1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
5 * The Regents of the University of California. All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
32 */
33
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 #include "opt_tcpdebug.h"
40
41 /* For debugging we want counters and BB logging */
42 /* #define TCP_REASS_COUNTERS 1 */
43 /* #define TCP_REASS_LOGGING 1 */
44
45 #include <sys/param.h>
46 #include <sys/kernel.h>
47 #include <sys/eventhandler.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/sysctl.h>
53 #include <sys/syslog.h>
54 #include <sys/systm.h>
55
56 #include <vm/uma.h>
57
58 #include <net/if.h>
59 #include <net/if_var.h>
60 #include <net/route.h>
61 #include <net/vnet.h>
62
63 #include <netinet/in.h>
64 #include <netinet/in_pcb.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/in_var.h>
67 #include <netinet/ip.h>
68 #include <netinet/ip_var.h>
69 #include <netinet/ip_options.h>
70 #include <netinet/ip6.h>
71 #include <netinet6/in6_pcb.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet6/nd6.h>
74 #include <netinet/tcp.h>
75 #include <netinet/tcp_fsm.h>
76 #include <netinet/tcp_seq.h>
77 #include <netinet/tcp_timer.h>
78 #include <netinet/tcp_var.h>
79 #ifdef TCP_REASS_LOGGING
80 #include <netinet/tcp_log_buf.h>
81 #include <netinet/tcp_hpts.h>
82 #endif
83 #include <netinet6/tcp6_var.h>
84 #include <netinet/tcpip.h>
85 #ifdef TCPDEBUG
86 #include <netinet/tcp_debug.h>
87 #endif /* TCPDEBUG */
88
89 #define TCP_R_LOG_ADD 1
90 #define TCP_R_LOG_LIMIT_REACHED 2
91 #define TCP_R_LOG_APPEND 3
92 #define TCP_R_LOG_PREPEND 4
93 #define TCP_R_LOG_REPLACE 5
94 #define TCP_R_LOG_MERGE_INTO 6
95 #define TCP_R_LOG_NEW_ENTRY 7
96 #define TCP_R_LOG_READ 8
97 #define TCP_R_LOG_ZERO 9
98 #define TCP_R_LOG_DUMP 10
99 #define TCP_R_LOG_TRIM 11
100
101 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass,
102 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
103 "TCP Segment Reassembly Queue");
104
105 static SYSCTL_NODE(_net_inet_tcp_reass, OID_AUTO, stats,
106 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
107 "TCP Segment Reassembly stats");
108
109 static int tcp_reass_maxseg = 0;
110 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN,
111 &tcp_reass_maxseg, 0,
112 "Global maximum number of TCP Segments in Reassembly Queue");
113
114 static uma_zone_t tcp_reass_zone;
115 SYSCTL_UMA_CUR(_net_inet_tcp_reass, OID_AUTO, cursegments, 0,
116 &tcp_reass_zone,
117 "Global number of TCP Segments currently in Reassembly Queue");
118
119 static u_int tcp_reass_maxqueuelen = 100;
120 SYSCTL_UINT(_net_inet_tcp_reass, OID_AUTO, maxqueuelen, CTLFLAG_RWTUN,
121 &tcp_reass_maxqueuelen, 0,
122 "Maximum number of TCP Segments per Reassembly Queue");
123
124 static int tcp_new_limits = 0;
125 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, new_limit, CTLFLAG_RWTUN,
126 &tcp_new_limits, 0,
127 "Do we use the new limit method we are discussing?");
128
129 static u_int tcp_reass_queue_guard = 16;
130 SYSCTL_UINT(_net_inet_tcp_reass, OID_AUTO, queueguard, CTLFLAG_RWTUN,
131 &tcp_reass_queue_guard, 16,
132 "Number of TCP Segments in Reassembly Queue where we flip over to guard mode");
133
134 #ifdef TCP_REASS_COUNTERS
135
136 counter_u64_t reass_entry;
137 SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, entry, CTLFLAG_RD,
138 &reass_entry, "A segment entered reassembly ");
139
140 counter_u64_t reass_path1;
141 SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, path1, CTLFLAG_RD,
142 &reass_path1, "Took path 1");
143
144 counter_u64_t reass_path2;
145 SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, path2, CTLFLAG_RD,
146 &reass_path2, "Took path 2");
147
148 counter_u64_t reass_path3;
149 SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, path3, CTLFLAG_RD,
150 &reass_path3, "Took path 3");
151
152 counter_u64_t reass_path4;
153 SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, path4, CTLFLAG_RD,
154 &reass_path4, "Took path 4");
155
156 counter_u64_t reass_path5;
157 SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, path5, CTLFLAG_RD,
158 &reass_path5, "Took path 5");
159
160 counter_u64_t reass_path6;
161 SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, path6, CTLFLAG_RD,
162 &reass_path6, "Took path 6");
163
164 counter_u64_t reass_path7;
