linux/net/sctp/associola.c
<<
>>
Prefs
   1/* SCTP kernel implementation
   2 * (C) Copyright IBM Corp. 2001, 2004
   3 * Copyright (c) 1999-2000 Cisco, Inc.
   4 * Copyright (c) 1999-2001 Motorola, Inc.
   5 * Copyright (c) 2001 Intel Corp.
   6 * Copyright (c) 2001 La Monte H.P. Yarroll
   7 *
   8 * This file is part of the SCTP kernel implementation
   9 *
  10 * This module provides the abstraction for an SCTP association.
  11 *
  12 * This SCTP implementation is free software;
  13 * you can redistribute it and/or modify it under the terms of
  14 * the GNU General Public License as published by
  15 * the Free Software Foundation; either version 2, or (at your option)
  16 * any later version.
  17 *
  18 * This SCTP implementation is distributed in the hope that it
  19 * will be useful, but WITHOUT ANY WARRANTY; without even the implied
  20 *                 ************************
  21 * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  22 * See the GNU General Public License for more details.
  23 *
  24 * You should have received a copy of the GNU General Public License
  25 * along with GNU CC; see the file COPYING.  If not, see
  26 * <http://www.gnu.org/licenses/>.
  27 *
  28 * Please send any bug reports or fixes you make to the
  29 * email address(es):
  30 *    lksctp developers <linux-sctp@vger.kernel.org>
  31 *
  32 * Written or modified by:
  33 *    La Monte H.P. Yarroll <piggy@acm.org>
  34 *    Karl Knutson          <karl@athena.chicago.il.us>
  35 *    Jon Grimm             <jgrimm@us.ibm.com>
  36 *    Xingang Guo           <xingang.guo@intel.com>
  37 *    Hui Huang             <hui.huang@nokia.com>
  38 *    Sridhar Samudrala     <sri@us.ibm.com>
  39 *    Daisy Chang           <daisyc@us.ibm.com>
  40 *    Ryan Layer            <rmlayer@us.ibm.com>
  41 *    Kevin Gao             <kevin.gao@intel.com>
  42 */
  43
  44#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  45
  46#include <linux/types.h>
  47#include <linux/fcntl.h>
  48#include <linux/poll.h>
  49#include <linux/init.h>
  50
  51#include <linux/slab.h>
  52#include <linux/in.h>
  53#include <net/ipv6.h>
  54#include <net/sctp/sctp.h>
  55#include <net/sctp/sm.h>
  56
  57/* Forward declarations for internal functions. */
  58static void sctp_select_active_and_retran_path(struct sctp_association *asoc);
  59static void sctp_assoc_bh_rcv(struct work_struct *work);
  60static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc);
  61static void sctp_assoc_free_asconf_queue(struct sctp_association *asoc);
  62
  63/* 1st Level Abstractions. */
  64
  65/* Initialize a new association from provided memory. */
  66static struct sctp_association *sctp_association_init(struct sctp_association *asoc,
  67                                          const struct sctp_endpoint *ep,
  68                                          const struct sock *sk,
  69                                          sctp_scope_t scope,
  70                                          gfp_t gfp)
  71{
  72        struct net *net = sock_net(sk);
  73        struct sctp_sock *sp;
  74        int i;
  75        sctp_paramhdr_t *p;
  76        int err;
  77
  78        /* Retrieve the SCTP per socket area.  */
  79        sp = sctp_sk((struct sock *)sk);
  80
  81        /* Discarding const is appropriate here.  */
  82        asoc->ep = (struct sctp_endpoint *)ep;
  83        asoc->base.sk = (struct sock *)sk;
  84
  85        sctp_endpoint_hold(asoc->ep);
  86        sock_hold(asoc->base.sk);
  87
  88        /* Initialize the common base substructure.  */
  89        asoc->base.type = SCTP_EP_TYPE_ASSOCIATION;
  90
  91        /* Initialize the object handling fields.  */
  92        atomic_set(&asoc->base.refcnt, 1);
  93
  94        /* Initialize the bind addr area.  */
  95        sctp_bind_addr_init(&asoc->base.bind_addr, ep->base.bind_addr.port);
  96
  97        asoc->state = SCTP_STATE_CLOSED;
  98        asoc->cookie_life = ms_to_ktime(sp->assocparams.sasoc_cookie_life);
  99        asoc->user_frag = sp->user_frag;
 100
 101        /* Set the association max_retrans and RTO values from the
 102         * socket values.
 103         */
 104        asoc->max_retrans = sp->assocparams.sasoc_asocmaxrxt;
 105        asoc->pf_retrans  = net->sctp.pf_retrans;
 106
 107        asoc->rto_initial = msecs_to_jiffies(sp->rtoinfo.srto_initial);
 108        asoc->rto_max = msecs_to_jiffies(sp->rtoinfo.srto_max);
 109        asoc->rto_min = msecs_to_jiffies(sp->rtoinfo.srto_min);
 110
 111        /* Initialize the association's heartbeat interval based on the
 112         * sock configured value.
 113         */
 114        asoc->hbinterval = msecs_to_jiffies(sp->hbinterval);
 115
 116        /* Initialize path max retrans value. */
 117        asoc->pathmaxrxt = sp->pathmaxrxt;
 118
 119        /* Initialize default path MTU. */
 120        asoc->pathmtu = sp->pathmtu;
 121
 122        /* Set association default SACK delay */
 123        asoc->sackdelay = msecs_to_jiffies(sp->sackdelay);
 124        asoc->sackfreq = sp->sackfreq;
 125
 126        /* Set the association default flags controlling
 127         * Heartbeat, SACK delay, and Path MTU Discovery.
 128         */
 129        asoc->param_flags = sp->param_flags;
 130
 131        /* Initialize the maximum number of new data packets that can be sent
 132         * in a burst.
 133         */
 134        asoc->max_burst = sp->max_burst;
 135
 136        /* initialize association timers */
 137        asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_COOKIE] = asoc->rto_initial;
 138        asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_INIT] = asoc->rto_initial;
 139        asoc->timeouts[SCTP_EVENT_TIMEOUT_T2_SHUTDOWN] = asoc->rto_initial;
 140
 141        /* sctpimpguide Section 2.12.2
 142         * If the 'T5-shutdown-guard' timer is used, it SHOULD be set to the
 143         * recommended value of 5 times 'RTO.Max'.
 144         */
 145        asoc->timeouts[SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]
 146                = 5 * asoc->rto_max;
 147
 148        asoc->timeouts[SCTP_EVENT_TIMEOUT_SACK] = asoc->sackdelay;
 149        asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE] = sp->autoclose * HZ;
 150
 151        /* Initializes the timers */
 152        for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i)
 153                setup_timer(&asoc->timers[i], sctp_timer_events[i],
 154                                (unsigned long)asoc);
 155
 156        /* Pull default initialization values from the sock options.
 157         * Note: This assumes that the values have already been
 158         * validated in the sock.
 159         */
 160        asoc->c.sinit_max_instreams = sp->initmsg.sinit_max_instreams;
 161        asoc->c.sinit_num_ostreams  = sp->initmsg.sinit_num_ostreams;
 162        asoc->max_init_attempts = sp->initmsg.sinit_max_attempts;
 163
 164        asoc->max_init_timeo =
 165                 msecs_to_jiffies(sp->initmsg.sinit_max_init_timeo);
 166
 167        /* Set the local window size for receive.
 168         * This is also the rcvbuf space per association.
 169         * RFC 6 - A SCTP receiver MUST be able to receive a minimum of
 170         * 1500 bytes in one SCTP packet.
 171         */
 172        if ((sk->sk_rcvbuf/2) < SCTP_DEFAULT_MINWINDOW)
 173                asoc->rwnd = SCTP_DEFAULT_MINWINDOW;
 174        else
 175                asoc->rwnd = sk->sk_rcvbuf/2;
 176
 177        asoc->a_rwnd = asoc->rwnd;
 178
 179        /* Use my own max window until I learn something better.  */
 180        asoc->peer.rwnd = SCTP_DEFAULT_MAXWINDOW;
 181
 182        /* Initialize the receive memory counter */
 183        atomic_set(&asoc->rmem_alloc, 0);
 184
 185        init_waitqueue_head(&asoc->wait);
 186
 187        asoc->c.my_vtag = sctp_generate_tag(ep);
 188        asoc->c.my_port = ep->base.bind_addr.port;
 189
 190        asoc->c.initial_tsn = sctp_generate_tsn(ep);
 191
 192        asoc->next_tsn = asoc->c.initial_tsn;
 193
 194        asoc->ctsn_ack_point = asoc->next_tsn - 1;
 195        asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
 196        asoc->highest_sacked = asoc->ctsn_ack_point;
 197        asoc->last_cwr_tsn = asoc->ctsn_ack_point;
 198
 199        /* ADDIP Section 4.1 Asconf Chunk Procedures
 200         *
 201         * When an endpoint has an ASCONF signaled change to be sent to the
 202         * remote endpoint it should do the following:
 203         * ...
