]>
Commit | Line | Data |
---|---|---|
8ad349bb | 1 | /* |
2d21ac55 | 2 | * Copyright (c) 2004,2007 Apple Inc. All rights reserved. |
5d5c5d0d | 3 | * |
2d21ac55 | 4 | * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ |
8ad349bb | 5 | * |
2d21ac55 A |
6 | * This file contains Original Code and/or Modifications of Original Code |
7 | * as defined in and that are subject to the Apple Public Source License | |
8 | * Version 2.0 (the 'License'). You may not use this file except in | |
9 | * compliance with the License. The rights granted to you under the License | |
10 | * may not be used to create, or enable the creation or redistribution of, | |
11 | * unlawful or unlicensed copies of an Apple operating system, or to | |
12 | * circumvent, violate, or enable the circumvention or violation of, any | |
13 | * terms of an Apple operating system software license agreement. | |
8f6c56a5 | 14 | * |
2d21ac55 A |
15 | * Please obtain a copy of the License at |
16 | * http://www.opensource.apple.com/apsl/ and read it before using this file. | |
17 | * | |
18 | * The Original Code and all software distributed under the License are | |
19 | * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER | |
8f6c56a5 A |
20 | * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, |
21 | * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, | |
2d21ac55 A |
22 | * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. |
23 | * Please see the License for the specific language governing rights and | |
24 | * limitations under the License. | |
8f6c56a5 | 25 | * |
2d21ac55 | 26 | * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ |
8ad349bb A |
27 | */ |
28 | /* | |
29 | * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 | |
30 | * The Regents of the University of California. All rights reserved. | |
31 | * | |
32 | * Redistribution and use in source and binary forms, with or without | |
33 | * modification, are permitted provided that the following conditions | |
34 | * are met: | |
35 | * 1. Redistributions of source code must retain the above copyright | |
36 | * notice, this list of conditions and the following disclaimer. | |
37 | * 2. Redistributions in binary form must reproduce the above copyright | |
38 | * notice, this list of conditions and the following disclaimer in the | |
39 | * documentation and/or other materials provided with the distribution. | |
40 | * 3. All advertising materials mentioning features or use of this software | |
41 | * must display the following acknowledgement: | |
42 | * This product includes software developed by the University of | |
43 | * California, Berkeley and its contributors. | |
44 | * 4. Neither the name of the University nor the names of its contributors | |
45 | * may be used to endorse or promote products derived from this software | |
46 | * without specific prior written permission. | |
47 | * | |
48 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND | |
49 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE | |
50 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE | |
51 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE | |
52 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL | |
53 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS | |
54 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) | |
55 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT | |
56 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY | |
57 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF | |
58 | * SUCH DAMAGE. | |
59 | * | |
60 | */ | |
61 | ||
62 | #define _IP_VHL | |
63 | ||
64 | ||
65 | #include <sys/param.h> | |
66 | #include <sys/systm.h> | |
67 | #include <sys/kernel.h> | |
68 | #include <sys/sysctl.h> | |
69 | #include <sys/mbuf.h> | |
70 | #include <sys/domain.h> | |
71 | #include <sys/protosw.h> | |
72 | #include <sys/socket.h> | |
73 | #include <sys/socketvar.h> | |
74 | ||
2d21ac55 A |
75 | #include <kern/zalloc.h> |
76 | ||
8ad349bb A |
77 | #include <net/route.h> |
78 | ||
79 | #include <netinet/in.h> | |
80 | #include <netinet/in_systm.h> | |
81 | #include <netinet/ip.h> | |
82 | #include <netinet/in_pcb.h> | |
83 | #include <netinet/ip_var.h> | |
84 | #if INET6 | |
85 | #include <netinet6/in6_pcb.h> | |
86 | #include <netinet/ip6.h> | |
87 | #include <netinet6/ip6_var.h> | |
88 | #endif | |
89 | #include <netinet/tcp.h> | |
90 | //#define TCPOUTFLAGS | |
91 | #include <netinet/tcp_fsm.h> | |
92 | #include <netinet/tcp_seq.h> | |
93 | #include <netinet/tcp_timer.h> | |
94 | #include <netinet/tcp_var.h> | |
95 | #include <netinet/tcpip.h> | |
96 | #if TCPDEBUG | |
97 | #include <netinet/tcp_debug.h> | |
98 | #endif | |
99 | #include <sys/kdebug.h> | |
100 | ||
101 | #if IPSEC | |
102 | #include <netinet6/ipsec.h> | |
103 | #endif /*IPSEC*/ | |
104 | ||
105 | int tcp_do_sack = 1; | |
6d2010ae | 106 | SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_do_sack, 0, |
8ad349bb A |
107 | "Enable/Disable TCP SACK support"); |
108 | static int tcp_sack_maxholes = 128; | |
6d2010ae | 109 | SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack_maxholes, CTLFLAG_RW | CTLFLAG_LOCKED, |
8ad349bb A |
110 | &tcp_sack_maxholes, 0, |
111 | "Maximum number of TCP SACK holes allowed per connection"); | |
112 | ||
113 | static int tcp_sack_globalmaxholes = 65536; | |
6d2010ae | 114 | SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack_globalmaxholes, CTLFLAG_RW | CTLFLAG_LOCKED, |
8ad349bb A |
115 | &tcp_sack_globalmaxholes, 0, |
116 | "Global maximum number of TCP SACK holes"); | |
117 | ||
118 | static int tcp_sack_globalholes = 0; | |
6d2010ae | 119 | SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack_globalholes, CTLFLAG_RD | CTLFLAG_LOCKED, |
8ad349bb A |
120 | &tcp_sack_globalholes, 0, |
121 | "Global number of TCP SACK holes currently allocated"); | |
122 | ||
123 | extern struct zone *sack_hole_zone; | |
124 | ||
125 | /* | |
126 | * This function is called upon receipt of new valid data (while not in header | |
127 | * prediction mode), and it updates the ordered list of sacks. | |
128 | */ | |
129 | void | |
130 | tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_start, tcp_seq rcv_end) | |
131 | { | |
132 | /* | |
133 | * First reported block MUST be the most recent one. Subsequent | |
134 | * blocks SHOULD be in the order in which they arrived at the | |
