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        asoc->prsctp_enable = ep->prsctp_enable;
 272
 273        /* Save the hmacs and chunks list into this association */
 274        if (ep->auth_hmacs_list)
 275                memcpy(asoc->c.auth_hmacs, ep->auth_hmacs_list,
 276                        ntohs(ep->auth_hmacs_list->param_hdr.length));
 277        if (ep->auth_chunk_list)
 278                memcpy(asoc->c.auth_chunks, ep->auth_chunk_list,
 279                        ntohs(ep->auth_chunk_list->param_hdr.length));
 280
 281        /* Get the AUTH random number for this association */
 282        p = (sctp_paramhdr_t *)asoc->c.auth_random;
 283        p->type = SCTP_PARAM_RANDOM;
 284        p->length = htons(sizeof(sctp_paramhdr_t) + SCTP_AUTH_RANDOM_LENGTH);
 285        get_random_bytes(p+1, SCTP_AUTH_RANDOM_LENGTH);
 286
 287        return asoc;
 288
 289fail_init:
 290        sock_put(asoc->base.sk);
 291        sctp_endpoint_put(asoc->ep);
 292        return NULL;
 293}
 294
 295/* Allocate and initialize a new association */
 296struct sctp_association *sctp_association_new(const struct sctp_endpoint *ep,
 297                                         const struct sock *sk,
 298                                         sctp_scope_t scope,
 299                                         gfp_t gfp)
 300{
 301        struct sctp_association *asoc;
 302
 303        asoc = kzalloc(sizeof(*asoc), gfp);
 304        if (!asoc)
 305                goto fail;
 306
 307        if (!sctp_association_init(asoc, ep, sk, scope, gfp))
 308                goto fail_init;
 309
 310        SCTP_DBG_OBJCNT_INC(assoc);
 311
 312        pr_debug("Created asoc %p\n", asoc);
 313
 314        return asoc;
 315
 316fail_init:
 317        kfree(asoc);
 318fail:
 319        return NULL;
 320}
 321
 322/* Free this association if possible.  There may still be users, so
 323 * the actual deallocation may be delayed.
 324 */
 325void sctp_association_free(struct sctp_association *asoc)
 326{
 327        struct sock *sk = asoc->base.sk;
 328        struct sctp_transport *transport;
 329        struct list_head *pos, *temp;
 330        int i;
 331
 332        /* Only real associations count against the endpoint, so
 333         * don't bother for if this is a temporary association.
 334         */
 335        if (!list_empty(&asoc->asocs)) {
 336                list_del(&asoc->asocs);
 337
 338                /* Decrement the backlog value for a TCP-style listening
 339                 * socket.
 340                 */
 341                if (sctp_style(sk, TCP) && sctp_sstate(sk, LISTENING))
 342                        sk->sk_ack_backlog--;
 343        }
 344
 345        /* Mark as dead, so other users can know this structure is
 346         * going away.
 347         */
 348        asoc->base.dead = true;
 349
 350        /* Dispose of any data lying around in the outqueue. */
 351        sctp_outq_free(&asoc->outqueue);
 352
 353        /* Dispose of any pending messages for the upper layer. */
 354        sctp_ulpq_free(&asoc->ulpq);
 355
 356        /* Dispose of any pending chunks on the inqueue. */
 357        sctp_inq_free(&asoc->base.inqueue);
 358
 359        sctp_tsnmap_free(&asoc->peer.tsn_map);
 360
 361        /* Free ssnmap storage. */
 362        sctp_ssnmap_free(asoc->ssnmap);
 363
 364        /* Clean up the bound address list. */
 365        sctp_bind_addr_free(&asoc->base.bind_addr);
 366
 367        /* Do we need to go through all of our timers and
 368         * delete them?   To be safe we will try to delete all, but we
 369         * should be able to go through and make a guess based
 370         * on our state.
 371         */
 372        for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i) {
 373                if (del_timer(&asoc->timers[i]))
 374                        sctp_association_put(asoc);
 375        }
 376
 377        /* Free peer's cached cookie. */
 378        kfree(asoc->peer.cookie);
 379        kfree(asoc->peer.peer_random);
 380        kfree(asoc->peer.peer_chunks);
 381        kfree(asoc->peer.peer_hmacs);
 382
 383        /* Release the transport structures. */
 384        list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
 385                transport = list_entry(pos, struct sctp_transport, transports);
 386                list_del_rcu(pos);
 387                sctp_unhash_transport(transport);
 388                sctp_transport_free(transport);
 389        }
 390
 391        asoc->peer.transport_count = 0;
 392
 393        sctp_asconf_queue_teardown(asoc);
 394
 395        /* Free pending address space being deleted */
 396        kfree(asoc->asconf_addr_del_pending);
 397
 398        /* AUTH - Free the endpoint shared keys */
 399        sctp_auth_destroy_keys(&asoc->endpoint_shared_keys);
 400
 401        /* AUTH - Free the association shared key */
 402        sctp_auth_key_put(asoc->asoc_shared_key);
 403
 404        sctp_association_put(asoc);
 405}
 406
 407/* Cleanup and free up an association. */
 408static void sctp_association_destroy(struct sctp_association *asoc)
 409{
 410        if (unlikely(!asoc->base.dead)) {
 411                WARN(1, "Attempt to destroy undead association %p!\n", asoc);
 412                return;
 413        }
 414
 415        sctp_endpoint_put(asoc->ep);
 416        sock_put(asoc->base.sk);
 417
 418        if (asoc->assoc_id != 0) {
 419                spin_lock_bh(&sctp_assocs_id_lock);
 420                idr_remove(&sctp_assocs_id, asoc->assoc_id);
 421                spin_unlock_bh(&sctp_assocs_id_lock);
 422        }
 423
 424        WARN_ON(atomic_read(&asoc->rmem_alloc));
 425
 426        kfree(asoc);
 427        SCTP_DBG_OBJCNT_DEC(assoc);
 428}
 429
 430/* Change the primary destination address for the peer. */
 431void sctp_assoc_set_primary(struct sctp_association *asoc,
 432                            struct sctp_transport *transport)
 433{
 434        int changeover = 0;
 435
 436        /* it's a changeover only if we already have a primary path
 437         * that we are changing
 438         */
 439        if (asoc->peer.primary_path != NULL &&
 440            asoc->peer.primary_path != transport)
 441                changeover = 1 ;
 442
 443        asoc->peer.primary_path = transport;
 444
 445        /* Set a default msg_name for events. */
 446        memcpy(&asoc->peer.primary_addr, &transport->ipaddr,
 447               sizeof(union sctp_addr));
 448
 449        /* If the primary path is changing, assume that the
 450         * user wants to use this new path.
