1 /*-
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)tcp_sack.c 8.12 (Berkeley) 5/24/95
30 * $FreeBSD$
31 */
32
33 /*-
34 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994
35 * The Regents of the University of California. All rights reserved.
36 *
37 * Redistribution and use in source and binary forms, with or without
38 * modification, are permitted provided that the following conditions
39 * are met:
40 * 1. Redistributions of source code must retain the above copyright
41 * notice, this list of conditions and the following disclaimer.
42 * 2. Redistributions in binary form must reproduce the above copyright
43 * notice, this list of conditions and the following disclaimer in the
44 * documentation and/or other materials provided with the distribution.
45 * 3. Neither the name of the University nor the names of its contributors
46 * may be used to endorse or promote products derived from this software
47 * without specific prior written permission.
48 *
49 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59 * SUCH DAMAGE.
60 *
61 * @@(#)COPYRIGHT 1.1 (NRL) 17 January 1995
62 *
63 * NRL grants permission for redistribution and use in source and binary
64 * forms, with or without modification, of the software and documentation
65 * created at NRL provided that the following conditions are met:
66 *
67 * 1. Redistributions of source code must retain the above copyright
68 * notice, this list of conditions and the following disclaimer.
69 * 2. Redistributions in binary form must reproduce the above copyright
70 * notice, this list of conditions and the following disclaimer in the
71 * documentation and/or other materials provided with the distribution.
72 * 3. All advertising materials mentioning features or use of this software
73 * must display the following acknowledgements:
74 * This product includes software developed by the University of
75 * California, Berkeley and its contributors.
76 * This product includes software developed at the Information
77 * Technology Division, US Naval Research Laboratory.
78 * 4. Neither the name of the NRL nor the names of its contributors
79 * may be used to endorse or promote products derived from this software
80 * without specific prior written permission.
81 *
82 * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS
83 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
84 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
85 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NRL OR
86 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
87 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
88 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
89 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
90 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
91 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
92 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
93 *
94 * The views and conclusions contained in the software and documentation
95 * are those of the authors and should not be interpreted as representing
96 * official policies, either expressed or implied, of the US Naval
97 * Research Laboratory (NRL).
98 */
99 #include "opt_inet.h"
100 #include "opt_inet6.h"
101 #include "opt_tcpdebug.h"
102 #include "opt_tcp_input.h"
103 #include "opt_tcp_sack.h"
104
105 #include <sys/param.h>
106 #include <sys/systm.h>
107 #include <sys/kernel.h>
108 #include <sys/sysctl.h>
109 #include <sys/malloc.h>
110 #include <sys/mbuf.h>
111 #include <sys/proc.h> /* for proc0 declaration */
112 #include <sys/protosw.h>
113 #include <sys/socket.h>
114 #include <sys/socketvar.h>
115 #include <sys/syslog.h>
116 #include <sys/systm.h>
117
118 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
119
120 #include <vm/uma.h>
121
122 #include <net/if.h>
123 #include <net/route.h>
124
125 #include <netinet/in.h>
126 #include <netinet/in_systm.h>
127 #include <netinet/ip.h>
128 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
129 #include <netinet/in_var.h>
130 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
131 #include <netinet/in_pcb.h>
132 #include <netinet/ip_var.h>
133 #include <netinet/ip6.h>
134 #include <netinet/icmp6.h>
135 #include <netinet6/nd6.h>
136 #include <netinet6/ip6_var.h>
137 #include <netinet6/in6_pcb.h>
138 #include <netinet/tcp.h>
139 #include <netinet/tcp_fsm.h>
140 #include <netinet/tcp_seq.h>
141 #include <netinet/tcp_timer.h>
142 #include <netinet/tcp_var.h>
143 #include <netinet6/tcp6_var.h>
144 #include <netinet/tcpip.h>
145 #ifdef TCPDEBUG
146 #include <netinet/tcp_debug.h>
147 #endif /* TCPDEBUG */
148
149 #include <machine/in_cksum.h>
150
151 extern struct uma_zone *sack_hole_zone;
152
153 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW, 0, "TCP SACK");
154 int tcp_do_sack = 1;
155 SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, enable, CTLFLAG_RW,
156 &tcp_do_sack, 0, "Enable/Disable TCP SACK support");
157 TUNABLE_INT("net.inet.tcp.sack.enable", &tcp_do_sack);
158
159 static int tcp_sack_maxholes = 128;
160 SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, maxholes, CTLFLAG_RW,
161 &tcp_sack_maxholes, 0,
162 "Maximum number of TCP SACK holes allowed per connection");
163
164 static int tcp_sack_globalmaxholes = 65536;
165 SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, globalmaxholes, CTLFLAG_RW,
166 &tcp_sack_globalmaxholes, 0,
167 "Global maximum number of TCP SACK holes");
168
169 static int tcp_sack_globalholes = 0;
170 SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, globalholes, CTLFLAG_RD,
171 &tcp_sack_globalholes, 0,
172 "Global number of TCP SACK holes currently allocated");
173
174 /*
175 * This function is called upon receipt of new valid data (while not in header
176 * prediction mode), and it updates the ordered list of sacks.