165 SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, path7, CTLFLAG_RD,
166 &reass_path7, "Took path 7");
167
168 counter_u64_t reass_fullwalk;
169 SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, fullwalk, CTLFLAG_RD,
170 &reass_fullwalk, "Took a full walk ");
171
172 counter_u64_t reass_nospace;
173 SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, nospace, CTLFLAG_RD,
174 &reass_nospace, "Had no mbuf capacity ");
175
176 counter_u64_t merge_fwd;
177 SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, merge_fwd, CTLFLAG_RD,
178 &merge_fwd, "Ran merge fwd");
179
180 counter_u64_t merge_into;
181 SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, merge_into, CTLFLAG_RD,
182 &merge_into, "Ran merge into");
183
184 counter_u64_t tcp_zero_input;
185 SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, zero_input, CTLFLAG_RD,
186 &tcp_zero_input, "The reassembly buffer saw a zero len segment etc");
187
188 #endif
189
190 /* Initialize TCP reassembly queue */
191 static void
tcp_reass_zone_change(void * tag)192 tcp_reass_zone_change(void *tag)
193 {
194
195 /* Set the zone limit and read back the effective value. */
196 tcp_reass_maxseg = nmbclusters / 16;
197 tcp_reass_maxseg = uma_zone_set_max(tcp_reass_zone,
198 tcp_reass_maxseg);
199 }
200
201 #ifdef TCP_REASS_LOGGING
202
203 static void
tcp_log_reassm(struct tcpcb * tp,struct tseg_qent * q,struct tseg_qent * p,tcp_seq seq,int len,uint8_t action,int instance)204 tcp_log_reassm(struct tcpcb *tp, struct tseg_qent *q, struct tseg_qent *p,
205 tcp_seq seq, int len, uint8_t action, int instance)
206 {
207 uint32_t cts;
208 struct timeval tv;
209
210 if (tp->t_logstate != TCP_LOG_STATE_OFF) {
211 union tcp_log_stackspecific log;
212
213 memset(&log, 0, sizeof(log));
214 cts = tcp_get_usecs(&tv);
215 log.u_bbr.flex1 = seq;
216 log.u_bbr.cur_del_rate = (uint64_t)q;
217 log.u_bbr.delRate = (uint64_t)p;
218 if (q != NULL) {
219 log.u_bbr.flex2 = q->tqe_start;
220 log.u_bbr.flex3 = q->tqe_len;
221 log.u_bbr.flex4 = q->tqe_mbuf_cnt;
222 log.u_bbr.hptsi_gain = q->tqe_flags;
223 }
224 if (p != NULL) {
225 log.u_bbr.flex5 = p->tqe_start;
226 log.u_bbr.pkts_out = p->tqe_len;
227 log.u_bbr.epoch = p->tqe_mbuf_cnt;
228 log.u_bbr.cwnd_gain = p->tqe_flags;
229 }
230 log.u_bbr.flex6 = tp->t_segqmbuflen;
231 log.u_bbr.flex7 = instance;
232 log.u_bbr.flex8 = action;
233 log.u_bbr.timeStamp = cts;
234 TCP_LOG_EVENTP(tp, NULL,
235 &tp->t_inpcb->inp_socket->so_rcv,
236 &tp->t_inpcb->inp_socket->so_snd,
237 TCP_LOG_REASS, 0,
238 len, &log, false, &tv);
239 }
240 }
241
242 static void
tcp_reass_log_dump(struct tcpcb * tp)243 tcp_reass_log_dump(struct tcpcb *tp)
244 {
245 struct tseg_qent *q;
246
247 if (tp->t_logstate != TCP_LOG_STATE_OFF) {
248 TAILQ_FOREACH(q, &tp->t_segq, tqe_q) {
249 tcp_log_reassm(tp, q, NULL, q->tqe_start, q->tqe_len, TCP_R_LOG_DUMP, 0);
250 }
251 };
252 }
253
254 static void
tcp_reass_log_new_in(struct tcpcb * tp,tcp_seq seq,int len,struct mbuf * m,int logval,struct tseg_qent * q)255 tcp_reass_log_new_in(struct tcpcb *tp, tcp_seq seq, int len, struct mbuf *m,
256 int logval, struct tseg_qent *q)
257 {
258 int cnt;
259 struct mbuf *t;
260
261 cnt = 0;
262 t = m;
263 while (t) {
264 cnt += t->m_len;
265 t = t->m_next;
266 }
267 tcp_log_reassm(tp, q, NULL, seq, len, logval, cnt);
268 }
269
270 #endif
271
272 void
tcp_reass_global_init(void)273 tcp_reass_global_init(void)
274 {
275
276 tcp_reass_maxseg = nmbclusters / 16;
277 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments",
278 &tcp_reass_maxseg);
279 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent),
280 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
281 /* Set the zone limit and read back the effective value. */
282 tcp_reass_maxseg = uma_zone_set_max(tcp_reass_zone,
283 tcp_reass_maxseg);
284 #ifdef TCP_REASS_COUNTERS
285 reass_path1 = counter_u64_alloc(M_WAITOK);
286 reass_path2 = counter_u64_alloc(M_WAITOK);
287 reass_path3 = counter_u64_alloc(M_WAITOK);
288 reass_path4 = counter_u64_alloc(M_WAITOK);
289 reass_path5 = counter_u64_alloc(M_WAITOK);
290 reass_path6 = counter_u64_alloc(M_WAITOK);
291 reass_path7 = counter_u64_alloc(M_WAITOK);
292 reass_fullwalk = counter_u64_alloc(M_WAITOK);
293 reass_nospace = counter_u64_alloc(M_WAITOK);
294 reass_entry = counter_u64_alloc(M_WAITOK);
295 merge_fwd = counter_u64_alloc(M_WAITOK);
