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_input.c 8.12 (Berkeley) 5/24/95 30 */ 31 32#include <sys/cdefs.h> 33__FBSDID("$FreeBSD: stable/11/sys/netinet/tcp_reass.c 344511 2019-02-25 10:38:37Z tuexen $"); 34 35#include "opt_inet.h" 36#include "opt_inet6.h" 37#include "opt_tcpdebug.h" 38 39/* For debugging we want counters and BB logging */ 40/* #define TCP_REASS_COUNTERS 1 */ 41/* #define TCP_REASS_LOGGING 1 */ 42 43#include <sys/param.h> 44#include <sys/kernel.h> 45#include <sys/eventhandler.h> 46#include <sys/malloc.h> 47#include <sys/mbuf.h> 48#include <sys/socket.h> 49#include <sys/socketvar.h> 50#include <sys/sysctl.h> 51#include <sys/syslog.h> 52#include <sys/systm.h> 53 54#include <vm/uma.h> 55 56#include <net/if.h> 57#include <net/if_var.h> 58#include <net/route.h> 59#include <net/vnet.h> 60 61#include <netinet/in.h> 62#include <netinet/in_pcb.h> 63#include <netinet/in_systm.h> 64#include <netinet/in_var.h> 65#include <netinet/ip.h> 66#include <netinet/ip_var.h> 67#include <netinet/ip_options.h> 68#include <netinet/ip6.h> 69#include <netinet6/in6_pcb.h> 70#include <netinet6/ip6_var.h> 71#include <netinet6/nd6.h> 72#include <netinet/tcp.h> 73#include <netinet/tcp_fsm.h> 74#include <netinet/tcp_seq.h> 75#include <netinet/tcp_timer.h> 76#include <netinet/tcp_var.h> 77#ifdef TCP_REASS_LOGGING 78#include <netinet/tcp_log_buf.h> 79#include <netinet/tcp_hpts.h> 80#endif 81#include <netinet6/tcp6_var.h> 82#include <netinet/tcpip.h> 83#ifdef TCPDEBUG 84#include <netinet/tcp_debug.h> 85#endif /* TCPDEBUG */ 86 87#define TCP_R_LOG_ADD 1 88#define TCP_R_LOG_LIMIT_REACHED 2 89#define TCP_R_LOG_APPEND 3 90#define TCP_R_LOG_PREPEND 4 91#define TCP_R_LOG_REPLACE 5 92#define TCP_R_LOG_MERGE_INTO 6 93#define TCP_R_LOG_NEW_ENTRY 7 94#define TCP_R_LOG_READ 8 95#define TCP_R_LOG_ZERO 9 96#define TCP_R_LOG_DUMP 10 97#define TCP_R_LOG_TRIM 11 98 99static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 100 "TCP Segment Reassembly Queue"); 101 102static SYSCTL_NODE(_net_inet_tcp_reass, OID_AUTO, stats, CTLFLAG_RW, 0, 103 "TCP Segment Reassembly stats"); 104 105 106static int tcp_reass_maxseg = 0; 107SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN, 108 &tcp_reass_maxseg, 0, 109 "Global maximum number of TCP Segments in Reassembly Queue"); 110 111static uma_zone_t tcp_reass_zone; 112SYSCTL_UMA_CUR(_net_inet_tcp_reass, OID_AUTO, cursegments, 0, 113 &tcp_reass_zone, 114 "Global number of TCP Segments currently in Reassembly Queue"); 115 116static u_int tcp_reass_maxqueuelen = 100; 117SYSCTL_UINT(_net_inet_tcp_reass, OID_AUTO, maxqueuelen, CTLFLAG_RWTUN, 118 &tcp_reass_maxqueuelen, 0, 119 "Maximum number of TCP Segments per Reassembly Queue"); 120 121static int tcp_new_limits = 0; 122SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, new_limit, CTLFLAG_RWTUN, 123 &tcp_new_limits, 0, 124 "Do we use the new limit method we are discussing?"); 125 126static u_int tcp_reass_queue_guard = 16; 127SYSCTL_UINT(_net_inet_tcp_reass, OID_AUTO, queueguard, CTLFLAG_RWTUN, 128 &tcp_reass_queue_guard, 16, 129 "Number of TCP Segments in Reassembly Queue where we flip over to guard mode"); 130 131#ifdef TCP_REASS_COUNTERS 132 133counter_u64_t reass_entry; 134SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, entry, CTLFLAG_RD, 135 &reass_entry, "A segment entered reassembly "); 136 137counter_u64_t reass_path1; 138SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, path1, CTLFLAG_RD, 139 &reass_path1, "Took path 1"); 140 141counter_u64_t reass_path2; 142SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, path2, CTLFLAG_RD, 143 &reass_path2, "Took path 2"); 144 145counter_u64_t reass_path3; 146SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, path3, CTLFLAG_RD, 147 &reass_path3, "Took path 3"); 148 149counter_u64_t reass_path4; 150SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, path4, CTLFLAG_RD, 151 &reass_path4, "Took path 4"); 