 204         * A2) a serial number should be assigned to the chunk. The serial
 205         * number SHOULD be a monotonically increasing number. The serial
 206         * numbers SHOULD be initialized at the start of the
 207         * association to the same value as the initial TSN.
 208         */
 209        asoc->addip_serial = asoc->c.initial_tsn;
 210
 211        INIT_LIST_HEAD(&asoc->addip_chunk_list);
 212        INIT_LIST_HEAD(&asoc->asconf_ack_list);
 213
 214        /* Make an empty list of remote transport addresses.  */
 215        INIT_LIST_HEAD(&asoc->peer.transport_addr_list);
 216
 217        /* RFC 2960 5.1 Normal Establishment of an Association
 218         *
 219         * After the reception of the first data chunk in an
 220         * association the endpoint must immediately respond with a
 221         * sack to acknowledge the data chunk.  Subsequent
 222         * acknowledgements should be done as described in Section
 223         * 6.2.
 224         *
 225         * [We implement this by telling a new association that it
 226         * already received one packet.]
 227         */
 228        asoc->peer.sack_needed = 1;
 229        asoc->peer.sack_generation = 1;
 230
 231        /* Assume that the peer will tell us if he recognizes ASCONF
 232         * as part of INIT exchange.
 233         * The sctp_addip_noauth option is there for backward compatibility
 234         * and will revert old behavior.
 235         */
 236        if (net->sctp.addip_noauth)
 237                asoc->peer.asconf_capable = 1;
 238
 239        /* Create an input queue.  */
 240        sctp_inq_init(&asoc->base.inqueue);
 241        sctp_inq_set_th_handler(&asoc->base.inqueue, sctp_assoc_bh_rcv);
 242
 243        /* Create an output queue.  */
 244        sctp_outq_init(asoc, &asoc->outqueue);
 245
 246        if (!sctp_ulpq_init(&asoc->ulpq, asoc))
 247                goto fail_init;
 248
 249        /* Assume that peer would support both address types unless we are
 250         * told otherwise.
 251         */
 252        asoc->peer.ipv4_address = 1;
 253        if (asoc->base.sk->sk_family == PF_INET6)
 254                asoc->peer.ipv6_address = 1;
 255        INIT_LIST_HEAD(&asoc->asocs);
 256
 257        asoc->default_stream = sp->default_stream;
 258        asoc->default_ppid = sp->default_ppid;
 259        asoc->default_flags = sp->default_flags;
 260        asoc->default_context = sp->default_context;
 261        asoc->default_timetolive = sp->default_timetolive;
 262        asoc->default_rcv_context = sp->default_rcv_context;
 263
 264        /* AUTH related initializations */
 265        INIT_LIST_HEAD(&asoc->endpoint_shared_keys);
 266        err = sctp_auth_asoc_copy_shkeys(ep, asoc, gfp);
 267        if (err)
 268                goto fail_init;
 269
 270        asoc->active_key_id = ep->active_key_id;
 271
 272        /* Save the hmacs and chunks list into this association */
 273        if (ep->auth_hmacs_list)
 274                memcpy(asoc->c.auth_hmacs, ep->auth_hmacs_list,
 275                        ntohs(ep->auth_hmacs_list->param_hdr.length));
 276        if (ep->auth_chunk_list)
 277                memcpy(asoc->c.auth_chunks, ep->auth_chunk_list,
 278                        ntohs(ep->auth_chunk_list->param_hdr.length));
 279
 280        /* Get the AUTH random number for this association */
 281        p = (sctp_paramhdr_t *)asoc->c.auth_random;
 282        p->type = SCTP_PARAM_RANDOM;
 283        p->length = htons(sizeof(sctp_paramhdr_t) + SCTP_AUTH_RANDOM_LENGTH);
 284        get_random_bytes(p+1, SCTP_AUTH_RANDOM_LENGTH);
 285
 286        return asoc;
 287
 288fail_init:
 289        sock_put(asoc->base.sk);
 290        sctp_endpoint_put(asoc->ep);
 291        return NULL;
 292}
 293
 294/* Allocate and initialize a new association */
 295struct sctp_association *sctp_association_new(const struct sctp_endpoint *ep,
 296                                         const struct sock *sk,
 297                                         sctp_scope_t scope,
 298                                         gfp_t gfp)
 299{
 300        struct sctp_association *asoc;
 301
 302        asoc = kzalloc(sizeof(*asoc), gfp);
 303        if (!asoc)
 304                goto fail;
 305
 306        if (!sctp_association_init(asoc, ep, sk, scope, gfp))
 307                goto fail_init;
 308
 309        SCTP_DBG_OBJCNT_INC(assoc);
 310
 311        pr_debug("Created asoc %p\n", asoc);
 312
 313        return asoc;
 314
 315fail_init:
 316        kfree(asoc);
 317fail:
 318        return NULL;
 319}
 320
 321/* Free this association if possible.  There may still be users, so
 322 * the actual deallocation may be delayed.
 323 */
 324void sctp_association_free(struct sctp_association *asoc)
 325{
 326        struct sock *sk = asoc->base.sk;
 327        struct sctp_transport *transport;
 328        struct list_head *pos, *temp;
 329        int i;
 330
 331        /* Only real associations count against the endpoint, so
 332         * don't bother for if this is a temporary association.
 333         */
 334        if (!list_empty(&asoc->asocs)) {
 335                list_del(&asoc->asocs);
 336
 337                /* Decrement the backlog value for a TCP-style listening
 338                 * socket.
 339                 */
 340                if (sctp_style(sk, TCP) && sctp_sstate(sk, LISTENING))
 341                        sk->sk_ack_backlog--;
 342        }
 343
 344        /* Mark as dead, so other users can know this structure is
 345         * going away.
 346         */
 347        asoc->base.dead = true;
 348
 349        /* Dispose of any data lying around in the outqueue. */
 350        sctp_outq_free(&asoc->outqueue);
 351
 352        /* Dispose of any pending messages for the upper layer. */
 353        sctp_ulpq_free(&asoc->ulpq);
 354
 355        /* Dispose of any pending chunks on the inqueue. */
 356        sctp_inq_free(&asoc->base.inqueue);
 357
 358        sctp_tsnmap_free(&asoc->peer.tsn_map);
 359
 360        /* Free ssnmap storage. */
 361        sctp_ssnmap_free(asoc->ssnmap);
 362
 363        /* Clean up the bound address list. */
 364        sctp_bind_addr_free(&asoc->base.bind_addr);
 365
 366        /* Do we need to go through all of our timers and
 367         * delete them?   To be safe we will try to delete all, but we
 368         * should be able to go through and make a guess based
 369         * on our state.
