1// SPDX-License-Identifier: GPL-2.0-or-later
   2/* SCTP kernel implementation
   3 * (C) Copyright IBM Corp. 2001, 2004
   4 * Copyright (c) 1999-2000 Cisco, Inc.
   5 * Copyright (c) 1999-2001 Motorola, Inc.
   6 * Copyright (c) 2001 Intel Corp.
   7 * Copyright (c) 2001 La Monte H.P. Yarroll
   8 *
   9 * This file is part of the SCTP kernel implementation
  10 *
  11 * This module provides the abstraction for an SCTP association.
  12 *
  13 * Please send any bug reports or fixes you make to the
  14 * email address(es):
  15 *    lksctp developers <linux-sctp@vger.kernel.org>
  16 *
  17 * Written or modified by:
  18 *    La Monte H.P. Yarroll <piggy@acm.org>
  19 *    Karl Knutson          <karl@athena.chicago.il.us>
  20 *    Jon Grimm             <jgrimm@us.ibm.com>
  21 *    Xingang Guo           <xingang.guo@intel.com>
  22 *    Hui Huang             <hui.huang@nokia.com>
  23 *    Sridhar Samudrala	    <sri@us.ibm.com>
  24 *    Daisy Chang	    <daisyc@us.ibm.com>
  25 *    Ryan Layer	    <rmlayer@us.ibm.com>
  26 *    Kevin Gao             <kevin.gao@intel.com>
  27 */
  28
  29#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  30
  31#include <linux/types.h>
  32#include <linux/fcntl.h>
  33#include <linux/poll.h>
  34#include <linux/init.h>
  35
  36#include <linux/slab.h>
  37#include <linux/in.h>
  38#include <net/ipv6.h>
  39#include <net/sctp/sctp.h>
  40#include <net/sctp/sm.h>
  41
  42/* Forward declarations for internal functions. */
  43static void sctp_select_active_and_retran_path(struct sctp_association *asoc);
  44static void sctp_assoc_bh_rcv(struct work_struct *work);
  45static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc);
  46static void sctp_assoc_free_asconf_queue(struct sctp_association *asoc);
  47
  48/* 1st Level Abstractions. */
  49
  50/* Initialize a new association from provided memory. */
  51static struct sctp_association *sctp_association_init(
  52					struct sctp_association *asoc,
  53					const struct sctp_endpoint *ep,
  54					const struct sock *sk,
  55					enum sctp_scope scope, gfp_t gfp)
  56{
  57	struct sctp_sock *sp;
  58	struct sctp_paramhdr *p;
  59	int i;
  60
  61	/* Retrieve the SCTP per socket area.  */
  62	sp = sctp_sk((struct sock *)sk);
  63
  64	/* Discarding const is appropriate here.  */
  65	asoc->ep = (struct sctp_endpoint *)ep;
  66	asoc->base.sk = (struct sock *)sk;
 
  67
  68	sctp_endpoint_hold(asoc->ep);
  69	sock_hold(asoc->base.sk);
  70
  71	/* Initialize the common base substructure.  */
  72	asoc->base.type = SCTP_EP_TYPE_ASSOCIATION;
  73
  74	/* Initialize the object handling fields.  */
  75	refcount_set(&asoc->base.refcnt, 1);
  76
  77	/* Initialize the bind addr area.  */
  78	sctp_bind_addr_init(&asoc->base.bind_addr, ep->base.bind_addr.port);
  79
  80	asoc->state = SCTP_STATE_CLOSED;
  81	asoc->cookie_life = ms_to_ktime(sp->assocparams.sasoc_cookie_life);
  82	asoc->user_frag = sp->user_frag;
  83
  84	/* Set the association max_retrans and RTO values from the
  85	 * socket values.
  86	 */
  87	asoc->max_retrans = sp->assocparams.sasoc_asocmaxrxt;
  88	asoc->pf_retrans  = sp->pf_retrans;
 
 
  89
  90	asoc->rto_initial = msecs_to_jiffies(sp->rtoinfo.srto_initial);
  91	asoc->rto_max = msecs_to_jiffies(sp->rtoinfo.srto_max);
  92	asoc->rto_min = msecs_to_jiffies(sp->rtoinfo.srto_min);
  93
  94	/* Initialize the association's heartbeat interval based on the
  95	 * sock configured value.
  96	 */
  97	asoc->hbinterval = msecs_to_jiffies(sp->hbinterval);
 
 
 
  98
  99	/* Initialize path max retrans value. */
 100	asoc->pathmaxrxt = sp->pathmaxrxt;
 101
 102	asoc->flowlabel = sp->flowlabel;
 103	asoc->dscp = sp->dscp;
 104
 105	/* Set association default SACK delay */
 106	asoc->sackdelay = msecs_to_jiffies(sp->sackdelay);
 107	asoc->sackfreq = sp->sackfreq;
 108
 109	/* Set the association default flags controlling
 110	 * Heartbeat, SACK delay, and Path MTU Discovery.
 111	 */
 112	asoc->param_flags = sp->param_flags;
 113
 114	/* Initialize the maximum number of new data packets that can be sent
 115	 * in a burst.
 116	 */
 117	asoc->max_burst = sp->max_burst;
 118
 119	asoc->subscribe = sp->subscribe;
 120
 121	/* initialize association timers */
 122	asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_COOKIE] = asoc->rto_initial;
 123	asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_INIT] = asoc->rto_initial;
 124	asoc->timeouts[SCTP_EVENT_TIMEOUT_T2_SHUTDOWN] = asoc->rto_initial;
 125
 126	/* sctpimpguide Section 2.12.2
 127	 * If the 'T5-shutdown-guard' timer is used, it SHOULD be set to the
 128	 * recommended value of 5 times 'RTO.Max'.
 129	 */
 130	asoc->timeouts[SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]
 131		= 5 * asoc->rto_max;
 132
 133	asoc->timeouts[SCTP_EVENT_TIMEOUT_SACK] = asoc->sackdelay;
 134	asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE] = sp->autoclose * HZ;
 