152 153counter_u64_t reass_path5; 154SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, path5, CTLFLAG_RD, 155 &reass_path5, "Took path 5"); 156 157counter_u64_t reass_path6; 158SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, path6, CTLFLAG_RD, 159 &reass_path6, "Took path 6"); 160 161counter_u64_t reass_path7; 162SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, path7, CTLFLAG_RD, 163 &reass_path7, "Took path 7"); 164 165counter_u64_t reass_fullwalk; 166SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, fullwalk, CTLFLAG_RD, 167 &reass_fullwalk, "Took a full walk "); 168 169counter_u64_t reass_nospace; 170SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, nospace, CTLFLAG_RD, 171 &reass_nospace, "Had no mbuf capacity "); 172 173counter_u64_t merge_fwd; 174SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, merge_fwd, CTLFLAG_RD, 175 &merge_fwd, "Ran merge fwd"); 176 177counter_u64_t merge_into; 178SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, merge_into, CTLFLAG_RD, 179 &merge_into, "Ran merge into"); 180 181counter_u64_t tcp_zero_input; 182SYSCTL_COUNTER_U64(_net_inet_tcp_reass_stats, OID_AUTO, zero_input, CTLFLAG_RD, 183 &tcp_zero_input, "The reassembly buffer saw a zero len segment etc"); 184 185#endif 186 187/* Initialize TCP reassembly queue */ 188static void 189tcp_reass_zone_change(void *tag) 190{ 191 192 /* Set the zone limit and read back the effective value. */ 193 tcp_reass_maxseg = nmbclusters / 16; 194 tcp_reass_maxseg = uma_zone_set_max(tcp_reass_zone, 195 tcp_reass_maxseg); 196} 197 198#ifdef TCP_REASS_LOGGING 199 200static void 201tcp_log_reassm(struct tcpcb *tp, struct tseg_qent *q, struct tseg_qent *p, 202 tcp_seq seq, int len, uint8_t action, int instance) 203{ 204 uint32_t cts; 205 struct timeval tv; 206 207 if (tp->t_logstate != TCP_LOG_STATE_OFF) { 208 union tcp_log_stackspecific log; 209 210 memset(&log, 0, sizeof(log)); 211 cts = tcp_get_usecs(&tv); 212 log.u_bbr.flex1 = seq; 213 log.u_bbr.cur_del_rate = (uint64_t)q; 214 log.u_bbr.delRate = (uint64_t)p; 215 if (q != NULL) { 216 log.u_bbr.flex2 = q->tqe_start; 217 log.u_bbr.flex3 = q->tqe_len; 218 log.u_bbr.flex4 = q->tqe_mbuf_cnt; 219 log.u_bbr.hptsi_gain = q->tqe_flags; 220 } 221 if (p != NULL) { 222 log.u_bbr.flex5 = p->tqe_start; 223 log.u_bbr.pkts_out = p->tqe_len; 224 log.u_bbr.epoch = p->tqe_mbuf_cnt; 225 log.u_bbr.cwnd_gain = p->tqe_flags; 226 } 227 log.u_bbr.flex6 = tp->t_segqmbuflen; 228 log.u_bbr.flex7 = instance; 229 log.u_bbr.flex8 = action; 230 log.u_bbr.timeStamp = cts; 231 TCP_LOG_EVENTP(tp, NULL, 232 &tp->t_inpcb->inp_socket->so_rcv, 233 &tp->t_inpcb->inp_socket->so_snd, 234 TCP_LOG_REASS, 0, 235 len, &log, false, &tv); 236 } 237} 238 239static void 240tcp_reass_log_dump(struct tcpcb *tp) 241{ 242 struct tseg_qent *q; 243 244 if (tp->t_logstate != TCP_LOG_STATE_OFF) { 245 TAILQ_FOREACH(q, &tp->t_segq, tqe_q) { 246 tcp_log_reassm(tp, q, NULL, q->tqe_start, q->tqe_len, TCP_R_LOG_DUMP, 0); 247 } 248 }; 249} 250 251static void 252tcp_reass_log_new_in(struct tcpcb *tp, tcp_seq seq, int len, struct mbuf *m, 253 int logval, struct tseg_qent *q) 254{ 255 int cnt; 256 struct mbuf *t; 257 258 cnt = 0; 259 t = m; 260 while (t) { 261 cnt += t->m_len; 262 t = t->m_next; 263 } 264 tcp_log_reassm(tp, q, NULL, seq, len, logval, cnt); 265} 266 267#endif 268 269void 270tcp_reass_global_init(void) 271{ 272 273 tcp_reass_maxseg = nmbclusters / 16; 274 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", 275 &tcp_reass_maxseg); 276 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent), 277 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 278 /* Set the zone limit and read back the effective value. */ 279 tcp_reass_maxseg = uma_zone_set_max(tcp_reass_zone, 280 tcp_reass_maxseg); 281#ifdef TCP_REASS_COUNTERS 282 reass_path1 = counter_u64_alloc(M_WAITOK); 283 reass_path2 = counter_u64_alloc(M_WAITOK); 284 reass_path3 = counter_u64_alloc(M_WAITOK); 285 reass_path4 = counter_u64_alloc(M_WAITOK); 286 reass_path5 = counter_u64_alloc(M_WAITOK); 287 reass_path6 = counter_u64_alloc(M_WAITOK); 288 reass_path7 = counter_u64_alloc(M_WAITOK); 289 reass_fullwalk = counter_u64_alloc(M_WAITOK); 290 reass_nospace = counter_u64_alloc(M_WAITOK); 291 reass_entry = counter_u64_alloc(M_WAITOK); 292 merge_fwd = counter_u64_alloc(M_WAITOK); 293 merge_into = counter_u64_alloc(M_WAITOK); 294 tcp_zero_input = counter_u64_alloc(M_WAITOK); 295#endif 296 EVENTHANDLER_REGISTER(nmbclusters_change, 297 tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY); 298 299} 300 301void 302tcp_reass_flush(struct tcpcb *tp) 303{ 304 struct tseg_qent *qe; 305 306 INP_WLOCK_ASSERT(tp->t_inpcb); 307 308 while ((qe = TAILQ_FIRST(&tp->t_segq)) != NULL) { 309 TAILQ_REMOVE(&tp->t_segq, qe, tqe_q); 310 m_freem(qe->tqe_m); 311 uma_zfree(tcp_reass_zone, qe); 312 tp->t_segqlen--; 313 } 314 tp->t_segqmbuflen = 0; 315 KASSERT((tp->t_segqlen == 0), 316 ("TCP reass queue %p segment count is %d instead of 0 after flush.", 317 tp, tp->t_segqlen)); 318} 319 320static void 321tcp_reass_append(struct tcpcb *tp, struct tseg_qent *last, 322 struct mbuf *m, struct tcphdr *th, int tlen, 323 struct mbuf *mlast, int lenofoh) 324{ 325 326#ifdef TCP_REASS_LOGGING 327 tcp_log_reassm(tp, last, NULL, th->th_seq, tlen, TCP_R_LOG_APPEND, 0); 328#endif 329 last->tqe_len += tlen; 330 last->tqe_m->m_pkthdr.len += tlen; 331 /* Preserve the FIN bit if its there */ 332 last->tqe_flags |= (th->th_flags & TH_FIN); 333 last->tqe_last->m_next = m; 334 last->tqe_last = mlast; 335 last->tqe_mbuf_cnt += lenofoh; 336 tp->t_rcvoopack++; 337 TCPSTAT_INC(tcps_rcvoopack); 338 TCPSTAT_ADD(tcps_rcvoobyte, tlen); 339#ifdef TCP_REASS_LOGGING 340 tcp_reass_log_new_in(tp, last->tqe_start, lenofoh, last->tqe_m, 341 TCP_R_LOG_APPEND, 342 last); 343#endif 344} 345 346static void 347tcp_reass_prepend(struct tcpcb *tp, struct tseg_qent *first, struct mbuf *m, struct tcphdr *th, 348 int tlen, struct mbuf *mlast, int lenofoh) 349{ 350 int i; 351 352#ifdef TCP_REASS_LOGGING 353 tcp_log_reassm(tp, first, NULL, th->th_seq, tlen, TCP_R_LOG_PREPEND, 0); 354#endif 355 if (SEQ_GT((th->th_seq + tlen), first->tqe_start)) { 356 /* The new data overlaps into the old */ 357 i = (th->th_seq + tlen) - first->tqe_start; 358#ifdef TCP_REASS_LOGGING 359 tcp_log_reassm(tp, first, NULL, 0, i, TCP_R_LOG_TRIM, 1); 360#endif 361 m_adj(first->tqe_m, i); 362 first->tqe_len -= i; 363 first->tqe_start += i; 364 } 365 /* Ok now setup our chain to point to the old first */ 366 mlast->m_next = first->tqe_m; 367 first->tqe_m = m; 368 first->tqe_len += tlen; 369 first->tqe_start = th->th_seq; 370 first->tqe_m->m_pkthdr.len = first->tqe_len; 371 first->tqe_mbuf_cnt += lenofoh; 372 tp->t_rcvoopack++; 373 TCPSTAT_INC(tcps_rcvoopack); 374 TCPSTAT_ADD(tcps_rcvoobyte, tlen); 375#ifdef TCP_REASS_LOGGING 376 tcp_reass_log_new_in(tp, first->tqe_start, lenofoh, first->tqe_m, 377 TCP_R_LOG_PREPEND, 378 first); 379#endif 380} 381 382static void 383tcp_reass_replace(struct tcpcb *tp, struct tseg_qent *q, struct mbuf *m, 384 tcp_seq seq, int len, struct mbuf *mlast, int mbufoh, uint8_t flags) 385{ 386 /* 387 * Free the data in q, and replace 388 * it with the new segment. 389 */ 390 int len_dif; 391 392#ifdef TCP_REASS_LOGGING 393 tcp_log_reassm(tp, q, NULL, seq, len, TCP_R_LOG_REPLACE, 0); 394#endif 395 m_freem(q->tqe_m); 396 KASSERT(tp->t_segqmbuflen >= q->tqe_mbuf_cnt, 397 ("Tp:%p seg queue goes negative", tp)); 398 tp->t_segqmbuflen -= q->tqe_mbuf_cnt; 399 q->tqe_mbuf_cnt = mbufoh; 400 q->tqe_m = m; 401 q->tqe_last = mlast; 402 q->tqe_start = seq; 403 if (len > q->tqe_len) 404 len_dif = len - q->tqe_len; 405 else 406 len_dif = 0; 407 tp->t_rcvoopack++; 408 TCPSTAT_INC(tcps_rcvoopack); 409 TCPSTAT_ADD(tcps_rcvoobyte, len_dif); 410 q->tqe_len = len; 411 q->tqe_flags = (flags & TH_FIN); 412 q->tqe_m->m_pkthdr.len = q->tqe_len; 413 tp->t_segqmbuflen += mbufoh; 414 415} 416 417static void 418tcp_reass_merge_into(struct tcpcb *tp, struct tseg_qent *ent, 419 struct tseg_qent *q) 420{ 421 /* 422 * Merge q into ent and free q from the list. 423 */ 424#ifdef TCP_REASS_LOGGING 425 tcp_log_reassm(tp, q, ent, 0, 0, TCP_R_LOG_MERGE_INTO, 0); 426#endif 427#ifdef TCP_REASS_COUNTERS 428 counter_u64_add(merge_into, 1); 429#endif 430 ent->tqe_last->m_next = q->tqe_m; 431 ent->tqe_last = q->tqe_last; 432 ent->tqe_len += q->tqe_len; 433 ent->tqe_mbuf_cnt += q->tqe_mbuf_cnt; 434 ent->tqe_m->m_pkthdr.len += q->tqe_len; 435 ent->tqe_flags |= (q->tqe_flags & TH_FIN); 436 TAILQ_REMOVE(&tp->t_segq, q, tqe_q); 437 uma_zfree(tcp_reass_zone, q); 438 tp->t_segqlen--; 439 440} 441 442static void 443tcp_reass_merge_forward(struct tcpcb *tp, struct tseg_qent *ent) 444{ 445 struct tseg_qent *q, *qtmp; 446 int i; 447 tcp_seq max; 448 /* 449 * Given an entry merge forward anyplace 450 * that ent overlaps forward. 451 */ 452 453 max = ent->tqe_start + ent->tqe_len; 454 q = TAILQ_NEXT(ent, tqe_q); 455 if (q == NULL) { 456 /* Nothing left */ 457 return; 458 } 459 TAILQ_FOREACH_FROM_SAFE(q, &tp->t_segq, tqe_q, qtmp) { 460 if (SEQ_GT(q->tqe_start, max)) { 461 /* Beyond q */ 462 break; 463 } 464 /* We have some or all that are overlapping */ 465 if (SEQ_GEQ(max, (q->tqe_start + q->tqe_len))) { 466 /* It consumes it all */ 467 tp->t_segqmbuflen -= q->tqe_mbuf_cnt; 468 m_freem(q->tqe_m); 469 TAILQ_REMOVE(&tp->t_segq, q, tqe_q); 470 uma_zfree(tcp_reass_zone, q); 471 tp->t_segqlen--; 472 continue; 473 } 474 /* 475 * Trim the q entry to dovetail to this one 476 * and then merge q into ent updating max 477 * in the process. 478 */ 479 i = max - q->tqe_start; 480#ifdef TCP_REASS_LOGGING 481 tcp_log_reassm(tp, q, NULL, 0, i, TCP_R_LOG_TRIM, 2); 482#endif 483 m_adj(q->tqe_m, i); 484 q->tqe_len -= i; 485 q->tqe_start += i; 486 tcp_reass_merge_into(tp, ent, q); 487 max = ent->tqe_start + ent->tqe_len; 488 } 489#ifdef TCP_REASS_COUNTERS 490 counter_u64_add(merge_fwd, 1); 491#endif 492} 493 494static int 495tcp_reass_overhead_of_chain(struct mbuf *m, struct mbuf **mlast) 496{ 497 int len = MSIZE; 498 499 if (m->m_flags & M_EXT) 500 len += m->m_ext.ext_size; 501 while (m->m_next != NULL) { 502 m = m->m_next; 503 len += MSIZE; 504 if (m->m_flags & M_EXT) 505 len += m->m_ext.ext_size; 506 } 507 *mlast = m; 508 return (len); 509} 510 511 512/* 513 * NOTE!!! the new tcp-reassembly code *must not* use 514 * m_adj() with a negative index. That alters the chain 515 * of mbufs (by possibly chopping trailing mbufs). At 516 * the front of tcp_reass we count the mbuf overhead 517 * and setup the tail pointer. If we use m_adj(m, -5) 518 * we could corrupt the tail pointer. Currently the 519 * code only uses m_adj(m, postive-num). If this 520 * changes appropriate changes to update mlast would 521 * be needed. 522 */ 523int 524tcp_reass(struct tcpcb *tp, struct tcphdr *th, tcp_seq *seq_start, 525 int *tlenp, struct mbuf *m) 526{ 527 struct tseg_qent *q, *last, *first; 528 struct tseg_qent *p = NULL; 529 struct tseg_qent *nq = NULL; 530 struct tseg_qent *te = NULL; 531 struct tseg_qent tqs; 532 struct mbuf *mlast = NULL; 533 struct sockbuf *sb; 534 struct socket *so = tp->t_inpcb->inp_socket; 535 char *s = NULL; 536 int flags, i, lenofoh; 537 538 INP_WLOCK_ASSERT(tp->t_inpcb); 539 /* 540 * XXX: tcp_reass() is rather inefficient with its data structures 541 * and should be rewritten (see NetBSD for optimizations). 542 */ 543 544 KASSERT(th == NULL || (seq_start != NULL && tlenp != NULL), 545 ("tcp_reass called with illegal parameter combination " 546 "(tp=%p, th=%p, seq_start=%p, tlenp=%p, m=%p)", 547 tp, th, seq_start, tlenp, m)); 548 /* 549 * Call with th==NULL after become established to 550 * force pre-ESTABLISHED data up to user socket. 