 370         */
 371        for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i) {
 372                if (del_timer(&asoc->timers[i]))
 373                        sctp_association_put(asoc);
 374        }
 375
 376        /* Free peer's cached cookie. */
 377        kfree(asoc->peer.cookie);
 378        kfree(asoc->peer.peer_random);
 379        kfree(asoc->peer.peer_chunks);
 380        kfree(asoc->peer.peer_hmacs);
 381
 382        /* Release the transport structures. */
 383        list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
 384                transport = list_entry(pos, struct sctp_transport, transports);
 385                list_del_rcu(pos);
 386                sctp_transport_free(transport);
 387        }
 388
 389        asoc->peer.transport_count = 0;
 390
 391        sctp_asconf_queue_teardown(asoc);
 392
 393        /* Free pending address space being deleted */
 394        kfree(asoc->asconf_addr_del_pending);
 395
 396        /* AUTH - Free the endpoint shared keys */
 397        sctp_auth_destroy_keys(&asoc->endpoint_shared_keys);
 398
 399        /* AUTH - Free the association shared key */
 400        sctp_auth_key_put(asoc->asoc_shared_key);
 401
 402        sctp_association_put(asoc);
 403}
 404
 405/* Cleanup and free up an association. */
 406static void sctp_association_destroy(struct sctp_association *asoc)
 407{
 408        if (unlikely(!asoc->base.dead)) {
 409                WARN(1, "Attempt to destroy undead association %p!\n", asoc);
 410                return;
 411        }
 412
 413        sctp_endpoint_put(asoc->ep);
 414        sock_put(asoc->base.sk);
 415
 416        if (asoc->assoc_id != 0) {
 417                spin_lock_bh(&sctp_assocs_id_lock);
 418                idr_remove(&sctp_assocs_id, asoc->assoc_id);
 419                spin_unlock_bh(&sctp_assocs_id_lock);
 420        }
 421
 422        WARN_ON(atomic_read(&asoc->rmem_alloc));
 423
 424        kfree(asoc);
 425        SCTP_DBG_OBJCNT_DEC(assoc);
 426}
 427
 428/* Change the primary destination address for the peer. */
 429void sctp_assoc_set_primary(struct sctp_association *asoc,
 430                            struct sctp_transport *transport)
 431{
 432        int changeover = 0;
 433
 434        /* it's a changeover only if we already have a primary path
 435         * that we are changing
 436         */
 437        if (asoc->peer.primary_path != NULL &&
 438            asoc->peer.primary_path != transport)
 439                changeover = 1 ;
 440
 441        asoc->peer.primary_path = transport;
 442
 443        /* Set a default msg_name for events. */
 444        memcpy(&asoc->peer.primary_addr, &transport->ipaddr,
 445               sizeof(union sctp_addr));
 446
 447        /* If the primary path is changing, assume that the
 448         * user wants to use this new path.
 449         */
 450        if ((transport->state == SCTP_ACTIVE) ||
 451            (transport->state == SCTP_UNKNOWN))
 452                asoc->peer.active_path = transport;
 453
 454        /*
 455         * SFR-CACC algorithm:
 456         * Upon the receipt of a request to change the primary
 457         * destination address, on the data structure for the new
 458         * primary destination, the sender MUST do the following:
 459         *
 460         * 1) If CHANGEOVER_ACTIVE is set, then there was a switch
 461         * to this destination address earlier. The sender MUST set
 462         * CYCLING_CHANGEOVER to indicate that this switch is a
 463         * double switch to the same destination address.
 464         *
 465         * Really, only bother is we have data queued or outstanding on
 466         * the association.
 467         */
 468        if (!asoc->outqueue.outstanding_bytes && !asoc->outqueue.out_qlen)
 469                return;
 470
 471        if (transport->cacc.changeover_active)
 472                transport->cacc.cycling_changeover = changeover;
 473
 474        /* 2) The sender MUST set CHANGEOVER_ACTIVE to indicate that
 475         * a changeover has occurred.
 476         */
 477        transport->cacc.changeover_active = changeover;
 478
 479        /* 3) The sender MUST store the next TSN to be sent in
 480         * next_tsn_at_change.
 481         */
 482        transport->cacc.next_tsn_at_change = asoc->next_tsn;
 483}
 484
 485/* Remove a transport from an association.  */
 486void sctp_assoc_rm_peer(struct sctp_association *asoc,
 487                        struct sctp_transport *peer)
 488{
 489        struct list_head        *pos;
 490        struct sctp_transport   *transport;
 491
 492        pr_debug("%s: association:%p addr:%pISpc\n",
 493                 __func__, asoc, &peer->ipaddr.sa);
 494
 495        /* If we are to remove the current retran_path, update it
 496         * to the next peer before removing this peer from the list.
 497         */
 498        if (asoc->peer.retran_path == peer)
 499                sctp_assoc_update_retran_path(asoc);
 500
 501        /* Remove this peer from the list. */
 502        list_del_rcu(&peer->transports);
 503
 504        /* Get the first transport of asoc. */
 505        pos = asoc->peer.transport_addr_list.next;
 506        transport = list_entry(pos, struct sctp_transport, transports);
 507
 508        /* Update any entries that match the peer to be deleted. */
 509        if (asoc->peer.primary_path == peer)
 510                sctp_assoc_set_primary(asoc, transport);
 511        if (asoc->peer.active_path == peer)
 512                asoc->peer.active_path = transport;
 513        if (asoc->peer.retran_path == peer)
 514                asoc->peer.retran_path = transport;
 515        if (asoc->peer.last_data_from == peer)
 516                asoc->peer.last_data_from = transport;
 517
 518        /* If we remove the transport an INIT was last sent to, set it to
 519         * NULL. Combined with the update of the retran path above, this
 520         * will cause the next INIT to be sent to the next available
 521         * transport, maintaining the cycle.
 522         */
 523        if (asoc->init_last_sent_to == peer)
 524                asoc->init_last_sent_to = NULL;
 525
 526        /* If we remove the transport an SHUTDOWN was last sent to, set it
 527         * to NULL. Combined with the update of the retran path above, this
 528         * will cause the next SHUTDOWN to be sent to the next available
 529         * transport, maintaining the cycle.
 530         */
 531        if (asoc->shutdown_last_sent_to == peer)
 532                asoc->shutdown_last_sent_to = NULL;
 533
 534        /* If we remove the transport an ASCONF was last sent to, set it to
 535         * NULL.
 536         */
 537        if (asoc->addip_last_asconf &&
 538            asoc->addip_last_asconf->transport == peer)
 539                asoc->addip_last_asconf->transport = NULL;
 540
 541        /* If we have something on the transmitted list, we have to
 542         * save it off.  The best place is the active path.
 543         */
 544        if (!list_empty(&peer->transmitted)) {
 545                struct sctp_transport *active = asoc->peer.active_path;
 546                struct sctp_chunk *ch;
 547
 548                /* Reset the transport of each chunk on this list */
 549                list_for_each_entry(ch, &peer->transmitted,
 550                                        transmitted_list) {
 551                        ch->transport = NULL;
 552                        ch->rtt_in_progress = 0;
 553                }
 554
 555                list_splice_tail_init(&peer->transmitted,
 556                                        &active->transmitted);
 557
 558                /* Start a T3 timer here in case it wasn't running so
 559                 * that these migrated packets have a chance to get
 560                 * retransmitted.
 561                 */
 562                if (!timer_pending(&active->T3_rtx_timer))
 563                        if (!mod_timer(&active->T3_rtx_timer,
 564                                        jiffies + active->rto))
 565                                sctp_transport_hold(active);
 566        }
 567
 568        asoc->peer.transport_count--;
 569
 570        sctp_transport_free(peer);
 571}
 572
 573/* Add a transport address to an association.  */
 574struct sctp_transport *sctp_assoc_add_peer(struct sctp_association *asoc,
 575                                           const union sctp_addr *addr,
 576                                           const gfp_t gfp,
 577                                           const int peer_state)
 578{
 579        struct net *net = sock_net(asoc->base.sk);
 580        struct sctp_transport *peer;
 581        struct sctp_sock *sp;
 582        unsigned short port;
 583
 584        sp = sctp_sk(asoc->base.sk);
 585
 586        /* AF_INET and AF_INET6 share common port field. */
 587        port = ntohs(addr->v4.sin_port);
 588
 589        pr_debug("%s: association:%p addr:%pISpc state:%d\n", __func__,
 590                 asoc, &addr->sa, peer_state);
 591
 592        /* Set the port if it has not been set yet.  */
 593        if (0 == asoc->peer.port)
 594                asoc->peer.port = port;
 595
 596        /* Check to see if this is a duplicate. */
 597        peer = sctp_assoc_lookup_paddr(asoc, addr);
 598        if (peer) {
 599                /* An UNKNOWN state is only set on transports added by
 600                 * user in sctp_connectx() call.  Such transports should be
 601                 * considered CONFIRMED per RFC 4960, Section 5.4.
 602                 */
 603                if (peer->state == SCTP_UNKNOWN) {
 604                        peer->state = SCTP_ACTIVE;
 605                }
 606                return peer;
 607        }
 608
 609        peer = sctp_transport_new(net, addr, gfp);
 610        if (!peer)
 611                return NULL;
 612
 613        sctp_transport_set_owner(peer, asoc);
 614
 615        /* Initialize the peer's heartbeat interval based on the
 616         * association configured value.
