net/sctp/transport.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* SCTP kernel implementation
   3  * Copyright (c) 1999-2000 Cisco, Inc.
   4  * Copyright (c) 1999-2001 Motorola, Inc.
   5  * Copyright (c) 2001-2003 International Business Machines Corp.
   6  * Copyright (c) 2001 Intel Corp.
   7  * Copyright (c) 2001 La Monte H.P. Yarroll
   8  *
   9  * This file is part of the SCTP kernel implementation
  10  *
  11  * This module provides the abstraction for an SCTP transport representing
  12  * a remote transport address.  For local transport addresses, we just use
  13  * union sctp_addr.
  14  *
  15  * Please send any bug reports or fixes you make to the
  16  * email address(es):
  17  *    lksctp developers <linux-sctp@vger.kernel.org>
  18  *
  19  * Written or modified by:
  20  *    La Monte H.P. Yarroll <piggy@acm.org>
  21  *    Karl Knutson          <karl@athena.chicago.il.us>
  22  *    Jon Grimm             <jgrimm@us.ibm.com>
  23  *    Xingang Guo           <xingang.guo@intel.com>
  24  *    Hui Huang             <hui.huang@nokia.com>
  25  *    Sridhar Samudrala     <sri@us.ibm.com>
  26  *    Ardelle Fan           <ardelle.fan@intel.com>
  27  */
  28
  29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  30
  31 #include <linux/slab.h>
  32 #include <linux/types.h>
  33 #include <linux/random.h>
  34 #include <net/sctp/sctp.h>
  35 #include <net/sctp/sm.h>
  36
  37 /* 1st Level Abstractions.  */
  38
  39 /* Initialize a new transport from provided memory.  */
  40 static struct sctp_transport *sctp_transport_init(struct net *net,
  41                                                   struct sctp_transport *peer,
  42                                                   const union sctp_addr *addr,
  43                                                   gfp_t gfp)
  44 {
  45         /* Copy in the address.  */
  46         peer->af_specific = sctp_get_af_specific(addr->sa.sa_family);
  47         memcpy(&peer->ipaddr, addr, peer->af_specific->sockaddr_len);
  48         memset(&peer->saddr, 0, sizeof(union sctp_addr));
  49
  50         peer->sack_generation = 0;
  51
  52         /* From 6.3.1 RTO Calculation:
  53          *
  54          * C1) Until an RTT measurement has been made for a packet sent to the
  55          * given destination transport address, set RTO to the protocol
  56          * parameter 'RTO.Initial'.
  57          */
  58         peer->rto = msecs_to_jiffies(net->sctp.rto_initial);
  59
  60         peer->last_time_heard = 0;
  61         peer->last_time_ecne_reduced = jiffies;
  62
  63         peer->param_flags = SPP_HB_DISABLE |
  64                             SPP_PMTUD_ENABLE |
  65                             SPP_SACKDELAY_ENABLE;
  66
  67         /* Initialize the default path max_retrans.  */
  68         peer->pathmaxrxt  = net->sctp.max_retrans_path;
  69         peer->pf_retrans  = net->sctp.pf_retrans;
  70
  71         INIT_LIST_HEAD(&peer->transmitted);
  72         INIT_LIST_HEAD(&peer->send_ready);
  73         INIT_LIST_HEAD(&peer->transports);
  74
  75         timer_setup(&peer->T3_rtx_timer, sctp_generate_t3_rtx_event, 0);
  76         timer_setup(&peer->hb_timer, sctp_generate_heartbeat_event, 0);
  77         timer_setup(&peer->reconf_timer, sctp_generate_reconf_event, 0);
  78         timer_setup(&peer->probe_timer, sctp_generate_probe_event, 0);
  79         timer_setup(&peer->proto_unreach_timer,
  80                     sctp_generate_proto_unreach_event, 0);
  81
  82         /* Initialize the 64-bit random nonce sent with heartbeat. */
  83         get_random_bytes(&peer->hb_nonce, sizeof(peer->hb_nonce));
  84
  85         refcount_set(&peer->refcnt, 1);
  86
  87         return peer;
  88 }
  89
  90 /* Allocate and initialize a new transport.  */
  91 struct sctp_transport *sctp_transport_new(struct net *net,
  92                                           const union sctp_addr *addr,
  93                                           gfp_t gfp)
  94 {
  95         struct sctp_transport *transport;
  96
  97         transport = kzalloc(sizeof(*transport), gfp);
  98         if (!transport)
  99                 goto fail;
 100
 101         if (!sctp_transport_init(net, transport, addr, gfp))
 102                 goto fail_init;
 103
 104         SCTP_DBG_OBJCNT_INC(transport);
 105
 106         return transport;
 107
 108 fail_init:
 109         kfree(transport);
 110
 111 fail:
 112         return NULL;
 113 }
 114
 115 /* This transport is no longer needed.  Free up if possible, or
 116  * delay until it last reference count.
