1 /* SCTP kernel implementation 2 * (C) Copyright IBM Corp. 2001, 2004 3 * Copyright (c) 1999-2000 Cisco, Inc. 4 * Copyright (c) 1999-2001 Motorola, Inc. 5 * Copyright (c) 2001 Intel Corp. 6 * Copyright (c) 2001 La Monte H.P. Yarroll 7 * 8 * This file is part of the SCTP kernel implementation 9 * 10 * This module provides the abstraction for an SCTP association. 11 * 12 * This SCTP implementation is free software; 13 * you can redistribute it and/or modify it under the terms of 14 * the GNU General Public License as published by 15 * the Free Software Foundation; either version 2, or (at your option) 16 * any later version. 17 * 18 * This SCTP implementation is distributed in the hope that it 19 * will be useful, but WITHOUT ANY WARRANTY; without even the implied 20 * ************************ 21 * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. 22 * See the GNU General Public License for more details. 23 * 24 * You should have received a copy of the GNU General Public License 25 * along with GNU CC; see the file COPYING. If not, see 26 * <http://www.gnu.org/licenses/>. 27 * 28 * Please send any bug reports or fixes you make to the 29 * email address(es): 30 * lksctp developers <linux-sctp@vger.kernel.org> 31 * 32 * Written or modified by: 33 * La Monte H.P. Yarroll <piggy@acm.org> 34 * Karl Knutson <karl@athena.chicago.il.us> 35 * Jon Grimm <jgrimm@us.ibm.com> 36 * Xingang Guo <xingang.guo@intel.com> 37 * Hui Huang <hui.huang@nokia.com> 38 * Sridhar Samudrala <sri@us.ibm.com> 39 * Daisy Chang <daisyc@us.ibm.com> 40 * Ryan Layer <rmlayer@us.ibm.com> 41 * Kevin Gao <kevin.gao@intel.com> 42 */ 43 44 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 45 46 #include <linux/types.h> 47 #include <linux/fcntl.h> 48 #include <linux/poll.h> 49 #include <linux/init.h> 50 51 #include <linux/slab.h> 52 #include <linux/in.h> 53 #include <net/ipv6.h> 54 #include <net/sctp/sctp.h> 55 #include <net/sctp/sm.h> 56 57 /* Forward declarations for internal functions. */ 58 static void sctp_select_active_and_retran_path(struct sctp_association *asoc); 59 static void sctp_assoc_bh_rcv(struct work_struct *work); 60 static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc); 61 static void sctp_assoc_free_asconf_queue(struct sctp_association *asoc); 62 63 /* 1st Level Abstractions. */ 64 65 /* Initialize a new association from provided memory. */ 66 static struct sctp_association *sctp_association_init(struct sctp_association *asoc, 67 const struct sctp_endpoint *ep, 68 const struct sock *sk, 69 sctp_scope_t scope, 70 gfp_t gfp) 71 { 72 struct net *net = sock_net(sk); 73 struct sctp_sock *sp; 74 struct sctp_paramhdr *p; 75 int i; 76 77 /* Retrieve the SCTP per socket area. */ 78 sp = sctp_sk((struct sock *)sk); 79 80 /* Discarding const is appropriate here. */ 81 asoc->ep = (struct sctp_endpoint *)ep; 82 asoc->base.sk = (struct sock *)sk; 83 84 sctp_endpoint_hold(asoc->ep); 85 sock_hold(asoc->base.sk); 86 87 /* Initialize the common base substructure. */ 88 asoc->base.type = SCTP_EP_TYPE_ASSOCIATION; 89 90 /* Initialize the object handling fields. */ 91 refcount_set(&asoc->base.refcnt, 1); 92 93 /* Initialize the bind addr area. */ 94 sctp_bind_addr_init(&asoc->base.bind_addr, ep->base.bind_addr.port); 95 96 asoc->state = SCTP_STATE_CLOSED; 97 asoc->cookie_life = ms_to_ktime(sp->assocparams.sasoc_cookie_life); 98 asoc->user_frag = sp->user_frag; 99 100 /* Set the association max_retrans and RTO values from the 101 * socket values. 102 */ 103 asoc->max_retrans = sp->assocparams.sasoc_asocmaxrxt; 104 asoc->pf_retrans = net->sctp.pf_retrans; 105 106 asoc->rto_initial = msecs_to_jiffies(sp->rtoinfo.srto_initial); 107 asoc->rto_max = msecs_to_jiffies(sp->rtoinfo.srto_max); 108 asoc->rto_min = msecs_to_jiffies(sp->rtoinfo.srto_min); 109 110 /* Initialize the association's heartbeat interval based on the 111 * sock configured value. 112 */ 113 asoc->hbinterval = msecs_to_jiffies(sp->hbinterval); 114 115 /* Initialize path max retrans value. */ 116 asoc->pathmaxrxt = sp->pathmaxrxt; 117 118 /* Initialize default path MTU. */ 119 asoc->pathmtu = sp->pathmtu; 120 121 /* Set association default SACK delay */ 122 asoc->sackdelay = msecs_to_jiffies(sp->sackdelay); 123 asoc->sackfreq = sp->sackfreq; 124 125 /* Set the association default flags controlling 126 * Heartbeat, SACK delay, and Path MTU Discovery. 127 */ 128 asoc->param_flags = sp->param_flags; 129 130 /* Initialize the maximum number of new data packets that can be sent 131 * in a burst. 132 */ 133 asoc->max_burst = sp->max_burst; 134 135 /* initialize association timers */ 136 asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_COOKIE] = asoc->rto_initial; 137 asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_INIT] = asoc->rto_initial; 138 asoc->timeouts[SCTP_EVENT_TIMEOUT_T2_SHUTDOWN] = asoc->rto_initial; 139 140 /* sctpimpguide Section 2.12.2 141 * If the 'T5-shutdown-guard' timer is used, it SHOULD be set to the 142 * recommended value of 5 times 'RTO.Max'. 143 */ 144 asoc->timeouts[SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD] 145 = 5 * asoc->rto_max; 146 147 asoc->timeouts[SCTP_EVENT_TIMEOUT_SACK] = asoc->sackdelay; 148 asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE] = sp->autoclose * HZ; 149 150 /* Initializes the timers */ 151 for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i) 152 setup_timer(&asoc->timers[i], sctp_timer_events[i], 153 (unsigned long)asoc); 154 155 /* Pull default initialization values from the sock options. 156 * Note: This assumes that the values have already been 157 * validated in the sock. 158 */ 159 asoc->c.sinit_max_instreams = sp->initmsg.sinit_max_instreams; 160 asoc->c.sinit_num_ostreams = sp->initmsg.sinit_num_ostreams; 161 asoc->max_init_attempts = sp->initmsg.sinit_max_attempts; 162 163 asoc->max_init_timeo = 164 msecs_to_jiffies(sp->initmsg.sinit_max_init_timeo); 165 166 /* Set the local window size for receive. 167 * This is also the rcvbuf space per association. 168 * RFC 6 - A SCTP receiver MUST be able to receive a minimum of 169 * 1500 bytes in one SCTP packet. 170 */ 171 if ((sk->sk_rcvbuf/2) < SCTP_DEFAULT_MINWINDOW) 172 asoc->rwnd = SCTP_DEFAULT_MINWINDOW; 173 else 174 asoc->rwnd = sk->sk_rcvbuf/2; 175 176 asoc->a_rwnd = asoc->rwnd; 177 178 /* Use my own max window until I learn something better. */ 179 asoc->peer.rwnd = SCTP_DEFAULT_MAXWINDOW; 180 181 /* Initialize the receive memory counter */ 182 atomic_set(&asoc->rmem_alloc, 0); 183 184 init_waitqueue_head(&asoc->wait); 185 186 asoc->c.my_vtag = sctp_generate_tag(ep); 187 asoc->c.my_port = ep->base.bind_addr.port; 188 189 asoc->c.initial_tsn = sctp_generate_tsn(ep); 190 191 asoc->next_tsn = asoc->c.initial_tsn; 192 193 asoc->ctsn_ack_point = asoc->next_tsn - 1; 194 asoc->adv_peer_ack_point = asoc->ctsn_ack_point; 195 asoc->highest_sacked = asoc->ctsn_ack_point; 196 asoc->last_cwr_tsn = asoc->ctsn_ack_point; 197 198 /* ADDIP Section 4.1 Asconf Chunk Procedures 199 * 200 * When an endpoint has an ASCONF signaled change to be sent to the 201 * remote endpoint it should do the following: 202 * ... 