 617         */
 618        peer->hbinterval = asoc->hbinterval;
 619
 620        /* Set the path max_retrans.  */
 621        peer->pathmaxrxt = asoc->pathmaxrxt;
 622
 623        /* And the partial failure retrans threshold */
 624        peer->pf_retrans = asoc->pf_retrans;
 625
 626        /* Initialize the peer's SACK delay timeout based on the
 627         * association configured value.
 628         */
 629        peer->sackdelay = asoc->sackdelay;
 630        peer->sackfreq = asoc->sackfreq;
 631
 632        /* Enable/disable heartbeat, SACK delay, and path MTU discovery
 633         * based on association setting.
 634         */
 635        peer->param_flags = asoc->param_flags;
 636
 637        sctp_transport_route(peer, NULL, sp);
 638
 639        /* Initialize the pmtu of the transport. */
 640        if (peer->param_flags & SPP_PMTUD_DISABLE) {
 641                if (asoc->pathmtu)
 642                        peer->pathmtu = asoc->pathmtu;
 643                else
 644                        peer->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
 645        }
 646
 647        /* If this is the first transport addr on this association,
 648         * initialize the association PMTU to the peer's PMTU.
 649         * If not and the current association PMTU is higher than the new
 650         * peer's PMTU, reset the association PMTU to the new peer's PMTU.
 651         */
 652        if (asoc->pathmtu)
 653                asoc->pathmtu = min_t(int, peer->pathmtu, asoc->pathmtu);
 654        else
 655                asoc->pathmtu = peer->pathmtu;
 656
 657        pr_debug("%s: association:%p PMTU set to %d\n", __func__, asoc,
 658                 asoc->pathmtu);
 659
 660        peer->pmtu_pending = 0;
 661
 662        asoc->frag_point = sctp_frag_point(asoc, asoc->pathmtu);
 663
 664        /* The asoc->peer.port might not be meaningful yet, but
 665         * initialize the packet structure anyway.
 666         */
 667        sctp_packet_init(&peer->packet, peer, asoc->base.bind_addr.port,
 668                         asoc->peer.port);
 669
 670        /* 7.2.1 Slow-Start
 671         *
 672         * o The initial cwnd before DATA transmission or after a sufficiently
 673         *   long idle period MUST be set to
 674         *      min(4*MTU, max(2*MTU, 4380 bytes))
 675         *
 676         * o The initial value of ssthresh MAY be arbitrarily high
 677         *   (for example, implementations MAY use the size of the
 678         *   receiver advertised window).
 679         */
 680        peer->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
 681
 682        /* At this point, we may not have the receiver's advertised window,
 683         * so initialize ssthresh to the default value and it will be set
 684         * later when we process the INIT.
 685         */
 686        peer->ssthresh = SCTP_DEFAULT_MAXWINDOW;
 687
 688        peer->partial_bytes_acked = 0;
 689        peer->flight_size = 0;
 690        peer->burst_limited = 0;
 691
 692        /* Set the transport's RTO.initial value */
 693        peer->rto = asoc->rto_initial;
 694        sctp_max_rto(asoc, peer);
 695
 696        /* Set the peer's active state. */
 697        peer->state = peer_state;
 698
 699        /* Attach the remote transport to our asoc.  */
 700        list_add_tail_rcu(&peer->transports, &asoc->peer.transport_addr_list);
 701        asoc->peer.transport_count++;
 702
 703        /* If we do not yet have a primary path, set one.  */
 704        if (!asoc->peer.primary_path) {
 705                sctp_assoc_set_primary(asoc, peer);
 706                asoc->peer.retran_path = peer;
 707        }
 708
 709        if (asoc->peer.active_path == asoc->peer.retran_path &&
 710            peer->state != SCTP_UNCONFIRMED) {
 711                asoc->peer.retran_path = peer;
 712        }
 713
 714        return peer;
 715}
 716
 717/* Delete a transport address from an association.  */
 718void sctp_assoc_del_peer(struct sctp_association *asoc,
 719                         const union sctp_addr *addr)
 720{
 721        struct list_head        *pos;
 722        struct list_head        *temp;
 723        struct sctp_transport   *transport;
 724
 725        list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
 726                transport = list_entry(pos, struct sctp_transport, transports);
 727                if (sctp_cmp_addr_exact(addr, &transport->ipaddr)) {
 728                        /* Do book keeping for removing the peer and free it. */
 729                        sctp_assoc_rm_peer(asoc, transport);
 730                        break;
 731                }
 732        }
 733}
 734
 735/* Lookup a transport by address. */
 736struct sctp_transport *sctp_assoc_lookup_paddr(
 737                                        const struct sctp_association *asoc,
 738                                        const union sctp_addr *address)
 739{
 740        struct sctp_transport *t;
 741
 742        /* Cycle through all transports searching for a peer address. */
 743
 744        list_for_each_entry(t, &asoc->peer.transport_addr_list,
 745                        transports) {
 746                if (sctp_cmp_addr_exact(address, &t->ipaddr))
 747                        return t;
 748        }
 749
 750        return NULL;
 751}
 752
 753/* Remove all transports except a give one */
 754void sctp_assoc_del_nonprimary_peers(struct sctp_association *asoc,
 755                                     struct sctp_transport *primary)
 756{
 757        struct sctp_transport   *temp;
 758        struct sctp_transport   *t;
 759
 760        list_for_each_entry_safe(t, temp, &asoc->peer.transport_addr_list,
 761                                 transports) {
 762                /* if the current transport is not the primary one, delete it */
 763                if (t != primary)
 764                        sctp_assoc_rm_peer(asoc, t);
 765        }
 766}
 767
 768/* Engage in transport control operations.
 769 * Mark the transport up or down and send a notification to the user.
 770 * Select and update the new active and retran paths.
 771 */
 772void sctp_assoc_control_transport(struct sctp_association *asoc,
 773                                  struct sctp_transport *transport,
 774                                  sctp_transport_cmd_t command,
 775                                  sctp_sn_error_t error)
 776{
 777        struct sctp_ulpevent *event;
 778        struct sockaddr_storage addr;
 779        int spc_state = 0;
 780        bool ulp_notify = true;
 781
 782        /* Record the transition on the transport.  */
 783        switch (command) {
 784        case SCTP_TRANSPORT_UP:
 785                /* If we are moving from UNCONFIRMED state due
 786                 * to heartbeat success, report the SCTP_ADDR_CONFIRMED
 787                 * state to the user, otherwise report SCTP_ADDR_AVAILABLE.
 788                 */
 789                if (SCTP_UNCONFIRMED == transport->state &&
 790                    SCTP_HEARTBEAT_SUCCESS == error)
 791                        spc_state = SCTP_ADDR_CONFIRMED;
 792                else
 793                        spc_state = SCTP_ADDR_AVAILABLE;
 794                /* Don't inform ULP about transition from PF to
 795                 * active state and set cwnd to 1 MTU, see SCTP
 796                 * Quick failover draft section 5.1, point 5
 797                 */
 798                if (transport->state == SCTP_PF) {
 799                        ulp_notify = false;
 800                        transport->cwnd = asoc->pathmtu;
 801                }
 802                transport->state = SCTP_ACTIVE;
 803                break;
 804
 805        case SCTP_TRANSPORT_DOWN:
 806                /* If the transport was never confirmed, do not transition it
 807                 * to inactive state.  Also, release the cached route since
 808                 * there may be a better route next time.
 809                 */
 810                if (transport->state != SCTP_UNCONFIRMED)
 811                        transport->state = SCTP_INACTIVE;
 812                else {
 813                        dst_release(transport->dst);
 814                        transport->dst = NULL;
 815                        ulp_notify = false;
 816                }
 817
 818                spc_state = SCTP_ADDR_UNREACHABLE;
 819                break;
 820
 821        case SCTP_TRANSPORT_PF:
 822                transport->state = SCTP_PF;
 823                ulp_notify = false;
 824                break;
 825
 826        default:
 827                return;
 828        }
 829
 830        /* Generate and send a SCTP_PEER_ADDR_CHANGE notification
 831         * to the user.