 135
 136	/* Initializes the timers */
 137	for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i)
 138		timer_setup(&asoc->timers[i], sctp_timer_events[i], 0);
 139
 140	/* Pull default initialization values from the sock options.
 141	 * Note: This assumes that the values have already been
 142	 * validated in the sock.
 143	 */
 144	asoc->c.sinit_max_instreams = sp->initmsg.sinit_max_instreams;
 145	asoc->c.sinit_num_ostreams  = sp->initmsg.sinit_num_ostreams;
 146	asoc->max_init_attempts	= sp->initmsg.sinit_max_attempts;
 147
 148	asoc->max_init_timeo =
 149		 msecs_to_jiffies(sp->initmsg.sinit_max_init_timeo);
 150
 151	/* Set the local window size for receive.
 152	 * This is also the rcvbuf space per association.
 153	 * RFC 6 - A SCTP receiver MUST be able to receive a minimum of
 154	 * 1500 bytes in one SCTP packet.
 155	 */
 156	if ((sk->sk_rcvbuf/2) < SCTP_DEFAULT_MINWINDOW)
 157		asoc->rwnd = SCTP_DEFAULT_MINWINDOW;
 158	else
 159		asoc->rwnd = sk->sk_rcvbuf/2;
 160
 161	asoc->a_rwnd = asoc->rwnd;
 162
 163	/* Use my own max window until I learn something better.  */
 164	asoc->peer.rwnd = SCTP_DEFAULT_MAXWINDOW;
 165
 166	/* Initialize the receive memory counter */
 167	atomic_set(&asoc->rmem_alloc, 0);
 168
 169	init_waitqueue_head(&asoc->wait);
 170
 171	asoc->c.my_vtag = sctp_generate_tag(ep);
 172	asoc->c.my_port = ep->base.bind_addr.port;
 173
 174	asoc->c.initial_tsn = sctp_generate_tsn(ep);
 175
 176	asoc->next_tsn = asoc->c.initial_tsn;
 177
 178	asoc->ctsn_ack_point = asoc->next_tsn - 1;
 179	asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
 180	asoc->highest_sacked = asoc->ctsn_ack_point;
 181	asoc->last_cwr_tsn = asoc->ctsn_ack_point;
 182
 183	/* ADDIP Section 4.1 Asconf Chunk Procedures
 184	 *
 185	 * When an endpoint has an ASCONF signaled change to be sent to the
 186	 * remote endpoint it should do the following:
 187	 * ...
 188	 * A2) a serial number should be assigned to the chunk. The serial
 189	 * number SHOULD be a monotonically increasing number. The serial
 190	 * numbers SHOULD be initialized at the start of the
 191	 * association to the same value as the initial TSN.
 192	 */
 193	asoc->addip_serial = asoc->c.initial_tsn;
 194	asoc->strreset_outseq = asoc->c.initial_tsn;
 195
 196	INIT_LIST_HEAD(&asoc->addip_chunk_list);
 197	INIT_LIST_HEAD(&asoc->asconf_ack_list);
 198
 199	/* Make an empty list of remote transport addresses.  */
 200	INIT_LIST_HEAD(&asoc->peer.transport_addr_list);
 201
 202	/* RFC 2960 5.1 Normal Establishment of an Association
 203	 *
 204	 * After the reception of the first data chunk in an
 205	 * association the endpoint must immediately respond with a
 206	 * sack to acknowledge the data chunk.  Subsequent
 207	 * acknowledgements should be done as described in Section
 208	 * 6.2.
 209	 *
 210	 * [We implement this by telling a new association that it
 211	 * already received one packet.]
 212	 */
 213	asoc->peer.sack_needed = 1;
 214	asoc->peer.sack_generation = 1;
 215
 216	/* Create an input queue.  */
 217	sctp_inq_init(&asoc->base.inqueue);
 218	sctp_inq_set_th_handler(&asoc->base.inqueue, sctp_assoc_bh_rcv);
 219
 220	/* Create an output queue.  */
 221	sctp_outq_init(asoc, &asoc->outqueue);
 222
 223	if (!sctp_ulpq_init(&asoc->ulpq, asoc))
 224		goto fail_init;
 225
 226	if (sctp_stream_init(&asoc->stream, asoc->c.sinit_num_ostreams,
 227			     0, gfp))
 228		goto fail_init;
 229
 230	/* Initialize default path MTU. */
 231	asoc->pathmtu = sp->pathmtu;
 232	sctp_assoc_update_frag_point(asoc);
 233
 234	/* Assume that peer would support both address types unless we are
 235	 * told otherwise.
 236	 */
 237	asoc->peer.ipv4_address = 1;
 238	if (asoc->base.sk->sk_family == PF_INET6)
 239		asoc->peer.ipv6_address = 1;
 240	INIT_LIST_HEAD(&asoc->asocs);
 241
 242	asoc->default_stream = sp->default_stream;
 243	asoc->default_ppid = sp->default_ppid;
 244	asoc->default_flags = sp->default_flags;
 245	asoc->default_context = sp->default_context;
 246	asoc->default_timetolive = sp->default_timetolive;
 247	asoc->default_rcv_context = sp->default_rcv_context;
 248
 249	/* AUTH related initializations */
 250	INIT_LIST_HEAD(&asoc->endpoint_shared_keys);
 251	if (sctp_auth_asoc_copy_shkeys(ep, asoc, gfp))
 252		goto stream_free;
 253
 254	asoc->active_key_id = ep->active_key_id;
 255	asoc->strreset_enable = ep->strreset_enable;
 256
 257	/* Save the hmacs and chunks list into this association */
 258	if (ep->auth_hmacs_list)
 259		memcpy(asoc->c.auth_hmacs, ep->auth_hmacs_list,
 260			ntohs(ep->auth_hmacs_list->param_hdr.length));
 261	if (ep->auth_chunk_list)
 262		memcpy(asoc->c.auth_chunks, ep->auth_chunk_list,
 263			ntohs(ep->auth_chunk_list->param_hdr.length));
 264
 265	/* Get the AUTH random number for this association */
 266	p = (struct sctp_paramhdr *)asoc->c.auth_random;
 267	p->type = SCTP_PARAM_RANDOM;
 268	p->length = htons(sizeof(*p) + SCTP_AUTH_RANDOM_LENGTH);
 269	get_random_bytes(p+1, SCTP_AUTH_RANDOM_LENGTH);
 270
 271	return asoc;
 272
 273stream_free:
 274	sctp_stream_free(&asoc->stream);
 275fail_init:
 276	sock_put(asoc->base.sk);
 277	sctp_endpoint_put(asoc->ep);
 278	return NULL;
 279}
 280
 281/* Allocate and initialize a new association */
 282struct sctp_association *sctp_association_new(const struct sctp_endpoint *ep,
 283					      const struct sock *sk,
 284					      enum sctp_scope scope, gfp_t gfp)
 285{
 286	struct sctp_association *asoc;
 287
 288	asoc = kzalloc(sizeof(*asoc), gfp);
 289	if (!asoc)
 290		goto fail;
 291
 292	if (!sctp_association_init(asoc, ep, sk, scope, gfp))
 293		goto fail_init;
 294
 295	SCTP_DBG_OBJCNT_INC(assoc);
 296
 297	pr_debug("Created asoc %p\n", asoc);
 298
 299	return asoc;
 300
 301fail_init:
 302	kfree(asoc);
 303fail:
 304	return NULL;
 305}
 306
 307/* Free this association if possible.  There may still be users, so
 308 * the actual deallocation may be delayed.
 309 */
 310void sctp_association_free(struct sctp_association *asoc)
 311{
 312	struct sock *sk = asoc->base.sk;
 313	struct sctp_transport *transport;
 314	struct list_head *pos, *temp;
 315	int i;
 316
 317	/* Only real associations count against the endpoint, so
 318	 * don't bother for if this is a temporary association.
 319	 */
 320	if (!list_empty(&asoc->asocs)) {
 321		list_del(&asoc->asocs);
 322
 323		/* Decrement the backlog value for a TCP-style listening
 324		 * socket.
 325		 */
 326		if (sctp_style(sk, TCP) && sctp_sstate(sk, LISTENING))
 327			sk->sk_ack_backlog--;
 328	}
 329
 330	/* Mark as dead, so other users can know this structure is
 331	 * going away.
 332	 */
 333	asoc->base.dead = true;
 334
 335	/* Dispose of any data lying around in the outqueue. */
 336	sctp_outq_free(&asoc->outqueue);
 337
 338	/* Dispose of any pending messages for the upper layer. */
 339	sctp_ulpq_free(&asoc->ulpq);
 340
 341	/* Dispose of any pending chunks on the inqueue. */
 342	sctp_inq_free(&asoc->base.inqueue);
 343
 344	sctp_tsnmap_free(&asoc->peer.tsn_map);
 345
 346	/* Free stream information. */
 347	sctp_stream_free(&asoc->stream);
 348
 349	if (asoc->strreset_chunk)
 350		sctp_chunk_free(asoc->strreset_chunk);
 351
 352	/* Clean up the bound address list. */
 353	sctp_bind_addr_free(&asoc->base.bind_addr);
 354
 355	/* Do we need to go through all of our timers and
 356	 * delete them?   To be safe we will try to delete all, but we
 357	 * should be able to go through and make a guess based
 358	 * on our state.
 359	 */
 360	for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i) {
 361		if (del_timer(&asoc->timers[i]))
 362			sctp_association_put(asoc);
 363	}
 364
 365	/* Free peer's cached cookie. */
 366	kfree(asoc->peer.cookie);
 367	kfree(asoc->peer.peer_random);
 368	kfree(asoc->peer.peer_chunks);
 369	kfree(asoc->peer.peer_hmacs);
 370
 371	/* Release the transport structures. */
 372	list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
 373		transport = list_entry(pos, struct sctp_transport, transports);
 374		list_del_rcu(pos);
 375		sctp_unhash_transport(transport);
 376		sctp_transport_free(transport);
 377	}
 378
 379	asoc->peer.transport_count = 0;
 380
 381	sctp_asconf_queue_teardown(asoc);
 382
 383	/* Free pending address space being deleted */
 384	kfree(asoc->asconf_addr_del_pending);
 385
 386	/* AUTH - Free the endpoint shared keys */
 387	sctp_auth_destroy_keys(&asoc->endpoint_shared_keys);
 388
 389	/* AUTH - Free the association shared key */
 390	sctp_auth_key_put(asoc->asoc_shared_key);
 391
 392	sctp_association_put(asoc);
 393}
 394
 395/* Cleanup and free up an association. */
 396static void sctp_association_destroy(struct sctp_association *asoc)
 397{
 398	if (unlikely(!asoc->base.dead)) {
 399		WARN(1, "Attempt to destroy undead association %p!\n", asoc);
 400		return;
 401	}
 402
 403	sctp_endpoint_put(asoc->ep);
 404	sock_put(asoc->base.sk);
 405
 406	if (asoc->assoc_id != 0) {
 407		spin_lock_bh(&sctp_assocs_id_lock);
 408		idr_remove(&sctp_assocs_id, asoc->assoc_id);
 409		spin_unlock_bh(&sctp_assocs_id_lock);
 410	}
 411
 412	WARN_ON(atomic_read(&asoc->rmem_alloc));
 413
 414	kfree_rcu(asoc, rcu);
 415	SCTP_DBG_OBJCNT_DEC(assoc);
 416}
 417
 418/* Change the primary destination address for the peer. */
 419void sctp_assoc_set_primary(struct sctp_association *asoc,
 420			    struct sctp_transport *transport)
 421{
 422	int changeover = 0;
 423
 424	/* it's a changeover only if we already have a primary path
 425	 * that we are changing
 426	 */
 427	if (asoc->peer.primary_path != NULL &&
 428	    asoc->peer.primary_path != transport)
 429		changeover = 1 ;
 430
 431	asoc->peer.primary_path = transport;
 