177 */
178 void
179 tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_start, tcp_seq rcv_end)
180 {
181 /*
182 * First reported block MUST be the most recent one. Subsequent
183 * blocks SHOULD be in the order in which they arrived at the
184 * receiver. These two conditions make the implementation fully
185 * compliant with RFC 2018.
186 */
187 struct sackblk head_blk, saved_blks[MAX_SACK_BLKS];
188 int num_head, num_saved, i;
189
190 INP_LOCK_ASSERT(tp->t_inpcb);
191
192 /* Check arguments */
193 KASSERT(SEQ_LT(rcv_start, rcv_end), ("rcv_start < rcv_end"));
194
195 /* SACK block for the received segment. */
196 head_blk.start = rcv_start;
197 head_blk.end = rcv_end;
198
199 /*
200 * Merge updated SACK blocks into head_blk, and
201 * save unchanged SACK blocks into saved_blks[].
202 * num_saved will have the number of the saved SACK blocks.
203 */
204 num_saved = 0;
205 for (i = 0; i < tp->rcv_numsacks; i++) {
206 tcp_seq start = tp->sackblks[i].start;
207 tcp_seq end = tp->sackblks[i].end;
208 if (SEQ_GEQ(start, end) || SEQ_LEQ(start, tp->rcv_nxt)) {
209 /*
210 * Discard this SACK block.
211 */
212 } else if (SEQ_LEQ(head_blk.start, end) &&
213 SEQ_GEQ(head_blk.end, start)) {
214 /*
215 * Merge this SACK block into head_blk.
216 * This SACK block itself will be discarded.
217 */
218 if (SEQ_GT(head_blk.start, start))
219 head_blk.start = start;
220 if (SEQ_LT(head_blk.end, end))
221 head_blk.end = end;
222 } else {
223 /*
224 * Save this SACK block.
225 */
226 saved_blks[num_saved].start = start;
227 saved_blks[num_saved].end = end;
228 num_saved++;
229 }
230 }
231
232 /*
233 * Update SACK list in tp->sackblks[].
234 */
235 num_head = 0;
236 if (SEQ_GT(head_blk.start, tp->rcv_nxt)) {
237 /*
238 * The received data segment is an out-of-order segment.
239 * Put head_blk at the top of SACK list.
240 */
241 tp->sackblks[0] = head_blk;
242 num_head = 1;
243 /*
244 * If the number of saved SACK blocks exceeds its limit,
245 * discard the last SACK block.
246 */
247 if (num_saved >= MAX_SACK_BLKS)
248 num_saved--;
249 }
250 if (num_saved > 0) {
251 /*
252 * Copy the saved SACK blocks back.
253 */
254 bcopy(saved_blks, &tp->sackblks[num_head],
255 sizeof(struct sackblk) * num_saved);
256 }
257
258 /* Save the number of SACK blocks. */
259 tp->rcv_numsacks = num_head + num_saved;
260 }
261
262 /*
263 * Delete all receiver-side SACK information.
264 */
265 void
266 tcp_clean_sackreport(tp)
267 struct tcpcb *tp;
268 {
269 int i;
270
271 INP_LOCK_ASSERT(tp->t_inpcb);
272 tp->rcv_numsacks = 0;
273 for (i = 0; i < MAX_SACK_BLKS; i++)
274 tp->sackblks[i].start = tp->sackblks[i].end=0;
275 }
276
277 /*
278 * Allocate struct sackhole.