 832         */
 833        if (ulp_notify) {
 834                memset(&addr, 0, sizeof(struct sockaddr_storage));
 835                memcpy(&addr, &transport->ipaddr,
 836                       transport->af_specific->sockaddr_len);
 837
 838                event = sctp_ulpevent_make_peer_addr_change(asoc, &addr,
 839                                        0, spc_state, error, GFP_ATOMIC);
 840                if (event)
 841                        sctp_ulpq_tail_event(&asoc->ulpq, event);
 842        }
 843
 844        /* Select new active and retran paths. */
 845        sctp_select_active_and_retran_path(asoc);
 846}
 847
 848/* Hold a reference to an association. */
 849void sctp_association_hold(struct sctp_association *asoc)
 850{
 851        atomic_inc(&asoc->base.refcnt);
 852}
 853
 854/* Release a reference to an association and cleanup
 855 * if there are no more references.
 856 */
 857void sctp_association_put(struct sctp_association *asoc)
 858{
 859        if (atomic_dec_and_test(&asoc->base.refcnt))
 860                sctp_association_destroy(asoc);
 861}
 862
 863/* Allocate the next TSN, Transmission Sequence Number, for the given
 864 * association.
 865 */
 866__u32 sctp_association_get_next_tsn(struct sctp_association *asoc)
 867{
 868        /* From Section 1.6 Serial Number Arithmetic:
 869         * Transmission Sequence Numbers wrap around when they reach
 870         * 2**32 - 1.  That is, the next TSN a DATA chunk MUST use
 871         * after transmitting TSN = 2*32 - 1 is TSN = 0.
 872         */
 873        __u32 retval = asoc->next_tsn;
 874        asoc->next_tsn++;
 875        asoc->unack_data++;
 876
 877        return retval;
 878}
 879
 880/* Compare two addresses to see if they match.  Wildcard addresses
 881 * only match themselves.
 882 */
 883int sctp_cmp_addr_exact(const union sctp_addr *ss1,
 884                        const union sctp_addr *ss2)
 885{
 886        struct sctp_af *af;
 887
 888        af = sctp_get_af_specific(ss1->sa.sa_family);
 889        if (unlikely(!af))
 890                return 0;
 891
 892        return af->cmp_addr(ss1, ss2);
 893}
 894
 895/* Return an ecne chunk to get prepended to a packet.
 896 * Note:  We are sly and return a shared, prealloced chunk.  FIXME:
 897 * No we don't, but we could/should.
 898 */
 899struct sctp_chunk *sctp_get_ecne_prepend(struct sctp_association *asoc)
 900{
 901        if (!asoc->need_ecne)
 902                return NULL;
 903
 904        /* Send ECNE if needed.
 905         * Not being able to allocate a chunk here is not deadly.
 906         */
 907        return sctp_make_ecne(asoc, asoc->last_ecne_tsn);
 908}
 909
 910/*
 911 * Find which transport this TSN was sent on.
 912 */
 913struct sctp_transport *sctp_assoc_lookup_tsn(struct sctp_association *asoc,
 914                                             __u32 tsn)
 915{
 916        struct sctp_transport *active;
 917        struct sctp_transport *match;
 918        struct sctp_transport *transport;
 919        struct sctp_chunk *chunk;
 920        __be32 key = htonl(tsn);
 921
 922        match = NULL;
 923
 924        /*
 925         * FIXME: In general, find a more efficient data structure for
 926         * searching.
 927         */
 928
 929        /*
 930         * The general strategy is to search each transport's transmitted
 931         * list.   Return which transport this TSN lives on.
 932         *
 933         * Let's be hopeful and check the active_path first.
 934         * Another optimization would be to know if there is only one
 935         * outbound path and not have to look for the TSN at all.
 936         *
 937         */
 938
 939        active = asoc->peer.active_path;
 940
 941        list_for_each_entry(chunk, &active->transmitted,
 942                        transmitted_list) {
 943
 944                if (key == chunk->subh.data_hdr->tsn) {
 945                        match = active;
 946                        goto out;
 947                }
 948        }
 949
 950        /* If not found, go search all the other transports. */
 951        list_for_each_entry(transport, &asoc->peer.transport_addr_list,
 952                        transports) {
 953
 954                if (transport == active)
 955                        continue;
 956                list_for_each_entry(chunk, &transport->transmitted,
 957                                transmitted_list) {
 958                        if (key == chunk->subh.data_hdr->tsn) {
 959                                match = transport;
 960                                goto out;
 961                        }
 962                }
 963        }
 964out:
 965        return match;
 966}
 967
 968/* Is this the association we are looking for? */
 969struct sctp_transport *sctp_assoc_is_match(struct sctp_association *asoc,
 970                                           struct net *net,
 971                                           const union sctp_addr *laddr,
 972                                           const union sctp_addr *paddr)
 973{
 974        struct sctp_transport *transport;
 975
 976        if ((htons(asoc->base.bind_addr.port) == laddr->v4.sin_port) &&
 977            (htons(asoc->peer.port) == paddr->v4.sin_port) &&
 978            net_eq(sock_net(asoc->base.sk), net)) {
 979                transport = sctp_assoc_lookup_paddr(asoc, paddr);
 980                if (!transport)
 981                        goto out;
 982
 983                if (sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
 984                                         sctp_sk(asoc->base.sk)))
 985                        goto out;
 986        }
 987        transport = NULL;
 988
 989out:
 990        return transport;
 991}
 992
 993/* Do delayed input processing.  This is scheduled by sctp_rcv(). */
 994static void sctp_assoc_bh_rcv(struct work_struct *work)
 995{
 996        struct sctp_association *asoc =
 997                container_of(work, struct sctp_association,
 998                             base.inqueue.immediate);
 999        struct net *net = sock_net(asoc->base.sk);
1000        struct sctp_endpoint *ep;
1001        struct sctp_chunk *chunk;
1002        struct sctp_inq *inqueue;
1003        int state;
1004        sctp_subtype_t subtype;
1005        int error = 0;
1006
1007        /* The association should be held so we should be safe. */
1008        ep = asoc->ep;
1009
1010        inqueue = &asoc->base.inqueue;
1011        sctp_association_hold(asoc);
1012        while (NULL != (chunk = sctp_inq_pop(inqueue))) {
1013                state = asoc->state;
1014                subtype = SCTP_ST_CHUNK(chunk->chunk_hdr->type);
1015
1016                /* SCTP-AUTH, Section 6.3:
1017                 *    The receiver has a list of chunk types which it expects
1018                 *    to be received only after an AUTH-chunk.  This list has
1019                 *    been sent to the peer during the association setup.  It
1020                 *    MUST silently discard these chunks if they are not placed
1021                 *    after an AUTH chunk in the packet.
1022                 */
1023                if (sctp_auth_recv_cid(subtype.chunk, asoc) && !chunk->auth)
1024                        continue;
1025
1026                /* Remember where the last DATA chunk came from so we
1027                 * know where to send the SACK.
1028                 */
1029                if (sctp_chunk_is_data(chunk))
1030                        asoc->peer.last_data_from = chunk->transport;
1031                else {
1032                        SCTP_INC_STATS(net, SCTP_MIB_INCTRLCHUNKS);
1033                        asoc->stats.ictrlchunks++;
1034                        if (chunk->chunk_hdr->type == SCTP_CID_SACK)
1035                                asoc->stats.isacks++;
1036                }
1037
1038                if (chunk->transport)
1039                        chunk->transport->last_time_heard = ktime_get();
1040
1041                /* Run through the state machine. */
1042                error = sctp_do_sm(net, SCTP_EVENT_T_CHUNK, subtype,
1043                                   state, ep, asoc, chunk, GFP_ATOMIC);
1044
1045                /* Check to see if the association is freed in response to
1046                 * the incoming chunk.  If so, get out of the while loop.
1047                 */
1048                if (asoc->base.dead)
1049                        break;
1050
1051                /* If there is an error on chunk, discard this packet. */
1052                if (error && chunk)
1053                        chunk->pdiscard = 1;
1054        }
1055        sctp_association_put(asoc);
1056}
1057
1058/* This routine moves an association from its old sk to a new sk.  */
1059void sctp_assoc_migrate(struct sctp_association *assoc, struct sock *newsk)
1060{
1061        struct sctp_sock *newsp = sctp_sk(newsk);
1062        struct sock *oldsk = assoc->base.sk;
1063
1064        /* Delete the association from the old endpoint's list of
1065         * associations.