551 */ 552 if (th == NULL) 553 goto present; 554 KASSERT(SEQ_GEQ(th->th_seq, tp->rcv_nxt), 555 ("Attempt to add old entry to reassembly queue (th=%p, tp=%p)", 556 th, tp)); 557#ifdef TCP_REASS_LOGGING 558 tcp_reass_log_new_in(tp, th->th_seq, *tlenp, m, TCP_R_LOG_ADD, NULL); 559#endif 560#ifdef TCP_REASS_COUNTERS 561 counter_u64_add(reass_entry, 1); 562#endif 563 /* 564 * Check for zero length data. 565 */ 566 if ((*tlenp == 0) && ((th->th_flags & TH_FIN) == 0)) { 567 /* 568 * A zero length segment does no 569 * one any good. We could check 570 * the rcv_nxt <-> rcv_wnd but thats 571 * already done for us by the caller. 572 */ 573#ifdef TCP_REASS_COUNTERS 574 counter_u64_add(tcp_zero_input, 1); 575#endif 576 m_freem(m); 577#ifdef TCP_REASS_LOGGING 578 tcp_reass_log_dump(tp); 579#endif 580 return (0); 581 } 582 /* 583 * Will it fit? 584 */ 585 lenofoh = tcp_reass_overhead_of_chain(m, &mlast); 586 sb = &tp->t_inpcb->inp_socket->so_rcv; 587 if ((th->th_seq != tp->rcv_nxt || !TCPS_HAVEESTABLISHED(tp->t_state)) && 588 (sb->sb_mbcnt + tp->t_segqmbuflen + lenofoh) > sb->sb_mbmax) { 589 /* No room */ 590 TCPSTAT_INC(tcps_rcvreassfull); 591#ifdef TCP_REASS_COUNTERS 592 counter_u64_add(reass_nospace, 1); 593#endif 594#ifdef TCP_REASS_LOGGING 595 tcp_log_reassm(tp, NULL, NULL, th->th_seq, lenofoh, TCP_R_LOG_LIMIT_REACHED, 0); 596#endif 597 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) { 598 log(LOG_DEBUG, "%s; %s: mbuf count limit reached, " 599 "segment dropped\n", s, __func__); 600 free(s, M_TCPLOG); 601 } 602 m_freem(m); 603 *tlenp = 0; 604#ifdef TCP_REASS_LOGGING 605 tcp_reass_log_dump(tp); 606#endif 607 return (0); 608 } 609 /* 610 * First lets deal with two common cases, the 611 * segment appends to the back of our collected 612 * segments. Or the segment is the next in line. 613 */ 614 last = TAILQ_LAST_FAST(&tp->t_segq, tseg_qent, tqe_q); 615 if (last != NULL) { 616 if ((th->th_flags & TH_FIN) && 617 SEQ_LT((th->th_seq + *tlenp), (last->tqe_start + last->tqe_len))) { 618 /* 619 * Someone is trying to game us, dump 620 * the segment. 621 */ 622 *tlenp = 0; 623 m_freem(m); 624 return (0); 625 } 626 if ((SEQ_GEQ(th->th_seq, last->tqe_start)) && 627 (SEQ_GEQ((last->tqe_start + last->tqe_len), th->th_seq))) { 628 /* Common case, trailing segment is added */ 629 /** 630 * +--last 631 * v 632 * reassembly buffer |---| |---| |---| 633 * new segment |---| 634 */ 635#ifdef TCP_REASS_COUNTERS 636 counter_u64_add(reass_path1, 1); 637#endif 638 if (SEQ_GT((last->tqe_start + last->tqe_len), th->th_seq)) { 639 i = (last->tqe_start + last->tqe_len) - th->th_seq; 640 if (i < *tlenp) { 641#ifdef TCP_REASS_LOGGING 642 tcp_log_reassm(tp, last, NULL, 0, i, TCP_R_LOG_TRIM, 3); 643 th->th_seq += i; 644#endif 645 m_adj(m, i); 646 *tlenp -= i; 647 } else { 648 /* Complete overlap */ 649 TCPSTAT_INC(tcps_rcvduppack); 650 TCPSTAT_ADD(tcps_rcvdupbyte, *tlenp); 651 m_freem(m); 652 *tlenp = last->tqe_len; 653 *seq_start = last->tqe_start; 654 return (0); 655 } 656 } 657 if (last->tqe_flags & TH_FIN) { 658 /* 659 * We have data after the FIN on the last? 660 */ 661 *tlenp = 0; 662 m_freem(m); 663 return(0); 664 } 665 tcp_reass_append(tp, last, m, th, *tlenp, mlast, lenofoh); 666 tp->t_segqmbuflen += lenofoh; 667 *seq_start = last->tqe_start; 668 *tlenp = last->tqe_len; 669 return (0); 670 } else if (SEQ_GT(th->th_seq, (last->tqe_start + last->tqe_len))) { 671 /* 672 * Second common case, we missed 673 * another one and have something more 674 * for the end. 675 */ 676 /** 677 * +--last 678 * v 679 * reassembly buffer |---| |---| |---| 680 * new segment |---| 681 */ 682 if (last->tqe_flags & TH_FIN) { 683 /* 684 * We have data after the FIN on the last? 685 */ 686 *tlenp = 0; 687 m_freem(m); 688 return(0); 689 } 690#ifdef TCP_REASS_COUNTERS 691 counter_u64_add(reass_path2, 1); 692#endif 693 p = last; 694 goto new_entry; 695 } 696 } else { 697 /* First segment (it's NULL). */ 698 goto new_entry; 699 } 700 first = TAILQ_FIRST(&tp->t_segq); 701 if (SEQ_LT(th->th_seq, first->tqe_start) && 702 SEQ_GEQ((th->th_seq + *tlenp),first->tqe_start) && 703 SEQ_LT((th->th_seq + *tlenp), (first->tqe_start + first->tqe_len))) { 704 /* 705 * The head of the queue is prepended by this and 706 * it may be the one I want most. 707 */ 708 /** 709 * first-------+ 710 * v 711 * rea: |---| |---| |---| 712 * new: |---| 713 * Note the case we do not deal with here is: 714 * rea= |---| |---| |---| 715 * new= |----| 716 * Due to the fact that it could be 717 * new |--------------------| 718 * And we might need to merge forward. 