 117  */
 118 void sctp_transport_free(struct sctp_transport *transport)
 119 {
 120         /* Try to delete the heartbeat timer.  */
 121         if (del_timer(&transport->hb_timer))
 122                 sctp_transport_put(transport);
 123
 124         /* Delete the T3_rtx timer if it's active.
 125          * There is no point in not doing this now and letting
 126          * structure hang around in memory since we know
 127          * the transport is going away.
 128          */
 129         if (del_timer(&transport->T3_rtx_timer))
 130                 sctp_transport_put(transport);
 131
 132         if (del_timer(&transport->reconf_timer))
 133                 sctp_transport_put(transport);
 134
 135         if (del_timer(&transport->probe_timer))
 136                 sctp_transport_put(transport);
 137
 138         /* Delete the ICMP proto unreachable timer if it's active. */
 139         if (del_timer(&transport->proto_unreach_timer))
 140                 sctp_transport_put(transport);
 141
 142         sctp_transport_put(transport);
 143 }
 144
 145 static void sctp_transport_destroy_rcu(struct rcu_head *head)
 146 {
 147         struct sctp_transport *transport;
 148
 149         transport = container_of(head, struct sctp_transport, rcu);
 150
 151         dst_release(transport->dst);
 152         kfree(transport);
 153         SCTP_DBG_OBJCNT_DEC(transport);
 154 }
 155
 156 /* Destroy the transport data structure.
 157  * Assumes there are no more users of this structure.
 158  */
 159 static void sctp_transport_destroy(struct sctp_transport *transport)
 160 {
 161         if (unlikely(refcount_read(&transport->refcnt))) {
 162                 WARN(1, "Attempt to destroy undead transport %p!\n", transport);
 163                 return;
 164         }
 165
 166         sctp_packet_free(&transport->packet);
 167
 168         if (transport->asoc)
 169                 sctp_association_put(transport->asoc);
 170
 171         call_rcu(&transport->rcu, sctp_transport_destroy_rcu);
 172 }
 173
 174 /* Start T3_rtx timer if it is not already running and update the heartbeat
 175  * timer.  This routine is called every time a DATA chunk is sent.
 176  */
 177 void sctp_transport_reset_t3_rtx(struct sctp_transport *transport)
 178 {
 179         /* RFC 2960 6.3.2 Retransmission Timer Rules
 180          *
 181          * R1) Every time a DATA chunk is sent to any address(including a
 182          * retransmission), if the T3-rtx timer of that address is not running
 183          * start it running so that it will expire after the RTO of that
 184          * address.
 185          */
 186
 187         if (!timer_pending(&transport->T3_rtx_timer))
 188                 if (!mod_timer(&transport->T3_rtx_timer,
 189                                jiffies + transport->rto))
 190                         sctp_transport_hold(transport);
 191 }
 192
 193 void sctp_transport_reset_hb_timer(struct sctp_transport *transport)
 194 {
 195         unsigned long expires;
 196
 197         /* When a data chunk is sent, reset the heartbeat interval.  */
 198         expires = jiffies + sctp_transport_timeout(transport);
 199         if ((time_before(transport->hb_timer.expires, expires) ||
 200              !timer_pending(&transport->hb_timer)) &&
 201             !mod_timer(&transport->hb_timer,
 202                        expires + prandom_u32_max(transport->rto)))
 203                 sctp_transport_hold(transport);
 204 }
 205
 206 void sctp_transport_reset_reconf_timer(struct sctp_transport *transport)
 207 {
 208         if (!timer_pending(&transport->reconf_timer))
 209                 if (!mod_timer(&transport->reconf_timer,
 210                                jiffies + transport->rto))
 211                         sctp_transport_hold(transport);
 212 }
 213
 214 void sctp_transport_reset_probe_timer(struct sctp_transport *transport)
 215 {
 216         int scale = 1;
 217
 218         if (timer_pending(&transport->probe_timer))
 219                 return;
 220         if (transport->pl.state == SCTP_PL_COMPLETE &&
 221             transport->pl.probe_count == 1)
 222                 scale = 30; /* works as PMTU_RAISE_TIMER */
 223         if (!mod_timer(&transport->probe_timer,
 224                        jiffies + transport->probe_interval * scale))
 225                 sctp_transport_hold(transport);
 226 }
 227
 228 /* This transport has been assigned to an association.
