| 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, see |
| 28 | * <http://www.gnu.org/licenses/>. |
| 29 | * |
| 30 | * Please send any bug reports or fixes you make to the |
| 31 | * email address(es): |
| 32 | * lksctp developers <linux-sctp@vger.kernel.org> |
| 33 | * |
| 34 | * Written or modified by: |
| 35 | * La Monte H.P. Yarroll <piggy@acm.org> |
| 36 | * Karl Knutson <karl@athena.chicago.il.us> |
| 37 | * Jon Grimm <jgrimm@us.ibm.com> |
| 38 | * Xingang Guo <xingang.guo@intel.com> |
| 39 | * Hui Huang <hui.huang@nokia.com> |
| 40 | * Sridhar Samudrala <sri@us.ibm.com> |
| 41 | * Ardelle Fan <ardelle.fan@intel.com> |
| 42 | */ |
| 43 | |
| 44 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
| 45 | |
| 46 | #include <linux/slab.h> |
| 47 | #include <linux/types.h> |
| 48 | #include <linux/random.h> |
| 49 | #include <net/sctp/sctp.h> |
| 50 | #include <net/sctp/sm.h> |
| 51 | |
| 52 | /* 1st Level Abstractions. */ |
| 53 | |
| 54 | /* Initialize a new transport from provided memory. */ |
| 55 | static struct sctp_transport *sctp_transport_init(struct net *net, |
| 56 | struct sctp_transport *peer, |
| 57 | const union sctp_addr *addr, |
| 58 | gfp_t gfp) |
| 59 | { |
| 60 | /* Copy in the address. */ |
| 61 | peer->ipaddr = *addr; |
| 62 | peer->af_specific = sctp_get_af_specific(addr->sa.sa_family); |
| 63 | memset(&peer->saddr, 0, sizeof(union sctp_addr)); |
| 64 | |
| 65 | peer->sack_generation = 0; |
| 66 | |
| 67 | /* From 6.3.1 RTO Calculation: |
| 68 | * |
| 69 | * C1) Until an RTT measurement has been made for a packet sent to the |
| 70 | * given destination transport address, set RTO to the protocol |
| 71 | * parameter 'RTO.Initial'. |
| 72 | */ |
| 73 | peer->rto = msecs_to_jiffies(net->sctp.rto_initial); |
| 74 | |
| 75 | peer->last_time_heard = 0; |
| 76 | peer->last_time_ecne_reduced = jiffies; |
| 77 | |
| 78 | peer->param_flags = SPP_HB_DISABLE | |
| 79 | SPP_PMTUD_ENABLE | |
| 80 | SPP_SACKDELAY_ENABLE; |
| 81 | |
| 82 | /* Initialize the default path max_retrans. */ |
| 83 | peer->pathmaxrxt = net->sctp.max_retrans_path; |
| 84 | peer->pf_retrans = net->sctp.pf_retrans; |
| 85 | |
| 86 | INIT_LIST_HEAD(&peer->transmitted); |
| 87 | INIT_LIST_HEAD(&peer->send_ready); |
| 88 | INIT_LIST_HEAD(&peer->transports); |
| 89 | |
| 90 | setup_timer(&peer->T3_rtx_timer, sctp_generate_t3_rtx_event, |
| 91 | (unsigned long)peer); |
| 92 | setup_timer(&peer->hb_timer, sctp_generate_heartbeat_event, |
| 93 | (unsigned long)peer); |
| 94 | setup_timer(&peer->reconf_timer, sctp_generate_reconf_event, |
| 95 | (unsigned long)peer); |
| 96 | setup_timer(&peer->proto_unreach_timer, |
| 97 | sctp_generate_proto_unreach_event, (unsigned long)peer); |
| 98 | |
| 99 | /* Initialize the 64-bit random nonce sent with heartbeat. */ |
| 100 | get_random_bytes(&peer->hb_nonce, sizeof(peer->hb_nonce)); |
| 101 | |
| 102 | refcount_set(&peer->refcnt, 1); |
| 103 | |
| 104 | return peer; |
| 105 | } |
| 106 | |
| 107 | /* Allocate and initialize a new transport. */ |
| 108 | struct sctp_transport *sctp_transport_new(struct net *net, |
| 109 | const union sctp_addr *addr, |
| 110 | gfp_t gfp) |
| 111 | { |
| 112 | struct sctp_transport *transport; |
| 113 | |
| 114 | transport = kzalloc(sizeof(*transport), gfp); |