719 */ 720#ifdef INVARIANTS 721 struct mbuf *firstmbuf; 722#endif 723 724#ifdef TCP_REASS_COUNTERS 725 counter_u64_add(reass_path3, 1); 726#endif 727 if (SEQ_LT(th->th_seq, tp->rcv_nxt)) { 728 /* 729 * The resend was even before 730 * what we have. We need to trim it. 731 * Note TSNH (it should be trimmed 732 * before the call to tcp_reass()). 733 */ 734#ifdef INVARIANTS 735 panic("th->th_seq:%u rcv_nxt:%u tp:%p not pre-trimmed", 736 th->th_seq, tp->rcv_nxt, tp); 737#else 738 i = tp->rcv_nxt - th->th_seq; 739#ifdef TCP_REASS_LOGGING 740 tcp_log_reassm(tp, first, NULL, 0, i, TCP_R_LOG_TRIM, 4); 741#endif 742 m_adj(m, i); 743 th->th_seq += i; 744 *tlenp -= i; 745#endif 746 } 747#ifdef INVARIANTS 748 firstmbuf = first->tqe_m; 749#endif 750 tcp_reass_prepend(tp, first, m, th, *tlenp, mlast, lenofoh); 751#ifdef INVARIANTS 752 if (firstmbuf == first->tqe_m) { 753 panic("First stayed same m:%p foobar:%p first->tqe_m:%p tp:%p first:%p", 754 m, firstmbuf, first->tqe_m, tp, first); 755 } else if (first->tqe_m != m) { 756 panic("First did not change to m:%p foobar:%p first->tqe_m:%p tp:%p first:%p", 757 m, firstmbuf, first->tqe_m, tp, first); 758 } 759#endif 760 tp->t_segqmbuflen += lenofoh; 761 *seq_start = first->tqe_start; 762 *tlenp = first->tqe_len; 763 goto present; 764 } else if (SEQ_LT((th->th_seq + *tlenp), first->tqe_start)) { 765 /* New segment is before our earliest segment. */ 766 /** 767 * first---->+ 768 * v 769 * rea= |---| .... 770 * new" |---| 771 * 772 */ 773 goto new_entry; 774 } 775 /* 776 * Find a segment which begins after this one does. 777 */ 778#ifdef TCP_REASS_COUNTERS 779 counter_u64_add(reass_fullwalk, 1); 780#endif 781 TAILQ_FOREACH(q, &tp->t_segq, tqe_q) { 782 if (SEQ_GT(q->tqe_start, th->th_seq)) 783 break; 784 } 785 p = TAILQ_PREV(q, tsegqe_head, tqe_q); 786 /** 787 * Now is this fit just in-between only? 788 * i.e.: 789 * p---+ +----q 790 * v v 791 * res= |--| |--| |--| 792 * nee |-| 793 */ 794 if (SEQ_LT((th->th_seq + *tlenp), q->tqe_start) && 795 ((p == NULL) || (SEQ_GT(th->th_seq, (p->tqe_start + p->tqe_len))))) { 796 /* Yep no overlap */ 797 goto new_entry; 798 } 799 /** 800 * If we reach here we have some (possibly all) overlap 801 * such as: 802 * res= |--| |--| |--| 803 * new= |----| 804 * or new= |-----------------| 805 * or new= |--------| 806 * or new= |---| 807 * or new= |-----------| 808 */ 809 if ((p != NULL) && 810 (SEQ_LEQ(th->th_seq, (p->tqe_start + p->tqe_len)))) { 811 /* conversion to int (in i) handles seq wraparound */ 812 813#ifdef TCP_REASS_COUNTERS 814 counter_u64_add(reass_path4, 1); 815#endif 816 i = p->tqe_start + p->tqe_len - th->th_seq; 817 if (i >= 0) { 818 if (i >= *tlenp) { 819 /** 820 * prev seg---->+ 821 * v 822 * reassembly buffer |---| 823 * new segment |-| 824 */ 825 TCPSTAT_INC(tcps_rcvduppack); 826 TCPSTAT_ADD(tcps_rcvdupbyte, *tlenp); 827 *tlenp = p->tqe_len; 828 *seq_start = p->tqe_start; 829 m_freem(m); 830 /* 831 * Try to present any queued data 832 * at the left window edge to the user. 833 * This is needed after the 3-WHS 834 * completes. Note this probably 835 * will not work and we will return. 