1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright 2007 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27 #pragma ident "%Z%%M% %I% %E% SMI" 28 29 #include <sys/types.h> 30 #include <sys/systm.h> 31 #include <sys/stream.h> 32 #include <sys/cmn_err.h> 33 #include <sys/kmem.h> 34 #define _SUN_TPI_VERSION 2 35 #include <sys/tihdr.h> 36 #include <sys/socket.h> 37 #include <sys/strsun.h> 38 #include <sys/strsubr.h> 39 40 #include <netinet/in.h> 41 #include <netinet/ip6.h> 42 #include <netinet/tcp_seq.h> 43 #include <netinet/sctp.h> 44 45 #include <inet/common.h> 46 #include <inet/ip.h> 47 #include <inet/ip6.h> 48 #include <inet/mib2.h> 49 #include <inet/ipclassifier.h> 50 #include <inet/ipp_common.h> 51 #include <inet/ipsec_impl.h> 52 #include <inet/sctp_ip.h> 53 54 #include "sctp_impl.h" 55 #include "sctp_asconf.h" 56 #include "sctp_addr.h" 57 58 static struct kmem_cache *sctp_kmem_set_cache; 59 60 /* 61 * PR-SCTP comments. 62 * 63 * When we get a valid Forward TSN chunk, we check the fragment list for this 64 * SSN and preceeding SSNs free all them. Further, if this Forward TSN causes 65 * the next expected SSN to be present in the stream queue, we deliver any 66 * such stranded messages upstream. We also update the SACK info. appropriately. 67 * When checking for advancing the cumulative ack (in sctp_cumack()) we must 68 * check for abandoned chunks and messages. While traversing the tramsmit 69 * list if we come across an abandoned chunk, we can skip the message (i.e. 70 * take it out of the (re)transmit list) since this message, and hence this 71 * chunk, has been marked abandoned by sctp_rexmit(). If we come across an 72 * unsent chunk for a message this now abandoned we need to check if a 73 * Forward TSN needs to be sent, this could be a case where we deferred sending 74 * a Forward TSN in sctp_get_msg_to_send(). Further, after processing a 75 * SACK we check if the Advanced peer ack point can be moved ahead, i.e. 76 * if we can send a Forward TSN via sctp_check_abandoned_data(). 77 */ 78 void 79 sctp_free_set(sctp_set_t *s) 80 { 81 sctp_set_t *p; 82 83 while (s) { 84 p = s->next; 85 kmem_cache_free(sctp_kmem_set_cache, s); 86 s = p; 87 } 88 } 89 90 static void 91 sctp_ack_add(sctp_set_t **head, uint32_t tsn, int *num) 92 { 93 sctp_set_t *p, *t; 94 95 if (head == NULL || num == NULL) 96 return; 97 98 ASSERT(*num >= 0); 99 ASSERT((*num == 0 && *head == NULL) || (*num > 0 && *head != NULL)); 100 101 if (*head == NULL) { 102 *head = kmem_cache_alloc(sctp_kmem_set_cache, KM_NOSLEEP); 103 if (*head == NULL) 104 return; 105 (*head)->prev = (*head)->next = NULL; 106 (*head)->begin = tsn; 107 (*head)->end = tsn; 108 *num = 1; 109 return; 110 } 111 112 ASSERT((*head)->prev == NULL); 113 114 /* 115 * Handle this special case here so we don't have to check 116 * for it each time in the loop. 117 */ 118 if (SEQ_LT(tsn + 1, (*head)->begin)) { 119 /* add a new set, and move the head pointer */ 120 t = kmem_cache_alloc(sctp_kmem_set_cache, KM_NOSLEEP); 121 if (t == NULL) 122 return; 123 t->next = *head; 124 t->prev = NULL; 125 (*head)->prev = t; 126 t->begin = tsn; 127 t->end = tsn; 128 (*num)++; 129 *head = t; 130 return; 131 } 132 133 /* 134 * We need to handle the following cases, where p points to 135 * the current set (as we walk through the loop): 136 * 137 * 1. tsn is entirely less than p; create a new set before p. 138 * 2. tsn borders p from less; coalesce p with tsn. 139 * 3. tsn is withing p; do nothing. 140 * 4. tsn borders p from greater; coalesce p with tsn. 141 * 4a. p may now border p->next from less; if so, coalesce those 142 * two sets. 143 * 5. tsn is entirely greater then all sets; add a new set at 144 * the end. 145 */ 146 for (p = *head; ; p = p->next) { 147 if (SEQ_LT(tsn + 1, p->begin)) { 148 /* 1: add a new set before p. */ 149 t = kmem_cache_alloc(sctp_kmem_set_cache, KM_NOSLEEP); 150 if (t == NULL) 151 return; 152 t->next = p; 153 t->prev = NULL; 154 t->begin = tsn; 155 t->end = tsn; 156 if (p->prev) { 157 t->prev = p->prev; 158 p->prev->next = t; 159 } 160 p->prev = t; 161 (*num)++; 162 return; 163 } 164 165 if ((tsn + 1) == p->begin) { 166 /* 2: adjust p->begin */ 167 p->begin = tsn; 168 return; 169 } 170 171 if (SEQ_GEQ(tsn, p->begin) && SEQ_LEQ(tsn, p->end)) { 172 /* 3; do nothing */ 173 return; 174 } 175 176 if ((p->end + 1) == tsn) { 177 /* 4; adjust p->end */ 178 p->end = tsn; 179 180 if (p->next != NULL && (tsn + 1) == p->next->begin) { 181 /* 4a: coalesce p and p->next */ 182 t = p->next; 183 p->end = t->end; 184 p->next = t->next; 185 if (t->next != NULL) 186 t->next->prev = p; 187 kmem_cache_free(sctp_kmem_set_cache, t); 188 (*num)--; 189 } 190 return; 191 } 192 193 if (p->next == NULL) { 194 /* 5: add new set at the end */ 195 t = kmem_cache_alloc(sctp_kmem_set_cache, KM_NOSLEEP); 196 if (t == NULL) 197 return; 198 t->next = NULL; 199 t->prev = p; 200 t->begin = tsn; 201 t->end = tsn; 202 p->next = t; 203 (*num)++; 204 return; 205 } 206 207 if (SEQ_GT(tsn, p->end + 1)) 208 continue; 209 } 210 } 211 212 static void 213 sctp_ack_rem(sctp_set_t **head, uint32_t end, int *num) 214 { 215 sctp_set_t *p, *t; 216 217 if (head == NULL || *head == NULL || num == NULL) 218 return; 219 220 /* Nothing to remove */ 221 if (SEQ_LT(end, (*head)->begin)) 222 return; 223 224 /* Find out where to start removing sets */ 225 for (p = *head; p->next; p = p->next) { 226 if (SEQ_LEQ(end, p->end)) 227 break; 228 } 229 230 if (SEQ_LT(end, p->end) && SEQ_GEQ(end, p->begin)) { 231 /* adjust p */ 232 p->begin = end + 1; 233 /* all done */ 234 if (p == *head) 235 return; 236 } else if (SEQ_GEQ(end, p->end)) { 237 /* remove this set too */ 238 p = p->next; 239 } 240 241 /* unlink everything before this set */ 242 t = *head; 243 *head = p; 244 if (p != NULL && p->prev != NULL) { 245 p->prev->next = NULL; 246 p->prev = NULL; 247 } 248 249 sctp_free_set(t); 250 251 /* recount the number of sets */ 252 *num = 0; 253 254 for (p = *head; p != NULL; p = p->next) 255 (*num)++; 256 } 257 258 void 259 sctp_sets_init() 260 { 261 sctp_kmem_set_cache = kmem_cache_create("sctp_set_cache", 262 sizeof (sctp_set_t), 0, NULL, NULL, NULL, NULL, 263 NULL, 0); 264 } 265 266 void 267 sctp_sets_fini() 268 { 269 kmem_cache_destroy(sctp_kmem_set_cache); 270 } 271 272 sctp_chunk_hdr_t * 273 sctp_first_chunk(uchar_t *rptr, ssize_t remaining) 274 { 275 sctp_chunk_hdr_t *ch; 276 uint16_t ch_len; 277 278 if (remaining < sizeof (*ch)) { 279 return (NULL); 280 } 281 282 ch = (sctp_chunk_hdr_t *)rptr; 283 ch_len = ntohs(ch->sch_len); 284 285 if (ch_len < sizeof (*ch) || remaining < ch_len) { 286 return (NULL); 287 } 288 289 return (ch); 290 } 291 292 sctp_chunk_hdr_t * 293 sctp_next_chunk(sctp_chunk_hdr_t *ch, ssize_t *remaining) 294 { 295 int pad; 296 uint16_t ch_len; 297 298 if (!ch) { 299 return (NULL); 300 } 301 302 ch_len = ntohs(ch->sch_len); 303 304 if ((pad = ch_len & (SCTP_ALIGN - 1)) != 0) { 305 pad = SCTP_ALIGN - pad; 306 } 307 308 *remaining -= (ch_len + pad); 309 ch = (sctp_chunk_hdr_t *)((char *)ch + ch_len + pad); 310 311 return (sctp_first_chunk((uchar_t *)ch, *remaining)); 312 } 313 314 /* 315 * Attach ancillary data to a received SCTP segments. 316 * If the source address (fp) is not the primary, send up a 317 * unitdata_ind so recvfrom() can populate the msg_name field. 318 * If ancillary data is also requested, we append it to the 319 * unitdata_req. Otherwise, we just send up an optdata_ind. 320 */ 321 static int 322 sctp_input_add_ancillary(sctp_t *sctp, mblk_t **mp, sctp_data_hdr_t *dcp, 323 sctp_faddr_t *fp, ip6_pkt_t *ipp) 324 { 325 struct T_unitdata_ind *tudi; 326 int optlen; 327 int hdrlen; 328 uchar_t *optptr; 329 struct cmsghdr *cmsg; 330 mblk_t *mp1; 331 struct sockaddr_in6 sin_buf[1]; 332 struct sockaddr_in6 *sin6; 333 struct sockaddr_in *sin4; 334 uint_t addflag = 0; 335 336 sin4 = NULL; 337 sin6 = NULL; 338 339 optlen = hdrlen = 0; 340 341 /* Figure out address size */ 342 if (sctp->sctp_ipversion == IPV4_VERSION) { 343 sin4 = (struct sockaddr_in *)sin_buf; 344 sin4->sin_family = AF_INET; 345 sin4->sin_port = sctp->sctp_fport; 346 IN6_V4MAPPED_TO_IPADDR(&fp->faddr, sin4->sin_addr.s_addr); 347 hdrlen = sizeof (*tudi) + sizeof (*sin4); 348 } else { 349 sin6 = sin_buf; 350 sin6->sin6_family = AF_INET6; 351 sin6->sin6_port = sctp->sctp_fport; 352 sin6->sin6_addr = fp->faddr; 353 hdrlen = sizeof (*tudi) + sizeof (*sin6); 354 } 355 356 /* If app asked to receive send / recv info */ 357 if (sctp->sctp_recvsndrcvinfo) { 358 optlen += sizeof (*cmsg) + sizeof (struct sctp_sndrcvinfo); 359 if (hdrlen == 0) 360 hdrlen = sizeof (struct T_optdata_ind); 361 } 362 363 if (sctp->sctp_ipv6_recvancillary == 0) 364 goto noancillary; 365 366 if ((ipp->ipp_fields & IPPF_IFINDEX) && 367 ipp->ipp_ifindex != sctp->sctp_recvifindex && 368 (sctp->sctp_ipv6_recvancillary & SCTP_IPV6_RECVPKTINFO)) { 369 optlen += sizeof (*cmsg) + sizeof (struct in6_pktinfo); 370 if (hdrlen == 0) 371 hdrlen = sizeof (struct T_unitdata_ind); 372 addflag |= SCTP_IPV6_RECVPKTINFO; 373 } 374 /* If app asked for hoplimit and it has changed ... */ 375 if ((ipp->ipp_fields & IPPF_HOPLIMIT) && 376 ipp->ipp_hoplimit != sctp->sctp_recvhops && 377 (sctp->sctp_ipv6_recvancillary & SCTP_IPV6_RECVHOPLIMIT)) { 378 optlen += sizeof (*cmsg) + sizeof (uint_t); 379 if (hdrlen == 0) 380 hdrlen = sizeof (struct T_unitdata_ind); 381 addflag |= SCTP_IPV6_RECVHOPLIMIT; 382 } 383 /* If app asked for hopbyhop headers and it has changed ... */ 384 if ((sctp->sctp_ipv6_recvancillary & SCTP_IPV6_RECVHOPOPTS) && 385 ip_cmpbuf(sctp->sctp_hopopts, sctp->sctp_hopoptslen, 386 (ipp->ipp_fields & IPPF_HOPOPTS), 387 ipp->ipp_hopopts, ipp->ipp_hopoptslen)) { 388 optlen += sizeof (*cmsg) + ipp->ipp_hopoptslen - 389 sctp->sctp_v6label_len; 390 if (hdrlen == 0) 391 hdrlen = sizeof (struct T_unitdata_ind); 392 addflag |= SCTP_IPV6_RECVHOPOPTS; 393 if (!ip_allocbuf((void **)&sctp->sctp_hopopts, 394 &sctp->sctp_hopoptslen, 395 (ipp->ipp_fields & IPPF_HOPOPTS), 396 ipp->ipp_hopopts, ipp->ipp_hopoptslen)) 397 return (-1); 398 } 399 /* If app asked for dst headers before routing headers ... */ 400 if ((sctp->sctp_ipv6_recvancillary & SCTP_IPV6_RECVRTDSTOPTS) && 401 ip_cmpbuf(sctp->sctp_rtdstopts, sctp->sctp_rtdstoptslen, 402 (ipp->ipp_fields & IPPF_RTDSTOPTS), 403 ipp->ipp_rtdstopts, ipp->ipp_rtdstoptslen)) { 404 optlen += sizeof (*cmsg) + ipp->ipp_rtdstoptslen; 405 if (hdrlen == 0) 406 hdrlen = sizeof (struct T_unitdata_ind); 407 addflag |= SCTP_IPV6_RECVRTDSTOPTS; 408 if (!ip_allocbuf((void **)&sctp->sctp_rtdstopts, 409 &sctp->sctp_rtdstoptslen, 410 (ipp->ipp_fields & IPPF_RTDSTOPTS), 411 ipp->ipp_rtdstopts, ipp->ipp_rtdstoptslen)) 412 return (-1); 413 } 414 /* If app asked for routing headers and it has changed ... */ 415 if (sctp->sctp_ipv6_recvancillary & SCTP_IPV6_RECVRTHDR) { 416 if (ip_cmpbuf(sctp->sctp_rthdr, sctp->sctp_rthdrlen, 417 (ipp->ipp_fields & IPPF_RTHDR), 418 ipp->ipp_rthdr, ipp->ipp_rthdrlen)) { 419 optlen += sizeof (*cmsg) + ipp->ipp_rthdrlen; 420 if (hdrlen == 0) 421 hdrlen = sizeof (struct T_unitdata_ind); 422 addflag |= SCTP_IPV6_RECVRTHDR; 423 if (!ip_allocbuf((void **)&sctp->sctp_rthdr, 424 &sctp->sctp_rthdrlen, 425 (ipp->ipp_fields & IPPF_RTHDR), 426 ipp->ipp_rthdr, ipp->ipp_rthdrlen)) 427 return (-1); 428 } 429 } 430 /* If app asked for dest headers and it has changed ... */ 431 if ((sctp->sctp_ipv6_recvancillary & SCTP_IPV6_RECVDSTOPTS) && 432 ip_cmpbuf(sctp->sctp_dstopts, sctp->sctp_dstoptslen, 433 (ipp->ipp_fields & IPPF_DSTOPTS), 434 ipp->ipp_dstopts, ipp->ipp_dstoptslen)) { 435 optlen += sizeof (*cmsg) + ipp->ipp_dstoptslen; 436 if (hdrlen == 0) 437 hdrlen = sizeof (struct T_unitdata_ind); 438 addflag |= SCTP_IPV6_RECVDSTOPTS; 439 if (!ip_allocbuf((void **)&sctp->sctp_dstopts, 440 &sctp->sctp_dstoptslen, 441 (ipp->ipp_fields & IPPF_DSTOPTS), 442 ipp->ipp_dstopts, ipp->ipp_dstoptslen)) 443 return (-1); 444 } 445 noancillary: 446 /* Nothing to add */ 447 if (hdrlen == 0) 448 return (-1); 449 450 mp1 = allocb(hdrlen + optlen + sizeof (void *), BPRI_MED); 451 if (mp1 == NULL) 452 return (-1); 453 mp1->b_cont = *mp; 454 *mp = mp1; 455 mp1->b_rptr += sizeof (void *); /* pointer worth of padding */ 456 mp1->b_wptr = mp1->b_rptr + hdrlen + optlen; 457 DB_TYPE(mp1) = M_PROTO; 458 tudi = (struct T_unitdata_ind *)mp1->b_rptr; 459 tudi->PRIM_type = T_UNITDATA_IND; 460 tudi->SRC_length = sin4 ? sizeof (*sin4) : sizeof (*sin6); 461 tudi->SRC_offset = sizeof (*tudi); 462 tudi->OPT_offset = sizeof (*tudi) + tudi->SRC_length; 463 tudi->OPT_length = optlen; 464 if (sin4) { 465 bcopy(sin4, tudi + 1, sizeof (*sin4)); 466 } else { 467 bcopy(sin6, tudi + 1, sizeof (*sin6)); 468 } 469 optptr = (uchar_t *)tudi + tudi->OPT_offset; 470 471 if (sctp->sctp_recvsndrcvinfo) { 472 /* XXX need backout method if memory allocation fails. */ 473 struct sctp_sndrcvinfo *sri; 474 475 cmsg = (struct cmsghdr *)optptr; 476 cmsg->cmsg_level = IPPROTO_SCTP; 477 cmsg->cmsg_type = SCTP_SNDRCV; 478 cmsg->cmsg_len = sizeof (*cmsg) + sizeof (*sri); 479 optptr += sizeof (*cmsg); 480 481 sri = (struct sctp_sndrcvinfo *)(cmsg + 1); 482 ASSERT(OK_32PTR(sri)); 483 sri->sinfo_stream = ntohs(dcp->sdh_sid); 484 sri->sinfo_ssn = ntohs(dcp->sdh_ssn); 485 if (SCTP_DATA_GET_UBIT(dcp)) { 486 sri->sinfo_flags = MSG_UNORDERED; 487 } else { 488 sri->sinfo_flags = 0; 489 } 490 sri->sinfo_ppid = dcp->sdh_payload_id; 491 sri->sinfo_context = 0; 492 sri->sinfo_timetolive = 0; 493 sri->sinfo_tsn = ntohl(dcp->sdh_tsn); 494 sri->sinfo_cumtsn = sctp->sctp_ftsn; 495 sri->sinfo_assoc_id = 0; 496 497 optptr += sizeof (*sri); 498 } 499 500 /* 501 * If app asked for pktinfo and the index has changed ... 502 * Note that the local address never changes for the connection. 