135 | * receiver. These two conditions make the implementation fully | |
136 | * compliant with RFC 2018. | |
137 | */ | |
138 | struct sackblk head_blk, saved_blks[MAX_SACK_BLKS]; | |
139 | int num_head, num_saved, i; | |
140 | ||
141 | /* SACK block for the received segment. */ | |
142 | head_blk.start = rcv_start; | |
143 | head_blk.end = rcv_end; | |
144 | ||
145 | /* | |
146 | * Merge updated SACK blocks into head_blk, and | |
147 | * save unchanged SACK blocks into saved_blks[]. | |
148 | * num_saved will have the number of the saved SACK blocks. | |
149 | */ | |
150 | num_saved = 0; | |
151 | for (i = 0; i < tp->rcv_numsacks; i++) { | |
152 | tcp_seq start = tp->sackblks[i].start; | |
153 | tcp_seq end = tp->sackblks[i].end; | |
154 | if (SEQ_GEQ(start, end) || SEQ_LEQ(start, tp->rcv_nxt)) { | |
155 | /* | |
156 | * Discard this SACK block. | |
157 | */ | |
158 | } else if (SEQ_LEQ(head_blk.start, end) && | |
159 | SEQ_GEQ(head_blk.end, start)) { | |
160 | /* | |
161 | * Merge this SACK block into head_blk. | |
162 | * This SACK block itself will be discarded. | |
163 | */ | |
164 | if (SEQ_GT(head_blk.start, start)) | |
165 | head_blk.start = start; | |
166 | if (SEQ_LT(head_blk.end, end)) | |
167 | head_blk.end = end; | |
168 | } else { | |
169 | /* | |
170 | * Save this SACK block. | |
171 | */ | |
172 | saved_blks[num_saved].start = start; | |
173 | saved_blks[num_saved].end = end; | |
174 | num_saved++; | |
175 | } | |
176 | } | |
177 | ||
178 | /* | |
179 | * Update SACK list in tp->sackblks[]. | |
180 | */ | |
181 | num_head = 0; | |
182 | if (SEQ_GT(head_blk.start, tp->rcv_nxt)) { | |
183 | /* | |
184 | * The received data segment is an out-of-order segment. | |
185 | * Put head_blk at the top of SACK list. | |
186 | */ | |
187 | tp->sackblks[0] = head_blk; | |
188 | num_head = 1; | |
189 | /* | |
190 | * If the number of saved SACK blocks exceeds its limit, | |
191 | * discard the last SACK block. | |
192 | */ | |
193 | if (num_saved >= MAX_SACK_BLKS) | |
194 | num_saved--; | |
195 | } | |
196 | if (num_saved > 0) { | |
197 | /* | |
198 | * Copy the saved SACK blocks back. | |
199 | */ | |
200 | bcopy(saved_blks, &tp->sackblks[num_head], | |
201 | sizeof(struct sackblk) * num_saved); | |
202 | } | |
203 | ||
204 | /* Save the number of SACK blocks. */ | |
205 | tp->rcv_numsacks = num_head + num_saved; | |
6d2010ae A |
206 | |
207 | /* If we are requesting SACK recovery, reset the stretch-ack state | |
208 | * so that connection will generate more acks after recovery and | |
209 | * sender's cwnd will open. | |
210 | */ | |
211 | if ((tp->t_flags & TF_STRETCHACK) != 0 && tp->rcv_numsacks > 0) | |
212 | tcp_reset_stretch_ack(tp); | |
213 | ||
214 | #if TRAFFIC_MGT | |
215 | if (tp->acc_iaj > 0 && tp->rcv_numsacks > 0) | |
216 | reset_acc_iaj(tp); | |
217 | #endif /* TRAFFIC_MGT */ | |
8ad349bb A |
218 | } |
219 | ||
220 | /* | |
221 | * Delete all receiver-side SACK information. | |
222 | */ | |
223 | void | |
224 | tcp_clean_sackreport( struct tcpcb *tp) | |
225 | { | |
8ad349bb A |
226 | |
227 | tp->rcv_numsacks = 0; | |
8ad349bb A |
228 | bzero(&tp->sackblks[0], sizeof (struct sackblk) * MAX_SACK_BLKS); |
229 | } | |
230 | ||
