1/* SCTP kernel implementation 2 * Copyright (c) 1999-2000 Cisco, Inc. 3 * Copyright (c) 1999-2001 Motorola, Inc. 4 * Copyright (c) 2001-2003 International Business Machines Corp. 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 tranport representing 11 * a remote transport address. For local transport addresses, we just use 12 * union sctp_addr. 13 * 14 * This SCTP implementation is free software; 15 * you can redistribute it and/or modify it under the terms of 16 * the GNU General Public License as published by 17 * the Free Software Foundation; either version 2, or (at your option) 18 * any later version. 19 * 20 * This SCTP implementation is distributed in the hope that it 21 * will be useful, but WITHOUT ANY WARRANTY; without even the implied 22 * ************************ 23 * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. 24 * See the GNU General Public License for more details. 25 * 26 * You should have received a copy of the GNU General Public License 27 * along with GNU CC; see the file COPYING. If not, write to 28 * the Free Software Foundation, 59 Temple Place - Suite 330, 29 * Boston, MA 02111-1307, USA. 30 * 31 * Please send any bug reports or fixes you make to the 32 * email address(es): 33 * lksctp developers <lksctp-developers@lists.sourceforge.net> 34 * 35 * Or submit a bug report through the following website: 36 * http://www.sf.net/projects/lksctp 37 * 38 * Written or modified by: 39 * La Monte H.P. Yarroll <piggy@acm.org> 40 * Karl Knutson <karl@athena.chicago.il.us> 41 * Jon Grimm <jgrimm@us.ibm.com> 42 * Xingang Guo <xingang.guo@intel.com> 43 * Hui Huang <hui.huang@nokia.com> 44 * Sridhar Samudrala <sri@us.ibm.com> 45 * Ardelle Fan <ardelle.fan@intel.com> 46 * 47 * Any bugs reported given to us we will try to fix... any fixes shared will 48 * be incorporated into the next SCTP release. 49 */ 50 51#include <linux/slab.h> 52#include <linux/types.h> 53#include <linux/random.h> 54#include <net/sctp/sctp.h> 55#include <net/sctp/sm.h> 56 57/* 1st Level Abstractions. */ 58 59/* Initialize a new transport from provided memory. */ 60static struct sctp_transport *sctp_transport_init(struct sctp_transport *peer, 61 const union sctp_addr *addr, 62 gfp_t gfp) 63{ 64 /* Copy in the address. */ 65 peer->ipaddr = *addr; 66 peer->af_specific = sctp_get_af_specific(addr->sa.sa_family); 67 memset(&peer->saddr, 0, sizeof(union sctp_addr)); 68 69 /* From 6.3.1 RTO Calculation: 70 * 71 * C1) Until an RTT measurement has been made for a packet sent to the 72 * given destination transport address, set RTO to the protocol 73 * parameter 'RTO.Initial'. 74 */ 75 peer->rto = msecs_to_jiffies(sctp_rto_initial); 76 77 peer->last_time_heard = jiffies; 78 peer->last_time_ecne_reduced = jiffies; 79 80 peer->param_flags = SPP_HB_DISABLE | 81 SPP_PMTUD_ENABLE | 82 SPP_SACKDELAY_ENABLE; 83 84 /* Initialize the default path max_retrans. */ 85 peer->pathmaxrxt = sctp_max_retrans_path; 86 87 INIT_LIST_HEAD(&peer->transmitted); 88 INIT_LIST_HEAD(&peer->send_ready); 89 INIT_LIST_HEAD(&peer->transports); 90 91 setup_timer(&peer->T3_rtx_timer, sctp_generate_t3_rtx_event, 92 (unsigned long)peer); 93 setup_timer(&peer->hb_timer, sctp_generate_heartbeat_event, 94 (unsigned long)peer); 95 setup_timer(&peer->proto_unreach_timer, 96 sctp_generate_proto_unreach_event, (unsigned long)peer); 97 98 /* Initialize the 64-bit random nonce sent with heartbeat. */ 99 get_random_bytes(&peer->hb_nonce, sizeof(peer->hb_nonce)); 