279 */
280 static struct sackhole *
281 tcp_sackhole_alloc(struct tcpcb *tp, tcp_seq start, tcp_seq end)
282 {
283 struct sackhole *hole;
284
285 if (tp->snd_numholes >= tcp_sack_maxholes ||
286 tcp_sack_globalholes >= tcp_sack_globalmaxholes) {
287 tcpstat.tcps_sack_sboverflow++;
288 return NULL;
289 }
290
291 hole = (struct sackhole *)uma_zalloc(sack_hole_zone, M_NOWAIT);
292 if (hole == NULL)
293 return NULL;
294
295 hole->start = start;
296 hole->end = end;
297 hole->rxmit = start;
298
299 tp->snd_numholes++;
300 tcp_sack_globalholes++;
301
302 return hole;
303 }
304
305 /*
306 * Free struct sackhole.
307 */
308 static void
309 tcp_sackhole_free(struct tcpcb *tp, struct sackhole *hole)
310 {
311 uma_zfree(sack_hole_zone, hole);
312
313 tp->snd_numholes--;
314 tcp_sack_globalholes--;
315
316 KASSERT(tp->snd_numholes >= 0, ("tp->snd_numholes >= 0"));
317 KASSERT(tcp_sack_globalholes >= 0, ("tcp_sack_globalholes >= 0"));
318 }
319
320 /*
321 * Insert new SACK hole into scoreboard.
322 */
323 static struct sackhole *
324 tcp_sackhole_insert(struct tcpcb *tp, tcp_seq start, tcp_seq end,
325 struct sackhole *after)
326 {
327 struct sackhole *hole;
328
329 /* Allocate a new SACK hole. */
330 hole = tcp_sackhole_alloc(tp, start, end);
331 if (hole == NULL)
332 return NULL;
333
334 /* Insert the new SACK hole into scoreboard */
335 if (after != NULL)
336 TAILQ_INSERT_AFTER(&tp->snd_holes, after, hole, scblink);
337 else
338 TAILQ_INSERT_TAIL(&tp->snd_holes, hole, scblink);
339
340 /* Update SACK hint. */
341 if (tp->sackhint.nexthole == NULL)
342 tp->sackhint.nexthole = hole;
343
344 return hole;
345 }
346
347 /*
348 * Remove SACK hole from scoreboard.
349 */
350 static void
351 tcp_sackhole_remove(struct tcpcb *tp, struct sackhole *hole)
352 {
353 /* Update SACK hint. */
354 if (tp->sackhint.nexthole == hole)
355 tp->sackhint.nexthole = TAILQ_NEXT(hole, scblink);
356
357 /* Remove this SACK hole. */
358 TAILQ_REMOVE(&tp->snd_holes, hole, scblink);
359
360 /* Free this SACK hole. */
361 tcp_sackhole_free(tp, hole);
362 }
363
364 /*
365 * Process cumulative ACK and the TCP SACK option to update the scoreboard.
366 * tp->snd_holes is an ordered list of holes (oldest to newest, in terms of
367 * the sequence space).
368 */
369 void
370 tcp_sack_doack(struct tcpcb *tp, struct tcpopt *to, tcp_seq th_ack)
371 {
372 struct sackhole *cur, *temp;
373 struct sackblk sack, sack_blocks[TCP_MAX_SACK + 1], *sblkp;
374 int i, j, num_sack_blks;
375
376 INP_LOCK_ASSERT(tp->t_inpcb);
377
378 num_sack_blks = 0;
379 /*
380 * If SND.UNA will be advanced by SEG.ACK, and if SACK holes exist,
381 * treat [SND.UNA, SEG.ACK) as if it is a SACK block.
382 */
383 if (SEQ_LT(tp->snd_una, th_ack) && !TAILQ_EMPTY(&tp->snd_holes)) {
384 sack_blocks[num_sack_blks].start = tp->snd_una;
385 sack_blocks[num_sack_blks++].end = th_ack;
386 }
387 /*
388 * Append received valid SACK blocks to sack_blocks[].
389 */
390 for (i = 0; i < to->to_nsacks; i++) {
391 bcopy((to->to_sacks + i * TCPOLEN_SACK), &sack, sizeof(sack));
392 sack.start = ntohl(sack.start);
393 sack.end = ntohl(sack.end);
394 if (SEQ_GT(sack.end, sack.start) &&
395 SEQ_GT(sack.start, tp->snd_una) &&
396 SEQ_GT(sack.start, th_ack) &&
397 SEQ_LT(sack.start, tp->snd_max) &&
398 SEQ_GT(sack.end, tp->snd_una) &&
399 SEQ_LEQ(sack.end, tp->snd_max))
400 sack_blocks[num_sack_blks++] = sack;
401 }
402
403 /*
404 * Return if SND.UNA is not advanced and no valid SACK block
405 * is received.
