ntp_adjtime.2: Use the correct function name.
[dragonfly.git] / sys / netinet / tcp_sack.c
CommitLineData
91489f6b
JH
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 *
12f68b86 33 * $DragonFly: src/sys/netinet/tcp_sack.c,v 1.8 2008/08/15 21:37:16 nth Exp $
91489f6b
JH
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
46#include <net/if.h>
47
48#include <netinet/in.h>
49#include <netinet/in_systm.h>
50#include <netinet/ip.h>
51#include <netinet/in_var.h>
52#include <netinet/in_pcb.h>
53#include <netinet/ip_var.h>
54#include <netinet/tcp.h>
55#include <netinet/tcp_seq.h>
56#include <netinet/tcp_var.h>
57
bf55dc6d
SZ
58/*
59 * Implemented:
60 *
61 * RFC 2018
62 * RFC 2883
63 * RFC 3517
64 */
65
91489f6b
JH
66struct sackblock {
67 tcp_seq sblk_start;
68 tcp_seq sblk_end;
69 TAILQ_ENTRY(sackblock) sblk_list;
70};
71
72#define MAXSAVEDBLOCKS 8 /* per connection limit */
73
01d244e6
SZ
74static int insert_block(struct scoreboard *scb,
75 const struct raw_sackblock *raw_sb);
91489f6b
JH
76static void update_lostseq(struct scoreboard *scb, tcp_seq snd_una,
77 u_int maxseg);
78
12f68b86 79static MALLOC_DEFINE(M_SACKBLOCK, "sblk", "sackblock struct");
91489f6b
JH
80
81/*
82 * Per-tcpcb initialization.
83 */
84void
85tcp_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 */
99static boolean_t
100sack_block_lookup(struct scoreboard *scb, tcp_seq seq, struct sackblock **sb)
101{
102 struct sackblock *hint = scb->lastfound;
103 struct sackblock *cur, *last, *prev;
104
105 if (TAILQ_EMPTY(&scb->sackblocks)) {
106 *sb = NULL;
107 return FALSE;
108 }
109
110 if (hint == NULL) {
111 /* No hint. Search from start to end. */
112 cur = TAILQ_FIRST(&scb->sackblocks);
113 last = NULL;
114 prev = TAILQ_LAST(&scb->sackblocks, sackblock_list);
115 } else {
116 if (SEQ_GEQ(seq, hint->sblk_start)) {
117 /* Search from hint to end of list. */
118 cur = hint;
119 last = NULL;
120 prev = TAILQ_LAST(&scb->sackblocks, sackblock_list);
121 } else {
122 /* Search from front of list to hint. */
123 cur = TAILQ_FIRST(&scb->sackblocks);
124 last = hint;
125 prev = TAILQ_PREV(hint, sackblock_list, sblk_list);
126 }
127 }
128
129 do {
130 if (SEQ_GT(cur->sblk_end, seq)) {
131 if (SEQ_GEQ(seq, cur->sblk_start)) {
132 *sb = scb->lastfound = cur;
133 return TRUE;
134 } else {
135 *sb = scb->lastfound =
136 TAILQ_PREV(cur, sackblock_list, sblk_list);
137 return FALSE;
138 }
139 }
140 cur = TAILQ_NEXT(cur, sblk_list);
141 } while (cur != last);
142
143 *sb = scb->lastfound = prev;
144 return FALSE;
145}
146
147/*
148 * Allocate a SACK block.
149 */
150static __inline struct sackblock *
006af0db 151alloc_sackblock(struct scoreboard *scb, const struct raw_sackblock *raw_sb)
91489f6b 152{
26d3f1fb
SZ
153 struct sackblock *sb;
154
006af0db
SZ
155 if (scb->freecache != NULL) {
156 sb = scb->freecache;
157 scb->freecache = NULL;
158 tcpstat.tcps_sacksbfast++;
01d244e6 159 } else {
006af0db
SZ
160 sb = kmalloc(sizeof(struct sackblock), M_SACKBLOCK, M_NOWAIT);
161 if (sb == NULL) {
162 tcpstat.tcps_sacksbfailed++;
163 return NULL;
164 }
26d3f1fb 165 }
006af0db
SZ
166 sb->sblk_start = raw_sb->rblk_start;
167 sb->sblk_end = raw_sb->rblk_end;
26d3f1fb 168 return sb;
91489f6b
JH
169}
170
01d244e6
SZ
171static __inline struct sackblock *
172alloc_sackblock_limit(struct scoreboard *scb,
173 const struct raw_sackblock *raw_sb)
174{
175 if (scb->nblocks == MAXSAVEDBLOCKS) {
176 /*
177 * Should try to kick out older blocks XXX JH
178 * May be able to coalesce with existing block.