203 * A2) a serial number should be assigned to the chunk. The serial 204 * number SHOULD be a monotonically increasing number. The serial 205 * numbers SHOULD be initialized at the start of the 206 * association to the same value as the initial TSN. 207 */ 208 asoc->addip_serial = asoc->c.initial_tsn; 209 asoc->strreset_outseq = asoc->c.initial_tsn; 210 211 INIT_LIST_HEAD(&asoc->addip_chunk_list); 212 INIT_LIST_HEAD(&asoc->asconf_ack_list); 213 214 /* Make an empty list of remote transport addresses. */ 215 INIT_LIST_HEAD(&asoc->peer.transport_addr_list); 216 217 /* RFC 2960 5.1 Normal Establishment of an Association 218 * 219 * After the reception of the first data chunk in an 220 * association the endpoint must immediately respond with a 221 * sack to acknowledge the data chunk. Subsequent 222 * acknowledgements should be done as described in Section 223 * 6.2. 224 * 225 * [We implement this by telling a new association that it 226 * already received one packet.] 227 */ 228 asoc->peer.sack_needed = 1; 229 asoc->peer.sack_generation = 1; 230 231 /* Assume that the peer will tell us if he recognizes ASCONF 232 * as part of INIT exchange. 233 * The sctp_addip_noauth option is there for backward compatibility 234 * and will revert old behavior. 235 */ 236 if (net->sctp.addip_noauth) 237 asoc->peer.asconf_capable = 1; 238 239 /* Create an input queue. */ 240 sctp_inq_init(&asoc->base.inqueue); 241 sctp_inq_set_th_handler(&asoc->base.inqueue, sctp_assoc_bh_rcv); 242 243 /* Create an output queue. */ 244 sctp_outq_init(asoc, &asoc->outqueue); 245 246 if (!sctp_ulpq_init(&asoc->ulpq, asoc)) 247 goto fail_init; 248 249 if (sctp_stream_init(&asoc->stream, asoc->c.sinit_num_ostreams, 250 0, gfp)) 251 goto fail_init; 252 253 /* Assume that peer would support both address types unless we are 254 * told otherwise. 255 */ 256 asoc->peer.ipv4_address = 1; 257 if (asoc->base.sk->sk_family == PF_INET6) 258 asoc->peer.ipv6_address = 1; 259 INIT_LIST_HEAD(&asoc->asocs); 260 261 asoc->default_stream = sp->default_stream; 262 asoc->default_ppid = sp->default_ppid; 263 asoc->default_flags = sp->default_flags; 264 asoc->default_context = sp->default_context; 265 asoc->default_timetolive = sp->default_timetolive; 266 asoc->default_rcv_context = sp->default_rcv_context; 267 268 /* AUTH related initializations */ 269 INIT_LIST_HEAD(&asoc->endpoint_shared_keys); 270 if (sctp_auth_asoc_copy_shkeys(ep, asoc, gfp)) 271 goto stream_free; 272 273 asoc->active_key_id = ep->active_key_id; 274 asoc->prsctp_enable = ep->prsctp_enable; 275 asoc->reconf_enable = ep->reconf_enable; 276 asoc->strreset_enable = ep->strreset_enable; 277 278 /* Save the hmacs and chunks list into this association */ 279 if (ep->auth_hmacs_list) 280 memcpy(asoc->c.auth_hmacs, ep->auth_hmacs_list, 281 ntohs(ep->auth_hmacs_list->param_hdr.length)); 282 if (ep->auth_chunk_list) 283 memcpy(asoc->c.auth_chunks, ep->auth_chunk_list, 284 ntohs(ep->auth_chunk_list->param_hdr.length)); 285 286 /* Get the AUTH random number for this association */ 287 p = (struct sctp_paramhdr *)asoc->c.auth_random; 288 p->type = SCTP_PARAM_RANDOM; 289 p->length = htons(sizeof(*p) + SCTP_AUTH_RANDOM_LENGTH); 290 get_random_bytes(p+1, SCTP_AUTH_RANDOM_LENGTH); 291 292 return asoc; 293 294 stream_free: 295 sctp_stream_free(&asoc->stream); 296 fail_init: 297 sock_put(asoc->base.sk); 298 sctp_endpoint_put(asoc->ep); 299 return NULL; 300 } 301 302 /* Allocate and initialize a new association */ 303 struct sctp_association *sctp_association_new(const struct sctp_endpoint *ep, 304 const struct sock *sk, 305 sctp_scope_t scope, 306 gfp_t gfp) 307 { 308 struct sctp_association *asoc; 309 310 asoc = kzalloc(sizeof(*asoc), gfp); 311 if (!asoc) 312 goto fail; 313 314 if (!sctp_association_init(asoc, ep, sk, scope, gfp)) 315 goto fail_init; 316 317 SCTP_DBG_OBJCNT_INC(assoc); 318 319 pr_debug("Created asoc %p\n", asoc); 320 321 return asoc; 322 323 fail_init: 324 kfree(asoc); 325 fail: 326 return NULL; 327 } 328 329 /* Free this association if possible. There may still be users, so 330 * the actual deallocation may be delayed. 331 */ 332 void sctp_association_free(struct sctp_association *asoc) 333 { 334 struct sock *sk = asoc->base.sk; 335 struct sctp_transport *transport; 336 struct list_head *pos, *temp; 337 int i; 338 339 /* Only real associations count against the endpoint, so 340 * don't bother for if this is a temporary association. 341 */ 342 if (!list_empty(&asoc->asocs)) { 343 list_del(&asoc->asocs); 344 345 /* Decrement the backlog value for a TCP-style listening 346 * socket. 347 */ 348 if (sctp_style(sk, TCP) && sctp_sstate(sk, LISTENING)) 349 sk->sk_ack_backlog--; 350 } 351 352 /* Mark as dead, so other users can know this structure is 353 * going away. 354 */ 355 asoc->base.dead = true; 356 357 /* Dispose of any data lying around in the outqueue. */ 358 sctp_outq_free(&asoc->outqueue); 359 360 /* Dispose of any pending messages for the upper layer. */ 361 sctp_ulpq_free(&asoc->ulpq); 362 363 /* Dispose of any pending chunks on the inqueue. */ 364 sctp_inq_free(&asoc->base.inqueue); 365 366 sctp_tsnmap_free(&asoc->peer.tsn_map); 367 368 /* Free stream information. */ 369 sctp_stream_free(&asoc->stream); 370 371 if (asoc->strreset_chunk) 372 sctp_chunk_free(asoc->strreset_chunk); 373 374 /* Clean up the bound address list. */ 375 sctp_bind_addr_free(&asoc->base.bind_addr); 376 377 /* Do we need to go through all of our timers and 378 * delete them? To be safe we will try to delete all, but we 379 * should be able to go through and make a guess based 380 * on our state. 381 */ 382 for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i) { 383 if (del_timer(&asoc->timers[i])) 384 sctp_association_put(asoc); 385 } 386 387 /* Free peer's cached cookie. */ 388 kfree(asoc->peer.cookie); 389 kfree(asoc->peer.peer_random); 390 kfree(asoc->peer.peer_chunks); 391 kfree(asoc->peer.peer_hmacs); 392 393 /* Release the transport structures. */ 394 list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) { 395 transport = list_entry(pos, struct sctp_transport, transports); 396 list_del_rcu(pos); 397 sctp_unhash_transport(transport); 398 sctp_transport_free(transport); 399 } 400 401 asoc->peer.transport_count = 0; 402 403 sctp_asconf_queue_teardown(asoc); 404 405 /* Free pending address space being deleted */ 406 kfree(asoc->asconf_addr_del_pending); 407 408 /* AUTH - Free the endpoint shared keys */ 409 sctp_auth_destroy_keys(&asoc->endpoint_shared_keys); 410 411 /* AUTH - Free the association shared key */ 412 sctp_auth_key_put(asoc->asoc_shared_key); 413 414 sctp_association_put(asoc); 415 } 416 417 /* Cleanup and free up an association. */ 418 static void sctp_association_destroy(struct sctp_association *asoc) 419 { 420 if (unlikely(!asoc->base.dead)) { 421 WARN(1, "Attempt to destroy undead association %p!