xref: /dragonfly/sys/netinet/tcp_sack.c (revision 63f17add)
1 /*
2  * Copyright (c) 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 2003, 2004 The DragonFly Project.  All rights reserved.
4  *
5  * This code is derived from software contributed to The DragonFly Project
6  * by Jeffrey M. Hsu.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of The DragonFly Project nor the names of its
17  *    contributors may be used to endorse or promote products derived
18  *    from this software without specific, prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  * $DragonFly: src/sys/netinet/tcp_sack.c,v 1.8 2008/08/15 21:37:16 nth Exp $
34  */
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/kernel.h>
39 #include <sys/malloc.h>
40 #include <sys/queue.h>
41 #include <sys/thread.h>
42 #include <sys/types.h>
43 #include <sys/socket.h>
44 #include <sys/socketvar.h>
45 #include <sys/resourcevar.h>
46 
47 #include <net/if.h>
48 
49 #include <netinet/in.h>
50 #include <netinet/in_systm.h>
51 #include <netinet/ip.h>
52 #include <netinet/in_var.h>
53 #include <netinet/in_pcb.h>
54 #include <netinet/ip_var.h>
55 #include <netinet/tcp.h>
56 #include <netinet/tcp_seq.h>
57 #include <netinet/tcp_var.h>
58 
59 /*
60  * Implemented:
61  *
62  * RFC 2018
63  * RFC 2883
64  * RFC 3517
65  * RFC 6675
66  */
67 
68 struct sackblock {
69 	tcp_seq			sblk_start;
70 	tcp_seq			sblk_end;
71 	TAILQ_ENTRY(sackblock)	sblk_list;
72 };
73 
74 #define	MAXSAVEDBLOCKS	8			/* per connection limit */
75 
76 static int insert_block(struct scoreboard *scb,
77 			const struct raw_sackblock *raw_sb, boolean_t *update);
78 
79 static MALLOC_DEFINE(M_SACKBLOCK, "sblk", "sackblock struct");
80 
81 /*
82  * Per-tcpcb initialization.
83  */
84 void
tcp_sack_tcpcb_init(struct tcpcb * tp)85 tcp_sack_tcpcb_init(struct tcpcb *tp)
86 {
87 	struct scoreboard *scb = &tp->scb;
88 
89 	scb->nblocks = 0;
90 	TAILQ_INIT(&scb->sackblocks);
91 	scb->lastfound = NULL;
92 }
93 
94 /*
95  * Find the SACK block containing or immediately preceding "seq".
96  * The boolean result indicates whether the sequence is actually
97  * contained in the SACK block.
98  */
99 static boolean_t
sack_block_lookup(struct scoreboard * scb,tcp_seq seq,struct sackblock ** sb)100 sack_block_lookup(struct scoreboard *scb, tcp_seq seq, struct sackblock **sb)
101 {
102 	static struct krate sackkrate = { .freq = 1 };
103 	struct sackblock *hint = scb->lastfound;
104 	struct sackblock *cur, *last, *prev;
105 
106 	if (TAILQ_EMPTY(&scb->sackblocks)) {
107 		*sb = NULL;
108 		return FALSE;
109 	}
110 
111 	if (hint == NULL) {
112 		/* No hint.  Search from start to end. */
113 		cur = TAILQ_FIRST(&scb->sackblocks);
114 		last = NULL;
115 		prev = TAILQ_LAST(&scb->sackblocks, sackblock_list);
116 	} else  {
117 		if (SEQ_GEQ(seq, hint->sblk_start)) {
118 			/* Search from hint to end of list. */
119 			cur = hint;
120 			last = NULL;
121 			prev = TAILQ_LAST(&scb->sackblocks, sackblock_list);
122 		} else {
123 			/* Search from front of list to hint. */
124 			cur = TAILQ_FIRST(&scb->sackblocks);
125 			last = hint;
126 			prev = TAILQ_PREV(hint, sackblock_list, sblk_list);
127 		}
128 	}
129 
130 	do {
131 		/*
132 		 * Ensure we can't crash if the list really blows up due to
133 		 * delta sign wraps when comparing seq against sblk_start vs
134 		 * sblk_end.
135 		 */
136 		if (cur == NULL) {
137 			krateprintf(&sackkrate,
138 				    "tcp_sack: fatal corrupt seq\n");
139 			*sb = NULL;
140 			return FALSE;
141 		}
142 
143 		/*
144 		 * Check completion
145 		 */
146 		if (SEQ_GT(cur->sblk_end, seq)) {
147 			if (SEQ_GEQ(seq, cur->sblk_start)) {
148 				*sb = scb->lastfound = cur;
149 				return TRUE;
150 			} else {
151 				*sb = scb->lastfound =
152 				    TAILQ_PREV(cur, sackblock_list, sblk_list);
153 				return FALSE;
154 			}
155 		}
156 
157 		/*
158 		 * seq is greater than sblk_end, nominally proceed to the
159 		 * next block.
160 		 *
161 		 * It is possible for an overflow to cause the comparison
162 		 * between seq an sblk_start vs sblk_end to make it appear
163 		 * that seq is less than sblk_start and also greater than
164 		 * sblk_end.  If we allow the case to fall through we can
165 		 * end up with cur == NULL on the next loop.