296 merge_into = counter_u64_alloc(M_WAITOK);
297 tcp_zero_input = counter_u64_alloc(M_WAITOK);
298 #endif
299 EVENTHANDLER_REGISTER(nmbclusters_change,
300 tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY);
301
302 }
303
304 void
tcp_reass_flush(struct tcpcb * tp)305 tcp_reass_flush(struct tcpcb *tp)
306 {
307 struct tseg_qent *qe;
308
309 INP_WLOCK_ASSERT(tp->t_inpcb);
310
311 while ((qe = TAILQ_FIRST(&tp->t_segq)) != NULL) {
312 TAILQ_REMOVE(&tp->t_segq, qe, tqe_q);
313 m_freem(qe->tqe_m);
314 uma_zfree(tcp_reass_zone, qe);
315 tp->t_segqlen--;
316 }
317 tp->t_segqmbuflen = 0;
318 KASSERT((tp->t_segqlen == 0),
319 ("TCP reass queue %p segment count is %d instead of 0 after flush.",
320 tp, tp->t_segqlen));
321 }
322
323 static void
tcp_reass_append(struct tcpcb * tp,struct tseg_qent * last,struct mbuf * m,struct tcphdr * th,int tlen,struct mbuf * mlast,int lenofoh)324 tcp_reass_append(struct tcpcb *tp, struct tseg_qent *last,
325 struct mbuf *m, struct tcphdr *th, int tlen,
326 struct mbuf *mlast, int lenofoh)
327 {
328
329 #ifdef TCP_REASS_LOGGING
330 tcp_log_reassm(tp, last, NULL, th->th_seq, tlen, TCP_R_LOG_APPEND, 0);
331 #endif
332 last->tqe_len += tlen;
333 last->tqe_m->m_pkthdr.len += tlen;
334 /* Preserve the FIN bit if its there */
335 last->tqe_flags |= (th->th_flags & TH_FIN);
336 last->tqe_last->m_next = m;
337 last->tqe_last = mlast;
338 last->tqe_mbuf_cnt += lenofoh;
339 tp->t_rcvoopack++;
340 TCPSTAT_INC(tcps_rcvoopack);
341 TCPSTAT_ADD(tcps_rcvoobyte, tlen);
342 #ifdef TCP_REASS_LOGGING
343 tcp_reass_log_new_in(tp, last->tqe_start, lenofoh, last->tqe_m,
344 TCP_R_LOG_APPEND,
345 last);
346 #endif
347 }
348
349 static void
tcp_reass_prepend(struct tcpcb * tp,struct tseg_qent * first,struct mbuf * m,struct tcphdr * th,int tlen,struct mbuf * mlast,int lenofoh)350 tcp_reass_prepend(struct tcpcb *tp, struct tseg_qent *first, struct mbuf *m, struct tcphdr *th,
351 int tlen, struct mbuf *mlast, int lenofoh)
352 {
353 int i;
354
355 #ifdef TCP_REASS_LOGGING
356 tcp_log_reassm(tp, first, NULL, th->th_seq, tlen, TCP_R_LOG_PREPEND, 0);
357 #endif
358 if (SEQ_GT((th->th_seq + tlen), first->tqe_start)) {
359 /* The new data overlaps into the old */
360 i = (th->th_seq + tlen) - first->tqe_start;
361 #ifdef TCP_REASS_LOGGING
362 tcp_log_reassm(tp, first, NULL, 0, i, TCP_R_LOG_TRIM, 1);
363 #endif
364 m_adj(first->tqe_m, i);
365 first->tqe_len -= i;
366 first->tqe_start += i;
367 }
368 /* Ok now setup our chain to point to the old first */
369 mlast->m_next = first->tqe_m;
370 first->tqe_m = m;
371 first->tqe_len += tlen;
372 first->tqe_start = th->th_seq;
373 first->tqe_m->m_pkthdr.len = first->tqe_len;
374 first->tqe_mbuf_cnt += lenofoh;
375 tp->t_rcvoopack++;
376 TCPSTAT_INC(tcps_rcvoopack);
377 TCPSTAT_ADD(tcps_rcvoobyte, tlen);
378 #ifdef TCP_REASS_LOGGING
379 tcp_reass_log_new_in(tp, first->tqe_start, lenofoh, first->tqe_m,
380 TCP_R_LOG_PREPEND,
381 first);
382 #endif
383 }
384
385 static void
tcp_reass_replace(struct tcpcb * tp,struct tseg_qent * q,struct mbuf * m,tcp_seq seq,int len,struct mbuf * mlast,int mbufoh,uint8_t flags)386 tcp_reass_replace(struct tcpcb *tp, struct tseg_qent *q, struct mbuf *m,
387 tcp_seq seq, int len, struct mbuf *mlast, int mbufoh, uint8_t flags)
388 {
389 /*
390 * Free the data in q, and replace
391 * it with the new segment.
392 */
393 int len_dif;
394
395 #ifdef TCP_REASS_LOGGING
396 tcp_log_reassm(tp, q, NULL, seq, len, TCP_R_LOG_REPLACE, 0);
397 #endif
398 m_freem(q->tqe_m);
399 KASSERT(tp->t_segqmbuflen >= q->tqe_mbuf_cnt,
400 ("Tp:%p seg queue goes negative", tp));
401 tp->t_segqmbuflen -= q->tqe_mbuf_cnt;
402 q->tqe_mbuf_cnt = mbufoh;
403 q->tqe_m = m;
404 q->tqe_last = mlast;
405 q->tqe_start = seq;
406 if (len > q->tqe_len)
407 len_dif = len - q->tqe_len;
408 else
409 len_dif = 0;
410 tp->t_rcvoopack++;
411 TCPSTAT_INC(tcps_rcvoopack);
412 TCPSTAT_ADD(tcps_rcvoobyte, len_dif);
413 q->tqe_len = len;
414 q->tqe_flags = (flags & TH_FIN);
415 q->tqe_m->m_pkthdr.len = q->tqe_len;
416 tp->t_segqmbuflen += mbufoh;
417
418 }
419
420 static void
tcp_reass_merge_into(struct tcpcb * tp,struct tseg_qent * ent,struct tseg_qent * q)421 tcp_reass_merge_into(struct tcpcb *tp, struct tseg_qent *ent,
422 struct tseg_qent *q)
423 {
424 /*
425 * Merge q into ent and free q from the list.