 451         */
 452        if ((transport->state == SCTP_ACTIVE) ||
 453            (transport->state == SCTP_UNKNOWN))
 454                asoc->peer.active_path = transport;
 455
 456        /*
 457         * SFR-CACC algorithm:
 458         * Upon the receipt of a request to change the primary
 459         * destination address, on the data structure for the new
 460         * primary destination, the sender MUST do the following:
 461         *
 462         * 1) If CHANGEOVER_ACTIVE is set, then there was a switch
 463         * to this destination address earlier. The sender MUST set
 464         * CYCLING_CHANGEOVER to indicate that this switch is a
 465         * double switch to the same destination address.
 466         *
 467         * Really, only bother is we have data queued or outstanding on
 468         * the association.
 469         */
 470        if (!asoc->outqueue.outstanding_bytes && !asoc->outqueue.out_qlen)
 471                return;
 472
 473        if (transport->cacc.changeover_active)
 474                transport->cacc.cycling_changeover = changeover;
 475
 476        /* 2) The sender MUST set CHANGEOVER_ACTIVE to indicate that
 477         * a changeover has occurred.
 478         */
 479        transport->cacc.changeover_active = changeover;
 480
 481        /* 3) The sender MUST store the next TSN to be sent in
 482         * next_tsn_at_change.
 483         */
 484        transport->cacc.next_tsn_at_change = asoc->next_tsn;
 485}
 486
 487/* Remove a transport from an association.  */
 488void sctp_assoc_rm_peer(struct sctp_association *asoc,
 489                        struct sctp_transport *peer)
 490{
 491        struct list_head        *pos;
 492        struct sctp_transport   *transport;
 493
 494        pr_debug("%s: association:%p addr:%pISpc\n",
 495                 __func__, asoc, &peer->ipaddr.sa);
 496
 497        /* If we are to remove the current retran_path, update it
 498         * to the next peer before removing this peer from the list.
 499         */
 500        if (asoc->peer.retran_path == peer)
 501                sctp_assoc_update_retran_path(asoc);
 502
 503        /* Remove this peer from the list. */
 504        list_del_rcu(&peer->transports);
 505        /* Remove this peer from the transport hashtable */
 506        sctp_unhash_transport(peer);
 507
 508        /* Get the first transport of asoc. */
 509        pos = asoc->peer.transport_addr_list.next;
 510        transport = list_entry(pos, struct sctp_transport, transports);
 511
 512        /* Update any entries that match the peer to be deleted. */
 513        if (asoc->peer.primary_path == peer)
 514                sctp_assoc_set_primary(asoc, transport);
 515        if (asoc->peer.active_path == peer)
 516                asoc->peer.active_path = transport;
 517        if (asoc->peer.retran_path == peer)
 518                asoc->peer.retran_path = transport;
 519        if (asoc->peer.last_data_from == peer)
 520                asoc->peer.last_data_from = transport;
 521
 522        /* If we remove the transport an INIT was last sent to, set it to
 523         * NULL. Combined with the update of the retran path above, this
 524         * will cause the next INIT to be sent to the next available
 525         * transport, maintaining the cycle.
 526         */
 527        if (asoc->init_last_sent_to == peer)
 528                asoc->init_last_sent_to = NULL;
 529
 530        /* If we remove the transport an SHUTDOWN was last sent to, set it
 531         * to NULL. Combined with the update of the retran path above, this
 532         * will cause the next SHUTDOWN to be sent to the next available
 533         * transport, maintaining the cycle.
 534         */
 535        if (asoc->shutdown_last_sent_to == peer)
 536                asoc->shutdown_last_sent_to = NULL;
 537
 538        /* If we remove the transport an ASCONF was last sent to, set it to
 539         * NULL.
 540         */
 541        if (asoc->addip_last_asconf &&
 542            asoc->addip_last_asconf->transport == peer)
 543                asoc->addip_last_asconf->transport = NULL;
 544
 545        /* If we have something on the transmitted list, we have to
 546         * save it off.  The best place is the active path.
 547         */
 548        if (!list_empty(&peer->transmitted)) {
 549                struct sctp_transport *active = asoc->peer.active_path;
 550                struct sctp_chunk *ch;
 551
 552                /* Reset the transport of each chunk on this list */
 553                list_for_each_entry(ch, &peer->transmitted,
 554                                        transmitted_list) {
 555                        ch->transport = NULL;
 556                        ch->rtt_in_progress = 0;
 557                }
 558
 559                list_splice_tail_init(&peer->transmitted,
 560                                        &active->transmitted);
 561
 562                /* Start a T3 timer here in case it wasn't running so
 563                 * that these migrated packets have a chance to get
 564                 * retransmitted.
 565                 */
 566                if (!timer_pending(&active->T3_rtx_timer))
 567                        if (!mod_timer(&active->T3_rtx_timer,
 568                                        jiffies + active->rto))
 569                                sctp_transport_hold(active);
 570        }
 571
 572        asoc->peer.transport_count--;
 573
 574        sctp_transport_free(peer);
 575}
 576
 577/* Add a transport address to an association.  */
 578struct sctp_transport *sctp_assoc_add_peer(struct sctp_association *asoc,
 579                                           const union sctp_addr *addr,
 580                                           const gfp_t gfp,
 581                                           const int peer_state)
 582{
 583        struct net *net = sock_net(asoc->base.sk);
 584        struct sctp_transport *peer;
 585        struct sctp_sock *sp;
 586        unsigned short port;
 587
 588        sp = sctp_sk(asoc->base.sk);
 589
 590        /* AF_INET and AF_INET6 share common port field. */
 591        port = ntohs(addr->v4.sin_port);
 592
 593        pr_debug("%s: association:%p addr:%pISpc state:%d\n", __func__,
 594                 asoc, &addr->sa, peer_state);
 595
 596        /* Set the port if it has not been set yet.  */
 597        if (0 == asoc->peer.port)
 598                asoc->peer.port = port;
 599
 600        /* Check to see if this is a duplicate. */
 601        peer = sctp_assoc_lookup_paddr(asoc, addr);
 602        if (peer) {
 603                /* An UNKNOWN state is only set on transports added by
 604                 * user in sctp_connectx() call.  Such transports should be
 605                 * considered CONFIRMED per RFC 4960, Section 5.4.
 606                 */
 607                if (peer->state == SCTP_UNKNOWN) {
 608                        peer->state = SCTP_ACTIVE;
 609                }
 610                return peer;
 611        }
 612
 613        peer = sctp_transport_new(net, addr, gfp);
 614        if (!peer)
 615                return NULL;
 616
 617        sctp_transport_set_owner(peer, asoc);
 618
 619        /* Initialize the peer's heartbeat interval based on the
 620         * association configured value.
 621         */
 622        peer->hbinterval = asoc->hbinterval;
 623
 624        /* Set the path max_retrans.  */
 625        peer->pathmaxrxt = asoc->pathmaxrxt;
 626
 627        /* And the partial failure retrans threshold */
 628        peer->pf_retrans = asoc->pf_retrans;
 629
 630        /* Initialize the peer's SACK delay timeout based on the
 631         * association configured value.