| 115 | if (!transport) |
| 116 | goto fail; |
| 117 | |
| 118 | if (!sctp_transport_init(net, transport, addr, gfp)) |
| 119 | goto fail_init; |
| 120 | |
| 121 | SCTP_DBG_OBJCNT_INC(transport); |
| 122 | |
| 123 | return transport; |
| 124 | |
| 125 | fail_init: |
| 126 | kfree(transport); |
| 127 | |
| 128 | fail: |
| 129 | return NULL; |
| 130 | } |
| 131 | |
| 132 | /* This transport is no longer needed. Free up if possible, or |
| 133 | * delay until it last reference count. |
| 134 | */ |
| 135 | void sctp_transport_free(struct sctp_transport *transport) |
| 136 | { |
| 137 | /* Try to delete the heartbeat timer. */ |
| 138 | if (del_timer(&transport->hb_timer)) |
| 139 | sctp_transport_put(transport); |
| 140 | |
| 141 | /* Delete the T3_rtx timer if it's active. |
| 142 | * There is no point in not doing this now and letting |
| 143 | * structure hang around in memory since we know |
| 144 | * the tranport is going away. |
| 145 | */ |
| 146 | if (del_timer(&transport->T3_rtx_timer)) |
| 147 | sctp_transport_put(transport); |
| 148 | |
| 149 | if (del_timer(&transport->reconf_timer)) |
| 150 | sctp_transport_put(transport); |
| 151 | |
| 152 | /* Delete the ICMP proto unreachable timer if it's active. */ |
| 153 | if (del_timer(&transport->proto_unreach_timer)) |
| 154 | sctp_association_put(transport->asoc); |
| 155 | |
| 156 | sctp_transport_put(transport); |
| 157 | } |
| 158 | |
| 159 | static void sctp_transport_destroy_rcu(struct rcu_head *head) |
| 160 | { |
| 161 | struct sctp_transport *transport; |
| 162 | |
| 163 | transport = container_of(head, struct sctp_transport, rcu); |
| 164 | |
| 165 | dst_release(transport->dst); |
| 166 | kfree(transport); |
| 167 | SCTP_DBG_OBJCNT_DEC(transport); |
| 168 | } |
| 169 | |
| 170 | /* Destroy the transport data structure. |
| 171 | * Assumes there are no more users of this structure. |
| 172 | */ |
| 173 | static void sctp_transport_destroy(struct sctp_transport *transport) |
| 174 | { |
| 175 | if (unlikely(refcount_read(&transport->refcnt))) { |
| 176 | WARN(1, "Attempt to destroy undead transport %p!\n", transport); |
| 177 | return; |
| 178 | } |
| 179 | |
| 180 | sctp_packet_free(&transport->packet); |
| 181 | |
| 182 | if (transport->asoc) |
| 183 | sctp_association_put(transport->asoc); |
| 184 | |
| 185 | call_rcu(&transport->rcu, sctp_transport_destroy_rcu); |
| 186 | } |
| 187 | |
| 188 | /* Start T3_rtx timer if it is not already running and update the heartbeat |
| 189 | * timer. This routine is called every time a DATA chunk is sent. |
| 190 | */ |
| 191 | void sctp_transport_reset_t3_rtx(struct sctp_transport *transport) |
| 192 | { |
| 193 | /* RFC 2960 6.3.2 Retransmission Timer Rules |
| 194 | * |
| 195 | * R1) Every time a DATA chunk is sent to any address(including a |
| 196 | * retransmission), if the T3-rtx timer of that address is not running |
| 197 | * start it running so that it will expire after the RTO of that |
| 198 | * address. |
| 199 | */ |
| 200 | |
| 201 | if (!timer_pending(&transport->T3_rtx_timer)) |
| 202 | if (!mod_timer(&transport->T3_rtx_timer, |
| 203 | jiffies + transport->rto)) |
| 204 | sctp_transport_hold(transport); |
| 205 | } |
| 206 | |
| 207 | void sctp_transport_reset_hb_timer(struct sctp_transport *transport) |