179 * Or, go other way and free all blocks if we hit
180 * this limit.
181 */
182 tcpstat.tcps_sacksboverflow++;
183 return NULL;
184 }
006af0db 185 return alloc_sackblock(scb, raw_sb);
01d244e6
SZ
186}
187
91489f6b
JH
188/*
189 * Free a SACK block.
190 */
191static __inline void
006af0db 192free_sackblock(struct scoreboard *scb, struct sackblock *s)
91489f6b 193{
006af0db
SZ
194 if (scb->freecache == NULL) {
195 /* YYY Maybe use the latest freed block? */
196 scb->freecache = s;
197 return;
198 }
12f68b86 199 kfree(s, M_SACKBLOCK);
91489f6b
JH
200}
201
202/*
203 * Free up SACK blocks for data that's been acked.
204 */
205static void
206tcp_sack_ack_blocks(struct scoreboard *scb, tcp_seq th_ack)
207{
208 struct sackblock *sb, *nb;
209
210 sb = TAILQ_FIRST(&scb->sackblocks);
211 while (sb && SEQ_LEQ(sb->sblk_end, th_ack)) {
212 nb = TAILQ_NEXT(sb, sblk_list);
9e3d6c96 213 if (scb->lastfound == sb)
91489f6b
JH
214 scb->lastfound = NULL;
215 TAILQ_REMOVE(&scb->sackblocks, sb, sblk_list);
006af0db 216 free_sackblock(scb, sb);
91489f6b
JH
217 --scb->nblocks;
218 KASSERT(scb->nblocks >= 0,
219 ("SACK block count underflow: %d < 0", scb->nblocks));
220 sb = nb;
221 }
222 if (sb && SEQ_GT(th_ack, sb->sblk_start))
223 sb->sblk_start = th_ack; /* other side reneged? XXX */
224}
225
226/*
227 * Delete and free SACK blocks saved in scoreboard.
228 */
229void
230tcp_sack_cleanup(struct scoreboard *scb)
231{
232 struct sackblock *sb, *nb;
233
234 TAILQ_FOREACH_MUTABLE(sb, &scb->sackblocks, sblk_list, nb) {
006af0db 235 free_sackblock(scb, sb);
91489f6b
JH
236 --scb->nblocks;
237 }
238 KASSERT(scb->nblocks == 0,
239 ("SACK block %d count not zero", scb->nblocks));
240 TAILQ_INIT(&scb->sackblocks);
241 scb->lastfound = NULL;
242}
243
006af0db
SZ
244/*
245 * Delete and free SACK blocks saved in scoreboard.
246 * Delete the one slot block cache.
247 */
248void
249tcp_sack_destroy(struct scoreboard *scb)
250{
251 tcp_sack_cleanup(scb);
252 if (scb->freecache != NULL) {
253 kfree(scb->freecache, M_SACKBLOCK);
254 scb->freecache = NULL;
255 }
256}
257
d58ca578
SZ
258/*
259 * Cleanup the reported SACK block information
260 */
261void
262tcp_sack_report_cleanup(struct tcpcb *tp)
263{
264 tp->t_flags &= ~(TF_DUPSEG | TF_ENCLOSESEG | TF_SACKLEFT);
265 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
266}
267
91489f6b
JH
268/*
269 * Returns 0 if not D-SACK block,
270 * 1 if D-SACK,
271 * 2 if duplicate of out-of-order D-SACK block.
272 */
273int
274tcp_sack_ndsack_blocks(struct raw_sackblock *blocks, const int numblocks,
275 tcp_seq snd_una)
276{
277 if (numblocks == 0)
278 return 0;
279
280 if (SEQ_LT(blocks[0].rblk_start, snd_una))
281 return 1;
282
283 /* block 0 inside block 1 */
284 if (numblocks > 1 &&
285 SEQ_GEQ(blocks[0].rblk_start, blocks[1].rblk_start) &&
286 SEQ_LEQ(blocks[0].rblk_end, blocks[1].rblk_end))
287 return 2;
288
289 return 0;
290}
291
292/*
293 * Update scoreboard on new incoming ACK.