166 		 */
167 		if (SEQ_LT(seq, cur->sblk_start)) {
168 			krateprintf(&sackkrate,
169 				    "tcp_sack: corrupt seq "
170 				    "0x%08x vs 0x%08x-0x%08x\n",
171 				    seq, cur->sblk_start, cur->sblk_end);
172 			if (SEQ_GEQ(seq, cur->sblk_start)) {
173 				*sb = scb->lastfound = cur;
174 				return TRUE;
175 			} else {
176 				*sb = scb->lastfound =
177 				    TAILQ_PREV(cur, sackblock_list, sblk_list);
178 				return FALSE;
179 			}
180 		}
181 		cur = TAILQ_NEXT(cur, sblk_list);
182 	} while (cur != last);
183 
184 	*sb = scb->lastfound = prev;
185 	return FALSE;
186 }
187 
188 /*
189  * Allocate a SACK block.
190  */
191 static __inline struct sackblock *
alloc_sackblock(struct scoreboard * scb,const struct raw_sackblock * raw_sb)192 alloc_sackblock(struct scoreboard *scb, const struct raw_sackblock *raw_sb)
193 {
194 	struct sackblock *sb;
195 
196 	if (scb->freecache != NULL) {
197 		sb = scb->freecache;
198 		scb->freecache = NULL;
199 		tcpstat.tcps_sacksbfast++;
200 	} else {
201 		sb = kmalloc(sizeof(struct sackblock), M_SACKBLOCK, M_NOWAIT);
202 		if (sb == NULL) {
203 			tcpstat.tcps_sacksbfailed++;
204 			return NULL;
205 		}
206 	}
207 	sb->sblk_start = raw_sb->rblk_start;
208 	sb->sblk_end = raw_sb->rblk_end;
209 	return sb;
210 }
211 
212 static __inline struct sackblock *
alloc_sackblock_limit(struct scoreboard * scb,const struct raw_sackblock * raw_sb)213 alloc_sackblock_limit(struct scoreboard *scb,
214     const struct raw_sackblock *raw_sb)
215 {
216 	if (scb->nblocks == MAXSAVEDBLOCKS) {
217 		/*
218 		 * Should try to kick out older blocks XXX JH
219 		 * May be able to coalesce with existing block.
220 		 * Or, go other way and free all blocks if we hit
221 		 * this limit.
222 		 */
223 		tcpstat.tcps_sacksboverflow++;
224 		return NULL;
225 	}
226 	return alloc_sackblock(scb, raw_sb);
227 }
228 
229 /*
230  * Free a SACK block.
231  */
232 static __inline void
free_sackblock(struct scoreboard * scb,struct sackblock * s)233 free_sackblock(struct scoreboard *scb, struct sackblock *s)
234 {
235 	if (scb->freecache == NULL) {
236 		/* YYY Maybe use the latest freed block? */
237 		scb->freecache = s;
238 		return;
239 	}
240 	kfree(s, M_SACKBLOCK);
241 }
242 
243 /*
244  * Free up SACK blocks for data that's been acked.
245  */
246 static void
tcp_sack_ack_blocks(struct tcpcb * tp,tcp_seq th_ack)247 tcp_sack_ack_blocks(struct tcpcb *tp, tcp_seq th_ack)
248 {
249 	struct scoreboard *scb = &tp->scb;
250 	struct sackblock *sb, *nb;
251 
252 	sb = TAILQ_FIRST(&scb->sackblocks);
253 	while (sb && SEQ_LEQ(sb->sblk_end, th_ack)) {
254 		nb = TAILQ_NEXT(sb, sblk_list);
255 		if (scb->lastfound == sb)
256 			scb->lastfound = NULL;
257 		TAILQ_REMOVE(&scb->sackblocks, sb, sblk_list);
258 		free_sackblock(scb, sb);
259 		--scb->nblocks;
260 		KASSERT(scb->nblocks >= 0,
261 		    ("SACK block count underflow: %d < 0", scb->nblocks));
262 		sb = nb;
263 	}
264 	if (sb && SEQ_GEQ(th_ack, sb->sblk_start)) {
265 		/* Other side reneged? XXX */
266 		tcpstat.tcps_sackrenege++;
267 		tcp_sack_discard(tp);
268 	}
269 }
270 
271 /*
272  * Delete and free SACK blocks saved in scoreboard.
273  */
274 static void
tcp_sack_cleanup(struct scoreboard * scb)275 tcp_sack_cleanup(struct scoreboard *scb)
276 {
277 	struct sackblock *sb, *nb;
278 
279 	TAILQ_FOREACH_MUTABLE(sb, &scb->sackblocks, sblk_list, nb) {
280 		free_sackblock(scb, sb);
281 		--scb->nblocks;
282 	}
283 	KASSERT(scb->nblocks == 0,
284 	    ("SACK block %d count not zero", scb->nblocks));
285 	TAILQ_INIT(&scb->sackblocks);
286 	scb->lastfound = NULL;
287 }
288 
289 /*
290  * Discard SACK scoreboard, HighRxt, RescueRxt and LostSeq.