503 */ 504 if (addflag & SCTP_IPV6_RECVPKTINFO) { 505 struct in6_pktinfo *pkti; 506 507 cmsg = (struct cmsghdr *)optptr; 508 cmsg->cmsg_level = IPPROTO_IPV6; 509 cmsg->cmsg_type = IPV6_PKTINFO; 510 cmsg->cmsg_len = sizeof (*cmsg) + sizeof (*pkti); 511 optptr += sizeof (*cmsg); 512 513 pkti = (struct in6_pktinfo *)optptr; 514 if (sctp->sctp_ipversion == IPV6_VERSION) 515 pkti->ipi6_addr = sctp->sctp_ip6h->ip6_src; 516 else 517 IN6_IPADDR_TO_V4MAPPED(sctp->sctp_ipha->ipha_src, 518 &pkti->ipi6_addr); 519 pkti->ipi6_ifindex = ipp->ipp_ifindex; 520 optptr += sizeof (*pkti); 521 ASSERT(OK_32PTR(optptr)); 522 /* Save as "last" value */ 523 sctp->sctp_recvifindex = ipp->ipp_ifindex; 524 } 525 /* If app asked for hoplimit and it has changed ... */ 526 if (addflag & SCTP_IPV6_RECVHOPLIMIT) { 527 cmsg = (struct cmsghdr *)optptr; 528 cmsg->cmsg_level = IPPROTO_IPV6; 529 cmsg->cmsg_type = IPV6_HOPLIMIT; 530 cmsg->cmsg_len = sizeof (*cmsg) + sizeof (uint_t); 531 optptr += sizeof (*cmsg); 532 533 *(uint_t *)optptr = ipp->ipp_hoplimit; 534 optptr += sizeof (uint_t); 535 ASSERT(OK_32PTR(optptr)); 536 /* Save as "last" value */ 537 sctp->sctp_recvhops = ipp->ipp_hoplimit; 538 } 539 if (addflag & SCTP_IPV6_RECVHOPOPTS) { 540 cmsg = (struct cmsghdr *)optptr; 541 cmsg->cmsg_level = IPPROTO_IPV6; 542 cmsg->cmsg_type = IPV6_HOPOPTS; 543 cmsg->cmsg_len = sizeof (*cmsg) + ipp->ipp_hopoptslen; 544 optptr += sizeof (*cmsg); 545 546 bcopy(ipp->ipp_hopopts, optptr, ipp->ipp_hopoptslen); 547 optptr += ipp->ipp_hopoptslen; 548 ASSERT(OK_32PTR(optptr)); 549 /* Save as last value */ 550 ip_savebuf((void **)&sctp->sctp_hopopts, 551 &sctp->sctp_hopoptslen, 552 (ipp->ipp_fields & IPPF_HOPOPTS), 553 ipp->ipp_hopopts, ipp->ipp_hopoptslen); 554 } 555 if (addflag & SCTP_IPV6_RECVRTDSTOPTS) { 556 cmsg = (struct cmsghdr *)optptr; 557 cmsg->cmsg_level = IPPROTO_IPV6; 558 cmsg->cmsg_type = IPV6_RTHDRDSTOPTS; 559 cmsg->cmsg_len = sizeof (*cmsg) + ipp->ipp_rtdstoptslen; 560 optptr += sizeof (*cmsg); 561 562 bcopy(ipp->ipp_rtdstopts, optptr, ipp->ipp_rtdstoptslen); 563 optptr += ipp->ipp_rtdstoptslen; 564 ASSERT(OK_32PTR(optptr)); 565 /* Save as last value */ 566 ip_savebuf((void **)&sctp->sctp_rtdstopts, 567 &sctp->sctp_rtdstoptslen, 568 (ipp->ipp_fields & IPPF_RTDSTOPTS), 569 ipp->ipp_rtdstopts, ipp->ipp_rtdstoptslen); 570 } 571 if (addflag & SCTP_IPV6_RECVRTHDR) { 572 cmsg = (struct cmsghdr *)optptr; 573 cmsg->cmsg_level = IPPROTO_IPV6; 574 cmsg->cmsg_type = IPV6_RTHDR; 575 cmsg->cmsg_len = sizeof (*cmsg) + ipp->ipp_rthdrlen; 576 optptr += sizeof (*cmsg); 577 578 bcopy(ipp->ipp_rthdr, optptr, ipp->ipp_rthdrlen); 579 optptr += ipp->ipp_rthdrlen; 580 ASSERT(OK_32PTR(optptr)); 581 /* Save as last value */ 582 ip_savebuf((void **)&sctp->sctp_rthdr, 583 &sctp->sctp_rthdrlen, 584 (ipp->ipp_fields & IPPF_RTHDR), 585 ipp->ipp_rthdr, ipp->ipp_rthdrlen); 586 } 587 if (addflag & SCTP_IPV6_RECVDSTOPTS) { 588 cmsg = (struct cmsghdr *)optptr; 589 cmsg->cmsg_level = IPPROTO_IPV6; 590 cmsg->cmsg_type = IPV6_DSTOPTS; 591 cmsg->cmsg_len = sizeof (*cmsg) + ipp->ipp_dstoptslen; 592 optptr += sizeof (*cmsg); 593 594 bcopy(ipp->ipp_dstopts, optptr, ipp->ipp_dstoptslen); 595 optptr += ipp->ipp_dstoptslen; 596 ASSERT(OK_32PTR(optptr)); 597 /* Save as last value */ 598 ip_savebuf((void **)&sctp->sctp_dstopts, 599 &sctp->sctp_dstoptslen, 600 (ipp->ipp_fields & IPPF_DSTOPTS), 601 ipp->ipp_dstopts, ipp->ipp_dstoptslen); 602 } 603 604 ASSERT(optptr == mp1->b_wptr); 605 606 return (0); 607 } 608 609 void 610 sctp_free_reass(sctp_instr_t *sip) 611 { 612 mblk_t *mp, *mpnext, *mctl; 613 614 for (mp = sip->istr_reass; mp != NULL; mp = mpnext) { 615 mpnext = mp->b_next; 616 mp->b_next = NULL; 617 mp->b_prev = NULL; 618 if (DB_TYPE(mp) == M_CTL) { 619 mctl = mp; 620 ASSERT(mp->b_cont != NULL); 621 mp = mp->b_cont; 622 mctl->b_cont = NULL; 623 freeb(mctl); 624 } 625 freemsg(mp); 626 } 627 } 628 629 /* 630 * If the series of data fragments of which dmp is a part is successfully 631 * reassembled, the first mblk in the series is returned. dc is adjusted 632 * to point at the data chunk in the lead mblk, and b_rptr also points to 633 * the data chunk; the following mblk's b_rptr's point at the actual payload. 634 * 635 * If the series is not yet reassembled, NULL is returned. dc is not changed. 636 * XXX should probably move this up into the state machine. 637 */ 638 639 /* Fragment list for un-ordered messages. Partial delivery is not supported */ 640 static mblk_t * 641 sctp_uodata_frag(sctp_t *sctp, mblk_t *dmp, sctp_data_hdr_t **dc) 642 { 643 mblk_t *hmp; 644 mblk_t *begin = NULL; 645 mblk_t *end = NULL; 646 sctp_data_hdr_t *qdc; 647 uint32_t ntsn; 648 uint32_t tsn = ntohl((*dc)->sdh_tsn); 649 #ifdef DEBUG 650 mblk_t *mp1; 651 #endif 652 653 /* First frag. */ 654 if (sctp->sctp_uo_frags == NULL) { 655 sctp->sctp_uo_frags = dmp; 656 return (NULL); 657 } 658 hmp = sctp->sctp_uo_frags; 659 /* 660 * Insert the segment according to the TSN, fragmented unordered 661 * chunks are sequenced by TSN. 662 */ 663 while (hmp != NULL) { 664 qdc = (sctp_data_hdr_t *)hmp->b_rptr; 665 ntsn = ntohl(qdc->sdh_tsn); 666 if (SEQ_GT(ntsn, tsn)) { 667 if (hmp->b_prev == NULL) { 668 dmp->b_next = hmp; 669 hmp->b_prev = dmp; 670 sctp->sctp_uo_frags = dmp; 671 } else { 672 dmp->b_next = hmp; 673 dmp->b_prev = hmp->b_prev; 674 hmp->b_prev->b_next = dmp; 675 hmp->b_prev = dmp; 676 } 677 break; 678 } 679 if (hmp->b_next == NULL) { 680 hmp->b_next = dmp; 681 dmp->b_prev = hmp; 682 break; 683 } 684 hmp = hmp->b_next; 685 } 686 /* check if we completed a msg */ 687 if (SCTP_DATA_GET_BBIT(*dc)) { 688 begin = dmp; 689 } else if (SCTP_DATA_GET_EBIT(*dc)) { 690 end = dmp; 691 } 692 /* 693 * We walk consecutive TSNs backwards till we get a seg. with 694 * the B bit 695 */ 696 if (begin == NULL) { 697 for (hmp = dmp->b_prev; hmp != NULL; hmp = hmp->b_prev) { 698 qdc = (sctp_data_hdr_t *)hmp->b_rptr; 699 ntsn = ntohl(qdc->sdh_tsn); 700 if ((int32_t)(tsn - ntsn) > 1) { 701 return (NULL); 702 } 703 if (SCTP_DATA_GET_BBIT(qdc)) { 704 begin = hmp; 705 break; 706 } 707 tsn = ntsn; 708 } 709 } 710 tsn = ntohl((*dc)->sdh_tsn); 711 /* 712 * We walk consecutive TSNs till we get a seg. with the E bit 713 */ 714 if (end == NULL) { 715 for (hmp = dmp->b_next; hmp != NULL; hmp = hmp->b_next) { 716 qdc = (sctp_data_hdr_t *)hmp->b_rptr; 717 ntsn = ntohl(qdc->sdh_tsn); 718 if ((int32_t)(ntsn - tsn) > 1) { 719 return (NULL); 720 } 721 if (SCTP_DATA_GET_EBIT(qdc)) { 722 end = hmp; 723 break; 724 } 725 tsn = ntsn; 726 } 727 } 728 if (begin == NULL || end == NULL) { 729 return (NULL); 730 } 731 /* Got one!, Remove the msg from the list */ 732 if (sctp->sctp_uo_frags == begin) { 733 ASSERT(begin->b_prev == NULL); 734 sctp->sctp_uo_frags = end->b_next; 735 if (end->b_next != NULL) 736 end->b_next->b_prev = NULL; 737 } else { 738 begin->b_prev->b_next = end->b_next; 739 if (end->b_next != NULL) 740 end->b_next->b_prev = begin->b_prev; 741 } 742 begin->b_prev = NULL; 743 end->b_next = NULL; 744 745 /* 746 * Null out b_next and b_prev and chain using b_cont. 747 */ 748 dmp = end = begin; 749 hmp = begin->b_next; 750 *dc = (sctp_data_hdr_t *)begin->b_rptr; 751 begin->b_next = NULL; 752 while (hmp != NULL) { 753 qdc = (sctp_data_hdr_t *)hmp->b_rptr; 754 hmp->b_rptr = (uchar_t *)(qdc + 1); 755 end = hmp->b_next; 756 dmp->b_cont = hmp; 757 dmp = hmp; 758 759 if (end != NULL) 760 hmp->b_next = NULL; 761 hmp->b_prev = NULL; 762 hmp = end; 763 } 764 BUMP_LOCAL(sctp->sctp_reassmsgs); 765 #ifdef DEBUG 766 mp1 = begin; 767 while (mp1 != NULL) { 768 ASSERT(mp1->b_next == NULL); 769 ASSERT(mp1->b_prev == NULL); 770 mp1 = mp1->b_cont; 771 } 772 #endif 773 return (begin); 774 } 775 /* 776 * Fragment list for ordered messages. 777 * If no error occures, error is set to 0. If we run out of memory, error 778 * is set to 1. If the peer commits a fatal error (like using different 779 * sequence numbers for the same data fragment series), the association is 780 * aborted and error is set to 2. 781 */ 782 static mblk_t * 783 sctp_data_frag(sctp_t *sctp, mblk_t *dmp, sctp_data_hdr_t **dc, int *error, 784 sctp_instr_t *sip, int trypartial, int *tpfinished) 785 { 786 mblk_t *hmp; 787 mblk_t *pmp; 788 mblk_t *qmp; 789 mblk_t *mp; 790 mblk_t *prev; 791 mblk_t *prevprev; 792 mblk_t *first_mp; 793 sctp_reass_t *srp; 794 sctp_data_hdr_t *qdc; 795 sctp_data_hdr_t *bdc; 796 sctp_data_hdr_t *edc; 797 uint32_t tsn; 798 799 /* 800 * We can overwrite the Link Layer + IP header here, I suppose. 801 * The M_CTL does not leave this function. We need to check 802 * DB_REF(dmp) before using DB_BASE(dmp), since there could be 803 * two fragments for different ssns in the same mblk. 804 */ 805 #define SCTP_NEW_REASS(nmp, dmp, srp, seterror) \ 806 if ((DB_REF(dmp) == 2) && (MBLKHEAD(dmp) >= \ 807 (sizeof (*(srp)) + sizeof (sctp_hdr_t))) && \ 808 (IS_P2ALIGNED(DB_BASE(dmp), sizeof (uintptr_t)))) { \ 809 (nmp) = (dmp); \ 810 } else { \ 811 (nmp) = allocb(sizeof (*(srp)), BPRI_MED); \ 812 if ((nmp) == NULL) { \ 813 switch (seterror) { \ 814 case B_TRUE: \ 815 *error = 1; \ 816 break; \ 817 } \ 818 return (NULL); \ 819 } \ 820 DB_TYPE(nmp) = M_CTL; \ 821 (nmp)->b_cont = dmp; \ 822 } \ 823 (srp) = (sctp_reass_t *)DB_BASE(nmp); 824 825 *error = 0; 826 827 /* find the reassembly queue for this data chunk */ 828 hmp = qmp = sip->istr_reass; 829 for (; hmp != NULL; hmp = hmp->b_next) { 830 srp = (sctp_reass_t *)DB_BASE(hmp); 831 if (ntohs((*dc)->sdh_ssn) == srp->ssn) 832 goto foundit; 833 else if (SSN_GT(srp->ssn, ntohs((*dc)->sdh_ssn))) 834 break; 835 qmp = hmp; 836 } 837 838 SCTP_NEW_REASS(pmp, dmp, srp, B_TRUE); 839 srp->ssn = ntohs((*dc)->sdh_ssn); 840 srp->needed = 0; 841 srp->got = 1; 842 srp->tail = dmp; 843 srp->partial_delivered = B_FALSE; 844 845 if (hmp != NULL) { 846 if (sip->istr_reass == hmp) { 847 sip->istr_reass = pmp; 848 pmp->b_next = hmp; 849 pmp->b_prev = NULL; 850 hmp->b_prev = pmp; 851 } else { 852 qmp->b_next = pmp; 853 pmp->b_prev = qmp; 854 pmp->b_next = hmp; 855 hmp->b_prev = pmp; 856 } 857 } else { 858 /* make a new reass head and stick it on the end */ 859 if (sip->istr_reass == NULL) { 860 sip->istr_reass = pmp; 861 pmp->b_prev = NULL; 862 } else { 863 qmp->b_next = pmp; 864 pmp->b_prev = qmp; 865 } 866 pmp->b_next = NULL; 867 } 868 return (NULL); 869 foundit: 870 /* 871 * else already have a reassembly queue. Insert the new data chunk 872 * in the reassemble queue. Try the tail first, on the assumption 873 * that the fragments are coming in in order. 874 */ 875 876 qmp = srp->tail; 877 qdc = (sctp_data_hdr_t *)qmp->b_rptr; 878 ASSERT(qmp->b_cont == NULL); 879 880 /* XXXIs it fine to do this just here? */ 881 if ((*dc)->sdh_sid != qdc->sdh_sid) { 882 /* our peer is fatally confused; XXX abort the assc */ 883 *error = 2; 884 return (NULL); 885 } 886 if (SEQ_GT(ntohl((*dc)->sdh_tsn), ntohl(qdc->sdh_tsn))) { 887 qmp->b_cont = dmp; 888 srp->tail = dmp; 889 dmp->b_cont = NULL; 890 goto inserted; 891 } 892 893 /* Next check for insertion at the beginning */ 894 qmp = (DB_TYPE(hmp) == M_DATA) ? hmp : hmp->b_cont; 895 qdc = (sctp_data_hdr_t *)qmp->b_rptr; 896 if (SEQ_LT(ntohl((*dc)->sdh_tsn), ntohl(qdc->sdh_tsn))) { 897 if (DB_TYPE(hmp) == M_DATA) { 898 sctp_reass_t *srp1 = srp; 899 900 SCTP_NEW_REASS(pmp, dmp, srp, B_TRUE); 901 ASSERT(pmp->b_prev == NULL && pmp->b_next == NULL); 902 if (sip->istr_reass == hmp) { 903 sip->istr_reass = pmp; 904 if (hmp->b_next != NULL) { 905 hmp->b_next->b_prev = pmp; 906 pmp->b_next = hmp->b_next; 907 } 908 } else { 909 hmp->b_prev->b_next = pmp; 910 pmp->b_prev = hmp->b_prev; 911 if (hmp->b_next != NULL) { 912 hmp->b_next->b_prev = pmp; 913 pmp->b_next = hmp->b_next; 914 } 915 } 916 srp->ssn = srp1->ssn; 917 srp->needed = srp1->needed; 918 srp->got = srp1->got; 919 srp->tail = srp1->tail; 920 srp->partial_delivered = srp1->partial_delivered; 921 hmp->b_next = hmp->b_prev = NULL; 922 dmp->b_cont = hmp; 923 hmp = pmp; 924 } else { 925 ASSERT(DB_TYPE(hmp) == M_CTL); 926 dmp->b_cont = qmp; 927 hmp->b_cont = dmp; 928 } 929 goto inserted; 930 } 931 932 /* Insert somewhere in the middle */ 933 for (;;) { 934 /* Tail check above should have caught this */ 935 ASSERT(qmp->b_cont != NULL); 936 937 qdc = (sctp_data_hdr_t *)qmp->b_cont->b_rptr; 938 if (SEQ_LT(ntohl((*dc)->sdh_tsn), ntohl(qdc->sdh_tsn))) { 939 /* insert here */ 940 dmp->b_cont = qmp->b_cont; 941 qmp->b_cont = dmp; 942 break; 943 } 944 qmp = qmp->b_cont; 945 } 946 947 inserted: 948 (srp->got)++; 949 first_mp = (DB_TYPE(hmp) == M_DATA) ? hmp : hmp->b_cont; 950 if (srp->needed == 0) { 951 /* check if we have the first and last fragments */ 952 bdc = (sctp_data_hdr_t *)first_mp->b_rptr; 953 edc = (sctp_data_hdr_t *)srp->tail->b_rptr; 954 955 /* calculate how many fragments are needed, if possible */ 956 if (SCTP_DATA_GET_BBIT(bdc) && SCTP_DATA_GET_EBIT(edc)) 957 srp->needed = ntohl(edc->sdh_tsn) - 958 ntohl(bdc->sdh_tsn) + 1; 959 } 960 961 if (srp->needed != srp->got) { 962 if (!trypartial) 963 return (NULL); 964 /* 965 * Try partial delivery. We need a consecutive run of 966 * at least two chunks, starting from the first chunk 967 * (which may have been the last + 1 chunk from a 968 * previous partial delivery). 969 */ 970 dprint(4, ("trypartial: got=%d, needed=%d\n", 971 (int)(srp->got), (int)(srp->needed))); 972 mp = first_mp; 973 if (mp->b_cont == NULL) { 974 /* need at least two chunks */ 975 dprint(4, ("trypartial: only 1 chunk\n")); 976 return (NULL); 977 } 978 979 qdc = (sctp_data_hdr_t *)mp->b_rptr; 980 if (!SCTP_DATA_GET_BBIT(qdc)) { 981 /* don't have first chunk; can't do it. */ 982 dprint(4, ("trypartial: no beginning\n")); 983 return (NULL); 984 } 985 986 tsn = ntohl(qdc->sdh_tsn) + 1; 987 988 /* 989 * This loop has two exit conditions: the 990 * end of received chunks has been reached, or 991 * there is a break in the sequence. We want 992 * to chop the reassembly list as follows (the 993 * numbers are TSNs): 994 * 10 -> 11 -> | 12 (end of chunks) 995 * 10 -> 11 -> | 12 -> 14 (break in sequence) 996 */ 997 prevprev = prev = mp; 998 mp = mp->b_cont; 999 while (mp != NULL) { 1000 qdc = (sctp_data_hdr_t *)mp->b_rptr; 1001 if (ntohl(qdc->sdh_tsn) != tsn) { 1002 /* 1003 * break in sequence. 1004 * 1st and 2nd chunks are not sequntial. 1005 */ 1006 if (mp == first_mp->b_cont) 1007 return (NULL); 1008 /* Back up mp and prev */ 1009 mp = prev; 1010 prev = prevprev; 1011 break; 1012 } 1013 1014 /* end of sequence */ 1015 if (mp->b_cont == NULL) 1016 break; 1017 1018 prevprev = prev; 1019 prev = mp; 1020 mp = mp->b_cont; 1021 tsn++; 1022 } 1023 if (DB_TYPE(hmp) == M_DATA) { 1024 sctp_reass_t *srp1 = srp; 1025 1026 SCTP_NEW_REASS(pmp, mp, srp, B_FALSE); 1027 ASSERT(pmp->b_prev == NULL && pmp->b_next == NULL); 1028 if (sip->istr_reass == hmp) { 1029 sip->istr_reass = pmp; 1030 if (hmp->b_next != NULL) { 1031 hmp->b_next->b_prev = pmp; 1032 pmp->b_next = hmp->b_next; 1033 } 1034 } else { 1035 hmp->b_prev->b_next = pmp; 1036 pmp->b_prev = hmp->b_prev; 1037 if (hmp->b_next != NULL) { 1038 hmp->b_next->b_prev = pmp; 1039 pmp->b_next = hmp->b_next; 1040 } 1041 } 1042 srp->ssn = srp1->ssn; 1043 srp->needed = srp1->needed; 1044 srp->got = srp1->got; 1045 srp->tail = srp1->tail; 1046 hmp->b_next = hmp->b_prev = NULL; 1047 dmp = hmp; 1048 hmp = pmp; 1049 } else { 1050 ASSERT(DB_TYPE(hmp) == M_CTL); 1051 dmp = hmp->b_cont; 1052 hmp->b_cont = mp; 1053 } 1054 /* 1055 * mp now points at the last chunk in the sequence, 1056 * and prev points to mp's previous in the list. 1057 * We chop the list at prev, and convert mp into the 1058 * new list head by setting the B bit. Subsequence 1059 * fragment deliveries will follow the normal reassembly 1060 * path. 1061 */ 1062 prev->b_cont = NULL; 1063 bdc = (sctp_data_hdr_t *)mp->b_rptr; 1064 SCTP_DATA_SET_BBIT(bdc); 1065 *tpfinished = 0; 1066 srp->partial_delivered = B_TRUE; 1067 1068 dprint(4, ("trypartial: got some, got=%d, needed=%d\n", 1069 (int)(srp->got), (int)(srp->needed))); 1070 goto fixup; 1071 } 1072 1073 /* 1074 * else reassembly done; prepare the data for delivery. 