 632         */
 633        peer->sackdelay = asoc->sackdelay;
 634        peer->sackfreq = asoc->sackfreq;
 635
 636        /* Enable/disable heartbeat, SACK delay, and path MTU discovery
 637         * based on association setting.
 638         */
 639        peer->param_flags = asoc->param_flags;
 640
 641        sctp_transport_route(peer, NULL, sp);
 642
 643        /* Initialize the pmtu of the transport. */
 644        if (peer->param_flags & SPP_PMTUD_DISABLE) {
 645                if (asoc->pathmtu)
 646                        peer->pathmtu = asoc->pathmtu;
 647                else
 648                        peer->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
 649        }
 650
 651        /* If this is the first transport addr on this association,
 652         * initialize the association PMTU to the peer's PMTU.
 653         * If not and the current association PMTU is higher than the new
 654         * peer's PMTU, reset the association PMTU to the new peer's PMTU.
 655         */
 656        if (asoc->pathmtu)
 657                asoc->pathmtu = min_t(int, peer->pathmtu, asoc->pathmtu);
 658        else
 659                asoc->pathmtu = peer->pathmtu;
 660
 661        pr_debug("%s: association:%p PMTU set to %d\n", __func__, asoc,
 662                 asoc->pathmtu);
 663
 664        peer->pmtu_pending = 0;
 665
 666        asoc->frag_point = sctp_frag_point(asoc, asoc->pathmtu);
 667
 668        /* The asoc->peer.port might not be meaningful yet, but
 669         * initialize the packet structure anyway.
 670         */
 671        sctp_packet_init(&peer->packet, peer, asoc->base.bind_addr.port,
 672                         asoc->peer.port);
 673
 674        /* 7.2.1 Slow-Start
 675         *
 676         * o The initial cwnd before DATA transmission or after a sufficiently
 677         *   long idle period MUST be set to
 678         *      min(4*MTU, max(2*MTU, 4380 bytes))
 679         *
 680         * o The initial value of ssthresh MAY be arbitrarily high
 681         *   (for example, implementations MAY use the size of the
 682         *   receiver advertised window).
 683         */
 684        peer->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
 685
 686        /* At this point, we may not have the receiver's advertised window,
 687         * so initialize ssthresh to the default value and it will be set
 688         * later when we process the INIT.
 689         */
 690        peer->ssthresh = SCTP_DEFAULT_MAXWINDOW;
 691
 692        peer->partial_bytes_acked = 0;
 693        peer->flight_size = 0;
 694        peer->burst_limited = 0;
 695
 696        /* Set the transport's RTO.initial value */
 697        peer->rto = asoc->rto_initial;
 698        sctp_max_rto(asoc, peer);
 699
 700        /* Set the peer's active state. */
 701        peer->state = peer_state;
 702
 703        /* Attach the remote transport to our asoc.  */
 704        list_add_tail_rcu(&peer->transports, &asoc->peer.transport_addr_list);
 705        asoc->peer.transport_count++;
 706        /* Add this peer into the transport hashtable */
 707        sctp_hash_transport(peer);
 708
 709        /* If we do not yet have a primary path, set one.  */
 710        if (!asoc->peer.primary_path) {
 711                sctp_assoc_set_primary(asoc, peer);
 712                asoc->peer.retran_path = peer;
 713        }
 714
 715        if (asoc->peer.active_path == asoc->peer.retran_path &&
 716            peer->state != SCTP_UNCONFIRMED) {
 717                asoc->peer.retran_path = peer;
 718        }
 719
 720        return peer;
 721}
 722
 723/* Delete a transport address from an association.  */
 724void sctp_assoc_del_peer(struct sctp_association *asoc,
 725                         const union sctp_addr *addr)
 726{
 727        struct list_head        *pos;
 728        struct list_head        *temp;
 729        struct sctp_transport   *transport;
 730
 731        list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
 732                transport = list_entry(pos, struct sctp_transport, transports);
 733                if (sctp_cmp_addr_exact(addr, &transport->ipaddr)) {
 734                        /* Do book keeping for removing the peer and free it. */
 735                        sctp_assoc_rm_peer(asoc, transport);
 736                        break;
 737                }
 738        }
 739}
 740
 741/* Lookup a transport by address. */
 742struct sctp_transport *sctp_assoc_lookup_paddr(
 743                                        const struct sctp_association *asoc,
 744                                        const union sctp_addr *address)
 745{
 746        struct sctp_transport *t;
 747
 748        /* Cycle through all transports searching for a peer address. */
 749
 750        list_for_each_entry(t, &asoc->peer.transport_addr_list,
 751                        transports) {
 752                if (sctp_cmp_addr_exact(address, &t->ipaddr))
 753                        return t;
 754        }
 755
 756        return NULL;
 757}
 758
 759/* Remove all transports except a give one */
 760void sctp_assoc_del_nonprimary_peers(struct sctp_association *asoc,
 761                                     struct sctp_transport *primary)
 762{
 763        struct sctp_transport   *temp;
 764        struct sctp_transport   *t;
 765
 766        list_for_each_entry_safe(t, temp, &asoc->peer.transport_addr_list,
 767                                 transports) {
 768                /* if the current transport is not the primary one, delete it */
 769                if (t != primary)
 770                        sctp_assoc_rm_peer(asoc, t);
 771        }
 772}
 773
 774/* Engage in transport control operations.
 775 * Mark the transport up or down and send a notification to the user.
 776 * Select and update the new active and retran paths.
 777 */
 778void sctp_assoc_control_transport(struct sctp_association *asoc,
 779                                  struct sctp_transport *transport,
 780                                  sctp_transport_cmd_t command,
 781                                  sctp_sn_error_t error)
 782{
 783        struct sctp_ulpevent *event;
 784        struct sockaddr_storage addr;
 785        int spc_state = 0;
 786        bool ulp_notify = true;
 787
 788        /* Record the transition on the transport.  */
 789        switch (command) {
 790        case SCTP_TRANSPORT_UP:
 791                /* If we are moving from UNCONFIRMED state due
 792                 * to heartbeat success, report the SCTP_ADDR_CONFIRMED
 793                 * state to the user, otherwise report SCTP_ADDR_AVAILABLE.
 794                 */
 795                if (SCTP_UNCONFIRMED == transport->state &&
 796                    SCTP_HEARTBEAT_SUCCESS == error)
 797                        spc_state = SCTP_ADDR_CONFIRMED;
 798                else
 799                        spc_state = SCTP_ADDR_AVAILABLE;
 800                /* Don't inform ULP about transition from PF to
 801                 * active state and set cwnd to 1 MTU, see SCTP
 802                 * Quick failover draft section 5.1, point 5
 803                 */
 804                if (transport->state == SCTP_PF) {
 805                        ulp_notify = false;
 806                        transport->cwnd = asoc->pathmtu;
 807                }
 808                transport->state = SCTP_ACTIVE;
 809                break;
 810
 811        case SCTP_TRANSPORT_DOWN:
 812                /* If the transport was never confirmed, do not transition it
 813                 * to inactive state.  Also, release the cached route since
 814                 * there may be a better route next time.