1066         */
1067        list_del_init(&assoc->asocs);
1068
1069        /* Decrement the backlog value for a TCP-style socket. */
1070        if (sctp_style(oldsk, TCP))
1071                oldsk->sk_ack_backlog--;
1072
1073        /* Release references to the old endpoint and the sock.  */
1074        sctp_endpoint_put(assoc->ep);
1075        sock_put(assoc->base.sk);
1076
1077        /* Get a reference to the new endpoint.  */
1078        assoc->ep = newsp->ep;
1079        sctp_endpoint_hold(assoc->ep);
1080
1081        /* Get a reference to the new sock.  */
1082        assoc->base.sk = newsk;
1083        sock_hold(assoc->base.sk);
1084
1085        /* Add the association to the new endpoint's list of associations.  */
1086        sctp_endpoint_add_asoc(newsp->ep, assoc);
1087}
1088
1089/* Update an association (possibly from unexpected COOKIE-ECHO processing).  */
1090void sctp_assoc_update(struct sctp_association *asoc,
1091                       struct sctp_association *new)
1092{
1093        struct sctp_transport *trans;
1094        struct list_head *pos, *temp;
1095
1096        /* Copy in new parameters of peer. */
1097        asoc->c = new->c;
1098        asoc->peer.rwnd = new->peer.rwnd;
1099        asoc->peer.sack_needed = new->peer.sack_needed;
1100        asoc->peer.auth_capable = new->peer.auth_capable;
1101        asoc->peer.i = new->peer.i;
1102        sctp_tsnmap_init(&asoc->peer.tsn_map, SCTP_TSN_MAP_INITIAL,
1103                         asoc->peer.i.initial_tsn, GFP_ATOMIC);
1104
1105        /* Remove any peer addresses not present in the new association. */
1106        list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
1107                trans = list_entry(pos, struct sctp_transport, transports);
1108                if (!sctp_assoc_lookup_paddr(new, &trans->ipaddr)) {
1109                        sctp_assoc_rm_peer(asoc, trans);
1110                        continue;
1111                }
1112
1113                if (asoc->state >= SCTP_STATE_ESTABLISHED)
1114                        sctp_transport_reset(trans);
1115        }
1116
1117        /* If the case is A (association restart), use
1118         * initial_tsn as next_tsn. If the case is B, use
1119         * current next_tsn in case data sent to peer
1120         * has been discarded and needs retransmission.
1121         */
1122        if (asoc->state >= SCTP_STATE_ESTABLISHED) {
1123                asoc->next_tsn = new->next_tsn;
1124                asoc->ctsn_ack_point = new->ctsn_ack_point;
1125                asoc->adv_peer_ack_point = new->adv_peer_ack_point;
1126
1127                /* Reinitialize SSN for both local streams
1128                 * and peer's streams.
1129                 */
1130                sctp_ssnmap_clear(asoc->ssnmap);
1131
1132                /* Flush the ULP reassembly and ordered queue.
1133                 * Any data there will now be stale and will
1134                 * cause problems.
1135                 */
1136                sctp_ulpq_flush(&asoc->ulpq);
1137
1138                /* reset the overall association error count so
1139                 * that the restarted association doesn't get torn
1140                 * down on the next retransmission timer.
1141                 */
1142                asoc->overall_error_count = 0;
1143
1144        } else {
1145                /* Add any peer addresses from the new association. */
1146                list_for_each_entry(trans, &new->peer.transport_addr_list,
1147                                transports) {
1148                        if (!sctp_assoc_lookup_paddr(asoc, &trans->ipaddr))
1149                                sctp_assoc_add_peer(asoc, &trans->ipaddr,
1150                                                    GFP_ATOMIC, trans->state);
1151                }
1152
1153                asoc->ctsn_ack_point = asoc->next_tsn - 1;
1154                asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
1155                if (!asoc->ssnmap) {
1156                        /* Move the ssnmap. */
1157                        asoc->ssnmap = new->ssnmap;
1158                        new->ssnmap = NULL;
1159                }
1160
1161                if (!asoc->assoc_id) {
1162                        /* get a new association id since we don't have one
1163                         * yet.
1164                         */
1165                        sctp_assoc_set_id(asoc, GFP_ATOMIC);
1166                }
1167        }
1168
1169        /* SCTP-AUTH: Save the peer parameters from the new associations
1170         * and also move the association shared keys over
1171         */
1172        kfree(asoc->peer.peer_random);
1173        asoc->peer.peer_random = new->peer.peer_random;
1174        new->peer.peer_random = NULL;
1175
1176        kfree(asoc->peer.peer_chunks);
1177        asoc->peer.peer_chunks = new->peer.peer_chunks;
1178        new->peer.peer_chunks = NULL;
1179
1180        kfree(asoc->peer.peer_hmacs);
1181        asoc->peer.peer_hmacs = new->peer.peer_hmacs;
1182        new->peer.peer_hmacs = NULL;
1183
1184        sctp_auth_asoc_init_active_key(asoc, GFP_ATOMIC);
1185}
1186
1187/* Update the retran path for sending a retransmitted packet.
1188 * See also RFC4960, 6.4. Multi-Homed SCTP Endpoints:
1189 *
1190 *   When there is outbound data to send and the primary path
1191 *   becomes inactive (e.g., due to failures), or where the
1192 *   SCTP user explicitly requests to send data to an
1193 *   inactive destination transport address, before reporting
1194 *   an error to its ULP, the SCTP endpoint should try to send
1195 *   the data to an alternate active destination transport
1196 *   address if one exists.
1197 *
1198 *   When retransmitting data that timed out, if the endpoint
1199 *   is multihomed, it should consider each source-destination
1200 *   address pair in its retransmission selection policy.
1201 *   When retransmitting timed-out data, the endpoint should
1202 *   attempt to pick the most divergent source-destination
1203 *   pair from the original source-destination pair to which
1204 *   the packet was transmitted.
1205 *
1206 *   Note: Rules for picking the most divergent source-destination
1207 *   pair are an implementation decision and are not specified
1208 *   within this document.
1209 *
1210 * Our basic strategy is to round-robin transports in priorities
1211 * according to sctp_state_prio_map[] e.g., if no such
1212 * transport with state SCTP_ACTIVE exists, round-robin through
1213 * SCTP_UNKNOWN, etc. You get the picture.
1214 */
1215static const u8 sctp_trans_state_to_prio_map[] = {
1216        [SCTP_ACTIVE]   = 3,    /* best case */
1217        [SCTP_UNKNOWN]  = 2,
1218        [SCTP_PF]       = 1,
1219        [SCTP_INACTIVE] = 0,    /* worst case */
1220};
1221
1222static u8 sctp_trans_score(const struct sctp_transport *trans)
1223{
1224        return sctp_trans_state_to_prio_map[trans->state];
1225}
1226
1227static struct sctp_transport *sctp_trans_elect_tie(struct sctp_transport *trans1,
1228                                                   struct sctp_transport *trans2)
1229{
1230        if (trans1->error_count > trans2->error_count) {
1231                return trans2;
1232        } else if (trans1->error_count == trans2->error_count &&
1233                   ktime_after(trans2->last_time_heard,
1234                               trans1->last_time_heard)) {
1235                return trans2;
1236        } else {
1237                return trans1;
1238        }
1239}
1240
1241static struct sctp_transport *sctp_trans_elect_best(struct sctp_transport *curr,
1242                                                    struct sctp_transport *best)
1243{
1244        u8 score_curr, score_best;
1245
1246        if (best == NULL || curr == best)
1247                return curr;
1248
1249        score_curr = sctp_trans_score(curr);
1250        score_best = sctp_trans_score(best);
1251
1252        /* First, try a score-based selection if both transport states
1253         * differ. If we're in a tie, lets try to make a more clever
1254         * decision here based on error counts and last time heard.
1255         */
1256        if (score_curr > score_best)
1257                return curr;
1258        else if (score_curr == score_best)
1259                return sctp_trans_elect_tie(curr, best);
1260        else
1261                return best;
1262}
1263
1264void sctp_assoc_update_retran_path(struct sctp_association *asoc)
1265{
1266        struct sctp_transport *trans = asoc->peer.retran_path;
1267        struct sctp_transport *trans_next = NULL;
1268
1269        /* We're done as we only have the one and only path. */
1270        if (asoc->peer.transport_count == 1)
1271                return;
1272        /* If active_path and retran_path are the same and active,
1273         * then this is the only active path. Use it.