1075 * First unlink hmp from the ssn list. 1076 */ 1077 if (sip->istr_reass == hmp) { 1078 sip->istr_reass = hmp->b_next; 1079 if (hmp->b_next) { 1080 hmp->b_next->b_prev = NULL; 1081 } 1082 } else { 1083 ASSERT(hmp->b_prev != NULL); 1084 hmp->b_prev->b_next = hmp->b_next; 1085 if (hmp->b_next) { 1086 hmp->b_next->b_prev = hmp->b_prev; 1087 } 1088 } 1089 1090 /* 1091 * Using b_prev and b_next was a little sinful, but OK since 1092 * this mblk is never put*'d. However, freeb() will still 1093 * ASSERT that they are unused, so we need to NULL them out now. 1094 */ 1095 hmp->b_next = NULL; 1096 hmp->b_prev = NULL; 1097 dmp = hmp; 1098 if (DB_TYPE(hmp) == M_CTL) { 1099 dmp = dmp->b_cont; 1100 hmp->b_cont = NULL; 1101 freeb(hmp); 1102 } 1103 *tpfinished = 1; 1104 1105 fixup: 1106 /* 1107 * Adjust all mblk's except the lead so their rptr's point to the 1108 * payload. sctp_data_chunk() will need to process the lead's 1109 * data chunk section, so leave it's rptr pointing at the data chunk. 1110 */ 1111 *dc = (sctp_data_hdr_t *)dmp->b_rptr; 1112 if (trypartial && !(*tpfinished)) { 1113 (srp->got)--; 1114 ASSERT(srp->got != 0); 1115 if (srp->needed != 0) { 1116 (srp->needed)--; 1117 ASSERT(srp->needed != 0); 1118 } 1119 } 1120 for (qmp = dmp->b_cont; qmp; qmp = qmp->b_cont) { 1121 qdc = (sctp_data_hdr_t *)qmp->b_rptr; 1122 qmp->b_rptr = (uchar_t *)(qdc + 1); 1123 1124 /* 1125 * If in partial delivery, deduct the balance from got 1126 * and needed here, now that we know we are actually 1127 * delivering these data. 1128 */ 1129 if (trypartial && !(*tpfinished)) { 1130 (srp->got)--; 1131 ASSERT(srp->got != 0); 1132 if (srp->needed != 0) { 1133 (srp->needed)--; 1134 ASSERT(srp->needed != 0); 1135 } 1136 } 1137 } 1138 BUMP_LOCAL(sctp->sctp_reassmsgs); 1139 1140 return (dmp); 1141 } 1142 1143 static void 1144 sctp_add_dup(uint32_t tsn, mblk_t **dups) 1145 { 1146 mblk_t *mp; 1147 size_t bsize = SCTP_DUP_MBLK_SZ * sizeof (tsn); 1148 1149 if (dups == NULL) { 1150 return; 1151 } 1152 1153 /* first time? */ 1154 if (*dups == NULL) { 1155 *dups = allocb(bsize, BPRI_MED); 1156 if (*dups == NULL) { 1157 return; 1158 } 1159 } 1160 1161 mp = *dups; 1162 if ((mp->b_wptr - mp->b_rptr) >= bsize) { 1163 /* maximum reached */ 1164 return; 1165 } 1166 1167 /* add the duplicate tsn */ 1168 bcopy(&tsn, mp->b_wptr, sizeof (tsn)); 1169 mp->b_wptr += sizeof (tsn); 1170 ASSERT((mp->b_wptr - mp->b_rptr) <= bsize); 1171 } 1172 1173 static void 1174 sctp_data_chunk(sctp_t *sctp, sctp_chunk_hdr_t *ch, mblk_t *mp, mblk_t **dups, 1175 sctp_faddr_t *fp, ip6_pkt_t *ipp) 1176 { 1177 sctp_data_hdr_t *dc; 1178 mblk_t *dmp, *pmp; 1179 mblk_t *errmp; 1180 sctp_instr_t *instr; 1181 int ubit; 1182 int isfrag; 1183 uint16_t ssn; 1184 uint32_t oftsn; 1185 boolean_t can_deliver = B_TRUE; 1186 uint32_t tsn; 1187 int dlen; 1188 int trypartial = 0; 1189 int tpfinished = 1; 1190 int32_t new_rwnd; 1191 sctp_stack_t *sctps = sctp->sctp_sctps; 1192 1193 /* The following are used multiple times, so we inline them */ 1194 #define SCTP_ACK_IT(sctp, tsn) \ 1195 if (tsn == sctp->sctp_ftsn) { \ 1196 dprint(2, ("data_chunk: acking next %x\n", tsn)); \ 1197 (sctp)->sctp_ftsn++; \ 1198 if ((sctp)->sctp_sack_gaps > 0) \ 1199 (sctp)->sctp_force_sack = 1; \ 1200 } else if (SEQ_GT(tsn, sctp->sctp_ftsn)) { \ 1201 /* Got a gap; record it */ \ 1202 dprint(2, ("data_chunk: acking gap %x\n", tsn)); \ 1203 sctp_ack_add(&sctp->sctp_sack_info, tsn, \ 1204 &sctp->sctp_sack_gaps); \ 1205 sctp->sctp_force_sack = 1; \ 1206 } 1207 1208 errmp = NULL; 1209 dmp = NULL; 1210 1211 dc = (sctp_data_hdr_t *)ch; 1212 tsn = ntohl(dc->sdh_tsn); 1213 1214 dprint(3, ("sctp_data_chunk: mp=%p tsn=%x\n", (void *)mp, tsn)); 1215 1216 /* Check for duplicates */ 1217 if (SEQ_LT(tsn, sctp->sctp_ftsn)) { 1218 dprint(4, ("sctp_data_chunk: dropping duplicate\n")); 1219 sctp->sctp_force_sack = 1; 1220 sctp_add_dup(dc->sdh_tsn, dups); 1221 return; 1222 } 1223 1224 if (sctp->sctp_sack_info != NULL) { 1225 sctp_set_t *sp; 1226 1227 for (sp = sctp->sctp_sack_info; sp; sp = sp->next) { 1228 if (SEQ_GEQ(tsn, sp->begin) && SEQ_LEQ(tsn, sp->end)) { 1229 dprint(4, 1230 ("sctp_data_chunk: dropping dup > cumtsn\n")); 1231 sctp->sctp_force_sack = 1; 1232 sctp_add_dup(dc->sdh_tsn, dups); 1233 return; 1234 } 1235 } 1236 } 1237 1238 /* We cannot deliver anything up now but we still need to handle it. */ 1239 if (SCTP_IS_DETACHED(sctp)) { 1240 BUMP_MIB(&sctps->sctps_mib, sctpInClosed); 1241 can_deliver = B_FALSE; 1242 } 1243 1244 dlen = ntohs(dc->sdh_len) - sizeof (*dc); 1245 1246 /* Check for buffer space */ 1247 if (sctp->sctp_rwnd - sctp->sctp_rxqueued < dlen) { 1248 /* Drop and SACK, but don't advance the cumulative TSN. */ 1249 sctp->sctp_force_sack = 1; 1250 dprint(0, ("sctp_data_chunk: exceed rwnd %d rxqueued %d " 1251 "ssn %d tsn %x\n", sctp->sctp_rwnd, 1252 sctp->sctp_rxqueued, dc->sdh_ssn, ntohl(dc->sdh_tsn))); 1253 return; 1254 } 1255 1256 if (ntohs(dc->sdh_sid) >= sctp->sctp_num_istr) { 1257 uint16_t inval_parm[2]; 1258 1259 inval_parm[0] = dc->sdh_sid; 1260 /* RESERVED to be ignored at the receiving end */ 1261 inval_parm[1] = 0; 1262 /* ack and drop it */ 1263 errmp = sctp_make_err(sctp, SCTP_ERR_BAD_SID, 1264 (char *)inval_parm, sizeof (inval_parm)); 1265 SCTP_ACK_IT(sctp, tsn); 1266 if (errmp != NULL) 1267 sctp_send_err(sctp, errmp, NULL); 1268 return; 1269 } 1270 1271 ubit = SCTP_DATA_GET_UBIT(dc); 1272 ASSERT(sctp->sctp_instr != NULL); 1273 instr = &sctp->sctp_instr[ntohs(dc->sdh_sid)]; 1274 /* Initialize the stream, if not yet used */ 1275 if (instr->sctp == NULL) 1276 instr->sctp = sctp; 1277 /* 1278 * If we are getting low on buffers set trypartial to try 1279 * a partial delivery if we are reassembling a fragmented 1280 * message. Only do this if we can immediately deliver the 1281 * partially assembled message, and only partially deliver 1282 * one message at a time (i.e. messages cannot be intermixed 1283 * arriving at the upper layer). A simple way to enforce 1284 * this is to only try partial delivery if this TSN is 1285 * the next expected TSN. Partial Delivery not supported 1286 * for un-ordered message. 1287 */ 1288 isfrag = !(SCTP_DATA_GET_BBIT(dc) && SCTP_DATA_GET_EBIT(dc)); 1289 ssn = ntohs(dc->sdh_ssn); 1290 if ((sctp->sctp_rwnd - sctp->sctp_rxqueued < SCTP_RECV_LOWATER) && 1291 !ubit && isfrag && (tsn == sctp->sctp_ftsn)) { 1292 trypartial = 1; 1293 } 1294 1295 dmp = dupb(mp); 1296 if (dmp == NULL) { 1297 /* drop it and don't ack it, causing the peer to retransmit */ 1298 return; 1299 } 1300 dmp->b_wptr = (uchar_t *)ch + ntohs(ch->sch_len); 1301 1302 sctp->sctp_rxqueued += dlen; 1303 1304 oftsn = sctp->sctp_ftsn; 1305 1306 if (isfrag) { 1307 int error = 0; 1308 1309 /* fragmented data chunk */ 1310 dmp->b_rptr = (uchar_t *)dc; 1311 if (ubit) { 1312 dmp = sctp_uodata_frag(sctp, dmp, &dc); 1313 #if DEBUG 1314 if (dmp != NULL) { 1315 ASSERT(instr == 1316 &sctp->sctp_instr[ntohs(dc->sdh_sid)]); 1317 } 1318 #endif 1319 } else { 1320 dmp = sctp_data_frag(sctp, dmp, &dc, &error, instr, 1321 trypartial, &tpfinished); 1322 } 1323 if (error != 0) { 1324 sctp->sctp_rxqueued -= dlen; 1325 if (error == 1) { 1326 /* 1327 * out of memory; don't ack it so 1328 * the peer retransmits 1329 */ 1330 return; 1331 } else if (error == 2) { 1332 /* 1333 * fatal error (i.e. peer used different 1334 * ssn's for same fragmented data) -- 1335 * the association has been aborted. 1336 * XXX need to return errval so state 1337 * machine can also abort processing. 1338 */ 1339 dprint(0, ("error 2: must not happen!\n")); 1340 return; 1341 } 1342 } 1343 1344 if (dmp == NULL) { 1345 /* 1346 * Can't process this data now, but the cumulative 1347 * TSN may be advanced, so do the checks at done. 1348 */ 1349 SCTP_ACK_IT(sctp, tsn); 1350 goto done; 1351 } 1352 } 1353 1354 if (!ubit && !trypartial && ssn != instr->nextseq) { 1355 /* Adjust rptr to point at the data chunk for compares */ 1356 dmp->b_rptr = (uchar_t *)dc; 1357 1358 dprint(2, 1359 ("data_chunk: inserted %x in pq (ssn %d expected %d)\n", 1360 ntohl(dc->sdh_tsn), (int)(ssn), (int)(instr->nextseq))); 1361 1362 if (instr->istr_msgs == NULL) { 1363 instr->istr_msgs = dmp; 1364 ASSERT(dmp->b_prev == NULL && dmp->b_next == NULL); 1365 } else { 1366 mblk_t *imblk = instr->istr_msgs; 1367 sctp_data_hdr_t *idc; 1368 1369 /* 1370 * XXXNeed to take sequence wraps into account, 1371 * ... and a more efficient insertion algo. 1372 */ 1373 for (;;) { 1374 idc = (sctp_data_hdr_t *)imblk->b_rptr; 1375 if (SSN_GT(ntohs(idc->sdh_ssn), 1376 ntohs(dc->sdh_ssn))) { 1377 if (instr->istr_msgs == imblk) { 1378 instr->istr_msgs = dmp; 1379 dmp->b_next = imblk; 1380 imblk->b_prev = dmp; 1381 } else { 1382 ASSERT(imblk->b_prev != NULL); 1383 imblk->b_prev->b_next = dmp; 1384 dmp->b_prev = imblk->b_prev; 1385 imblk->b_prev = dmp; 1386 dmp->b_next = imblk; 1387 } 1388 break; 1389 } 1390 if (imblk->b_next == NULL) { 1391 imblk->b_next = dmp; 1392 dmp->b_prev = imblk; 1393 break; 1394 } 1395 imblk = imblk->b_next; 1396 } 1397 } 1398 (instr->istr_nmsgs)++; 1399 (sctp->sctp_istr_nmsgs)++; 1400 SCTP_ACK_IT(sctp, tsn); 1401 return; 1402 } 1403 1404 /* 1405 * Else we can deliver the data directly. Recalculate 1406 * dlen now since we may have reassembled data. 1407 */ 1408 dlen = dmp->b_wptr - (uchar_t *)dc - sizeof (*dc); 1409 for (pmp = dmp->b_cont; pmp != NULL; pmp = pmp->b_cont) 1410 dlen += pmp->b_wptr - pmp->b_rptr; 1411 ASSERT(sctp->sctp_rxqueued >= dlen); 1412 ASSERT(sctp->sctp_rwnd >= dlen); 1413 1414 /* Deliver the message. */ 1415 sctp->sctp_rxqueued -= dlen; 1416 1417 if (can_deliver) { 1418 dmp->b_rptr = (uchar_t *)(dc + 1); 1419 if (sctp_input_add_ancillary(sctp, &dmp, dc, fp, ipp) == 0) { 1420 dprint(1, ("sctp_data_chunk: delivering %lu bytes\n", 1421 msgdsize(dmp))); 1422 sctp->sctp_rwnd -= dlen; 1423 new_rwnd = sctp->sctp_ulp_recv(sctp->sctp_ulpd, dmp, 1424 tpfinished ? 0 : SCTP_PARTIAL_DATA); 1425 if (new_rwnd > sctp->sctp_rwnd) { 1426 sctp->sctp_rwnd = new_rwnd; 1427 } 1428 SCTP_ACK_IT(sctp, tsn); 1429 } else { 1430 /* Just free the message if we don't have memory. */ 1431 freemsg(dmp); 1432 return; 1433 } 1434 } else { 1435 /* About to free the data */ 1436 freemsg(dmp); 1437 SCTP_ACK_IT(sctp, tsn); 1438 } 1439 1440 /* 1441 * data, now enqueued, may already have been processed and free'd 1442 * by the ULP (or we may have just freed it above, if we could not 1443 * deliver it), so we must not reference it (this is why we kept 1444 * the ssn and ubit above). 1445 */ 1446 if (ubit != 0) { 1447 BUMP_LOCAL(sctp->sctp_iudchunks); 1448 goto done; 1449 } 1450 BUMP_LOCAL(sctp->sctp_idchunks); 1451 1452 /* 1453 * If there was a partial delivery and it has not finished, 1454 * don't pull anything from the pqueues. 1455 */ 1456 if (!tpfinished) { 1457 goto done; 1458 } 1459 1460 instr->nextseq = ssn + 1; 1461 /* Deliver any successive data chunks in the instr queue */ 1462 while (instr->istr_nmsgs > 0) { 1463 dmp = (mblk_t *)instr->istr_msgs; 1464 dc = (sctp_data_hdr_t *)dmp->b_rptr; 1465 ssn = ntohs(dc->sdh_ssn); 1466 /* Gap in the sequence */ 1467 if (ssn != instr->nextseq) 1468 break; 1469 1470 /* Else deliver the data */ 1471 (instr->istr_nmsgs)--; 1472 (instr->nextseq)++; 1473 (sctp->sctp_istr_nmsgs)--; 1474 1475 instr->istr_msgs = instr->istr_msgs->b_next; 1476 if (instr->istr_msgs != NULL) 1477 instr->istr_msgs->b_prev = NULL; 1478 dmp->b_next = dmp->b_prev = NULL; 1479 1480 dprint(2, ("data_chunk: pulling %x from pq (ssn %d)\n", 1481 ntohl(dc->sdh_tsn), (int)ssn)); 1482 1483 /* 1484 * If this chunk was reassembled, each b_cont represents 1485 * another TSN; advance ftsn now. 1486 */ 1487 dlen = dmp->b_wptr - dmp->b_rptr - sizeof (*dc); 1488 for (pmp = dmp->b_cont; pmp; pmp = pmp->b_cont) 1489 dlen += pmp->b_wptr - pmp->b_rptr; 1490 1491 ASSERT(sctp->sctp_rxqueued >= dlen); 1492 ASSERT(sctp->sctp_rwnd >= dlen); 1493 1494 sctp->sctp_rxqueued -= dlen; 1495 if (can_deliver) { 1496 dmp->b_rptr = (uchar_t *)(dc + 1); 1497 if (sctp_input_add_ancillary(sctp, &dmp, dc, fp, 1498 ipp) == 0) { 1499 dprint(1, ("sctp_data_chunk: delivering %lu " 1500 "bytes\n", msgdsize(dmp))); 1501 sctp->sctp_rwnd -= dlen; 1502 new_rwnd = sctp->sctp_ulp_recv(sctp->sctp_ulpd, 1503 dmp, tpfinished ? 0 : SCTP_PARTIAL_DATA); 1504 if (new_rwnd > sctp->sctp_rwnd) { 1505 sctp->sctp_rwnd = new_rwnd; 1506 } 1507 SCTP_ACK_IT(sctp, tsn); 1508 } else { 1509 freemsg(dmp); 1510 return; 1511 } 1512 } else { 1513 /* About to free the data */ 1514 freemsg(dmp); 1515 SCTP_ACK_IT(sctp, tsn); 1516 } 1517 } 1518 1519 done: 1520 1521 /* 1522 * If there are gap reports pending, check if advancing 1523 * the ftsn here closes a gap. If so, we can advance 1524 * ftsn to the end of the set. 1525 */ 1526 if (sctp->sctp_sack_info != NULL && 1527 sctp->sctp_ftsn == sctp->sctp_sack_info->begin) { 1528 sctp->sctp_ftsn = sctp->sctp_sack_info->end + 1; 1529 } 1530 /* 1531 * If ftsn has moved forward, maybe we can remove gap reports. 