 229  * Initialize fields from the association or from the sock itself.
 230  * Register the reference count in the association.
 231  */
 232 void sctp_transport_set_owner(struct sctp_transport *transport,
 233                               struct sctp_association *asoc)
 234 {
 235         transport->asoc = asoc;
 236         sctp_association_hold(asoc);
 237 }
 238
 239 /* Initialize the pmtu of a transport. */
 240 void sctp_transport_pmtu(struct sctp_transport *transport, struct sock *sk)
 241 {
 242         /* If we don't have a fresh route, look one up */
 243         if (!transport->dst || transport->dst->obsolete) {
 244                 sctp_transport_dst_release(transport);
 245                 transport->af_specific->get_dst(transport, &transport->saddr,
 246                                                 &transport->fl, sk);
 247         }
 248
 249         if (transport->param_flags & SPP_PMTUD_DISABLE) {
 250                 struct sctp_association *asoc = transport->asoc;
 251
 252                 if (!transport->pathmtu && asoc && asoc->pathmtu)
 253                         transport->pathmtu = asoc->pathmtu;
 254                 if (transport->pathmtu)
 255                         return;
 256         }
 257
 258         if (transport->dst)
 259                 transport->pathmtu = sctp_dst_mtu(transport->dst);
 260         else
 261                 transport->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
 262 }
 263
 264 void sctp_transport_pl_send(struct sctp_transport *t)
 265 {
 266         pr_debug("%s: PLPMTUD: transport: %p, state: %d, pmtu: %d, size: %d, high: %d\n",
 267                  __func__, t, t->pl.state, t->pl.pmtu, t->pl.probe_size, t->pl.probe_high);
 268
 269         if (t->pl.probe_count < SCTP_MAX_PROBES) {
 270                 t->pl.probe_count++;
 271                 return;
 272         }
 273
 274         if (t->pl.state == SCTP_PL_BASE) {
 275                 if (t->pl.probe_size == SCTP_BASE_PLPMTU) { /* BASE_PLPMTU Confirmation Failed */
 276                         t->pl.state = SCTP_PL_ERROR; /* Base -> Error */
 277
 278                         t->pl.pmtu = SCTP_MIN_PLPMTU;
 279                         t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
 280                         sctp_assoc_sync_pmtu(t->asoc);
 281                 }
 282         } else if (t->pl.state == SCTP_PL_SEARCH) {
 283                 if (t->pl.pmtu == t->pl.probe_size) { /* Black Hole Detected */
 284                         t->pl.state = SCTP_PL_BASE;  /* Search -> Base */
 285                         t->pl.probe_size = SCTP_BASE_PLPMTU;
 286                         t->pl.probe_high = 0;
 287
 288                         t->pl.pmtu = SCTP_BASE_PLPMTU;
 289                         t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
 290                         sctp_assoc_sync_pmtu(t->asoc);
 291                 } else { /* Normal probe failure. */
 292                         t->pl.probe_high = t->pl.probe_size;
 293                         t->pl.probe_size = t->pl.pmtu;
 294                 }
 295         } else if (t->pl.state == SCTP_PL_COMPLETE) {
 296                 if (t->pl.pmtu == t->pl.probe_size) { /* Black Hole Detected */
 297                         t->pl.state = SCTP_PL_BASE;  /* Search Complete -> Base */
 298                         t->pl.probe_size = SCTP_BASE_PLPMTU;
 299
 300                         t->pl.pmtu = SCTP_BASE_PLPMTU;
 301                         t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
 302                         sctp_assoc_sync_pmtu(t->asoc);
 303                 }
 304         }
 305         t->pl.probe_count = 1;
 306 }
 307
 308 void sctp_transport_pl_recv(struct sctp_transport *t)
 309 {
 310         pr_debug("%s: PLPMTUD: transport: %p, state: %d, pmtu: %d, size: %d, high: %d\n",
 311                  __func__, t, t->pl.state, t->pl.pmtu, t->pl.probe_size, t->pl.probe_high);
 312
 313         t->pl.pmtu = t->pl.probe_size;
 314         t->pl.probe_count = 0;