| 208 | { |
| 209 | unsigned long expires; |
| 210 | |
| 211 | /* When a data chunk is sent, reset the heartbeat interval. */ |
| 212 | expires = jiffies + sctp_transport_timeout(transport); |
| 213 | if (time_before(transport->hb_timer.expires, expires) && |
| 214 | !mod_timer(&transport->hb_timer, |
| 215 | expires + prandom_u32_max(transport->rto))) |
| 216 | sctp_transport_hold(transport); |
| 217 | } |
| 218 | |
| 219 | void sctp_transport_reset_reconf_timer(struct sctp_transport *transport) |
| 220 | { |
| 221 | if (!timer_pending(&transport->reconf_timer)) |
| 222 | if (!mod_timer(&transport->reconf_timer, |
| 223 | jiffies + transport->rto)) |
| 224 | sctp_transport_hold(transport); |
| 225 | } |
| 226 | |
| 227 | /* This transport has been assigned to an association. |
| 228 | * Initialize fields from the association or from the sock itself. |
| 229 | * Register the reference count in the association. |
| 230 | */ |
| 231 | void sctp_transport_set_owner(struct sctp_transport *transport, |
| 232 | struct sctp_association *asoc) |
| 233 | { |
| 234 | transport->asoc = asoc; |
| 235 | sctp_association_hold(asoc); |
| 236 | } |
| 237 | |
| 238 | /* Initialize the pmtu of a transport. */ |
| 239 | void sctp_transport_pmtu(struct sctp_transport *transport, struct sock *sk) |
| 240 | { |
| 241 | /* If we don't have a fresh route, look one up */ |
| 242 | if (!transport->dst || transport->dst->obsolete) { |
| 243 | sctp_transport_dst_release(transport); |
| 244 | transport->af_specific->get_dst(transport, &transport->saddr, |
| 245 | &transport->fl, sk); |
| 246 | } |
| 247 | |
| 248 | if (transport->dst) { |
| 249 | transport->pathmtu = SCTP_TRUNC4(dst_mtu(transport->dst)); |
| 250 | } else |
| 251 | transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT; |
| 252 | } |
| 253 | |
| 254 | bool sctp_transport_update_pmtu(struct sctp_transport *t, u32 pmtu) |
| 255 | { |
| 256 | struct dst_entry *dst = sctp_transport_dst_check(t); |
| 257 | struct sock *sk = t->asoc->base.sk; |
| 258 | bool change = true; |
| 259 | |
| 260 | if (unlikely(pmtu < SCTP_DEFAULT_MINSEGMENT)) { |
| 261 | pr_warn_ratelimited("%s: Reported pmtu %d too low, using default minimum of %d\n", |
| 262 | __func__, pmtu, SCTP_DEFAULT_MINSEGMENT); |
| 263 | /* Use default minimum segment instead */ |
| 264 | pmtu = SCTP_DEFAULT_MINSEGMENT; |
| 265 | } |
| 266 | pmtu = SCTP_TRUNC4(pmtu); |
| 267 | |
| 268 | if (dst) { |
| 269 | struct sctp_pf *pf = sctp_get_pf_specific(dst->ops->family); |
| 270 | union sctp_addr addr; |
| 271 | |
| 272 | pf->af->from_sk(&addr, sk); |
| 273 | pf->to_sk_daddr(&t->ipaddr, sk); |
| 274 | dst->ops->update_pmtu(dst, sk, NULL, pmtu); |
| 275 | pf->to_sk_daddr(&addr, sk); |
| 276 | |
| 277 | dst = sctp_transport_dst_check(t); |
| 278 | } |
| 279 | |
| 280 | if (!dst) { |
| 281 | t->af_specific->get_dst(t, &t->saddr, &t->fl, sk); |
| 282 | dst = t->dst; |
| 283 | } |
| 284 | |
| 285 | if (dst) { |
| 286 | /* Re-fetch, as under layers may have a higher minimum size */ |
| 287 | pmtu = SCTP_TRUNC4(dst_mtu(dst)); |
| 288 | change = t->pathmtu != pmtu; |
| 289 | } |
| 290 | t->pathmtu = pmtu; |
| 291 | |
| 292 | return change; |
| 293 | } |
| 294 | |
| 295 | /* Caches the dst entry and source address for a transport's destination |