231 | /* | |
232 | * Allocate struct sackhole. | |
233 | */ | |
234 | static struct sackhole * | |
235 | tcp_sackhole_alloc(struct tcpcb *tp, tcp_seq start, tcp_seq end) | |
236 | { | |
237 | struct sackhole *hole; | |
238 | ||
239 | if (tp->snd_numholes >= tcp_sack_maxholes || | |
240 | tcp_sack_globalholes >= tcp_sack_globalmaxholes) { | |
241 | tcpstat.tcps_sack_sboverflow++; | |
242 | return NULL; | |
243 | } | |
244 | ||
245 | hole = (struct sackhole *)zalloc_noblock(sack_hole_zone); | |
246 | if (hole == NULL) | |
247 | return NULL; | |
248 | ||
249 | hole->start = start; | |
250 | hole->end = end; | |
251 | hole->rxmit = start; | |
252 | ||
253 | tp->snd_numholes++; | |
254 | tcp_sack_globalholes++; | |
255 | ||
256 | return hole; | |
257 | } | |
258 | ||
259 | /* | |
260 | * Free struct sackhole. | |
261 | */ | |
262 | static void | |
263 | tcp_sackhole_free(struct tcpcb *tp, struct sackhole *hole) | |
264 | { | |
265 | zfree(sack_hole_zone, hole); | |
266 | ||
267 | tp->snd_numholes--; | |
268 | tcp_sack_globalholes--; | |
269 | } | |
270 | ||
271 | /* | |
272 | * Insert new SACK hole into scoreboard. | |
273 | */ | |
274 | static struct sackhole * | |
275 | tcp_sackhole_insert(struct tcpcb *tp, tcp_seq start, tcp_seq end, | |
276 | struct sackhole *after) | |
277 | { | |
278 | struct sackhole *hole; | |
279 | ||
280 | /* Allocate a new SACK hole. */ | |
281 | hole = tcp_sackhole_alloc(tp, start, end); | |
282 | if (hole == NULL) | |
283 | return NULL; | |
284 | ||
285 | /* Insert the new SACK hole into scoreboard */ | |
286 | if (after != NULL) | |
287 | TAILQ_INSERT_AFTER(&tp->snd_holes, after, hole, scblink); | |
288 | else | |
289 | TAILQ_INSERT_TAIL(&tp->snd_holes, hole, scblink); | |
290 | ||
291 | /* Update SACK hint. */ | |
292 | if (tp->sackhint.nexthole == NULL) | |
293 | tp->sackhint.nexthole = hole; | |
294 | ||
295 | return hole; | |
296 | } | |
297 | ||
298 | /* | |
299 | * Remove SACK hole from scoreboard. | |
300 | */ | |
301 | static void | |
302 | tcp_sackhole_remove(struct tcpcb *tp, struct sackhole *hole) | |
303 | { | |
304 | /* Update SACK hint. */ | |
305 | if (tp->sackhint.nexthole == hole) | |
306 | tp->sackhint.nexthole = TAILQ_NEXT(hole, scblink); | |
307 | ||
308 | /* Remove this SACK hole. */ | |
309 | TAILQ_REMOVE(&tp->snd_holes, hole, scblink); | |
310 | ||
311 | /* Free this SACK hole. */ | |
312 | tcp_sackhole_free(tp, hole); | |
313 | } | |
314 | ||
315 | /* | |
316 | * Process cumulative ACK and the TCP SACK option to update the scoreboard. | |
317 | * tp->snd_holes is an ordered list of holes (oldest to newest, in terms of | |
318 | * the sequence space). | |
319 | */ | |
320 | void | |
321 | tcp_sack_doack(struct tcpcb *tp, struct tcpopt *to, tcp_seq th_ack) | |
322 | { | |
323 | struct sackhole *cur, *temp; | |
324 | struct sackblk sack, sack_blocks[TCP_MAX_SACK + 1], *sblkp; | |
325 | int i, j, num_sack_blks; | |
326 | ||
327 | num_sack_blks = 0; | |
328 | /* | |
329 | * If SND.UNA will be advanced by SEG.ACK, and if SACK holes exist, | |
330 | * treat [SND.UNA, SEG.ACK) as if it is a SACK block. | |
331 | */ | |
332 | if (SEQ_LT(tp->snd_una, th_ack) && !TAILQ_EMPTY(&tp->snd_holes)) { | |