 315         if (t->pl.state == SCTP_PL_BASE) {
 316                 t->pl.state = SCTP_PL_SEARCH; /* Base -> Search */
 317                 t->pl.probe_size += SCTP_PL_BIG_STEP;
 318         } else if (t->pl.state == SCTP_PL_ERROR) {
 319                 t->pl.state = SCTP_PL_SEARCH; /* Error -> Search */
 320
 321                 t->pl.pmtu = t->pl.probe_size;
 322                 t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
 323                 sctp_assoc_sync_pmtu(t->asoc);
 324                 t->pl.probe_size += SCTP_PL_BIG_STEP;
 325         } else if (t->pl.state == SCTP_PL_SEARCH) {
 326                 if (!t->pl.probe_high) {
 327                         t->pl.probe_size = min(t->pl.probe_size + SCTP_PL_BIG_STEP,
 328                                                SCTP_MAX_PLPMTU);
 329                         return;
 330                 }
 331                 t->pl.probe_size += SCTP_PL_MIN_STEP;
 332                 if (t->pl.probe_size >= t->pl.probe_high) {
 333                         t->pl.probe_high = 0;
 334                         t->pl.state = SCTP_PL_COMPLETE; /* Search -> Search Complete */
 335
 336                         t->pl.probe_size = t->pl.pmtu;
 337                         t->pathmtu = t->pl.pmtu + sctp_transport_pl_hlen(t);
 338                         sctp_assoc_sync_pmtu(t->asoc);
 339                 }
 340         } else if (t->pl.state == SCTP_PL_COMPLETE) {
 341                 t->pl.state = SCTP_PL_SEARCH; /* Search Complete -> Search */
 342                 t->pl.probe_size += SCTP_PL_MIN_STEP;
 343         }
 344 }
 345
 346 bool sctp_transport_update_pmtu(struct sctp_transport *t, u32 pmtu)
 347 {
 348         struct dst_entry *dst = sctp_transport_dst_check(t);
 349         struct sock *sk = t->asoc->base.sk;
 350         bool change = true;
 351
 352         if (unlikely(pmtu < SCTP_DEFAULT_MINSEGMENT)) {
 353                 pr_warn_ratelimited("%s: Reported pmtu %d too low, using default minimum of %d\n",
 354                                     __func__, pmtu, SCTP_DEFAULT_MINSEGMENT);
 355                 /* Use default minimum segment instead */
 356                 pmtu = SCTP_DEFAULT_MINSEGMENT;
 357         }
 358         pmtu = SCTP_TRUNC4(pmtu);
 359
 360         if (dst) {
 361                 struct sctp_pf *pf = sctp_get_pf_specific(dst->ops->family);
 362                 union sctp_addr addr;
 363
 364                 pf->af->from_sk(&addr, sk);
 365                 pf->to_sk_daddr(&t->ipaddr, sk);
 366                 dst->ops->update_pmtu(dst, sk, NULL, pmtu, true);
 367                 pf->to_sk_daddr(&addr, sk);
 368
 369                 dst = sctp_transport_dst_check(t);
 370         }
 371
 372         if (!dst) {
 373                 t->af_specific->get_dst(t, &t->saddr, &t->fl, sk);
 374                 dst = t->dst;
 375         }
 376
 377         if (dst) {
 378                 /* Re-fetch, as under layers may have a higher minimum size */
 379                 pmtu = sctp_dst_mtu(dst);
 380                 change = t->pathmtu != pmtu;
 381         }
 382         t->pathmtu = pmtu;
 383
 384         return change;
 385 }
 386
 387 /* Caches the dst entry and source address for a transport's destination
 388  * address.
 389  */
 390 void sctp_transport_route(struct sctp_transport *transport,
 391                           union sctp_addr *saddr, struct sctp_sock *opt)
 392 {
 393         struct sctp_association *asoc = transport->asoc;
 394         struct sctp_af *af = transport->af_specific;
 395
 396         sctp_transport_dst_release(transport);
 397         af->get_dst(transport, saddr, &transport->fl, sctp_opt2sk(opt));
 398
 399         if (saddr)
 400                 memcpy(&transport->saddr, saddr, sizeof(union sctp_addr));
 401         else
 402                 af->get_saddr(opt, transport, &transport->fl);
 403
 404         sctp_transport_pmtu(transport, sctp_opt2sk(opt));
 405
 406         /* Initialize sk->sk_rcv_saddr, if the transport is the
 407          * association's active path for getsockname().