294 */
295static void
296tcp_sack_add_blocks(struct tcpcb *tp, struct tcpopt *to)
297{
298 const int numblocks = to->to_nsackblocks;
299 struct raw_sackblock *blocks = to->to_sackblocks;
300 struct scoreboard *scb = &tp->scb;
91489f6b
JH
301 int startblock;
302 int i;
303
304 if (tcp_sack_ndsack_blocks(blocks, numblocks, tp->snd_una) > 0)
305 startblock = 1;
306 else
307 startblock = 0;
308
309 for (i = startblock; i < numblocks; i++) {
310 struct raw_sackblock *newsackblock = &blocks[i];
311
312 /* don't accept bad SACK blocks */
02cc2f35
SZ
313 if (SEQ_GT(newsackblock->rblk_end, tp->snd_max)) {
314 tcpstat.tcps_rcvbadsackopt++;
91489f6b 315 break; /* skip all other blocks */
02cc2f35
SZ
316 }
317 tcpstat.tcps_sacksbupdate++;
91489f6b 318
01d244e6
SZ
319 if (insert_block(scb, newsackblock))
320 break;
91489f6b
JH
321 }
322}
323
324void
325tcp_sack_update_scoreboard(struct tcpcb *tp, struct tcpopt *to)
326{
327 struct scoreboard *scb = &tp->scb;
328
329 tcp_sack_ack_blocks(scb, tp->snd_una);
330 tcp_sack_add_blocks(tp, to);
331 update_lostseq(scb, tp->snd_una, tp->t_maxseg);
332 if (SEQ_LT(tp->rexmt_high, tp->snd_una))
333 tp->rexmt_high = tp->snd_una;
334}
335
336/*
337 * Insert SACK block into sender's scoreboard.
338 */
01d244e6
SZ
339static int
340insert_block(struct scoreboard *scb, const struct raw_sackblock *raw_sb)
91489f6b
JH
341{
342 struct sackblock *sb, *workingblock;
343 boolean_t overlap_front;
344
2ed615e0 345 if (TAILQ_EMPTY(&scb->sackblocks)) {
01d244e6
SZ
346 struct sackblock *newblock;
347
2ed615e0 348 KASSERT(scb->nblocks == 0, ("emply scb w/ blocks"));
01d244e6 349
006af0db 350 newblock = alloc_sackblock(scb, raw_sb);
01d244e6
SZ
351 if (newblock == NULL)
352 return ENOMEM;
2ed615e0 353 TAILQ_INSERT_HEAD(&scb->sackblocks, newblock, sblk_list);
01d244e6
SZ
354 scb->nblocks = 1;
355 return 0;
2ed615e0 356 }
91489f6b 357
01d244e6
SZ
358 KASSERT(scb->nblocks > 0, ("insert_block() called w/ no blocks"));
359 KASSERT(scb->nblocks <= MAXSAVEDBLOCKS,
91489f6b
JH
360 ("too many SACK blocks %d", scb->nblocks));
361
01d244e6 362 overlap_front = sack_block_lookup(scb, raw_sb->rblk_start, &sb);
91489f6b
JH
363
364 if (sb == NULL) {
01d244e6
SZ
365 workingblock = alloc_sackblock_limit(scb, raw_sb);
366 if (workingblock == NULL)
367 return ENOMEM;
368 TAILQ_INSERT_HEAD(&scb->sackblocks, workingblock, sblk_list);
91489f6b
JH
369 ++scb->nblocks;
370 } else {
01d244e6
SZ
371 if (overlap_front || sb->sblk_end == raw_sb->rblk_start) {
372 /* Extend old block */
91489f6b 373 workingblock = sb;
01d244e6
SZ
374 if (SEQ_GT(raw_sb->rblk_end, sb->sblk_end))
375 sb->sblk_end = raw_sb->rblk_end;
02cc2f35 376 tcpstat.tcps_sacksbreused++;
91489f6b 377 } else {
01d244e6
SZ
378 workingblock = alloc_sackblock_limit(scb, raw_sb);
379 if (workingblock == NULL)
380 return ENOMEM;
381 TAILQ_INSERT_AFTER(&scb->sackblocks, sb, workingblock,
382 sblk_list);
91489f6b
JH
383 ++scb->nblocks;
384 }
385 }
386
387 /* Consolidate right-hand side. */
388 sb = TAILQ_NEXT(workingblock, sblk_list);
389 while (sb != NULL &&
390 SEQ_GEQ(workingblock->sblk_end, sb->sblk_end)) {
391 struct sackblock *nextblock;
392