291  */
292 void
tcp_sack_discard(struct tcpcb * tp)293 tcp_sack_discard(struct tcpcb *tp)
294 {
295 	tcp_sack_cleanup(&tp->scb);
296 	tp->rexmt_high = tp->snd_una;
297 	tp->sack_flags &= ~TSACK_F_SACKRESCUED;
298 	tp->scb.lostseq = tp->snd_una;
299 }
300 
301 /*
302  * Delete and free SACK blocks saved in scoreboard.
303  * Delete the one slot block cache.
304  */
305 void
tcp_sack_destroy(struct scoreboard * scb)306 tcp_sack_destroy(struct scoreboard *scb)
307 {
308 	tcp_sack_cleanup(scb);
309 	if (scb->freecache != NULL) {
310 		kfree(scb->freecache, M_SACKBLOCK);
311 		scb->freecache = NULL;
312 	}
313 }
314 
315 /*
316  * Cleanup the reported SACK block information
317  */
318 void
tcp_sack_report_cleanup(struct tcpcb * tp)319 tcp_sack_report_cleanup(struct tcpcb *tp)
320 {
321 	tp->sack_flags &=
322 	    ~(TSACK_F_DUPSEG | TSACK_F_ENCLOSESEG | TSACK_F_SACKLEFT);
323 	tp->reportblk.rblk_start = tp->reportblk.rblk_end;
324 }
325 
326 /*
327  * Whether SACK report is needed or not
328  */
329 boolean_t
tcp_sack_report_needed(const struct tcpcb * tp)330 tcp_sack_report_needed(const struct tcpcb *tp)
331 {
332 	if ((tp->sack_flags &
333 	     (TSACK_F_DUPSEG | TSACK_F_ENCLOSESEG | TSACK_F_SACKLEFT)) ||
334 	    tp->reportblk.rblk_start != tp->reportblk.rblk_end)
335 		return TRUE;
336 	else
337 		return FALSE;
338 }
339 
340 /*
341  * Returns	0 if not D-SACK block,
342  *		1 if D-SACK,
343  *		2 if duplicate of out-of-order D-SACK block.
344  */
345 int
tcp_sack_ndsack_blocks(const struct raw_sackblock * blocks,const int numblocks,tcp_seq snd_una)346 tcp_sack_ndsack_blocks(const struct raw_sackblock *blocks, const int numblocks,
347     tcp_seq snd_una)
348 {
349 	if (numblocks == 0)
350 		return 0;
351 
352 	if (SEQ_LT(blocks[0].rblk_start, snd_una))
353 		return 1;
354 
355 	/* block 0 inside block 1 */
356 	if (numblocks > 1 &&
357 	    SEQ_GEQ(blocks[0].rblk_start, blocks[1].rblk_start) &&
358 	    SEQ_LEQ(blocks[0].rblk_end, blocks[1].rblk_end))
359 		return 2;
360 
361 	return 0;
362 }
363 
364 /*
365  * Update scoreboard on new incoming ACK.
366  */
367 static void
tcp_sack_add_blocks(struct tcpcb * tp,struct tcpopt * to)368 tcp_sack_add_blocks(struct tcpcb *tp, struct tcpopt *to)
369 {
370 	const int numblocks = to->to_nsackblocks;
371 	struct raw_sackblock *blocks = to->to_sackblocks;
372 	struct scoreboard *scb = &tp->scb;
373 	int startblock, i;
374 
375 	if (tcp_sack_ndsack_blocks(blocks, numblocks, tp->snd_una) > 0)
376 		startblock = 1;
377 	else
378 		startblock = 0;
379 
380 	to->to_flags |= TOF_SACK_REDUNDANT;
381 	for (i = startblock; i < numblocks; i++) {
382 		struct raw_sackblock *newsackblock = &blocks[i];
383 		boolean_t update;
384 		int error;
385 
386 		/* Guard against ACK reordering */
387 		if (SEQ_LEQ(newsackblock->rblk_start, tp->snd_una))
388 			continue;
389 
390 		/* Don't accept bad SACK blocks */
391 		if (SEQ_GT(newsackblock->rblk_end, tp->snd_max)) {
392 			tcpstat.tcps_rcvbadsackopt++;
393 			break;		/* skip all other blocks */
394 		}
395 		tcpstat.tcps_sacksbupdate++;
396 
397 		error = insert_block(scb, newsackblock, &update);
398 		if (update)
399 			to->to_flags &= ~TOF_SACK_REDUNDANT;
400 		if (error)
401 			break;
402 	}
403 }
404 
405 void
tcp_sack_update_scoreboard(struct tcpcb * tp,struct tcpopt * to)406 tcp_sack_update_scoreboard(struct tcpcb *tp, struct tcpopt *to)
407 {
408 	struct scoreboard *scb = &tp->scb;
409 	int rexmt_high_update = 0;
410 
411 	tcp_sack_ack_blocks(tp, tp->snd_una);
412 	tcp_sack_add_blocks(tp, to);
413 	tcp_sack_update_lostseq(scb, tp->snd_una, tp->t_maxseg,
414 	    tp->t_rxtthresh);
415 	if (SEQ_LT(tp->rexmt_high, tp->snd_una)) {
416 		tp->rexmt_high = tp->snd_una;
417 		rexmt_high_update = 1;
418 	}
419 	if (tp->sack_flags & TSACK_F_SACKRESCUED) {
420 		if (SEQ_LEQ(tp->rexmt_rescue, tp->snd_una)) {
421 			tp->sack_flags &= ~TSACK_F_SACKRESCUED;
422 		} else if (tcp_aggressive_rescuesack && rexmt_high_update &&
423 		    SEQ_LT(tp->rexmt_rescue, tp->rexmt_high)) {
424 			/* Drag RescueRxt along with HighRxt */
425 			tp->rexmt_rescue = tp->rexmt_high;
426 		}
427 	}
428 }
429 
430 /*
431  * Insert SACK block into sender's scoreboard.