| 296 | * address. |
| 297 | */ |
| 298 | void sctp_transport_route(struct sctp_transport *transport, |
| 299 | union sctp_addr *saddr, struct sctp_sock *opt) |
| 300 | { |
| 301 | struct sctp_association *asoc = transport->asoc; |
| 302 | struct sctp_af *af = transport->af_specific; |
| 303 | |
| 304 | af->get_dst(transport, saddr, &transport->fl, sctp_opt2sk(opt)); |
| 305 | |
| 306 | if (saddr) |
| 307 | memcpy(&transport->saddr, saddr, sizeof(union sctp_addr)); |
| 308 | else |
| 309 | af->get_saddr(opt, transport, &transport->fl); |
| 310 | |
| 311 | if ((transport->param_flags & SPP_PMTUD_DISABLE) && transport->pathmtu) { |
| 312 | return; |
| 313 | } |
| 314 | if (transport->dst) { |
| 315 | transport->pathmtu = SCTP_TRUNC4(dst_mtu(transport->dst)); |
| 316 | |
| 317 | /* Initialize sk->sk_rcv_saddr, if the transport is the |
| 318 | * association's active path for getsockname(). |
| 319 | */ |
| 320 | if (asoc && (!asoc->peer.primary_path || |
| 321 | (transport == asoc->peer.active_path))) |
| 322 | opt->pf->to_sk_saddr(&transport->saddr, |
| 323 | asoc->base.sk); |
| 324 | } else |
| 325 | transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT; |
| 326 | } |
| 327 | |
| 328 | /* Hold a reference to a transport. */ |
| 329 | int sctp_transport_hold(struct sctp_transport *transport) |
| 330 | { |
| 331 | return refcount_inc_not_zero(&transport->refcnt); |
| 332 | } |
| 333 | |
| 334 | /* Release a reference to a transport and clean up |
| 335 | * if there are no more references. |
| 336 | */ |
| 337 | void sctp_transport_put(struct sctp_transport *transport) |
| 338 | { |
| 339 | if (refcount_dec_and_test(&transport->refcnt)) |
| 340 | sctp_transport_destroy(transport); |
| 341 | } |
| 342 | |
| 343 | /* Update transport's RTO based on the newly calculated RTT. */ |
| 344 | void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt) |
| 345 | { |
| 346 | if (unlikely(!tp->rto_pending)) |
| 347 | /* We should not be doing any RTO updates unless rto_pending is set. */ |
| 348 | pr_debug("%s: rto_pending not set on transport %p!\n", __func__, tp); |
| 349 | |
| 350 | if (tp->rttvar || tp->srtt) { |
| 351 | struct net *net = sock_net(tp->asoc->base.sk); |
| 352 | /* 6.3.1 C3) When a new RTT measurement R' is made, set |
| 353 | * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'| |
| 354 | * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R' |
| 355 | */ |
| 356 | |
| 357 | /* Note: The above algorithm has been rewritten to |
| 358 | * express rto_beta and rto_alpha as inverse powers |
| 359 | * of two. |
| 360 | * For example, assuming the default value of RTO.Alpha of |
| 361 | * 1/8, rto_alpha would be expressed as 3. |
| 362 | */ |
| 363 | tp->rttvar = tp->rttvar - (tp->rttvar >> net->sctp.rto_beta) |
| 364 | + (((__u32)abs((__s64)tp->srtt - (__s64)rtt)) >> net->sctp.rto_beta); |
| 365 | tp->srtt = tp->srtt - (tp->srtt >> net->sctp.rto_alpha) |
| 366 | + (rtt >> net->sctp.rto_alpha); |
| 367 | } else { |
| 368 | /* 6.3.1 C2) When the first RTT measurement R is made, set |
| 369 | * SRTT <- R, RTTVAR <- R/2. |
| 370 | */ |
| 371 | tp->srtt = rtt; |
| 372 | tp->rttvar = rtt >> 1; |
| 373 | } |
| 374 | |
| 375 | /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then |