426 */
427 #ifdef TCP_REASS_LOGGING
428 tcp_log_reassm(tp, q, ent, 0, 0, TCP_R_LOG_MERGE_INTO, 0);
429 #endif
430 #ifdef TCP_REASS_COUNTERS
431 counter_u64_add(merge_into, 1);
432 #endif
433 ent->tqe_last->m_next = q->tqe_m;
434 ent->tqe_last = q->tqe_last;
435 ent->tqe_len += q->tqe_len;
436 ent->tqe_mbuf_cnt += q->tqe_mbuf_cnt;
437 ent->tqe_m->m_pkthdr.len += q->tqe_len;
438 ent->tqe_flags |= (q->tqe_flags & TH_FIN);
439 TAILQ_REMOVE(&tp->t_segq, q, tqe_q);
440 uma_zfree(tcp_reass_zone, q);
441 tp->t_segqlen--;
442
443 }
444
445 static void
tcp_reass_merge_forward(struct tcpcb * tp,struct tseg_qent * ent)446 tcp_reass_merge_forward(struct tcpcb *tp, struct tseg_qent *ent)
447 {
448 struct tseg_qent *q, *qtmp;
449 int i;
450 tcp_seq max;
451 /*
452 * Given an entry merge forward anyplace
453 * that ent overlaps forward.
454 */
455
456 max = ent->tqe_start + ent->tqe_len;
457 q = TAILQ_NEXT(ent, tqe_q);
458 if (q == NULL) {
459 /* Nothing left */
460 return;
461 }
462 TAILQ_FOREACH_FROM_SAFE(q, &tp->t_segq, tqe_q, qtmp) {
463 if (SEQ_GT(q->tqe_start, max)) {
464 /* Beyond q */
465 break;
466 }
467 /* We have some or all that are overlapping */
468 if (SEQ_GEQ(max, (q->tqe_start + q->tqe_len))) {
469 /* It consumes it all */
470 tp->t_segqmbuflen -= q->tqe_mbuf_cnt;
471 m_freem(q->tqe_m);
472 TAILQ_REMOVE(&tp->t_segq, q, tqe_q);
473 uma_zfree(tcp_reass_zone, q);
474 tp->t_segqlen--;
475 continue;
476 }
477 /*
478 * Trim the q entry to dovetail to this one
479 * and then merge q into ent updating max
480 * in the process.
481 */
482 i = max - q->tqe_start;
483 #ifdef TCP_REASS_LOGGING
484 tcp_log_reassm(tp, q, NULL, 0, i, TCP_R_LOG_TRIM, 2);
485 #endif
486 m_adj(q->tqe_m, i);
487 q->tqe_len -= i;
488 q->tqe_start += i;
489 tcp_reass_merge_into(tp, ent, q);
490 max = ent->tqe_start + ent->tqe_len;
491 }
492 #ifdef TCP_REASS_COUNTERS
493 counter_u64_add(merge_fwd, 1);
494 #endif
495 }
496
497 static int
tcp_reass_overhead_of_chain(struct mbuf * m,struct mbuf ** mlast)498 tcp_reass_overhead_of_chain(struct mbuf *m, struct mbuf **mlast)
499 {
500 int len = MSIZE;
501
502 if (m->m_flags & M_EXT)
503 len += m->m_ext.ext_size;
504 while (m->m_next != NULL) {
505 m = m->m_next;
506 len += MSIZE;
507 if (m->m_flags & M_EXT)
508 len += m->m_ext.ext_size;
509 }
510 *mlast = m;
511 return (len);
512 }
513
514 /*
515 * NOTE!!! the new tcp-reassembly code *must not* use
516 * m_adj() with a negative index. That alters the chain
517 * of mbufs (by possibly chopping trailing mbufs). At
518 * the front of tcp_reass we count the mbuf overhead
519 * and setup the tail pointer. If we use m_adj(m, -5)
520 * we could corrupt the tail pointer. Currently the
521 * code only uses m_adj(m, postive-num). If this
522 * changes appropriate changes to update mlast would
523 * be needed.
524 */
525 int
tcp_reass(struct tcpcb * tp,struct tcphdr * th,tcp_seq * seq_start,int * tlenp,struct mbuf * m)526 tcp_reass(struct tcpcb *tp, struct tcphdr *th, tcp_seq *seq_start,
527 int *tlenp, struct mbuf *m)
528 {
529 struct tseg_qent *q, *last, *first;
530 struct tseg_qent *p = NULL;
531 struct tseg_qent *nq = NULL;
532 struct tseg_qent *te = NULL;
533 struct mbuf *mlast = NULL;
534 struct sockbuf *sb;
535 struct socket *so = tp->t_inpcb->inp_socket;
536 char *s = NULL;
537 int flags, i, lenofoh;
538
539 INP_WLOCK_ASSERT(tp->t_inpcb);
540 /*
541 * XXX: tcp_reass() is rather inefficient with its data structures
542 * and should be rewritten (see NetBSD for optimizations).
543 */
544
545 KASSERT(th == NULL || (seq_start != NULL && tlenp != NULL),
546 ("tcp_reass called with illegal parameter combination "
547 "(tp=%p, th=%p, seq_start=%p, tlenp=%p, m=%p)",
548 tp, th, seq_start, tlenp, m));
549 /*
550 * Call with th==NULL after become established to
551 * force pre-ESTABLISHED data up to user socket.