100 101 atomic_set(&peer->refcnt, 1); 102 103 return peer; 104} 105 106/* Allocate and initialize a new transport. */ 107struct sctp_transport *sctp_transport_new(const union sctp_addr *addr, 108 gfp_t gfp) 109{ 110 struct sctp_transport *transport; 111 112 transport = t_new(struct sctp_transport, gfp); 113 if (!transport) 114 goto fail; 115 116 if (!sctp_transport_init(transport, addr, gfp)) 117 goto fail_init; 118 119 transport->malloced = 1; 120 SCTP_DBG_OBJCNT_INC(transport); 121 122 return transport; 123 124fail_init: 125 kfree(transport); 126 127fail: 128 return NULL; 129} 130 131/* This transport is no longer needed. Free up if possible, or 132 * delay until it last reference count. 133 */ 134void sctp_transport_free(struct sctp_transport *transport) 135{ 136 transport->dead = 1; 137 138 /* Try to delete the heartbeat timer. */ 139 if (del_timer(&transport->hb_timer)) 140 sctp_transport_put(transport); 141 142 /* Delete the T3_rtx timer if it's active. 143 * There is no point in not doing this now and letting 144 * structure hang around in memory since we know 145 * the tranport is going away. 146 */ 147 if (timer_pending(&transport->T3_rtx_timer) && 148 del_timer(&transport->T3_rtx_timer)) 149 sctp_transport_put(transport); 150 151 /* Delete the ICMP proto unreachable timer if it's active. */ 152 if (timer_pending(&transport->proto_unreach_timer) && 153 del_timer(&transport->proto_unreach_timer)) 154 sctp_association_put(transport->asoc); 155 156 sctp_transport_put(transport); 157} 158 159/* Destroy the transport data structure. 160 * Assumes there are no more users of this structure. 161 */ 162static void sctp_transport_destroy(struct sctp_transport *transport) 163{ 164 SCTP_ASSERT(transport->dead, "Transport is not dead", return); 165 166 if (transport->asoc) 167 sctp_association_put(transport->asoc); 168 169 sctp_packet_free(&transport->packet); 170 171 dst_release(transport->dst); 172 kfree(transport); 173 SCTP_DBG_OBJCNT_DEC(transport); 174} 175 176/* Start T3_rtx timer if it is not already running and update the heartbeat 177 * timer. This routine is called every time a DATA chunk is sent. 178 */ 179void sctp_transport_reset_timers(struct sctp_transport *transport) 180{ 181 /* RFC 2960 6.3.2 Retransmission Timer Rules 182 * 183 * R1) Every time a DATA chunk is sent to any address(including a 184 * retransmission), if the T3-rtx timer of that address is not running 185 * start it running so that it will expire after the RTO of that 186 * address. 187 */ 188 189 if (!timer_pending(&transport->T3_rtx_timer)) 190 if (!mod_timer(&transport->T3_rtx_timer, 191 jiffies + transport->rto)) 192 sctp_transport_hold(transport); 193 194 /* When a data chunk is sent, reset the heartbeat interval. */ 195 if (!mod_timer(&transport->hb_timer, 196 sctp_transport_timeout(transport))) 197 sctp_transport_hold(transport); 198} 199 200/* This transport has been assigned to an association. 201 * Initialize fields from the association or from the sock itself. 202 * Register the reference count in the association. 