 408          */
 409         if (transport->dst && asoc &&
 410             (!asoc->peer.primary_path || transport == asoc->peer.active_path))
 411                 opt->pf->to_sk_saddr(&transport->saddr, asoc->base.sk);
 412 }
 413
 414 /* Hold a reference to a transport.  */
 415 int sctp_transport_hold(struct sctp_transport *transport)
 416 {
 417         return refcount_inc_not_zero(&transport->refcnt);
 418 }
 419
 420 /* Release a reference to a transport and clean up
 421  * if there are no more references.
 422  */
 423 void sctp_transport_put(struct sctp_transport *transport)
 424 {
 425         if (refcount_dec_and_test(&transport->refcnt))
 426                 sctp_transport_destroy(transport);
 427 }
 428
 429 /* Update transport's RTO based on the newly calculated RTT. */
 430 void sctp_transport_update_rto(struct sctp_transport *tp, __u32 rtt)
 431 {
 432         if (unlikely(!tp->rto_pending))
 433                 /* We should not be doing any RTO updates unless rto_pending is set.  */
 434                 pr_debug("%s: rto_pending not set on transport %p!\n", __func__, tp);
 435
 436         if (tp->rttvar || tp->srtt) {
 437                 struct net *net = tp->asoc->base.net;
 438                 /* 6.3.1 C3) When a new RTT measurement R' is made, set
 439                  * RTTVAR <- (1 - RTO.Beta) * RTTVAR + RTO.Beta * |SRTT - R'|
 440                  * SRTT <- (1 - RTO.Alpha) * SRTT + RTO.Alpha * R'
 441                  */
 442
 443                 /* Note:  The above algorithm has been rewritten to
 444                  * express rto_beta and rto_alpha as inverse powers
 445                  * of two.
 446                  * For example, assuming the default value of RTO.Alpha of
 447                  * 1/8, rto_alpha would be expressed as 3.
 448                  */
 449                 tp->rttvar = tp->rttvar - (tp->rttvar >> net->sctp.rto_beta)
 450                         + (((__u32)abs((__s64)tp->srtt - (__s64)rtt)) >> net->sctp.rto_beta);
 451                 tp->srtt = tp->srtt - (tp->srtt >> net->sctp.rto_alpha)
 452                         + (rtt >> net->sctp.rto_alpha);
 453         } else {
 454                 /* 6.3.1 C2) When the first RTT measurement R is made, set
 455                  * SRTT <- R, RTTVAR <- R/2.
 456                  */
 457                 tp->srtt = rtt;
 458                 tp->rttvar = rtt >> 1;
 459         }
 460
 461         /* 6.3.1 G1) Whenever RTTVAR is computed, if RTTVAR = 0, then
 462          * adjust RTTVAR <- G, where G is the CLOCK GRANULARITY.
 463          */
 464         if (tp->rttvar == 0)
 465                 tp->rttvar = SCTP_CLOCK_GRANULARITY;
 466
 467         /* 6.3.1 C3) After the computation, update RTO <- SRTT + 4 * RTTVAR. */
 468         tp->rto = tp->srtt + (tp->rttvar << 2);
 469
 470         /* 6.3.1 C6) Whenever RTO is computed, if it is less than RTO.Min
 471          * seconds then it is rounded up to RTO.Min seconds.
 472          */
 473         if (tp->rto < tp->asoc->rto_min)
 474                 tp->rto = tp->asoc->rto_min;
 475
 476         /* 6.3.1 C7) A maximum value may be placed on RTO provided it is
 477          * at least RTO.max seconds.
 478          */
 479         if (tp->rto > tp->asoc->rto_max)
 480                 tp->rto = tp->asoc->rto_max;
 481
 482         sctp_max_rto(tp->asoc, tp);
 483         tp->rtt = rtt;
 484
 485         /* Reset rto_pending so that a new RTT measurement is started when a
 486          * new data chunk is sent.
 487          */
 488         tp->rto_pending = 0;
 489
 490         pr_debug("%s: transport:%p, rtt:%d, srtt:%d rttvar:%d, rto:%ld\n",
 491                  __func__, tp, rtt, tp->srtt, tp->rttvar, tp->rto);
 492 }
 493
 494 /* This routine updates the transport's cwnd and partial_bytes_acked
 495  * parameters based on the bytes acked in the received SACK.
