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_unhash_transport(transport);
 387                sctp_transport_free(transport);
 388        }
 389
 390        asoc->peer.transport_count = 0;
 391
 392        sctp_asconf_queue_teardown(asoc);
 393
 394        /* Free pending address space being deleted */
 395        kfree(asoc->asconf_addr_del_pending);
 396
 397        /* AUTH - Free the endpoint shared keys */
 398        sctp_auth_destroy_keys(&asoc->endpoint_shared_keys);
 399
 400        /* AUTH - Free the association shared key */
 401        sctp_auth_key_put(asoc->asoc_shared_key);
 402
 403        sctp_association_put(asoc);
 404}
 405
 406/* Cleanup and free up an association. */
 407static void sctp_association_destroy(struct sctp_association *asoc)
 408{
 409        if (unlikely(!asoc->base.dead)) {
 410                WARN(1, "Attempt to destroy undead association %p!\n", asoc);
 411                return;
 412        }
 413
 414        sctp_endpoint_put(asoc->ep);
 415        sock_put(asoc->base.sk);
 416
 417        if (asoc->assoc_id != 0) {
 418                spin_lock_bh(&sctp_assocs_id_lock);
 419                idr_remove(&sctp_assocs_id, asoc->assoc_id);
 420                spin_unlock_bh(&sctp_assocs_id_lock);
 421        }
 422
 423        WARN_ON(atomic_read(&asoc->rmem_alloc));
 424
 425        kfree(asoc);
 426        SCTP_DBG_OBJCNT_DEC(assoc);
 427}
 428
 429/* Change the primary destination address for the peer. */
 430void sctp_assoc_set_primary(struct sctp_association *asoc,
 431                            struct sctp_transport *transport)
 432{
 433        int changeover = 0;
 434
 435        /* it's a changeover only if we already have a primary path
 436         * that we are changing
 437         */
 438        if (asoc->peer.primary_path != NULL &&
 439            asoc->peer.primary_path != transport)
 440                changeover = 1 ;
 441
 442        asoc->peer.primary_path = transport;
 443
 444        /* Set a default msg_name for events. */
 445        memcpy(&asoc->peer.primary_addr, &transport->ipaddr,
 446               sizeof(union sctp_addr));
 447
 448        /* If the primary path is changing, assume that the
 449         * user wants to use this new path.
 450         */
 451        if ((transport->state == SCTP_ACTIVE) ||
 452            (transport->state == SCTP_UNKNOWN))
 453                asoc->peer.active_path = transport;
 454
 455        /*
 456         * SFR-CACC algorithm:
 457         * Upon the receipt of a request to change the primary
 458         * destination address, on the data structure for the new
 459         * primary destination, the sender MUST do the following:
 460         *
 461         * 1) If CHANGEOVER_ACTIVE is set, then there was a switch
 462         * to this destination address earlier. The sender MUST set
 463         * CYCLING_CHANGEOVER to indicate that this switch is a
 464         * double switch to the same destination address.
 465         *
 466         * Really, only bother is we have data queued or outstanding on
 467         * the association.
 468         */
 469        if (!asoc->outqueue.outstanding_bytes && !asoc->outqueue.out_qlen)
 470                return;
 471
 472        if (transport->cacc.changeover_active)
 473                transport->cacc.cycling_changeover = changeover;
 474
 475        /* 2) The sender MUST set CHANGEOVER_ACTIVE to indicate that
 476         * a changeover has occurred.
 477         */
 478        transport->cacc.changeover_active = changeover;
 479
 480        /* 3) The sender MUST store the next TSN to be sent in
 481         * next_tsn_at_change.
 482         */
 483        transport->cacc.next_tsn_at_change = asoc->next_tsn;
 484}
 485
 486/* Remove a transport from an association.  */
 487void sctp_assoc_rm_peer(struct sctp_association *asoc,
 488                        struct sctp_transport *peer)
 489{
 490        struct list_head        *pos;
 491        struct sctp_transport   *transport;
 492
 493        pr_debug("%s: association:%p addr:%pISpc\n",
 494                 __func__, asoc, &peer->ipaddr.sa);
 495
 496        /* If we are to remove the current retran_path, update it
 497         * to the next peer before removing this peer from the list.
 498         */
 499        if (asoc->peer.retran_path == peer)
 500                sctp_assoc_update_retran_path(asoc);
 501
 502        /* Remove this peer from the list. */
 503        list_del_rcu(&peer->transports);
 504        /* Remove this peer from the transport hashtable */
 505        sctp_unhash_transport(peer);
 506
 507        /* Get the first transport of asoc. */
 508        pos = asoc->peer.transport_addr_list.next;
 509        transport = list_entry(pos, struct sctp_transport, transports);
 510
 511        /* Update any entries that match the peer to be deleted. */
 512        if (asoc->peer.primary_path == peer)
 513                sctp_assoc_set_primary(asoc, transport);
 514        if (asoc->peer.active_path == peer)
 515                asoc->peer.active_path = transport;
 516        if (asoc->peer.retran_path == peer)
 517                asoc->peer.retran_path = transport;
 518        if (asoc->peer.last_data_from == peer)
 519                asoc->peer.last_data_from = transport;
 520
 521        /* If we remove the transport an INIT was last sent to, set it to
 522         * NULL. Combined with the update of the retran path above, this
 523         * will cause the next INIT to be sent to the next available
 524         * transport, maintaining the cycle.
 525         */
 526        if (asoc->init_last_sent_to == peer)
 527                asoc->init_last_sent_to = NULL;
 528
 529        /* If we remove the transport an SHUTDOWN was last sent to, set it
 530         * to NULL. Combined with the update of the retran path above, this
 531         * will cause the next SHUTDOWN to be sent to the next available
 532         * transport, maintaining the cycle.
 533         */
 534        if (asoc->shutdown_last_sent_to == peer)
 535                asoc->shutdown_last_sent_to = NULL;
 536
 537        /* If we remove the transport an ASCONF was last sent to, set it to
 538         * NULL.
 539         */
 540        if (asoc->addip_last_asconf &&
 541            asoc->addip_last_asconf->transport == peer)
 542                asoc->addip_last_asconf->transport = NULL;
 543
 544        /* If we have something on the transmitted list, we have to
 545         * save it off.  The best place is the active path.