 815                 */
 816                if (transport->state != SCTP_UNCONFIRMED)
 817                        transport->state = SCTP_INACTIVE;
 818                else {
 819                        dst_release(transport->dst);
 820                        transport->dst = NULL;
 821                        ulp_notify = false;
 822                }
 823
 824                spc_state = SCTP_ADDR_UNREACHABLE;
 825                break;
 826
 827        case SCTP_TRANSPORT_PF:
 828                transport->state = SCTP_PF;
 829                ulp_notify = false;
 830                break;
 831
 832        default:
 833                return;
 834        }
 835
 836        /* Generate and send a SCTP_PEER_ADDR_CHANGE notification
 837         * to the user.
 838         */
 839        if (ulp_notify) {
 840                memset(&addr, 0, sizeof(struct sockaddr_storage));
 841                memcpy(&addr, &transport->ipaddr,
 842                       transport->af_specific->sockaddr_len);
 843
 844                event = sctp_ulpevent_make_peer_addr_change(asoc, &addr,
 845                                        0, spc_state, error, GFP_ATOMIC);
 846                if (event)
 847                        sctp_ulpq_tail_event(&asoc->ulpq, event);
 848        }
 849
 850        /* Select new active and retran paths. */
 851        sctp_select_active_and_retran_path(asoc);
 852}
 853
 854/* Hold a reference to an association. */
 855void sctp_association_hold(struct sctp_association *asoc)
 856{
 857        atomic_inc(&asoc->base.refcnt);
 858}
 859
 860/* Release a reference to an association and cleanup
 861 * if there are no more references.
 862 */
 863void sctp_association_put(struct sctp_association *asoc)
 864{
 865        if (atomic_dec_and_test(&asoc->base.refcnt))
 866                sctp_association_destroy(asoc);
 867}
 868
 869/* Allocate the next TSN, Transmission Sequence Number, for the given
 870 * association.
 871 */
 872__u32 sctp_association_get_next_tsn(struct sctp_association *asoc)
 873{
 874        /* From Section 1.6 Serial Number Arithmetic:
 875         * Transmission Sequence Numbers wrap around when they reach
 876         * 2**32 - 1.  That is, the next TSN a DATA chunk MUST use
 877         * after transmitting TSN = 2*32 - 1 is TSN = 0.
 878         */
 879        __u32 retval = asoc->next_tsn;
 880        asoc->next_tsn++;
 881        asoc->unack_data++;
 882
 883        return retval;
 884}
 885
 886/* Compare two addresses to see if they match.  Wildcard addresses
 887 * only match themselves.
 888 */
 889int sctp_cmp_addr_exact(const union sctp_addr *ss1,
 890                        const union sctp_addr *ss2)
 891{
 892        struct sctp_af *af;
 893
 894        af = sctp_get_af_specific(ss1->sa.sa_family);
 895        if (unlikely(!af))
 896                return 0;
 897
 898        return af->cmp_addr(ss1, ss2);
 899}
 900
 901/* Return an ecne chunk to get prepended to a packet.
 902 * Note:  We are sly and return a shared, prealloced chunk.  FIXME:
 903 * No we don't, but we could/should.
 904 */
 905struct sctp_chunk *sctp_get_ecne_prepend(struct sctp_association *asoc)
 906{
 907        if (!asoc->need_ecne)
 908                return NULL;
 909
 910        /* Send ECNE if needed.
 911         * Not being able to allocate a chunk here is not deadly.
 912         */
 913        return sctp_make_ecne(asoc, asoc->last_ecne_tsn);
 914}
 915
 916/*
 917 * Find which transport this TSN was sent on.
 918 */
 919struct sctp_transport *sctp_assoc_lookup_tsn(struct sctp_association *asoc,
 920                                             __u32 tsn)
 921{
 922        struct sctp_transport *active;
 923        struct sctp_transport *match;
 924        struct sctp_transport *transport;
 925        struct sctp_chunk *chunk;
 926        __be32 key = htonl(tsn);
 927
 928        match = NULL;
 929
 930        /*
 931         * FIXME: In general, find a more efficient data structure for
 932         * searching.
 933         */
 934
 935        /*
 936         * The general strategy is to search each transport's transmitted
 937         * list.   Return which transport this TSN lives on.
 938         *
 939         * Let's be hopeful and check the active_path first.
 940         * Another optimization would be to know if there is only one
 941         * outbound path and not have to look for the TSN at all.
 942         *
 943         */
 944
 945        active = asoc->peer.active_path;
 946
 947        list_for_each_entry(chunk, &active->transmitted,
 948                        transmitted_list) {
 949
 950                if (key == chunk->subh.data_hdr->tsn) {
 951                        match = active;
 952                        goto out;
 953                }
 954        }
 955
 956        /* If not found, go search all the other transports. */
 957        list_for_each_entry(transport, &asoc->peer.transport_addr_list,
 958                        transports) {
 959
 960                if (transport == active)
 961                        continue;
 962                list_for_each_entry(chunk, &transport->transmitted,
 963                                transmitted_list) {
 964                        if (key == chunk->subh.data_hdr->tsn) {
 965                                match = transport;
 966                                goto out;
 967                        }
 968                }
 969        }
 970out:
 971        return match;
 972}
 973
 974/* Is this the association we are looking for? */
 975struct sctp_transport *sctp_assoc_is_match(struct sctp_association *asoc,
 976                                           struct net *net,
 977                                           const union sctp_addr *laddr,
 978                                           const union sctp_addr *paddr)
 979{
 980        struct sctp_transport *transport;
 981
 982        if ((htons(asoc->base.bind_addr.port) == laddr->v4.sin_port) &&
 983            (htons(asoc->peer.port) == paddr->v4.sin_port) &&
 984            net_eq(sock_net(asoc->base.sk), net)) {
 985                transport = sctp_assoc_lookup_paddr(asoc, paddr);
 986                if (!transport)
 987                        goto out;
 988
 989                if (sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
 990                                         sctp_sk(asoc->base.sk)))
 991                        goto out;
 992        }
 993        transport = NULL;
 994
 995out:
 996        return transport;
 997}
 998
 999/* Do delayed input processing.  This is scheduled by sctp_rcv(). */
1000static void sctp_assoc_bh_rcv(struct work_struct *work)
1001{
1002        struct sctp_association *asoc =
1003                container_of(work, struct sctp_association,
1004                             base.inqueue.immediate);
1005        struct net *net = sock_net(asoc->base.sk);
1006        struct sctp_endpoint *ep;
1007        struct sctp_chunk *chunk;
1008        struct sctp_inq *inqueue;
1009        int state;
1010        sctp_subtype_t subtype;
1011        int error = 0;
1012
1013        /* The association should be held so we should be safe. */
1014        ep = asoc->ep;
1015
1016        inqueue = &asoc->base.inqueue;
1017        sctp_association_hold(asoc);
1018        while (NULL != (chunk = sctp_inq_pop(inqueue))) {
1019                state = asoc->state;
1020                subtype = SCTP_ST_CHUNK(chunk->chunk_hdr->type);
1021
1022                /* SCTP-AUTH, Section 6.3:
1023                 *    The receiver has a list of chunk types which it expects
1024                 *    to be received only after an AUTH-chunk.  This list has
1025                 *    been sent to the peer during the association setup.  It
1026                 *    MUST silently discard these chunks if they are not placed
1027                 *    after an AUTH chunk in the packet.