 496  */
 497 void sctp_transport_raise_cwnd(struct sctp_transport *transport,
 498                                __u32 sack_ctsn, __u32 bytes_acked)
 499 {
 500         struct sctp_association *asoc = transport->asoc;
 501         __u32 cwnd, ssthresh, flight_size, pba, pmtu;
 502
 503         cwnd = transport->cwnd;
 504         flight_size = transport->flight_size;
 505
 506         /* See if we need to exit Fast Recovery first */
 507         if (asoc->fast_recovery &&
 508             TSN_lte(asoc->fast_recovery_exit, sack_ctsn))
 509                 asoc->fast_recovery = 0;
 510
 511         ssthresh = transport->ssthresh;
 512         pba = transport->partial_bytes_acked;
 513         pmtu = transport->asoc->pathmtu;
 514
 515         if (cwnd <= ssthresh) {
 516                 /* RFC 4960 7.2.1
 517                  * o  When cwnd is less than or equal to ssthresh, an SCTP
 518                  *    endpoint MUST use the slow-start algorithm to increase
 519                  *    cwnd only if the current congestion window is being fully
 520                  *    utilized, an incoming SACK advances the Cumulative TSN
 521                  *    Ack Point, and the data sender is not in Fast Recovery.
 522                  *    Only when these three conditions are met can the cwnd be
 523                  *    increased; otherwise, the cwnd MUST not be increased.
 524                  *    If these conditions are met, then cwnd MUST be increased
 525                  *    by, at most, the lesser of 1) the total size of the
 526                  *    previously outstanding DATA chunk(s) acknowledged, and
 527                  *    2) the destination's path MTU.  This upper bound protects
 528                  *    against the ACK-Splitting attack outlined in [SAVAGE99].
 529                  */
 530                 if (asoc->fast_recovery)
 531                         return;
 532
 533                 /* The appropriate cwnd increase algorithm is performed
 534                  * if, and only if the congestion window is being fully
 535                  * utilized.  Note that RFC4960 Errata 3.22 removed the
 536                  * other condition on ctsn moving.
 537                  */
 538                 if (flight_size < cwnd)
 539                         return;
 540
 541                 if (bytes_acked > pmtu)
 542                         cwnd += pmtu;
 543                 else
 544                         cwnd += bytes_acked;
 545
 546                 pr_debug("%s: slow start: transport:%p, bytes_acked:%d, "
 547                          "cwnd:%d, ssthresh:%d, flight_size:%d, pba:%d\n",
 548                          __func__, transport, bytes_acked, cwnd, ssthresh,
 549                          flight_size, pba);
 550         } else {
 551                 /* RFC 2960 7.2.2 Whenever cwnd is greater than ssthresh,
 552                  * upon each SACK arrival, increase partial_bytes_acked
 553                  * by the total number of bytes of all new chunks
 554                  * acknowledged in that SACK including chunks
 555                  * acknowledged by the new Cumulative TSN Ack and by Gap
 556                  * Ack Blocks. (updated by RFC4960 Errata 3.22)
 557                  *
 558                  * When partial_bytes_acked is greater than cwnd and
 559                  * before the arrival of the SACK the sender had less
 560                  * bytes of data outstanding than cwnd (i.e., before
 561                  * arrival of the SACK, flightsize was less than cwnd),
 562                  * reset partial_bytes_acked to cwnd. (RFC 4960 Errata
 563                  * 3.26)
 564                  *
 565                  * When partial_bytes_acked is equal to or greater than
 566                  * cwnd and before the arrival of the SACK the sender
 567                  * had cwnd or more bytes of data outstanding (i.e.,
 568                  * before arrival of the SACK, flightsize was greater
 569                  * than or equal to cwnd), partial_bytes_acked is reset
 570                  * to (partial_bytes_acked - cwnd). Next, cwnd is
 571                  * increased by MTU. (RFC 4960 Errata 3.12)
 572                  */
 573                 pba += bytes_acked;
 574                 if (pba > cwnd && flight_size < cwnd)
 575                         pba = cwnd;
 576                 if (pba >= cwnd && flight_size >= cwnd) {
 577                         pba = pba - cwnd;
 578                         cwnd += pmtu;
 579                 }
 580
 581                 pr_debug("%s: congestion avoidance: transport:%p, "
 582                          "bytes_acked:%d, cwnd:%d, ssthresh:%d, "
 583                          "flight_size:%d, pba:%d\n", __func__,
 584                          transport, bytes_acked, cwnd, ssthresh,
 585                          flight_size, pba);
 586         }
 587
 588         transport->cwnd = cwnd;
 589         transport->partial_bytes_acked = pba;
 590 }
 591
 592 /* This routine is used to lower the transport's cwnd when congestion is
 593  * detected.