 546         */
 547        if (!list_empty(&peer->transmitted)) {
 548                struct sctp_transport *active = asoc->peer.active_path;
 549                struct sctp_chunk *ch;
 550
 551                /* Reset the transport of each chunk on this list */
 552                list_for_each_entry(ch, &peer->transmitted,
 553                                        transmitted_list) {
 554                        ch->transport = NULL;
 555                        ch->rtt_in_progress = 0;
 556                }
 557
 558                list_splice_tail_init(&peer->transmitted,
 559                                        &active->transmitted);
 560
 561                /* Start a T3 timer here in case it wasn't running so
 562                 * that these migrated packets have a chance to get
 563                 * retransmitted.
 564                 */
 565                if (!timer_pending(&active->T3_rtx_timer))
 566                        if (!mod_timer(&active->T3_rtx_timer,
 567                                        jiffies + active->rto))
 568                                sctp_transport_hold(active);
 569        }
 570
 571        asoc->peer.transport_count--;
 572
 573        sctp_transport_free(peer);
 574}
 575
 576/* Add a transport address to an association.  */
 577struct sctp_transport *sctp_assoc_add_peer(struct sctp_association *asoc,
 578                                           const union sctp_addr *addr,
 579                                           const gfp_t gfp,
 580                                           const int peer_state)
 581{
 582        struct net *net = sock_net(asoc->base.sk);
 583        struct sctp_transport *peer;
 584        struct sctp_sock *sp;
 585        unsigned short port;
 586
 587        sp = sctp_sk(asoc->base.sk);
 588
 589        /* AF_INET and AF_INET6 share common port field. */
 590        port = ntohs(addr->v4.sin_port);
 591
 592        pr_debug("%s: association:%p addr:%pISpc state:%d\n", __func__,
 593                 asoc, &addr->sa, peer_state);
 594
 595        /* Set the port if it has not been set yet.  */
 596        if (0 == asoc->peer.port)
 597                asoc->peer.port = port;
 598
 599        /* Check to see if this is a duplicate. */
 600        peer = sctp_assoc_lookup_paddr(asoc, addr);
 601        if (peer) {
 602                /* An UNKNOWN state is only set on transports added by
 603                 * user in sctp_connectx() call.  Such transports should be
 604                 * considered CONFIRMED per RFC 4960, Section 5.4.
 605                 */
 606                if (peer->state == SCTP_UNKNOWN) {
 607                        peer->state = SCTP_ACTIVE;
 608                }
 609                return peer;
 610        }
 611
 612        peer = sctp_transport_new(net, addr, gfp);
 613        if (!peer)
 614                return NULL;
 615
 616        sctp_transport_set_owner(peer, asoc);
 617
 618        /* Initialize the peer's heartbeat interval based on the
 619         * association configured value.
 620         */
 621        peer->hbinterval = asoc->hbinterval;
 622
 623        /* Set the path max_retrans.  */
 624        peer->pathmaxrxt = asoc->pathmaxrxt;
 625
 626        /* And the partial failure retrans threshold */
 627        peer->pf_retrans = asoc->pf_retrans;
 628
 629        /* Initialize the peer's SACK delay timeout based on the
 630         * association configured value.
 631         */
 632        peer->sackdelay = asoc->sackdelay;
 633        peer->sackfreq = asoc->sackfreq;
 634
 635        /* Enable/disable heartbeat, SACK delay, and path MTU discovery
 636         * based on association setting.
 637         */
 638        peer->param_flags = asoc->param_flags;
 639
 640        sctp_transport_route(peer, NULL, sp);
 641
 642        /* Initialize the pmtu of the transport. */
 643        if (peer->param_flags & SPP_PMTUD_DISABLE) {
 644                if (asoc->pathmtu)
 645                        peer->pathmtu = asoc->pathmtu;
 646                else
 647                        peer->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
 648        }
 649
 650        /* If this is the first transport addr on this association,
 651         * initialize the association PMTU to the peer's PMTU.
 652         * If not and the current association PMTU is higher than the new
 653         * peer's PMTU, reset the association PMTU to the new peer's PMTU.
 654         */
 655        if (asoc->pathmtu)
 656                asoc->pathmtu = min_t(int, peer->pathmtu, asoc->pathmtu);
 657        else
 658                asoc->pathmtu = peer->pathmtu;
 659
 660        pr_debug("%s: association:%p PMTU set to %d\n", __func__, asoc,
 661                 asoc->pathmtu);
 662
 663        peer->pmtu_pending = 0;
 664
 665        asoc->frag_point = sctp_frag_point(asoc, asoc->pathmtu);
 666
 667        /* The asoc->peer.port might not be meaningful yet, but
 668         * initialize the packet structure anyway.
 669         */
 670        sctp_packet_init(&peer->packet, peer, asoc->base.bind_addr.port,
 671                         asoc->peer.port);
 672
 673        /* 7.2.1 Slow-Start
 674         *
 675         * o The initial cwnd before DATA transmission or after a sufficiently
 676         *   long idle period MUST be set to
 677         *      min(4*MTU, max(2*MTU, 4380 bytes))
 678         *
 679         * o The initial value of ssthresh MAY be arbitrarily high
 680         *   (for example, implementations MAY use the size of the
 681         *   receiver advertised window).
 682         */
 683        peer->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
 684
 685        /* At this point, we may not have the receiver's advertised window,
 686         * so initialize ssthresh to the default value and it will be set
 687         * later when we process the INIT.