203 */ 204void sctp_transport_set_owner(struct sctp_transport *transport, 205 struct sctp_association *asoc) 206{ 207 transport->asoc = asoc; 208 sctp_association_hold(asoc); 209} 210 211/* Initialize the pmtu of a transport. */ 212void sctp_transport_pmtu(struct sctp_transport *transport) 213{ 214 struct dst_entry *dst; 215 216 dst = transport->af_specific->get_dst(NULL, &transport->ipaddr, NULL); 217 218 if (dst) { 219 transport->pathmtu = dst_mtu(dst); 220 dst_release(dst); 221 } else 222 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT; 223} 224 225/* this is a complete rip-off from __sk_dst_check 226 * the cookie is always 0 since this is how it's used in the 227 * pmtu code 228 */ 229static struct dst_entry *sctp_transport_dst_check(struct sctp_transport *t) 230{ 231 struct dst_entry *dst = t->dst; 232 233 if (dst && dst->obsolete && dst->ops->check(dst, 0) == NULL) { 234 dst_release(t->dst); 235 t->dst = NULL; 236 return NULL; 237 } 238 239 return dst; 240} 241 242void sctp_transport_update_pmtu(struct sctp_transport *t, u32 pmtu) 243{ 244 struct dst_entry *dst; 245 246 if (unlikely(pmtu < SCTP_DEFAULT_MINSEGMENT)) { 247 printk(KERN_WARNING "%s: Reported pmtu %d too low, " 248 "using default minimum of %d\n", 249 __func__, pmtu, 250 SCTP_DEFAULT_MINSEGMENT); 251 /* Use default minimum segment size and disable 252 * pmtu discovery on this transport. 253 */ 254 t->pathmtu = SCTP_DEFAULT_MINSEGMENT; 255 } else { 256 t->pathmtu = pmtu; 257 } 258 259 dst = sctp_transport_dst_check(t); 260 if (dst) 261 dst->ops->update_pmtu(dst, pmtu); 262} 263 264/* Caches the dst entry and source address for a transport's destination 265 * address. 266 */ 267void sctp_transport_route(struct sctp_transport *transport, 268 union sctp_addr *saddr, struct sctp_sock *opt) 269{ 270 struct sctp_association *asoc = transport->asoc; 271 struct sctp_af *af = transport->af_specific; 272 union sctp_addr *daddr = &transport->ipaddr; 273 struct dst_entry *dst; 274 275 dst = af->get_dst(asoc, daddr, saddr); 276 277 if (saddr) 278 memcpy(&transport->saddr, saddr, sizeof(union sctp_addr)); 279 else 280 af->get_saddr(opt, asoc, dst, daddr, &transport->saddr); 281 282 transport->dst = dst; 283 if ((transport->param_flags & SPP_PMTUD_DISABLE) && transport->pathmtu) { 284 return; 285 } 286 if (dst) { 287 transport->pathmtu = dst_mtu(dst); 288 289 /* Initialize sk->sk_rcv_saddr, if the transport is the 290 * association's active path for getsockname(). 291 */ 292 if (asoc && (!asoc->peer.primary_path || 293 (transport == asoc->peer.active_path))) 294 opt->pf->af->to_sk_saddr(&transport->saddr, 295 asoc->base.sk); 296 } else 297 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT; 298} 299 300/* Hold a reference to a transport. */ 301void sctp_transport_hold(struct sctp_transport *transport) 302{ 303 atomic_inc(&transport->refcnt); 304} 305 306/* Release a reference to a transport and clean up 307 * if there are no more references. 308 */ 309void sctp_transport_put(struct sctp_transport *transport) 310{ 311 if (atomic_dec_and_test(&transport->refcnt)) 312 sctp_transport_destroy(transport); 313} 314 315/* Update transport's RTO based on the newly calculated RTT. */ 316void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt) 317{ 318 /* Check for valid transport. */ 319 SCTP_ASSERT(tp, "NULL transport", return); 320 321 /* We should not be doing any RTO updates unless rto_pending is set. */ 322 SCTP_ASSERT(tp->rto_pending, "rto_pending not set", return); 323 324 if (tp->rttvar || tp->srtt) { 325 /* 6.3.1 C3) When a new RTT measurement R' is made, set 326 * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'| 327 * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R' 328 */ 329 330 /* Note: The above algorithm has been rewritten to 331 * express rto_beta and rto_alpha as inverse powers 332 * of two. 333 * For example, assuming the default value of RTO.Alpha of 334 * 1/8, rto_alpha would be expressed as 3. 