\n", asoc); 422 return; 423 } 424 425 sctp_endpoint_put(asoc->ep); 426 sock_put(asoc->base.sk); 427 428 if (asoc->assoc_id != 0) { 429 spin_lock_bh(&sctp_assocs_id_lock); 430 idr_remove(&sctp_assocs_id, asoc->assoc_id); 431 spin_unlock_bh(&sctp_assocs_id_lock); 432 } 433 434 WARN_ON(atomic_read(&asoc->rmem_alloc)); 435 436 kfree(asoc); 437 SCTP_DBG_OBJCNT_DEC(assoc); 438 } 439 440 /* Change the primary destination address for the peer. */ 441 void sctp_assoc_set_primary(struct sctp_association *asoc, 442 struct sctp_transport *transport) 443 { 444 int changeover = 0; 445 446 /* it's a changeover only if we already have a primary path 447 * that we are changing 448 */ 449 if (asoc->peer.primary_path != NULL && 450 asoc->peer.primary_path != transport) 451 changeover = 1 ; 452 453 asoc->peer.primary_path = transport; 454 455 /* Set a default msg_name for events. */ 456 memcpy(&asoc->peer.primary_addr, &transport->ipaddr, 457 sizeof(union sctp_addr)); 458 459 /* If the primary path is changing, assume that the 460 * user wants to use this new path. 461 */ 462 if ((transport->state == SCTP_ACTIVE) || 463 (transport->state == SCTP_UNKNOWN)) 464 asoc->peer.active_path = transport; 465 466 /* 467 * SFR-CACC algorithm: 468 * Upon the receipt of a request to change the primary 469 * destination address, on the data structure for the new 470 * primary destination, the sender MUST do the following: 471 * 472 * 1) If CHANGEOVER_ACTIVE is set, then there was a switch 473 * to this destination address earlier. The sender MUST set 474 * CYCLING_CHANGEOVER to indicate that this switch is a 475 * double switch to the same destination address. 476 * 477 * Really, only bother is we have data queued or outstanding on 478 * the association. 479 */ 480 if (!asoc->outqueue.outstanding_bytes && !asoc->outqueue.out_qlen) 481 return; 482 483 if (transport->cacc.changeover_active) 484 transport->cacc.cycling_changeover = changeover; 485 486 /* 2) The sender MUST set CHANGEOVER_ACTIVE to indicate that 487 * a changeover has occurred. 488 */ 489 transport->cacc.changeover_active = changeover; 490 491 /* 3) The sender MUST store the next TSN to be sent in 492 * next_tsn_at_change. 493 */ 494 transport->cacc.next_tsn_at_change = asoc->next_tsn; 495 } 496 497 /* Remove a transport from an association. */ 498 void sctp_assoc_rm_peer(struct sctp_association *asoc, 499 struct sctp_transport *peer) 500 { 501 struct list_head *pos; 502 struct sctp_transport *transport; 503 504 pr_debug("%s: association:%p addr:%pISpc\n", 505 __func__, asoc, &peer->ipaddr.sa); 506 507 /* If we are to remove the current retran_path, update it 508 * to the next peer before removing this peer from the list. 509 */ 510 if (asoc->peer.retran_path == peer) 511 sctp_assoc_update_retran_path(asoc); 512 513 /* Remove this peer from the list. */ 514 list_del_rcu(&peer->transports); 515 /* Remove this peer from the transport hashtable */ 516 sctp_unhash_transport(peer); 517 518 /* Get the first transport of asoc. */ 519 pos = asoc->peer.transport_addr_list.next; 520 transport = list_entry(pos, struct sctp_transport, transports); 521 522 /* Update any entries that match the peer to be deleted. */ 523 if (asoc->peer.primary_path == peer) 524 sctp_assoc_set_primary(asoc, transport); 525 if (asoc->peer.active_path == peer) 526 asoc->peer.active_path = transport; 527 if (asoc->peer.retran_path == peer) 528 asoc->peer.retran_path = transport; 529 if (asoc->peer.last_data_from == peer) 530 asoc->peer.last_data_from = transport; 531 532 if (asoc->strreset_chunk && 533 asoc->strreset_chunk->transport == peer) { 534 asoc->strreset_chunk->transport = transport; 535 sctp_transport_reset_reconf_timer(transport); 536 } 537 538 /* If we remove the transport an INIT was last sent to, set it to 539 * NULL. Combined with the update of the retran path above, this 540 * will cause the next INIT to be sent to the next available 541 * transport, maintaining the cycle. 542 */ 543 if (asoc->init_last_sent_to == peer) 544 asoc->init_last_sent_to = NULL; 545 546 /* If we remove the transport an SHUTDOWN was last sent to, set it 547 * to NULL. Combined with the update of the retran path above, this 548 * will cause the next SHUTDOWN to be sent to the next available 549 * transport, maintaining the cycle. 550 */ 551 if (asoc->shutdown_last_sent_to == peer) 552 asoc->shutdown_last_sent_to = NULL; 553 554 /* If we remove the transport an ASCONF was last sent to, set it to 555 * NULL. 556 */ 557 if (asoc->addip_last_asconf && 558 asoc->addip_last_asconf->transport == peer) 559 asoc->addip_last_asconf->transport = NULL; 560 561 /* If we have something on the transmitted list, we have to 562 * save it off. The best place is the active path. 563 */ 564 if (!list_empty(&peer->transmitted)) { 565 struct sctp_transport *active = asoc->peer.active_path; 566 struct sctp_chunk *ch; 567 568 /* Reset the transport of each chunk on this list */ 569 list_for_each_entry(ch, &peer->transmitted, 570 transmitted_list) { 571 ch->transport = NULL; 572 ch->rtt_in_progress = 0; 573 } 574 575 list_splice_tail_init(&peer->transmitted, 576 &active->transmitted); 577 578 /* Start a T3 timer here in case it wasn't running so 579 * that these migrated packets have a chance to get 580 * retransmitted. 581 */ 582 if (!timer_pending(&active->T3_rtx_timer)) 583 if (!mod_timer(&active->T3_rtx_timer, 584 jiffies + active->rto)) 585 sctp_transport_hold(active); 586 } 587 588 asoc->peer.transport_count--; 589 590 sctp_transport_free(peer); 591 } 592 593 /* Add a transport address to an association. */ 594 struct sctp_transport *sctp_assoc_add_peer(struct sctp_association *asoc, 595 const union sctp_addr *addr, 596 const gfp_t gfp, 597 const int peer_state) 598 { 599 struct net *net = sock_net(asoc->base.sk); 600 struct sctp_transport *peer; 601 struct sctp_sock *sp; 602 unsigned short port; 603 604 sp = sctp_sk(asoc->base.sk); 605 606 /* AF_INET and AF_INET6 share common port field. */ 607 port = ntohs(addr->v4.sin_port); 608 609 pr_debug("%s: association:%p addr:%pISpc state:%d\n", __func__, 610 asoc, &addr->sa, peer_state); 611 612 /* Set the port if it has not been set yet. */ 613 if (0 == asoc->peer.port) 614 asoc->peer.port = port; 615 616 /* Check to see if this is a duplicate. */ 617 peer = sctp_assoc_lookup_paddr(asoc, addr); 618 if (peer) { 619 /* An UNKNOWN state is only set on transports added by 620 * user in sctp_connectx() call. Such transports should be 621 * considered CONFIRMED per RFC 4960, Section 5.4. 