transport.c revision d448388bdaca946aa2b07973cb72a9b834e530bf
1/* SCTP kernel reference 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 reference 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 * The SCTP reference 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 * The SCTP reference 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/types.h> 52#include <linux/random.h> 53#include <net/sctp/sctp.h> 54#include <net/sctp/sm.h> 55 56/* 1st Level Abstractions. */ 57 58/* Initialize a new transport from provided memory. */ 59static struct sctp_transport *sctp_transport_init(struct sctp_transport *peer, 60 const union sctp_addr *addr, 61 gfp_t gfp) 62{ 63 /* Copy in the address. */ 64 peer->ipaddr = *addr; 65 flip_to_h(&peer->ipaddr_h, &peer->ipaddr); 66 peer->af_specific = sctp_get_af_specific(addr->sa.sa_family); 67 peer->asoc = NULL; 68 69 peer->dst = NULL; 70 memset(&peer->saddr, 0, sizeof(union sctp_addr)); 71 memset(&peer->saddr_h, 0, sizeof(union sctp_addr)); 72 73 /* From 6.3.1 RTO Calculation: 74 * 75 * C1) Until an RTT measurement has been made for a packet sent to the 76 * given destination transport address, set RTO to the protocol 77 * parameter 'RTO.Initial'. 78 */ 79 peer->rtt = 0; 80 peer->rto = msecs_to_jiffies(sctp_rto_initial); 81 peer->rttvar = 0; 82 peer->srtt = 0; 83 peer->rto_pending = 0; 84 85 peer->last_time_heard = jiffies; 86 peer->last_time_used = jiffies; 87 peer->last_time_ecne_reduced = jiffies; 88 89 peer->init_sent_count = 0; 90 91 peer->param_flags = SPP_HB_DISABLE | 92 SPP_PMTUD_ENABLE | 93 SPP_SACKDELAY_ENABLE; 94 peer->hbinterval = 0; 95 96 /* Initialize the default path max_retrans. */ 97 peer->pathmaxrxt = sctp_max_retrans_path; 98 peer->error_count = 0; 99 100 INIT_LIST_HEAD(&peer->transmitted); 101 INIT_LIST_HEAD(&peer->send_ready); 102 INIT_LIST_HEAD(&peer->transports); 103 104 /* Set up the retransmission timer. */ 105 init_timer(&peer->T3_rtx_timer); 106 peer->T3_rtx_timer.function = sctp_generate_t3_rtx_event; 107 peer->T3_rtx_timer.data = (unsigned long)peer; 108 109 /* Set up the heartbeat timer. */ 110 init_timer(&peer->hb_timer); 111 peer->hb_timer.function = sctp_generate_heartbeat_event; 112 peer->hb_timer.data = (unsigned long)peer; 113 114 /* Initialize the 64-bit random nonce sent with heartbeat. */ 115 get_random_bytes(&peer->hb_nonce, sizeof(peer->hb_nonce)); 116 117 atomic_set(&peer->refcnt, 1); 118 peer->dead = 0; 119 120 peer->malloced = 0; 121 122 /* Initialize the state information for SFR-CACC */ 123 peer->cacc.changeover_active = 0; 124 peer->cacc.cycling_changeover = 0; 125 peer->cacc.next_tsn_at_change = 0; 126 peer->cacc.cacc_saw_newack = 0; 127 128 return peer; 129} 130 131/* Allocate and initialize a new transport. */ 132struct sctp_transport *sctp_transport_new(const union sctp_addr *addr, 133 gfp_t gfp) 134{ 135 struct sctp_transport *transport; 136 137 transport = t_new(struct sctp_transport, gfp); 138 if (!transport) 139 goto fail; 140 141 if (!sctp_transport_init(transport, addr, gfp)) 142 goto fail_init; 143 144 transport->malloced = 1; 145 SCTP_DBG_OBJCNT_INC(transport); 146 147 return transport; 148 149fail_init: 150 kfree(transport); 151 152fail: 153 return NULL; 154} 155 156/* This transport is no longer needed. Free up if possible, or 157 * delay until it last reference count. 158 */ 159void sctp_transport_free(struct sctp_transport *transport) 160{ 161 transport->dead = 1; 162 163 /* Try to delete the heartbeat timer. */ 164 if (del_timer(&transport->hb_timer)) 165 sctp_transport_put(transport); 166 167 /* Delete the T3_rtx timer if it's active. 168 * There is no point in not doing this now and letting 169 * structure hang around in memory since we know 170 * the tranport is going away. 171 */ 172 if (timer_pending(&transport->T3_rtx_timer) && 173 del_timer(&transport->T3_rtx_timer)) 174 sctp_transport_put(transport); 175 176 177 sctp_transport_put(transport); 178} 179 180/* Destroy the transport data structure. 