 
 432
 433	/* Set a default msg_name for events. */
 434	memcpy(&asoc->peer.primary_addr, &transport->ipaddr,
 435	       sizeof(union sctp_addr));
 436
 437	/* If the primary path is changing, assume that the
 438	 * user wants to use this new path.
 439	 */
 440	if ((transport->state == SCTP_ACTIVE) ||
 441	    (transport->state == SCTP_UNKNOWN))
 442		asoc->peer.active_path = transport;
 443
 444	/*
 445	 * SFR-CACC algorithm:
 446	 * Upon the receipt of a request to change the primary
 447	 * destination address, on the data structure for the new
 448	 * primary destination, the sender MUST do the following:
 449	 *
 450	 * 1) If CHANGEOVER_ACTIVE is set, then there was a switch
 451	 * to this destination address earlier. The sender MUST set
 452	 * CYCLING_CHANGEOVER to indicate that this switch is a
 453	 * double switch to the same destination address.
 454	 *
 455	 * Really, only bother is we have data queued or outstanding on
 456	 * the association.
 457	 */
 458	if (!asoc->outqueue.outstanding_bytes && !asoc->outqueue.out_qlen)
 459		return;
 460
 461	if (transport->cacc.changeover_active)
 462		transport->cacc.cycling_changeover = changeover;
 463
 464	/* 2) The sender MUST set CHANGEOVER_ACTIVE to indicate that
 465	 * a changeover has occurred.
 466	 */
 467	transport->cacc.changeover_active = changeover;
 468
 469	/* 3) The sender MUST store the next TSN to be sent in
 470	 * next_tsn_at_change.
 471	 */
 472	transport->cacc.next_tsn_at_change = asoc->next_tsn;
 473}
 474
 475/* Remove a transport from an association.  */
 476void sctp_assoc_rm_peer(struct sctp_association *asoc,
 477			struct sctp_transport *peer)
 478{
 479	struct sctp_transport *transport;
 480	struct list_head *pos;
 481	struct sctp_chunk *ch;
 482
 483	pr_debug("%s: association:%p addr:%pISpc\n",
 484		 __func__, asoc, &peer->ipaddr.sa);
 485
 486	/* If we are to remove the current retran_path, update it
 487	 * to the next peer before removing this peer from the list.
 488	 */
 489	if (asoc->peer.retran_path == peer)
 490		sctp_assoc_update_retran_path(asoc);
 491
 492	/* Remove this peer from the list. */
 493	list_del_rcu(&peer->transports);
 494	/* Remove this peer from the transport hashtable */
 495	sctp_unhash_transport(peer);
 496
 497	/* Get the first transport of asoc. */
 498	pos = asoc->peer.transport_addr_list.next;
 499	transport = list_entry(pos, struct sctp_transport, transports);
 500
 501	/* Update any entries that match the peer to be deleted. */
 502	if (asoc->peer.primary_path == peer)
 503		sctp_assoc_set_primary(asoc, transport);
 504	if (asoc->peer.active_path == peer)
 505		asoc->peer.active_path = transport;
 506	if (asoc->peer.retran_path == peer)
 507		asoc->peer.retran_path = transport;
 508	if (asoc->peer.last_data_from == peer)
 509		asoc->peer.last_data_from = transport;
 510
 511	if (asoc->strreset_chunk &&
 512	    asoc->strreset_chunk->transport == peer) {
 513		asoc->strreset_chunk->transport = transport;
 514		sctp_transport_reset_reconf_timer(transport);
 515	}
 516
 517	/* If we remove the transport an INIT was last sent to, set it to
 518	 * NULL. Combined with the update of the retran path above, this
 519	 * will cause the next INIT to be sent to the next available
 520	 * transport, maintaining the cycle.
 521	 */
 522	if (asoc->init_last_sent_to == peer)
 523		asoc->init_last_sent_to = NULL;
 524
 525	/* If we remove the transport an SHUTDOWN was last sent to, set it
 526	 * to NULL. Combined with the update of the retran path above, this
 527	 * will cause the next SHUTDOWN to be sent to the next available
 528	 * transport, maintaining the cycle.
 529	 */
 530	if (asoc->shutdown_last_sent_to == peer)
 531		asoc->shutdown_last_sent_to = NULL;
 532
 533	/* If we remove the transport an ASCONF was last sent to, set it to
 534	 * NULL.
 535	 */
 536	if (asoc->addip_last_asconf &&
 537	    asoc->addip_last_asconf->transport == peer)
 538		asoc->addip_last_asconf->transport = NULL;
 539
 540	/* If we have something on the transmitted list, we have to
 541	 * save it off.  The best place is the active path.
 542	 */
 543	if (!list_empty(&peer->transmitted)) {
 544		struct sctp_transport *active = asoc->peer.active_path;
 545
 546		/* Reset the transport of each chunk on this list */
 547		list_for_each_entry(ch, &peer->transmitted,
 548					transmitted_list) {
 549			ch->transport = NULL;
 550			ch->rtt_in_progress = 0;
 551		}
 552
 553		list_splice_tail_init(&peer->transmitted,
 554					&active->transmitted);
 555
 556		/* Start a T3 timer here in case it wasn't running so
 557		 * that these migrated packets have a chance to get
 558		 * retransmitted.
 559		 */
 560		if (!timer_pending(&active->T3_rtx_timer))
 561			if (!mod_timer(&active->T3_rtx_timer,
 562					jiffies + active->rto))
 563				sctp_transport_hold(active);
 564	}
 565
 566	list_for_each_entry(ch, &asoc->outqueue.out_chunk_list, list)
 567		if (ch->transport == peer)
 568			ch->transport = NULL;
 569
 570	asoc->peer.transport_count--;
 571
 