622 */ 623 if (peer->state == SCTP_UNKNOWN) { 624 peer->state = SCTP_ACTIVE; 625 } 626 return peer; 627 } 628 629 peer = sctp_transport_new(net, addr, gfp); 630 if (!peer) 631 return NULL; 632 633 sctp_transport_set_owner(peer, asoc); 634 635 /* Initialize the peer's heartbeat interval based on the 636 * association configured value. 637 */ 638 peer->hbinterval = asoc->hbinterval; 639 640 /* Set the path max_retrans. */ 641 peer->pathmaxrxt = asoc->pathmaxrxt; 642 643 /* And the partial failure retrans threshold */ 644 peer->pf_retrans = asoc->pf_retrans; 645 646 /* Initialize the peer's SACK delay timeout based on the 647 * association configured value. 648 */ 649 peer->sackdelay = asoc->sackdelay; 650 peer->sackfreq = asoc->sackfreq; 651 652 /* Enable/disable heartbeat, SACK delay, and path MTU discovery 653 * based on association setting. 654 */ 655 peer->param_flags = asoc->param_flags; 656 657 sctp_transport_route(peer, NULL, sp); 658 659 /* Initialize the pmtu of the transport. */ 660 if (peer->param_flags & SPP_PMTUD_DISABLE) { 661 if (asoc->pathmtu) 662 peer->pathmtu = asoc->pathmtu; 663 else 664 peer->pathmtu = SCTP_DEFAULT_MAXSEGMENT; 665 } 666 667 /* If this is the first transport addr on this association, 668 * initialize the association PMTU to the peer's PMTU. 669 * If not and the current association PMTU is higher than the new 670 * peer's PMTU, reset the association PMTU to the new peer's PMTU. 671 */ 672 if (asoc->pathmtu) 673 asoc->pathmtu = min_t(int, peer->pathmtu, asoc->pathmtu); 674 else 675 asoc->pathmtu = peer->pathmtu; 676 677 pr_debug("%s: association:%p PMTU set to %d\n", __func__, asoc, 678 asoc->pathmtu); 679 680 peer->pmtu_pending = 0; 681 682 asoc->frag_point = sctp_frag_point(asoc, asoc->pathmtu); 683 684 /* The asoc->peer.port might not be meaningful yet, but 685 * initialize the packet structure anyway. 686 */ 687 sctp_packet_init(&peer->packet, peer, asoc->base.bind_addr.port, 688 asoc->peer.port); 689 690 /* 7.2.1 Slow-Start 691 * 692 * o The initial cwnd before DATA transmission or after a sufficiently 693 * long idle period MUST be set to 694 * min(4*MTU, max(2*MTU, 4380 bytes)) 695 * 696 * o The initial value of ssthresh MAY be arbitrarily high 697 * (for example, implementations MAY use the size of the 698 * receiver advertised window). 699 */ 700 peer->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380)); 701 702 /* At this point, we may not have the receiver's advertised window, 703 * so initialize ssthresh to the default value and it will be set 704 * later when we process the INIT. 705 */ 706 peer->ssthresh = SCTP_DEFAULT_MAXWINDOW; 707 708 peer->partial_bytes_acked = 0; 709 peer->flight_size = 0; 710 peer->burst_limited = 0; 711 712 /* Set the transport's RTO.initial value */ 713 peer->rto = asoc->rto_initial; 714 sctp_max_rto(asoc, peer); 715 716 /* Set the peer's active state. */ 717 peer->state = peer_state; 718 719 /* Add this peer into the transport hashtable */ 720 if (sctp_hash_transport(peer)) { 721 sctp_transport_free(peer); 722 return NULL; 723 } 724 725 /* Attach the remote transport to our asoc. */ 726 list_add_tail_rcu(&peer->transports, &asoc->peer.transport_addr_list); 727 asoc->peer.transport_count++; 728 729 /* If we do not yet have a primary path, set one. */ 730 if (!asoc->peer.primary_path) { 731 sctp_assoc_set_primary(asoc, peer); 732 asoc->peer.retran_path = peer; 733 } 734 735 if (asoc->peer.active_path == asoc->peer.retran_path && 736 peer->state != SCTP_UNCONFIRMED) { 737 asoc->peer.retran_path = peer; 738 } 739 740 return peer; 741 } 742 743 /* Delete a transport address from an association. */ 744 void sctp_assoc_del_peer(struct sctp_association *asoc, 745 const union sctp_addr *addr) 746 { 747 struct list_head *pos; 748 struct list_head *temp; 749 struct sctp_transport *transport; 750 751 list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) { 752 transport = list_entry(pos, struct sctp_transport, transports); 753 if (sctp_cmp_addr_exact(addr, &transport->ipaddr)) { 754 /* Do book keeping for removing the peer and free it. */ 755 sctp_assoc_rm_peer(asoc, transport); 756 break; 757 } 758 } 759 } 760 761 /* Lookup a transport by address. */ 762 struct sctp_transport *sctp_assoc_lookup_paddr( 763 const struct sctp_association *asoc, 764 const union sctp_addr *address) 765 { 766 struct sctp_transport *t; 767 768 /* Cycle through all transports searching for a peer address. */ 769 770 list_for_each_entry(t, &asoc->peer.transport_addr_list, 771 transports) { 772 if (sctp_cmp_addr_exact(address, &t->ipaddr)) 773 return t; 774 } 775 776 return NULL; 777 } 778 779 /* Remove all transports except a give one */ 780 void sctp_assoc_del_nonprimary_peers(struct sctp_association *asoc, 781 struct sctp_transport *primary) 782 { 783 struct sctp_transport *temp; 784 struct sctp_transport *t; 785 786 list_for_each_entry_safe(t, temp, &asoc->peer.transport_addr_list, 787 transports) { 788 /* if the current transport is not the primary one, delete it */ 789 if (t != primary) 790 sctp_assoc_rm_peer(asoc, t); 791 } 792 } 793 794 /* Engage in transport control operations. 795 * Mark the transport up or down and send a notification to the user. 796 * Select and update the new active and retran paths. 797 */ 798 void sctp_assoc_control_transport(struct sctp_association *asoc, 799 struct sctp_transport *transport, 800 sctp_transport_cmd_t command, 801 sctp_sn_error_t error) 802 { 803 struct sctp_ulpevent *event; 804 struct sockaddr_storage addr; 805 int spc_state = 0; 806 bool ulp_notify = true; 807 808 /* Record the transition on the transport. */ 809 switch (command) { 810 case SCTP_TRANSPORT_UP: 811 /* If we are moving from UNCONFIRMED state due 812 * to heartbeat success, report the SCTP_ADDR_CONFIRMED 813 * state to the user, otherwise report SCTP_ADDR_AVAILABLE. 814 */ 815 if (SCTP_UNCONFIRMED == transport->state && 816 SCTP_HEARTBEAT_SUCCESS == error) 817 spc_state = SCTP_ADDR_CONFIRMED; 818 else 819 spc_state = SCTP_ADDR_AVAILABLE; 820 /* Don't inform ULP about transition from PF to 821 * active state and set cwnd to 1 MTU, see SCTP 822 * Quick failover draft section 5.1, point 5 823 */ 824 if (transport->state == SCTP_PF) { 825 ulp_notify = false; 826 transport->cwnd = asoc->pathmtu; 827 } 828 transport->state = SCTP_ACTIVE; 829 break; 830 831 case SCTP_TRANSPORT_DOWN: 832 /* If the transport was never confirmed, do not transition it 833 * to inactive state. Also, release the cached route since 834 * there may be a better route next time. 835 */ 836 if (transport->state != SCTP_UNCONFIRMED) 837 transport->state = SCTP_INACTIVE; 838 else { 839 sctp_transport_dst_release(transport); 840 ulp_notify = false; 841 } 842 843 spc_state = SCTP_ADDR_UNREACHABLE; 844 break; 845 846 case SCTP_TRANSPORT_PF: 847 transport->state = SCTP_PF; 848 ulp_notify = false; 849 break; 850 851 default: 852 return; 853 } 854 855 /* Generate and send a SCTP_PEER_ADDR_CHANGE notification 856 * to the user. 857 */ 858 if (ulp_notify) { 859 memset(&addr, 0, sizeof(struct sockaddr_storage)); 860 memcpy(&addr, &transport->ipaddr, 861 transport->af_specific->sockaddr_len); 862 863 event = sctp_ulpevent_make_peer_addr_change(asoc, &addr, 864 0, spc_state, error, GFP_ATOMIC); 865 if (event) 866 sctp_ulpq_tail_event(&asoc->ulpq, event); 867 } 868 869 /* Select new active and retran paths. */ 870 sctp_select_active_and_retran_path(asoc); 871 } 872 873 /* Hold a reference to an association. */ 874 void sctp_association_hold(struct sctp_association *asoc) 875 { 876 refcount_inc(&asoc->base.refcnt); 877 } 878 879 /* Release a reference to an association and cleanup 880 * if there are no more references. 