432  */
433 static int
insert_block(struct scoreboard * scb,const struct raw_sackblock * raw_sb,boolean_t * update)434 insert_block(struct scoreboard *scb, const struct raw_sackblock *raw_sb,
435     boolean_t *update)
436 {
437 	struct sackblock *sb, *workingblock;
438 	boolean_t overlap_front;
439 
440 	*update = TRUE;
441 	if (TAILQ_EMPTY(&scb->sackblocks)) {
442 		struct sackblock *newblock;
443 
444 		KASSERT(scb->nblocks == 0, ("emply scb w/ blocks"));
445 
446 		newblock = alloc_sackblock(scb, raw_sb);
447 		if (newblock == NULL)
448 			return ENOMEM;
449 		TAILQ_INSERT_HEAD(&scb->sackblocks, newblock, sblk_list);
450 		scb->nblocks = 1;
451 		return 0;
452 	}
453 
454 	KASSERT(scb->nblocks > 0, ("insert_block() called w/ no blocks"));
455 	KASSERT(scb->nblocks <= MAXSAVEDBLOCKS,
456 	    ("too many SACK blocks %d", scb->nblocks));
457 
458 	overlap_front = sack_block_lookup(scb, raw_sb->rblk_start, &sb);
459 
460 	if (sb == NULL) {
461 		workingblock = alloc_sackblock_limit(scb, raw_sb);
462 		if (workingblock == NULL)
463 			return ENOMEM;
464 		TAILQ_INSERT_HEAD(&scb->sackblocks, workingblock, sblk_list);
465 		++scb->nblocks;
466 	} else {
467 		if (overlap_front || sb->sblk_end == raw_sb->rblk_start) {
468 			tcpstat.tcps_sacksbreused++;
469 
470 			/* Extend old block */
471 			workingblock = sb;
472 			if (SEQ_GT(raw_sb->rblk_end, sb->sblk_end)) {
473 				sb->sblk_end = raw_sb->rblk_end;
474 			} else {
475 				/* Exact match, nothing to consolidate */
476 				*update = FALSE;
477 				return 0;
478 			}
479 		} else {
480 			workingblock = alloc_sackblock_limit(scb, raw_sb);
481 			if (workingblock == NULL)
482 				return ENOMEM;
483 			TAILQ_INSERT_AFTER(&scb->sackblocks, sb, workingblock,
484 			    sblk_list);
485 			++scb->nblocks;
486 		}
487 	}
488 
489 	/* Consolidate right-hand side. */
490 	sb = TAILQ_NEXT(workingblock, sblk_list);
491 	while (sb != NULL &&
492 	    SEQ_GEQ(workingblock->sblk_end, sb->sblk_end)) {
493 		struct sackblock *nextblock;
494 
495 		nextblock = TAILQ_NEXT(sb, sblk_list);
496 		if (scb->lastfound == sb)
497 			scb->lastfound = NULL;
498 		/* Remove completely overlapped block */
499 		TAILQ_REMOVE(&scb->sackblocks, sb, sblk_list);
500 		free_sackblock(scb, sb);
501 		--scb->nblocks;
502 		KASSERT(scb->nblocks > 0,
503 		    ("removed overlapped block: %d blocks left", scb->nblocks));
504 		sb = nextblock;
505 	}
506 	if (sb != NULL &&
507 	    SEQ_GEQ(workingblock->sblk_end, sb->sblk_start)) {
508 		/* Extend new block to cover partially overlapped old block. */
509 		workingblock->sblk_end = sb->sblk_end;
510 		if (scb->lastfound == sb)
511 			scb->lastfound = NULL;
512 		TAILQ_REMOVE(&scb->sackblocks, sb, sblk_list);
513 		free_sackblock(scb, sb);
514 		--scb->nblocks;
515 		KASSERT(scb->nblocks > 0,
516 		    ("removed partial right: %d blocks left", scb->nblocks));
517 	}
518 	return 0;
519 }
520 
521 #ifdef DEBUG_SACK_BLOCKS
522 static void
tcp_sack_dump_blocks(const struct scoreboard * scb)523 tcp_sack_dump_blocks(const struct scoreboard *scb)
524 {
525 	const struct sackblock *sb;
526 
527 	kprintf("%d blocks:", scb->nblocks);
528 	TAILQ_FOREACH(sb, &scb->sackblocks, sblk_list)
529 		kprintf(" [%u, %u)", sb->sblk_start, sb->sblk_end);
530 	kprintf("\n");
531 }
532 #else
533 static __inline void
tcp_sack_dump_blocks(const struct scoreboard * scb)534 tcp_sack_dump_blocks(const struct scoreboard *scb)
535 {
536 }
537 #endif
538 
539 /*
540  * Optimization to quickly determine which packets are lost.