 688         */
 689        peer->ssthresh = SCTP_DEFAULT_MAXWINDOW;
 690
 691        peer->partial_bytes_acked = 0;
 692        peer->flight_size = 0;
 693        peer->burst_limited = 0;
 694
 695        /* Set the transport's RTO.initial value */
 696        peer->rto = asoc->rto_initial;
 697        sctp_max_rto(asoc, peer);
 698
 699        /* Set the peer's active state. */
 700        peer->state = peer_state;
 701
 702        /* Attach the remote transport to our asoc.  */
 703        list_add_tail_rcu(&peer->transports, &asoc->peer.transport_addr_list);
 704        asoc->peer.transport_count++;
 705        /* Add this peer into the transport hashtable */
 706        sctp_hash_transport(peer);
 707
 708        /* If we do not yet have a primary path, set one.  */
 709        if (!asoc->peer.primary_path) {
 710                sctp_assoc_set_primary(asoc, peer);
 711                asoc->peer.retran_path = peer;
 712        }
 713
 714        if (asoc->peer.active_path == asoc->peer.retran_path &&
 715            peer->state != SCTP_UNCONFIRMED) {
 716                asoc->peer.retran_path = peer;
 717        }
 718
 719        return peer;
 720}
 721
 722/* Delete a transport address from an association.  */
 723void sctp_assoc_del_peer(struct sctp_association *asoc,
 724                         const union sctp_addr *addr)
 725{
 726        struct list_head        *pos;
 727        struct list_head        *temp;
 728        struct sctp_transport   *transport;
 729
 730        list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
 731                transport = list_entry(pos, struct sctp_transport, transports);
 732                if (sctp_cmp_addr_exact(addr, &transport->ipaddr)) {
 733                        /* Do book keeping for removing the peer and free it. */
 734                        sctp_assoc_rm_peer(asoc, transport);
 735                        break;
 736                }
 737        }
 738}
 739
 740/* Lookup a transport by address. */
 741struct sctp_transport *sctp_assoc_lookup_paddr(
 742                                        const struct sctp_association *asoc,
 743                                        const union sctp_addr *address)
 744{
 745        struct sctp_transport *t;
 746
 747        /* Cycle through all transports searching for a peer address. */
 748
 749        list_for_each_entry(t, &asoc->peer.transport_addr_list,
 750                        transports) {
 751                if (sctp_cmp_addr_exact(address, &t->ipaddr))
 752                        return t;
 753        }
 754
 755        return NULL;
 756}
 757
 758/* Remove all transports except a give one */
 759void sctp_assoc_del_nonprimary_peers(struct sctp_association *asoc,
 760                                     struct sctp_transport *primary)
 761{
 762        struct sctp_transport   *temp;
 763        struct sctp_transport   *t;
 764
 765        list_for_each_entry_safe(t, temp, &asoc->peer.transport_addr_list,
 766                                 transports) {
 767                /* if the current transport is not the primary one, delete it */
 768                if (t != primary)
 769                        sctp_assoc_rm_peer(asoc, t);
 770        }
 771}
 772
 773/* Engage in transport control operations.
 774 * Mark the transport up or down and send a notification to the user.
 775 * Select and update the new active and retran paths.
 776 */
 777void sctp_assoc_control_transport(struct sctp_association *asoc,
 778                                  struct sctp_transport *transport,
 779                                  sctp_transport_cmd_t command,
 780                                  sctp_sn_error_t error)
 781{
 782        struct sctp_ulpevent *event;
 783        struct sockaddr_storage addr;
 784        int spc_state = 0;
 785        bool ulp_notify = true;
 786
 787        /* Record the transition on the transport.  */
 788        switch (command) {
 789        case SCTP_TRANSPORT_UP:
 790                /* If we are moving from UNCONFIRMED state due
 791                 * to heartbeat success, report the SCTP_ADDR_CONFIRMED
 792                 * state to the user, otherwise report SCTP_ADDR_AVAILABLE.
 793                 */
 794                if (SCTP_UNCONFIRMED == transport->state &&
 795                    SCTP_HEARTBEAT_SUCCESS == error)
 796                        spc_state = SCTP_ADDR_CONFIRMED;
 797                else
 798                        spc_state = SCTP_ADDR_AVAILABLE;
 799                /* Don't inform ULP about transition from PF to
 800                 * active state and set cwnd to 1 MTU, see SCTP
 801                 * Quick failover draft section 5.1, point 5
 802                 */
 803                if (transport->state == SCTP_PF) {
 804                        ulp_notify = false;
 805                        transport->cwnd = asoc->pathmtu;
 806                }
 807                transport->state = SCTP_ACTIVE;
 808                break;
 809
 810        case SCTP_TRANSPORT_DOWN:
 811                /* If the transport was never confirmed, do not transition it
 812                 * to inactive state.  Also, release the cached route since
 813                 * there may be a better route next time.
 814                 */
 815                if (transport->state != SCTP_UNCONFIRMED)
 816                        transport->state = SCTP_INACTIVE;
 817                else {
 818                        dst_release(transport->dst);
 819                        transport->dst = NULL;
 820                        ulp_notify = false;
 821                }
 822
 823                spc_state = SCTP_ADDR_UNREACHABLE;
 824                break;
 825
 826        case SCTP_TRANSPORT_PF:
 827                transport->state = SCTP_PF;
 828                ulp_notify = false;
 829                break;
 830
 831        default:
 832                return;
 833        }
 834
 835        /* Generate and send a SCTP_PEER_ADDR_CHANGE notification
 836         * to the user.
 837         */
 838        if (ulp_notify) {
 839                memset(&addr, 0, sizeof(struct sockaddr_storage));
 840                memcpy(&addr, &transport->ipaddr,
 841                       transport->af_specific->sockaddr_len);
 842
 843                event = sctp_ulpevent_make_peer_addr_change(asoc, &addr,
 844                                        0, spc_state, error, GFP_ATOMIC);
 845                if (event)
 846                        sctp_ulpq_tail_event(&asoc->ulpq, event);
 847        }
 848
 849        /* Select new active and retran paths. */
 850        sctp_select_active_and_retran_path(asoc);
 851}
 852
 853/* Hold a reference to an association. */
 854void sctp_association_hold(struct sctp_association *asoc)
 855{
 856        atomic_inc(&asoc->base.refcnt);
 857}
 858
 859/* Release a reference to an association and cleanup
 860 * if there are no more references.
 861 */
 862void sctp_association_put(struct sctp_association *asoc)
 863{
 864        if (atomic_dec_and_test(&asoc->base.refcnt))
 865                sctp_association_destroy(asoc);
 866}
 867
 868/* Allocate the next TSN, Transmission Sequence Number, for the given
 869 * association.
 870 */
 871__u32 sctp_association_get_next_tsn(struct sctp_association *asoc)
 872{
 873        /* From Section 1.6 Serial Number Arithmetic:
 874         * Transmission Sequence Numbers wrap around when they reach
 875         * 2**32 - 1.  That is, the next TSN a DATA chunk MUST use
 876         * after transmitting TSN = 2*32 - 1 is TSN = 0.
 877         */
 878        __u32 retval = asoc->next_tsn;
 879        asoc->next_tsn++;
 880        asoc->unack_data++;
 881
 882        return retval;
 883}
 884
 885/* Compare two addresses to see if they match.  Wildcard addresses
 886 * only match themselves.