1532 * NB: dmp may now be NULL, so don't dereference it here. 1533 */ 1534 if (oftsn != sctp->sctp_ftsn && sctp->sctp_sack_info != NULL) { 1535 sctp_ack_rem(&sctp->sctp_sack_info, sctp->sctp_ftsn - 1, 1536 &sctp->sctp_sack_gaps); 1537 dprint(2, ("data_chunk: removed acks before %x (num=%d)\n", 1538 sctp->sctp_ftsn - 1, sctp->sctp_sack_gaps)); 1539 } 1540 1541 #ifdef DEBUG 1542 if (sctp->sctp_sack_info != NULL) { 1543 ASSERT(sctp->sctp_ftsn != sctp->sctp_sack_info->begin); 1544 } 1545 #endif 1546 1547 #undef SCTP_ACK_IT 1548 } 1549 1550 void 1551 sctp_fill_sack(sctp_t *sctp, unsigned char *dst, int sacklen) 1552 { 1553 sctp_chunk_hdr_t *sch; 1554 sctp_sack_chunk_t *sc; 1555 sctp_sack_frag_t *sf; 1556 uint16_t num_gaps = sctp->sctp_sack_gaps; 1557 sctp_set_t *sp; 1558 1559 /* Chunk hdr */ 1560 sch = (sctp_chunk_hdr_t *)dst; 1561 sch->sch_id = CHUNK_SACK; 1562 sch->sch_flags = 0; 1563 sch->sch_len = htons(sacklen); 1564 1565 /* SACK chunk */ 1566 sctp->sctp_lastacked = sctp->sctp_ftsn - 1; 1567 1568 sc = (sctp_sack_chunk_t *)(sch + 1); 1569 sc->ssc_cumtsn = htonl(sctp->sctp_lastacked); 1570 if (sctp->sctp_rxqueued < sctp->sctp_rwnd) { 1571 sc->ssc_a_rwnd = htonl(sctp->sctp_rwnd - sctp->sctp_rxqueued); 1572 } else { 1573 sc->ssc_a_rwnd = 0; 1574 } 1575 sc->ssc_numfrags = htons(num_gaps); 1576 sc->ssc_numdups = 0; 1577 1578 /* lay in gap reports */ 1579 sf = (sctp_sack_frag_t *)(sc + 1); 1580 for (sp = sctp->sctp_sack_info; sp; sp = sp->next) { 1581 uint16_t offset; 1582 1583 /* start */ 1584 if (sp->begin > sctp->sctp_lastacked) { 1585 offset = (uint16_t)(sp->begin - sctp->sctp_lastacked); 1586 } else { 1587 /* sequence number wrap */ 1588 offset = (uint16_t)(UINT32_MAX - sctp->sctp_lastacked + 1589 sp->begin); 1590 } 1591 sf->ssf_start = htons(offset); 1592 1593 /* end */ 1594 if (sp->end >= sp->begin) { 1595 offset += (uint16_t)(sp->end - sp->begin); 1596 } else { 1597 /* sequence number wrap */ 1598 offset += (uint16_t)(UINT32_MAX - sp->begin + sp->end); 1599 } 1600 sf->ssf_end = htons(offset); 1601 1602 sf++; 1603 /* This is just for debugging (a la the following assertion) */ 1604 num_gaps--; 1605 } 1606 1607 ASSERT(num_gaps == 0); 1608 1609 /* If the SACK timer is running, stop it */ 1610 if (sctp->sctp_ack_timer_running) { 1611 sctp_timer_stop(sctp->sctp_ack_mp); 1612 sctp->sctp_ack_timer_running = B_FALSE; 1613 } 1614 1615 BUMP_LOCAL(sctp->sctp_obchunks); 1616 } 1617 1618 mblk_t * 1619 sctp_make_sack(sctp_t *sctp, sctp_faddr_t *sendto, mblk_t *dups) 1620 { 1621 mblk_t *smp; 1622 size_t slen; 1623 sctp_chunk_hdr_t *sch; 1624 sctp_sack_chunk_t *sc; 1625 int32_t acks_max; 1626 sctp_stack_t *sctps = sctp->sctp_sctps; 1627 1628 if (sctp->sctp_force_sack) { 1629 sctp->sctp_force_sack = 0; 1630 goto checks_done; 1631 } 1632 1633 acks_max = sctps->sctps_deferred_acks_max; 1634 if (sctp->sctp_state == SCTPS_ESTABLISHED) { 1635 if (sctp->sctp_sack_toggle < acks_max) { 1636 /* no need to SACK right now */ 1637 dprint(2, ("sctp_make_sack: %p no sack (toggle)\n", 1638 (void *)sctp)); 1639 return (NULL); 1640 } else if (sctp->sctp_sack_toggle >= acks_max) { 1641 sctp->sctp_sack_toggle = 0; 1642 } 1643 } 1644 1645 if (sctp->sctp_ftsn == sctp->sctp_lastacked + 1) { 1646 dprint(2, ("sctp_make_sack: %p no sack (already)\n", 1647 (void *)sctp)); 1648 return (NULL); 1649 } 1650 1651 checks_done: 1652 dprint(2, ("sctp_make_sack: acking %x\n", sctp->sctp_ftsn - 1)); 1653 1654 slen = sizeof (*sch) + sizeof (*sc) + 1655 (sizeof (sctp_sack_frag_t) * sctp->sctp_sack_gaps); 1656 smp = sctp_make_mp(sctp, sendto, slen); 1657 if (smp == NULL) { 1658 SCTP_KSTAT(sctps, sctp_send_sack_failed); 1659 return (NULL); 1660 } 1661 sch = (sctp_chunk_hdr_t *)smp->b_wptr; 1662 1663 sctp_fill_sack(sctp, smp->b_wptr, slen); 1664 smp->b_wptr += slen; 1665 if (dups) { 1666 sc = (sctp_sack_chunk_t *)(sch + 1); 1667 sc->ssc_numdups = htons((dups->b_wptr - dups->b_rptr) 1668 / sizeof (uint32_t)); 1669 sch->sch_len = htons(slen + (dups->b_wptr - dups->b_rptr)); 1670 smp->b_cont = dups; 1671 } 1672 1673 return (smp); 1674 } 1675 1676 void 1677 sctp_sack(sctp_t *sctp, mblk_t *dups) 1678 { 1679 mblk_t *smp; 1680 sctp_stack_t *sctps = sctp->sctp_sctps; 1681 1682 /* If we are shutting down, let send_shutdown() bundle the SACK */ 1683 if (sctp->sctp_state == SCTPS_SHUTDOWN_SENT) { 1684 sctp_send_shutdown(sctp, 0); 1685 } 1686 1687 ASSERT(sctp->sctp_lastdata != NULL); 1688 1689 if ((smp = sctp_make_sack(sctp, sctp->sctp_lastdata, dups)) == NULL) { 1690 /* The caller of sctp_sack() will not free the dups mblk. */ 1691 if (dups != NULL) 1692 freeb(dups); 1693 return; 1694 } 1695 1696 sctp_set_iplen(sctp, smp); 1697 1698 dprint(2, ("sctp_sack: sending to %p %x:%x:%x:%x\n", 1699 (void *)sctp->sctp_lastdata, 1700 SCTP_PRINTADDR(sctp->sctp_lastdata->faddr))); 1701 1702 sctp->sctp_active = lbolt64; 1703 1704 BUMP_MIB(&sctps->sctps_mib, sctpOutAck); 1705 sctp_add_sendq(sctp, smp); 1706 } 1707 1708 /* 1709 * This is called if we have a message that was partially sent and is 1710 * abandoned. The cum TSN will be the last chunk sent for this message, 1711 * subsequent chunks will be marked ABANDONED. We send a Forward TSN 1712 * chunk in this case with the TSN of the last sent chunk so that the 1713 * peer can clean up its fragment list for this message. This message 1714 * will be removed from the transmit list when the peer sends a SACK 1715 * back. 1716 */ 1717 int 1718 sctp_check_abandoned_msg(sctp_t *sctp, mblk_t *meta) 1719 { 1720 sctp_data_hdr_t *dh; 1721 mblk_t *nmp; 1722 mblk_t *head; 1723 int32_t unsent = 0; 1724 mblk_t *mp1 = meta->b_cont; 1725 uint32_t adv_pap = sctp->sctp_adv_pap; 1726 sctp_faddr_t *fp = sctp->sctp_current; 1727 sctp_stack_t *sctps = sctp->sctp_sctps; 1728 1729 dh = (sctp_data_hdr_t *)mp1->b_rptr; 1730 if (SEQ_GEQ(sctp->sctp_lastack_rxd, ntohl(dh->sdh_tsn))) { 1731 sctp_ftsn_set_t *sets = NULL; 1732 uint_t nsets = 0; 1733 uint32_t seglen = sizeof (uint32_t); 1734 boolean_t ubit = SCTP_DATA_GET_UBIT(dh); 1735 1736 while (mp1->b_next != NULL && SCTP_CHUNK_ISSENT(mp1->b_next)) 1737 mp1 = mp1->b_next; 1738 dh = (sctp_data_hdr_t *)mp1->b_rptr; 1739 sctp->sctp_adv_pap = ntohl(dh->sdh_tsn); 1740 if (!ubit && 1741 !sctp_add_ftsn_set(&sets, fp, meta, &nsets, &seglen)) { 1742 sctp->sctp_adv_pap = adv_pap; 1743 return (ENOMEM); 1744 } 1745 nmp = sctp_make_ftsn_chunk(sctp, fp, sets, nsets, seglen); 1746 sctp_free_ftsn_set(sets); 1747 if (nmp == NULL) { 1748 sctp->sctp_adv_pap = adv_pap; 1749 return (ENOMEM); 1750 } 1751 head = sctp_add_proto_hdr(sctp, fp, nmp, 0, NULL); 1752 if (head == NULL) { 1753 sctp->sctp_adv_pap = adv_pap; 1754 freemsg(nmp); 1755 SCTP_KSTAT(sctps, sctp_send_ftsn_failed); 1756 return (ENOMEM); 1757 } 1758 SCTP_MSG_SET_ABANDONED(meta); 1759 sctp_set_iplen(sctp, head); 1760 sctp_add_sendq(sctp, head); 1761 if (!fp->timer_running) 1762 SCTP_FADDR_TIMER_RESTART(sctp, fp, fp->rto); 1763 mp1 = mp1->b_next; 1764 while (mp1 != NULL) { 1765 ASSERT(!SCTP_CHUNK_ISSENT(mp1)); 1766 ASSERT(!SCTP_CHUNK_ABANDONED(mp1)); 1767 SCTP_ABANDON_CHUNK(mp1); 1768 dh = (sctp_data_hdr_t *)mp1->b_rptr; 1769 unsent += ntohs(dh->sdh_len) - sizeof (*dh); 1770 mp1 = mp1->b_next; 1771 } 1772 ASSERT(sctp->sctp_unsent >= unsent); 1773 sctp->sctp_unsent -= unsent; 1774 /* 1775 * Update ULP the amount of queued data, which is 1776 * sent-unack'ed + unsent. 1777 */ 1778 if (!SCTP_IS_DETACHED(sctp)) { 1779 sctp->sctp_ulp_xmitted(sctp->sctp_ulpd, 1780 sctp->sctp_unacked + sctp->sctp_unsent); 1781 } 1782 return (0); 1783 } 1784 return (-1); 1785 } 1786 1787 uint32_t 1788 sctp_cumack(sctp_t *sctp, uint32_t tsn, mblk_t **first_unacked) 1789 { 1790 mblk_t *ump, *nump, *mp = NULL; 1791 uint16_t chunklen; 1792 uint32_t xtsn; 1793 sctp_faddr_t *fp; 1794 sctp_data_hdr_t *sdc; 1795 uint32_t cumack_forward = 0; 1796 sctp_msg_hdr_t *mhdr; 1797 sctp_stack_t *sctps = sctp->sctp_sctps; 1798 1799 ump = sctp->sctp_xmit_head; 1800 1801 /* 1802 * Free messages only when they're completely acked. 1803 */ 1804 while (ump != NULL) { 1805 mhdr = (sctp_msg_hdr_t *)ump->b_rptr; 1806 for (mp = ump->b_cont; mp != NULL; mp = mp->b_next) { 1807 if (SCTP_CHUNK_ABANDONED(mp)) { 1808 ASSERT(SCTP_IS_MSG_ABANDONED(ump)); 1809 mp = NULL; 1810 break; 1811 } 1812 /* 1813 * We check for abandoned message if we are PR-SCTP 1814 * aware, if this is not the first chunk in the 1815 * message (b_cont) and if the message is marked 1816 * abandoned. 1817 */ 1818 if (!SCTP_CHUNK_ISSENT(mp)) { 1819 if (sctp->sctp_prsctp_aware && 1820 mp != ump->b_cont && 1821 (SCTP_IS_MSG_ABANDONED(ump) || 1822 SCTP_MSG_TO_BE_ABANDONED(ump, mhdr, 1823 sctp))) { 1824 (void) sctp_check_abandoned_msg(sctp, 1825 ump); 1826 } 1827 goto cum_ack_done; 1828 } 1829 sdc = (sctp_data_hdr_t *)mp->b_rptr; 1830 xtsn = ntohl(sdc->sdh_tsn); 1831 if (SEQ_GEQ(sctp->sctp_lastack_rxd, xtsn)) 1832 continue; 1833 if (SEQ_GEQ(tsn, xtsn)) { 1834 fp = SCTP_CHUNK_DEST(mp); 1835 chunklen = ntohs(sdc->sdh_len); 1836 1837 if (sctp->sctp_out_time != 0 && 1838 xtsn == sctp->sctp_rtt_tsn) { 1839 /* Got a new RTT measurement */ 1840 sctp_update_rtt(sctp, fp, 1841 lbolt64 - sctp->sctp_out_time); 1842 sctp->sctp_out_time = 0; 1843 } 1844 if (SCTP_CHUNK_ISACKED(mp)) 1845 continue; 1846 SCTP_CHUNK_SET_SACKCNT(mp, 0); 1847 SCTP_CHUNK_ACKED(mp); 1848 ASSERT(fp->suna >= chunklen); 1849 fp->suna -= chunklen; 1850 fp->acked += chunklen; 1851 cumack_forward += chunklen; 1852 ASSERT(sctp->sctp_unacked >= 1853 (chunklen - sizeof (*sdc))); 1854 sctp->sctp_unacked -= 1855 (chunklen - sizeof (*sdc)); 1856 if (fp->suna == 0) { 1857 /* all outstanding data acked */ 1858 fp->pba = 0; 1859 SCTP_FADDR_TIMER_STOP(fp); 1860 } else { 1861 SCTP_FADDR_TIMER_RESTART(sctp, fp, 1862 fp->rto); 1863 } 1864 } else { 1865 goto cum_ack_done; 1866 } 1867 } 1868 nump = ump->b_next; 1869 if (nump != NULL) 1870 nump->b_prev = NULL; 1871 if (ump == sctp->sctp_xmit_tail) 1872 sctp->sctp_xmit_tail = nump; 1873 if (SCTP_IS_MSG_ABANDONED(ump)) { 1874 BUMP_LOCAL(sctp->sctp_prsctpdrop); 1875 ump->b_next = NULL; 1876 sctp_sendfail_event(sctp, ump, 0, B_TRUE); 1877 } else { 1878 sctp_free_msg(ump); 1879 } 1880 sctp->sctp_xmit_head = ump = nump; 1881 } 1882 cum_ack_done: 1883 *first_unacked = mp; 1884 if (cumack_forward > 0) { 1885 BUMP_MIB(&sctps->sctps_mib, sctpInAck); 1886 if (SEQ_GT(sctp->sctp_lastack_rxd, sctp->sctp_recovery_tsn)) { 1887 sctp->sctp_recovery_tsn = sctp->sctp_lastack_rxd; 1888 } 1889 1890 /* 1891 * Update ULP the amount of queued data, which is 1892 * sent-unack'ed + unsent. 1893 */ 1894 if (!SCTP_IS_DETACHED(sctp)) { 1895 sctp->sctp_ulp_xmitted(sctp->sctp_ulpd, 1896 sctp->sctp_unacked + sctp->sctp_unsent); 1897 } 1898 1899 /* Time to send a shutdown? */ 1900 if (sctp->sctp_state == SCTPS_SHUTDOWN_PENDING) { 1901 sctp_send_shutdown(sctp, 0); 1902 } 1903 sctp->sctp_xmit_unacked = mp; 1904 } else { 1905 /* dup ack */ 1906 BUMP_MIB(&sctps->sctps_mib, sctpInDupAck); 1907 } 1908 sctp->sctp_lastack_rxd = tsn; 1909 if (SEQ_LT(sctp->sctp_adv_pap, sctp->sctp_lastack_rxd)) 1910 sctp->sctp_adv_pap = sctp->sctp_lastack_rxd; 1911 ASSERT(sctp->sctp_xmit_head || sctp->sctp_unacked == 0); 1912 1913 return (cumack_forward); 1914 } 1915 1916 static int 1917 sctp_set_frwnd(sctp_t *sctp, uint32_t frwnd) 1918 { 1919 uint32_t orwnd; 1920 1921 if (sctp->sctp_unacked > frwnd) { 1922 sctp->sctp_frwnd = 0; 1923 return (0); 1924 } 1925 orwnd = sctp->sctp_frwnd; 1926 sctp->sctp_frwnd = frwnd - sctp->sctp_unacked; 1927 if (orwnd < sctp->sctp_frwnd) { 1928 return (1); 1929 } else { 1930 return (0); 1931 } 1932 } 1933 1934 /* 1935 * For un-ordered messages. 1936 * Walk the sctp->sctp_uo_frag list and remove any fragments with TSN 1937 * less than/equal to ftsn. Fragments for un-ordered messages are 1938 * strictly in sequence (w.r.t TSN). 1939 */ 1940 static int 1941 sctp_ftsn_check_uo_frag(sctp_t *sctp, uint32_t ftsn) 1942 { 1943 mblk_t *hmp; 1944 mblk_t *hmp_next; 1945 sctp_data_hdr_t *dc; 1946 int dlen = 0; 1947 1948 hmp = sctp->sctp_uo_frags; 1949 while (hmp != NULL) { 1950 hmp_next = hmp->b_next; 1951 dc = (sctp_data_hdr_t *)hmp->b_rptr; 1952 if (SEQ_GT(ntohl(dc->sdh_tsn), ftsn)) 1953 return (dlen); 1954 sctp->sctp_uo_frags = hmp_next; 1955 if (hmp_next != NULL) 1956 hmp_next->b_prev = NULL; 1957 hmp->b_next = NULL; 1958 dlen += ntohs(dc->sdh_len) - sizeof (*dc); 1959 freeb(hmp); 1960 hmp = hmp_next; 1961 } 1962 return (dlen); 1963 } 1964 1965 /* 1966 * For ordered messages. 1967 * Check for existing fragments for an sid-ssn pair reported as abandoned, 1968 * hence will not receive, in the Forward TSN. If there are fragments, then 1969 * we just nuke them. If and when Partial Delivery API is supported, we 1970 * would need to send a notification to the upper layer about this. 1971 */ 1972 static int 1973 sctp_ftsn_check_frag(sctp_t *sctp, uint16_t ssn, sctp_instr_t *sip) 1974 { 1975 sctp_reass_t *srp; 1976 mblk_t *hmp; 1977 mblk_t *dmp; 1978 mblk_t *hmp_next; 1979 sctp_data_hdr_t *dc; 1980 int dlen = 0; 1981 1982 hmp = sip->istr_reass; 1983 while (hmp != NULL) { 1984 hmp_next = hmp->b_next; 1985 srp = (sctp_reass_t *)DB_BASE(hmp); 1986 if (SSN_GT(srp->ssn, ssn)) 1987 return (dlen); 1988 /* 1989 * If we had sent part of this message up, send a partial 1990 * delivery event. Since this is ordered delivery, we should 1991 * have sent partial message only for the next in sequence, 1992 * hence the ASSERT. See comments in sctp_data_chunk() for 1993 * trypartial. 1994 */ 1995 if (srp->partial_delivered) { 1996 ASSERT(sip->nextseq == srp->ssn); 1997 sctp_partial_delivery_event(sctp); 1998 } 1999 /* Take it out of the reass queue */ 2000 sip->istr_reass = hmp_next; 2001 if (hmp_next != NULL) 2002 hmp_next->b_prev = NULL; 2003 hmp->b_next = NULL; 2004 ASSERT(hmp->b_prev == NULL); 2005 dmp = hmp; 2006 if (DB_TYPE(hmp) == M_CTL) { 2007 dmp = hmp->b_cont; 2008 hmp->b_cont = NULL; 2009 freeb(hmp); 2010 hmp = dmp; 2011 } 2012 while (dmp != NULL) { 2013 dc = (sctp_data_hdr_t *)dmp->b_rptr; 2014 dlen += ntohs(dc->sdh_len) - sizeof (*dc); 2015 dmp = dmp->b_cont; 2016 } 2017 freemsg(hmp); 2018 hmp = hmp_next; 2019 } 2020 return (dlen); 2021 } 2022 2023 /* 2024 * Update sctp_ftsn to the cumulative TSN from the Forward TSN chunk. Remove 2025 * any SACK gaps less than the newly updated sctp_ftsn. Walk through the 2026 * sid-ssn pair in the Forward TSN and for each, clean the fragment list 2027 * for this pair, if needed, and check if we can deliver subsequent 2028 * messages, if any, from the instream queue (that were waiting for this 2029 * sid-ssn message to show up). Once we are done try to update the SACK 2030 * info. We could get a duplicate Forward TSN, in which case just send 2031 * a SACK. If any of the sid values in the the Forward TSN is invalid, 2032 * send back an "Invalid Stream Identifier" error and continue processing 2033 * the rest. 2034 */ 2035 static void 2036 sctp_process_forward_tsn(sctp_t *sctp, sctp_chunk_hdr_t *ch, sctp_faddr_t *fp, 2037 ip6_pkt_t *ipp) 2038 { 2039 uint32_t *ftsn = (uint32_t *)(ch + 1); 2040 ftsn_entry_t *ftsn_entry; 2041 sctp_instr_t *instr; 2042 boolean_t can_deliver = B_TRUE; 2043 size_t dlen; 2044 int flen; 2045 mblk_t *dmp; 2046 mblk_t *pmp; 2047 sctp_data_hdr_t *dc; 2048 ssize_t remaining; 2049 sctp_stack_t *sctps = sctp->sctp_sctps; 2050 2051 *ftsn = ntohl(*ftsn); 2052 remaining = ntohs(ch->sch_len) - sizeof (*ch) - sizeof (*ftsn); 2053 2054 if (SCTP_IS_DETACHED(sctp)) { 2055 BUMP_MIB(&sctps->sctps_mib, sctpInClosed); 2056 can_deliver = B_FALSE; 2057 } 2058 /* 2059 * un-ordered messages don't have SID-SSN pair entries, we check 2060 * for any fragments (for un-ordered message) to be discarded using 2061 * the cumulative FTSN. 2062 */ 2063 flen = sctp_ftsn_check_uo_frag(sctp, *ftsn); 2064 if (flen > 0) { 2065 ASSERT(sctp->sctp_rxqueued >= flen); 2066 sctp->sctp_rxqueued -= flen; 2067 } 2068 ftsn_entry = (ftsn_entry_t *)(ftsn + 1); 2069 while (remaining >= sizeof (*ftsn_entry)) { 2070 ftsn_entry->ftsn_sid = ntohs(ftsn_entry->ftsn_sid); 2071 ftsn_entry->ftsn_ssn = ntohs(ftsn_entry->ftsn_ssn); 2072 if (ftsn_entry->ftsn_sid >= sctp->sctp_num_istr) { 2073 uint16_t inval_parm[2]; 2074 mblk_t *errmp; 2075 2076 inval_parm[0] = htons(ftsn_entry->ftsn_sid); 2077 /* RESERVED to be ignored at the receiving end */ 2078 inval_parm[1] = 0; 2079 errmp = sctp_make_err(sctp, SCTP_ERR_BAD_SID, 2080 (char *)inval_parm, sizeof (inval_parm)); 2081 if (errmp != NULL) 2082 sctp_send_err(sctp, errmp, NULL); 2083 ftsn_entry++; 2084 remaining -= sizeof (*ftsn_entry); 2085 continue; 2086 } 2087 instr = &sctp->sctp_instr[ftsn_entry->ftsn_sid]; 2088 flen = sctp_ftsn_check_frag(sctp, ftsn_entry->ftsn_ssn, instr); 2089 /* Indicates frags were nuked, update rxqueued */ 2090 if (flen > 0) { 2091 ASSERT(sctp->sctp_rxqueued >= flen); 2092 sctp->sctp_rxqueued -= flen; 2093 } 2094 /* 2095 * It is possible to receive an FTSN chunk with SSN smaller 2096 * than then nextseq if this chunk is a retransmission because 2097 * of incomplete processing when it was first processed. 