541  */
542 void
tcp_sack_update_lostseq(struct scoreboard * scb,tcp_seq snd_una,u_int maxseg,int rxtthresh)543 tcp_sack_update_lostseq(struct scoreboard *scb, tcp_seq snd_una, u_int maxseg,
544     int rxtthresh)
545 {
546 	struct sackblock *sb;
547 	int nsackblocks = 0;
548 	int bytes_sacked = 0;
549 	int rxtthresh_bytes;
550 
551 	if (tcp_do_rfc6675)
552 		rxtthresh_bytes = (rxtthresh - 1) * maxseg;
553 	else
554 		rxtthresh_bytes = rxtthresh * maxseg;
555 
556 	sb = TAILQ_LAST(&scb->sackblocks, sackblock_list);
557 	while (sb != NULL) {
558 		++nsackblocks;
559 		bytes_sacked += sb->sblk_end - sb->sblk_start;
560 		if (nsackblocks == rxtthresh ||
561 		    bytes_sacked >= rxtthresh_bytes) {
562 			scb->lostseq = sb->sblk_start;
563 			return;
564 		}
565 		sb = TAILQ_PREV(sb, sackblock_list, sblk_list);
566 	}
567 	scb->lostseq = snd_una;
568 }
569 
570 /*
571  * Return whether the given sequence number is considered lost.
572  */
573 boolean_t
tcp_sack_islost(const struct scoreboard * scb,tcp_seq seqnum)574 tcp_sack_islost(const struct scoreboard *scb, tcp_seq seqnum)
575 {
576 	return SEQ_LT(seqnum, scb->lostseq);
577 }
578 
579 /*
580  * True if at least "amount" has been SACKed.  Used by Early Retransmit.
581  */
582 boolean_t
tcp_sack_has_sacked(const struct scoreboard * scb,u_int amount)583 tcp_sack_has_sacked(const struct scoreboard *scb, u_int amount)
584 {
585 	const struct sackblock *sb;
586 	int bytes_sacked = 0;
587 
588 	TAILQ_FOREACH(sb, &scb->sackblocks, sblk_list) {
589 		bytes_sacked += sb->sblk_end - sb->sblk_start;
590 		if (bytes_sacked >= amount)
591 			return TRUE;
592 	}
593 	return FALSE;
594 }
595 
596 /*
597  * Number of bytes SACKed below seq.
598  */
599 int
tcp_sack_bytes_below(const struct scoreboard * scb,tcp_seq seq)600 tcp_sack_bytes_below(const struct scoreboard *scb, tcp_seq seq)
601 {
602 	const struct sackblock *sb;
603 	int bytes_sacked = 0;
604 
605 	sb = TAILQ_FIRST(&scb->sackblocks);
606 	while (sb && SEQ_GT(seq, sb->sblk_start)) {
607 		bytes_sacked += seq_min(seq, sb->sblk_end) - sb->sblk_start;
608 		sb = TAILQ_NEXT(sb, sblk_list);
609 	}
610 	return bytes_sacked;
611 }
612 
613 /*
614  * Return estimate of the number of bytes outstanding in the network.
615  */
616 uint32_t
tcp_sack_compute_pipe(const struct tcpcb * tp)617 tcp_sack_compute_pipe(const struct tcpcb *tp)
618 {
619 	const struct scoreboard *scb = &tp->scb;
620 	const struct sackblock *sb;
621 	int nlost, nretransmitted;
622 	tcp_seq end;
623 
624 	nlost = tp->snd_max - scb->lostseq;
625 	nretransmitted = tp->rexmt_high - tp->snd_una;
626 
627 	TAILQ_FOREACH(sb, &scb->sackblocks, sblk_list) {
628 		if (SEQ_LT(sb->sblk_start, tp->rexmt_high)) {
629 			end = seq_min(sb->sblk_end, tp->rexmt_high);
630 			nretransmitted -= end - sb->sblk_start;
631 		}
632 		if (SEQ_GEQ(sb->sblk_start, scb->lostseq))
633 			nlost -= sb->sblk_end - sb->sblk_start;
634 	}
635 
636 	return (nlost + nretransmitted);
637 }
638 
639 /*
640  * Return the sequence number and length of the next segment to transmit
641  * when in Fast Recovery.