| 376 | * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY. |
| 377 | */ |
| 378 | if (tp->rttvar == 0) |
| 379 | tp->rttvar = SCTP_CLOCK_GRANULARITY; |
| 380 | |
| 381 | /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */ |
| 382 | tp->rto = tp->srtt + (tp->rttvar << 2); |
| 383 | |
| 384 | /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min |
| 385 | * seconds then it is rounded up to RTO.Min seconds. |
| 386 | */ |
| 387 | if (tp->rto < tp->asoc->rto_min) |
| 388 | tp->rto = tp->asoc->rto_min; |
| 389 | |
| 390 | /* 6.3.1 C7) A maximum value may be placed on RTO provided it is |
| 391 | * at least RTO.max seconds. |
| 392 | */ |
| 393 | if (tp->rto > tp->asoc->rto_max) |
| 394 | tp->rto = tp->asoc->rto_max; |
| 395 | |
| 396 | sctp_max_rto(tp->asoc, tp); |
| 397 | tp->rtt = rtt; |
| 398 | |
| 399 | /* Reset rto_pending so that a new RTT measurement is started when a |
| 400 | * new data chunk is sent. |
| 401 | */ |
| 402 | tp->rto_pending = 0; |
| 403 | |
| 404 | pr_debug("%s: transport:%p, rtt:%d, srtt:%d rttvar:%d, rto:%ld\n", |
| 405 | __func__, tp, rtt, tp->srtt, tp->rttvar, tp->rto); |
| 406 | } |
| 407 | |
| 408 | /* This routine updates the transport's cwnd and partial_bytes_acked |
| 409 | * parameters based on the bytes acked in the received SACK. |
| 410 | */ |
| 411 | void sctp_transport_raise_cwnd(struct sctp_transport *transport, |
| 412 | __u32 sack_ctsn, __u32 bytes_acked) |
| 413 | { |
| 414 | struct sctp_association *asoc = transport->asoc; |
| 415 | __u32 cwnd, ssthresh, flight_size, pba, pmtu; |
| 416 | |
| 417 | cwnd = transport->cwnd; |
| 418 | flight_size = transport->flight_size; |
| 419 | |
| 420 | /* See if we need to exit Fast Recovery first */ |
| 421 | if (asoc->fast_recovery && |
| 422 | TSN_lte(asoc->fast_recovery_exit, sack_ctsn)) |
| 423 | asoc->fast_recovery = 0; |
| 424 | |
| 425 | ssthresh = transport->ssthresh; |
| 426 | pba = transport->partial_bytes_acked; |
| 427 | pmtu = transport->asoc->pathmtu; |
| 428 | |
| 429 | if (cwnd <= ssthresh) { |
| 430 | /* RFC 4960 7.2.1 |
| 431 | * o When cwnd is less than or equal to ssthresh, an SCTP |
| 432 | * endpoint MUST use the slow-start algorithm to increase |
| 433 | * cwnd only if the current congestion window is being fully |
| 434 | * utilized, an incoming SACK advances the Cumulative TSN |
| 435 | * Ack Point, and the data sender is not in Fast Recovery. |
| 436 | * Only when these three conditions are met can the cwnd be |
| 437 | * increased; otherwise, the cwnd MUST not be increased. |
| 438 | * If these conditions are met, then cwnd MUST be increased |
| 439 | * by, at most, the lesser of 1) the total size of the |
| 440 | * previously outstanding DATA chunk(s) acknowledged, and |
| 441 | * 2) the destination's path MTU. This upper bound protects |
| 442 | * against the ACK-Splitting attack outlined in [SAVAGE99]. |
| 443 | */ |
| 444 | if (asoc->fast_recovery) |
| 445 | return; |
| 446 | |
| 447 | /* The appropriate cwnd increase algorithm is performed |
| 448 | * if, and only if the congestion window is being fully |
| 449 | * utilized. Note that RFC4960 Errata 3.22 removed the |
| 450 | * other condition on ctsn moving. |
| 451 | */ |
| 452 | if (flight_size < cwnd) |
| 453 | return; |
| 454 | |
| 455 | if (bytes_acked > pmtu) |
| 456 | cwnd += pmtu; |