335 */ 336 tp->rttvar = tp->rttvar - (tp->rttvar >> sctp_rto_beta) 337 + ((abs(tp->srtt - rtt)) >> sctp_rto_beta); 338 tp->srtt = tp->srtt - (tp->srtt >> sctp_rto_alpha) 339 + (rtt >> sctp_rto_alpha); 340 } else { 341 /* 6.3.1 C2) When the first RTT measurement R is made, set 342 * SRTT <- R, RTTVAR <- R/2. 343 */ 344 tp->srtt = rtt; 345 tp->rttvar = rtt >> 1; 346 } 347 348 /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then 349 * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY. 350 */ 351 if (tp->rttvar == 0) 352 tp->rttvar = SCTP_CLOCK_GRANULARITY; 353 354 /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */ 355 tp->rto = tp->srtt + (tp->rttvar << 2); 356 357 /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min 358 * seconds then it is rounded up to RTO.Min seconds. 359 */ 360 if (tp->rto < tp->asoc->rto_min) 361 tp->rto = tp->asoc->rto_min; 362 363 /* 6.3.1 C7) A maximum value may be placed on RTO provided it is 364 * at least RTO.max seconds. 365 */ 366 if (tp->rto > tp->asoc->rto_max) 367 tp->rto = tp->asoc->rto_max; 368 369 tp->rtt = rtt; 370 371 /* Reset rto_pending so that a new RTT measurement is started when a 372 * new data chunk is sent. 373 */ 374 tp->rto_pending = 0; 375 376 SCTP_DEBUG_PRINTK("%s: transport: %p, rtt: %d, srtt: %d " 377 "rttvar: %d, rto: %ld\n", __func__, 378 tp, rtt, tp->srtt, tp->rttvar, tp->rto); 379} 380 381/* This routine updates the transport's cwnd and partial_bytes_acked 382 * parameters based on the bytes acked in the received SACK. 383 */ 384void sctp_transport_raise_cwnd(struct sctp_transport *transport, 385 __u32 sack_ctsn, __u32 bytes_acked) 386{ 387 struct sctp_association *asoc = transport->asoc; 388 __u32 cwnd, ssthresh, flight_size, pba, pmtu; 389 390 cwnd = transport->cwnd; 391 flight_size = transport->flight_size; 392 393 /* See if we need to exit Fast Recovery first */ 394 if (asoc->fast_recovery && 395 TSN_lte(asoc->fast_recovery_exit, sack_ctsn)) 396 asoc->fast_recovery = 0; 397 398 /* The appropriate cwnd increase algorithm is performed if, and only 399 * if the cumulative TSN whould advanced and the congestion window is 400 * being fully utilized. 401 */ 402 if (TSN_lte(sack_ctsn, transport->asoc->ctsn_ack_point) || 403 (flight_size < cwnd)) 404 return; 405 406 ssthresh = transport->ssthresh; 407 pba = transport->partial_bytes_acked; 408 pmtu = transport->asoc->pathmtu; 409 410 if (cwnd <= ssthresh) { 411 /* RFC 4960 7.2.1 412 * o When cwnd is less than or equal to ssthresh, an SCTP 413 * endpoint MUST use the slow-start algorithm to increase 414 * cwnd only if the current congestion window is being fully 415 * utilized, an incoming SACK advances the Cumulative TSN 416 * Ack Point, and the data sender is not in Fast Recovery. 417 * Only when these three conditions are met can the cwnd be 418 * increased; otherwise, the cwnd MUST not be increased. 419 * If these conditions are met, then cwnd MUST be increased 420 * by, at most, the lesser of 1) the total size of the 421 * previously outstanding DATA chunk(s) acknowledged, and 422 * 2) the destination's path MTU. This upper bound protects 423 * against the ACK-Splitting attack outlined in [SAVAGE99]. 