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