406 */
407 if (num_sack_blks == 0)
408 return;
409
410 /*
411 * Sort the SACK blocks so we can update the scoreboard
412 * with just one pass. The overhead of sorting upto 4+1 elements
413 * is less than making upto 4+1 passes over the scoreboard.
414 */
415 for (i = 0; i < num_sack_blks; i++) {
416 for (j = i + 1; j < num_sack_blks; j++) {
417 if (SEQ_GT(sack_blocks[i].end, sack_blocks[j].end)) {
418 sack = sack_blocks[i];
419 sack_blocks[i] = sack_blocks[j];
420 sack_blocks[j] = sack;
421 }
422 }
423 }
424 if (TAILQ_EMPTY(&tp->snd_holes))
425 /*
426 * Empty scoreboard. Need to initialize snd_fack (it may be
427 * uninitialized or have a bogus value). Scoreboard holes
428 * (from the sack blocks received) are created later below (in
429 * the logic that adds holes to the tail of the scoreboard).
430 */
431 tp->snd_fack = SEQ_MAX(tp->snd_una, th_ack);
432 /*
433 * In the while-loop below, incoming SACK blocks (sack_blocks[])
434 * and SACK holes (snd_holes) are traversed from their tails with
435 * just one pass in order to reduce the number of compares especially
436 * when the bandwidth-delay product is large.
437 * Note: Typically, in the first RTT of SACK recovery, the highest
438 * three or four SACK blocks with the same ack number are received.
439 * In the second RTT, if retransmitted data segments are not lost,
440 * the highest three or four SACK blocks with ack number advancing
441 * are received.
442 */
443 sblkp = &sack_blocks[num_sack_blks - 1]; /* Last SACK block */
444 if (SEQ_LT(tp->snd_fack, sblkp->start)) {
445 /*
446 * The highest SACK block is beyond fack.
447 * Append new SACK hole at the tail.
448 * If the second or later highest SACK blocks are also
449 * beyond the current fack, they will be inserted by
450 * way of hole splitting in the while-loop below.
451 */
452 temp = tcp_sackhole_insert(tp, tp->snd_fack,sblkp->start,NULL);
453 if (temp != NULL) {
454 tp->snd_fack = sblkp->end;
455 /* Go to the previous sack block. */
456 sblkp--;
457 } else {
458 /*
459 * We failed to add a new hole based on the current
460 * sack block. Skip over all the sack blocks that
461 * fall completely to the right of snd_fack and proceed
462 * to trim the scoreboard based on the remaining sack
463 * blocks. This also trims the scoreboard for th_ack
464 * (which is sack_blocks[0]).
465 */
466 while (sblkp >= sack_blocks &&
467 SEQ_LT(tp->snd_fack, sblkp->start))
468 sblkp--;
469 if (sblkp >= sack_blocks &&
470 SEQ_LT(tp->snd_fack, sblkp->end))
471 tp->snd_fack = sblkp->end;
472 }
473 } else if (SEQ_LT(tp->snd_fack, sblkp->end))
474 /* fack is advanced. */
475 tp->snd_fack = sblkp->end;
476 /* We must have at least one SACK hole in scoreboard */
477 KASSERT(!TAILQ_EMPTY(&tp->snd_holes), ("SACK scoreboard must not be empty"));
478 cur = TAILQ_LAST(&tp->snd_holes, sackhole_head); /* Last SACK hole */
479 /*
480 * Since the incoming sack blocks are sorted, we can process them
481 * making one sweep of the scoreboard.
482 */
483 while (sblkp >= sack_blocks && cur != NULL) {
484 if (SEQ_GEQ(sblkp->start, cur->end)) {
485 /*
486 * SACKs data beyond the current hole.
487 * Go to the previous sack block.
488 */
489 sblkp--;
490 continue;
491 }
492 if (SEQ_LEQ(sblkp->end, cur->start)) {
493 /*
494 * SACKs data before the current hole.
495 * Go to the previous hole.