 887 */
 888int sctp_cmp_addr_exact(const union sctp_addr *ss1,
 889                        const union sctp_addr *ss2)
 890{
 891        struct sctp_af *af;
 892
 893        af = sctp_get_af_specific(ss1->sa.sa_family);
 894        if (unlikely(!af))
 895                return 0;
 896
 897        return af->cmp_addr(ss1, ss2);
 898}
 899
 900/* Return an ecne chunk to get prepended to a packet.
 901 * Note:  We are sly and return a shared, prealloced chunk.  FIXME:
 902 * No we don't, but we could/should.
 903 */
 904struct sctp_chunk *sctp_get_ecne_prepend(struct sctp_association *asoc)
 905{
 906        if (!asoc->need_ecne)
 907                return NULL;
 908
 909        /* Send ECNE if needed.
 910         * Not being able to allocate a chunk here is not deadly.
 911         */
 912        return sctp_make_ecne(asoc, asoc->last_ecne_tsn);
 913}
 914
 915/*
 916 * Find which transport this TSN was sent on.
 917 */
 918struct sctp_transport *sctp_assoc_lookup_tsn(struct sctp_association *asoc,
 919                                             __u32 tsn)
 920{
 921        struct sctp_transport *active;
 922        struct sctp_transport *match;
 923        struct sctp_transport *transport;
 924        struct sctp_chunk *chunk;
 925        __be32 key = htonl(tsn);
 926
 927        match = NULL;
 928
 929        /*
 930         * FIXME: In general, find a more efficient data structure for
 931         * searching.
 932         */
 933
 934        /*
 935         * The general strategy is to search each transport's transmitted
 936         * list.   Return which transport this TSN lives on.
 937         *
 938         * Let's be hopeful and check the active_path first.
 939         * Another optimization would be to know if there is only one
 940         * outbound path and not have to look for the TSN at all.
 941         *
 942         */
 943
 944        active = asoc->peer.active_path;
 945
 946        list_for_each_entry(chunk, &active->transmitted,
 947                        transmitted_list) {
 948
 949                if (key == chunk->subh.data_hdr->tsn) {
 950                        match = active;
 951                        goto out;
 952                }
 953        }
 954
 955        /* If not found, go search all the other transports. */
 956        list_for_each_entry(transport, &asoc->peer.transport_addr_list,
 957                        transports) {
 958
 959                if (transport == active)
 960                        continue;
 961                list_for_each_entry(chunk, &transport->transmitted,
 962                                transmitted_list) {
 963                        if (key == chunk->subh.data_hdr->tsn) {
 964                                match = transport;
 965                                goto out;
 966                        }
 967                }
 968        }
 969out:
 970        return match;
 971}
 972
 973/* Is this the association we are looking for? */
 974struct sctp_transport *sctp_assoc_is_match(struct sctp_association *asoc,
 975                                           struct net *net,
 976                                           const union sctp_addr *laddr,
 977                                           const union sctp_addr *paddr)
 978{
 979        struct sctp_transport *transport;
 980
 981        if ((htons(asoc->base.bind_addr.port) == laddr->v4.sin_port) &&
 982            (htons(asoc->peer.port) == paddr->v4.sin_port) &&
 983            net_eq(sock_net(asoc->base.sk), net)) {
 984                transport = sctp_assoc_lookup_paddr(asoc, paddr);
 985                if (!transport)
 986                        goto out;
 987
 988                if (sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
 989                                         sctp_sk(asoc->base.sk)))
 990                        goto out;
 991        }
 992        transport = NULL;
 993
 994out:
 995        return transport;
 996}
 997
 998/* Do delayed input processing.  This is scheduled by sctp_rcv(). */
 999static void sctp_assoc_bh_rcv(struct work_struct *work)
1000{
1001        struct sctp_association *asoc =
1002                container_of(work, struct sctp_association,
1003                             base.inqueue.immediate);
1004        struct net *net = sock_net(asoc->base.sk);
1005        struct sctp_endpoint *ep;
1006        struct sctp_chunk *chunk;
1007        struct sctp_inq *inqueue;
1008        int state;
1009        sctp_subtype_t subtype;
1010        int error = 0;
1011
1012        /* The association should be held so we should be safe. */
1013        ep = asoc->ep;
1014
1015        inqueue = &asoc->base.inqueue;
1016        sctp_association_hold(asoc);
1017        while (NULL != (chunk = sctp_inq_pop(inqueue))) {
1018                state = asoc->state;
1019                subtype = SCTP_ST_CHUNK(chunk->chunk_hdr->type);
1020
1021                /* SCTP-AUTH, Section 6.3:
1022                 *    The receiver has a list of chunk types which it expects
1023                 *    to be received only after an AUTH-chunk.  This list has
1024                 *    been sent to the peer during the association setup.  It
1025                 *    MUST silently discard these chunks if they are not placed
1026                 *    after an AUTH chunk in the packet.
1027                 */
1028                if (sctp_auth_recv_cid(subtype.chunk, asoc) && !chunk->auth)
1029                        continue;
1030
1031                /* Remember where the last DATA chunk came from so we
1032                 * know where to send the SACK.
1033                 */
1034                if (sctp_chunk_is_data(chunk))
1035                        asoc->peer.last_data_from = chunk->transport;
1036                else {
1037                        SCTP_INC_STATS(net, SCTP_MIB_INCTRLCHUNKS);
1038                        asoc->stats.ictrlchunks++;
1039                        if (chunk->chunk_hdr->type == SCTP_CID_SACK)
1040                                asoc->stats.isacks++;
1041                }
1042
1043                if (chunk->transport)
1044                        chunk->transport->last_time_heard = ktime_get();
1045
1046                /* Run through the state machine. */
1047                error = sctp_do_sm(net, SCTP_EVENT_T_CHUNK, subtype,
1048                                   state, ep, asoc, chunk, GFP_ATOMIC);
1049
1050                /* Check to see if the association is freed in response to
1051                 * the incoming chunk.  If so, get out of the while loop.
1052                 */
1053                if (asoc->base.dead)
1054                        break;
1055
1056                /* If there is an error on chunk, discard this packet. */
1057                if (error && chunk)
1058                        chunk->pdiscard = 1;
1059        }
1060        sctp_association_put(asoc);
1061}
1062
1063/* This routine moves an association from its old sk to a new sk.  */
1064void sctp_assoc_migrate(struct sctp_association *assoc, struct sock *newsk)
1065{
1066        struct sctp_sock *newsp = sctp_sk(newsk);
1067        struct sock *oldsk = assoc->base.sk;
1068
1069        /* Delete the association from the old endpoint's list of
1070         * associations.