393 nextblock = TAILQ_NEXT(sb, sblk_list);
9e3d6c96
MD
394 if (scb->lastfound == sb)
395 scb->lastfound = NULL;
91489f6b
JH
396 /* Remove completely overlapped block */
397 TAILQ_REMOVE(&scb->sackblocks, sb, sblk_list);
006af0db 398 free_sackblock(scb, sb);
91489f6b
JH
399 --scb->nblocks;
400 KASSERT(scb->nblocks > 0,
401 ("removed overlapped block: %d blocks left", scb->nblocks));
402 sb = nextblock;
403 }
404 if (sb != NULL &&
405 SEQ_GEQ(workingblock->sblk_end, sb->sblk_start)) {
406 /* Extend new block to cover partially overlapped old block. */
407 workingblock->sblk_end = sb->sblk_end;
9e3d6c96
MD
408 if (scb->lastfound == sb)
409 scb->lastfound = NULL;
91489f6b 410 TAILQ_REMOVE(&scb->sackblocks, sb, sblk_list);
006af0db 411 free_sackblock(scb, sb);
91489f6b
JH
412 --scb->nblocks;
413 KASSERT(scb->nblocks > 0,
414 ("removed partial right: %d blocks left", scb->nblocks));
415 }
01d244e6 416 return 0;
91489f6b
JH
417}
418
419#ifdef DEBUG_SACK_BLOCKS
420static void
421tcp_sack_dump_blocks(struct scoreboard *scb)
422{
423 struct sackblock *sb;
424
a6ec04bc 425 kprintf("%d blocks:", scb->nblocks);
91489f6b 426 TAILQ_FOREACH(sb, &scb->sackblocks, sblk_list)
a6ec04bc
SW
427 kprintf(" [%u, %u)", sb->sblk_start, sb->sblk_end);
428 kprintf("\n");
91489f6b
JH
429}
430#else
431static __inline void
432tcp_sack_dump_blocks(struct scoreboard *scb)
433{
434}
435#endif
436
437/*
438 * Optimization to quickly determine which packets are lost.
439 */
440static void
441update_lostseq(struct scoreboard *scb, tcp_seq snd_una, u_int maxseg)
442{
443 struct sackblock *sb;
444 int nsackblocks = 0;
445 int bytes_sacked = 0;
446
447 sb = TAILQ_LAST(&scb->sackblocks, sackblock_list);
448 while (sb != NULL) {
449 ++nsackblocks;
450 bytes_sacked += sb->sblk_end - sb->sblk_start;
451 if (nsackblocks == tcprexmtthresh ||
452 bytes_sacked >= tcprexmtthresh * maxseg) {
453 scb->lostseq = sb->sblk_start;
454 return;
455 }
456 sb = TAILQ_PREV(sb, sackblock_list, sblk_list);
457 }
458 scb->lostseq = snd_una;
459}
460
461/*
462 * Return whether the given sequence number is considered lost.
463 */
464static boolean_t
465scb_islost(struct scoreboard *scb, tcp_seq seqnum)
466{
467 return SEQ_LT(seqnum, scb->lostseq);
468}
469
470/*
471 * True if at least "amount" has been SACKed. Used by Early Retransmit.
472 */
473boolean_t
474tcp_sack_has_sacked(struct scoreboard *scb, u_int amount)
475{
476 struct sackblock *sb;
477 int bytes_sacked = 0;
478
479 TAILQ_FOREACH(sb, &scb->sackblocks, sblk_list) {
480 bytes_sacked += sb->sblk_end - sb->sblk_start;
481 if (bytes_sacked >= amount)
482 return TRUE;
483 }
484 return FALSE;
485}
486
9a5b142f
JH
487/*
488 * Number of bytes SACKed below seq.
489 */
490int
491tcp_sack_bytes_below(struct scoreboard *scb, tcp_seq seq)
492{
493 struct sackblock *sb;
494 int bytes_sacked = 0;
495
496 sb = TAILQ_FIRST(&scb->sackblocks);
497 while (sb && SEQ_GT(seq, sb->sblk_start)) {
498 bytes_sacked += seq_min(seq, sb->sblk_end) - sb->sblk_start;
499 sb = TAILQ_NEXT(sb, sblk_list);
500 }
501 return bytes_sacked;
502}
503
91489f6b
JH
504/*
505 * Return estimate of the number of bytes outstanding in the network.