836 */ 837 return (0); 838 } 839 if (i > 0) { 840 /** 841 * prev seg---->+ 842 * v 843 * reassembly buffer |---| 844 * new segment |-----| 845 */ 846#ifdef TCP_REASS_COUNTERS 847 counter_u64_add(reass_path5, 1); 848#endif 849#ifdef TCP_REASS_LOGGING 850 tcp_log_reassm(tp, p, NULL, 0, i, TCP_R_LOG_TRIM, 5); 851#endif 852 m_adj(m, i); 853 *tlenp -= i; 854 th->th_seq += i; 855 } 856 } 857 if (th->th_seq == (p->tqe_start + p->tqe_len)) { 858 /* 859 * If dovetails in with this one 860 * append it. 861 */ 862 /** 863 * prev seg---->+ 864 * v 865 * reassembly buffer |--| |---| 866 * new segment |--| 867 * (note: it was trimmed above if it overlapped) 868 */ 869 tcp_reass_append(tp, p, m, th, *tlenp, mlast, lenofoh); 870 tp->t_segqmbuflen += lenofoh; 871 } else { 872#ifdef INVARIANTS 873 panic("Impossible cut th_seq:%u p->seq:%u(%d) p:%p tp:%p", 874 th->th_seq, p->tqe_start, p->tqe_len, 875 p, tp); 876#endif 877 *tlenp = 0; 878 m_freem(m); 879 return (0); 880 } 881 q = p; 882 } else { 883 /* 884 * The new data runs over the 885 * top of previously sack'd data (in q). 886 * It may be partially overlapping, or 887 * it may overlap the entire segment. 888 */ 889#ifdef TCP_REASS_COUNTERS 890 counter_u64_add(reass_path6, 1); 891#endif 892 if (SEQ_GEQ((th->th_seq + *tlenp), (q->tqe_start + q->tqe_len))) { 893 /* It consumes it all */ 894 /** 895 * next seg---->+ 896 * v 897 * reassembly buffer |--| |---| 898 * new segment |----------| 899 */ 900#ifdef TCP_REASS_COUNTERS 901 counter_u64_add(reass_path7, 1); 902#endif 903 tcp_reass_replace(tp, q, m, th->th_seq, *tlenp, mlast, lenofoh, th->th_flags); 904 } else { 905 /* 906 * We just need to prepend the data 907 * to this. It does not overrun 908 * the end. 909 */ 910 /** 911 * next seg---->+ 912 * v 913 * reassembly buffer |--| |---| 914 * new segment |----------| 915 */ 916 tcp_reass_prepend(tp, q, m, th, *tlenp, mlast, lenofoh); 917 tp->t_segqmbuflen += lenofoh; 918 } 919 } 920 /* Now does it go further than that? */ 921 tcp_reass_merge_forward(tp, q); 922 *seq_start = q->tqe_start; 923 *tlenp = q->tqe_len; 924 goto present; 925 926 /* 927 * When we reach here we can't combine it 928 * with any existing segment. 929 * 930 * Limit the number of segments that can be queued to reduce the 931 * potential for mbuf exhaustion. For best performance, we want to be 932 * able to queue a full window's worth of segments. The size of the 933 * socket receive buffer determines our advertised window and grows 934 * automatically when socket buffer autotuning is enabled. Use it as the 935 * basis for our queue limit. 936 * 937 * However, allow the user to specify a ceiling for the number of 938 * segments in each queue. 939 * 940 * Always let the missing segment through which caused this queue. 941 * NB: Access to the socket buffer is left intentionally unlocked as we 942 * can tolerate stale information here. 943 * 944 * XXXLAS: Using sbspace(so->so_rcv) instead of so->so_rcv.sb_hiwat 945 * should work but causes packets to be dropped when they shouldn't. 946 * Investigate why and re-evaluate the below limit after the behaviour 947 * is understood. 948 */ 949new_entry: 950 if (th->th_seq == tp->rcv_nxt && TCPS_HAVEESTABLISHED(tp->t_state)) { 951 tp->rcv_nxt += *tlenp; 952 flags = th->th_flags & TH_FIN; 953 TCPSTAT_INC(tcps_rcvoopack); 954 TCPSTAT_ADD(tcps_rcvoobyte, *tlenp); 955 SOCKBUF_LOCK(&so->so_rcv); 956 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 957 m_freem(m); 958 } else { 959 sbappendstream_locked(&so->so_rcv, m, 0); 960 } 961 sorwakeup_locked(so); 962 return (flags); 963 } 964 if (tcp_new_limits) { 965 if ((tp->t_segqlen > tcp_reass_queue_guard) && 966 (*tlenp < MSIZE)) { 967 /* 968 * This is really a lie, we are not full but 969 * are getting a segment that is above 970 * guard threshold. If it is and its below 971 * a mbuf size (256) we drop it if it 972 * can't fill in some place. 