2098 */ 2099 if (SSN_GE(ftsn_entry->ftsn_ssn, instr->nextseq)) 2100 instr->nextseq = ftsn_entry->ftsn_ssn + 1; 2101 while (instr->istr_nmsgs > 0) { 2102 mblk_t *next; 2103 2104 dmp = (mblk_t *)instr->istr_msgs; 2105 dc = (sctp_data_hdr_t *)dmp->b_rptr; 2106 if (ntohs(dc->sdh_ssn) != instr->nextseq) 2107 break; 2108 2109 next = dmp->b_next; 2110 dlen = dmp->b_wptr - dmp->b_rptr - sizeof (*dc); 2111 for (pmp = dmp->b_cont; pmp != NULL; 2112 pmp = pmp->b_cont) { 2113 dlen += pmp->b_wptr - pmp->b_rptr; 2114 } 2115 if (can_deliver) { 2116 int32_t nrwnd; 2117 2118 dmp->b_rptr = (uchar_t *)(dc + 1); 2119 dmp->b_next = NULL; 2120 ASSERT(dmp->b_prev == NULL); 2121 if (sctp_input_add_ancillary(sctp, 2122 &dmp, dc, fp, ipp) == 0) { 2123 sctp->sctp_rxqueued -= dlen; 2124 sctp->sctp_rwnd -= dlen; 2125 nrwnd = sctp->sctp_ulp_recv( 2126 sctp->sctp_ulpd, dmp, 0); 2127 if (nrwnd > sctp->sctp_rwnd) 2128 sctp->sctp_rwnd = nrwnd; 2129 } else { 2130 /* 2131 * We will resume processing when 2132 * the FTSN chunk is re-xmitted. 2133 */ 2134 dmp->b_rptr = (uchar_t *)dc; 2135 dmp->b_next = next; 2136 dprint(0, 2137 ("FTSN dequeuing %u failed\n", 2138 ntohs(dc->sdh_ssn))); 2139 return; 2140 } 2141 } else { 2142 sctp->sctp_rxqueued -= dlen; 2143 ASSERT(dmp->b_prev == NULL); 2144 dmp->b_next = NULL; 2145 freemsg(dmp); 2146 } 2147 instr->istr_nmsgs--; 2148 instr->nextseq++; 2149 sctp->sctp_istr_nmsgs--; 2150 if (next != NULL) 2151 next->b_prev = NULL; 2152 instr->istr_msgs = next; 2153 } 2154 ftsn_entry++; 2155 remaining -= sizeof (*ftsn_entry); 2156 } 2157 /* Duplicate FTSN */ 2158 if (*ftsn <= (sctp->sctp_ftsn - 1)) { 2159 sctp->sctp_force_sack = 1; 2160 return; 2161 } 2162 /* Advance cum TSN to that reported in the Forward TSN chunk */ 2163 sctp->sctp_ftsn = *ftsn + 1; 2164 2165 /* Remove all the SACK gaps before the new cum TSN */ 2166 if (sctp->sctp_sack_info != NULL) { 2167 sctp_ack_rem(&sctp->sctp_sack_info, sctp->sctp_ftsn - 1, 2168 &sctp->sctp_sack_gaps); 2169 } 2170 /* 2171 * If there are gap reports pending, check if advancing 2172 * the ftsn here closes a gap. If so, we can advance 2173 * ftsn to the end of the set. 2174 * If ftsn has moved forward, maybe we can remove gap reports. 2175 */ 2176 if (sctp->sctp_sack_info != NULL && 2177 sctp->sctp_ftsn == sctp->sctp_sack_info->begin) { 2178 sctp->sctp_ftsn = sctp->sctp_sack_info->end + 1; 2179 sctp_ack_rem(&sctp->sctp_sack_info, sctp->sctp_ftsn - 1, 2180 &sctp->sctp_sack_gaps); 2181 } 2182 } 2183 2184 /* 2185 * When we have processed a SACK we check to see if we can advance the 2186 * cumulative TSN if there are abandoned chunks immediately following 2187 * the updated cumulative TSN. If there are, we attempt to send a 2188 * Forward TSN chunk. 2189 */ 2190 static void 2191 sctp_check_abandoned_data(sctp_t *sctp, sctp_faddr_t *fp) 2192 { 2193 mblk_t *meta = sctp->sctp_xmit_head; 2194 mblk_t *mp; 2195 mblk_t *nmp; 2196 uint32_t seglen; 2197 uint32_t adv_pap = sctp->sctp_adv_pap; 2198 2199 /* 2200 * We only check in the first meta since otherwise we can't 2201 * advance the cumulative ack point. We just look for chunks 2202 * marked for retransmission, else we might prematurely 2203 * send an FTSN for a sent, but unacked, chunk. 2204 */ 2205 for (mp = meta->b_cont; mp != NULL; mp = mp->b_next) { 2206 if (!SCTP_CHUNK_ISSENT(mp)) 2207 return; 2208 if (SCTP_CHUNK_WANT_REXMIT(mp)) 2209 break; 2210 } 2211 if (mp == NULL) 2212 return; 2213 sctp_check_adv_ack_pt(sctp, meta, mp); 2214 if (SEQ_GT(sctp->sctp_adv_pap, adv_pap)) { 2215 sctp_make_ftsns(sctp, meta, mp, &nmp, fp, &seglen); 2216 if (nmp == NULL) { 2217 sctp->sctp_adv_pap = adv_pap; 2218 if (!fp->timer_running) 2219 SCTP_FADDR_TIMER_RESTART(sctp, fp, fp->rto); 2220 return; 2221 } 2222 sctp_set_iplen(sctp, nmp); 2223 sctp_add_sendq(sctp, nmp); 2224 if (!fp->timer_running) 2225 SCTP_FADDR_TIMER_RESTART(sctp, fp, fp->rto); 2226 } 2227 } 2228 2229 /* 2230 * The processing here follows the same logic in sctp_got_sack(), the reason 2231 * we do this separately is because, usually, gap blocks are ordered and 2232 * we can process it in sctp_got_sack(). However if they aren't we would 2233 * need to do some additional non-optimal stuff when we start processing the 2234 * unordered gaps. To that effect sctp_got_sack() does the processing in the 2235 * simple case and this does the same in the more involved case. 2236 */ 2237 static uint32_t 2238 sctp_process_uo_gaps(sctp_t *sctp, uint32_t ctsn, sctp_sack_frag_t *ssf, 2239 int num_gaps, mblk_t *umphead, mblk_t *mphead, int *trysend, 2240 boolean_t *fast_recovery, uint32_t fr_xtsn) 2241 { 2242 uint32_t xtsn; 2243 uint32_t gapstart = 0; 2244 uint32_t gapend = 0; 2245 int gapcnt; 2246 uint16_t chunklen; 2247 sctp_data_hdr_t *sdc; 2248 int gstart; 2249 mblk_t *ump = umphead; 2250 mblk_t *mp = mphead; 2251 sctp_faddr_t *fp; 2252 uint32_t acked = 0; 2253 sctp_stack_t *sctps = sctp->sctp_sctps; 2254 2255 /* 2256 * gstart tracks the last (in the order of TSN) gapstart that 2257 * we process in this SACK gaps walk. 2258 */ 2259 gstart = ctsn; 2260 2261 sdc = (sctp_data_hdr_t *)mp->b_rptr; 2262 xtsn = ntohl(sdc->sdh_tsn); 2263 for (gapcnt = 0; gapcnt < num_gaps; gapcnt++, ssf++) { 2264 if (gapstart != 0) { 2265 /* 2266 * If we have reached the end of the transmit list or 2267 * hit an unsent chunk or encountered an unordered gap 2268 * block start from the ctsn again. 2269 */ 2270 if (ump == NULL || !SCTP_CHUNK_ISSENT(mp) || 2271 SEQ_LT(ctsn + ntohs(ssf->ssf_start), xtsn)) { 2272 ump = umphead; 2273 mp = mphead; 2274 sdc = (sctp_data_hdr_t *)mp->b_rptr; 2275 xtsn = ntohl(sdc->sdh_tsn); 2276 } 2277 } 2278 2279 gapstart = ctsn + ntohs(ssf->ssf_start); 2280 gapend = ctsn + ntohs(ssf->ssf_end); 2281 2282 /* SACK for TSN we have not sent - ABORT */ 2283 if (SEQ_GT(gapstart, sctp->sctp_ltsn - 1) || 2284 SEQ_GT(gapend, sctp->sctp_ltsn - 1)) { 2285 BUMP_MIB(&sctps->sctps_mib, sctpInAckUnsent); 2286 *trysend = -1; 2287 return (acked); 2288 } else if (SEQ_LT(gapend, gapstart)) { 2289 break; 2290 } 2291 /* 2292 * The xtsn can be the TSN processed for the last gap 2293 * (gapend) or it could be the cumulative TSN. We continue 2294 * with the last xtsn as long as the gaps are ordered, when 2295 * we hit an unordered gap, we re-start from the cumulative 2296 * TSN. For the first gap it is always the cumulative TSN. 2297 */ 2298 while (xtsn != gapstart) { 2299 /* 2300 * We can't reliably check for reneged chunks 2301 * when walking the unordered list, so we don't. 2302 * In case the peer reneges then we will end up 2303 * sending the reneged chunk via timeout. 2304 */ 2305 mp = mp->b_next; 2306 if (mp == NULL) { 2307 ump = ump->b_next; 2308 /* 2309 * ump can't be NULL because of the sanity 2310 * check above. 2311 */ 2312 ASSERT(ump != NULL); 2313 mp = ump->b_cont; 2314 } 2315 /* 2316 * mp can't be unsent because of the sanity check 2317 * above. 2318 */ 2319 ASSERT(SCTP_CHUNK_ISSENT(mp)); 2320 sdc = (sctp_data_hdr_t *)mp->b_rptr; 2321 xtsn = ntohl(sdc->sdh_tsn); 2322 } 2323 /* 2324 * Now that we have found the chunk with TSN == 'gapstart', 2325 * let's walk till we hit the chunk with TSN == 'gapend'. 2326 * All intermediate chunks will be marked ACKED, if they 2327 * haven't already been. 2328 */ 2329 while (SEQ_LEQ(xtsn, gapend)) { 2330 /* 2331 * SACKed 2332 */ 2333 SCTP_CHUNK_SET_SACKCNT(mp, 0); 2334 if (!SCTP_CHUNK_ISACKED(mp)) { 2335 SCTP_CHUNK_ACKED(mp); 2336 2337 fp = SCTP_CHUNK_DEST(mp); 2338 chunklen = ntohs(sdc->sdh_len); 2339 ASSERT(fp->suna >= chunklen); 2340 fp->suna -= chunklen; 2341 if (fp->suna == 0) { 2342 /* All outstanding data acked. */ 2343 fp->pba = 0; 2344 SCTP_FADDR_TIMER_STOP(fp); 2345 } 2346 fp->acked += chunklen; 2347 acked += chunklen; 2348 sctp->sctp_unacked -= chunklen - sizeof (*sdc); 2349 ASSERT(sctp->sctp_unacked >= 0); 2350 } 2351 /* 2352 * Move to the next message in the transmit list 2353 * if we are done with all the chunks from the current 2354 * message. Note, it is possible to hit the end of the 2355 * transmit list here, i.e. if we have already completed 2356 * processing the gap block. 2357 */ 2358 mp = mp->b_next; 2359 if (mp == NULL) { 2360 ump = ump->b_next; 2361 if (ump == NULL) { 2362 ASSERT(xtsn == gapend); 2363 break; 2364 } 2365 mp = ump->b_cont; 2366 } 2367 /* 2368 * Likewise, we can hit an unsent chunk once we have 2369 * completed processing the gap block. 2370 */ 2371 if (!SCTP_CHUNK_ISSENT(mp)) { 2372 ASSERT(xtsn == gapend); 2373 break; 2374 } 2375 sdc = (sctp_data_hdr_t *)mp->b_rptr; 2376 xtsn = ntohl(sdc->sdh_tsn); 2377 } 2378 /* 2379 * We keep track of the last gap we successfully processed 2380 * so that we can terminate the walk below for incrementing 2381 * the SACK count. 2382 */ 2383 if (SEQ_LT(gstart, gapstart)) 2384 gstart = gapstart; 2385 } 2386 /* 2387 * Check if have incremented the SACK count for all unacked TSNs in 2388 * sctp_got_sack(), if so we are done. 2389 */ 2390 if (SEQ_LEQ(gstart, fr_xtsn)) 2391 return (acked); 2392 2393 ump = umphead; 2394 mp = mphead; 2395 sdc = (sctp_data_hdr_t *)mp->b_rptr; 2396 xtsn = ntohl(sdc->sdh_tsn); 2397 while (SEQ_LT(xtsn, gstart)) { 2398 /* 2399 * We have incremented SACK count for TSNs less than fr_tsn 2400 * in sctp_got_sack(), so don't increment them again here. 2401 */ 2402 if (SEQ_GT(xtsn, fr_xtsn) && !SCTP_CHUNK_ISACKED(mp)) { 2403 SCTP_CHUNK_SET_SACKCNT(mp, SCTP_CHUNK_SACKCNT(mp) + 1); 2404 if (SCTP_CHUNK_SACKCNT(mp) == 2405 sctps->sctps_fast_rxt_thresh) { 2406 SCTP_CHUNK_REXMIT(mp); 2407 sctp->sctp_chk_fast_rexmit = B_TRUE; 2408 *trysend = 1; 2409 if (!*fast_recovery) { 2410 /* 2411 * Entering fast recovery. 2412 */ 2413 fp = SCTP_CHUNK_DEST(mp); 2414 fp->ssthresh = fp->cwnd / 2; 2415 if (fp->ssthresh < 2 * fp->sfa_pmss) { 2416 fp->ssthresh = 2417 2 * fp->sfa_pmss; 2418 } 2419 fp->cwnd = fp->ssthresh; 2420 fp->pba = 0; 2421 sctp->sctp_recovery_tsn = 2422 sctp->sctp_ltsn - 1; 2423 *fast_recovery = B_TRUE; 2424 } 2425 } 2426 } 2427 mp = mp->b_next; 2428 if (mp == NULL) { 2429 ump = ump->b_next; 2430 /* We can't get to the end of the transmit list here */ 2431 ASSERT(ump != NULL); 2432 mp = ump->b_cont; 2433 } 2434 /* We can't hit an unsent chunk here */ 2435 ASSERT(SCTP_CHUNK_ISSENT(mp)); 2436 sdc = (sctp_data_hdr_t *)mp->b_rptr; 2437 xtsn = ntohl(sdc->sdh_tsn); 2438 } 2439 return (acked); 2440 } 2441 2442 static int 2443 sctp_got_sack(sctp_t *sctp, sctp_chunk_hdr_t *sch) 2444 { 2445 sctp_sack_chunk_t *sc; 2446 sctp_data_hdr_t *sdc; 2447 sctp_sack_frag_t *ssf; 2448 mblk_t *ump; 2449 mblk_t *mp; 2450 mblk_t *mp1; 2451 uint32_t cumtsn; 2452 uint32_t xtsn; 2453 uint32_t gapstart = 0; 2454 uint32_t gapend = 0; 2455 uint32_t acked = 0; 2456 uint16_t chunklen; 2457 sctp_faddr_t *fp; 2458 int num_gaps; 2459 int trysend = 0; 2460 int i; 2461 boolean_t fast_recovery = B_FALSE; 2462 boolean_t cumack_forward = B_FALSE; 2463 boolean_t fwd_tsn = B_FALSE; 2464 sctp_stack_t *sctps = sctp->sctp_sctps; 2465 2466 BUMP_LOCAL(sctp->sctp_ibchunks); 2467 chunklen = ntohs(sch->sch_len); 2468 if (chunklen < (sizeof (*sch) + sizeof (*sc))) 2469 return (0); 2470 2471 sc = (sctp_sack_chunk_t *)(sch + 1); 2472 cumtsn = ntohl(sc->ssc_cumtsn); 2473 2474 dprint(2, ("got sack cumtsn %x -> %x\n", sctp->sctp_lastack_rxd, 2475 cumtsn)); 2476 2477 /* out of order */ 2478 if (SEQ_LT(cumtsn, sctp->sctp_lastack_rxd)) 2479 return (0); 2480 2481 if (SEQ_GT(cumtsn, sctp->sctp_ltsn - 1)) { 2482 BUMP_MIB(&sctps->sctps_mib, sctpInAckUnsent); 2483 /* Send an ABORT */ 2484 return (-1); 2485 } 2486 2487 /* 2488 * Cwnd only done when not in fast recovery mode. 2489 */ 2490 if (SEQ_LT(sctp->sctp_lastack_rxd, sctp->sctp_recovery_tsn)) 2491 fast_recovery = B_TRUE; 2492 2493 /* 2494 * .. and if the cum TSN is not moving ahead on account Forward TSN 2495 */ 2496 if (SEQ_LT(sctp->sctp_lastack_rxd, sctp->sctp_adv_pap)) 2497 fwd_tsn = B_TRUE; 2498 2499 if (cumtsn == sctp->sctp_lastack_rxd && 2500 (sctp->sctp_xmit_unacked == NULL || 2501 !SCTP_CHUNK_ABANDONED(sctp->sctp_xmit_unacked))) { 2502 if (sctp->sctp_xmit_unacked != NULL) 2503 mp = sctp->sctp_xmit_unacked; 2504 else if (sctp->sctp_xmit_head != NULL) 2505 mp = sctp->sctp_xmit_head->b_cont; 2506 else 2507 mp = NULL; 2508 BUMP_MIB(&sctps->sctps_mib, sctpInDupAck); 2509 /* 2510 * If we were doing a zero win probe and the win 2511 * has now opened to at least MSS, re-transmit the 2512 * zero win probe via sctp_rexmit_packet(). 2513 */ 2514 if (mp != NULL && sctp->sctp_zero_win_probe && 2515 ntohl(sc->ssc_a_rwnd) >= sctp->sctp_current->sfa_pmss) { 2516 mblk_t *pkt; 2517 uint_t pkt_len; 2518 mblk_t *mp1 = mp; 2519 mblk_t *meta = sctp->sctp_xmit_head; 2520 2521 /* 2522 * Reset the RTO since we have been backing-off 2523 * to send the ZWP. 2524 */ 2525 fp = sctp->sctp_current; 2526 fp->rto = fp->srtt + 4 * fp->rttvar; 2527 /* Resend the ZWP */ 2528 pkt = sctp_rexmit_packet(sctp, &meta, &mp1, fp, 2529 &pkt_len); 2530 if (pkt == NULL) { 2531 SCTP_KSTAT(sctps, sctp_ss_rexmit_failed); 2532 return (0); 2533 } 2534 ASSERT(pkt_len <= fp->sfa_pmss); 2535 sctp->sctp_zero_win_probe = B_FALSE; 2536 sctp->sctp_rxt_nxttsn = sctp->sctp_ltsn; 2537 sctp->sctp_rxt_maxtsn = sctp->sctp_ltsn; 2538 sctp_set_iplen(sctp, pkt); 2539 sctp_add_sendq(sctp, pkt); 2540 } 2541 } else { 2542 if (sctp->sctp_zero_win_probe) { 2543 /* 2544 * Reset the RTO since we have been backing-off 2545 * to send the ZWP. 2546 */ 2547 fp = sctp->sctp_current; 2548 fp->rto = fp->srtt + 4 * fp->rttvar; 2549 sctp->sctp_zero_win_probe = B_FALSE; 2550 /* This is probably not required */ 2551 if (!sctp->sctp_rexmitting) { 2552 sctp->sctp_rxt_nxttsn = sctp->sctp_ltsn; 2553 sctp->sctp_rxt_maxtsn = sctp->sctp_ltsn; 2554 } 2555 } 2556 acked = sctp_cumack(sctp, cumtsn, &mp); 2557 sctp->sctp_xmit_unacked = mp; 2558 if (acked > 0) { 2559 trysend = 1; 2560 cumack_forward = B_TRUE; 2561 if (fwd_tsn && SEQ_GEQ(sctp->sctp_lastack_rxd, 2562 sctp->sctp_adv_pap)) { 2563 cumack_forward = B_FALSE; 2564 } 2565 } 2566 } 2567 num_gaps = ntohs(sc->ssc_numfrags); 2568 if (num_gaps == 0 || mp == NULL || !SCTP_CHUNK_ISSENT(mp) || 2569 chunklen < (sizeof (*sch) + sizeof (*sc) + 2570 num_gaps * sizeof (*ssf))) { 2571 goto ret; 2572 } 2573 #ifdef DEBUG 2574 /* 2575 * Since we delete any message that has been acked completely, 2576 * the unacked chunk must belong to sctp_xmit_head (as 2577 * we don't have a back pointer from the mp to the meta data 2578 * we do this). 2579 */ 2580 { 2581 mblk_t *mp2 = sctp->sctp_xmit_head->b_cont; 2582 2583 while (mp2 != NULL) { 2584 if (mp2 == mp) 2585 break; 2586 mp2 = mp2->b_next; 2587 } 2588 ASSERT(mp2 != NULL); 2589 } 2590 #endif 2591 ump = sctp->sctp_xmit_head; 2592 2593 /* 2594 * Just remember where we started from, in case we need to call 2595 * sctp_process_uo_gaps() if the gap blocks are unordered. 