506 */
507uint32_t
508tcp_sack_compute_pipe(struct tcpcb *tp)
509{
510 struct scoreboard *scb = &tp->scb;
511 struct sackblock *sb;
512 int nlost, nretransmitted;
513 tcp_seq end;
514
515 nlost = tp->snd_max - scb->lostseq;
516 nretransmitted = tp->rexmt_high - tp->snd_una;
517
518 TAILQ_FOREACH(sb, &scb->sackblocks, sblk_list) {
519 if (SEQ_LT(sb->sblk_start, tp->rexmt_high)) {
520 end = seq_min(sb->sblk_end, tp->rexmt_high);
521 nretransmitted -= end - sb->sblk_start;
522 }
523 if (SEQ_GEQ(sb->sblk_start, scb->lostseq))
524 nlost -= sb->sblk_end - sb->sblk_start;
525 }
526
527 return (nlost + nretransmitted);
528}
529
530/*
531 * Return the sequence number and length of the next segment to transmit
532 * when in Fast Recovery.
533 */
534boolean_t
535tcp_sack_nextseg(struct tcpcb *tp, tcp_seq *nextrexmt, uint32_t *plen,
536 boolean_t *lostdup)
537{
538 struct scoreboard *scb = &tp->scb;
539 struct socket *so = tp->t_inpcb->inp_socket;
540 struct sackblock *sb;
541 const struct sackblock *lastblock =
542 TAILQ_LAST(&scb->sackblocks, sackblock_list);
543 tcp_seq torexmt;
544 long len, off;
545
546 /* skip SACKed data */
547 tcp_sack_skip_sacked(scb, &tp->rexmt_high);
548
549 /* Look for lost data. */
550 torexmt = tp->rexmt_high;
551 *lostdup = FALSE;
552 if (lastblock != NULL) {
553 if (SEQ_LT(torexmt, lastblock->sblk_end) &&
554 scb_islost(scb, torexmt)) {
555sendunsacked:
556 *nextrexmt = torexmt;
557 /* If the left-hand edge has been SACKed, pull it in. */
558 if (sack_block_lookup(scb, torexmt + tp->t_maxseg, &sb))
559 *plen = sb->sblk_start - torexmt;
560 else
561 *plen = tp->t_maxseg;
562 return TRUE;
563 }
564 }
565
566 /* See if unsent data available within send window. */
567 off = tp->snd_max - tp->snd_una;
6d49aa6f 568 len = (long) ulmin(so->so_snd.ssb_cc, tp->snd_wnd) - off;
91489f6b
JH
569 if (len > 0) {
570 *nextrexmt = tp->snd_max; /* Send new data. */
571 *plen = tp->t_maxseg;
572 return TRUE;
573 }
574
575 /* We're less certain this data has been lost. */
576 if (lastblock == NULL || SEQ_LT(torexmt, lastblock->sblk_end))
577 goto sendunsacked;
578
579 return FALSE;
580}
581
582/*
583 * Return the next sequence number higher than "*prexmt" that has
584 * not been SACKed.
585 */
586void
587tcp_sack_skip_sacked(struct scoreboard *scb, tcp_seq *prexmt)
588{
589 struct sackblock *sb;
590
591 /* skip SACKed data */
592 if (sack_block_lookup(scb, *prexmt, &sb))
593 *prexmt = sb->sblk_end;
594}
595
596#ifdef later
597void
598tcp_sack_save_scoreboard(struct scoreboard *scb)
599{
600 struct scoreboard *scb = &tp->scb;
601
602 scb->sackblocks_prev = scb->sackblocks;
603 TAILQ_INIT(&scb->sackblocks);
604}
605
606void
607tcp_sack_revert_scoreboard(struct scoreboard *scb, tcp_seq snd_una,
608 u_int maxseg)
609{
610 struct sackblock *sb;
611
612 scb->sackblocks = scb->sackblocks_prev;
613 scb->nblocks = 0;
614 TAILQ_FOREACH(sb, &scb->sackblocks, sblk_list)
615 ++scb->nblocks;
616 tcp_sack_ack_blocks(scb, snd_una);
617 scb->lastfound = NULL;
618}
619#endif
620
621#ifdef DEBUG_SACK_HISTORY
622static void
623tcp_sack_dump_history(char *msg, struct tcpcb *tp)
624{
625 int i;
626 static int ndumped;
627
628 /* only need a couple of these to debug most problems */
629 if (++ndumped > 900)
630 return;
631
a6ec04bc 632 kprintf("%s:\tnsackhistory %d: ", msg, tp->nsackhistory);
91489f6b 633 for (i = 0; i < tp->nsackhistory; ++i)
a6ec04bc 634 kprintf("[%u, %u) ", tp->sackhistory[i].rblk_start,
91489f6b 635 tp->sackhistory[i].rblk_end);
a6ec04bc 636 kprintf("\n");
91489f6b
JH
637}
638#else
639static __inline void
640tcp_sack_dump_history(char *msg, struct tcpcb *tp)
641{
642}
643#endif
644
645/*
646 * Remove old SACK blocks from the SACK history that have already been ACKed.