424 */ 425 if (asoc->fast_recovery) 426 return; 427 428 if (bytes_acked > pmtu) 429 cwnd += pmtu; 430 else 431 cwnd += bytes_acked; 432 SCTP_DEBUG_PRINTK("%s: SLOW START: transport: %p, " 433 "bytes_acked: %d, cwnd: %d, ssthresh: %d, " 434 "flight_size: %d, pba: %d\n", 435 __func__, 436 transport, bytes_acked, cwnd, 437 ssthresh, flight_size, pba); 438 } else { 439 /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh, 440 * upon each SACK arrival that advances the Cumulative TSN Ack 441 * Point, increase partial_bytes_acked by the total number of 442 * bytes of all new chunks acknowledged in that SACK including 443 * chunks acknowledged by the new Cumulative TSN Ack and by 444 * Gap Ack Blocks. 445 * 446 * When partial_bytes_acked is equal to or greater than cwnd 447 * and before the arrival of the SACK the sender had cwnd or 448 * more bytes of data outstanding (i.e., before arrival of the 449 * SACK, flightsize was greater than or equal to cwnd), 450 * increase cwnd by MTU, and reset partial_bytes_acked to 451 * (partial_bytes_acked - cwnd). 452 */ 453 pba += bytes_acked; 454 if (pba >= cwnd) { 455 cwnd += pmtu; 456 pba = ((cwnd < pba) ? (pba - cwnd) : 0); 457 } 458 SCTP_DEBUG_PRINTK("%s: CONGESTION AVOIDANCE: " 459 "transport: %p, bytes_acked: %d, cwnd: %d, " 460 "ssthresh: %d, flight_size: %d, pba: %d\n", 461 __func__, 462 transport, bytes_acked, cwnd, 463 ssthresh, flight_size, pba); 464 } 465 466 transport->cwnd = cwnd; 467 transport->partial_bytes_acked = pba; 468} 469 470/* This routine is used to lower the transport's cwnd when congestion is 471 * detected. 472 */ 473void sctp_transport_lower_cwnd(struct sctp_transport *transport, 474 sctp_lower_cwnd_t reason) 475{ 476 struct sctp_association *asoc = transport->asoc; 477 478 switch (reason) { 479 case SCTP_LOWER_CWND_T3_RTX: 480 /* RFC 2960 Section 7.2.3, sctpimpguide 481 * When the T3-rtx timer expires on an address, SCTP should 482 * perform slow start by: 483 * ssthresh = max(cwnd/2, 4*MTU) 484 * cwnd = 1*MTU 485 * partial_bytes_acked = 0 486 */ 487 transport->ssthresh = max(transport->cwnd/2, 488 4*asoc->pathmtu); 489 transport->cwnd = asoc->pathmtu; 490 491 /* T3-rtx also clears fast recovery */ 492 asoc->fast_recovery = 0; 493 break; 494 495 case SCTP_LOWER_CWND_FAST_RTX: 496 /* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the 497 * destination address(es) to which the missing DATA chunks 498 * were last sent, according to the formula described in 499 * Section 7.2.3. 500 * 501 * RFC 2960 7.2.3, sctpimpguide Upon detection of packet 502 * losses from SACK (see Section 7.2.4), An endpoint 503 * should do the following: 504 * ssthresh = max(cwnd/2, 4*MTU) 505 * cwnd = ssthresh 506 * partial_bytes_acked = 0 507 */ 508 if (asoc->fast_recovery) 509 return; 510 511 /* Mark Fast recovery */ 512 asoc->fast_recovery = 1; 513 asoc->fast_recovery_exit = asoc->next_tsn - 1; 514 515 transport->ssthresh = max(transport->cwnd/2, 516 4*asoc->pathmtu); 517 transport->cwnd = transport->ssthresh; 518 break; 519 520 case SCTP_LOWER_CWND_ECNE: 521 /* RFC 2481 Section 6.1.2. 522 * If the sender receives an ECN-Echo ACK packet 523 * then the sender knows that congestion was encountered in the 524 * network on the path from the sender to the receiver. The 525 * indication of congestion should be treated just as a 526 * congestion loss in non-ECN Capable TCP. That is, the TCP 527 * source halves the congestion window "cwnd" and reduces the 528 * slow start threshold "ssthresh". 529 * A critical condition is that TCP does not react to 530 * congestion indications more than once every window of 531 * data (or more loosely more than once every round-trip time). 