| 457 | else |
| 458 | cwnd += bytes_acked; |
| 459 | |
| 460 | pr_debug("%s: slow start: transport:%p, bytes_acked:%d, " |
| 461 | "cwnd:%d, ssthresh:%d, flight_size:%d, pba:%d\n", |
| 462 | __func__, transport, bytes_acked, cwnd, ssthresh, |
| 463 | flight_size, pba); |
| 464 | } else { |
| 465 | /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh, |
| 466 | * upon each SACK arrival, increase partial_bytes_acked |
| 467 | * by the total number of bytes of all new chunks |
| 468 | * acknowledged in that SACK including chunks |
| 469 | * acknowledged by the new Cumulative TSN Ack and by Gap |
| 470 | * Ack Blocks. (updated by RFC4960 Errata 3.22) |
| 471 | * |
| 472 | * When partial_bytes_acked is greater than cwnd and |
| 473 | * before the arrival of the SACK the sender had less |
| 474 | * bytes of data outstanding than cwnd (i.e., before |
| 475 | * arrival of the SACK, flightsize was less than cwnd), |
| 476 | * reset partial_bytes_acked to cwnd. (RFC 4960 Errata |
| 477 | * 3.26) |
| 478 | * |
| 479 | * When partial_bytes_acked is equal to or greater than |
| 480 | * cwnd and before the arrival of the SACK the sender |
| 481 | * had cwnd or more bytes of data outstanding (i.e., |
| 482 | * before arrival of the SACK, flightsize was greater |
| 483 | * than or equal to cwnd), partial_bytes_acked is reset |
| 484 | * to (partial_bytes_acked - cwnd). Next, cwnd is |
| 485 | * increased by MTU. (RFC 4960 Errata 3.12) |
| 486 | */ |
| 487 | pba += bytes_acked; |
| 488 | if (pba > cwnd && flight_size < cwnd) |
| 489 | pba = cwnd; |
| 490 | if (pba >= cwnd && flight_size >= cwnd) { |
| 491 | pba = pba - cwnd; |
| 492 | cwnd += pmtu; |
| 493 | } |
| 494 | |
| 495 | pr_debug("%s: congestion avoidance: transport:%p, " |
| 496 | "bytes_acked:%d, cwnd:%d, ssthresh:%d, " |
| 497 | "flight_size:%d, pba:%d\n", __func__, |
| 498 | transport, bytes_acked, cwnd, ssthresh, |
| 499 | flight_size, pba); |
| 500 | } |
| 501 | |
| 502 | transport->cwnd = cwnd; |
| 503 | transport->partial_bytes_acked = pba; |
| 504 | } |
| 505 | |
| 506 | /* This routine is used to lower the transport's cwnd when congestion is |
| 507 | * detected. |
| 508 | */ |
| 509 | void sctp_transport_lower_cwnd(struct sctp_transport *transport, |
| 510 | enum sctp_lower_cwnd reason) |
| 511 | { |
| 512 | struct sctp_association *asoc = transport->asoc; |
| 513 | |
| 514 | switch (reason) { |
| 515 | case SCTP_LOWER_CWND_T3_RTX: |
| 516 | /* RFC 2960 Section 7.2.3, sctpimpguide |
| 517 | * When the T3-rtx timer expires on an address, SCTP should |
| 518 | * perform slow start by: |
| 519 | * ssthresh = max(cwnd/2, 4*MTU) |
| 520 | * cwnd = 1*MTU |
| 521 | * partial_bytes_acked = 0 |
| 522 | */ |
| 523 | transport->ssthresh = max(transport->cwnd/2, |
| 524 | 4*asoc->pathmtu); |
| 525 | transport->cwnd = asoc->pathmtu; |
| 526 | |
| 527 | /* T3-rtx also clears fast recovery */ |
| 528 | asoc->fast_recovery = 0; |
| 529 | break; |
| 530 | |
| 531 | case SCTP_LOWER_CWND_FAST_RTX: |
| 532 | /* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the |
| 533 | * destination address(es) to which the missing DATA chunks |
| 534 | * were last sent, according to the formula described in |
| 535 | * Section 7.2.3. |
| 536 | * |
| 537 | * RFC 2960 7.2.3, sctpimpguide Upon detection of packet |