 572	sctp_transport_free(peer);
 573}
 574
 575/* Add a transport address to an association.  */
 576struct sctp_transport *sctp_assoc_add_peer(struct sctp_association *asoc,
 577					   const union sctp_addr *addr,
 578					   const gfp_t gfp,
 579					   const int peer_state)
 580{
 581	struct net *net = sock_net(asoc->base.sk);
 582	struct sctp_transport *peer;
 583	struct sctp_sock *sp;
 584	unsigned short port;
 585
 586	sp = sctp_sk(asoc->base.sk);
 587
 588	/* AF_INET and AF_INET6 share common port field. */
 589	port = ntohs(addr->v4.sin_port);
 590
 591	pr_debug("%s: association:%p addr:%pISpc state:%d\n", __func__,
 592		 asoc, &addr->sa, peer_state);
 593
 594	/* Set the port if it has not been set yet.  */
 595	if (0 == asoc->peer.port)
 596		asoc->peer.port = port;
 597
 598	/* Check to see if this is a duplicate. */
 599	peer = sctp_assoc_lookup_paddr(asoc, addr);
 600	if (peer) {
 601		/* An UNKNOWN state is only set on transports added by
 602		 * user in sctp_connectx() call.  Such transports should be
 603		 * considered CONFIRMED per RFC 4960, Section 5.4.
 604		 */
 605		if (peer->state == SCTP_UNKNOWN) {
 606			peer->state = SCTP_ACTIVE;
 607		}
 608		return peer;
 609	}
 610
 611	peer = sctp_transport_new(net, addr, gfp);
 612	if (!peer)
 613		return NULL;
 614
 615	sctp_transport_set_owner(peer, asoc);
 616
 617	/* Initialize the peer's heartbeat interval based on the
 618	 * association configured value.
 619	 */
 620	peer->hbinterval = asoc->hbinterval;
 