1274         */
1275        if (asoc->peer.active_path == asoc->peer.retran_path &&
1276            asoc->peer.active_path->state == SCTP_ACTIVE)
1277                return;
1278
1279        /* Iterate from retran_path's successor back to retran_path. */
1280        for (trans = list_next_entry(trans, transports); 1;
1281             trans = list_next_entry(trans, transports)) {
1282                /* Manually skip the head element. */
1283                if (&trans->transports == &asoc->peer.transport_addr_list)
1284                        continue;
1285                if (trans->state == SCTP_UNCONFIRMED)
1286                        continue;
1287                trans_next = sctp_trans_elect_best(trans, trans_next);
1288                /* Active is good enough for immediate return. */
1289                if (trans_next->state == SCTP_ACTIVE)
1290                        break;
1291                /* We've reached the end, time to update path. */
1292                if (trans == asoc->peer.retran_path)
1293                        break;
1294        }
1295
1296        asoc->peer.retran_path = trans_next;
1297
1298        pr_debug("%s: association:%p updated new path to addr:%pISpc\n",
1299                 __func__, asoc, &asoc->peer.retran_path->ipaddr.sa);
1300}
1301
1302static void sctp_select_active_and_retran_path(struct sctp_association *asoc)
1303{
1304        struct sctp_transport *trans, *trans_pri = NULL, *trans_sec = NULL;
1305        struct sctp_transport *trans_pf = NULL;
1306
1307        /* Look for the two most recently used active transports. */
1308        list_for_each_entry(trans, &asoc->peer.transport_addr_list,
1309                            transports) {
1310                /* Skip uninteresting transports. */
1311                if (trans->state == SCTP_INACTIVE ||
1312                    trans->state == SCTP_UNCONFIRMED)
1313                        continue;
1314                /* Keep track of the best PF transport from our
1315                 * list in case we don't find an active one.
1316                 */
1317                if (trans->state == SCTP_PF) {
1318                        trans_pf = sctp_trans_elect_best(trans, trans_pf);
1319                        continue;
1320                }
1321                /* For active transports, pick the most recent ones. */
1322                if (trans_pri == NULL ||
1323                    ktime_after(trans->last_time_heard,
1324                                trans_pri->last_time_heard)) {
1325                        trans_sec = trans_pri;
1326                        trans_pri = trans;
1327                } else if (trans_sec == NULL ||
1328                           ktime_after(trans->last_time_heard,
1329                                       trans_sec->last_time_heard)) {
1330                        trans_sec = trans;
1331                }
1332        }
1333
1334        /* RFC 2960 6.4 Multi-Homed SCTP Endpoints
1335         *
1336         * By default, an endpoint should always transmit to the primary
1337         * path, unless the SCTP user explicitly specifies the
1338         * destination transport address (and possibly source transport
1339         * address) to use. [If the primary is active but not most recent,
1340         * bump the most recently used transport.]
1341         */
1342        if ((asoc->peer.primary_path->state == SCTP_ACTIVE ||
1343             asoc->peer.primary_path->state == SCTP_UNKNOWN) &&
1344             asoc->peer.primary_path != trans_pri) {
1345                trans_sec = trans_pri;
1346                trans_pri = asoc->peer.primary_path;
1347        }
1348
1349        /* We did not find anything useful for a possible retransmission
1350         * path; either primary path that we found is the the same as
1351         * the current one, or we didn't generally find an active one.
1352         */
1353        if (trans_sec == NULL)
1354                trans_sec = trans_pri;
1355
1356        /* If we failed to find a usable transport, just camp on the
1357         * active or pick a PF iff it's the better choice.
1358         */
1359        if (trans_pri == NULL) {
1360                trans_pri = sctp_trans_elect_best(asoc->peer.active_path, trans_pf);
1361                trans_sec = trans_pri;
1362        }
1363
1364        /* Set the active and retran transports. */
1365        asoc->peer.active_path = trans_pri;
1366        asoc->peer.retran_path = trans_sec;
1367}
1368
1369struct sctp_transport *
1370sctp_assoc_choose_alter_transport(struct sctp_association *asoc,
1371                                  struct sctp_transport *last_sent_to)
1372{
1373        /* If this is the first time packet is sent, use the active path,
1374         * else use the retran path. If the last packet was sent over the
1375         * retran path, update the retran path and use it.
1376         */
1377        if (last_sent_to == NULL) {
1378                return asoc->peer.active_path;
1379        } else {
1380                if (last_sent_to == asoc->peer.retran_path)
1381                        sctp_assoc_update_retran_path(asoc);
1382
1383                return asoc->peer.retran_path;
1384        }
1385}
1386
1387/* Update the association's pmtu and frag_point by going through all the
1388 * transports. This routine is called when a transport's PMTU has changed.
1389 */
1390void sctp_assoc_sync_pmtu(struct sock *sk, struct sctp_association *asoc)
1391{
1392        struct sctp_transport *t;
1393        __u32 pmtu = 0;
1394
1395        if (!asoc)
1396                return;
1397
1398        /* Get the lowest pmtu of all the transports. */
1399        list_for_each_entry(t, &asoc->peer.transport_addr_list,
1400                                transports) {
1401                if (t->pmtu_pending && t->dst) {
1402                        sctp_transport_update_pmtu(sk, t, dst_mtu(t->dst));
1403                        t->pmtu_pending = 0;
1404                }
1405                if (!pmtu || (t->pathmtu < pmtu))
1406                        pmtu = t->pathmtu;
1407        }
1408
1409        if (pmtu) {
1410                asoc->pathmtu = pmtu;
1411                asoc->frag_point = sctp_frag_point(asoc, pmtu);
1412        }
1413
1414        pr_debug("%s: asoc:%p, pmtu:%d, frag_point:%d\n", __func__, asoc,
1415                 asoc->pathmtu, asoc->frag_point);
1416}
1417
1418/* Should we send a SACK to update our peer? */
1419static inline bool sctp_peer_needs_update(struct sctp_association *asoc)
1420{
1421        struct net *net = sock_net(asoc->base.sk);
1422        switch (asoc->state) {
1423        case SCTP_STATE_ESTABLISHED:
1424        case SCTP_STATE_SHUTDOWN_PENDING:
1425        case SCTP_STATE_SHUTDOWN_RECEIVED:
1426        case SCTP_STATE_SHUTDOWN_SENT:
1427                if ((asoc->rwnd > asoc->a_rwnd) &&
1428                    ((asoc->rwnd - asoc->a_rwnd) >= max_t(__u32,
1429                           (asoc->base.sk->sk_rcvbuf >> net->sctp.rwnd_upd_shift),
1430                           asoc->pathmtu)))
1431                        return true;
1432                break;
1433        default:
1434                break;
1435        }
1436        return false;
1437}
1438
1439/* Increase asoc's rwnd by len and send any window update SACK if needed. */
1440void sctp_assoc_rwnd_increase(struct sctp_association *asoc, unsigned int len)
1441{
1442        struct sctp_chunk *sack;
1443        struct timer_list *timer;
1444
1445        if (asoc->rwnd_over) {
1446                if (asoc->rwnd_over >= len) {
1447                        asoc->rwnd_over -= len;
1448                } else {
1449                        asoc->rwnd += (len - asoc->rwnd_over);
1450                        asoc->rwnd_over = 0;
1451                }
1452        } else {
1453                asoc->rwnd += len;
1454        }
1455
1456        /* If we had window pressure, start recovering it
1457         * once our rwnd had reached the accumulated pressure
1458         * threshold.  The idea is to recover slowly, but up
1459         * to the initial advertised window.
1460         */
1461        if (asoc->rwnd_press && asoc->rwnd >= asoc->rwnd_press) {
1462                int change = min(asoc->pathmtu, asoc->rwnd_press);
1463                asoc->rwnd += change;
1464                asoc->rwnd_press -= change;
1465        }
1466
1467        pr_debug("%s: asoc:%p rwnd increased by %d to (%u, %u) - %u\n",
1468                 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
1469                 asoc->a_rwnd);
1470
1471        /* Send a window update SACK if the rwnd has increased by at least the
1472         * minimum of the association's PMTU and half of the receive buffer.