1071         */
1072        list_del_init(&assoc->asocs);
1073
1074        /* Decrement the backlog value for a TCP-style socket. */
1075        if (sctp_style(oldsk, TCP))
1076                oldsk->sk_ack_backlog--;
1077
1078        /* Release references to the old endpoint and the sock.  */
1079        sctp_endpoint_put(assoc->ep);
1080        sock_put(assoc->base.sk);
1081
1082        /* Get a reference to the new endpoint.  */
1083        assoc->ep = newsp->ep;
1084        sctp_endpoint_hold(assoc->ep);
1085
1086        /* Get a reference to the new sock.  */
1087        assoc->base.sk = newsk;
1088        sock_hold(assoc->base.sk);
1089
1090        /* Add the association to the new endpoint's list of associations.  */
1091        sctp_endpoint_add_asoc(newsp->ep, assoc);
1092}
1093
1094/* Update an association (possibly from unexpected COOKIE-ECHO processing).  */
1095void sctp_assoc_update(struct sctp_association *asoc,
1096                       struct sctp_association *new)
1097{
1098        struct sctp_transport *trans;
1099        struct list_head *pos, *temp;
1100
1101        /* Copy in new parameters of peer. */
1102        asoc->c = new->c;
1103        asoc->peer.rwnd = new->peer.rwnd;
1104        asoc->peer.sack_needed = new->peer.sack_needed;
1105        asoc->peer.auth_capable = new->peer.auth_capable;
1106        asoc->peer.i = new->peer.i;
1107        sctp_tsnmap_init(&asoc->peer.tsn_map, SCTP_TSN_MAP_INITIAL,
1108                         asoc->peer.i.initial_tsn, GFP_ATOMIC);
1109
1110        /* Remove any peer addresses not present in the new association. */
1111        list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
1112                trans = list_entry(pos, struct sctp_transport, transports);
1113                if (!sctp_assoc_lookup_paddr(new, &trans->ipaddr)) {
1114                        sctp_assoc_rm_peer(asoc, trans);
1115                        continue;
1116                }
1117
1118                if (asoc->state >= SCTP_STATE_ESTABLISHED)
1119                        sctp_transport_reset(trans);
1120        }
1121
1122        /* If the case is A (association restart), use
1123         * initial_tsn as next_tsn. If the case is B, use
1124         * current next_tsn in case data sent to peer
1125         * has been discarded and needs retransmission.
1126         */
1127        if (asoc->state >= SCTP_STATE_ESTABLISHED) {
1128                asoc->next_tsn = new->next_tsn;
1129                asoc->ctsn_ack_point = new->ctsn_ack_point;
1130                asoc->adv_peer_ack_point = new->adv_peer_ack_point;
1131
1132                /* Reinitialize SSN for both local streams
1133                 * and peer's streams.
1134                 */
1135                sctp_ssnmap_clear(asoc->ssnmap);
1136
1137                /* Flush the ULP reassembly and ordered queue.
1138                 * Any data there will now be stale and will
1139                 * cause problems.
1140                 */
1141                sctp_ulpq_flush(&asoc->ulpq);
1142
1143                /* reset the overall association error count so
1144                 * that the restarted association doesn't get torn
1145                 * down on the next retransmission timer.
1146                 */
1147                asoc->overall_error_count = 0;
1148
1149        } else {
1150                /* Add any peer addresses from the new association. */
1151                list_for_each_entry(trans, &new->peer.transport_addr_list,
1152                                transports) {
1153                        if (!sctp_assoc_lookup_paddr(asoc, &trans->ipaddr))
1154                                sctp_assoc_add_peer(asoc, &trans->ipaddr,
1155                                                    GFP_ATOMIC, trans->state);
1156                }
1157
1158                asoc->ctsn_ack_point = asoc->next_tsn - 1;
1159                asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
1160                if (!asoc->ssnmap) {
1161                        /* Move the ssnmap. */
1162                        asoc->ssnmap = new->ssnmap;
1163                        new->ssnmap = NULL;
1164                }
1165
1166                if (!asoc->assoc_id) {
1167                        /* get a new association id since we don't have one
1168                         * yet.
1169                         */
1170                        sctp_assoc_set_id(asoc, GFP_ATOMIC);
1171                }
1172        }
1173
1174        /* SCTP-AUTH: Save the peer parameters from the new associations
1175         * and also move the association shared keys over
1176         */
1177        kfree(asoc->peer.peer_random);
1178        asoc->peer.peer_random = new->peer.peer_random;
1179        new->peer.peer_random = NULL;
1180
1181        kfree(asoc->peer.peer_chunks);
1182        asoc->peer.peer_chunks = new->peer.peer_chunks;
1183        new->peer.peer_chunks = NULL;
1184
1185        kfree(asoc->peer.peer_hmacs);
1186        asoc->peer.peer_hmacs = new->peer.peer_hmacs;
1187        new->peer.peer_hmacs = NULL;
1188
1189        sctp_auth_asoc_init_active_key(asoc, GFP_ATOMIC);
1190}
1191
1192/* Update the retran path for sending a retransmitted packet.
1193 * See also RFC4960, 6.4. Multi-Homed SCTP Endpoints:
1194 *
1195 *   When there is outbound data to send and the primary path
1196 *   becomes inactive (e.g., due to failures), or where the
1197 *   SCTP user explicitly requests to send data to an
1198 *   inactive destination transport address, before reporting
1199 *   an error to its ULP, the SCTP endpoint should try to send
1200 *   the data to an alternate active destination transport
1201 *   address if one exists.
1202 *
1203 *   When retransmitting data that timed out, if the endpoint
1204 *   is multihomed, it should consider each source-destination
1205 *   address pair in its retransmission selection policy.
1206 *   When retransmitting timed-out data, the endpoint should
1207 *   attempt to pick the most divergent source-destination
1208 *   pair from the original source-destination pair to which
1209 *   the packet was transmitted.
1210 *
1211 *   Note: Rules for picking the most divergent source-destination
1212 *   pair are an implementation decision and are not specified
1213 *   within this document.