 
 
 621
 622	/* Set the path max_retrans.  */
 623	peer->pathmaxrxt = asoc->pathmaxrxt;
 624
 625	/* And the partial failure retrans threshold */
 626	peer->pf_retrans = asoc->pf_retrans;
 
 
 627
 628	/* Initialize the peer's SACK delay timeout based on the
 629	 * association configured value.
 630	 */
 631	peer->sackdelay = asoc->sackdelay;
 632	peer->sackfreq = asoc->sackfreq;
 633
 634	if (addr->sa.sa_family == AF_INET6) {
 635		__be32 info = addr->v6.sin6_flowinfo;
 636
 637		if (info) {
 638			peer->flowlabel = ntohl(info & IPV6_FLOWLABEL_MASK);
 639			peer->flowlabel |= SCTP_FLOWLABEL_SET_MASK;
 640		} else {
 641			peer->flowlabel = asoc->flowlabel;
 642		}
 643	}
 644	peer->dscp = asoc->dscp;
 645
 646	/* Enable/disable heartbeat, SACK delay, and path MTU discovery
 647	 * based on association setting.
 648	 */
 649	peer->param_flags = asoc->param_flags;
 650
 651	/* Initialize the pmtu of the transport. */
 652	sctp_transport_route(peer, NULL, sp);
 653
 654	/* If this is the first transport addr on this association,
 655	 * initialize the association PMTU to the peer's PMTU.
 656	 * If not and the current association PMTU is higher than the new
 657	 * peer's PMTU, reset the association PMTU to the new peer's PMTU.
 658	 */
 659	sctp_assoc_set_pmtu(asoc, asoc->pathmtu ?
 660				  min_t(int, peer->pathmtu, asoc->pathmtu) :
 661				  peer->pathmtu);
 662
 663	peer->pmtu_pending = 0;
 664
 665	/* The asoc->peer.port might not be meaningful yet, but
 666	 * initialize the packet structure anyway.
 667	 */
 668	sctp_packet_init(&peer->packet, peer, asoc->base.bind_addr.port,
 669			 asoc->peer.port);
 670
 671	/* 7.2.1 Slow-Start
 672	 *
 673	 * o The initial cwnd before DATA transmission or after a sufficiently
 674	 *   long idle period MUST be set to
 675	 *      min(4*MTU, max(2*MTU, 4380 bytes))
 676	 *
 677	 * o The initial value of ssthresh MAY be arbitrarily high
 678	 *   (for example, implementations MAY use the size of the
 679	 *   receiver advertised window).
 680	 */
 681	peer->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
 682
 683	/* At this point, we may not have the receiver's advertised window,
 684	 * so initialize ssthresh to the default value and it will be set
 685	 * later when we process the INIT.
 686	 */
 687	peer->ssthresh = SCTP_DEFAULT_MAXWINDOW;
 688
 689	peer->partial_bytes_acked = 0;
 690	peer->flight_size = 0;
 691	peer->burst_limited = 0;
 692
 693	/* Set the transport's RTO.initial value */
 694	peer->rto = asoc->rto_initial;
 695	sctp_max_rto(asoc, peer);
 696
 697	/* Set the peer's active state. */
 698	peer->state = peer_state;
 699
 700	/* Add this peer into the transport hashtable */
 701	if (sctp_hash_transport(peer)) {
 702		sctp_transport_free(peer);
 703		return NULL;
 704	}
 705
 