1473         * The algorithm used is similar to the one described in
1474         * Section 4.2.3.3 of RFC 1122.
1475         */
1476        if (sctp_peer_needs_update(asoc)) {
1477                asoc->a_rwnd = asoc->rwnd;
1478
1479                pr_debug("%s: sending window update SACK- asoc:%p rwnd:%u "
1480                         "a_rwnd:%u\n", __func__, asoc, asoc->rwnd,
1481                         asoc->a_rwnd);
1482
1483                sack = sctp_make_sack(asoc);
1484                if (!sack)
1485                        return;
1486
1487                asoc->peer.sack_needed = 0;
1488
1489                sctp_outq_tail(&asoc->outqueue, sack);
1490
1491                /* Stop the SACK timer.  */
1492                timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
1493                if (del_timer(timer))
1494                        sctp_association_put(asoc);
1495        }
1496}
1497
1498/* Decrease asoc's rwnd by len. */
1499void sctp_assoc_rwnd_decrease(struct sctp_association *asoc, unsigned int len)
1500{
1501        int rx_count;
1502        int over = 0;
1503
1504        if (unlikely(!asoc->rwnd || asoc->rwnd_over))
1505                pr_debug("%s: association:%p has asoc->rwnd:%u, "
1506                         "asoc->rwnd_over:%u!\n", __func__, asoc,
1507                         asoc->rwnd, asoc->rwnd_over);
1508
1509        if (asoc->ep->rcvbuf_policy)
1510                rx_count = atomic_read(&asoc->rmem_alloc);
1511        else
1512                rx_count = atomic_read(&asoc->base.sk->sk_rmem_alloc);
1513
1514        /* If we've reached or overflowed our receive buffer, announce
1515         * a 0 rwnd if rwnd would still be positive.  Store the
1516         * the potential pressure overflow so that the window can be restored
1517         * back to original value.
1518         */
1519        if (rx_count >= asoc->base.sk->sk_rcvbuf)
1520                over = 1;
1521
1522        if (asoc->rwnd >= len) {
1523                asoc->rwnd -= len;
1524                if (over) {
1525                        asoc->rwnd_press += asoc->rwnd;
1526                        asoc->rwnd = 0;
1527                }
1528        } else {
1529                asoc->rwnd_over = len - asoc->rwnd;
1530                asoc->rwnd = 0;
1531        }
1532
1533        pr_debug("%s: asoc:%p rwnd decreased by %d to (%u, %u, %u)\n",
1534                 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
1535                 asoc->rwnd_press);
1536}
1537
1538/* Build the bind address list for the association based on info from the
1539 * local endpoint and the remote peer.
1540 */
1541int sctp_assoc_set_bind_addr_from_ep(struct sctp_association *asoc,
1542                                     sctp_scope_t scope, gfp_t gfp)
1543{
1544        int flags;
1545
1546        /* Use scoping rules to determine the subset of addresses from
1547         * the endpoint.
1548         */
1549        flags = (PF_INET6 == asoc->base.sk->sk_family) ? SCTP_ADDR6_ALLOWED : 0;
1550        if (asoc->peer.ipv4_address)
1551                flags |= SCTP_ADDR4_PEERSUPP;
1552        if (asoc->peer.ipv6_address)
1553                flags |= SCTP_ADDR6_PEERSUPP;
1554
1555        return sctp_bind_addr_copy(sock_net(asoc->base.sk),
1556                                   &asoc->base.bind_addr,
1557                                   &asoc->ep->base.bind_addr,
1558                                   scope, gfp, flags);
1559}
1560
1561/* Build the association's bind address list from the cookie.  */
1562int sctp_assoc_set_bind_addr_from_cookie(struct sctp_association *asoc,
1563                                         struct sctp_cookie *cookie,
1564                                         gfp_t gfp)
1565{
1566        int var_size2 = ntohs(cookie->peer_init->chunk_hdr.length);
1567        int var_size3 = cookie->raw_addr_list_len;
1568        __u8 *raw = (__u8 *)cookie->peer_init + var_size2;
1569
1570        return sctp_raw_to_bind_addrs(&asoc->base.bind_addr, raw, var_size3,
1571                                      asoc->ep->base.bind_addr.port, gfp);
1572}
1573
1574/* Lookup laddr in the bind address list of an association. */
1575int sctp_assoc_lookup_laddr(struct sctp_association *asoc,
1576                            const union sctp_addr *laddr)
1577{
1578        int found = 0;
1579
1580        if ((asoc->base.bind_addr.port == ntohs(laddr->v4.sin_port)) &&
1581            sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
1582                                 sctp_sk(asoc->base.sk)))
1583                found = 1;
1584
1585        return found;
1586}
1587
1588/* Set an association id for a given association */
1589int sctp_assoc_set_id(struct sctp_association *asoc, gfp_t gfp)
1590{
1591        bool preload = !!(gfp & __GFP_WAIT);
1592        int ret;
1593
1594        /* If the id is already assigned, keep it. */
1595        if (asoc->assoc_id)
1596                return 0;
1597
1598        if (preload)
1599                idr_preload(gfp);
1600        spin_lock_bh(&sctp_assocs_id_lock);
1601        /* 0 is not a valid assoc_id, must be >= 1 */
1602        ret = idr_alloc_cyclic(&sctp_assocs_id, asoc, 1, 0, GFP_NOWAIT);
1603        spin_unlock_bh(&sctp_assocs_id_lock);
1604        if (preload)
1605                idr_preload_end();
1606        if (ret < 0)
1607                return ret;
1608
1609        asoc->assoc_id = (sctp_assoc_t)ret;
1610        return 0;
1611}
1612
1613/* Free the ASCONF queue */
1614static void sctp_assoc_free_asconf_queue(struct sctp_association *asoc)
1615{
1616        struct sctp_chunk *asconf;
1617        struct sctp_chunk *tmp;
1618
1619        list_for_each_entry_safe(asconf, tmp, &asoc->addip_chunk_list, list) {
1620                list_del_init(&asconf->list);
1621                sctp_chunk_free(asconf);
1622        }
1623}
1624
1625/* Free asconf_ack cache */
1626static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc)
1627{
1628        struct sctp_chunk *ack;
1629        struct sctp_chunk *tmp;
1630
1631        list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
1632                                transmitted_list) {
1633                list_del_init(&ack->transmitted_list);
1634                sctp_chunk_free(ack);
1635        }
1636}
1637
1638/* Clean up the ASCONF_ACK queue */
1639void sctp_assoc_clean_asconf_ack_cache(const struct sctp_association *asoc)
1640{
1641        struct sctp_chunk *ack;
1642        struct sctp_chunk *tmp;
1643
1644        /* We can remove all the entries from the queue up to
1645         * the "Peer-Sequence-Number".
1646         */
1647        list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
1648                                transmitted_list) {
1649                if (ack->subh.addip_hdr->serial ==
1650                                htonl(asoc->peer.addip_serial))
1651                        break;
1652
1653                list_del_init(&ack->transmitted_list);
1654                sctp_chunk_free(ack);
1655        }
1656}
1657
1658/* Find the ASCONF_ACK whose serial number matches ASCONF */
1659struct sctp_chunk *sctp_assoc_lookup_asconf_ack(
1660                                        const struct sctp_association *asoc,
1661                                        __be32 serial)
1662{
1663        struct sctp_chunk *ack;
1664
1665        /* Walk through the list of cached ASCONF-ACKs and find the
1666         * ack chunk whose serial number matches that of the request.
1667         */
1668        list_for_each_entry(ack, &asoc->asconf_ack_list, transmitted_list) {
1669                if (sctp_chunk_pending(ack))
1670                        continue;
1671                if (ack->subh.addip_hdr->serial == serial) {
1672                        sctp_chunk_hold(ack);
1673                        return ack;
1674                }
1675        }
1676
1677        return NULL;
1678}
1679
1680void sctp_asconf_queue_teardown(struct sctp_association *asoc)
1681{
1682        /* Free any cached ASCONF_ACK chunk. */
1683        sctp_assoc_free_asconf_acks(asoc);
1684
1685        /* Free the ASCONF queue. */
1686        sctp_assoc_free_asconf_queue(asoc);
1687
1688        /* Free any cached ASCONF chunk. */
1689        if (asoc->addip_last_asconf)
1690                sctp_chunk_free(asoc->addip_last_asconf);
1691}
1692