642  */
643 boolean_t
tcp_sack_nextseg(struct tcpcb * tp,tcp_seq * nextrexmt,uint32_t * plen,boolean_t * rescue)644 tcp_sack_nextseg(struct tcpcb *tp, tcp_seq *nextrexmt, uint32_t *plen,
645     boolean_t *rescue)
646 {
647 	struct scoreboard *scb = &tp->scb;
648 	struct socket *so = tp->t_inpcb->inp_socket;
649 	struct sackblock *sb;
650 	const struct sackblock *lastblock =
651 	    TAILQ_LAST(&scb->sackblocks, sackblock_list);
652 	tcp_seq torexmt;
653 	long len, off, sendwin;
654 
655 	/* skip SACKed data */
656 	tcp_sack_skip_sacked(scb, &tp->rexmt_high);
657 
658 	/* Look for lost data. */
659 	torexmt = tp->rexmt_high;
660 	*rescue = FALSE;
661 	if (lastblock != NULL) {
662 		if (SEQ_LT(torexmt, lastblock->sblk_end) &&
663 		    tcp_sack_islost(scb, torexmt)) {
664 sendunsacked:
665 			*nextrexmt = torexmt;
666 			/* If the left-hand edge has been SACKed, pull it in. */
667 			if (sack_block_lookup(scb, torexmt + tp->t_maxseg, &sb))
668 				*plen = sb->sblk_start - torexmt;
669 			else
670 				*plen = tp->t_maxseg;
671 			return TRUE;
672 		}
673 	}
674 
675 	/* See if unsent data available within send window. */
676 	off = tp->snd_max - tp->snd_una;
677 	sendwin = min(tp->snd_wnd, tp->snd_bwnd);
678 	len = (long) ulmin(so->so_snd.ssb_cc, sendwin) - off;
679 	if (len > 0) {
680 		*nextrexmt = tp->snd_max;	/* Send new data. */
681 		*plen = tp->t_maxseg;
682 		return TRUE;
683 	}
684 
685 	/* We're less certain this data has been lost. */
686 	if (lastblock != NULL && SEQ_LT(torexmt, lastblock->sblk_end))
687 		goto sendunsacked;
688 
689 	/* Rescue retransmission */
690 	if (tcp_do_rescuesack || tcp_do_rfc6675) {
691 		tcpstat.tcps_sackrescue_try++;
692 		if (tp->sack_flags & TSACK_F_SACKRESCUED) {
693 			if (!tcp_aggressive_rescuesack)
694 				return FALSE;
695 
696 			/*
697 			 * Aggressive variant of the rescue retransmission.
698 			 *
699 			 * The idea of the rescue retransmission is to sustain
700 			 * the ACK clock thus to avoid timeout retransmission.
701 			 *
702 			 * Under some situations, the conservative approach
703 			 * suggested in the draft
704  			 * http://tools.ietf.org/html/
705 			 * draft-nishida-tcpm-rescue-retransmission-00
706 			 * could not sustain ACK clock, since it only allows
707 			 * one rescue retransmission before a cumulative ACK
708 			 * covers the segement transmitted by rescue
709 			 * retransmission.
710 			 *
711 			 * We try to locate the next unSACKed segment which
712 			 * follows the previously sent rescue segment.  If
713 			 * there is no such segment, we loop back to the first
714 			 * unacknowledged segment.
715 			 */
716 
717 			/*
718 			 * Skip SACKed data, but here we follow
719 			 * the last transmitted rescue segment.
720 			 */
721 			torexmt = tp->rexmt_rescue;
722 			tcp_sack_skip_sacked(scb, &torexmt);
723 		}
724 		if (torexmt == tp->snd_max) {
725 			/* Nothing left to retransmit; restart */
726 			torexmt = tp->snd_una;
727 		}
728 		*rescue = TRUE;
729 		goto sendunsacked;
730 	} else if (tcp_do_smartsack && lastblock == NULL) {
731 		tcpstat.tcps_sackrescue_try++;
732 		*rescue = TRUE;
733 		goto sendunsacked;
734 	}
735 
736 	return FALSE;
737 }
738 
739 /*
740  * Return the next sequence number higher than "*prexmt" that has
741  * not been SACKed.