552 */
553 if (th == NULL)
554 goto present;
555 KASSERT(SEQ_GEQ(th->th_seq, tp->rcv_nxt),
556 ("Attempt to add old entry to reassembly queue (th=%p, tp=%p)",
557 th, tp));
558 #ifdef TCP_REASS_LOGGING
559 tcp_reass_log_new_in(tp, th->th_seq, *tlenp, m, TCP_R_LOG_ADD, NULL);
560 #endif
561 #ifdef TCP_REASS_COUNTERS
562 counter_u64_add(reass_entry, 1);
563 #endif
564 /*
565 * Check for zero length data.
566 */
567 if ((*tlenp == 0) && ((th->th_flags & TH_FIN) == 0)) {
568 /*
569 * A zero length segment does no
570 * one any good. We could check
571 * the rcv_nxt <-> rcv_wnd but thats
572 * already done for us by the caller.
573 */
574 strip_fin:
575 #ifdef TCP_REASS_COUNTERS
576 counter_u64_add(tcp_zero_input, 1);
577 #endif
578 m_freem(m);
579 #ifdef TCP_REASS_LOGGING
580 tcp_reass_log_dump(tp);
581 #endif
582 return (0);
583 } else if ((*tlenp == 0) &&
584 (th->th_flags & TH_FIN) &&
585 !TCPS_HAVEESTABLISHED(tp->t_state)) {
586 /*
587 * We have not established, and we
588 * have a FIN and no data. Lets treat
589 * this as the same as if the FIN were
590 * not present. We don't want to save
591 * the FIN bit in a reassembly buffer
592 * we want to get established first before
593 * we do that (the peer will retransmit).
594 */
595 goto strip_fin;
596 }
597 /*
598 * Will it fit?
599 */
600 lenofoh = tcp_reass_overhead_of_chain(m, &mlast);
601 sb = &tp->t_inpcb->inp_socket->so_rcv;
602 if ((th->th_seq != tp->rcv_nxt || !TCPS_HAVEESTABLISHED(tp->t_state)) &&
603 (sb->sb_mbcnt + tp->t_segqmbuflen + lenofoh) > sb->sb_mbmax) {
604 /* No room */
605 TCPSTAT_INC(tcps_rcvreassfull);
606 #ifdef TCP_REASS_COUNTERS
607 counter_u64_add(reass_nospace, 1);
608 #endif
609 #ifdef TCP_REASS_LOGGING
610 tcp_log_reassm(tp, NULL, NULL, th->th_seq, lenofoh, TCP_R_LOG_LIMIT_REACHED, 0);
611 #endif
612 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) {
613 log(LOG_DEBUG, "%s; %s: mbuf count limit reached, "
614 "segment dropped\n", s, __func__);
615 free(s, M_TCPLOG);
616 }
617 m_freem(m);
618 *tlenp = 0;
619 #ifdef TCP_REASS_LOGGING
620 tcp_reass_log_dump(tp);
621 #endif
622 return (0);
623 }
624 /*
625 * First lets deal with two common cases, the
626 * segment appends to the back of our collected
627 * segments. Or the segment is the next in line.
628 */
629 last = TAILQ_LAST_FAST(&tp->t_segq, tseg_qent, tqe_q);
630 if (last != NULL) {
631 if ((th->th_flags & TH_FIN) &&
632 SEQ_LT((th->th_seq + *tlenp), (last->tqe_start + last->tqe_len))) {
633 /*
634 * Someone is trying to game us, dump
635 * the segment.
636 */
637 *tlenp = 0;
638 m_freem(m);
639 return (0);
640 }
641 if ((SEQ_GEQ(th->th_seq, last->tqe_start)) &&
642 (SEQ_GEQ((last->tqe_start + last->tqe_len), th->th_seq))) {
643 /* Common case, trailing segment is added */
644 /**
645 * +--last
646 * v
647 * reassembly buffer |---| |---| |---|
648 * new segment |---|
649 */
650 #ifdef TCP_REASS_COUNTERS
651 counter_u64_add(reass_path1, 1);
652 #endif
653 if (SEQ_GT((last->tqe_start + last->tqe_len), th->th_seq)) {
654 i = (last->tqe_start + last->tqe_len) - th->th_seq;
655 if (i < *tlenp) {
656 #ifdef TCP_REASS_LOGGING
657 tcp_log_reassm(tp, last, NULL, 0, i, TCP_R_LOG_TRIM, 3);
658 th->th_seq += i;
659 #endif
660 m_adj(m, i);
661 *tlenp -= i;
662 } else {
663 /* Complete overlap */
664 TCPSTAT_INC(tcps_rcvduppack);
665 TCPSTAT_ADD(tcps_rcvdupbyte, *tlenp);
666 m_freem(m);
667 *tlenp = last->tqe_len;
668 *seq_start = last->tqe_start;
669 return (0);
670 }
671 }
672 if (last->tqe_flags & TH_FIN) {
673 /*
674 * We have data after the FIN on the last?
675 */
676 *tlenp = 0;
677 m_freem(m);
678 return(0);
679 }
680 tcp_reass_append(tp, last, m, th, *tlenp, mlast, lenofoh);
681 tp->t_segqmbuflen += lenofoh;
682 *seq_start = last->tqe_start;
683 *tlenp = last->tqe_len;
684 return (0);
685 } else if (SEQ_GT(th->th_seq, (last->tqe_start + last->tqe_len))) {
686 /*
687 * Second common case, we missed
688 * another one and have something more
689 * for the end.
690 */
691 /**
692 * +--last
693 * v
694 * reassembly buffer |---| |---| |---|
695 * new segment |---|
696 */
697 if (last->tqe_flags & TH_FIN) {
698 /*
699 * We have data after the FIN on the last?