881 */ 882 void sctp_association_put(struct sctp_association *asoc) 883 { 884 if (refcount_dec_and_test(&asoc->base.refcnt)) 885 sctp_association_destroy(asoc); 886 } 887 888 /* Allocate the next TSN, Transmission Sequence Number, for the given 889 * association. 890 */ 891 __u32 sctp_association_get_next_tsn(struct sctp_association *asoc) 892 { 893 /* From Section 1.6 Serial Number Arithmetic: 894 * Transmission Sequence Numbers wrap around when they reach 895 * 2**32 - 1. That is, the next TSN a DATA chunk MUST use 896 * after transmitting TSN = 2*32 - 1 is TSN = 0. 897 */ 898 __u32 retval = asoc->next_tsn; 899 asoc->next_tsn++; 900 asoc->unack_data++; 901 902 return retval; 903 } 904 905 /* Compare two addresses to see if they match. Wildcard addresses 906 * only match themselves. 907 */ 908 int sctp_cmp_addr_exact(const union sctp_addr *ss1, 909 const union sctp_addr *ss2) 910 { 911 struct sctp_af *af; 912 913 af = sctp_get_af_specific(ss1->sa.sa_family); 914 if (unlikely(!af)) 915 return 0; 916 917 return af->cmp_addr(ss1, ss2); 918 } 919 920 /* Return an ecne chunk to get prepended to a packet. 921 * Note: We are sly and return a shared, prealloced chunk. FIXME: 922 * No we don't, but we could/should. 923 */ 924 struct sctp_chunk *sctp_get_ecne_prepend(struct sctp_association *asoc) 925 { 926 if (!asoc->need_ecne) 927 return NULL; 928 929 /* Send ECNE if needed. 930 * Not being able to allocate a chunk here is not deadly. 931 */ 932 return sctp_make_ecne(asoc, asoc->last_ecne_tsn); 933 } 934 935 /* 936 * Find which transport this TSN was sent on. 937 */ 938 struct sctp_transport *sctp_assoc_lookup_tsn(struct sctp_association *asoc, 939 __u32 tsn) 940 { 941 struct sctp_transport *active; 942 struct sctp_transport *match; 943 struct sctp_transport *transport; 944 struct sctp_chunk *chunk; 945 __be32 key = htonl(tsn); 946 947 match = NULL; 948 949 /* 950 * FIXME: In general, find a more efficient data structure for 951 * searching. 952 */ 953 954 /* 955 * The general strategy is to search each transport's transmitted 956 * list. Return which transport this TSN lives on. 957 * 958 * Let's be hopeful and check the active_path first. 959 * Another optimization would be to know if there is only one 960 * outbound path and not have to look for the TSN at all. 961 * 962 */ 963 964 active = asoc->peer.active_path; 965 966 list_for_each_entry(chunk, &active->transmitted, 967 transmitted_list) { 968 969 if (key == chunk->subh.data_hdr->tsn) { 970 match = active; 971 goto out; 972 } 973 } 974 975 /* If not found, go search all the other transports. */ 976 list_for_each_entry(transport, &asoc->peer.transport_addr_list, 977 transports) { 978 979 if (transport == active) 980 continue; 981 list_for_each_entry(chunk, &transport->transmitted, 982 transmitted_list) { 983 if (key == chunk->subh.data_hdr->tsn) { 984 match = transport; 985 goto out; 986 } 987 } 988 } 989 out: 990 return match; 991 } 992 993 /* Is this the association we are looking for? */ 994 struct sctp_transport *sctp_assoc_is_match(struct sctp_association *asoc, 995 struct net *net, 996 const union sctp_addr *laddr, 997 const union sctp_addr *paddr) 998 { 999 struct sctp_transport *transport; 1000 1001 if ((htons(asoc->base.bind_addr.port) == laddr->v4.sin_port) && 1002 (htons(asoc->peer.port) == paddr->v4.sin_port) && 1003 net_eq(sock_net(asoc->base.sk), net)) { 1004 transport = sctp_assoc_lookup_paddr(asoc, paddr); 1005 if (!transport) 1006 goto out; 1007 1008 if (sctp_bind_addr_match(&asoc->base.bind_addr, laddr, 1009 sctp_sk(asoc->base.sk))) 1010 goto out; 1011 } 1012 transport = NULL; 1013 1014 out: 1015 return transport; 1016 } 1017 1018 /* Do delayed input processing. This is scheduled by sctp_rcv(). */ 1019 static void sctp_assoc_bh_rcv(struct work_struct *work) 1020 { 1021 struct sctp_association *asoc = 1022 container_of(work, struct sctp_association, 1023 base.inqueue.immediate); 1024 struct net *net = sock_net(asoc->base.sk); 1025 struct sctp_endpoint *ep; 1026 struct sctp_chunk *chunk; 1027 struct sctp_inq *inqueue; 1028 int state; 1029 sctp_subtype_t subtype; 1030 int error = 0; 1031 1032 /* The association should be held so we should be safe. */ 1033 ep = asoc->ep; 1034 1035 inqueue = &asoc->base.inqueue; 1036 sctp_association_hold(asoc); 1037 while (NULL != (chunk = sctp_inq_pop(inqueue))) { 1038 state = asoc->state; 1039 subtype = SCTP_ST_CHUNK(chunk->chunk_hdr->type); 1040 1041 /* SCTP-AUTH, Section 6.3: 1042 * The receiver has a list of chunk types which it expects 1043 * to be received only after an AUTH-chunk. This list has 1044 * been sent to the peer during the association setup. It 1045 * MUST silently discard these chunks if they are not placed 1046 * after an AUTH chunk in the packet. 1047 */ 1048 if (sctp_auth_recv_cid(subtype.chunk, asoc) && !chunk->auth) 1049 continue; 1050 1051 /* Remember where the last DATA chunk came from so we 1052 * know where to send the SACK. 1053 */ 1054 if (sctp_chunk_is_data(chunk)) 1055 asoc->peer.last_data_from = chunk->transport; 1056 else { 1057 SCTP_INC_STATS(net, SCTP_MIB_INCTRLCHUNKS); 1058 asoc->stats.ictrlchunks++; 1059 if (chunk->chunk_hdr->type == SCTP_CID_SACK) 1060 asoc->stats.isacks++; 1061 } 1062 1063 if (chunk->transport) 1064 chunk->transport->last_time_heard = ktime_get(); 1065 1066 /* Run through the state machine. */ 1067 error = sctp_do_sm(net, SCTP_EVENT_T_CHUNK, subtype, 1068 state, ep, asoc, chunk, GFP_ATOMIC); 1069 1070 /* Check to see if the association is freed in response to 1071 * the incoming chunk. If so, get out of the while loop. 1072 */ 1073 if (asoc->base.dead) 1074 break; 1075 1076 /* If there is an error on chunk, discard this packet. */ 1077 if (error && chunk) 1078 chunk->pdiscard = 1; 1079 } 1080 sctp_association_put(asoc); 1081 } 1082 1083 /* This routine moves an association from its old sk to a new sk. */ 1084 void sctp_assoc_migrate(struct sctp_association *assoc, struct sock *newsk) 1085 { 1086 struct sctp_sock *newsp = sctp_sk(newsk); 1087 struct sock *oldsk = assoc->base.sk; 1088 1089 /* Delete the association from the old endpoint's list of 1090 * associations. 