 
 706	/* Attach the remote transport to our asoc.  */
 707	list_add_tail_rcu(&peer->transports, &asoc->peer.transport_addr_list);
 708	asoc->peer.transport_count++;
 709
 
 
 710	/* If we do not yet have a primary path, set one.  */
 711	if (!asoc->peer.primary_path) {
 712		sctp_assoc_set_primary(asoc, peer);
 713		asoc->peer.retran_path = peer;
 714	}
 715
 716	if (asoc->peer.active_path == asoc->peer.retran_path &&
 717	    peer->state != SCTP_UNCONFIRMED) {
 718		asoc->peer.retran_path = peer;
 719	}
 720
 721	return peer;
 722}
 723
 724/* Delete a transport address from an association.  */
 725void sctp_assoc_del_peer(struct sctp_association *asoc,
 726			 const union sctp_addr *addr)
 727{
 728	struct list_head	*pos;
 729	struct list_head	*temp;
 730	struct sctp_transport	*transport;
 731
 732	list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
 733		transport = list_entry(pos, struct sctp_transport, transports);
 734		if (sctp_cmp_addr_exact(addr, &transport->ipaddr)) {
 735			/* Do book keeping for removing the peer and free it. */
 736			sctp_assoc_rm_peer(asoc, transport);
 737			break;
 738		}
 739	}
 740}
 741
 742/* Lookup a transport by address. */
 743struct sctp_transport *sctp_assoc_lookup_paddr(
 744					const struct sctp_association *asoc,
 745					const union sctp_addr *address)
 746{
 747	struct sctp_transport *t;
 748
 749	/* Cycle through all transports searching for a peer address. */
 750
 751	list_for_each_entry(t, &asoc->peer.transport_addr_list,
 752			transports) {
 753		if (sctp_cmp_addr_exact(address, &t->ipaddr))
 754			return t;
 755	}
 756
 757	return NULL;
 758}
 759
 760/* Remove all transports except a give one */
 761void sctp_assoc_del_nonprimary_peers(struct sctp_association *asoc,
 762				     struct sctp_transport *primary)
 763{
 764	struct sctp_transport	*temp;
 765	struct sctp_transport	*t;
 766
 767	list_for_each_entry_safe(t, temp, &asoc->peer.transport_addr_list,
 768				 transports) {
 769		/* if the current transport is not the primary one, delete it */
 770		if (t != primary)
 771			sctp_assoc_rm_peer(asoc, t);
 772	}
 773}
 774
 775/* Engage in transport control operations.
 776 * Mark the transport up or down and send a notification to the user.
 777 * Select and update the new active and retran paths.
 778 */
 779void sctp_assoc_control_transport(struct sctp_association *asoc,
 780				  struct sctp_transport *transport,
 781				  enum sctp_transport_cmd command,
 782				  sctp_sn_error_t error)
 783{
 784	struct sctp_ulpevent *event;
 785	struct sockaddr_storage addr;
 786	int spc_state = 0;
 787	bool ulp_notify = true;
 788
 789	/* Record the transition on the transport.  */
 790	switch (command) {
 791	case SCTP_TRANSPORT_UP:
 792		/* If we are moving from UNCONFIRMED state due
 793		 * to heartbeat success, report the SCTP_ADDR_CONFIRMED
 794		 * state to the user, otherwise report SCTP_ADDR_AVAILABLE.
 795		 */
 796		if (SCTP_UNCONFIRMED == transport->state &&
 797		    SCTP_HEARTBEAT_SUCCESS == error)
 798			spc_state = SCTP_ADDR_CONFIRMED;
 799		else
 800			spc_state = SCTP_ADDR_AVAILABLE;
 801		/* Don't inform ULP about transition from PF to
 802		 * active state and set cwnd to 1 MTU, see SCTP
 803		 * Quick failover draft section 5.1, point 5
 804		 */
 805		if (transport->state == SCTP_PF) {
 806			ulp_notify = false;
 807			transport->cwnd = asoc->pathmtu;
 808		}
 
 
 809		transport->state = SCTP_ACTIVE;
 
 810		break;
 811
 812	case SCTP_TRANSPORT_DOWN:
 813		/* If the transport was never confirmed, do not transition it
 814		 * to inactive state.  Also, release the cached route since
 815		 * there may be a better route next time.
 816		 */
 817		if (transport->state != SCTP_UNCONFIRMED)
 818			transport->state = SCTP_INACTIVE;
 819		else {
 