| 538 | * losses from SACK (see Section 7.2.4), An endpoint |
| 539 | * should do the following: |
| 540 | * ssthresh = max(cwnd/2, 4*MTU) |
| 541 | * cwnd = ssthresh |
| 542 | * partial_bytes_acked = 0 |
| 543 | */ |
| 544 | if (asoc->fast_recovery) |
| 545 | return; |
| 546 | |
| 547 | /* Mark Fast recovery */ |
| 548 | asoc->fast_recovery = 1; |
| 549 | asoc->fast_recovery_exit = asoc->next_tsn - 1; |
| 550 | |
| 551 | transport->ssthresh = max(transport->cwnd/2, |
| 552 | 4*asoc->pathmtu); |
| 553 | transport->cwnd = transport->ssthresh; |
| 554 | break; |
| 555 | |
| 556 | case SCTP_LOWER_CWND_ECNE: |
| 557 | /* RFC 2481 Section 6.1.2. |
| 558 | * If the sender receives an ECN-Echo ACK packet |
| 559 | * then the sender knows that congestion was encountered in the |
| 560 | * network on the path from the sender to the receiver. The |
| 561 | * indication of congestion should be treated just as a |
| 562 | * congestion loss in non-ECN Capable TCP. That is, the TCP |
| 563 | * source halves the congestion window "cwnd" and reduces the |
| 564 | * slow start threshold "ssthresh". |
| 565 | * A critical condition is that TCP does not react to |
| 566 | * congestion indications more than once every window of |
| 567 | * data (or more loosely more than once every round-trip time). |
| 568 | */ |
| 569 | if (time_after(jiffies, transport->last_time_ecne_reduced + |
| 570 | transport->rtt)) { |
| 571 | transport->ssthresh = max(transport->cwnd/2, |
| 572 | 4*asoc->pathmtu); |
| 573 | transport->cwnd = transport->ssthresh; |
| 574 | transport->last_time_ecne_reduced = jiffies; |
| 575 | } |
| 576 | break; |
| 577 | |
| 578 | case SCTP_LOWER_CWND_INACTIVE: |
| 579 | /* RFC 2960 Section 7.2.1, sctpimpguide |
| 580 | * When the endpoint does not transmit data on a given |
| 581 | * transport address, the cwnd of the transport address |
| 582 | * should be adjusted to max(cwnd/2, 4*MTU) per RTO. |
| 583 | * NOTE: Although the draft recommends that this check needs |
| 584 | * to be done every RTO interval, we do it every hearbeat |
| 585 | * interval. |
| 586 | */ |
| 587 | transport->cwnd = max(transport->cwnd/2, |
| 588 | 4*asoc->pathmtu); |
| 589 | /* RFC 4960 Errata 3.27.2: also adjust sshthresh */ |
| 590 | transport->ssthresh = transport->cwnd; |
| 591 | break; |
| 592 | } |
| 593 | |
| 594 | transport->partial_bytes_acked = 0; |
| 595 | |
| 596 | pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d\n", |
| 597 | __func__, transport, reason, transport->cwnd, |
| 598 | transport->ssthresh); |
| 599 | } |
| 600 | |
| 601 | /* Apply Max.Burst limit to the congestion window: |
| 602 | * sctpimpguide-05 2.14.2 |
| 603 | * D) When the time comes for the sender to |
| 604 | * transmit new DATA chunks, the protocol parameter Max.Burst MUST |
| 605 | * first be applied to limit how many new DATA chunks may be sent. |
| 606 | * The limit is applied by adjusting cwnd as follows: |
| 607 | * if ((flightsize+ Max.Burst * MTU) < cwnd) |
| 608 | * cwnd = flightsize + Max.Burst * MTU |
| 609 | */ |
| 610 | |
| 611 | void sctp_transport_burst_limited(struct sctp_transport *t) |
| 612 | { |
| 613 | struct sctp_association *asoc = t->asoc; |
| 614 | u32 old_cwnd = t->cwnd; |
| 615 | u32 max_burst_bytes; |