1028                 */
1029                if (sctp_auth_recv_cid(subtype.chunk, asoc) && !chunk->auth)
1030                        continue;
1031
1032                /* Remember where the last DATA chunk came from so we
1033                 * know where to send the SACK.
1034                 */
1035                if (sctp_chunk_is_data(chunk))
1036                        asoc->peer.last_data_from = chunk->transport;
1037                else {
1038                        SCTP_INC_STATS(net, SCTP_MIB_INCTRLCHUNKS);
1039                        asoc->stats.ictrlchunks++;
1040                        if (chunk->chunk_hdr->type == SCTP_CID_SACK)
1041                                asoc->stats.isacks++;
1042                }
1043
1044                if (chunk->transport)
1045                        chunk->transport->last_time_heard = ktime_get();
1046
1047                /* Run through the state machine. */
1048                error = sctp_do_sm(net, SCTP_EVENT_T_CHUNK, subtype,
1049                                   state, ep, asoc, chunk, GFP_ATOMIC);
1050
1051                /* Check to see if the association is freed in response to
1052                 * the incoming chunk.  If so, get out of the while loop.
1053                 */
1054                if (asoc->base.dead)
1055                        break;
1056
1057                /* If there is an error on chunk, discard this packet. */
1058                if (error && chunk)
1059                        chunk->pdiscard = 1;
1060        }
1061        sctp_association_put(asoc);
1062}
1063
1064/* This routine moves an association from its old sk to a new sk.  */
1065void sctp_assoc_migrate(struct sctp_association *assoc, struct sock *newsk)
1066{
1067        struct sctp_sock *newsp = sctp_sk(newsk);
1068        struct sock *oldsk = assoc->base.sk;
1069
1070        /* Delete the association from the old endpoint's list of
1071         * associations.
1072         */
1073        list_del_init(&assoc->asocs);
1074
1075        /* Decrement the backlog value for a TCP-style socket. */
1076        if (sctp_style(oldsk, TCP))
1077                oldsk->sk_ack_backlog--;
1078
1079        /* Release references to the old endpoint and the sock.  */
1080        sctp_endpoint_put(assoc->ep);
1081        sock_put(assoc->base.sk);
1082
1083        /* Get a reference to the new endpoint.  */
1084        assoc->ep = newsp->ep;
1085        sctp_endpoint_hold(assoc->ep);
1086
1087        /* Get a reference to the new sock.  */
1088        assoc->base.sk = newsk;
1089        sock_hold(assoc->base.sk);
1090
1091        /* Add the association to the new endpoint's list of associations.  */
1092        sctp_endpoint_add_asoc(newsp->ep, assoc);
1093}
1094
1095/* Update an association (possibly from unexpected COOKIE-ECHO processing).  */
1096void sctp_assoc_update(struct sctp_association *asoc,
1097                       struct sctp_association *new)
1098{
1099        struct sctp_transport *trans;
1100        struct list_head *pos, *temp;
1101
1102        /* Copy in new parameters of peer. */
1103        asoc->c = new->c;
1104        asoc->peer.rwnd = new->peer.rwnd;
1105        asoc->peer.sack_needed = new->peer.sack_needed;
1106        asoc->peer.auth_capable = new->peer.auth_capable;
1107        asoc->peer.i = new->peer.i;
1108        sctp_tsnmap_init(&asoc->peer.tsn_map, SCTP_TSN_MAP_INITIAL,
1109                         asoc->peer.i.initial_tsn, GFP_ATOMIC);
1110
1111        /* Remove any peer addresses not present in the new association. */
1112        list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
1113                trans = list_entry(pos, struct sctp_transport, transports);
1114                if (!sctp_assoc_lookup_paddr(new, &trans->ipaddr)) {
1115                        sctp_assoc_rm_peer(asoc, trans);
1116                        continue;
1117                }
1118
1119                if (asoc->state >= SCTP_STATE_ESTABLISHED)
1120                        sctp_transport_reset(trans);
1121        }
1122
1123        /* If the case is A (association restart), use
1124         * initial_tsn as next_tsn. If the case is B, use
1125         * current next_tsn in case data sent to peer
1126         * has been discarded and needs retransmission.
1127         */
1128        if (asoc->state >= SCTP_STATE_ESTABLISHED) {
1129                asoc->next_tsn = new->next_tsn;
1130                asoc->ctsn_ack_point = new->ctsn_ack_point;
1131                asoc->adv_peer_ack_point = new->adv_peer_ack_point;
1132
1133                /* Reinitialize SSN for both local streams
1134                 * and peer's streams.
1135                 */
1136                sctp_ssnmap_clear(asoc->ssnmap);
1137
1138                /* Flush the ULP reassembly and ordered queue.
1139                 * Any data there will now be stale and will
1140                 * cause problems.
1141                 */
1142                sctp_ulpq_flush(&asoc->ulpq);
1143
1144                /* reset the overall association error count so
1145                 * that the restarted association doesn't get torn
1146                 * down on the next retransmission timer.
1147                 */
1148                asoc->overall_error_count = 0;
1149
1150        } else {
1151                /* Add any peer addresses from the new association. */
1152                list_for_each_entry(trans, &new->peer.transport_addr_list,
1153                                transports) {
1154                        if (!sctp_assoc_lookup_paddr(asoc, &trans->ipaddr))
1155                                sctp_assoc_add_peer(asoc, &trans->ipaddr,
1156                                                    GFP_ATOMIC, trans->state);
1157                }
1158
1159                asoc->ctsn_ack_point = asoc->next_tsn - 1;
1160                asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
1161                if (!asoc->ssnmap) {
1162                        /* Move the ssnmap. */
1163                        asoc->ssnmap = new->ssnmap;
1164                        new->ssnmap = NULL;
1165                }
1166
1167                if (!asoc->assoc_id) {
1168                        /* get a new association id since we don't have one
1169                         * yet.