333 | sack_blocks[num_sack_blks].start = tp->snd_una; | |
334 | sack_blocks[num_sack_blks++].end = th_ack; | |
335 | } | |
336 | /* | |
337 | * Append received valid SACK blocks to sack_blocks[]. | |
b0d623f7 | 338 | * Check that the SACK block range is valid. |
8ad349bb | 339 | */ |
b0d623f7 A |
340 | for (i = 0; i < to->to_nsacks; i++) { |
341 | bcopy((to->to_sacks + i * TCPOLEN_SACK), | |
342 | &sack, sizeof(sack)); | |
343 | sack.start = ntohl(sack.start); | |
344 | sack.end = ntohl(sack.end); | |
345 | if (SEQ_GT(sack.end, sack.start) && | |
346 | SEQ_GT(sack.start, tp->snd_una) && | |
347 | SEQ_GT(sack.start, th_ack) && | |
348 | SEQ_LT(sack.start, tp->snd_max) && | |
349 | SEQ_GT(sack.end, tp->snd_una) && | |
350 | SEQ_LEQ(sack.end, tp->snd_max)) | |
351 | sack_blocks[num_sack_blks++] = sack; | |
8ad349bb A |
352 | } |
353 | ||
354 | /* | |
355 | * Return if SND.UNA is not advanced and no valid SACK block | |
356 | * is received. | |
357 | */ | |
358 | if (num_sack_blks == 0) | |
359 | return; | |
360 | ||
361 | /* | |
362 | * Sort the SACK blocks so we can update the scoreboard | |
363 | * with just one pass. The overhead of sorting upto 4+1 elements | |
364 | * is less than making upto 4+1 passes over the scoreboard. | |
365 | */ | |
366 | for (i = 0; i < num_sack_blks; i++) { | |
367 | for (j = i + 1; j < num_sack_blks; j++) { | |
368 | if (SEQ_GT(sack_blocks[i].end, sack_blocks[j].end)) { | |
369 | sack = sack_blocks[i]; | |
370 | sack_blocks[i] = sack_blocks[j]; | |
371 | sack_blocks[j] = sack; | |
372 | } | |
373 | } | |
374 | } | |
375 | if (TAILQ_EMPTY(&tp->snd_holes)) | |
376 | /* | |
377 | * Empty scoreboard. Need to initialize snd_fack (it may be | |
378 | * uninitialized or have a bogus value). Scoreboard holes | |
379 | * (from the sack blocks received) are created later below (in | |
380 | * the logic that adds holes to the tail of the scoreboard). | |
381 | */ | |
382 | tp->snd_fack = SEQ_MAX(tp->snd_una, th_ack); | |
383 | /* | |
384 | * In the while-loop below, incoming SACK blocks (sack_blocks[]) | |
385 | * and SACK holes (snd_holes) are traversed from their tails with | |
386 | * just one pass in order to reduce the number of compares especially | |
387 | * when the bandwidth-delay product is large. | |
388 | * Note: Typically, in the first RTT of SACK recovery, the highest | |
389 | * three or four SACK blocks with the same ack number are received. | |
390 | * In the second RTT, if retransmitted data segments are not lost, | |
391 | * the highest three or four SACK blocks with ack number advancing | |
392 | * are received. | |
393 | */ | |
394 | sblkp = &sack_blocks[num_sack_blks - 1]; /* Last SACK block */ | |
395 | if (SEQ_LT(tp->snd_fack, sblkp->start)) { | |
396 | /* | |
397 | * The highest SACK block is beyond fack. | |
398 | * Append new SACK hole at the tail. | |
399 | * If the second or later highest SACK blocks are also | |
400 | * beyond the current fack, they will be inserted by | |
401 | * way of hole splitting in the while-loop below. | |
402 | */ | |
403 | temp = tcp_sackhole_insert(tp, tp->snd_fack,sblkp->start,NULL); | |
404 | if (temp != NULL) { | |
405 | tp->snd_fack = sblkp->end; | |