700 */
701 *tlenp = 0;
702 m_freem(m);
703 return(0);
704 }
705 #ifdef TCP_REASS_COUNTERS
706 counter_u64_add(reass_path2, 1);
707 #endif
708 p = last;
709 goto new_entry;
710 }
711 } else {
712 /* First segment (it's NULL). */
713 goto new_entry;
714 }
715 first = TAILQ_FIRST(&tp->t_segq);
716 if (SEQ_LT(th->th_seq, first->tqe_start) &&
717 SEQ_GEQ((th->th_seq + *tlenp),first->tqe_start) &&
718 SEQ_LT((th->th_seq + *tlenp), (first->tqe_start + first->tqe_len))) {
719 /*
720 * The head of the queue is prepended by this and
721 * it may be the one I want most.
722 */
723 /**
724 * first-------+
725 * v
726 * rea: |---| |---| |---|
727 * new: |---|
728 * Note the case we do not deal with here is:
729 * rea= |---| |---| |---|
730 * new= |----|
731 * Due to the fact that it could be
732 * new |--------------------|
733 * And we might need to merge forward.
734 */
735 #ifdef INVARIANTS
736 struct mbuf *firstmbuf;
737 #endif
738
739 #ifdef TCP_REASS_COUNTERS
740 counter_u64_add(reass_path3, 1);
741 #endif
742 if (SEQ_LT(th->th_seq, tp->rcv_nxt)) {
743 /*
744 * The resend was even before
745 * what we have. We need to trim it.
746 * Note TSNH (it should be trimmed
747 * before the call to tcp_reass()).
748 */
749 #ifdef INVARIANTS
750 panic("th->th_seq:%u rcv_nxt:%u tp:%p not pre-trimmed",
751 th->th_seq, tp->rcv_nxt, tp);
752 #else
753 i = tp->rcv_nxt - th->th_seq;
754 #ifdef TCP_REASS_LOGGING
755 tcp_log_reassm(tp, first, NULL, 0, i, TCP_R_LOG_TRIM, 4);
756 #endif
757 m_adj(m, i);
758 th->th_seq += i;
759 *tlenp -= i;
760 #endif
761 }
762 #ifdef INVARIANTS
763 firstmbuf = first->tqe_m;
764 #endif
765 tcp_reass_prepend(tp, first, m, th, *tlenp, mlast, lenofoh);
766 #ifdef INVARIANTS
767 if (firstmbuf == first->tqe_m) {
768 panic("First stayed same m:%p foobar:%p first->tqe_m:%p tp:%p first:%p",
769 m, firstmbuf, first->tqe_m, tp, first);
770 } else if (first->tqe_m != m) {
771 panic("First did not change to m:%p foobar:%p first->tqe_m:%p tp:%p first:%p",
772 m, firstmbuf, first->tqe_m, tp, first);
773 }
774 #endif
775 tp->t_segqmbuflen += lenofoh;
776 *seq_start = first->tqe_start;
777 *tlenp = first->tqe_len;
778 goto present;
779 } else if (SEQ_LT((th->th_seq + *tlenp), first->tqe_start)) {
780 /* New segment is before our earliest segment. */
781 /**
782 * first---->+
783 * v
784 * rea= |---| ....
785 * new" |---|
786 *
787 */
788 goto new_entry;
789 }
790 /*
791 * Find a segment which begins after this one does.
792 */
793 #ifdef TCP_REASS_COUNTERS
794 counter_u64_add(reass_fullwalk, 1);
795 #endif
796 TAILQ_FOREACH(q, &tp->t_segq, tqe_q) {
797 if (SEQ_GT(q->tqe_start, th->th_seq))
798 break;
799 }
800 p = TAILQ_PREV(q, tsegqe_head, tqe_q);
801 /**
802 * Now is this fit just in-between only?
803 * i.e.:
804 * p---+ +----q
805 * v v
806 * res= |--| |--| |--|
807 * nee |-|
808 */
809 if (SEQ_LT((th->th_seq + *tlenp), q->tqe_start) &&
810 ((p == NULL) || (SEQ_GT(th->th_seq, (p->tqe_start + p->tqe_len))))) {
811 /* Yep no overlap */
812 goto new_entry;
813 }
814 /**
815 * If we reach here we have some (possibly all) overlap
816 * such as:
817 * res= |--| |--| |--|
818 * new= |----|
819 * or new= |-----------------|
820 * or new= |--------|
821 * or new= |---|
822 * or new= |-----------|
823 */
824 if ((p != NULL) &&
825 (SEQ_LEQ(th->th_seq, (p->tqe_start + p->tqe_len)))) {
826 /* conversion to int (in i) handles seq wraparound */
827
828 #ifdef TCP_REASS_COUNTERS
829 counter_u64_add(reass_path4, 1);
830 #endif
831 i = p->tqe_start + p->tqe_len - th->th_seq;
832 if (i >= 0) {
833 if (i >= *tlenp) {
834 /**
835 * prev seg---->+
836 * v
837 * reassembly buffer |---|
838 * new segment |-|
839 */
840 TCPSTAT_INC(tcps_rcvduppack);
841 TCPSTAT_ADD(tcps_rcvdupbyte, *tlenp);
842 *tlenp = p->tqe_len;
843 *seq_start = p->tqe_start;
844 m_freem(m);
845 /*
846 * Try to present any queued data
847 * at the left window edge to the user.
848 * This is needed after the 3-WHS
849 * completes. Note this probably
850 * will not work and we will return.