1170                         */
1171                        sctp_assoc_set_id(asoc, GFP_ATOMIC);
1172                }
1173        }
1174
1175        /* SCTP-AUTH: Save the peer parameters from the new associations
1176         * and also move the association shared keys over
1177         */
1178        kfree(asoc->peer.peer_random);
1179        asoc->peer.peer_random = new->peer.peer_random;
1180        new->peer.peer_random = NULL;
1181
1182        kfree(asoc->peer.peer_chunks);
1183        asoc->peer.peer_chunks = new->peer.peer_chunks;
1184        new->peer.peer_chunks = NULL;
1185
1186        kfree(asoc->peer.peer_hmacs);
1187        asoc->peer.peer_hmacs = new->peer.peer_hmacs;
1188        new->peer.peer_hmacs = NULL;
1189
1190        sctp_auth_asoc_init_active_key(asoc, GFP_ATOMIC);
1191}
1192
1193/* Update the retran path for sending a retransmitted packet.
1194 * See also RFC4960, 6.4. Multi-Homed SCTP Endpoints:
1195 *
1196 *   When there is outbound data to send and the primary path
1197 *   becomes inactive (e.g., due to failures), or where the
1198 *   SCTP user explicitly requests to send data to an
1199 *   inactive destination transport address, before reporting
1200 *   an error to its ULP, the SCTP endpoint should try to send
1201 *   the data to an alternate active destination transport
1202 *   address if one exists.
1203 *
1204 *   When retransmitting data that timed out, if the endpoint
1205 *   is multihomed, it should consider each source-destination
1206 *   address pair in its retransmission selection policy.
1207 *   When retransmitting timed-out data, the endpoint should
1208 *   attempt to pick the most divergent source-destination
1209 *   pair from the original source-destination pair to which
1210 *   the packet was transmitted.
1211 *
1212 *   Note: Rules for picking the most divergent source-destination
1213 *   pair are an implementation decision and are not specified
1214 *   within this document.
1215 *
1216 * Our basic strategy is to round-robin transports in priorities
1217 * according to sctp_trans_score() e.g., if no such
1218 * transport with state SCTP_ACTIVE exists, round-robin through
1219 * SCTP_UNKNOWN, etc. You get the picture.
1220 */
1221static u8 sctp_trans_score(const struct sctp_transport *trans)
1222{
1223        switch (trans->state) {
1224        case SCTP_ACTIVE:
1225                return 3;       /* best case */
1226        case SCTP_UNKNOWN:
1227                return 2;
1228        case SCTP_PF:
1229                return 1;
1230        default: /* case SCTP_INACTIVE */
1231                return 0;       /* worst case */
1232        }
1233}
1234
1235static struct sctp_transport *sctp_trans_elect_tie(struct sctp_transport *trans1,
1236                                                   struct sctp_transport *trans2)
1237{
1238        if (trans1->error_count > trans2->error_count) {
1239                return trans2;
1240        } else if (trans1->error_count == trans2->error_count &&
1241                   ktime_after(trans2->last_time_heard,
1242                               trans1->last_time_heard)) {
1243                return trans2;
1244        } else {
1245                return trans1;
1246        }
1247}
1248
1249static struct sctp_transport *sctp_trans_elect_best(struct sctp_transport *curr,
1250                                                    struct sctp_transport *best)
1251{
1252        u8 score_curr, score_best;
1253
1254        if (best == NULL || curr == best)
1255                return curr;
1256
1257        score_curr = sctp_trans_score(curr);
1258        score_best = sctp_trans_score(best);
1259
1260        /* First, try a score-based selection if both transport states
1261         * differ. If we're in a tie, lets try to make a more clever
1262         * decision here based on error counts and last time heard.
1263         */
1264        if (score_curr > score_best)
1265                return curr;
1266        else if (score_curr == score_best)
1267                return sctp_trans_elect_tie(best, curr);
1268        else
1269                return best;
1270}
1271
1272void sctp_assoc_update_retran_path(struct sctp_association *asoc)
1273{
1274        struct sctp_transport *trans = asoc->peer.retran_path;
1275        struct sctp_transport *trans_next = NULL;
1276
1277        /* We're done as we only have the one and only path. */
1278        if (asoc->peer.transport_count == 1)
1279                return;
1280        /* If active_path and retran_path are the same and active,
1281         * then this is the only active path. Use it.
1282         */
1283        if (asoc->peer.active_path == asoc->peer.retran_path &&
1284            asoc->peer.active_path->state == SCTP_ACTIVE)
1285                return;
1286
1287        /* Iterate from retran_path's successor back to retran_path. */
1288        for (trans = list_next_entry(trans, transports); 1;
1289             trans = list_next_entry(trans, transports)) {
1290                /* Manually skip the head element. */
1291                if (&trans->transports == &asoc->peer.transport_addr_list)
1292                        continue;
1293                if (trans->state == SCTP_UNCONFIRMED)
1294                        continue;
1295                trans_next = sctp_trans_elect_best(trans, trans_next);
1296                /* Active is good enough for immediate return. */
1297                if (trans_next->state == SCTP_ACTIVE)
1298                        break;
1299                /* We've reached the end, time to update path. */
1300                if (trans == asoc->peer.retran_path)
1301                        break;
1302        }
1303
1304        asoc->peer.retran_path = trans_next;
1305
1306        pr_debug("%s: association:%p updated new path to addr:%pISpc\n",
1307                 __func__, asoc, &asoc->peer.retran_path->ipaddr.sa);
1308}
1309
1310static void sctp_select_active_and_retran_path(struct sctp_association *asoc)
1311{
1312        struct sctp_transport *trans, *trans_pri = NULL, *trans_sec = NULL;
1313        struct sctp_transport *trans_pf = NULL;
1314
1315        /* Look for the two most recently used active transports. */
1316        list_for_each_entry(trans, &asoc->peer.transport_addr_list,
1317                            transports) {
1318                /* Skip uninteresting transports. */
1319                if (trans->state == SCTP_INACTIVE ||
1320                    trans->state == SCTP_UNCONFIRMED)
1321                        continue;
1322                /* Keep track of the best PF transport from our
1323                 * list in case we don't find an active one.
1324                 */
1325                if (trans->state == SCTP_PF) {
1326                        trans_pf = sctp_trans_elect_best(trans, trans_pf);
1327                        continue;
1328                }
1329                /* For active transports, pick the most recent ones. */
1330                if (trans_pri == NULL ||
1331                    ktime_after(trans->last_time_heard,
1332                                trans_pri->last_time_heard)) {
1333                        trans_sec = trans_pri;
1334                        trans_pri = trans;
1335                } else if (trans_sec == NULL ||
1336                           ktime_after(trans->last_time_heard,
1337                                       trans_sec->last_time_heard)) {
1338                        trans_sec = trans;
1339                }
1340        }
1341
1342        /* RFC 2960 6.4 Multi-Homed SCTP Endpoints
1343         *
1344         * By default, an endpoint should always transmit to the primary
1345         * path, unless the SCTP user explicitly specifies the
1346         * destination transport address (and possibly source transport
1347         * address) to use. [If the primary is active but not most recent,
1348         * bump the most recently used transport.]