496 */
497 cur = TAILQ_PREV(cur, sackhole_head, scblink);
498 continue;
499 }
500 tp->sackhint.sack_bytes_rexmit -= (cur->rxmit - cur->start);
501 KASSERT(tp->sackhint.sack_bytes_rexmit >= 0,
502 ("sackhint bytes rtx >= 0"));
503 if (SEQ_LEQ(sblkp->start, cur->start)) {
504 /* Data acks at least the beginning of hole */
505 if (SEQ_GEQ(sblkp->end, cur->end)) {
506 /* Acks entire hole, so delete hole */
507 temp = cur;
508 cur = TAILQ_PREV(cur, sackhole_head, scblink);
509 tcp_sackhole_remove(tp, temp);
510 /*
511 * The sack block may ack all or part of the next
512 * hole too, so continue onto the next hole.
513 */
514 continue;
515 } else {
516 /* Move start of hole forward */
517 cur->start = sblkp->end;
518 cur->rxmit = SEQ_MAX(cur->rxmit, cur->start);
519 }
520 } else {
521 /* Data acks at least the end of hole */
522 if (SEQ_GEQ(sblkp->end, cur->end)) {
523 /* Move end of hole backward */
524 cur->end = sblkp->start;
525 cur->rxmit = SEQ_MIN(cur->rxmit, cur->end);
526 } else {
527 /*
528 * ACKs some data in middle of a hole; need to
529 * split current hole
530 */
531 temp = tcp_sackhole_insert(tp, sblkp->end,
532 cur->end, cur);
533 if (temp != NULL) {
534 if (SEQ_GT(cur->rxmit, temp->rxmit)) {
535 temp->rxmit = cur->rxmit;
536 tp->sackhint.sack_bytes_rexmit
537 += (temp->rxmit
538 - temp->start);
539 }
540 cur->end = sblkp->start;
541 cur->rxmit = SEQ_MIN(cur->rxmit,
542 cur->end);
543 }
544 }
545 }
546 tp->sackhint.sack_bytes_rexmit += (cur->rxmit - cur->start);
547 /*
548 * Testing sblkp->start against cur->start tells us whether
549 * we're done with the sack block or the sack hole.
550 * Accordingly, we advance one or the other.
551 */
552 if (SEQ_LEQ(sblkp->start, cur->start))
553 cur = TAILQ_PREV(cur, sackhole_head, scblink);
554 else
555 sblkp--;
556 }
557 }
558
559 /*
560 * Free all SACK holes to clear the scoreboard.
561 */
562 void
563 tcp_free_sackholes(struct tcpcb *tp)
564 {
565 struct sackhole *q;
566
567 INP_LOCK_ASSERT(tp->t_inpcb);
568 while ((q = TAILQ_FIRST(&tp->snd_holes)) != NULL)
569 tcp_sackhole_remove(tp, q);
570 tp->sackhint.sack_bytes_rexmit = 0;
571
572 KASSERT(tp->snd_numholes == 0, ("tp->snd_numholes == 0"));
573 KASSERT(tp->sackhint.nexthole == NULL,
574 ("tp->sackhint.nexthole == NULL"));
575 }
576
577 /*
578 * Partial ack handling within a sack recovery episode.
579 * Keeping this very simple for now. When a partial ack
580 * is received, force snd_cwnd to a value that will allow
581 * the sender to transmit no more than 2 segments.
582 * If necessary, a better scheme can be adopted at a
583 * later point, but for now, the goal is to prevent the
584 * sender from bursting a large amount of data in the midst
585 * of sack recovery.
586 */
587 void
588 tcp_sack_partialack(tp, th)
589 struct tcpcb *tp;
590 struct tcphdr *th;
591 {
592 int num_segs = 1;
593
594 INP_LOCK_ASSERT(tp->t_inpcb);
595 callout_stop(tp->tt_rexmt);
596 tp->t_rtttime = 0;
597 /* send one or 2 segments based on how much new data was acked */
598 if (((th->th_ack - tp->snd_una) / tp->t_maxseg) > 2)
599 num_segs = 2;
600 tp->snd_cwnd = (tp->sackhint.sack_bytes_rexmit +
601 (tp->snd_nxt - tp->sack_newdata) +
602 num_segs * tp->t_maxseg);
603 if (tp->snd_cwnd > tp->snd_ssthresh)
604 tp->snd_cwnd = tp->snd_ssthresh;
605 tp->t_flags |= TF_ACKNOW;
606 (void) tcp_output(tp);
607 }
608
609 /*
610 * Debug version of tcp_sack_output() that walks the scoreboard. Used for
611 * now to sanity check the hint.