647 */
648static void
649tcp_sack_ack_history(struct tcpcb *tp)
650{
651 int i, nblocks, openslot;
652
653 tcp_sack_dump_history("before tcp_sack_ack_history", tp);
654 nblocks = tp->nsackhistory;
655 for (i = openslot = 0; i < nblocks; ++i) {
656 if (SEQ_LEQ(tp->sackhistory[i].rblk_end, tp->rcv_nxt)) {
657 --tp->nsackhistory;
658 continue;
659 }
660 if (SEQ_LT(tp->sackhistory[i].rblk_start, tp->rcv_nxt))
661 tp->sackhistory[i].rblk_start = tp->rcv_nxt;
662 if (i == openslot)
663 ++openslot;
664 else
665 tp->sackhistory[openslot++] = tp->sackhistory[i];
666 }
667 tcp_sack_dump_history("after tcp_sack_ack_history", tp);
668 KASSERT(openslot == tp->nsackhistory,
669 ("tcp_sack_ack_history miscounted: %d != %d",
670 openslot, tp->nsackhistory));
671}
672
673/*
674 * Add or merge newblock into reported history.
675 * Also remove or update SACK blocks that will be acked.
676 */
677static void
678tcp_sack_update_reported_history(struct tcpcb *tp, tcp_seq start, tcp_seq end)
679{
680 struct raw_sackblock copy[MAX_SACK_REPORT_BLOCKS];
681 int i, cindex;
682
683 tcp_sack_dump_history("before tcp_sack_update_reported_history", tp);
684 /*
685 * Six cases:
686 * 0) no overlap
687 * 1) newblock == oldblock
688 * 2) oldblock contains newblock
689 * 3) newblock contains oldblock
690 * 4) tail of oldblock overlaps or abuts start of newblock
691 * 5) tail of newblock overlaps or abuts head of oldblock
692 */
693 for (i = cindex = 0; i < tp->nsackhistory; ++i) {
694 struct raw_sackblock *oldblock = &tp->sackhistory[i];
695 tcp_seq old_start = oldblock->rblk_start;
696 tcp_seq old_end = oldblock->rblk_end;
697
698 if (SEQ_LT(end, old_start) || SEQ_GT(start, old_end)) {
699 /* Case 0: no overlap. Copy old block. */
700 copy[cindex++] = *oldblock;
701 continue;
702 }
703
704 if (SEQ_GEQ(start, old_start) && SEQ_LEQ(end, old_end)) {
705 /* Cases 1 & 2. Move block to front of history. */
706 int j;
707
708 start = old_start;
709 end = old_end;
710 /* no need to check rest of blocks */
711 for (j = i + 1; j < tp->nsackhistory; ++j)
712 copy[cindex++] = tp->sackhistory[j];
713 break;
714 }
715
716 if (SEQ_GEQ(old_end, start) && SEQ_LT(old_start, start)) {
717 /* Case 4: extend start of new block. */
718 start = old_start;
719 } else if (SEQ_GEQ(end, old_start) && SEQ_GT(old_end, end)) {
720 /* Case 5: extend end of new block */
721 end = old_end;
722 } else {
723 /* Case 3. Delete old block by not copying it. */
724 KASSERT(SEQ_LEQ(start, old_start) &&
725 SEQ_GEQ(end, old_end),
726 ("bad logic: old [%u, %u), new [%u, %u)",
727 old_start, old_end, start, end));
728 }
729 }
730
731 /* insert new block */
732 tp->sackhistory[0].rblk_start = start;
733 tp->sackhistory[0].rblk_end = end;
734 cindex = min(cindex, MAX_SACK_REPORT_BLOCKS - 1);
735 for (i = 0; i < cindex; ++i)
736 tp->sackhistory[i + 1] = copy[i];
737 tp->nsackhistory = cindex + 1;
738 tcp_sack_dump_history("after tcp_sack_update_reported_history", tp);
739}
740
741/*
742 * Fill in SACK report to return to data sender.