 
 820			sctp_transport_dst_release(transport);
 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			asoc->stream.si->enqueue_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	refcount_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 (refcount_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/* Do delayed input processing.  This is scheduled by sctp_rcv(). */
 975static void sctp_assoc_bh_rcv(struct work_struct *work)
 976{
 977	struct sctp_association *asoc =
 978		container_of(work, struct sctp_association,
 979			     base.inqueue.immediate);
 980	struct net *net = sock_net(asoc->base.sk);
 981	union sctp_subtype subtype;
 982	struct sctp_endpoint *ep;
 983	struct sctp_chunk *chunk;
 984	struct sctp_inq *inqueue;
 985	int first_time = 1;	/* is this the first time through the loop */
 986	int error = 0;
 987	int state;
 988
 989	/* The association should be held so we should be safe. */
 990	ep = asoc->ep;
 991
 992	inqueue = &asoc->base.inqueue;
 993	sctp_association_hold(asoc);
 994	while (NULL != (chunk = sctp_inq_pop(inqueue))) {
 995		state = asoc->state;
 996		subtype = SCTP_ST_CHUNK(chunk->chunk_hdr->type);
 997
 998		/* If the first chunk in the packet is AUTH, do special
 999		 * processing specified in Section 6.3 of SCTP-AUTH spec
1000		 */
1001		if (first_time && subtype.chunk == SCTP_CID_AUTH) {
1002			struct sctp_chunkhdr *next_hdr;
1003
1004			next_hdr = sctp_inq_peek(inqueue);
1005			if (!next_hdr)
1006				goto normal;
1007
1008			/* If the next chunk is COOKIE-ECHO, skip the AUTH
1009			 * chunk while saving a pointer to it so we can do
1010			 * Authentication later (during cookie-echo
1011			 * processing).
1012			 */
1013			if (next_hdr->type == SCTP_CID_COOKIE_ECHO) {
1014				chunk->auth_chunk = skb_clone(chunk->skb,
1015							      GFP_ATOMIC);
1016				chunk->auth = 1;
1017				continue;
1018			}
1019		}
1020
1021normal:
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		if (first_time)
1062			first_time = 0;
1063	}
1064	sctp_association_put(asoc);
1065}
1066
1067/* This routine moves an association from its old sk to a new sk.  */
1068void sctp_assoc_migrate(struct sctp_association *assoc, struct sock *newsk)
1069{
1070	struct sctp_sock *newsp = sctp_sk(newsk);
1071	struct sock *oldsk = assoc->base.sk;
1072
1073	/* Delete the association from the old endpoint's list of
1074	 * associations.
1075	 */
1076	list_del_init(&assoc->asocs);
1077
1078	/* Decrement the backlog value for a TCP-style socket. */
1079	if (sctp_style(oldsk, TCP))
1080		oldsk->sk_ack_backlog--;
1081
1082	/* Release references to the old endpoint and the sock.  */
1083	sctp_endpoint_put(assoc->ep);
1084	sock_put(assoc->base.sk);
1085
1086	/* Get a reference to the new endpoint.  */
1087	assoc->ep = newsp->ep;
1088	sctp_endpoint_hold(assoc->ep);
1089
1090	/* Get a reference to the new sock.  */
1091	assoc->base.sk = newsk;
1092	sock_hold(assoc->base.sk);
1093
1094	/* Add the association to the new endpoint's list of associations.  */
1095	sctp_endpoint_add_asoc(newsp->ep, assoc);
1096}
1097
1098/* Update an association (possibly from unexpected COOKIE-ECHO processing).  */
1099int sctp_assoc_update(struct sctp_association *asoc,
1100		      struct sctp_association *new)
1101{
1102	struct sctp_transport *trans;
1103	struct list_head *pos, *temp;
1104
1105	/* Copy in new parameters of peer. */
1106	asoc->c = new->c;
1107	asoc->peer.rwnd = new->peer.rwnd;
1108	asoc->peer.sack_needed = new->peer.sack_needed;
1109	asoc->peer.auth_capable = new->peer.auth_capable;
1110	asoc->peer.i = new->peer.i;
1111
1112	if (!sctp_tsnmap_init(&asoc->peer.tsn_map, SCTP_TSN_MAP_INITIAL,
1113			      asoc->peer.i.initial_tsn, GFP_ATOMIC))
1114		return -ENOMEM;
1115
1116	/* Remove any peer addresses not present in the new association. */
1117	list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
1118		trans = list_entry(pos, struct sctp_transport, transports);
1119		if (!sctp_assoc_lookup_paddr(new, &trans->ipaddr)) {
1120			sctp_assoc_rm_peer(asoc, trans);
1121			continue;
1122		}
1123
1124		if (asoc->state >= SCTP_STATE_ESTABLISHED)
1125			sctp_transport_reset(trans);
1126	}
1127
1128	/* If the case is A (association restart), use
1129	 * initial_tsn as next_tsn. If the case is B, use
1130	 * current next_tsn in case data sent to peer
1131	 * has been discarded and needs retransmission.
1132	 */
1133	if (asoc->state >= SCTP_STATE_ESTABLISHED) {
1134		asoc->next_tsn = new->next_tsn;
1135		asoc->ctsn_ack_point = new->ctsn_ack_point;
1136		asoc->adv_peer_ack_point = new->adv_peer_ack_point;
1137
1138		/* Reinitialize SSN for both local streams
1139		 * and peer's streams.
1140		 */
1141		sctp_stream_clear(&asoc->stream);
1142
1143		/* Flush the ULP reassembly and ordered queue.
1144		 * Any data there will now be stale and will
1145		 * cause problems.
1146		 */
1147		sctp_ulpq_flush(&asoc->ulpq);
1148
1149		/* reset the overall association error count so
1150		 * that the restarted association doesn't get torn
1151		 * down on the next retransmission timer.
1152		 */
1153		asoc->overall_error_count = 0;
1154
1155	} else {
1156		/* Add any peer addresses from the new association. */
1157		list_for_each_entry(trans, &new->peer.transport_addr_list,
1158				    transports)
1159			if (!sctp_assoc_lookup_paddr(asoc, &trans->ipaddr) &&
1160			    !sctp_assoc_add_peer(asoc, &trans->ipaddr,
1161						 GFP_ATOMIC, trans->state))
1162				return -ENOMEM;
1163
1164		asoc->ctsn_ack_point = asoc->next_tsn - 1;
1165		asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
1166
1167		if (sctp_state(asoc, COOKIE_WAIT))
1168			sctp_stream_update(&asoc->stream, &new->stream);
1169
1170		/* get a new assoc id if we don't have one yet. */
1171		if (sctp_assoc_set_id(asoc, GFP_ATOMIC))
1172			return -ENOMEM;
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	return 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
1395void sctp_assoc_update_frag_point(struct sctp_association *asoc)
1396{
1397	int frag = sctp_mtu_payload(sctp_sk(asoc->base.sk), asoc->pathmtu,
1398				    sctp_datachk_len(&asoc->stream));
1399
1400	if (asoc->user_frag)
1401		frag = min_t(int, frag, asoc->user_frag);
1402
1403	frag = min_t(int, frag, SCTP_MAX_CHUNK_LEN -
1404				sctp_datachk_len(&asoc->stream));
1405
1406	asoc->frag_point = SCTP_TRUNC4(frag);
1407}
1408
1409void sctp_assoc_set_pmtu(struct sctp_association *asoc, __u32 pmtu)
1410{
1411	if (asoc->pathmtu != pmtu) {
1412		asoc->pathmtu = pmtu;
1413		sctp_assoc_update_frag_point(asoc);
1414	}
1415
1416	pr_debug("%s: asoc:%p, pmtu:%d, frag_point:%d\n", __func__, asoc,
1417		 asoc->pathmtu, asoc->frag_point);
1418}
1419
1420/* Update the association's pmtu and frag_point by going through all the
1421 * transports. This routine is called when a transport's PMTU has changed.
1422 */
1423void sctp_assoc_sync_pmtu(struct sctp_association *asoc)
1424{
1425	struct sctp_transport *t;
1426	__u32 pmtu = 0;
1427
1428	if (!asoc)
1429		return;
1430
1431	/* Get the lowest pmtu of all the transports. */
1432	list_for_each_entry(t, &asoc->peer.transport_addr_list, transports) {
1433		if (t->pmtu_pending && t->dst) {
1434			sctp_transport_update_pmtu(t,
1435						   atomic_read(&t->mtu_info));
1436			t->pmtu_pending = 0;
1437		}
1438		if (!pmtu || (t->pathmtu < pmtu))
1439			pmtu = t->pathmtu;
1440	}
1441
1442	sctp_assoc_set_pmtu(asoc, pmtu);
1443}
1444
1445/* Should we send a SACK to update our peer? */
1446static inline bool sctp_peer_needs_update(struct sctp_association *asoc)
1447{
1448	struct net *net = sock_net(asoc->base.sk);
 