742  */
743 void
tcp_sack_skip_sacked(struct scoreboard * scb,tcp_seq * prexmt)744 tcp_sack_skip_sacked(struct scoreboard *scb, tcp_seq *prexmt)
745 {
746 	struct sackblock *sb;
747 
748 	/* skip SACKed data */
749 	if (sack_block_lookup(scb, *prexmt, &sb))
750 		*prexmt = sb->sblk_end;
751 }
752 
753 /*
754  * The length of the first amount of unSACKed data
755  */
756 uint32_t
tcp_sack_first_unsacked_len(const struct tcpcb * tp)757 tcp_sack_first_unsacked_len(const struct tcpcb *tp)
758 {
759 	const struct sackblock *sb;
760 
761 	sb = TAILQ_FIRST(&tp->scb.sackblocks);
762 	if (sb == NULL)
763 		return tp->t_maxseg;
764 
765 	KASSERT(SEQ_LT(tp->snd_una, sb->sblk_start),
766 	    ("invalid sb start %u, snd_una %u",
767 	     sb->sblk_start, tp->snd_una));
768 	return (sb->sblk_start - tp->snd_una);
769 }
770 
771 #ifdef later
772 void
tcp_sack_save_scoreboard(struct scoreboard * scb)773 tcp_sack_save_scoreboard(struct scoreboard *scb)
774 {
775 	struct scoreboard *scb = &tp->scb;
776 
777 	scb->sackblocks_prev = scb->sackblocks;
778 	TAILQ_INIT(&scb->sackblocks);
779 }
780 
781 void
tcp_sack_revert_scoreboard(struct scoreboard * scb,tcp_seq snd_una,u_int maxseg)782 tcp_sack_revert_scoreboard(struct scoreboard *scb, tcp_seq snd_una,
783 			   u_int maxseg)
784 {
785 	struct sackblock *sb;
786 
787 	scb->sackblocks = scb->sackblocks_prev;
788 	scb->nblocks = 0;
789 	TAILQ_FOREACH(sb, &scb->sackblocks, sblk_list)
790 		++scb->nblocks;
791 	tcp_sack_ack_blocks(scb, snd_una);
792 	scb->lastfound = NULL;
793 }
794 #endif
795 
796 #ifdef DEBUG_SACK_HISTORY
797 static void
tcp_sack_dump_history(const char * msg,const struct tcpcb * tp)798 tcp_sack_dump_history(const char *msg, const struct tcpcb *tp)
799 {
800 	int i;
801 	static int ndumped;
802 
803 	/* only need a couple of these to debug most problems */
804 	if (++ndumped > 900)
805 		return;
806 
807 	kprintf("%s:\tnsackhistory %d: ", msg, tp->nsackhistory);
808 	for (i = 0; i < tp->nsackhistory; ++i)
809 		kprintf("[%u, %u) ", tp->sackhistory[i].rblk_start,
810 		    tp->sackhistory[i].rblk_end);
811 	kprintf("\n");
812 }
813 #else
814 static __inline void
tcp_sack_dump_history(const char * msg,const struct tcpcb * tp)815 tcp_sack_dump_history(const char *msg, const struct tcpcb *tp)
816 {
817 }
818 #endif
819 
820 /*
821  * Remove old SACK blocks from the SACK history that have already been ACKed.
822  */
823 static void
tcp_sack_ack_history(struct tcpcb * tp)824 tcp_sack_ack_history(struct tcpcb *tp)
825 {
826 	int i, nblocks, openslot;
827 
828 	tcp_sack_dump_history("before tcp_sack_ack_history", tp);
829 	nblocks = tp->nsackhistory;
830 	for (i = openslot = 0; i < nblocks; ++i) {
831 		if (SEQ_LEQ(tp->sackhistory[i].rblk_end, tp->rcv_nxt)) {
832 			--tp->nsackhistory;
833 			continue;
834 		}
835 		if (SEQ_LT(tp->sackhistory[i].rblk_start, tp->rcv_nxt))
836 			tp->sackhistory[i].rblk_start = tp->rcv_nxt;
837 		if (i == openslot)
838 			++openslot;
839 		else
840 			tp->sackhistory[openslot++] = tp->sackhistory[i];
841 	}
842 	tcp_sack_dump_history("after tcp_sack_ack_history", tp);
843 	KASSERT(openslot == tp->nsackhistory,
844 	    ("tcp_sack_ack_history miscounted: %d != %d",
845 	    openslot, tp->nsackhistory));
846 }
847 
848 /*
849  * Add or merge newblock into reported history.
850  * Also remove or update SACK blocks that will be acked.
851  */
852 static void
tcp_sack_update_reported_history(struct tcpcb * tp,tcp_seq start,tcp_seq end)853 tcp_sack_update_reported_history(struct tcpcb *tp, tcp_seq start, tcp_seq end)
854 {
855 	struct raw_sackblock copy[MAX_SACK_REPORT_BLOCKS];
856 	int i, cindex;
857 
858 	tcp_sack_dump_history("before tcp_sack_update_reported_history", tp);
859 	/*
860 	 * Six cases:
861 	 *	0) no overlap
862 	 *	1) newblock == oldblock
863 	 *	2) oldblock contains newblock
864 	 *	3) newblock contains oldblock
865 	 *	4) tail of oldblock overlaps or abuts start of newblock
866 	 *	5) tail of newblock overlaps or abuts head of oldblock
867 	 */
868 	for (i = cindex = 0; i < tp->nsackhistory; ++i) {
869 		struct raw_sackblock *oldblock = &tp->sackhistory[i];
870 		tcp_seq old_start = oldblock->rblk_start;
871 		tcp_seq old_end = oldblock->rblk_end;
872 
873 		if (SEQ_LT(end, old_start) || SEQ_GT(start, old_end)) {
874 			/* Case 0:  no overlap.  Copy old block. */
875 			copy[cindex++] = *oldblock;
876 			continue;
877 		}
878 
879 		if (SEQ_GEQ(start, old_start) && SEQ_LEQ(end, old_end)) {
880 			/* Cases 1 & 2.  Move block to front of history. */
881 			int j;
882 
883 			start = old_start;
884 			end = old_end;
885 			/* no need to check rest of blocks */
886 			for (j = i + 1; j < tp->nsackhistory; ++j)
887 				copy[cindex++] = tp->sackhistory[j];
888 			break;
889 		}
890 
891 		if (SEQ_GEQ(old_end, start) && SEQ_LT(old_start, start)) {
892 			/* Case 4:  extend start of new block. */
893 			start = old_start;
894 		} else if (SEQ_GEQ(end, old_start) && SEQ_GT(old_end, end)) {
895 			/* Case 5: extend end of new block */
896 			end = old_end;
897 		} else {
898 			/* Case 3.  Delete old block by not copying it. */
899 			KASSERT(SEQ_LEQ(start, old_start) &&
900 				SEQ_GEQ(end, old_end),
901 			    ("bad logic: old [%u, %u), new [%u, %u)",
902 			     old_start, old_end, start, end));
903 		}
904 	}
905 
906 	/* insert new block */
907 	tp->sackhistory[0].rblk_start = start;
908 	tp->sackhistory[0].rblk_end = end;
909 	cindex = min(cindex, MAX_SACK_REPORT_BLOCKS - 1);
910 	for (i = 0; i < cindex; ++i)
911 		tp->sackhistory[i + 1] = copy[i];
912 	tp->nsackhistory = cindex + 1;
913 	tcp_sack_dump_history("after tcp_sack_update_reported_history", tp);
914 }
915 
916 /*
917  * Fill in SACK report to return to data sender.