 594  */
 595 void sctp_transport_lower_cwnd(struct sctp_transport *transport,
 596                                enum sctp_lower_cwnd reason)
 597 {
 598         struct sctp_association *asoc = transport->asoc;
 599
 600         switch (reason) {
 601         case SCTP_LOWER_CWND_T3_RTX:
 602                 /* RFC 2960 Section 7.2.3, sctpimpguide
 603                  * When the T3-rtx timer expires on an address, SCTP should
 604                  * perform slow start by:
 605                  *      ssthresh = max(cwnd/2, 4*MTU)
 606                  *      cwnd = 1*MTU
 607                  *      partial_bytes_acked = 0
 608                  */
 609                 transport->ssthresh = max(transport->cwnd/2,
 610                                           4*asoc->pathmtu);
 611                 transport->cwnd = asoc->pathmtu;
 612
 613                 /* T3-rtx also clears fast recovery */
 614                 asoc->fast_recovery = 0;
 615                 break;
 616
 617         case SCTP_LOWER_CWND_FAST_RTX:
 618                 /* RFC 2960 7.2.4 Adjust the ssthresh and cwnd of the
 619                  * destination address(es) to which the missing DATA chunks
 620                  * were last sent, according to the formula described in
 621                  * Section 7.2.3.
 622                  *
 623                  * RFC 2960 7.2.3, sctpimpguide Upon detection of packet
 624                  * losses from SACK (see Section 7.2.4), An endpoint
 625                  * should do the following:
 626                  *      ssthresh = max(cwnd/2, 4*MTU)
 627                  *      cwnd = ssthresh
 628                  *      partial_bytes_acked = 0
 629                  */
 630                 if (asoc->fast_recovery)
 631                         return;
 632
 633                 /* Mark Fast recovery */
 634                 asoc->fast_recovery = 1;
 635                 asoc->fast_recovery_exit = asoc->next_tsn - 1;
 636
 637                 transport->ssthresh = max(transport->cwnd/2,
 638                                           4*asoc->pathmtu);
 639                 transport->cwnd = transport->ssthresh;
 640                 break;
 641
 642         case SCTP_LOWER_CWND_ECNE:
 643                 /* RFC 2481 Section 6.1.2.
 644                  * If the sender receives an ECN-Echo ACK packet
 645                  * then the sender knows that congestion was encountered in the
 646                  * network on the path from the sender to the receiver. The
 647                  * indication of congestion should be treated just as a
 648                  * congestion loss in non-ECN Capable TCP. That is, the TCP
 649                  * source halves the congestion window "cwnd" and reduces the
 650                  * slow start threshold "ssthresh".
 651                  * A critical condition is that TCP does not react to
 652                  * congestion indications more than once every window of
 653                  * data (or more loosely more than once every round-trip time).
 654                  */
 655                 if (time_after(jiffies, transport->last_time_ecne_reduced +
 656                                         transport->rtt)) {
 657                         transport->ssthresh = max(transport->cwnd/2,
 658                                                   4*asoc->pathmtu);
 659                         transport->cwnd = transport->ssthresh;
 660                         transport->last_time_ecne_reduced = jiffies;
 661                 }
 662                 break;
 663
 664         case SCTP_LOWER_CWND_INACTIVE:
 665                 /* RFC 2960 Section 7.2.1, sctpimpguide
 666                  * When the endpoint does not transmit data on a given
 667                  * transport address, the cwnd of the transport address
 668                  * should be adjusted to max(cwnd/2, 4*MTU) per RTO.
 669                  * NOTE: Although the draft recommends that this check needs
 670                  * to be done every RTO interval, we do it every hearbeat
 671                  * interval.
 672                  */
 673                 transport->cwnd = max(transport->cwnd/2,
 674                                          4*asoc->pathmtu);
 675                 /* RFC 4960 Errata 3.27.2: also adjust sshthresh */
 676                 transport->ssthresh = transport->cwnd;
 677                 break;
 678         }
 679
 680         transport->partial_bytes_acked = 0;
 681
 682         pr_debug("%s: transport:%p, reason:%d, cwnd:%d, ssthresh:%d\n",
 683                  __func__, transport, reason, transport->cwnd,
 684                  transport->ssthresh);
 685 }
 686
 687 /* Apply Max.Burst limit to the congestion window:
 688  * sctpimpguide-05 2.14.2
 689  * D) When the time comes for the sender to
 690  * transmit new DATA chunks, the protocol parameter Max.Burst MUST
 691  * first be applied to limit how many new DATA chunks may be sent.