532 */ 533 if (time_after(jiffies, transport->last_time_ecne_reduced + 534 transport->rtt)) { 535 transport->ssthresh = max(transport->cwnd/2, 536 4*asoc->pathmtu); 537 transport->cwnd = transport->ssthresh; 538 transport->last_time_ecne_reduced = jiffies; 539 } 540 break; 541 542 case SCTP_LOWER_CWND_INACTIVE: 543 /* RFC 2960 Section 7.2.1, sctpimpguide 544 * When the endpoint does not transmit data on a given 545 * transport address, the cwnd of the transport address 546 * should be adjusted to max(cwnd/2, 4*MTU) per RTO. 547 * NOTE: Although the draft recommends that this check needs 548 * to be done every RTO interval, we do it every hearbeat 549 * interval. 550 */ 551 transport->cwnd = max(transport->cwnd/2, 552 4*asoc->pathmtu); 553 break; 554 } 555 556 transport->partial_bytes_acked = 0; 557 SCTP_DEBUG_PRINTK("%s: transport: %p reason: %d cwnd: " 558 "%d ssthresh: %d\n", __func__, 559 transport, reason, 560 transport->cwnd, transport->ssthresh); 561} 562 563/* Apply Max.Burst limit to the congestion window: 564 * sctpimpguide-05 2.14.2 565 * D) When the time comes for the sender to 566 * transmit new DATA chunks, the protocol parameter Max.Burst MUST 567 * first be applied to limit how many new DATA chunks may be sent. 568 * The limit is applied by adjusting cwnd as follows: 569 * if ((flightsize+ Max.Burst * MTU) < cwnd) 570 * cwnd = flightsize + Max.Burst * MTU 571 */ 572 573void sctp_transport_burst_limited(struct sctp_transport *t) 574{ 575 struct sctp_association *asoc = t->asoc; 576 u32 old_cwnd = t->cwnd; 577 u32 max_burst_bytes; 578 579 if (t->burst_limited) 580 return; 581 582 max_burst_bytes = t->flight_size + (asoc->max_burst * asoc->pathmtu); 583 if (max_burst_bytes < old_cwnd) { 584 t->cwnd = max_burst_bytes; 585 t->burst_limited = old_cwnd; 586 } 587} 588 589/* Restore the old cwnd congestion window, after the burst had it's 590 * desired effect. 591 */ 592void sctp_transport_burst_reset(struct sctp_transport *t) 593{ 594 if (t->burst_limited) { 595 t->cwnd = t->burst_limited; 596 t->burst_limited = 0; 597 } 598} 599 600/* What is the next timeout value for this transport? */ 601unsigned long sctp_transport_timeout(struct sctp_transport *t) 602{ 603 unsigned long timeout; 604 timeout = t->rto + sctp_jitter(t->rto); 605 if (t->state != SCTP_UNCONFIRMED) 606 timeout += t->hbinterval; 607 timeout += jiffies; 608 return timeout; 609} 610 611/* Reset transport variables to their initial values */ 612void sctp_transport_reset(struct sctp_transport *t) 613{ 614 struct sctp_association *asoc = t->asoc; 615 616 /* RFC 2960 (bis), Section 5.2.4 617 * All the congestion control parameters (e.g., cwnd, ssthresh) 618 * related to this peer MUST be reset to their initial values 619 * (see Section 6.2.1) 620 */ 621 t->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380)); 622 t->burst_limited = 0; 623 t->ssthresh = asoc->peer.i.a_rwnd; 624 t->rto = asoc->rto_initial; 625 t->rtt = 0; 626 t->srtt = 0; 627 t->rttvar = 0; 628 629 /* Reset these additional varibles so that we have a clean 630 * slate. 631 */ 632 t->partial_bytes_acked = 0; 633 t->flight_size = 0; 634 t->error_count = 0; 635 t->rto_pending = 0; 636 t->hb_sent = 0; 637 638 /* Initialize the state information for SFR-CACC */ 639 t->cacc.changeover_active = 0; 640 t->cacc.cycling_changeover = 0; 641 t->cacc.next_tsn_at_change = 0; 642 t->cacc.cacc_saw_newack = 0; 643} 644