1214 *
1215 * Our basic strategy is to round-robin transports in priorities
1216 * according to sctp_trans_score() e.g., if no such
1217 * transport with state SCTP_ACTIVE exists, round-robin through
1218 * SCTP_UNKNOWN, etc. You get the picture.
1219 */
1220static u8 sctp_trans_score(const struct sctp_transport *trans)
1221{
1222        switch (trans->state) {
1223        case SCTP_ACTIVE:
1224                return 3;       /* best case */
1225        case SCTP_UNKNOWN:
1226                return 2;
1227        case SCTP_PF:
1228                return 1;
1229        default: /* case SCTP_INACTIVE */
1230                return 0;       /* worst case */
1231        }
1232}
1233
1234static struct sctp_transport *sctp_trans_elect_tie(struct sctp_transport *trans1,
1235                                                   struct sctp_transport *trans2)
1236{
1237        if (trans1->error_count > trans2->error_count) {
1238                return trans2;
1239        } else if (trans1->error_count == trans2->error_count &&
1240                   ktime_after(trans2->last_time_heard,
1241                               trans1->last_time_heard)) {
1242                return trans2;
1243        } else {
1244                return trans1;
1245        }
1246}
1247
1248static struct sctp_transport *sctp_trans_elect_best(struct sctp_transport *curr,
1249                                                    struct sctp_transport *best)
1250{
1251        u8 score_curr, score_best;
1252
1253        if (best == NULL || curr == best)
1254                return curr;
1255
1256        score_curr = sctp_trans_score(curr);
1257        score_best = sctp_trans_score(best);
1258
1259        /* First, try a score-based selection if both transport states
1260         * differ. If we're in a tie, lets try to make a more clever
1261         * decision here based on error counts and last time heard.
1262         */
1263        if (score_curr > score_best)
1264                return curr;
1265        else if (score_curr == score_best)
1266                return sctp_trans_elect_tie(best, curr);
1267        else
1268                return best;
1269}
1270
1271void sctp_assoc_update_retran_path(struct sctp_association *asoc)
1272{
1273        struct sctp_transport *trans = asoc->peer.retran_path;
1274        struct sctp_transport *trans_next = NULL;
1275
1276        /* We're done as we only have the one and only path. */
1277        if (asoc->peer.transport_count == 1)
1278                return;
1279        /* If active_path and retran_path are the same and active,
1280         * then this is the only active path. Use it.
1281         */
1282        if (asoc->peer.active_path == asoc->peer.retran_path &&
1283            asoc->peer.active_path->state == SCTP_ACTIVE)
1284                return;
1285
1286        /* Iterate from retran_path's successor back to retran_path. */
1287        for (trans = list_next_entry(trans, transports); 1;
1288             trans = list_next_entry(trans, transports)) {
1289                /* Manually skip the head element. */
1290                if (&trans->transports == &asoc->peer.transport_addr_list)
1291                        continue;
1292                if (trans->state == SCTP_UNCONFIRMED)
1293                        continue;
1294                trans_next = sctp_trans_elect_best(trans, trans_next);
1295                /* Active is good enough for immediate return. */
1296                if (trans_next->state == SCTP_ACTIVE)
1297                        break;
1298                /* We've reached the end, time to update path. */
1299                if (trans == asoc->peer.retran_path)
1300                        break;
1301        }
1302
1303        asoc->peer.retran_path = trans_next;
1304
1305        pr_debug("%s: association:%p updated new path to addr:%pISpc\n",
1306                 __func__, asoc, &asoc->peer.retran_path->ipaddr.sa);
1307}
1308
1309static void sctp_select_active_and_retran_path(struct sctp_association *asoc)
1310{
1311        struct sctp_transport *trans, *trans_pri = NULL, *trans_sec = NULL;
1312        struct sctp_transport *trans_pf = NULL;
1313
1314        /* Look for the two most recently used active transports. */
1315        list_for_each_entry(trans, &asoc->peer.transport_addr_list,
1316                            transports) {
1317                /* Skip uninteresting transports. */
1318                if (trans->state == SCTP_INACTIVE ||
1319                    trans->state == SCTP_UNCONFIRMED)
1320                        continue;
1321                /* Keep track of the best PF transport from our
1322                 * list in case we don't find an active one.
1323                 */
1324                if (trans->state == SCTP_PF) {
1325                        trans_pf = sctp_trans_elect_best(trans, trans_pf);
1326                        continue;
1327                }
1328                /* For active transports, pick the most recent ones. */
1329                if (trans_pri == NULL ||
1330                    ktime_after(trans->last_time_heard,
1331                                trans_pri->last_time_heard)) {
1332                        trans_sec = trans_pri;
1333                        trans_pri = trans;
1334                } else if (trans_sec == NULL ||
1335                           ktime_after(trans->last_time_heard,
1336                                       trans_sec->last_time_heard)) {
1337                        trans_sec = trans;
1338                }
1339        }
1340
1341        /* RFC 2960 6.4 Multi-Homed SCTP Endpoints
1342         *
1343         * By default, an endpoint should always transmit to the primary
1344         * path, unless the SCTP user explicitly specifies the
1345         * destination transport address (and possibly source transport
1346         * address) to use. [If the primary is active but not most recent,
1347         * bump the most recently used transport.]
1348         */
1349        if ((asoc->peer.primary_path->state == SCTP_ACTIVE ||
1350             asoc->peer.primary_path->state == SCTP_UNKNOWN) &&
1351             asoc->peer.primary_path != trans_pri) {
1352                trans_sec = trans_pri;
1353                trans_pri = asoc->peer.primary_path;
1354        }
1355
1356        /* We did not find anything useful for a possible retransmission
1357         * path; either primary path that we found is the the same as
1358         * the current one, or we didn't generally find an active one.
1359         */
1360        if (trans_sec == NULL)
1361                trans_sec = trans_pri;
1362
1363        /* If we failed to find a usable transport, just camp on the
1364         * active or pick a PF iff it's the better choice.
1365         */
1366        if (trans_pri == NULL) {
1367                trans_pri = sctp_trans_elect_best(asoc->peer.active_path, trans_pf);
1368                trans_sec = trans_pri;
1369        }
1370
1371        /* Set the active and retran transports. */
1372        asoc->peer.active_path = trans_pri;
1373        asoc->peer.retran_path = trans_sec;
1374}
1375
1376struct sctp_transport *
1377sctp_assoc_choose_alter_transport(struct sctp_association *asoc,
1378                                  struct sctp_transport *last_sent_to)
1379{
1380        /* If this is the first time packet is sent, use the active path,
1381         * else use the retran path. If the last packet was sent over the
1382         * retran path, update the retran path and use it.