 692  * The limit is applied by adjusting cwnd as follows:
 693  *      if ((flightsize+ Max.Burst * MTU) < cwnd)
 694  *              cwnd = flightsize + Max.Burst * MTU
 695  */
 696
 697 void sctp_transport_burst_limited(struct sctp_transport *t)
 698 {
 699         struct sctp_association *asoc = t->asoc;
 700         u32 old_cwnd = t->cwnd;
 701         u32 max_burst_bytes;
 702
 703         if (t->burst_limited || asoc->max_burst == 0)
 704                 return;
 705
 706         max_burst_bytes = t->flight_size + (asoc->max_burst * asoc->pathmtu);
 707         if (max_burst_bytes < old_cwnd) {
 708                 t->cwnd = max_burst_bytes;
 709                 t->burst_limited = old_cwnd;
 710         }
 711 }
 712
 713 /* Restore the old cwnd congestion window, after the burst had it's
 714  * desired effect.
 715  */
 716 void sctp_transport_burst_reset(struct sctp_transport *t)
 717 {
 718         if (t->burst_limited) {
 719                 t->cwnd = t->burst_limited;
 720                 t->burst_limited = 0;
 721         }
 722 }
 723
 724 /* What is the next timeout value for this transport? */
 725 unsigned long sctp_transport_timeout(struct sctp_transport *trans)
 726 {
 727         /* RTO + timer slack +/- 50% of RTO */
 728         unsigned long timeout = trans->rto >> 1;
 729
 730         if (trans->state != SCTP_UNCONFIRMED &&
 731             trans->state != SCTP_PF)
 732                 timeout += trans->hbinterval;
 733
 734         return max_t(unsigned long, timeout, HZ / 5);
 735 }
 736
 737 /* Reset transport variables to their initial values */
 738 void sctp_transport_reset(struct sctp_transport *t)
 739 {
 740         struct sctp_association *asoc = t->asoc;
 741
 742         /* RFC 2960 (bis), Section 5.2.4
 743          * All the congestion control parameters (e.g., cwnd, ssthresh)
 744          * related to this peer MUST be reset to their initial values
 745          * (see Section 6.2.1)
 746          */
 747         t->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
 748         t->burst_limited = 0;
 749         t->ssthresh = asoc->peer.i.a_rwnd;
 750         t->rto = asoc->rto_initial;
 751         sctp_max_rto(asoc, t);
 752         t->rtt = 0;
 753         t->srtt = 0;
 754         t->rttvar = 0;
 755
 756         /* Reset these additional variables so that we have a clean slate. */
 757         t->partial_bytes_acked = 0;
 758         t->flight_size = 0;
 759         t->error_count = 0;
 760         t->rto_pending = 0;
 761         t->hb_sent = 0;
 762
 763         /* Initialize the state information for SFR-CACC */
 764         t->cacc.changeover_active = 0;
 765         t->cacc.cycling_changeover = 0;
 766         t->cacc.next_tsn_at_change = 0;
 767         t->cacc.cacc_saw_newack = 0;
 768 }
 769
 770 /* Schedule retransmission on the given transport */
 771 void sctp_transport_immediate_rtx(struct sctp_transport *t)
 772 {
 773         /* Stop pending T3_rtx_timer */
 774         if (del_timer(&t->T3_rtx_timer))
 775                 sctp_transport_put(t);
 776
 777         sctp_retransmit(&t->asoc->outqueue, t, SCTP_RTXR_T3_RTX);
 778         if (!timer_pending(&t->T3_rtx_timer)) {
 779                 if (!mod_timer(&t->T3_rtx_timer, jiffies + t->rto))
 780                         sctp_transport_hold(t);
 781         }
 782 }
 783
 784 /* Drop dst */
 785 void sctp_transport_dst_release(struct sctp_transport *t)
 786 {
 787         dst_release(t->dst);
 788         t->dst = NULL;
 789         t->dst_pending_confirm = 0;
 790 }
 791
 792 /* Schedule neighbour confirm */
 793 void sctp_transport_dst_confirm(struct sctp_transport *t)
 794 {
 795         t->dst_pending_confirm = 1;
 796 }