181 * Assumes there are no more users of this structure. 182 */ 183static void sctp_transport_destroy(struct sctp_transport *transport) 184{ 185 SCTP_ASSERT(transport->dead, "Transport is not dead", return); 186 187 if (transport->asoc) 188 sctp_association_put(transport->asoc); 189 190 sctp_packet_free(&transport->packet); 191 192 dst_release(transport->dst); 193 kfree(transport); 194 SCTP_DBG_OBJCNT_DEC(transport); 195} 196 197/* Start T3_rtx timer if it is not already running and update the heartbeat 198 * timer. This routine is called every time a DATA chunk is sent. 199 */ 200void sctp_transport_reset_timers(struct sctp_transport *transport) 201{ 202 /* RFC 2960 6.3.2 Retransmission Timer Rules 203 * 204 * R1) Every time a DATA chunk is sent to any address(including a 205 * retransmission), if the T3-rtx timer of that address is not running 206 * start it running so that it will expire after the RTO of that 207 * address. 208 */ 209 210 if (!timer_pending(&transport->T3_rtx_timer)) 211 if (!mod_timer(&transport->T3_rtx_timer, 212 jiffies + transport->rto)) 213 sctp_transport_hold(transport); 214 215 /* When a data chunk is sent, reset the heartbeat interval. */ 216 if (!mod_timer(&transport->hb_timer, 217 sctp_transport_timeout(transport))) 218 sctp_transport_hold(transport); 219} 220 221/* This transport has been assigned to an association. 222 * Initialize fields from the association or from the sock itself. 223 * Register the reference count in the association. 224 */ 225void sctp_transport_set_owner(struct sctp_transport *transport, 226 struct sctp_association *asoc) 227{ 228 transport->asoc = asoc; 229 sctp_association_hold(asoc); 230} 231 232/* Initialize the pmtu of a transport. */ 233void sctp_transport_pmtu(struct sctp_transport *transport) 234{ 235 struct dst_entry *dst; 236 237 dst = transport->af_specific->get_dst(NULL, &transport->ipaddr, NULL); 238 239 if (dst) { 240 transport->pathmtu = dst_mtu(dst); 241 dst_release(dst); 242 } else 243 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT; 244} 245 246/* Caches the dst entry and source address for a transport's destination 247 * address. 248 */ 249void sctp_transport_route(struct sctp_transport *transport, 250 union sctp_addr *saddr, struct sctp_sock *opt) 251{ 252 struct sctp_association *asoc = transport->asoc; 253 struct sctp_af *af = transport->af_specific; 254 union sctp_addr *daddr = &transport->ipaddr; 255 struct dst_entry *dst; 256 257 dst = af->get_dst(asoc, daddr, saddr); 258 259 if (saddr) 260 memcpy(&transport->saddr, saddr, sizeof(union sctp_addr)); 261 else 262 af->get_saddr(asoc, dst, daddr, &transport->saddr); 263 264 flip_to_h(&transport->saddr_h, &transport->saddr); 265 transport->dst = dst; 266 if ((transport->param_flags & SPP_PMTUD_DISABLE) && transport->pathmtu) { 267 return; 268 } 269 if (dst) { 270 transport->pathmtu = dst_mtu(dst); 271 272 /* Initialize sk->sk_rcv_saddr, if the transport is the 273 * association's active path for getsockname(). 274 */ 275 if (asoc && (transport == asoc->peer.active_path)) 276 opt->pf->af->to_sk_saddr(&transport->saddr_h, 277 asoc->base.sk); 278 } else 279 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT; 280} 281 282/* Hold a reference to a transport. */ 283void sctp_transport_hold(struct sctp_transport *transport) 284{ 285 atomic_inc(&transport->refcnt); 286} 287 288/* Release a reference to a transport and clean up 289 * if there are no more references. 290 */ 291void sctp_transport_put(struct sctp_transport *transport) 292{ 293 if (atomic_dec_and_test(&transport->refcnt)) 294 sctp_transport_destroy(transport); 295} 296 297/* Update transport's RTO based on the newly calculated RTT. */ 298void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt) 299{ 300 /* Check for valid transport. */ 301 SCTP_ASSERT(tp, "NULL transport", return); 302 303 /* We should not be doing any RTO updates unless rto_pending is set. */ 304 SCTP_ASSERT(tp->rto_pending, "rto_pending not set", return); 305 306 if (tp->rttvar || tp->srtt) { 307 /* 6.3.1 C3) When a new RTT measurement R' is made, set 308 * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'| 309 * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R' 310 */ 311 312 /* Note: The above algorithm has been rewritten to 313 * express rto_beta and rto_alpha as inverse powers 314 * of two. 315 * For example, assuming the default value of RTO.Alpha of 316 * 1/8, rto_alpha would be expressed as 3. 