851 */
852 return (0);
853 }
854 if (i > 0) {
855 /**
856 * prev seg---->+
857 * v
858 * reassembly buffer |---|
859 * new segment |-----|
860 */
861 #ifdef TCP_REASS_COUNTERS
862 counter_u64_add(reass_path5, 1);
863 #endif
864 #ifdef TCP_REASS_LOGGING
865 tcp_log_reassm(tp, p, NULL, 0, i, TCP_R_LOG_TRIM, 5);
866 #endif
867 m_adj(m, i);
868 *tlenp -= i;
869 th->th_seq += i;
870 }
871 }
872 if (th->th_seq == (p->tqe_start + p->tqe_len)) {
873 /*
874 * If dovetails in with this one
875 * append it.
876 */
877 /**
878 * prev seg---->+
879 * v
880 * reassembly buffer |--| |---|
881 * new segment |--|
882 * (note: it was trimmed above if it overlapped)
883 */
884 tcp_reass_append(tp, p, m, th, *tlenp, mlast, lenofoh);
885 tp->t_segqmbuflen += lenofoh;
886 } else {
887 #ifdef INVARIANTS
888 panic("Impossible cut th_seq:%u p->seq:%u(%d) p:%p tp:%p",
889 th->th_seq, p->tqe_start, p->tqe_len,
890 p, tp);
891 #endif
892 *tlenp = 0;
893 m_freem(m);
894 return (0);
895 }
896 q = p;
897 } else {
898 /*
899 * The new data runs over the
900 * top of previously sack'd data (in q).
901 * It may be partially overlapping, or
902 * it may overlap the entire segment.
903 */
904 #ifdef TCP_REASS_COUNTERS
905 counter_u64_add(reass_path6, 1);
906 #endif
907 if (SEQ_GEQ((th->th_seq + *tlenp), (q->tqe_start + q->tqe_len))) {
908 /* It consumes it all */
909 /**
910 * next seg---->+
911 * v
912 * reassembly buffer |--| |---|
913 * new segment |----------|
914 */
915 #ifdef TCP_REASS_COUNTERS
916 counter_u64_add(reass_path7, 1);
917 #endif
918 tcp_reass_replace(tp, q, m, th->th_seq, *tlenp, mlast, lenofoh, th->th_flags);
919 } else {
920 /*
921 * We just need to prepend the data
922 * to this. It does not overrun
923 * the end.
924 */
925 /**
926 * next seg---->+
927 * v
928 * reassembly buffer |--| |---|
929 * new segment |----------|
930 */
931 tcp_reass_prepend(tp, q, m, th, *tlenp, mlast, lenofoh);
932 tp->t_segqmbuflen += lenofoh;
933 }
934 }
935 /* Now does it go further than that? */
936 tcp_reass_merge_forward(tp, q);
937 *seq_start = q->tqe_start;
938 *tlenp = q->tqe_len;
939 goto present;
940
941 /*
942 * When we reach here we can't combine it
943 * with any existing segment.
944 *
945 * Limit the number of segments that can be queued to reduce the
946 * potential for mbuf exhaustion. For best performance, we want to be
947 * able to queue a full window's worth of segments. The size of the
948 * socket receive buffer determines our advertised window and grows
949 * automatically when socket buffer autotuning is enabled. Use it as the
950 * basis for our queue limit.
951 *
952 * However, allow the user to specify a ceiling for the number of
953 * segments in each queue.
954 *
955 * Always let the missing segment through which caused this queue.
956 * NB: Access to the socket buffer is left intentionally unlocked as we
957 * can tolerate stale information here.
958 *
959 * XXXLAS: Using sbspace(so->so_rcv) instead of so->so_rcv.sb_hiwat
960 * should work but causes packets to be dropped when they shouldn't.
961 * Investigate why and re-evaluate the below limit after the behaviour
962 * is understood.
963 */
964 new_entry:
965 if (th->th_seq == tp->rcv_nxt && TCPS_HAVEESTABLISHED(tp->t_state)) {
966 tp->rcv_nxt += *tlenp;
967 flags = th->th_flags & TH_FIN;
968 TCPSTAT_INC(tcps_rcvoopack);
969 TCPSTAT_ADD(tcps_rcvoobyte, *tlenp);
970 SOCKBUF_LOCK(&so->so_rcv);
971 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
972 m_freem(m);
973 } else {
974 sbappendstream_locked(&so->so_rcv, m, 0);
975 }
976 tp->t_flags |= TF_WAKESOR;
977 return (flags);
978 }
979 if (tcp_new_limits) {
980 if ((tp->t_segqlen > tcp_reass_queue_guard) &&
981 (*tlenp < MSIZE)) {
982 /*
983 * This is really a lie, we are not full but
984 * are getting a segment that is above
985 * guard threshold. If it is and its below
986 * a mbuf size (256) we drop it if it
987 * can't fill in some place.
988 */
989 TCPSTAT_INC(tcps_rcvreassfull);
990 *tlenp = 0;
991 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) {
992 log(LOG_DEBUG, "%s; %s: queue limit reached, "
993 "segment dropped\n", s, __func__);
994 free(s, M_TCPLOG);
995 }
996 m_freem(m);
997 #ifdef TCP_REASS_LOGGING
998 tcp_reass_log_dump(tp);
999 #endif
1000 return (0);
1001 }
1002 } else {
1003 if (tp->t_segqlen >= min((so->so_rcv.sb_hiwat / tp->t_maxseg) + 1,
1004 tcp_reass_maxqueuelen)) {
1005 TCPSTAT_INC(tcps_rcvreassfull);
1006 *tlenp = 0;
1007 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) {
1008 log(LOG_DEBUG, "%s; %s: queue limit reached, "
1009 "segment dropped\n", s, __func__);
1010 free(s, M_TCPLOG);
1011 }
1012 m_freem(m);
1013 #ifdef TCP_REASS_LOGGING
1014 tcp_reass_log_dump(tp);
1015 #endif
1016 return (0);
1017 }
1018 }
1019 /*
1020 * Allocate a new queue entry. If we can't, or hit the zone limit
1021 * just drop the pkt.