918  */
919 void
tcp_sack_fill_report(struct tcpcb * tp,u_char * opt,u_int * plen)920 tcp_sack_fill_report(struct tcpcb *tp, u_char *opt, u_int *plen)
921 {
922 	u_int optlen = *plen;
923 	uint32_t *lp = (uint32_t *)(opt + optlen);
924 	uint32_t *olp;
925 	tcp_seq hstart = tp->rcv_nxt, hend;
926 	int nblocks;
927 
928 	KASSERT(TCP_MAXOLEN - optlen >=
929 	    TCPOLEN_SACK_ALIGNED + TCPOLEN_SACK_BLOCK,
930 	    ("no room for SACK header and one block: optlen %d", optlen));
931 
932 	if (tp->sack_flags & TSACK_F_DUPSEG)
933 		tcpstat.tcps_snddsackopt++;
934 	else
935 		tcpstat.tcps_sndsackopt++;
936 
937 	olp = lp++;
938 	optlen += TCPOLEN_SACK_ALIGNED;
939 
940 	tcp_sack_ack_history(tp);
941 	if (tp->reportblk.rblk_start != tp->reportblk.rblk_end) {
942 		*lp++ = htonl(tp->reportblk.rblk_start);
943 		*lp++ = htonl(tp->reportblk.rblk_end);
944 		optlen += TCPOLEN_SACK_BLOCK;
945 		hstart = tp->reportblk.rblk_start;
946 		hend = tp->reportblk.rblk_end;
947 		if (tp->sack_flags & TSACK_F_ENCLOSESEG) {
948 			KASSERT(TCP_MAXOLEN - optlen >= TCPOLEN_SACK_BLOCK,
949 			    ("no room for enclosing SACK block: oplen %d",
950 			    optlen));
951 			*lp++ = htonl(tp->encloseblk.rblk_start);
952 			*lp++ = htonl(tp->encloseblk.rblk_end);
953 			optlen += TCPOLEN_SACK_BLOCK;
954 			hstart = tp->encloseblk.rblk_start;
955 			hend = tp->encloseblk.rblk_end;
956 		}
957 		if (SEQ_GT(hstart, tp->rcv_nxt))
958 			tcp_sack_update_reported_history(tp, hstart, hend);
959 	}
960 	if (tcp_do_smartsack && (tp->sack_flags & TSACK_F_SACKLEFT)) {
961 		/* Fill in from left!  Walk re-assembly queue. */
962 		struct tseg_qent *q;
963 
964 		q = TAILQ_FIRST(&tp->t_segq);
965 		while (q != NULL &&
966 		    TCP_MAXOLEN - optlen >= TCPOLEN_SACK_BLOCK) {
967 			*lp++ = htonl(q->tqe_th->th_seq);
968 			*lp++ = htonl(TCP_SACK_BLKEND(
969 			    q->tqe_th->th_seq + q->tqe_len,
970 			    q->tqe_th->th_flags));
971 			optlen += TCPOLEN_SACK_BLOCK;
972 			q = TAILQ_NEXT(q, tqe_q);
973 		}
974 	} else {
975 		int n = 0;
976 
977 		/* Fill in SACK blocks from right side. */
978 		while (n < tp->nsackhistory &&
979 		    TCP_MAXOLEN - optlen >= TCPOLEN_SACK_BLOCK) {
980 			if (tp->sackhistory[n].rblk_start != hstart) {
981 				*lp++ = htonl(tp->sackhistory[n].rblk_start);
982 				*lp++ = htonl(tp->sackhistory[n].rblk_end);
983 				optlen += TCPOLEN_SACK_BLOCK;
984 			}
985 			++n;
986 		}
987 	}
988 	tp->reportblk.rblk_start = tp->reportblk.rblk_end;
989 	tp->sack_flags &=
990 	    ~(TSACK_F_DUPSEG | TSACK_F_ENCLOSESEG | TSACK_F_SACKLEFT);
991 	nblocks = (lp - olp - 1) / 2;
992 	*olp = htonl(TCPOPT_SACK_ALIGNED |
993 		     (TCPOLEN_SACK + nblocks * TCPOLEN_SACK_BLOCK));
994 	*plen = optlen;
995 }
996