317 */ 318 tp->rttvar = tp->rttvar - (tp->rttvar >> sctp_rto_beta) 319 + ((abs(tp->srtt - rtt)) >> sctp_rto_beta); 320 tp->srtt = tp->srtt - (tp->srtt >> sctp_rto_alpha) 321 + (rtt >> sctp_rto_alpha); 322 } else { 323 /* 6.3.1 C2) When the first RTT measurement R is made, set 324 * SRTT <- R, RTTVAR <- R/2. 325 */ 326 tp->srtt = rtt; 327 tp->rttvar = rtt >> 1; 328 } 329 330 /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then 331 * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY. 332 */ 333 if (tp->rttvar == 0) 334 tp->rttvar = SCTP_CLOCK_GRANULARITY; 335 336 /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */ 337 tp->rto = tp->srtt + (tp->rttvar << 2); 338 339 /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min 340 * seconds then it is rounded up to RTO.Min seconds. 341 */ 342 if (tp->rto < tp->asoc->rto_min) 343 tp->rto = tp->asoc->rto_min; 344 345 /* 6.3.1 C7) A maximum value may be placed on RTO provided it is 346 * at least RTO.max seconds. 347 */ 348 if (tp->rto > tp->asoc->rto_max) 349 tp->rto = tp->asoc->rto_max; 350 351 tp->rtt = rtt; 352 353 /* Reset rto_pending so that a new RTT measurement is started when a 354 * new data chunk is sent. 355 */ 356 tp->rto_pending = 0; 357 358 SCTP_DEBUG_PRINTK("%s: transport: %p, rtt: %d, srtt: %d " 359 "rttvar: %d, rto: %ld\n", __FUNCTION__, 360 tp, rtt, tp->srtt, tp->rttvar, tp->rto); 361} 362 363/* This routine updates the transport's cwnd and partial_bytes_acked 364 * parameters based on the bytes acked in the received SACK. 365 */ 366void sctp_transport_raise_cwnd(struct sctp_transport *transport, 367 __u32 sack_ctsn, __u32 bytes_acked) 368{ 369 __u32 cwnd, ssthresh, flight_size, pba, pmtu; 370 371 cwnd = transport->cwnd; 372 flight_size = transport->flight_size; 373 374 /* The appropriate cwnd increase algorithm is performed if, and only 375 * if the cumulative TSN has advanced and the congestion window is 376 * being fully utilized. 377 */ 378 if ((transport->asoc->ctsn_ack_point >= sack_ctsn) || 379 (flight_size < cwnd)) 380 return; 381 382 ssthresh = transport->ssthresh; 383 pba = transport->partial_bytes_acked; 384 pmtu = transport->asoc->pathmtu; 385 386 if (cwnd <= ssthresh) { 387 /* RFC 2960 7.2.1, sctpimpguide-05 2.14.2 When cwnd is less 388 * than or equal to ssthresh an SCTP endpoint MUST use the 389 * slow start algorithm to increase cwnd only if the current 390 * congestion window is being fully utilized and an incoming 391 * SACK advances the Cumulative TSN Ack Point. Only when these 392 * two conditions are met can the cwnd be increased otherwise 393 * the cwnd MUST not be increased. If these conditions are met 394 * then cwnd MUST be increased by at most the lesser of 395 * 1) the total size of the previously outstanding DATA 396 * chunk(s) acknowledged, and 2) the destination's path MTU. 397 */ 398 if (bytes_acked > pmtu) 399 cwnd += pmtu; 400 else 401 cwnd += bytes_acked; 402 SCTP_DEBUG_PRINTK("%s: SLOW START: transport: %p, " 403 "bytes_acked: %d, cwnd: %d, ssthresh: %d, " 404 "flight_size: %d, pba: %d\n", 405 __FUNCTION__, 406 transport, bytes_acked, cwnd, 407 ssthresh, flight_size, pba); 408 } else { 409 /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh, 410 * upon each SACK arrival that advances the Cumulative TSN Ack 411 * Point, increase partial_bytes_acked by the total number of 412 * bytes of all new chunks acknowledged in that SACK including 413 * chunks acknowledged by the new Cumulative TSN Ack and by 414 * Gap Ack Blocks. 