1349         */
1350        if ((asoc->peer.primary_path->state == SCTP_ACTIVE ||
1351             asoc->peer.primary_path->state == SCTP_UNKNOWN) &&
1352             asoc->peer.primary_path != trans_pri) {
1353                trans_sec = trans_pri;
1354                trans_pri = asoc->peer.primary_path;
1355        }
1356
1357        /* We did not find anything useful for a possible retransmission
1358         * path; either primary path that we found is the the same as
1359         * the current one, or we didn't generally find an active one.
1360         */
1361        if (trans_sec == NULL)
1362                trans_sec = trans_pri;
1363
1364        /* If we failed to find a usable transport, just camp on the
1365         * active or pick a PF iff it's the better choice.
1366         */
1367        if (trans_pri == NULL) {
1368                trans_pri = sctp_trans_elect_best(asoc->peer.active_path, trans_pf);
1369                trans_sec = trans_pri;
1370        }
1371
1372        /* Set the active and retran transports. */
1373        asoc->peer.active_path = trans_pri;
1374        asoc->peer.retran_path = trans_sec;
1375}
1376
1377struct sctp_transport *
1378sctp_assoc_choose_alter_transport(struct sctp_association *asoc,
1379                                  struct sctp_transport *last_sent_to)
1380{
1381        /* If this is the first time packet is sent, use the active path,
1382         * else use the retran path. If the last packet was sent over the
1383         * retran path, update the retran path and use it.
1384         */
1385        if (last_sent_to == NULL) {
1386                return asoc->peer.active_path;
1387        } else {
1388                if (last_sent_to == asoc->peer.retran_path)
1389                        sctp_assoc_update_retran_path(asoc);
1390
1391                return asoc->peer.retran_path;
1392        }
1393}
1394
1395/* Update the association's pmtu and frag_point by going through all the
1396 * transports. This routine is called when a transport's PMTU has changed.
1397 */
1398void sctp_assoc_sync_pmtu(struct sock *sk, struct sctp_association *asoc)
1399{
1400        struct sctp_transport *t;
1401        __u32 pmtu = 0;
1402
1403        if (!asoc)
1404                return;
1405
1406        /* Get the lowest pmtu of all the transports. */
1407        list_for_each_entry(t, &asoc->peer.transport_addr_list,
1408                                transports) {
1409                if (t->pmtu_pending && t->dst) {
1410                        sctp_transport_update_pmtu(sk, t,
1411                                                   SCTP_TRUNC4(dst_mtu(t->dst)));
1412                        t->pmtu_pending = 0;
1413                }
1414                if (!pmtu || (t->pathmtu < pmtu))
1415                        pmtu = t->pathmtu;
1416        }
1417
1418        if (pmtu) {
1419                asoc->pathmtu = pmtu;
1420                asoc->frag_point = sctp_frag_point(asoc, pmtu);
1421        }
1422
1423        pr_debug("%s: asoc:%p, pmtu:%d, frag_point:%d\n", __func__, asoc,
1424                 asoc->pathmtu, asoc->frag_point);
1425}
1426
1427/* Should we send a SACK to update our peer? */
1428static inline bool sctp_peer_needs_update(struct sctp_association *asoc)
1429{
1430        struct net *net = sock_net(asoc->base.sk);
1431        switch (asoc->state) {
1432        case SCTP_STATE_ESTABLISHED:
1433        case SCTP_STATE_SHUTDOWN_PENDING:
1434        case SCTP_STATE_SHUTDOWN_RECEIVED:
1435        case SCTP_STATE_SHUTDOWN_SENT:
1436                if ((asoc->rwnd > asoc->a_rwnd) &&
1437                    ((asoc->rwnd - asoc->a_rwnd) >= max_t(__u32,
1438                           (asoc->base.sk->sk_rcvbuf >> net->sctp.rwnd_upd_shift),
1439                           asoc->pathmtu)))
1440                        return true;
1441                break;
1442        default:
1443                break;
1444        }
1445        return false;
1446}
1447
1448/* Increase asoc's rwnd by len and send any window update SACK if needed. */
1449void sctp_assoc_rwnd_increase(struct sctp_association *asoc, unsigned int len)
1450{
1451        struct sctp_chunk *sack;
1452        struct timer_list *timer;
1453
1454        if (asoc->rwnd_over) {
1455                if (asoc->rwnd_over >= len) {
1456                        asoc->rwnd_over -= len;
1457                } else {
1458                        asoc->rwnd += (len - asoc->rwnd_over);
1459                        asoc->rwnd_over = 0;
1460                }
1461        } else {
1462                asoc->rwnd += len;
1463        }
1464
1465        /* If we had window pressure, start recovering it
1466         * once our rwnd had reached the accumulated pressure
1467         * threshold.  The idea is to recover slowly, but up
1468         * to the initial advertised window.
1469         */
1470        if (asoc->rwnd_press && asoc->rwnd >= asoc->rwnd_press) {
1471                int change = min(asoc->pathmtu, asoc->rwnd_press);
1472                asoc->rwnd += change;
1473                asoc->rwnd_press -= change;
1474        }
1475
1476        pr_debug("%s: asoc:%p rwnd increased by %d to (%u, %u) - %u\n",
1477                 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
1478                 asoc->a_rwnd);
1479
1480        /* Send a window update SACK if the rwnd has increased by at least the
1481         * minimum of the association's PMTU and half of the receive buffer.
1482         * The algorithm used is similar to the one described in
1483         * Section 4.2.3.3 of RFC 1122.
1484         */
1485        if (sctp_peer_needs_update(asoc)) {
1486                asoc->a_rwnd = asoc->rwnd;
1487
1488                pr_debug("%s: sending window update SACK- asoc:%p rwnd:%u "
1489                         "a_rwnd:%u\n", __func__, asoc, asoc->rwnd,
1490                         asoc->a_rwnd);
1491
1492                sack = sctp_make_sack(asoc);
1493                if (!sack)
1494                        return;
1495
1496                asoc->peer.sack_needed = 0;
1497
1498                sctp_outq_tail(&asoc->outqueue, sack, GFP_ATOMIC);
1499
1500                /* Stop the SACK timer.  */
1501                timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
1502                if (del_timer(timer))
1503                        sctp_association_put(asoc);
1504        }
1505}
1506
1507/* Decrease asoc's rwnd by len. */
1508void sctp_assoc_rwnd_decrease(struct sctp_association *asoc, unsigned int len)
1509{
1510        int rx_count;
1511        int over = 0;
1512
1513        if (unlikely(!asoc->rwnd || asoc->rwnd_over))
1514                pr_debug("%s: association:%p has asoc->rwnd:%u, "
1515                         "asoc->rwnd_over:%u!\n", __func__, asoc,
1516                         asoc->rwnd, asoc->rwnd_over);
1517
1518        if (asoc->ep->rcvbuf_policy)
1519                rx_count = atomic_read(&asoc->rmem_alloc);
1520        else
1521                rx_count = atomic_read(&asoc->base.sk->sk_rmem_alloc);
1522
1523        /* If we've reached or overflowed our receive buffer, announce
1524         * a 0 rwnd if rwnd would still be positive.  Store the
1525         * the potential pressure overflow so that the window can be restored
1526         * back to original value.