743 */
744void
745tcp_sack_fill_report(struct tcpcb *tp, u_char *opt, u_int *plen)
746{
747 u_int optlen = *plen;
748 uint32_t *lp = (uint32_t *)(opt + optlen);
749 uint32_t *olp;
750 tcp_seq hstart = tp->rcv_nxt, hend;
751 int nblocks;
752
753 KASSERT(TCP_MAXOLEN - optlen >=
754 TCPOLEN_SACK_ALIGNED + TCPOLEN_SACK_BLOCK,
755 ("no room for SACK header and one block: optlen %d", optlen));
756
02cc2f35
SZ
757 if (tp->t_flags & TF_DUPSEG)
758 tcpstat.tcps_snddsackopt++;
759 else
760 tcpstat.tcps_sndsackopt++;
761
91489f6b
JH
762 olp = lp++;
763 optlen += TCPOLEN_SACK_ALIGNED;
764
765 tcp_sack_ack_history(tp);
766 if (tp->reportblk.rblk_start != tp->reportblk.rblk_end) {
767 *lp++ = htonl(tp->reportblk.rblk_start);
768 *lp++ = htonl(tp->reportblk.rblk_end);
769 optlen += TCPOLEN_SACK_BLOCK;
770 hstart = tp->reportblk.rblk_start;
771 hend = tp->reportblk.rblk_end;
772 if (tp->t_flags & TF_ENCLOSESEG) {
773 KASSERT(TCP_MAXOLEN - optlen >= TCPOLEN_SACK_BLOCK,
774 ("no room for enclosing SACK block: oplen %d",
775 optlen));
776 *lp++ = htonl(tp->encloseblk.rblk_start);
777 *lp++ = htonl(tp->encloseblk.rblk_end);
778 optlen += TCPOLEN_SACK_BLOCK;
779 hstart = tp->encloseblk.rblk_start;
780 hend = tp->encloseblk.rblk_end;
781 }
782 if (SEQ_GT(hstart, tp->rcv_nxt))
783 tcp_sack_update_reported_history(tp, hstart, hend);
784 }
785 if (tcp_do_smartsack && (tp->t_flags & TF_SACKLEFT)) {
786 /* Fill in from left! Walk re-assembly queue. */
787 struct tseg_qent *q;
788
789 q = LIST_FIRST(&tp->t_segq);
790 while (q != NULL &&
791 TCP_MAXOLEN - optlen >= TCPOLEN_SACK_BLOCK) {
792 *lp++ = htonl(q->tqe_th->th_seq);
3a5d999b
SZ
793 *lp++ = htonl(TCP_SACK_BLKEND(
794 q->tqe_th->th_seq + q->tqe_len,
795 q->tqe_th->th_flags));
91489f6b
JH
796 optlen += TCPOLEN_SACK_BLOCK;
797 q = LIST_NEXT(q, tqe_q);
798 }
799 } else {
800 int n = 0;
801
802 /* Fill in SACK blocks from right side. */
803 while (n < tp->nsackhistory &&
804 TCP_MAXOLEN - optlen >= TCPOLEN_SACK_BLOCK) {
805 if (tp->sackhistory[n].rblk_start != hstart) {
806 *lp++ = htonl(tp->sackhistory[n].rblk_start);
807 *lp++ = htonl(tp->sackhistory[n].rblk_end);
808 optlen += TCPOLEN_SACK_BLOCK;
809 }
810 ++n;
811 }
812 }
813 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
814 tp->t_flags &= ~(TF_DUPSEG | TF_ENCLOSESEG | TF_SACKLEFT);
815 nblocks = (lp - olp - 1) / 2;
816 *olp = htonl(TCPOPT_SACK_ALIGNED |
817 (TCPOLEN_SACK + nblocks * TCPOLEN_SACK_BLOCK));
818 *plen = optlen;
819}