406 | /* Go to the previous sack block. */ | |
407 | sblkp--; | |
408 | } else { | |
409 | /* | |
410 | * We failed to add a new hole based on the current | |
411 | * sack block. Skip over all the sack blocks that | |
412 | * fall completely to the right of snd_fack and proceed | |
413 | * to trim the scoreboard based on the remaining sack | |
414 | * blocks. This also trims the scoreboard for th_ack | |
415 | * (which is sack_blocks[0]). | |
416 | */ | |
417 | while (sblkp >= sack_blocks && | |
418 | SEQ_LT(tp->snd_fack, sblkp->start)) | |
419 | sblkp--; | |
420 | if (sblkp >= sack_blocks && | |
421 | SEQ_LT(tp->snd_fack, sblkp->end)) | |
422 | tp->snd_fack = sblkp->end; | |
423 | } | |
424 | } else if (SEQ_LT(tp->snd_fack, sblkp->end)) | |
425 | /* fack is advanced. */ | |
426 | tp->snd_fack = sblkp->end; | |
427 | /* We must have at least one SACK hole in scoreboard */ | |
428 | cur = TAILQ_LAST(&tp->snd_holes, sackhole_head); /* Last SACK hole */ | |
429 | /* | |
430 | * Since the incoming sack blocks are sorted, we can process them | |
431 | * making one sweep of the scoreboard. | |
432 | */ | |
433 | while (sblkp >= sack_blocks && cur != NULL) { | |
434 | if (SEQ_GEQ(sblkp->start, cur->end)) { | |
435 | /* | |
436 | * SACKs data beyond the current hole. | |
437 | * Go to the previous sack block. | |
438 | */ | |
439 | sblkp--; | |
440 | continue; | |
441 | } | |
442 | if (SEQ_LEQ(sblkp->end, cur->start)) { | |
443 | /* | |
444 | * SACKs data before the current hole. | |
445 | * Go to the previous hole. | |
446 | */ | |
447 | cur = TAILQ_PREV(cur, sackhole_head, scblink); | |
448 | continue; | |
449 | } | |
450 | tp->sackhint.sack_bytes_rexmit -= (cur->rxmit - cur->start); | |
451 | if (SEQ_LEQ(sblkp->start, cur->start)) { | |
452 | /* Data acks at least the beginning of hole */ | |
453 | if (SEQ_GEQ(sblkp->end, cur->end)) { | |
454 | /* Acks entire hole, so delete hole */ | |
455 | temp = cur; | |
456 | cur = TAILQ_PREV(cur, sackhole_head, scblink); | |
457 | tcp_sackhole_remove(tp, temp); | |
458 | /* | |
459 | * The sack block may ack all or part of the next | |
460 | * hole too, so continue onto the next hole. | |
461 | */ | |
462 | continue; | |
463 | } else { | |
464 | /* Move start of hole forward */ | |
465 | cur->start = sblkp->end; | |
466 | cur->rxmit = SEQ_MAX(cur->rxmit, cur->start); | |
467 | } | |
468 | } else { | |
469 | /* Data acks at least the end of hole */ | |
470 | if (SEQ_GEQ(sblkp->end, cur->end)) { | |
471 | /* Move end of hole backward */ | |
472 | cur->end = sblkp->start; | |
473 | cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); | |
474 | } else { | |
475 | /* | |
476 | * ACKs some data in middle of a hole; need to | |
477 | * split current hole | |
478 | */ | |
479 | temp = tcp_sackhole_insert(tp, sblkp->end, | |
480 | cur->end, cur); | |
481 | if (temp != NULL) { | |
482 | if (SEQ_GT(cur->rxmit, temp->rxmit)) { | |
483 | temp->rxmit = cur->rxmit; | |
484 | tp->sackhint.sack_bytes_rexmit | |
485 | += (temp->rxmit | |
486 | - temp->start); | |
487 | } | |
488 | cur->end = sblkp->start; | |
489 | cur->rxmit = SEQ_MIN(cur->rxmit, | |
490 | cur->end); | |
491 | } | |
492 | } | |
493 | } | |