612 */
613 static struct sackhole *
614 tcp_sack_output_debug(struct tcpcb *tp, int *sack_bytes_rexmt)
615 {
616 struct sackhole *p;
617
618 INP_LOCK_ASSERT(tp->t_inpcb);
619 *sack_bytes_rexmt = 0;
620 TAILQ_FOREACH(p, &tp->snd_holes, scblink) {
621 if (SEQ_LT(p->rxmit, p->end)) {
622 if (SEQ_LT(p->rxmit, tp->snd_una)) {/* old SACK hole */
623 continue;
624 }
625 *sack_bytes_rexmt += (p->rxmit - p->start);
626 break;
627 }
628 *sack_bytes_rexmt += (p->rxmit - p->start);
629 }
630 return (p);
631 }
632
633 /*
634 * Returns the next hole to retransmit and the number of retransmitted bytes
635 * from the scoreboard. We store both the next hole and the number of
636 * retransmitted bytes as hints (and recompute these on the fly upon SACK/ACK
637 * reception). This avoids scoreboard traversals completely.
638 *
639 * The loop here will traverse *at most* one link. Here's the argument.
640 * For the loop to traverse more than 1 link before finding the next hole to
641 * retransmit, we would need to have at least 1 node following the current hint
642 * with (rxmit == end). But, for all holes following the current hint,
643 * (start == rxmit), since we have not yet retransmitted from them. Therefore,
644 * in order to traverse more 1 link in the loop below, we need to have at least
645 * one node following the current hint with (start == rxmit == end).
646 * But that can't happen, (start == end) means that all the data in that hole
647 * has been sacked, in which case, the hole would have been removed from the
648 * scoreboard.
649 */
650 struct sackhole *
651 tcp_sack_output(struct tcpcb *tp, int *sack_bytes_rexmt)
652 {
653 struct sackhole *hole = NULL, *dbg_hole = NULL;
654 int dbg_bytes_rexmt;
655
656 INP_LOCK_ASSERT(tp->t_inpcb);
657 dbg_hole = tcp_sack_output_debug(tp, &dbg_bytes_rexmt);
658 *sack_bytes_rexmt = tp->sackhint.sack_bytes_rexmit;
659 hole = tp->sackhint.nexthole;
660 if (hole == NULL || SEQ_LT(hole->rxmit, hole->end))
661 goto out;
662 while ((hole = TAILQ_NEXT(hole, scblink)) != NULL) {
663 if (SEQ_LT(hole->rxmit, hole->end)) {
664 tp->sackhint.nexthole = hole;
665 break;
666 }
667 }
668 out:
669 if (dbg_hole != hole) {
670 printf("%s: Computed sack hole not the same as cached value\n", __func__);
671 hole = dbg_hole;
672 }
673 if (*sack_bytes_rexmt != dbg_bytes_rexmt) {
674 printf("%s: Computed sack_bytes_retransmitted (%d) not "
675 "the same as cached value (%d)\n",
676 __func__, dbg_bytes_rexmt, *sack_bytes_rexmt);
677 *sack_bytes_rexmt = dbg_bytes_rexmt;
678 }
679 return (hole);
680 }
681
682 /*
683 * After a timeout, the SACK list may be rebuilt. This SACK information
684 * should be used to avoid retransmitting SACKed data. This function
685 * traverses the SACK list to see if snd_nxt should be moved forward.
686 */
687 void
688 tcp_sack_adjust(struct tcpcb *tp)
689 {
690 struct sackhole *p, *cur = TAILQ_FIRST(&tp->snd_holes);
691
692 INP_LOCK_ASSERT(tp->t_inpcb);
693 if (cur == NULL)
694 return; /* No holes */
695 if (SEQ_GEQ(tp->snd_nxt, tp->snd_fack))
696 return; /* We're already beyond any SACKed blocks */
697 /*
698 * Two cases for which we want to advance snd_nxt:
699 * i) snd_nxt lies between end of one hole and beginning of another
700 * ii) snd_nxt lies between end of last hole and snd_fack
701 */
702 while ((p = TAILQ_NEXT(cur, scblink)) != NULL) {
703 if (SEQ_LT(tp->snd_nxt, cur->end))
704 return;
705 if (SEQ_GEQ(tp->snd_nxt, p->start))
706 cur = p;
707 else {
708 tp->snd_nxt = p->start;
709 return;
710 }
711 }
712 if (SEQ_LT(tp->snd_nxt, cur->end))
713 return;
714 tp->snd_nxt = tp->snd_fack;
715 return;
716 }
Cache object: 61b177230c65bc4d190ebc3093fcc115
|