1091 */ 1092 list_del_init(&assoc->asocs); 1093 1094 /* Decrement the backlog value for a TCP-style socket. */ 1095 if (sctp_style(oldsk, TCP)) 1096 oldsk->sk_ack_backlog--; 1097 1098 /* Release references to the old endpoint and the sock. */ 1099 sctp_endpoint_put(assoc->ep); 1100 sock_put(assoc->base.sk); 1101 1102 /* Get a reference to the new endpoint. */ 1103 assoc->ep = newsp->ep; 1104 sctp_endpoint_hold(assoc->ep); 1105 1106 /* Get a reference to the new sock. */ 1107 assoc->base.sk = newsk; 1108 sock_hold(assoc->base.sk); 1109 1110 /* Add the association to the new endpoint's list of associations. */ 1111 sctp_endpoint_add_asoc(newsp->ep, assoc); 1112 } 1113 1114 /* Update an association (possibly from unexpected COOKIE-ECHO processing). */ 1115 int sctp_assoc_update(struct sctp_association *asoc, 1116 struct sctp_association *new) 1117 { 1118 struct sctp_transport *trans; 1119 struct list_head *pos, *temp; 1120 1121 /* Copy in new parameters of peer. */ 1122 asoc->c = new->c; 1123 asoc->peer.rwnd = new->peer.rwnd; 1124 asoc->peer.sack_needed = new->peer.sack_needed; 1125 asoc->peer.auth_capable = new->peer.auth_capable; 1126 asoc->peer.i = new->peer.i; 1127 1128 if (!sctp_tsnmap_init(&asoc->peer.tsn_map, SCTP_TSN_MAP_INITIAL, 1129 asoc->peer.i.initial_tsn, GFP_ATOMIC)) 1130 return -ENOMEM; 1131 1132 /* Remove any peer addresses not present in the new association. */ 1133 list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) { 1134 trans = list_entry(pos, struct sctp_transport, transports); 1135 if (!sctp_assoc_lookup_paddr(new, &trans->ipaddr)) { 1136 sctp_assoc_rm_peer(asoc, trans); 1137 continue; 1138 } 1139 1140 if (asoc->state >= SCTP_STATE_ESTABLISHED) 1141 sctp_transport_reset(trans); 1142 } 1143 1144 /* If the case is A (association restart), use 1145 * initial_tsn as next_tsn. If the case is B, use 1146 * current next_tsn in case data sent to peer 1147 * has been discarded and needs retransmission. 1148 */ 1149 if (asoc->state >= SCTP_STATE_ESTABLISHED) { 1150 asoc->next_tsn = new->next_tsn; 1151 asoc->ctsn_ack_point = new->ctsn_ack_point; 1152 asoc->adv_peer_ack_point = new->adv_peer_ack_point; 1153 1154 /* Reinitialize SSN for both local streams 1155 * and peer's streams. 1156 */ 1157 sctp_stream_clear(&asoc->stream); 1158 1159 /* Flush the ULP reassembly and ordered queue. 1160 * Any data there will now be stale and will 1161 * cause problems. 1162 */ 1163 sctp_ulpq_flush(&asoc->ulpq); 1164 1165 /* reset the overall association error count so 1166 * that the restarted association doesn't get torn 1167 * down on the next retransmission timer. 1168 */ 1169 asoc->overall_error_count = 0; 1170 1171 } else { 1172 /* Add any peer addresses from the new association. */ 1173 list_for_each_entry(trans, &new->peer.transport_addr_list, 1174 transports) 1175 if (!sctp_assoc_lookup_paddr(asoc, &trans->ipaddr) && 1176 !sctp_assoc_add_peer(asoc, &trans->ipaddr, 1177 GFP_ATOMIC, trans->state)) 1178 return -ENOMEM; 1179 1180 asoc->ctsn_ack_point = asoc->next_tsn - 1; 1181 asoc->adv_peer_ack_point = asoc->ctsn_ack_point; 1182 1183 if (sctp_state(asoc, COOKIE_WAIT)) 1184 sctp_stream_update(&asoc->stream, &new->stream); 1185 1186 /* get a new assoc id if we don't have one yet. */ 1187 if (sctp_assoc_set_id(asoc, GFP_ATOMIC)) 1188 return -ENOMEM; 1189 } 1190 1191 /* SCTP-AUTH: Save the peer parameters from the new associations 1192 * and also move the association shared keys over 1193 */ 1194 kfree(asoc->peer.peer_random); 1195 asoc->peer.peer_random = new->peer.peer_random; 1196 new->peer.peer_random = NULL; 1197 1198 kfree(asoc->peer.peer_chunks); 1199 asoc->peer.peer_chunks = new->peer.peer_chunks; 1200 new->peer.peer_chunks = NULL; 1201 1202 kfree(asoc->peer.peer_hmacs); 1203 asoc->peer.peer_hmacs = new->peer.peer_hmacs; 1204 new->peer.peer_hmacs = NULL; 1205 1206 return sctp_auth_asoc_init_active_key(asoc, GFP_ATOMIC); 1207 } 1208 1209 /* Update the retran path for sending a retransmitted packet. 1210 * See also RFC4960, 6.4. Multi-Homed SCTP Endpoints: 1211 * 1212 * When there is outbound data to send and the primary path 1213 * becomes inactive (e.g., due to failures), or where the 1214 * SCTP user explicitly requests to send data to an 1215 * inactive destination transport address, before reporting 1216 * an error to its ULP, the SCTP endpoint should try to send 1217 * the data to an alternate active destination transport 1218 * address if one exists. 1219 * 1220 * When retransmitting data that timed out, if the endpoint 1221 * is multihomed, it should consider each source-destination 1222 * address pair in its retransmission selection policy. 1223 * When retransmitting timed-out data, the endpoint should 1224 * attempt to pick the most divergent source-destination 1225 * pair from the original source-destination pair to which 1226 * the packet was transmitted. 1227 * 1228 * Note: Rules for picking the most divergent source-destination 1229 * pair are an implementation decision and are not specified 1230 * within this document. 1231 * 1232 * Our basic strategy is to round-robin transports in priorities 1233 * according to sctp_trans_score() e.g., if no such 1234 * transport with state SCTP_ACTIVE exists, round-robin through 1235 * SCTP_UNKNOWN, etc. You get the picture. 1236 */ 1237 static u8 sctp_trans_score(const struct sctp_transport *trans) 1238 { 1239 switch (trans->state) { 1240 case SCTP_ACTIVE: 1241 return 3; /* best case */ 1242 case SCTP_UNKNOWN: 1243 return 2; 1244 case SCTP_PF: 1245 return 1; 1246 default: /* case SCTP_INACTIVE */ 1247 return 0; /* worst case */ 1248 } 1249 } 1250 1251 static struct sctp_transport *sctp_trans_elect_tie(struct sctp_transport *trans1, 1252 struct sctp_transport *trans2) 1253 { 1254 if (trans1->error_count > trans2->error_count) { 1255 return trans2; 1256 } else if (trans1->error_count == trans2->error_count && 1257 ktime_after(trans2->last_time_heard, 1258 trans1->last_time_heard)) { 1259 return trans2; 1260 } else { 1261 return trans1; 1262 } 1263 } 1264 1265 static struct sctp_transport *sctp_trans_elect_best(struct sctp_transport *curr, 1266 struct sctp_transport *best) 1267 { 1268 u8 score_curr, score_best; 1269 1270 if (best == NULL || curr == best) 1271 return curr; 1272 1273 score_curr = sctp_trans_score(curr); 1274 score_best = sctp_trans_score(best); 1275 1276 /* First, try a score-based selection if both transport states 1277 * differ. If we're in a tie, lets try to make a more clever 1278 * decision here based on error counts and last time heard. 