973 */ 974 TCPSTAT_INC(tcps_rcvreassfull); 975 *tlenp = 0; 976 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) { 977 log(LOG_DEBUG, "%s; %s: queue limit reached, " 978 "segment dropped\n", s, __func__); 979 free(s, M_TCPLOG); 980 } 981 m_freem(m); 982#ifdef TCP_REASS_LOGGING 983 tcp_reass_log_dump(tp); 984#endif 985 return (0); 986 } 987 } else { 988 if (tp->t_segqlen >= min((so->so_rcv.sb_hiwat / tp->t_maxseg) + 1, 989 tcp_reass_maxqueuelen)) { 990 TCPSTAT_INC(tcps_rcvreassfull); 991 *tlenp = 0; 992 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) { 993 log(LOG_DEBUG, "%s; %s: queue limit reached, " 994 "segment dropped\n", s, __func__); 995 free(s, M_TCPLOG); 996 } 997 m_freem(m); 998#ifdef TCP_REASS_LOGGING 999 tcp_reass_log_dump(tp); 1000#endif 1001 return (0); 1002 } 1003 } 1004 /* 1005 * Allocate a new queue entry. If we can't, or hit the zone limit 1006 * just drop the pkt. 1007 */ 1008 te = uma_zalloc(tcp_reass_zone, M_NOWAIT); 1009 if (te == NULL) { 1010 TCPSTAT_INC(tcps_rcvmemdrop); 1011 m_freem(m); 1012 *tlenp = 0; 1013 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, 1014 NULL))) { 1015 log(LOG_DEBUG, "%s; %s: global zone limit " 1016 "reached, segment dropped\n", s, __func__); 1017 free(s, M_TCPLOG); 1018 } 1019 return (0); 1020 } 1021 tp->t_segqlen++; 1022 tp->t_rcvoopack++; 1023 TCPSTAT_INC(tcps_rcvoopack); 1024 TCPSTAT_ADD(tcps_rcvoobyte, *tlenp); 1025 /* Insert the new segment queue entry into place. */ 1026 te->tqe_m = m; 1027 te->tqe_flags = th->th_flags; 1028 te->tqe_len = *tlenp; 1029 te->tqe_start = th->th_seq; 1030 te->tqe_last = mlast; 1031 te->tqe_mbuf_cnt = lenofoh; 1032 tp->t_segqmbuflen += te->tqe_mbuf_cnt; 1033 if (p == NULL) { 1034 TAILQ_INSERT_HEAD(&tp->t_segq, te, tqe_q); 1035 } else { 1036 TAILQ_INSERT_AFTER(&tp->t_segq, p, te, tqe_q); 1037 } 1038#ifdef TCP_REASS_LOGGING 1039 tcp_reass_log_new_in(tp, th->th_seq, *tlenp, m, TCP_R_LOG_NEW_ENTRY, te); 1040#endif 1041present: 1042 /* 1043 * Present data to user, advancing rcv_nxt through 1044 * completed sequence space. 1045 */ 1046 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 1047 return (0); 1048 q = TAILQ_FIRST(&tp->t_segq); 1049 KASSERT(q == NULL || SEQ_GEQ(q->tqe_start, tp->rcv_nxt), 1050 ("Reassembly queue for %p has stale entry at head", tp)); 1051 if (!q || q->tqe_start != tp->rcv_nxt) { 1052#ifdef TCP_REASS_LOGGING 1053 tcp_reass_log_dump(tp); 1054#endif 1055 return (0); 1056 } 1057 SOCKBUF_LOCK(&so->so_rcv); 1058 do { 1059 tp->rcv_nxt += q->tqe_len; 1060 flags = q->tqe_flags & TH_FIN; 1061 nq = TAILQ_NEXT(q, tqe_q); 1062 TAILQ_REMOVE(&tp->t_segq, q, tqe_q); 1063 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1064 m_freem(q->tqe_m); 1065 } else { 1066#ifdef TCP_REASS_LOGGING 1067 tcp_reass_log_new_in(tp, q->tqe_start, q->tqe_len, q->tqe_m, TCP_R_LOG_READ, q); 1068 if (th != NULL) { 1069 tcp_log_reassm(tp, q, NULL, th->th_seq, *tlenp, TCP_R_LOG_READ, 1); 1070 } else { 1071 tcp_log_reassm(tp, q, NULL, 0, 0, TCP_R_LOG_READ, 1); 1072 } 1073#endif 1074 sbappendstream_locked(&so->so_rcv, q->tqe_m, 0); 1075 } 1076#ifdef TCP_REASS_LOGGING 1077 if (th != NULL) { 1078 tcp_log_reassm(tp, q, NULL, th->th_seq, *tlenp, TCP_R_LOG_READ, 2); 1079 } else { 1080 tcp_log_reassm(tp, q, NULL, 0, 0, TCP_R_LOG_READ, 2); 1081 } 1082#endif 1083 KASSERT(tp->t_segqmbuflen >= q->tqe_mbuf_cnt, 1084 ("tp:%p seg queue goes negative", tp)); 1085 tp->t_segqmbuflen -= q->tqe_mbuf_cnt; 1086 if (q != &tqs) 1087 uma_zfree(tcp_reass_zone, q); 1088 tp->t_segqlen--; 1089 q = nq; 1090 } while (q && q->tqe_start == tp->rcv_nxt); 1091 if (TAILQ_EMPTY(&tp->t_segq) && 1092 (tp->t_segqmbuflen != 0)) { 1093#ifdef INVARIANTS 1094 panic("tp:%p segq:%p len:%d queue empty", 1095 tp, &tp->t_segq, tp->t_segqmbuflen); 1096#else 1097#ifdef TCP_REASS_LOGGING 1098 if (th != NULL) { 1099 tcp_log_reassm(tp, NULL, NULL, th->th_seq, *tlenp, TCP_R_LOG_ZERO, 0); 1100 } else { 1101 tcp_log_reassm(tp, NULL, NULL, 0, 0, TCP_R_LOG_ZERO, 0); 1102 } 1103#endif 1104 tp->t_segqmbuflen = 0; 1105#endif 1106 } 1107#ifdef TCP_REASS_LOGGING 1108 tcp_reass_log_dump(tp); 1109#endif 1110 sorwakeup_locked(so); 1111 return (flags); 1112} 1113