415 * 416 * When partial_bytes_acked is equal to or greater than cwnd 417 * and before the arrival of the SACK the sender had cwnd or 418 * more bytes of data outstanding (i.e., before arrival of the 419 * SACK, flightsize was greater than or equal to cwnd), 420 * increase cwnd by MTU, and reset partial_bytes_acked to 421 * (partial_bytes_acked - cwnd). 422 */ 423 pba += bytes_acked; 424 if (pba >= cwnd) { 425 cwnd += pmtu; 426 pba = ((cwnd < pba) ? (pba - cwnd) : 0); 427 } 428 SCTP_DEBUG_PRINTK("%s: CONGESTION AVOIDANCE: " 429 "transport: %p, bytes_acked: %d, cwnd: %d, " 430 "ssthresh: %d, flight_size: %d, pba: %d\n", 431 __FUNCTION__, 432 transport, bytes_acked, cwnd, 433 ssthresh, flight_size, pba); 434 } 435 436 transport->cwnd = cwnd; 437 transport->partial_bytes_acked = pba; 438} 439 440/* This routine is used to lower the transport's cwnd when congestion is 441 * detected. 442 */ 443void sctp_transport_lower_cwnd(struct sctp_transport *transport, 444 sctp_lower_cwnd_t reason) 445{ 446 switch (reason) { 447 case SCTP_LOWER_CWND_T3_RTX: 448 /* RFC 2960 Section 7.2.3, sctpimpguide 449 * When the T3-rtx timer expires on an address, SCTP should 450 * perform slow start by: 451 * ssthresh = max(cwnd/2, 4*MTU) 452 * cwnd = 1*MTU 453 * partial_bytes_acked = 0 454 */ 455 transport->ssthresh = max(transport->cwnd/2, 456 4*transport->asoc->pathmtu); 457 transport->cwnd = transport->asoc->pathmtu; 458 break; 459 460 case SCTP_LOWER_CWND_FAST_RTX: 461 /* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the 462 * destination address(es) to which the missing DATA chunks 463 * were last sent, according to the formula described in 464 * Section 7.2.3. 465 * 466 * RFC 2960 7.2.3, sctpimpguide Upon detection of packet 467 * losses from SACK (see Section 7.2.4), An endpoint 468 * should do the following: 469 * ssthresh = max(cwnd/2, 4*MTU) 470 * cwnd = ssthresh 471 * partial_bytes_acked = 0 472 */ 473 transport->ssthresh = max(transport->cwnd/2, 474 4*transport->asoc->pathmtu); 475 transport->cwnd = transport->ssthresh; 476 break; 477 478 case SCTP_LOWER_CWND_ECNE: 479 /* RFC 2481 Section 6.1.2. 480 * If the sender receives an ECN-Echo ACK packet 481 * then the sender knows that congestion was encountered in the 482 * network on the path from the sender to the receiver. The 483 * indication of congestion should be treated just as a 484 * congestion loss in non-ECN Capable TCP. That is, the TCP 485 * source halves the congestion window "cwnd" and reduces the 486 * slow start threshold "ssthresh". 487 * A critical condition is that TCP does not react to 488 * congestion indications more than once every window of 489 * data (or more loosely more than once every round-trip time). 490 */ 491 if ((jiffies - transport->last_time_ecne_reduced) > 492 transport->rtt) { 493 transport->ssthresh = max(transport->cwnd/2, 494 4*transport->asoc->pathmtu); 495 transport->cwnd = transport->ssthresh; 496 transport->last_time_ecne_reduced = jiffies; 497 } 498 break; 499 500 case SCTP_LOWER_CWND_INACTIVE: 501 /* RFC 2960 Section 7.2.1, sctpimpguide 502 * When the endpoint does not transmit data on a given 503 * transport address, the cwnd of the transport address 504 * should be adjusted to max(cwnd/2, 4*MTU) per RTO. 505 * NOTE: Although the draft recommends that this check needs 506 * to be done every RTO interval, we do it every hearbeat 507 * interval. 508 */ 509 if ((jiffies - transport->last_time_used) > transport->rto) 510 transport->cwnd = max(transport->cwnd/2, 511 4*transport->asoc->pathmtu); 512 break; 513 }; 514 515 transport->partial_bytes_acked = 0; 516 SCTP_DEBUG_PRINTK("%s: transport: %p reason: %d cwnd: " 517 "%d ssthresh: %d\n", __FUNCTION__, 518 transport, reason, 519 transport->cwnd, transport->ssthresh); 520} 521 522/* What is the next timeout value for this transport? */ 523unsigned long sctp_transport_timeout(struct sctp_transport *t) 524{ 525 unsigned long timeout; 526 timeout = t->rto + sctp_jitter(t->rto); 527 if (t->state != SCTP_UNCONFIRMED) 528 timeout += t->hbinterval; 529 timeout += jiffies; 530 return timeout; 531} 532