1527         */
1528        if (rx_count >= asoc->base.sk->sk_rcvbuf)
1529                over = 1;
1530
1531        if (asoc->rwnd >= len) {
1532                asoc->rwnd -= len;
1533                if (over) {
1534                        asoc->rwnd_press += asoc->rwnd;
1535                        asoc->rwnd = 0;
1536                }
1537        } else {
1538                asoc->rwnd_over = len - asoc->rwnd;
1539                asoc->rwnd = 0;
1540        }
1541
1542        pr_debug("%s: asoc:%p rwnd decreased by %d to (%u, %u, %u)\n",
1543                 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
1544                 asoc->rwnd_press);
1545}
1546
1547/* Build the bind address list for the association based on info from the
1548 * local endpoint and the remote peer.
1549 */
1550int sctp_assoc_set_bind_addr_from_ep(struct sctp_association *asoc,
1551                                     sctp_scope_t scope, gfp_t gfp)
1552{
1553        int flags;
1554
1555        /* Use scoping rules to determine the subset of addresses from
1556         * the endpoint.
1557         */
1558        flags = (PF_INET6 == asoc->base.sk->sk_family) ? SCTP_ADDR6_ALLOWED : 0;
1559        if (asoc->peer.ipv4_address)
1560                flags |= SCTP_ADDR4_PEERSUPP;
1561        if (asoc->peer.ipv6_address)
1562                flags |= SCTP_ADDR6_PEERSUPP;
1563
1564        return sctp_bind_addr_copy(sock_net(asoc->base.sk),
1565                                   &asoc->base.bind_addr,
1566                                   &asoc->ep->base.bind_addr,
1567                                   scope, gfp, flags);
1568}
1569
1570/* Build the association's bind address list from the cookie.  */
1571int sctp_assoc_set_bind_addr_from_cookie(struct sctp_association *asoc,
1572                                         struct sctp_cookie *cookie,
1573                                         gfp_t gfp)
1574{
1575        int var_size2 = ntohs(cookie->peer_init->chunk_hdr.length);
1576        int var_size3 = cookie->raw_addr_list_len;
1577        __u8 *raw = (__u8 *)cookie->peer_init + var_size2;
1578
1579        return sctp_raw_to_bind_addrs(&asoc->base.bind_addr, raw, var_size3,
1580                                      asoc->ep->base.bind_addr.port, gfp);
1581}
1582
1583/* Lookup laddr in the bind address list of an association. */
1584int sctp_assoc_lookup_laddr(struct sctp_association *asoc,
1585                            const union sctp_addr *laddr)
1586{
1587        int found = 0;
1588
1589        if ((asoc->base.bind_addr.port == ntohs(laddr->v4.sin_port)) &&
1590            sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
1591                                 sctp_sk(asoc->base.sk)))
1592                found = 1;
1593
1594        return found;
1595}
1596
1597/* Set an association id for a given association */
1598int sctp_assoc_set_id(struct sctp_association *asoc, gfp_t gfp)
1599{
1600        bool preload = gfpflags_allow_blocking(gfp);
1601        int ret;
1602
1603        /* If the id is already assigned, keep it. */
1604        if (asoc->assoc_id)
1605                return 0;
1606
1607        if (preload)
1608                idr_preload(gfp);
1609        spin_lock_bh(&sctp_assocs_id_lock);
1610        /* 0 is not a valid assoc_id, must be >= 1 */
1611        ret = idr_alloc_cyclic(&sctp_assocs_id, asoc, 1, 0, GFP_NOWAIT);
1612        spin_unlock_bh(&sctp_assocs_id_lock);
1613        if (preload)
1614                idr_preload_end();
1615        if (ret < 0)
1616                return ret;
1617
1618        asoc->assoc_id = (sctp_assoc_t)ret;
1619        return 0;
1620}
1621
1622/* Free the ASCONF queue */
1623static void sctp_assoc_free_asconf_queue(struct sctp_association *asoc)
1624{
1625        struct sctp_chunk *asconf;
1626        struct sctp_chunk *tmp;
1627
1628        list_for_each_entry_safe(asconf, tmp, &asoc->addip_chunk_list, list) {
1629                list_del_init(&asconf->list);
1630                sctp_chunk_free(asconf);
1631        }
1632}
1633
1634/* Free asconf_ack cache */
1635static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc)
1636{
1637        struct sctp_chunk *ack;
1638        struct sctp_chunk *tmp;
1639
1640        list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
1641                                transmitted_list) {
1642                list_del_init(&ack->transmitted_list);
1643                sctp_chunk_free(ack);
1644        }
1645}
1646
1647/* Clean up the ASCONF_ACK queue */
1648void sctp_assoc_clean_asconf_ack_cache(const struct sctp_association *asoc)
1649{
1650        struct sctp_chunk *ack;
1651        struct sctp_chunk *tmp;
1652
1653        /* We can remove all the entries from the queue up to
1654         * the "Peer-Sequence-Number".
1655         */
1656        list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
1657                                transmitted_list) {
1658                if (ack->subh.addip_hdr->serial ==
1659                                htonl(asoc->peer.addip_serial))
1660                        break;
1661
1662                list_del_init(&ack->transmitted_list);
1663                sctp_chunk_free(ack);
1664        }
1665}
1666
1667/* Find the ASCONF_ACK whose serial number matches ASCONF */
1668struct sctp_chunk *sctp_assoc_lookup_asconf_ack(
1669                                        const struct sctp_association *asoc,
1670                                        __be32 serial)
1671{
1672        struct sctp_chunk *ack;
1673
1674        /* Walk through the list of cached ASCONF-ACKs and find the
1675         * ack chunk whose serial number matches that of the request.
1676         */
1677        list_for_each_entry(ack, &asoc->asconf_ack_list, transmitted_list) {
1678                if (sctp_chunk_pending(ack))
1679                        continue;
1680                if (ack->subh.addip_hdr->serial == serial) {
1681                        sctp_chunk_hold(ack);
1682                        return ack;
1683                }
1684        }
1685
1686        return NULL;
1687}
1688
1689void sctp_asconf_queue_teardown(struct sctp_association *asoc)
1690{
1691        /* Free any cached ASCONF_ACK chunk. */
1692        sctp_assoc_free_asconf_acks(asoc);
1693
1694        /* Free the ASCONF queue. */
1695        sctp_assoc_free_asconf_queue(asoc);
1696
1697        /* Free any cached ASCONF chunk. */
1698        if (asoc->addip_last_asconf)
1699                sctp_chunk_free(asoc->addip_last_asconf);
1700}
1701