2596 */ 2597 mp1 = mp; 2598 2599 sdc = (sctp_data_hdr_t *)mp->b_rptr; 2600 xtsn = ntohl(sdc->sdh_tsn); 2601 ASSERT(xtsn == cumtsn + 1); 2602 2603 /* 2604 * Go through SACK gaps. They are ordered based on start TSN. 2605 */ 2606 ssf = (sctp_sack_frag_t *)(sc + 1); 2607 for (i = 0; i < num_gaps; i++, ssf++) { 2608 if (gapstart != 0) { 2609 /* check for unordered gap */ 2610 if (SEQ_LEQ(cumtsn + ntohs(ssf->ssf_start), gapstart)) { 2611 acked += sctp_process_uo_gaps(sctp, 2612 cumtsn, ssf, num_gaps - i, 2613 sctp->sctp_xmit_head, mp1, 2614 &trysend, &fast_recovery, gapstart); 2615 if (trysend < 0) { 2616 BUMP_MIB(&sctps->sctps_mib, 2617 sctpInAckUnsent); 2618 return (-1); 2619 } 2620 break; 2621 } 2622 } 2623 gapstart = cumtsn + ntohs(ssf->ssf_start); 2624 gapend = cumtsn + ntohs(ssf->ssf_end); 2625 2626 /* SACK for TSN we have not sent - ABORT */ 2627 if (SEQ_GT(gapstart, sctp->sctp_ltsn - 1) || 2628 SEQ_GT(gapend, sctp->sctp_ltsn - 1)) { 2629 BUMP_MIB(&sctps->sctps_mib, sctpInAckUnsent); 2630 return (-1); 2631 } else if (SEQ_LT(gapend, gapstart)) { 2632 break; 2633 } 2634 /* 2635 * Let's start at the current TSN (for the 1st gap we start 2636 * from the cumulative TSN, for subsequent ones we start from 2637 * where the previous gapend was found - second while loop 2638 * below) and walk the transmit list till we find the TSN 2639 * corresponding to gapstart. All the unacked chunks till we 2640 * get to the chunk with TSN == gapstart will have their 2641 * SACKCNT incremented by 1. Note since the gap blocks are 2642 * ordered, we won't be incrementing the SACKCNT for an 2643 * unacked chunk by more than one while processing the gap 2644 * blocks. If the SACKCNT for any unacked chunk exceeds 2645 * the fast retransmit threshold, we will fast retransmit 2646 * after processing all the gap blocks. 2647 */ 2648 ASSERT(SEQ_LT(xtsn, gapstart)); 2649 while (xtsn != gapstart) { 2650 SCTP_CHUNK_SET_SACKCNT(mp, SCTP_CHUNK_SACKCNT(mp) + 1); 2651 if (SCTP_CHUNK_SACKCNT(mp) == 2652 sctps->sctps_fast_rxt_thresh) { 2653 SCTP_CHUNK_REXMIT(mp); 2654 sctp->sctp_chk_fast_rexmit = B_TRUE; 2655 trysend = 1; 2656 if (!fast_recovery) { 2657 /* 2658 * Entering fast recovery. 2659 */ 2660 fp = SCTP_CHUNK_DEST(mp); 2661 fp->ssthresh = fp->cwnd / 2; 2662 if (fp->ssthresh < 2 * fp->sfa_pmss) { 2663 fp->ssthresh = 2664 2 * fp->sfa_pmss; 2665 } 2666 fp->cwnd = fp->ssthresh; 2667 fp->pba = 0; 2668 sctp->sctp_recovery_tsn = 2669 sctp->sctp_ltsn - 1; 2670 fast_recovery = B_TRUE; 2671 } 2672 } 2673 2674 /* 2675 * Peer may have reneged on this chunk, so un-sack 2676 * it now. If the peer did renege, we need to 2677 * readjust unacked. 2678 */ 2679 if (SCTP_CHUNK_ISACKED(mp)) { 2680 chunklen = ntohs(sdc->sdh_len); 2681 fp = SCTP_CHUNK_DEST(mp); 2682 fp->suna += chunklen; 2683 sctp->sctp_unacked += chunklen - sizeof (*sdc); 2684 SCTP_CHUNK_CLEAR_ACKED(mp); 2685 if (!fp->timer_running) { 2686 SCTP_FADDR_TIMER_RESTART(sctp, fp, 2687 fp->rto); 2688 } 2689 } 2690 2691 mp = mp->b_next; 2692 if (mp == NULL) { 2693 ump = ump->b_next; 2694 /* 2695 * ump can't be NULL given the sanity check 2696 * above. 2697 */ 2698 ASSERT(ump != NULL); 2699 mp = ump->b_cont; 2700 } 2701 /* 2702 * mp can't be unsent given the sanity check above. 2703 */ 2704 ASSERT(SCTP_CHUNK_ISSENT(mp)); 2705 sdc = (sctp_data_hdr_t *)mp->b_rptr; 2706 xtsn = ntohl(sdc->sdh_tsn); 2707 } 2708 /* 2709 * Now that we have found the chunk with TSN == 'gapstart', 2710 * let's walk till we hit the chunk with TSN == 'gapend'. 2711 * All intermediate chunks will be marked ACKED, if they 2712 * haven't already been. 2713 */ 2714 while (SEQ_LEQ(xtsn, gapend)) { 2715 /* 2716 * SACKed 2717 */ 2718 SCTP_CHUNK_SET_SACKCNT(mp, 0); 2719 if (!SCTP_CHUNK_ISACKED(mp)) { 2720 SCTP_CHUNK_ACKED(mp); 2721 2722 fp = SCTP_CHUNK_DEST(mp); 2723 chunklen = ntohs(sdc->sdh_len); 2724 ASSERT(fp->suna >= chunklen); 2725 fp->suna -= chunklen; 2726 if (fp->suna == 0) { 2727 /* All outstanding data acked. */ 2728 fp->pba = 0; 2729 SCTP_FADDR_TIMER_STOP(fp); 2730 } 2731 fp->acked += chunklen; 2732 acked += chunklen; 2733 sctp->sctp_unacked -= chunklen - sizeof (*sdc); 2734 ASSERT(sctp->sctp_unacked >= 0); 2735 } 2736 /* Go to the next chunk of the current message */ 2737 mp = mp->b_next; 2738 /* 2739 * Move to the next message in the transmit list 2740 * if we are done with all the chunks from the current 2741 * message. Note, it is possible to hit the end of the 2742 * transmit list here, i.e. if we have already completed 2743 * processing the gap block. 2744 * Also, note that we break here, which means we 2745 * continue processing gap blocks, if any. In case of 2746 * ordered gap blocks there can't be any following 2747 * this (if there is it will fail the sanity check 2748 * above). In case of un-ordered gap blocks we will 2749 * switch to sctp_process_uo_gaps(). In either case 2750 * it should be fine to continue with NULL ump/mp, 2751 * but we just reset it to xmit_head. 2752 */ 2753 if (mp == NULL) { 2754 ump = ump->b_next; 2755 if (ump == NULL) { 2756 ASSERT(xtsn == gapend); 2757 ump = sctp->sctp_xmit_head; 2758 mp = mp1; 2759 sdc = (sctp_data_hdr_t *)mp->b_rptr; 2760 xtsn = ntohl(sdc->sdh_tsn); 2761 break; 2762 } 2763 mp = ump->b_cont; 2764 } 2765 /* 2766 * Likewise, we could hit an unsent chunk once we have 2767 * completed processing the gap block. Again, it is 2768 * fine to continue processing gap blocks with mp 2769 * pointing to the unsent chunk, because if there 2770 * are more ordered gap blocks, they will fail the 2771 * sanity check, and if there are un-ordered gap blocks, 2772 * we will continue processing in sctp_process_uo_gaps() 2773 * We just reset the mp to the one we started with. 2774 */ 2775 if (!SCTP_CHUNK_ISSENT(mp)) { 2776 ASSERT(xtsn == gapend); 2777 ump = sctp->sctp_xmit_head; 2778 mp = mp1; 2779 sdc = (sctp_data_hdr_t *)mp->b_rptr; 2780 xtsn = ntohl(sdc->sdh_tsn); 2781 break; 2782 } 2783 sdc = (sctp_data_hdr_t *)mp->b_rptr; 2784 xtsn = ntohl(sdc->sdh_tsn); 2785 } 2786 } 2787 if (sctp->sctp_prsctp_aware) 2788 sctp_check_abandoned_data(sctp, sctp->sctp_current); 2789 if (sctp->sctp_chk_fast_rexmit) 2790 sctp_fast_rexmit(sctp); 2791 ret: 2792 trysend += sctp_set_frwnd(sctp, ntohl(sc->ssc_a_rwnd)); 2793 2794 /* 2795 * If receive window is closed while there is unsent data, 2796 * set a timer for doing zero window probes. 2797 */ 2798 if (sctp->sctp_frwnd == 0 && sctp->sctp_unacked == 0 && 2799 sctp->sctp_unsent != 0) { 2800 SCTP_FADDR_TIMER_RESTART(sctp, sctp->sctp_current, 2801 sctp->sctp_current->rto); 2802 } 2803 2804 /* 2805 * Set cwnd for all destinations. 2806 * Congestion window gets increased only when cumulative 2807 * TSN moves forward, we're not in fast recovery, and 2808 * cwnd has been fully utilized (almost fully, need to allow 2809 * some leeway due to non-MSS sized messages). 2810 */ 2811 if (sctp->sctp_current->acked == acked) { 2812 /* 2813 * Fast-path, only data sent to sctp_current got acked. 2814 */ 2815 fp = sctp->sctp_current; 2816 if (cumack_forward && !fast_recovery && 2817 (fp->acked + fp->suna > fp->cwnd - fp->sfa_pmss)) { 2818 if (fp->cwnd < fp->ssthresh) { 2819 /* 2820 * Slow start 2821 */ 2822 if (fp->acked > fp->sfa_pmss) { 2823 fp->cwnd += fp->sfa_pmss; 2824 } else { 2825 fp->cwnd += fp->acked; 2826 } 2827 fp->cwnd = MIN(fp->cwnd, sctp->sctp_cwnd_max); 2828 } else { 2829 /* 2830 * Congestion avoidance 2831 */ 2832 fp->pba += fp->acked; 2833 if (fp->pba >= fp->cwnd) { 2834 fp->pba -= fp->cwnd; 2835 fp->cwnd += fp->sfa_pmss; 2836 fp->cwnd = MIN(fp->cwnd, 2837 sctp->sctp_cwnd_max); 2838 } 2839 } 2840 } 2841 /* 2842 * Limit the burst of transmitted data segments. 2843 */ 2844 if (fp->suna + sctps->sctps_maxburst * fp->sfa_pmss < 2845 fp->cwnd) { 2846 fp->cwnd = fp->suna + sctps->sctps_maxburst * 2847 fp->sfa_pmss; 2848 } 2849 fp->acked = 0; 2850 goto check_ss_rxmit; 2851 } 2852 for (fp = sctp->sctp_faddrs; fp != NULL; fp = fp->next) { 2853 if (cumack_forward && fp->acked && !fast_recovery && 2854 (fp->acked + fp->suna > fp->cwnd - fp->sfa_pmss)) { 2855 if (fp->cwnd < fp->ssthresh) { 2856 if (fp->acked > fp->sfa_pmss) { 2857 fp->cwnd += fp->sfa_pmss; 2858 } else { 2859 fp->cwnd += fp->acked; 2860 } 2861 fp->cwnd = MIN(fp->cwnd, sctp->sctp_cwnd_max); 2862 } else { 2863 fp->pba += fp->acked; 2864 if (fp->pba >= fp->cwnd) { 2865 fp->pba -= fp->cwnd; 2866 fp->cwnd += fp->sfa_pmss; 2867 fp->cwnd = MIN(fp->cwnd, 2868 sctp->sctp_cwnd_max); 2869 } 2870 } 2871 } 2872 if (fp->suna + sctps->sctps_maxburst * fp->sfa_pmss < 2873 fp->cwnd) { 2874 fp->cwnd = fp->suna + sctps->sctps_maxburst * 2875 fp->sfa_pmss; 2876 } 2877 fp->acked = 0; 2878 } 2879 check_ss_rxmit: 2880 /* 2881 * If this is a SACK following a timeout, check if there are 2882 * still unacked chunks (sent before the timeout) that we can 2883 * send. 2884 */ 2885 if (sctp->sctp_rexmitting) { 2886 if (SEQ_LT(sctp->sctp_lastack_rxd, sctp->sctp_rxt_maxtsn)) { 2887 /* 2888 * As we are in retransmission phase, we may get a 2889 * SACK which indicates some new chunks are received 2890 * but cum_tsn does not advance. During this 2891 * phase, the other side advances cum_tsn only because 2892 * it receives our retransmitted chunks. Only 2893 * this signals that some chunks are still 2894 * missing. 2895 */ 2896 if (cumack_forward) 2897 sctp_ss_rexmit(sctp); 2898 } else { 2899 sctp->sctp_rexmitting = B_FALSE; 2900 sctp->sctp_rxt_nxttsn = sctp->sctp_ltsn; 2901 sctp->sctp_rxt_maxtsn = sctp->sctp_ltsn; 2902 } 2903 } 2904 return (trysend); 2905 } 2906 2907 /* 2908 * Returns 0 if the caller should stop processing any more chunks, 2909 * 1 if the caller should skip this chunk and continue processing. 2910 */ 2911 static int 2912 sctp_strange_chunk(sctp_t *sctp, sctp_chunk_hdr_t *ch, sctp_faddr_t *fp) 2913 { 2914 mblk_t *errmp; 2915 size_t len; 2916 2917 BUMP_LOCAL(sctp->sctp_ibchunks); 2918 /* check top two bits for action required */ 2919 if (ch->sch_id & 0x40) { /* also matches 0xc0 */ 2920 len = ntohs(ch->sch_len); 2921 errmp = sctp_make_err(sctp, SCTP_ERR_UNREC_CHUNK, ch, len); 2922 if (errmp != NULL) 2923 sctp_send_err(sctp, errmp, fp); 2924 if ((ch->sch_id & 0xc0) == 0xc0) { 2925 /* skip and continue */ 2926 return (1); 2927 } else { 2928 /* stop processing */ 2929 return (0); 2930 } 2931 } 2932 if (ch->sch_id & 0x80) { 2933 /* skip and continue, no error */ 2934 return (1); 2935 } 2936 /* top two bits are clear; stop processing and no error */ 2937 return (0); 2938 } 2939 2940 /* 2941 * Basic sanity checks on all input chunks and parameters: they must 2942 * be of legitimate size for their purported type, and must follow 2943 * ordering conventions as defined in rfc2960. 2944 * 2945 * Returns 1 if the chunk and all encloded params are legitimate, 2946 * 0 otherwise. 2947 */ 2948 /*ARGSUSED*/ 2949 static int 2950 sctp_check_input(sctp_t *sctp, sctp_chunk_hdr_t *ch, ssize_t len, int first) 2951 { 2952 sctp_parm_hdr_t *ph; 2953 void *p = NULL; 2954 ssize_t clen; 2955 uint16_t ch_len; 2956 2957 ch_len = ntohs(ch->sch_len); 2958 if (ch_len > len) { 2959 return (0); 2960 } 2961 2962 switch (ch->sch_id) { 2963 case CHUNK_DATA: 2964 if (ch_len < sizeof (sctp_data_hdr_t)) { 2965 return (0); 2966 } 2967 return (1); 2968 case CHUNK_INIT: 2969 case CHUNK_INIT_ACK: 2970 { 2971 ssize_t remlen = len; 2972 2973 /* 2974 * INIT and INIT-ACK chunks must not be bundled with 2975 * any other. 2976 */ 2977 if (!first || sctp_next_chunk(ch, &remlen) != NULL || 2978 (ch_len < (sizeof (*ch) + 2979 sizeof (sctp_init_chunk_t)))) { 2980 return (0); 2981 } 2982 /* may have params that need checking */ 2983 p = (char *)(ch + 1) + sizeof (sctp_init_chunk_t); 2984 clen = ch_len - (sizeof (*ch) + 2985 sizeof (sctp_init_chunk_t)); 2986 } 2987 break; 2988 case CHUNK_SACK: 2989 if (ch_len < (sizeof (*ch) + sizeof (sctp_sack_chunk_t))) { 2990 return (0); 2991 } 2992 /* dup and gap reports checked by got_sack() */ 2993 return (1); 2994 case CHUNK_SHUTDOWN: 2995 if (ch_len < (sizeof (*ch) + sizeof (uint32_t))) { 2996 return (0); 2997 } 2998 return (1); 2999 case CHUNK_ABORT: 3000 case CHUNK_ERROR: 3001 if (ch_len < sizeof (*ch)) { 3002 return (0); 3003 } 3004 /* may have params that need checking */ 3005 p = ch + 1; 3006 clen = ch_len - sizeof (*ch); 3007 break; 3008 case CHUNK_ECNE: 3009 case CHUNK_CWR: 3010 case CHUNK_HEARTBEAT: 3011 case CHUNK_HEARTBEAT_ACK: 3012 /* Full ASCONF chunk and parameter checks are in asconf.c */ 3013 case CHUNK_ASCONF: 3014 case CHUNK_ASCONF_ACK: 3015 if (ch_len < sizeof (*ch)) { 3016 return (0); 3017 } 3018 /* heartbeat data checked by process_heartbeat() */ 3019 return (1); 3020 case CHUNK_SHUTDOWN_COMPLETE: 3021 { 3022 ssize_t remlen = len; 3023 3024 /* 3025 * SHUTDOWN-COMPLETE chunk must not be bundled with any 3026 * other 3027 */ 3028 if (!first || sctp_next_chunk(ch, &remlen) != NULL || 3029 ch_len < sizeof (*ch)) { 3030 return (0); 3031 } 3032 } 3033 return (1); 3034 case CHUNK_COOKIE: 3035 case CHUNK_COOKIE_ACK: 3036 case CHUNK_SHUTDOWN_ACK: 3037 if (ch_len < sizeof (*ch) || !first) { 3038 return (0); 3039 } 3040 return (1); 3041 case CHUNK_FORWARD_TSN: 3042 if (ch_len < (sizeof (*ch) + sizeof (uint32_t))) 3043 return (0); 3044 return (1); 3045 default: 3046 return (1); /* handled by strange_chunk() */ 3047 } 3048 3049 /* check and byteorder parameters */ 3050 if (clen <= 0) { 3051 return (1); 3052 } 3053 ASSERT(p != NULL); 3054 3055 ph = p; 3056 while (ph != NULL && clen > 0) { 3057 ch_len = ntohs(ph->sph_len); 3058 if (ch_len > len || ch_len < sizeof (*ph)) { 3059 return (0); 3060 } 3061 ph = sctp_next_parm(ph, &clen); 3062 } 3063 3064 /* All OK */ 3065 return (1); 3066 } 3067 3068 /* ARGSUSED */ 3069 static sctp_hdr_t * 3070 find_sctp_hdrs(mblk_t *mp, in6_addr_t *src, in6_addr_t *dst, 3071 uint_t *ifindex, uint_t *ip_hdr_len, ip6_pkt_t *ipp, ip_pktinfo_t *pinfo) 3072 { 3073 uchar_t *rptr; 3074 ipha_t *ip4h; 3075 ip6_t *ip6h; 3076 mblk_t *mp1; 3077 3078 rptr = mp->b_rptr; 3079 if (IPH_HDR_VERSION(rptr) == IPV4_VERSION) { 3080 *ip_hdr_len = IPH_HDR_LENGTH(rptr); 3081 ip4h = (ipha_t *)rptr; 3082 IN6_IPADDR_TO_V4MAPPED(ip4h->ipha_src, src); 3083 IN6_IPADDR_TO_V4MAPPED(ip4h->ipha_dst, dst); 3084 3085 ipp->ipp_fields |= IPPF_HOPLIMIT; 3086 ipp->ipp_hoplimit = ((ipha_t *)rptr)->ipha_ttl; 3087 if (pinfo != NULL && (pinfo->ip_pkt_flags & IPF_RECVIF)) { 3088 ipp->ipp_fields |= IPPF_IFINDEX; 3089 ipp->ipp_ifindex = pinfo->ip_pkt_ifindex; 3090 } 3091 } else { 3092 ASSERT(IPH_HDR_VERSION(rptr) == IPV6_VERSION); 3093 ip6h = (ip6_t *)rptr; 3094 ipp->ipp_fields = IPPF_HOPLIMIT; 3095 ipp->ipp_hoplimit = ip6h->ip6_hops; 3096 3097 if (ip6h->ip6_nxt != IPPROTO_SCTP) { 3098 /* Look for ifindex information */ 3099 if (ip6h->ip6_nxt == IPPROTO_RAW) { 3100 ip6i_t *ip6i = (ip6i_t *)ip6h; 3101 3102 if (ip6i->ip6i_flags & IP6I_IFINDEX) { 3103 ASSERT(ip6i->ip6i_ifindex != 0); 3104 ipp->ipp_fields |= IPPF_IFINDEX; 3105 ipp->ipp_ifindex = ip6i->ip6i_ifindex; 3106 } 3107 rptr = (uchar_t *)&ip6i[1]; 3108 mp->b_rptr = rptr; 3109 if (rptr == mp->b_wptr) { 3110 mp1 = mp->b_cont; 3111 freeb(mp); 3112 mp = mp1; 3113 rptr = mp->b_rptr; 3114 } 3115 ASSERT(mp->b_wptr - rptr >= 3116 IPV6_HDR_LEN + sizeof (sctp_hdr_t)); 3117 ip6h = (ip6_t *)rptr; 3118 } 3119 /* 3120 * Find any potentially interesting extension headers 3121 * as well as the length of the IPv6 + extension 3122 * headers. 