494 | tp->sackhint.sack_bytes_rexmit += (cur->rxmit - cur->start); | |
495 | /* | |
496 | * Testing sblkp->start against cur->start tells us whether | |
497 | * we're done with the sack block or the sack hole. | |
498 | * Accordingly, we advance one or the other. | |
499 | */ | |
500 | if (SEQ_LEQ(sblkp->start, cur->start)) | |
501 | cur = TAILQ_PREV(cur, sackhole_head, scblink); | |
502 | else | |
503 | sblkp--; | |
504 | } | |
505 | } | |
506 | ||
507 | /* | |
508 | * Free all SACK holes to clear the scoreboard. | |
509 | */ | |
510 | void | |
511 | tcp_free_sackholes(struct tcpcb *tp) | |
512 | { | |
513 | struct sackhole *q; | |
514 | ||
515 | while ((q = TAILQ_FIRST(&tp->snd_holes)) != NULL) | |
516 | tcp_sackhole_remove(tp, q); | |
517 | tp->sackhint.sack_bytes_rexmit = 0; | |
b0d623f7 A |
518 | tp->sackhint.nexthole = NULL; |
519 | tp->sack_newdata = 0; | |
8ad349bb A |
520 | |
521 | } | |
522 | ||
523 | /* | |
524 | * Partial ack handling within a sack recovery episode. | |
525 | * Keeping this very simple for now. When a partial ack | |
526 | * is received, force snd_cwnd to a value that will allow | |
527 | * the sender to transmit no more than 2 segments. | |
528 | * If necessary, a better scheme can be adopted at a | |
529 | * later point, but for now, the goal is to prevent the | |
530 | * sender from bursting a large amount of data in the midst | |
531 | * of sack recovery. | |
532 | */ | |
533 | void | |
534 | tcp_sack_partialack(tp, th) | |
535 | struct tcpcb *tp; | |
536 | struct tcphdr *th; | |
537 | { | |
538 | int num_segs = 1; | |
539 | ||
540 | tp->t_timer[TCPT_REXMT] = 0; | |
541 | tp->t_rtttime = 0; | |
542 | /* send one or 2 segments based on how much new data was acked */ | |
543 | if (((th->th_ack - tp->snd_una) / tp->t_maxseg) > 2) | |
544 | num_segs = 2; | |
545 | tp->snd_cwnd = (tp->sackhint.sack_bytes_rexmit + | |
546 | (tp->snd_nxt - tp->sack_newdata) + | |
547 | num_segs * tp->t_maxseg); | |
548 | if (tp->snd_cwnd > tp->snd_ssthresh) | |
549 | tp->snd_cwnd = tp->snd_ssthresh; | |
550 | tp->t_flags |= TF_ACKNOW; | |
551 | (void) tcp_output(tp); | |
552 | } | |
553 | ||
554 | /* | |
555 | * Debug version of tcp_sack_output() that walks the scoreboard. Used for | |
556 | * now to sanity check the hint. | |
557 | */ | |
558 | static struct sackhole * | |
559 | tcp_sack_output_debug(struct tcpcb *tp, int *sack_bytes_rexmt) | |
560 | { | |
561 | struct sackhole *p; | |
562 | ||
563 | *sack_bytes_rexmt = 0; | |
564 | TAILQ_FOREACH(p, &tp->snd_holes, scblink) { | |
565 | if (SEQ_LT(p->rxmit, p->end)) { | |
566 | if (SEQ_LT(p->rxmit, tp->snd_una)) {/* old SACK hole */ | |
567 | continue; | |
568 | } | |
569 | *sack_bytes_rexmt += (p->rxmit - p->start); | |
570 | break; | |
571 | } | |
572 | *sack_bytes_rexmt += (p->rxmit - p->start); | |
573 | } | |
574 | return (p); | |
575 | } | |
576 | ||
577 | /* | |
578 | * Returns the next hole to retransmit and the number of retransmitted bytes | |
579 | * from the scoreboard. We store both the next hole and the number of | |
580 | * retransmitted bytes as hints (and recompute these on the fly upon SACK/ACK | |
581 | * reception). This avoids scoreboard traversals completely. | |