1022 */
1023 te = uma_zalloc(tcp_reass_zone, M_NOWAIT);
1024 if (te == NULL) {
1025 TCPSTAT_INC(tcps_rcvmemdrop);
1026 m_freem(m);
1027 *tlenp = 0;
1028 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL,
1029 NULL))) {
1030 log(LOG_DEBUG, "%s; %s: global zone limit "
1031 "reached, segment dropped\n", s, __func__);
1032 free(s, M_TCPLOG);
1033 }
1034 return (0);
1035 }
1036 tp->t_segqlen++;
1037 tp->t_rcvoopack++;
1038 TCPSTAT_INC(tcps_rcvoopack);
1039 TCPSTAT_ADD(tcps_rcvoobyte, *tlenp);
1040 /* Insert the new segment queue entry into place. */
1041 te->tqe_m = m;
1042 te->tqe_flags = th->th_flags;
1043 te->tqe_len = *tlenp;
1044 te->tqe_start = th->th_seq;
1045 te->tqe_last = mlast;
1046 te->tqe_mbuf_cnt = lenofoh;
1047 tp->t_segqmbuflen += te->tqe_mbuf_cnt;
1048 if (p == NULL) {
1049 TAILQ_INSERT_HEAD(&tp->t_segq, te, tqe_q);
1050 } else {
1051 TAILQ_INSERT_AFTER(&tp->t_segq, p, te, tqe_q);
1052 }
1053 #ifdef TCP_REASS_LOGGING
1054 tcp_reass_log_new_in(tp, th->th_seq, *tlenp, m, TCP_R_LOG_NEW_ENTRY, te);
1055 #endif
1056 present:
1057 /*
1058 * Present data to user, advancing rcv_nxt through
1059 * completed sequence space.
1060 */
1061 if (!TCPS_HAVEESTABLISHED(tp->t_state))
1062 return (0);
1063 q = TAILQ_FIRST(&tp->t_segq);
1064 KASSERT(q == NULL || SEQ_GEQ(q->tqe_start, tp->rcv_nxt),
1065 ("Reassembly queue for %p has stale entry at head", tp));
1066 if (!q || q->tqe_start != tp->rcv_nxt) {
1067 #ifdef TCP_REASS_LOGGING
1068 tcp_reass_log_dump(tp);
1069 #endif
1070 return (0);
1071 }
1072 SOCKBUF_LOCK(&so->so_rcv);
1073 do {
1074 tp->rcv_nxt += q->tqe_len;
1075 flags = q->tqe_flags & TH_FIN;
1076 nq = TAILQ_NEXT(q, tqe_q);
1077 TAILQ_REMOVE(&tp->t_segq, q, tqe_q);
1078 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1079 m_freem(q->tqe_m);
1080 } else {
1081 #ifdef TCP_REASS_LOGGING
1082 tcp_reass_log_new_in(tp, q->tqe_start, q->tqe_len, q->tqe_m, TCP_R_LOG_READ, q);
1083 if (th != NULL) {
1084 tcp_log_reassm(tp, q, NULL, th->th_seq, *tlenp, TCP_R_LOG_READ, 1);
1085 } else {
1086 tcp_log_reassm(tp, q, NULL, 0, 0, TCP_R_LOG_READ, 1);
1087 }
1088 #endif
1089 sbappendstream_locked(&so->so_rcv, q->tqe_m, 0);
1090 }
1091 #ifdef TCP_REASS_LOGGING
1092 if (th != NULL) {
1093 tcp_log_reassm(tp, q, NULL, th->th_seq, *tlenp, TCP_R_LOG_READ, 2);
1094 } else {
1095 tcp_log_reassm(tp, q, NULL, 0, 0, TCP_R_LOG_READ, 2);
1096 }
1097 #endif
1098 KASSERT(tp->t_segqmbuflen >= q->tqe_mbuf_cnt,
1099 ("tp:%p seg queue goes negative", tp));
1100 tp->t_segqmbuflen -= q->tqe_mbuf_cnt;
1101 uma_zfree(tcp_reass_zone, q);
1102 tp->t_segqlen--;
1103 q = nq;
1104 } while (q && q->tqe_start == tp->rcv_nxt);
1105 if (TAILQ_EMPTY(&tp->t_segq) &&
1106 (tp->t_segqmbuflen != 0)) {
1107 #ifdef INVARIANTS
1108 panic("tp:%p segq:%p len:%d queue empty",
1109 tp, &tp->t_segq, tp->t_segqmbuflen);
1110 #else
1111 #ifdef TCP_REASS_LOGGING
1112 if (th != NULL) {
1113 tcp_log_reassm(tp, NULL, NULL, th->th_seq, *tlenp, TCP_R_LOG_ZERO, 0);
1114 } else {
1115 tcp_log_reassm(tp, NULL, NULL, 0, 0, TCP_R_LOG_ZERO, 0);
1116 }
1117 #endif
1118 tp->t_segqmbuflen = 0;
1119 #endif
1120 }
1121 #ifdef TCP_REASS_LOGGING
1122 tcp_reass_log_dump(tp);
1123 #endif
1124 tp->t_flags |= TF_WAKESOR;
1125 return (flags);
1126 }
1127