3123 */ 3124 *ip_hdr_len = ip_find_hdr_v6(mp, ip6h, ipp, NULL); 3125 } else { 3126 *ip_hdr_len = IPV6_HDR_LEN; 3127 } 3128 *src = ip6h->ip6_src; 3129 *dst = ip6h->ip6_dst; 3130 } 3131 ASSERT((uintptr_t)(mp->b_wptr - rptr) <= (uintptr_t)INT_MAX); 3132 return ((sctp_hdr_t *)&rptr[*ip_hdr_len]); 3133 #undef IPVER 3134 } 3135 3136 static mblk_t * 3137 sctp_check_in_policy(mblk_t *mp, mblk_t *ipsec_mp) 3138 { 3139 ipsec_in_t *ii; 3140 boolean_t check = B_TRUE; 3141 boolean_t policy_present; 3142 ipha_t *ipha; 3143 ip6_t *ip6h; 3144 netstack_t *ns; 3145 ipsec_stack_t *ipss; 3146 3147 ii = (ipsec_in_t *)ipsec_mp->b_rptr; 3148 ASSERT(ii->ipsec_in_type == IPSEC_IN); 3149 ns = ii->ipsec_in_ns; 3150 ipss = ns->netstack_ipsec; 3151 3152 if (ii->ipsec_in_dont_check) { 3153 check = B_FALSE; 3154 if (!ii->ipsec_in_secure) { 3155 freeb(ipsec_mp); 3156 ipsec_mp = NULL; 3157 } 3158 } 3159 if (IPH_HDR_VERSION(mp->b_rptr) == IPV4_VERSION) { 3160 policy_present = ipss->ipsec_inbound_v4_policy_present; 3161 ipha = (ipha_t *)mp->b_rptr; 3162 ip6h = NULL; 3163 } else { 3164 policy_present = ipss->ipsec_inbound_v6_policy_present; 3165 ipha = NULL; 3166 ip6h = (ip6_t *)mp->b_rptr; 3167 } 3168 3169 if (check && policy_present) { 3170 /* 3171 * The conn_t parameter is NULL because we already know 3172 * nobody's home. 3173 */ 3174 ipsec_mp = ipsec_check_global_policy(ipsec_mp, (conn_t *)NULL, 3175 ipha, ip6h, B_TRUE, ns); 3176 if (ipsec_mp == NULL) 3177 return (NULL); 3178 } 3179 if (ipsec_mp != NULL) 3180 freeb(ipsec_mp); 3181 return (mp); 3182 } 3183 3184 /* Handle out-of-the-blue packets */ 3185 void 3186 sctp_ootb_input(mblk_t *mp, ill_t *recv_ill, uint_t ipif_seqid, 3187 zoneid_t zoneid, boolean_t mctl_present) 3188 { 3189 sctp_t *sctp; 3190 sctp_chunk_hdr_t *ch; 3191 sctp_hdr_t *sctph; 3192 in6_addr_t src, dst; 3193 uint_t ip_hdr_len; 3194 uint_t ifindex; 3195 ip6_pkt_t ipp; 3196 ssize_t mlen; 3197 ip_pktinfo_t *pinfo = NULL; 3198 mblk_t *first_mp; 3199 sctp_stack_t *sctps; 3200 ip_stack_t *ipst; 3201 3202 ASSERT(recv_ill != NULL); 3203 ipst = recv_ill->ill_ipst; 3204 sctps = ipst->ips_netstack->netstack_sctp; 3205 3206 BUMP_MIB(&sctps->sctps_mib, sctpOutOfBlue); 3207 BUMP_MIB(&sctps->sctps_mib, sctpInSCTPPkts); 3208 3209 if (sctps->sctps_gsctp == NULL) { 3210 /* 3211 * For non-zero stackids the default queue isn't created 3212 * until the first open, thus there can be a need to send 3213 * an error before then. But we can't do that, hence we just 3214 * drop the packet. Later during boot, when the default queue 3215 * has been setup, a retransmitted packet from the peer 3216 * will result in a error. 3217 */ 3218 ASSERT(sctps->sctps_netstack->netstack_stackid != 3219 GLOBAL_NETSTACKID); 3220 freemsg(mp); 3221 return; 3222 } 3223 3224 first_mp = mp; 3225 if (mctl_present) 3226 mp = mp->b_cont; 3227 3228 /* Initiate IPPf processing, if needed. */ 3229 if (IPP_ENABLED(IPP_LOCAL_IN, ipst)) { 3230 ip_process(IPP_LOCAL_IN, &mp, 3231 recv_ill->ill_phyint->phyint_ifindex); 3232 if (mp == NULL) { 3233 if (mctl_present) 3234 freeb(first_mp); 3235 return; 3236 } 3237 } 3238 3239 if (mp->b_cont != NULL) { 3240 /* 3241 * All subsequent code is vastly simplified if it can 3242 * assume a single contiguous chunk of data. 3243 */ 3244 if (pullupmsg(mp, -1) == 0) { 3245 BUMP_MIB(recv_ill->ill_ip_mib, ipIfStatsInDiscards); 3246 freemsg(first_mp); 3247 return; 3248 } 3249 } 3250 3251 /* 3252 * We don't really need to call this function... Need to 3253 * optimize later. 3254 */ 3255 sctph = find_sctp_hdrs(mp, &src, &dst, &ifindex, &ip_hdr_len, 3256 &ipp, pinfo); 3257 mlen = mp->b_wptr - (uchar_t *)(sctph + 1); 3258 if ((ch = sctp_first_chunk((uchar_t *)(sctph + 1), mlen)) == NULL) { 3259 dprint(3, ("sctp_ootb_input: invalid packet\n")); 3260 BUMP_MIB(recv_ill->ill_ip_mib, ipIfStatsInDiscards); 3261 freemsg(first_mp); 3262 return; 3263 } 3264 3265 switch (ch->sch_id) { 3266 case CHUNK_INIT: 3267 /* no listener; send abort */ 3268 if (mctl_present && sctp_check_in_policy(mp, first_mp) == NULL) 3269 return; 3270 sctp_send_abort(sctps->sctps_gsctp, sctp_init2vtag(ch), 0, 3271 NULL, 0, mp, 0, B_TRUE); 3272 break; 3273 case CHUNK_INIT_ACK: 3274 /* check for changed src addr */ 3275 sctp = sctp_addrlist2sctp(mp, sctph, ch, ipif_seqid, zoneid, 3276 sctps); 3277 if (sctp != NULL) { 3278 /* success; proceed to normal path */ 3279 mutex_enter(&sctp->sctp_lock); 3280 if (sctp->sctp_running) { 3281 if (!sctp_add_recvq(sctp, mp, B_FALSE)) { 3282 BUMP_MIB(recv_ill->ill_ip_mib, 3283 ipIfStatsInDiscards); 3284 freemsg(mp); 3285 } 3286 mutex_exit(&sctp->sctp_lock); 3287 } else { 3288 /* 3289 * If the source address is changed, we 3290 * don't need to worry too much about 3291 * out of order processing. So we don't 3292 * check if the recvq is empty or not here. 3293 */ 3294 sctp->sctp_running = B_TRUE; 3295 mutex_exit(&sctp->sctp_lock); 3296 sctp_input_data(sctp, mp, NULL); 3297 WAKE_SCTP(sctp); 3298 sctp_process_sendq(sctp); 3299 } 3300 SCTP_REFRELE(sctp); 3301 return; 3302 } 3303 if (mctl_present) 3304 freeb(first_mp); 3305 /* else bogus init ack; drop it */ 3306 break; 3307 case CHUNK_SHUTDOWN_ACK: 3308 if (mctl_present && sctp_check_in_policy(mp, first_mp) == NULL) 3309 return; 3310 sctp_ootb_shutdown_ack(sctps->sctps_gsctp, mp, ip_hdr_len); 3311 sctp_process_sendq(sctps->sctps_gsctp); 3312 return; 3313 case CHUNK_ERROR: 3314 case CHUNK_ABORT: 3315 case CHUNK_COOKIE_ACK: 3316 case CHUNK_SHUTDOWN_COMPLETE: 3317 if (mctl_present) 3318 freeb(first_mp); 3319 break; 3320 default: 3321 if (mctl_present && sctp_check_in_policy(mp, first_mp) == NULL) 3322 return; 3323 sctp_send_abort(sctps->sctps_gsctp, sctph->sh_verf, 0, 3324 NULL, 0, mp, 0, B_TRUE); 3325 break; 3326 } 3327 sctp_process_sendq(sctps->sctps_gsctp); 3328 freemsg(mp); 3329 } 3330 3331 void 3332 sctp_input(conn_t *connp, ipha_t *ipha, mblk_t *mp, mblk_t *first_mp, 3333 ill_t *recv_ill, boolean_t isv4, boolean_t mctl_present) 3334 { 3335 sctp_t *sctp = CONN2SCTP(connp); 3336 ip_stack_t *ipst = recv_ill->ill_ipst; 3337 ipsec_stack_t *ipss = ipst->ips_netstack->netstack_ipsec; 3338 3339 /* 3340 * We check some fields in conn_t without holding a lock. 3341 * This should be fine. 3342 */ 3343 if (CONN_INBOUND_POLICY_PRESENT(connp, ipss) || mctl_present) { 3344 first_mp = ipsec_check_inbound_policy(first_mp, connp, 3345 ipha, NULL, mctl_present); 3346 if (first_mp == NULL) { 3347 BUMP_MIB(recv_ill->ill_ip_mib, ipIfStatsInDiscards); 3348 SCTP_REFRELE(sctp); 3349 return; 3350 } 3351 } 3352 3353 /* Initiate IPPF processing for fastpath */ 3354 if (IPP_ENABLED(IPP_LOCAL_IN, ipst)) { 3355 ip_process(IPP_LOCAL_IN, &mp, 3356 recv_ill->ill_phyint->phyint_ifindex); 3357 if (mp == NULL) { 3358 SCTP_REFRELE(sctp); 3359 if (mctl_present) 3360 freeb(first_mp); 3361 return; 3362 } else if (mctl_present) { 3363 /* 3364 * ip_process might return a new mp. 3365 */ 3366 ASSERT(first_mp != mp); 3367 first_mp->b_cont = mp; 3368 } else { 3369 first_mp = mp; 3370 } 3371 } 3372 3373 if (connp->conn_recvif || connp->conn_recvslla || 3374 connp->conn_ip_recvpktinfo) { 3375 int in_flags = 0; 3376 3377 if (connp->conn_recvif || connp->conn_ip_recvpktinfo) { 3378 in_flags = IPF_RECVIF; 3379 } 3380 if (connp->conn_recvslla) { 3381 in_flags |= IPF_RECVSLLA; 3382 } 3383 if (isv4) { 3384 mp = ip_add_info(mp, recv_ill, in_flags, 3385 IPCL_ZONEID(connp), ipst); 3386 } else { 3387 mp = ip_add_info_v6(mp, recv_ill, 3388 &(((ip6_t *)ipha)->ip6_dst)); 3389 } 3390 if (mp == NULL) { 3391 BUMP_MIB(recv_ill->ill_ip_mib, ipIfStatsInDiscards); 3392 SCTP_REFRELE(sctp); 3393 if (mctl_present) 3394 freeb(first_mp); 3395 return; 3396 } else if (mctl_present) { 3397 /* 3398 * ip_add_info might return a new mp. 3399 */ 3400 ASSERT(first_mp != mp); 3401 first_mp->b_cont = mp; 3402 } else { 3403 first_mp = mp; 3404 } 3405 } 3406 3407 mutex_enter(&sctp->sctp_lock); 3408 if (sctp->sctp_running) { 3409 if (mctl_present) 3410 mp->b_prev = first_mp; 3411 if (!sctp_add_recvq(sctp, mp, B_FALSE)) { 3412 BUMP_MIB(recv_ill->ill_ip_mib, ipIfStatsInDiscards); 3413 freemsg(first_mp); 3414 } 3415 mutex_exit(&sctp->sctp_lock); 3416 SCTP_REFRELE(sctp); 3417 return; 3418 } else { 3419 sctp->sctp_running = B_TRUE; 3420 mutex_exit(&sctp->sctp_lock); 3421 3422 mutex_enter(&sctp->sctp_recvq_lock); 3423 if (sctp->sctp_recvq != NULL) { 3424 if (mctl_present) 3425 mp->b_prev = first_mp; 3426 if (!sctp_add_recvq(sctp, mp, B_TRUE)) { 3427 BUMP_MIB(recv_ill->ill_ip_mib, 3428 ipIfStatsInDiscards); 3429 freemsg(first_mp); 3430 } 3431 mutex_exit(&sctp->sctp_recvq_lock); 3432 WAKE_SCTP(sctp); 3433 SCTP_REFRELE(sctp); 3434 return; 3435 } 3436 } 3437 mutex_exit(&sctp->sctp_recvq_lock); 3438 sctp_input_data(sctp, mp, (mctl_present ? first_mp : NULL)); 3439 WAKE_SCTP(sctp); 3440 sctp_process_sendq(sctp); 3441 SCTP_REFRELE(sctp); 3442 } 3443 3444 static void 3445 sctp_process_abort(sctp_t *sctp, sctp_chunk_hdr_t *ch, int err) 3446 { 3447 sctp_stack_t *sctps = sctp->sctp_sctps; 3448 3449 BUMP_MIB(&sctps->sctps_mib, sctpAborted); 3450 BUMP_LOCAL(sctp->sctp_ibchunks); 3451 3452 sctp_assoc_event(sctp, SCTP_COMM_LOST, 3453 ntohs(((sctp_parm_hdr_t *)(ch + 1))->sph_type), ch); 3454 sctp_clean_death(sctp, err); 3455 } 3456 3457 void 3458 sctp_input_data(sctp_t *sctp, mblk_t *mp, mblk_t *ipsec_mp) 3459 { 3460 sctp_chunk_hdr_t *ch; 3461 ssize_t mlen; 3462 int gotdata; 3463 int trysend; 3464 sctp_faddr_t *fp; 3465 sctp_init_chunk_t *iack; 3466 uint32_t tsn; 3467 sctp_data_hdr_t *sdc; 3468 ip6_pkt_t ipp; 3469 in6_addr_t src; 3470 in6_addr_t dst; 3471 uint_t ifindex; 3472 sctp_hdr_t *sctph; 3473 uint_t ip_hdr_len; 3474 mblk_t *dups = NULL; 3475 int recv_adaption; 3476 boolean_t wake_eager = B_FALSE; 3477 mblk_t *pinfo_mp; 3478 ip_pktinfo_t *pinfo = NULL; 3479 in6_addr_t peer_src; 3480 int64_t now; 3481 sctp_stack_t *sctps = sctp->sctp_sctps; 3482 ip_stack_t *ipst = sctps->sctps_netstack->netstack_ip; 3483 3484 if (DB_TYPE(mp) != M_DATA) { 3485 ASSERT(DB_TYPE(mp) == M_CTL); 3486 if (MBLKL(mp) == sizeof (ip_pktinfo_t) && 3487 ((ip_pktinfo_t *)mp->b_rptr)->ip_pkt_ulp_type == 3488 IN_PKTINFO) { 3489 pinfo = (ip_pktinfo_t *)mp->b_rptr; 3490 pinfo_mp = mp; 3491 mp = mp->b_cont; 3492 } else { 3493 if (ipsec_mp != NULL) 3494 freeb(ipsec_mp); 3495 sctp_icmp_error(sctp, mp); 3496 return; 3497 } 3498 } 3499 ASSERT(DB_TYPE(mp) == M_DATA); 3500 3501 if (mp->b_cont != NULL) { 3502 /* 3503 * All subsequent code is vastly simplified if it can 3504 * assume a single contiguous chunk of data. 3505 */ 3506 if (pullupmsg(mp, -1) == 0) { 3507 BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsInDiscards); 3508 if (ipsec_mp != NULL) 3509 freeb(ipsec_mp); 3510 if (pinfo != NULL) 3511 freeb(pinfo_mp); 3512 freemsg(mp); 3513 return; 3514 } 3515 } 3516 3517 BUMP_LOCAL(sctp->sctp_ipkts); 3518 sctph = find_sctp_hdrs(mp, &src, &dst, &ifindex, &ip_hdr_len, 3519 &ipp, pinfo); 3520 if (pinfo != NULL) 3521 freeb(pinfo_mp); 3522 mlen = mp->b_wptr - (uchar_t *)(sctph + 1); 3523 ch = sctp_first_chunk((uchar_t *)(sctph + 1), mlen); 3524 if (ch == NULL) { 3525 BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsInDiscards); 3526 if (ipsec_mp != NULL) 3527 freeb(ipsec_mp); 3528 freemsg(mp); 3529 return; 3530 } 3531 3532 if (!sctp_check_input(sctp, ch, mlen, 1)) { 3533 BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsInDiscards); 3534 goto done; 3535 } 3536 /* 3537 * Check verfication tag (special handling for INIT, 3538 * COOKIE, SHUTDOWN_COMPLETE and SHUTDOWN_ACK chunks). 3539 * ABORTs are handled in the chunk processing loop, since 3540 * may not appear first. All other checked chunks must 3541 * appear first, or will have been dropped by check_input(). 3542 */ 3543 switch (ch->sch_id) { 3544 case CHUNK_INIT: 3545 if (sctph->sh_verf != 0) { 3546 /* drop it */ 3547 goto done; 3548 } 3549 break; 3550 case CHUNK_SHUTDOWN_COMPLETE: 3551 if (sctph->sh_verf == sctp->sctp_lvtag) 3552 break; 3553 if (sctph->sh_verf == sctp->sctp_fvtag && 3554 SCTP_GET_TBIT(ch)) { 3555 break; 3556 } 3557 /* else drop it */ 3558 goto done; 3559 case CHUNK_ABORT: 3560 case CHUNK_COOKIE: 3561 /* handled below */ 3562 break; 3563 case CHUNK_SHUTDOWN_ACK: 3564 if (sctp->sctp_state > SCTPS_BOUND && 3565 sctp->sctp_state < SCTPS_ESTABLISHED) { 3566 /* treat as OOTB */ 3567 sctp_ootb_shutdown_ack(sctp, mp, ip_hdr_len); 3568 if (ipsec_mp != NULL) 3569 freeb(ipsec_mp); 3570 return; 3571 } 3572 /* else fallthru */ 3573 default: 3574 /* 3575 * All other packets must have a valid 3576 * verification tag, however if this is a 3577 * listener, we use a refined version of 3578 * out-of-the-blue logic. 3579 */ 3580 if (sctph->sh_verf != sctp->sctp_lvtag && 3581 sctp->sctp_state != SCTPS_LISTEN) { 3582 /* drop it */ 3583 goto done; 3584 } 3585 break; 3586 } 3587 3588 /* Have a valid sctp for this packet */ 3589 fp = sctp_lookup_faddr(sctp, &src); 3590 dprint(2, ("sctp_dispatch_rput: mp=%p fp=%p sctp=%p\n", (void *)mp, 3591 (void *)fp, (void *)sctp)); 3592 3593 gotdata = 0; 3594 trysend = 0; 3595 3596 now = lbolt64; 3597 /* Process the chunks */ 3598 do { 3599 dprint(3, ("sctp_dispatch_rput: state=%d, chunk id=%d\n", 3600 sctp->sctp_state, (int)(ch->sch_id))); 3601 3602 if (ch->sch_id == CHUNK_ABORT) { 3603 if (sctph->sh_verf != sctp->sctp_lvtag && 3604 sctph->sh_verf != sctp->sctp_fvtag) { 3605 /* drop it */ 3606 goto done; 3607 } 3608 } 3609 3610 switch (sctp->sctp_state) { 3611 3612 case SCTPS_ESTABLISHED: 3613 case SCTPS_SHUTDOWN_PENDING: 3614 case SCTPS_SHUTDOWN_SENT: 3615 switch (ch->sch_id) { 3616 case CHUNK_DATA: 3617 /* 0-length data chunks are not allowed */ 3618 if (ntohs(ch->sch_len) == sizeof (*sdc)) { 3619 sdc = (sctp_data_hdr_t *)ch; 3620 tsn = sdc->sdh_tsn; 3621 sctp_send_abort(sctp, sctp->sctp_fvtag, 3622 SCTP_ERR_NO_USR_DATA, (char *)&tsn, 3623 sizeof (tsn), mp, 0, B_FALSE); 3624 sctp_assoc_event(sctp, SCTP_COMM_LOST, 3625 0, NULL); 3626 sctp_clean_death(sctp, ECONNABORTED); 3627 goto done; 3628 } 3629 3630 ASSERT(fp != NULL); 3631 sctp->sctp_lastdata = fp; 3632 sctp_data_chunk(sctp, ch, mp, &dups, fp, &ipp); 3633 gotdata = 1; 3634 /* Restart shutdown timer if shutting down */ 3635 if (sctp->sctp_state == SCTPS_SHUTDOWN_SENT) { 3636 /* 3637 * If we have exceeded our max 3638 * wait bound for waiting for a 3639 * shutdown ack from the peer, 3640 * abort the association. 