582 | * | |
583 | * The loop here will traverse *at most* one link. Here's the argument. | |
584 | * For the loop to traverse more than 1 link before finding the next hole to | |
585 | * retransmit, we would need to have at least 1 node following the current hint | |
586 | * with (rxmit == end). But, for all holes following the current hint, | |
587 | * (start == rxmit), since we have not yet retransmitted from them. Therefore, | |
588 | * in order to traverse more 1 link in the loop below, we need to have at least | |
589 | * one node following the current hint with (start == rxmit == end). | |
590 | * But that can't happen, (start == end) means that all the data in that hole | |
591 | * has been sacked, in which case, the hole would have been removed from the | |
592 | * scoreboard. | |
593 | */ | |
594 | struct sackhole * | |
595 | tcp_sack_output(struct tcpcb *tp, int *sack_bytes_rexmt) | |
596 | { | |
597 | struct sackhole *hole = NULL, *dbg_hole = NULL; | |
598 | int dbg_bytes_rexmt; | |
599 | ||
600 | dbg_hole = tcp_sack_output_debug(tp, &dbg_bytes_rexmt); | |
601 | *sack_bytes_rexmt = tp->sackhint.sack_bytes_rexmit; | |
602 | hole = tp->sackhint.nexthole; | |
603 | if (hole == NULL || SEQ_LT(hole->rxmit, hole->end)) | |
604 | goto out; | |
605 | while ((hole = TAILQ_NEXT(hole, scblink)) != NULL) { | |
606 | if (SEQ_LT(hole->rxmit, hole->end)) { | |
607 | tp->sackhint.nexthole = hole; | |
608 | break; | |
609 | } | |
610 | } | |
611 | out: | |
612 | if (dbg_hole != hole) { | |
613 | printf("%s: Computed sack hole not the same as cached value\n", __func__); | |
614 | hole = dbg_hole; | |
615 | } | |
616 | if (*sack_bytes_rexmt != dbg_bytes_rexmt) { | |
617 | printf("%s: Computed sack_bytes_retransmitted (%d) not " | |
618 | "the same as cached value (%d)\n", | |
619 | __func__, dbg_bytes_rexmt, *sack_bytes_rexmt); | |
620 | *sack_bytes_rexmt = dbg_bytes_rexmt; | |
621 | } | |
622 | return (hole); | |
623 | } | |
624 | ||
625 | /* | |
626 | * After a timeout, the SACK list may be rebuilt. This SACK information | |
627 | * should be used to avoid retransmitting SACKed data. This function | |
628 | * traverses the SACK list to see if snd_nxt should be moved forward. | |
629 | */ | |
630 | void | |
631 | tcp_sack_adjust(struct tcpcb *tp) | |
632 | { | |
633 | struct sackhole *p, *cur = TAILQ_FIRST(&tp->snd_holes); | |
634 | ||
635 | if (cur == NULL) | |
636 | return; /* No holes */ | |
637 | if (SEQ_GEQ(tp->snd_nxt, tp->snd_fack)) | |
638 | return; /* We're already beyond any SACKed blocks */ | |
639 | /* | |
640 | * Two cases for which we want to advance snd_nxt: | |
641 | * i) snd_nxt lies between end of one hole and beginning of another | |
642 | * ii) snd_nxt lies between end of last hole and snd_fack | |
643 | */ | |
644 | while ((p = TAILQ_NEXT(cur, scblink)) != NULL) { | |
645 | if (SEQ_LT(tp->snd_nxt, cur->end)) | |
646 | return; | |
647 | if (SEQ_GEQ(tp->snd_nxt, p->start)) | |
648 | cur = p; | |
649 | else { | |
650 | tp->snd_nxt = p->start; | |
651 | return; | |
652 | } | |
653 | } | |
654 | if (SEQ_LT(tp->snd_nxt, cur->end)) | |
655 | return; | |
656 | tp->snd_nxt = tp->snd_fack; | |
657 | return; | |
658 | } |