 	 * "Advanced.Peer.Ack.Point" from 102 to 104 locally.
1678 	 */
1679 	list_for_each_safe(lchunk, temp, &q->abandoned) {
1680 		chunk = list_entry(lchunk, struct sctp_chunk,
1681 					transmitted_list);
1682 		tsn = ntohl(chunk->subh.data_hdr->tsn);
1683 
1684 		/* Remove any chunks in the abandoned queue that are acked by
1685 		 * the ctsn.
1686 		 */
1687 		if (TSN_lte(tsn, ctsn)) {
1688 			list_del_init(lchunk);
1689 			if (!chunk->tsn_gap_acked) {
1690 				chunk->transport->flight_size -=
1691 					sctp_data_size(chunk);
1692 				q->outstanding_bytes -= sctp_data_size(chunk);
1693 			}
1694 			sctp_chunk_free(chunk);
1695 		} else {
1696 			if (TSN_lte(tsn, asoc->adv_peer_ack_point+1)) {
1697 				asoc->adv_peer_ack_point = tsn;
1698 				if (chunk->chunk_hdr->flags &
1699 					 SCTP_DATA_UNORDERED)
1700 					continue;
1701 				skip_pos = sctp_get_skip_pos(&ftsn_skip_arr[0],
1702 						nskips,
1703 						chunk->subh.data_hdr->stream);
1704 				ftsn_skip_arr[skip_pos].stream =
1705 					chunk->subh.data_hdr->stream;
1706 				ftsn_skip_arr[skip_pos].ssn =
1707 					 chunk->subh.data_hdr->ssn;
1708 				if (skip_pos == nskips)
1709 					nskips++;
1710 				if (nskips == 10)
1711 					break;
1712 			} else
1713 				break;
1714 		}
1715 	}
1716 
1717 	/* PR-SCTP C3) If, after step C1 and C2, the "Advanced.Peer.Ack.Point"
1718 	 * is greater than the Cumulative TSN ACK carried in the received
1719 	 * SACK, the data sender MUST send the data receiver a FORWARD TSN
1720 	 * chunk containing the latest value of the
1721 	 * "Advanced.Peer.Ack.Point".
1722 	 *
1723 	 * C4) For each "abandoned" TSN the sender of the FORWARD TSN SHOULD
1724 	 * list each stream and sequence number in the forwarded TSN. This
1725 	 * information will enable the receiver to easily find any
1726 	 * stranded TSN's waiting on stream reorder queues. Each stream
1727 	 * SHOULD only be reported once; this means that if multiple
1728 	 * abandoned messages occur in the same stream then only the
1729 	 * highest abandoned stream sequence number is reported. If the
1730 	 * total size of the FORWARD TSN does NOT fit in a single MTU then
1731 	 * the sender of the FORWARD TSN SHOULD lower the
1732 	 * Advanced.Peer.Ack.Point to the last TSN that will fit in a
1733 	 * single MTU.
1734 	 */
1735 	if (asoc->adv_peer_ack_point > ctsn)
1736 		ftsn_chunk = sctp_make_fwdtsn(asoc, asoc->adv_peer_ack_point,
1737 					      nskips, &ftsn_skip_arr[0]);
1738 
1739 	if (ftsn_chunk) {
1740 		list_add_tail(&ftsn_chunk->list, &q->control_chunk_list);
1741 		SCTP_INC_STATS(SCTP_MIB_OUTCTRLCHUNKS);
1742 	}
1743 }
1744