3641 */ 3642 if (sctps->sctps_shutack_wait_bound != 3643 0 && 3644 TICK_TO_MSEC(now - 3645 sctp->sctp_out_time) > 3646 sctps->sctps_shutack_wait_bound) { 3647 sctp_send_abort(sctp, 3648 sctp->sctp_fvtag, 0, NULL, 3649 0, mp, 0, B_FALSE); 3650 sctp_assoc_event(sctp, 3651 SCTP_COMM_LOST, 0, NULL); 3652 sctp_clean_death(sctp, 3653 ECONNABORTED); 3654 goto done; 3655 } 3656 SCTP_FADDR_TIMER_RESTART(sctp, fp, 3657 fp->rto); 3658 } 3659 break; 3660 case CHUNK_SACK: 3661 ASSERT(fp != NULL); 3662 /* 3663 * Peer is real and alive if it can ack our 3664 * data. 3665 */ 3666 sctp_faddr_alive(sctp, fp); 3667 trysend = sctp_got_sack(sctp, ch); 3668 if (trysend < 0) { 3669 sctp_send_abort(sctp, sctph->sh_verf, 3670 0, NULL, 0, mp, 0, B_FALSE); 3671 sctp_assoc_event(sctp, 3672 SCTP_COMM_LOST, 0, NULL); 3673 sctp_clean_death(sctp, 3674 ECONNABORTED); 3675 goto done; 3676 } 3677 break; 3678 case CHUNK_HEARTBEAT: 3679 sctp_return_heartbeat(sctp, ch, mp); 3680 break; 3681 case CHUNK_HEARTBEAT_ACK: 3682 sctp_process_heartbeat(sctp, ch); 3683 break; 3684 case CHUNK_SHUTDOWN: 3685 sctp_shutdown_event(sctp); 3686 trysend = sctp_shutdown_received(sctp, ch, 3687 B_FALSE, B_FALSE, fp); 3688 BUMP_LOCAL(sctp->sctp_ibchunks); 3689 break; 3690 case CHUNK_SHUTDOWN_ACK: 3691 BUMP_LOCAL(sctp->sctp_ibchunks); 3692 if (sctp->sctp_state == SCTPS_SHUTDOWN_SENT) { 3693 sctp_shutdown_complete(sctp); 3694 BUMP_MIB(&sctps->sctps_mib, 3695 sctpShutdowns); 3696 sctp_assoc_event(sctp, 3697 SCTP_SHUTDOWN_COMP, 0, NULL); 3698 sctp_clean_death(sctp, 0); 3699 goto done; 3700 } 3701 break; 3702 case CHUNK_ABORT: { 3703 sctp_saddr_ipif_t *sp; 3704 3705 /* Ignore if delete pending */ 3706 sp = sctp_saddr_lookup(sctp, &dst, 0); 3707 ASSERT(sp != NULL); 3708 if (sp->saddr_ipif_delete_pending) { 3709 BUMP_LOCAL(sctp->sctp_ibchunks); 3710 break; 3711 } 3712 3713 sctp_process_abort(sctp, ch, ECONNRESET); 3714 goto done; 3715 } 3716 case CHUNK_INIT: 3717 sctp_send_initack(sctp, sctph, ch, mp); 3718 break; 3719 case CHUNK_COOKIE: 3720 if (sctp_process_cookie(sctp, ch, mp, &iack, 3721 sctph, &recv_adaption, NULL) != -1) { 3722 sctp_send_cookie_ack(sctp); 3723 sctp_assoc_event(sctp, SCTP_RESTART, 3724 0, NULL); 3725 if (recv_adaption) { 3726 sctp->sctp_recv_adaption = 1; 3727 sctp_adaption_event(sctp); 3728 } 3729 } else { 3730 BUMP_MIB(&sctps->sctps_mib, 3731 sctpInInvalidCookie); 3732 } 3733 break; 3734 case CHUNK_ERROR: { 3735 int error; 3736 3737 BUMP_LOCAL(sctp->sctp_ibchunks); 3738 error = sctp_handle_error(sctp, sctph, ch, mp); 3739 if (error != 0) { 3740 sctp_assoc_event(sctp, SCTP_COMM_LOST, 3741 0, NULL); 3742 sctp_clean_death(sctp, error); 3743 goto done; 3744 } 3745 break; 3746 } 3747 case CHUNK_ASCONF: 3748 ASSERT(fp != NULL); 3749 sctp_input_asconf(sctp, ch, fp); 3750 BUMP_LOCAL(sctp->sctp_ibchunks); 3751 break; 3752 case CHUNK_ASCONF_ACK: 3753 ASSERT(fp != NULL); 3754 sctp_faddr_alive(sctp, fp); 3755 sctp_input_asconf_ack(sctp, ch, fp); 3756 BUMP_LOCAL(sctp->sctp_ibchunks); 3757 break; 3758 case CHUNK_FORWARD_TSN: 3759 ASSERT(fp != NULL); 3760 sctp->sctp_lastdata = fp; 3761 sctp_process_forward_tsn(sctp, ch, fp, &ipp); 3762 gotdata = 1; 3763 BUMP_LOCAL(sctp->sctp_ibchunks); 3764 break; 3765 default: 3766 if (sctp_strange_chunk(sctp, ch, fp) == 0) { 3767 goto nomorechunks; 3768 } /* else skip and continue processing */ 3769 break; 3770 } 3771 break; 3772 3773 case SCTPS_LISTEN: 3774 switch (ch->sch_id) { 3775 case CHUNK_INIT: 3776 sctp_send_initack(sctp, sctph, ch, mp); 3777 break; 3778 case CHUNK_COOKIE: { 3779 sctp_t *eager; 3780 3781 if (sctp_process_cookie(sctp, ch, mp, &iack, 3782 sctph, &recv_adaption, &peer_src) == -1) { 3783 BUMP_MIB(&sctps->sctps_mib, 3784 sctpInInvalidCookie); 3785 goto done; 3786 } 3787 3788 /* 3789 * The cookie is good; ensure that 3790 * the peer used the verification 3791 * tag from the init ack in the header. 3792 */ 3793 if (iack->sic_inittag != sctph->sh_verf) 3794 goto done; 3795 3796 eager = sctp_conn_request(sctp, mp, ifindex, 3797 ip_hdr_len, iack, ipsec_mp); 3798 if (eager == NULL) { 3799 sctp_send_abort(sctp, sctph->sh_verf, 3800 SCTP_ERR_NO_RESOURCES, NULL, 0, mp, 3801 0, B_FALSE); 3802 goto done; 3803 } 3804 3805 /* 3806 * If there were extra chunks 3807 * bundled with the cookie, 3808 * they must be processed 3809 * on the eager's queue. We 3810 * accomplish this by refeeding 3811 * the whole packet into the 3812 * state machine on the right 3813 * q. The packet (mp) gets 3814 * there via the eager's 3815 * cookie_mp field (overloaded 3816 * with the active open role). 3817 * This is picked up when 3818 * processing the null bind 3819 * request put on the eager's 3820 * q by sctp_accept(). We must 3821 * first revert the cookie 3822 * chunk's length field to network 3823 * byteorder so it can be 3824 * properly reprocessed on the 3825 * eager's queue. 3826 */ 3827 BUMP_MIB(&sctps->sctps_mib, sctpPassiveEstab); 3828 if (mlen > ntohs(ch->sch_len)) { 3829 eager->sctp_cookie_mp = dupb(mp); 3830 mblk_setcred(eager->sctp_cookie_mp, 3831 CONN_CRED(eager->sctp_connp)); 3832 /* 3833 * If no mem, just let 3834 * the peer retransmit. 3835 */ 3836 } 3837 sctp_assoc_event(eager, SCTP_COMM_UP, 0, NULL); 3838 if (recv_adaption) { 3839 eager->sctp_recv_adaption = 1; 3840 eager->sctp_rx_adaption_code = 3841 sctp->sctp_rx_adaption_code; 3842 sctp_adaption_event(eager); 3843 } 3844 3845 eager->sctp_active = now; 3846 sctp_send_cookie_ack(eager); 3847 3848 wake_eager = B_TRUE; 3849 3850 /* 3851 * Process rest of the chunks with eager. 3852 */ 3853 sctp = eager; 3854 fp = sctp_lookup_faddr(sctp, &peer_src); 3855 /* 3856 * Confirm peer's original source. fp can 3857 * only be NULL if peer does not use the 3858 * original source as one of its addresses... 3859 */ 3860 if (fp == NULL) 3861 fp = sctp_lookup_faddr(sctp, &src); 3862 else 3863 sctp_faddr_alive(sctp, fp); 3864 3865 /* 3866 * Validate the peer addresses. It also starts 3867 * the heartbeat timer. 3868 */ 3869 sctp_validate_peer(sctp); 3870 break; 3871 } 3872 /* Anything else is considered out-of-the-blue */ 3873 case CHUNK_ERROR: 3874 case CHUNK_ABORT: 3875 case CHUNK_COOKIE_ACK: 3876 case CHUNK_SHUTDOWN_COMPLETE: 3877 BUMP_LOCAL(sctp->sctp_ibchunks); 3878 goto done; 3879 default: 3880 BUMP_LOCAL(sctp->sctp_ibchunks); 3881 sctp_send_abort(sctp, sctph->sh_verf, 0, NULL, 3882 0, mp, 0, B_TRUE); 3883 goto done; 3884 } 3885 break; 3886 3887 case SCTPS_COOKIE_WAIT: 3888 switch (ch->sch_id) { 3889 case CHUNK_INIT_ACK: 3890 sctp_stop_faddr_timers(sctp); 3891 sctp_faddr_alive(sctp, sctp->sctp_current); 3892 sctp_send_cookie_echo(sctp, ch, mp); 3893 BUMP_LOCAL(sctp->sctp_ibchunks); 3894 break; 3895 case CHUNK_ABORT: 3896 sctp_process_abort(sctp, ch, ECONNREFUSED); 3897 goto done; 3898 case CHUNK_INIT: 3899 sctp_send_initack(sctp, sctph, ch, mp); 3900 break; 3901 case CHUNK_COOKIE: 3902 if (sctp_process_cookie(sctp, ch, mp, &iack, 3903 sctph, &recv_adaption, NULL) == -1) { 3904 BUMP_MIB(&sctps->sctps_mib, 3905 sctpInInvalidCookie); 3906 break; 3907 } 3908 sctp_send_cookie_ack(sctp); 3909 sctp_stop_faddr_timers(sctp); 3910 if (!SCTP_IS_DETACHED(sctp)) { 3911 sctp->sctp_ulp_connected(sctp->sctp_ulpd); 3912 sctp_set_ulp_prop(sctp); 3913 } 3914 sctp->sctp_state = SCTPS_ESTABLISHED; 3915 sctp->sctp_assoc_start_time = (uint32_t)lbolt; 3916 BUMP_MIB(&sctps->sctps_mib, sctpActiveEstab); 3917 if (sctp->sctp_cookie_mp) { 3918 freemsg(sctp->sctp_cookie_mp); 3919 sctp->sctp_cookie_mp = NULL; 3920 } 3921 3922 /* Validate the peer addresses. */ 3923 sctp->sctp_active = now; 3924 sctp_validate_peer(sctp); 3925 3926 sctp_assoc_event(sctp, SCTP_COMM_UP, 0, NULL); 3927 if (recv_adaption) { 3928 sctp->sctp_recv_adaption = 1; 3929 sctp_adaption_event(sctp); 3930 } 3931 /* Try sending queued data, or ASCONFs */ 3932 trysend = 1; 3933 break; 3934 default: 3935 if (sctp_strange_chunk(sctp, ch, fp) == 0) { 3936 goto nomorechunks; 3937 } /* else skip and continue processing */ 3938 break; 3939 } 3940 break; 3941 3942 case SCTPS_COOKIE_ECHOED: 3943 switch (ch->sch_id) { 3944 case CHUNK_COOKIE_ACK: 3945 if (!SCTP_IS_DETACHED(sctp)) { 3946 sctp->sctp_ulp_connected(sctp->sctp_ulpd); 3947 sctp_set_ulp_prop(sctp); 3948 } 3949 if (sctp->sctp_unacked == 0) 3950 sctp_stop_faddr_timers(sctp); 3951 sctp->sctp_state = SCTPS_ESTABLISHED; 3952 sctp->sctp_assoc_start_time = (uint32_t)lbolt; 3953 BUMP_MIB(&sctps->sctps_mib, sctpActiveEstab); 3954 BUMP_LOCAL(sctp->sctp_ibchunks); 3955 if (sctp->sctp_cookie_mp) { 3956 freemsg(sctp->sctp_cookie_mp); 3957 sctp->sctp_cookie_mp = NULL; 3958 } 3959 sctp_faddr_alive(sctp, fp); 3960 /* Validate the peer addresses. */ 3961 sctp->sctp_active = now; 3962 sctp_validate_peer(sctp); 3963 3964 /* Try sending queued data, or ASCONFs */ 3965 trysend = 1; 3966 sctp_assoc_event(sctp, SCTP_COMM_UP, 0, NULL); 3967 sctp_adaption_event(sctp); 3968 break; 3969 case CHUNK_ABORT: 3970 sctp_process_abort(sctp, ch, ECONNREFUSED); 3971 goto done; 3972 case CHUNK_COOKIE: 3973 if (sctp_process_cookie(sctp, ch, mp, &iack, 3974 sctph, &recv_adaption, NULL) == -1) { 3975 BUMP_MIB(&sctps->sctps_mib, 3976 sctpInInvalidCookie); 3977 break; 3978 } 3979 sctp_send_cookie_ack(sctp); 3980 3981 if (!SCTP_IS_DETACHED(sctp)) { 3982 sctp->sctp_ulp_connected(sctp->sctp_ulpd); 3983 sctp_set_ulp_prop(sctp); 3984 } 3985 if (sctp->sctp_unacked == 0) 3986 sctp_stop_faddr_timers(sctp); 3987 sctp->sctp_state = SCTPS_ESTABLISHED; 3988 sctp->sctp_assoc_start_time = (uint32_t)lbolt; 3989 BUMP_MIB(&sctps->sctps_mib, sctpActiveEstab); 3990 if (sctp->sctp_cookie_mp) { 3991 freemsg(sctp->sctp_cookie_mp); 3992 sctp->sctp_cookie_mp = NULL; 3993 } 3994 /* Validate the peer addresses. */ 3995 sctp->sctp_active = now; 3996 sctp_validate_peer(sctp); 3997 3998 sctp_assoc_event(sctp, SCTP_COMM_UP, 0, NULL); 3999 if (recv_adaption) { 4000 sctp->sctp_recv_adaption = 1; 4001 sctp_adaption_event(sctp); 4002 } 4003 /* Try sending queued data, or ASCONFs */ 4004 trysend = 1; 4005 break; 4006 case CHUNK_INIT: 4007 sctp_send_initack(sctp, sctph, ch, mp); 4008 break; 4009 case CHUNK_ERROR: { 4010 sctp_parm_hdr_t *p; 4011 4012 BUMP_LOCAL(sctp->sctp_ibchunks); 4013 /* check for a stale cookie */ 4014 if (ntohs(ch->sch_len) >= 4015 (sizeof (*p) + sizeof (*ch)) + 4016 sizeof (uint32_t)) { 4017 4018 p = (sctp_parm_hdr_t *)(ch + 1); 4019 if (p->sph_type == 4020 htons(SCTP_ERR_STALE_COOKIE)) { 4021 BUMP_MIB(&sctps->sctps_mib, 4022 sctpAborted); 4023 sctp_error_event(sctp, ch); 4024 sctp_assoc_event(sctp, 4025 SCTP_COMM_LOST, 0, NULL); 4026 sctp_clean_death(sctp, 4027 ECONNREFUSED); 4028 goto done; 4029 } 4030 } 4031 break; 4032 } 4033 case CHUNK_HEARTBEAT: 4034 sctp_return_heartbeat(sctp, ch, mp); 4035 break; 4036 default: 4037 if (sctp_strange_chunk(sctp, ch, fp) == 0) { 4038 goto nomorechunks; 4039 } /* else skip and continue processing */ 4040 } /* switch (ch->sch_id) */ 4041 break; 4042 4043 case SCTPS_SHUTDOWN_ACK_SENT: 4044 switch (ch->sch_id) { 4045 case CHUNK_ABORT: 4046 /* Pass gathered wisdom to IP for keeping */ 4047 sctp_update_ire(sctp); 4048 sctp_process_abort(sctp, ch, 0); 4049 goto done; 4050 case CHUNK_SHUTDOWN_COMPLETE: 4051 BUMP_LOCAL(sctp->sctp_ibchunks); 4052 BUMP_MIB(&sctps->sctps_mib, sctpShutdowns); 4053 sctp_assoc_event(sctp, SCTP_SHUTDOWN_COMP, 0, 4054 NULL); 4055 4056 /* Pass gathered wisdom to IP for keeping */ 4057 sctp_update_ire(sctp); 4058 sctp_clean_death(sctp, 0); 4059 goto done; 4060 case CHUNK_SHUTDOWN_ACK: 4061 sctp_shutdown_complete(sctp); 4062 BUMP_LOCAL(sctp->sctp_ibchunks); 4063 BUMP_MIB(&sctps->sctps_mib, sctpShutdowns); 4064 sctp_assoc_event(sctp, SCTP_SHUTDOWN_COMP, 0, 4065 NULL); 4066 sctp_clean_death(sctp, 0); 4067 goto done; 4068 case CHUNK_COOKIE: 4069 (void) sctp_shutdown_received(sctp, NULL, 4070 B_TRUE, B_FALSE, fp); 4071 BUMP_LOCAL(sctp->sctp_ibchunks); 4072 break; 4073 case CHUNK_HEARTBEAT: 4074 sctp_return_heartbeat(sctp, ch, mp); 4075 break; 4076 default: 4077 if (sctp_strange_chunk(sctp, ch, fp) == 0) { 4078 goto nomorechunks; 4079 } /* else skip and continue processing */ 4080 break; 4081 } 4082 break; 4083 4084 case SCTPS_SHUTDOWN_RECEIVED: 4085 switch (ch->sch_id) { 4086 case CHUNK_SHUTDOWN: 4087 trysend = sctp_shutdown_received(sctp, ch, 4088 B_FALSE, B_FALSE, fp); 4089 break; 4090 case CHUNK_SACK: 4091 trysend = sctp_got_sack(sctp, ch); 4092 if (trysend < 0) { 4093 sctp_send_abort(sctp, sctph->sh_verf, 4094 0, NULL, 0, mp, 0, B_FALSE); 4095 sctp_assoc_event(sctp, 4096 SCTP_COMM_LOST, 0, NULL); 4097 sctp_clean_death(sctp, 4098 ECONNABORTED); 4099 goto done; 4100 } 4101 break; 4102 case CHUNK_ABORT: 4103 sctp_process_abort(sctp, ch, ECONNRESET); 4104 goto done; 4105 case CHUNK_HEARTBEAT: 4106 sctp_return_heartbeat(sctp, ch, mp); 4107 break; 4108 default: 4109 if (sctp_strange_chunk(sctp, ch, fp) == 0) { 4110 goto nomorechunks; 4111 } /* else skip and continue processing */ 4112 break; 4113 } 4114 break; 4115 4116 default: 4117 /* 4118 * The only remaining states are SCTPS_IDLE and 4119 * SCTPS_BOUND, and we should not be getting here 4120 * for these. 4121 */ 4122 ASSERT(0); 4123 } /* switch (sctp->sctp_state) */ 4124 4125 ch = sctp_next_chunk(ch, &mlen); 4126 if (ch != NULL && !sctp_check_input(sctp, ch, mlen, 0)) 4127 goto done; 4128 } while (ch != NULL); 4129 4130 /* Finished processing all chunks in packet */ 4131 4132 nomorechunks: 4133 /* SACK if necessary */ 4134 if (gotdata) { 4135 (sctp->sctp_sack_toggle)++; 4136 sctp_sack(sctp, dups); 4137 dups = NULL; 4138 4139 if (!sctp->sctp_ack_timer_running) { 4140 sctp->sctp_ack_timer_running = B_TRUE; 4141 sctp_timer(sctp, sctp->sctp_ack_mp, 4142 MSEC_TO_TICK(sctps->sctps_deferred_ack_interval)); 4143 } 4144 } 4145 4146 if (trysend) { 4147 sctp_output(sctp); 4148 if (sctp->sctp_cxmit_list != NULL) 4149 sctp_wput_asconf(sctp, NULL); 4150 } 4151 /* If there is unsent data, make sure a timer is running */ 4152 if (sctp->sctp_unsent > 0 && !sctp->sctp_current->timer_running) { 4153 SCTP_FADDR_TIMER_RESTART(sctp, sctp->sctp_current, 4154 sctp->sctp_current->rto); 4155 } 4156 4157 done: 4158 if (dups != NULL) 4159 freeb(dups); 4160 if (ipsec_mp != NULL) 4161 freeb(ipsec_mp); 4162 freemsg(mp); 4163 4164 if (wake_eager) { 4165 /* 4166 * sctp points to newly created control block, need to 4167 * release it before exiting. Before releasing it and 4168 * processing the sendq, need to grab a hold on it. 4169 * Otherwise, another thread can close it while processing 4170 * the sendq. 4171 */ 4172 SCTP_REFHOLD(sctp); 4173 WAKE_SCTP(sctp); 4174 sctp_process_sendq(sctp); 4175 SCTP_REFRELE(sctp); 4176 } 4177 } 4178 4179 /* 4180 * Some amount of data got removed from rx q. 4181 * Check if we should send a window update. 4182 * 4183 * Due to way sctp_rwnd updates are made, ULP can give reports out-of-order. 4184 * To keep from dropping incoming data due to this, we only update 4185 * sctp_rwnd when if it's larger than what we've reported to peer earlier. 4186 */ 4187 void 4188 sctp_recvd(sctp_t *sctp, int len) 4189 { 4190 int32_t old, new; 4191 sctp_stack_t *sctps = sctp->sctp_sctps; 4192 4193 ASSERT(sctp != NULL); 4194 RUN_SCTP(sctp); 4195 4196 if (len < sctp->sctp_rwnd) { 4197 WAKE_SCTP(sctp); 4198 return; 4199 } 4200 ASSERT(sctp->sctp_rwnd >= sctp->sctp_rxqueued); 4201 old = sctp->sctp_rwnd - sctp->sctp_rxqueued; 4202 new = len - sctp->sctp_rxqueued; 4203 sctp->sctp_rwnd = len; 4204 4205 if (sctp->sctp_state >= SCTPS_ESTABLISHED && 4206 ((old <= new >> 1) || (old < sctp->sctp_mss))) { 4207 sctp->sctp_force_sack = 1; 4208 BUMP_MIB(&sctps->sctps_mib, sctpOutWinUpdate); 4209 sctp_sack(sctp, NULL); 4210 old = 1; 4211 } else { 4212 old = 0; 4213 } 4214 WAKE_SCTP(sctp); 4215 if (old > 0) { 4216 sctp_process_sendq(sctp); 4217 } 4218 } 4219