1279 */ 1280 if (score_curr > score_best) 1281 return curr; 1282 else if (score_curr == score_best) 1283 return sctp_trans_elect_tie(best, curr); 1284 else 1285 return best; 1286 } 1287 1288 void sctp_assoc_update_retran_path(struct sctp_association *asoc) 1289 { 1290 struct sctp_transport *trans = asoc->peer.retran_path; 1291 struct sctp_transport *trans_next = NULL; 1292 1293 /* We're done as we only have the one and only path. */ 1294 if (asoc->peer.transport_count == 1) 1295 return; 1296 /* If active_path and retran_path are the same and active, 1297 * then this is the only active path. Use it. 1298 */ 1299 if (asoc->peer.active_path == asoc->peer.retran_path && 1300 asoc->peer.active_path->state == SCTP_ACTIVE) 1301 return; 1302 1303 /* Iterate from retran_path's successor back to retran_path. */ 1304 for (trans = list_next_entry(trans, transports); 1; 1305 trans = list_next_entry(trans, transports)) { 1306 /* Manually skip the head element. */ 1307 if (&trans->transports == &asoc->peer.transport_addr_list) 1308 continue; 1309 if (trans->state == SCTP_UNCONFIRMED) 1310 continue; 1311 trans_next = sctp_trans_elect_best(trans, trans_next); 1312 /* Active is good enough for immediate return. */ 1313 if (trans_next->state == SCTP_ACTIVE) 1314 break; 1315 /* We've reached the end, time to update path. */ 1316 if (trans == asoc->peer.retran_path) 1317 break; 1318 } 1319 1320 asoc->peer.retran_path = trans_next; 1321 1322 pr_debug("%s: association:%p updated new path to addr:%pISpc\n", 1323 __func__, asoc, &asoc->peer.retran_path->ipaddr.sa); 1324 } 1325 1326 static void sctp_select_active_and_retran_path(struct sctp_association *asoc) 1327 { 1328 struct sctp_transport *trans, *trans_pri = NULL, *trans_sec = NULL; 1329 struct sctp_transport *trans_pf = NULL; 1330 1331 /* Look for the two most recently used active transports. */ 1332 list_for_each_entry(trans, &asoc->peer.transport_addr_list, 1333 transports) { 1334 /* Skip uninteresting transports. */ 1335 if (trans->state == SCTP_INACTIVE || 1336 trans->state == SCTP_UNCONFIRMED) 1337 continue; 1338 /* Keep track of the best PF transport from our 1339 * list in case we don't find an active one. 1340 */ 1341 if (trans->state == SCTP_PF) { 1342 trans_pf = sctp_trans_elect_best(trans, trans_pf); 1343 continue; 1344 } 1345 /* For active transports, pick the most recent ones. */ 1346 if (trans_pri == NULL || 1347 ktime_after(trans->last_time_heard, 1348 trans_pri->last_time_heard)) { 1349 trans_sec = trans_pri; 1350 trans_pri = trans; 1351 } else if (trans_sec == NULL || 1352 ktime_after(trans->last_time_heard, 1353 trans_sec->last_time_heard)) { 1354 trans_sec = trans; 1355 } 1356 } 1357 1358 /* RFC 2960 6.4 Multi-Homed SCTP Endpoints 1359 * 1360 * By default, an endpoint should always transmit to the primary 1361 * path, unless the SCTP user explicitly specifies the 1362 * destination transport address (and possibly source transport 1363 * address) to use. [If the primary is active but not most recent, 1364 * bump the most recently used transport.] 1365 */ 1366 if ((asoc->peer.primary_path->state == SCTP_ACTIVE || 1367 asoc->peer.primary_path->state == SCTP_UNKNOWN) && 1368 asoc->peer.primary_path != trans_pri) { 1369 trans_sec = trans_pri; 1370 trans_pri = asoc->peer.primary_path; 1371 } 1372 1373 /* We did not find anything useful for a possible retransmission 1374 * path; either primary path that we found is the the same as 1375 * the current one, or we didn't generally find an active one. 1376 */ 1377 if (trans_sec == NULL) 1378 trans_sec = trans_pri; 1379 1380 /* If we failed to find a usable transport, just camp on the 1381 * active or pick a PF iff it's the better choice. 1382 */ 1383 if (trans_pri == NULL) { 1384 trans_pri = sctp_trans_elect_best(asoc->peer.active_path, trans_pf); 1385 trans_sec = trans_pri; 1386 } 1387 1388 /* Set the active and retran transports. */ 1389 asoc->peer.active_path = trans_pri; 1390 asoc->peer.retran_path = trans_sec; 1391 } 1392 1393 struct sctp_transport * 1394 sctp_assoc_choose_alter_transport(struct sctp_association *asoc, 1395 struct sctp_transport *last_sent_to) 1396 { 1397 /* If this is the first time packet is sent, use the active path, 1398 * else use the retran path. If the last packet was sent over the 1399 * retran path, update the retran path and use it. 1400 */ 1401 if (last_sent_to == NULL) { 1402 return asoc->peer.active_path; 1403 } else { 1404 if (last_sent_to == asoc->peer.retran_path) 1405 sctp_assoc_update_retran_path(asoc); 1406 1407 return asoc->peer.retran_path; 1408 } 1409 } 1410 1411 /* Update the association's pmtu and frag_point by going through all the 1412 * transports. This routine is called when a transport's PMTU has changed. 1413 */ 1414 void sctp_assoc_sync_pmtu(struct sctp_association *asoc) 1415 { 1416 struct sctp_transport *t; 1417 __u32 pmtu = 0; 1418 1419 if (!asoc) 1420 return; 1421 1422 /* Get the lowest pmtu of all the transports. */ 1423 list_for_each_entry(t, &asoc->peer.transport_addr_list, 1424 transports) { 1425 if (t->pmtu_pending && t->dst) { 1426 sctp_transport_update_pmtu( 1427 t, SCTP_TRUNC4(dst_mtu(t->dst))); 1428 t->pmtu_pending = 0; 1429 } 1430 if (!pmtu || (t->pathmtu < pmtu)) 1431 pmtu = t->pathmtu; 1432 } 1433 1434 if (pmtu) { 1435 asoc->pathmtu = pmtu; 1436 asoc->frag_point = sctp_frag_point(asoc, pmtu); 1437 } 1438 1439 pr_debug("%s: asoc:%p, pmtu:%d, frag_point:%d\n", __func__, asoc, 1440 asoc->pathmtu, asoc->frag_point); 1441 } 1442 1443 /* Should we send a SACK to update our peer? */ 1444 static inline bool sctp_peer_needs_update(struct sctp_association *asoc) 1445 { 1446 struct net *net = sock_net(asoc->base.sk); 1447 switch (asoc->state) { 1448 case SCTP_STATE_ESTABLISHED: 1449 case SCTP_STATE_SHUTDOWN_PENDING: 1450 case SCTP_STATE_SHUTDOWN_RECEIVED: 1451 case SCTP_STATE_SHUTDOWN_SENT: 1452 if ((asoc->rwnd > asoc->a_rwnd) && 1453 ((asoc->rwnd - asoc->a_rwnd) >= max_t(__u32, 1454 (asoc->base.sk->sk_rcvbuf >> net->sctp.rwnd_upd_shift), 1455 asoc->pathmtu))) 1456 return true; 1457 break; 1458 default: 1459 break; 1460 } 1461 return false; 1462 } 1463 1464 /* Increase asoc's rwnd by len and send any window update SACK if needed. */ 1465 void sctp_assoc_rwnd_increase(struct sctp_association *asoc, unsigned int len) 1466 { 1467 struct sctp_chunk *sack; 1468 struct timer_list *timer; 1469 1470 if (asoc->rwnd_over) { 1471 if (asoc->rwnd_over >= len) { 1472 asoc->rwnd_over -= len; 1473 } else { 1474 asoc->rwnd += (len - asoc->rwnd_over); 1475 asoc->rwnd_over = 0; 1476 } 1477 } else { 1478 asoc->rwnd += len; 1479 } 1480 1481 /* If we had window pressure, start recovering it 1482 * once our rwnd had reached the accumulated pressure 1483 * threshold. The idea is to recover slowly, but up 1484 * to the initial advertised window. 