1383         */
1384        if (last_sent_to == NULL) {
1385                return asoc->peer.active_path;
1386        } else {
1387                if (last_sent_to == asoc->peer.retran_path)
1388                        sctp_assoc_update_retran_path(asoc);
1389
1390                return asoc->peer.retran_path;
1391        }
1392}
1393
1394/* Update the association's pmtu and frag_point by going through all the
1395 * transports. This routine is called when a transport's PMTU has changed.
1396 */
1397void sctp_assoc_sync_pmtu(struct sock *sk, struct sctp_association *asoc)
1398{
1399        struct sctp_transport *t;
1400        __u32 pmtu = 0;
1401
1402        if (!asoc)
1403                return;
1404
1405        /* Get the lowest pmtu of all the transports. */
1406        list_for_each_entry(t, &asoc->peer.transport_addr_list,
1407                                transports) {
1408                if (t->pmtu_pending && t->dst) {
1409                        sctp_transport_update_pmtu(sk, t,
1410                                                   WORD_TRUNC(dst_mtu(t->dst)));
1411                        t->pmtu_pending = 0;
1412                }
1413                if (!pmtu || (t->pathmtu < pmtu))
1414                        pmtu = t->pathmtu;
1415        }
1416
1417        if (pmtu) {
1418                asoc->pathmtu = pmtu;
1419                asoc->frag_point = sctp_frag_point(asoc, pmtu);
1420        }
1421
1422        pr_debug("%s: asoc:%p, pmtu:%d, frag_point:%d\n", __func__, asoc,
1423                 asoc->pathmtu, asoc->frag_point);
1424}
1425
1426/* Should we send a SACK to update our peer? */
1427static inline bool sctp_peer_needs_update(struct sctp_association *asoc)
1428{
1429        struct net *net = sock_net(asoc->base.sk);
1430        switch (asoc->state) {
1431        case SCTP_STATE_ESTABLISHED:
1432        case SCTP_STATE_SHUTDOWN_PENDING:
1433        case SCTP_STATE_SHUTDOWN_RECEIVED:
1434        case SCTP_STATE_SHUTDOWN_SENT:
1435                if ((asoc->rwnd > asoc->a_rwnd) &&
1436                    ((asoc->rwnd - asoc->a_rwnd) >= max_t(__u32,
1437                           (asoc->base.sk->sk_rcvbuf >> net->sctp.rwnd_upd_shift),
1438                           asoc->pathmtu)))
1439                        return true;
1440                break;
1441        default:
1442                break;
1443        }
1444        return false;
1445}
1446
1447/* Increase asoc's rwnd by len and send any window update SACK if needed. */
1448void sctp_assoc_rwnd_increase(struct sctp_association *asoc, unsigned int len)
1449{
1450        struct sctp_chunk *sack;
1451        struct timer_list *timer;
1452
1453        if (asoc->rwnd_over) {
1454                if (asoc->rwnd_over >= len) {
1455                        asoc->rwnd_over -= len;
1456                } else {
1457                        asoc->rwnd += (len - asoc->rwnd_over);
1458                        asoc->rwnd_over = 0;
1459                }
1460        } else {
1461                asoc->rwnd += len;
1462        }
1463
1464        /* If we had window pressure, start recovering it
1465         * once our rwnd had reached the accumulated pressure
1466         * threshold.  The idea is to recover slowly, but up
1467         * to the initial advertised window.
1468         */
1469        if (asoc->rwnd_press && asoc->rwnd >= asoc->rwnd_press) {
1470                int change = min(asoc->pathmtu, asoc->rwnd_press);
1471                asoc->rwnd += change;
1472                asoc->rwnd_press -= change;
1473        }
1474
1475        pr_debug("%s: asoc:%p rwnd increased by %d to (%u, %u) - %u\n",
1476                 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
1477                 asoc->a_rwnd);
1478
1479        /* Send a window update SACK if the rwnd has increased by at least the
1480         * minimum of the association's PMTU and half of the receive buffer.
1481         * The algorithm used is similar to the one described in
1482         * Section 4.2.3.3 of RFC 1122.
1483         */
1484        if (sctp_peer_needs_update(asoc)) {
1485                asoc->a_rwnd = asoc->rwnd;
1486
1487                pr_debug("%s: sending window update SACK- asoc:%p rwnd:%u "
1488                         "a_rwnd:%u\n", __func__, asoc, asoc->rwnd,
1489                         asoc->a_rwnd);
1490
1491                sack = sctp_make_sack(asoc);
1492                if (!sack)
1493                        return;
1494
1495                asoc->peer.sack_needed = 0;
1496
1497                sctp_outq_tail(&asoc->outqueue, sack, GFP_ATOMIC);
1498
1499                /* Stop the SACK timer.  */
1500                timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
1501                if (del_timer(timer))
1502                        sctp_association_put(asoc);
1503        }
1504}
1505
1506/* Decrease asoc's rwnd by len. */
1507void sctp_assoc_rwnd_decrease(struct sctp_association *asoc, unsigned int len)
1508{
1509        int rx_count;
1510        int over = 0;
1511
1512        if (unlikely(!asoc->rwnd || asoc->rwnd_over))
1513                pr_debug("%s: association:%p has asoc->rwnd:%u, "
1514                         "asoc->rwnd_over:%u!\n", __func__, asoc,
1515                         asoc->rwnd, asoc->rwnd_over);
1516
1517        if (asoc->ep->rcvbuf_policy)
1518                rx_count = atomic_read(&asoc->rmem_alloc);
1519        else
1520                rx_count = atomic_read(&asoc->base.sk->sk_rmem_alloc);
1521
1522        /* If we've reached or overflowed our receive buffer, announce
1523         * a 0 rwnd if rwnd would still be positive.  Store the
1524         * the potential pressure overflow so that the window can be restored
1525         * back to original value.