| 616 | |
| 617 | if (t->burst_limited || asoc->max_burst == 0) |
| 618 | return; |
| 619 | |
| 620 | max_burst_bytes = t->flight_size + (asoc->max_burst * asoc->pathmtu); |
| 621 | if (max_burst_bytes < old_cwnd) { |
| 622 | t->cwnd = max_burst_bytes; |
| 623 | t->burst_limited = old_cwnd; |
| 624 | } |
| 625 | } |
| 626 | |
| 627 | /* Restore the old cwnd congestion window, after the burst had it's |
| 628 | * desired effect. |
| 629 | */ |
| 630 | void sctp_transport_burst_reset(struct sctp_transport *t) |
| 631 | { |
| 632 | if (t->burst_limited) { |
| 633 | t->cwnd = t->burst_limited; |
| 634 | t->burst_limited = 0; |
| 635 | } |
| 636 | } |
| 637 | |
| 638 | /* What is the next timeout value for this transport? */ |
| 639 | unsigned long sctp_transport_timeout(struct sctp_transport *trans) |
| 640 | { |
| 641 | /* RTO + timer slack +/- 50% of RTO */ |
| 642 | unsigned long timeout = trans->rto >> 1; |
| 643 | |
| 644 | if (trans->state != SCTP_UNCONFIRMED && |
| 645 | trans->state != SCTP_PF) |
| 646 | timeout += trans->hbinterval; |
| 647 | |
| 648 | return max_t(unsigned long, timeout, HZ / 5); |
| 649 | } |
| 650 | |
| 651 | /* Reset transport variables to their initial values */ |
| 652 | void sctp_transport_reset(struct sctp_transport *t) |
| 653 | { |
| 654 | struct sctp_association *asoc = t->asoc; |
| 655 | |
| 656 | /* RFC 2960 (bis), Section 5.2.4 |
| 657 | * All the congestion control parameters (e.g., cwnd, ssthresh) |
| 658 | * related to this peer MUST be reset to their initial values |
| 659 | * (see Section 6.2.1) |
| 660 | */ |
| 661 | t->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380)); |
| 662 | t->burst_limited = 0; |
| 663 | t->ssthresh = asoc->peer.i.a_rwnd; |
| 664 | t->rto = asoc->rto_initial; |
| 665 | sctp_max_rto(asoc, t); |
| 666 | t->rtt = 0; |
| 667 | t->srtt = 0; |
| 668 | t->rttvar = 0; |
| 669 | |
| 670 | /* Reset these additional variables so that we have a clean slate. */ |
| 671 | t->partial_bytes_acked = 0; |
| 672 | t->flight_size = 0; |
| 673 | t->error_count = 0; |
| 674 | t->rto_pending = 0; |
| 675 | t->hb_sent = 0; |
| 676 | |
| 677 | /* Initialize the state information for SFR-CACC */ |
| 678 | t->cacc.changeover_active = 0; |
| 679 | t->cacc.cycling_changeover = 0; |
| 680 | t->cacc.next_tsn_at_change = 0; |
| 681 | t->cacc.cacc_saw_newack = 0; |
| 682 | } |
| 683 | |
| 684 | /* Schedule retransmission on the given transport */ |
| 685 | void sctp_transport_immediate_rtx(struct sctp_transport *t) |
| 686 | { |
| 687 | /* Stop pending T3_rtx_timer */ |
| 688 | if (del_timer(&t->T3_rtx_timer)) |
| 689 | sctp_transport_put(t); |
| 690 | |
| 691 | sctp_retransmit(&t->asoc->outqueue, t, SCTP_RTXR_T3_RTX); |
| 692 | if (!timer_pending(&t->T3_rtx_timer)) { |
| 693 | if (!mod_timer(&t->T3_rtx_timer, jiffies + t->rto)) |
| 694 | sctp_transport_hold(t); |
| 695 | } |
| 696 | } |
| 697 | |
| 698 | /* Drop dst */ |
| 699 | void sctp_transport_dst_release(struct sctp_transport *t) |
| 700 | { |
| 701 | dst_release(t->dst); |
| 702 | t->dst = NULL; |
| 703 | t->dst_pending_confirm = 0; |
| 704 | } |
| 705 | |
| 706 | /* Schedule neighbour confirm */ |
| 707 | void sctp_transport_dst_confirm(struct sctp_transport *t) |
| 708 | { |
| 709 | t->dst_pending_confirm = 1; |
| 710 | } |