1485 */ 1486 if (asoc->rwnd_press) { 1487 int change = min(asoc->pathmtu, asoc->rwnd_press); 1488 asoc->rwnd += change; 1489 asoc->rwnd_press -= change; 1490 } 1491 1492 pr_debug("%s: asoc:%p rwnd increased by %d to (%u, %u) - %u\n", 1493 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over, 1494 asoc->a_rwnd); 1495 1496 /* Send a window update SACK if the rwnd has increased by at least the 1497 * minimum of the association's PMTU and half of the receive buffer. 1498 * The algorithm used is similar to the one described in 1499 * Section 4.2.3.3 of RFC 1122. 1500 */ 1501 if (sctp_peer_needs_update(asoc)) { 1502 asoc->a_rwnd = asoc->rwnd; 1503 1504 pr_debug("%s: sending window update SACK- asoc:%p rwnd:%u " 1505 "a_rwnd:%u\n", __func__, asoc, asoc->rwnd, 1506 asoc->a_rwnd); 1507 1508 sack = sctp_make_sack(asoc); 1509 if (!sack) 1510 return; 1511 1512 asoc->peer.sack_needed = 0; 1513 1514 sctp_outq_tail(&asoc->outqueue, sack, GFP_ATOMIC); 1515 1516 /* Stop the SACK timer. */ 1517 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK]; 1518 if (del_timer(timer)) 1519 sctp_association_put(asoc); 1520 } 1521 } 1522 1523 /* Decrease asoc's rwnd by len. */ 1524 void sctp_assoc_rwnd_decrease(struct sctp_association *asoc, unsigned int len) 1525 { 1526 int rx_count; 1527 int over = 0; 1528 1529 if (unlikely(!asoc->rwnd || asoc->rwnd_over)) 1530 pr_debug("%s: association:%p has asoc->rwnd:%u, " 1531 "asoc->rwnd_over:%u!\n", __func__, asoc, 1532 asoc->rwnd, asoc->rwnd_over); 1533 1534 if (asoc->ep->rcvbuf_policy) 1535 rx_count = atomic_read(&asoc->rmem_alloc); 1536 else 1537 rx_count = atomic_read(&asoc->base.sk->sk_rmem_alloc); 1538 1539 /* If we've reached or overflowed our receive buffer, announce 1540 * a 0 rwnd if rwnd would still be positive. Store the 1541 * the potential pressure overflow so that the window can be restored 1542 * back to original value. 1543 */ 1544 if (rx_count >= asoc->base.sk->sk_rcvbuf) 1545 over = 1; 1546 1547 if (asoc->rwnd >= len) { 1548 asoc->rwnd -= len; 1549 if (over) { 1550 asoc->rwnd_press += asoc->rwnd; 1551 asoc->rwnd = 0; 1552 } 1553 } else { 1554 asoc->rwnd_over += len - asoc->rwnd; 1555 asoc->rwnd = 0; 1556 } 1557 1558 pr_debug("%s: asoc:%p rwnd decreased by %d to (%u, %u, %u)\n", 1559 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over, 1560 asoc->rwnd_press); 1561 } 1562 1563 /* Build the bind address list for the association based on info from the 1564 * local endpoint and the remote peer. 1565 */ 1566 int sctp_assoc_set_bind_addr_from_ep(struct sctp_association *asoc, 1567 sctp_scope_t scope, gfp_t gfp) 1568 { 1569 int flags; 1570 1571 /* Use scoping rules to determine the subset of addresses from 1572 * the endpoint. 1573 */ 1574 flags = (PF_INET6 == asoc->base.sk->sk_family) ? SCTP_ADDR6_ALLOWED : 0; 1575 if (asoc->peer.ipv4_address) 1576 flags |= SCTP_ADDR4_PEERSUPP; 1577 if (asoc->peer.ipv6_address) 1578 flags |= SCTP_ADDR6_PEERSUPP; 1579 1580 return sctp_bind_addr_copy(sock_net(asoc->base.sk), 1581 &asoc->base.bind_addr, 1582 &asoc->ep->base.bind_addr, 1583 scope, gfp, flags); 1584 } 1585 1586 /* Build the association's bind address list from the cookie. */ 1587 int sctp_assoc_set_bind_addr_from_cookie(struct sctp_association *asoc, 1588 struct sctp_cookie *cookie, 1589 gfp_t gfp) 1590 { 1591 int var_size2 = ntohs(cookie->peer_init->chunk_hdr.length); 1592 int var_size3 = cookie->raw_addr_list_len; 1593 __u8 *raw = (__u8 *)cookie->peer_init + var_size2; 1594 1595 return sctp_raw_to_bind_addrs(&asoc->base.bind_addr, raw, var_size3, 1596 asoc->ep->base.bind_addr.port, gfp); 1597 } 1598 1599 /* Lookup laddr in the bind address list of an association. */ 1600 int sctp_assoc_lookup_laddr(struct sctp_association *asoc, 1601 const union sctp_addr *laddr) 1602 { 1603 int found = 0; 1604 1605 if ((asoc->base.bind_addr.port == ntohs(laddr->v4.sin_port)) && 1606 sctp_bind_addr_match(&asoc->base.bind_addr, laddr, 1607 sctp_sk(asoc->base.sk))) 1608 found = 1; 1609 1610 return found; 1611 } 1612 1613 /* Set an association id for a given association */ 1614 int sctp_assoc_set_id(struct sctp_association *asoc, gfp_t gfp) 1615 { 1616 bool preload = gfpflags_allow_blocking(gfp); 1617 int ret; 1618 1619 /* If the id is already assigned, keep it. */ 1620 if (asoc->assoc_id) 1621 return 0; 1622 1623 if (preload) 1624 idr_preload(gfp); 1625 spin_lock_bh(&sctp_assocs_id_lock); 1626 /* 0 is not a valid assoc_id, must be >= 1 */ 1627 ret = idr_alloc_cyclic(&sctp_assocs_id, asoc, 1, 0, GFP_NOWAIT); 1628 spin_unlock_bh(&sctp_assocs_id_lock); 1629 if (preload) 1630 idr_preload_end(); 1631 if (ret < 0) 1632 return ret; 1633 1634 asoc->assoc_id = (sctp_assoc_t)ret; 1635 return 0; 1636 } 1637 1638 /* Free the ASCONF queue */ 1639 static void sctp_assoc_free_asconf_queue(struct sctp_association *asoc) 1640 { 1641 struct sctp_chunk *asconf; 1642 struct sctp_chunk *tmp; 1643 1644 list_for_each_entry_safe(asconf, tmp, &asoc->addip_chunk_list, list) { 1645 list_del_init(&asconf->list); 1646 sctp_chunk_free(asconf); 1647 } 1648 } 1649 1650 /* Free asconf_ack cache */ 1651 static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc) 1652 { 1653 struct sctp_chunk *ack; 1654 struct sctp_chunk *tmp; 1655 1656 list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list, 1657 transmitted_list) { 1658 list_del_init(&ack->transmitted_list); 1659 sctp_chunk_free(ack); 1660 } 1661 } 1662 1663 /* Clean up the ASCONF_ACK queue */ 1664 void sctp_assoc_clean_asconf_ack_cache(const struct sctp_association *asoc) 1665 { 1666 struct sctp_chunk *ack; 1667 struct sctp_chunk *tmp; 1668 1669 /* We can remove all the entries from the queue up to 1670 * the "Peer-Sequence-Number". 1671 */ 1672 list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list, 1673 transmitted_list) { 1674 if (ack->subh.addip_hdr->serial == 1675 htonl(asoc->peer.addip_serial)) 1676 break; 1677 1678 list_del_init(&ack->transmitted_list); 1679 sctp_chunk_free(ack); 1680 } 1681 } 1682 1683 /* Find the ASCONF_ACK whose serial number matches ASCONF */ 1684 struct sctp_chunk *sctp_assoc_lookup_asconf_ack( 1685 const struct sctp_association *asoc, 1686 __be32 serial) 1687 { 1688 struct sctp_chunk *ack; 1689 1690 /* Walk through the list of cached ASCONF-ACKs and find the 1691 * ack chunk whose serial number matches that of the request. 1692 */ 1693 list_for_each_entry(ack, &asoc->asconf_ack_list, transmitted_list) { 1694 if (sctp_chunk_pending(ack)) 1695 continue; 1696 if (ack->subh.addip_hdr->serial == serial) { 1697 sctp_chunk_hold(ack); 1698 return ack; 1699 } 1700 } 1701 1702 return NULL; 1703 } 1704 1705 void sctp_asconf_queue_teardown(struct sctp_association *asoc) 1706 { 1707 /* Free any cached ASCONF_ACK chunk. */ 1708 sctp_assoc_free_asconf_acks(asoc); 1709 1710 /* Free the ASCONF queue. */ 1711 sctp_assoc_free_asconf_queue(asoc); 1712 1713 /* Free any cached ASCONF chunk. */ 1714 if (asoc->addip_last_asconf) 1715 sctp_chunk_free(asoc->addip_last_asconf); 1716 } 1717