1526         */
1527        if (rx_count >= asoc->base.sk->sk_rcvbuf)
1528                over = 1;
1529
1530        if (asoc->rwnd >= len) {
1531                asoc->rwnd -= len;
1532                if (over) {
1533                        asoc->rwnd_press += asoc->rwnd;
1534                        asoc->rwnd = 0;
1535                }
1536        } else {
1537                asoc->rwnd_over = len - asoc->rwnd;
1538                asoc->rwnd = 0;
1539        }
1540
1541        pr_debug("%s: asoc:%p rwnd decreased by %d to (%u, %u, %u)\n",
1542                 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
1543                 asoc->rwnd_press);
1544}
1545
1546/* Build the bind address list for the association based on info from the
1547 * local endpoint and the remote peer.
1548 */
1549int sctp_assoc_set_bind_addr_from_ep(struct sctp_association *asoc,
1550                                     sctp_scope_t scope, gfp_t gfp)
1551{
1552        int flags;
1553
1554        /* Use scoping rules to determine the subset of addresses from
1555         * the endpoint.
1556         */
1557        flags = (PF_INET6 == asoc->base.sk->sk_family) ? SCTP_ADDR6_ALLOWED : 0;
1558        if (asoc->peer.ipv4_address)
1559                flags |= SCTP_ADDR4_PEERSUPP;
1560        if (asoc->peer.ipv6_address)
1561                flags |= SCTP_ADDR6_PEERSUPP;
1562
1563        return sctp_bind_addr_copy(sock_net(asoc->base.sk),
1564                                   &asoc->base.bind_addr,
1565                                   &asoc->ep->base.bind_addr,
1566                                   scope, gfp, flags);
1567}
1568
1569/* Build the association's bind address list from the cookie.  */
1570int sctp_assoc_set_bind_addr_from_cookie(struct sctp_association *asoc,
1571                                         struct sctp_cookie *cookie,
1572                                         gfp_t gfp)
1573{
1574        int var_size2 = ntohs(cookie->peer_init->chunk_hdr.length);
1575        int var_size3 = cookie->raw_addr_list_len;
1576        __u8 *raw = (__u8 *)cookie->peer_init + var_size2;
1577
1578        return sctp_raw_to_bind_addrs(&asoc->base.bind_addr, raw, var_size3,
1579                                      asoc->ep->base.bind_addr.port, gfp);
1580}
1581
1582/* Lookup laddr in the bind address list of an association. */
1583int sctp_assoc_lookup_laddr(struct sctp_association *asoc,
1584                            const union sctp_addr *laddr)
1585{
1586        int found = 0;
1587
1588        if ((asoc->base.bind_addr.port == ntohs(laddr->v4.sin_port)) &&
1589            sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
1590                                 sctp_sk(asoc->base.sk)))
1591                found = 1;
1592
1593        return found;
1594}
1595
1596/* Set an association id for a given association */
1597int sctp_assoc_set_id(struct sctp_association *asoc, gfp_t gfp)
1598{
1599        bool preload = gfpflags_allow_blocking(gfp);
1600        int ret;
1601
1602        /* If the id is already assigned, keep it. */
1603        if (asoc->assoc_id)
1604                return 0;
1605
1606        if (preload)
1607                idr_preload(gfp);
1608        spin_lock_bh(&sctp_assocs_id_lock);
1609        /* 0 is not a valid assoc_id, must be >= 1 */
1610        ret = idr_alloc_cyclic(&sctp_assocs_id, asoc, 1, 0, GFP_NOWAIT);
1611        spin_unlock_bh(&sctp_assocs_id_lock);
1612        if (preload)
1613                idr_preload_end();
1614        if (ret < 0)
1615                return ret;
1616
1617        asoc->assoc_id = (sctp_assoc_t)ret;
1618        return 0;
1619}
1620
1621/* Free the ASCONF queue */
1622static void sctp_assoc_free_asconf_queue(struct sctp_association *asoc)
1623{
1624        struct sctp_chunk *asconf;
1625        struct sctp_chunk *tmp;
1626
1627        list_for_each_entry_safe(asconf, tmp, &asoc->addip_chunk_list, list) {
1628                list_del_init(&asconf->list);
1629                sctp_chunk_free(asconf);
1630        }
1631}
1632
1633/* Free asconf_ack cache */
1634static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc)
1635{
1636        struct sctp_chunk *ack;
1637        struct sctp_chunk *tmp;
1638
1639        list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
1640                                transmitted_list) {
1641                list_del_init(&ack->transmitted_list);
1642                sctp_chunk_free(ack);
1643        }
1644}
1645
1646/* Clean up the ASCONF_ACK queue */
1647void sctp_assoc_clean_asconf_ack_cache(const struct sctp_association *asoc)
1648{
1649        struct sctp_chunk *ack;
1650        struct sctp_chunk *tmp;
1651
1652        /* We can remove all the entries from the queue up to
1653         * the "Peer-Sequence-Number".
1654         */
1655        list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
1656                                transmitted_list) {
1657                if (ack->subh.addip_hdr->serial ==
1658                                htonl(asoc->peer.addip_serial))
1659                        break;
1660
1661                list_del_init(&ack->transmitted_list);
1662                sctp_chunk_free(ack);
1663        }
1664}
1665
1666/* Find the ASCONF_ACK whose serial number matches ASCONF */
1667struct sctp_chunk *sctp_assoc_lookup_asconf_ack(
1668                                        const struct sctp_association *asoc,
1669                                        __be32 serial)
1670{
1671        struct sctp_chunk *ack;
1672
1673        /* Walk through the list of cached ASCONF-ACKs and find the
1674         * ack chunk whose serial number matches that of the request.
1675         */
1676        list_for_each_entry(ack, &asoc->asconf_ack_list, transmitted_list) {
1677                if (sctp_chunk_pending(ack))
1678                        continue;
1679                if (ack->subh.addip_hdr->serial == serial) {
1680                        sctp_chunk_hold(ack);
1681                        return ack;
1682                }
1683        }
1684
1685        return NULL;
1686}
1687
1688void sctp_asconf_queue_teardown(struct sctp_association *asoc)
1689{
1690        /* Free any cached ASCONF_ACK chunk. */
1691        sctp_assoc_free_asconf_acks(asoc);
1692
1693        /* Free the ASCONF queue. */
1694        sctp_assoc_free_asconf_queue(asoc);
1695
1696        /* Free any cached ASCONF chunk. */
1697        if (asoc->addip_last_asconf)
1698                sctp_chunk_free(asoc->addip_last_asconf);
1699}
1700