1449	switch (asoc->state) {
1450	case SCTP_STATE_ESTABLISHED:
1451	case SCTP_STATE_SHUTDOWN_PENDING:
1452	case SCTP_STATE_SHUTDOWN_RECEIVED:
1453	case SCTP_STATE_SHUTDOWN_SENT:
1454		if ((asoc->rwnd > asoc->a_rwnd) &&
1455		    ((asoc->rwnd - asoc->a_rwnd) >= max_t(__u32,
1456			   (asoc->base.sk->sk_rcvbuf >> net->sctp.rwnd_upd_shift),
1457			   asoc->pathmtu)))
1458			return true;
1459		break;
1460	default:
1461		break;
1462	}
1463	return false;
1464}
1465
1466/* Increase asoc's rwnd by len and send any window update SACK if needed. */
1467void sctp_assoc_rwnd_increase(struct sctp_association *asoc, unsigned int len)
1468{
1469	struct sctp_chunk *sack;
1470	struct timer_list *timer;
1471
1472	if (asoc->rwnd_over) {
1473		if (asoc->rwnd_over >= len) {
1474			asoc->rwnd_over -= len;
1475		} else {
1476			asoc->rwnd += (len - asoc->rwnd_over);
1477			asoc->rwnd_over = 0;
1478		}
1479	} else {
1480		asoc->rwnd += len;
1481	}
1482
1483	/* If we had window pressure, start recovering it
1484	 * once our rwnd had reached the accumulated pressure
1485	 * threshold.  The idea is to recover slowly, but up
1486	 * to the initial advertised window.
1487	 */
1488	if (asoc->rwnd_press) {
1489		int change = min(asoc->pathmtu, asoc->rwnd_press);
1490		asoc->rwnd += change;
1491		asoc->rwnd_press -= change;
1492	}
1493
1494	pr_debug("%s: asoc:%p rwnd increased by %d to (%u, %u) - %u\n",
1495		 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
1496		 asoc->a_rwnd);
1497
1498	/* Send a window update SACK if the rwnd has increased by at least the
1499	 * minimum of the association's PMTU and half of the receive buffer.
1500	 * The algorithm used is similar to the one described in
1501	 * Section 4.2.3.3 of RFC 1122.
1502	 */
1503	if (sctp_peer_needs_update(asoc)) {
1504		asoc->a_rwnd = asoc->rwnd;
1505
1506		pr_debug("%s: sending window update SACK- asoc:%p rwnd:%u "
1507			 "a_rwnd:%u\n", __func__, asoc, asoc->rwnd,
1508			 asoc->a_rwnd);
1509
1510		sack = sctp_make_sack(asoc);
1511		if (!sack)
1512			return;
1513
1514		asoc->peer.sack_needed = 0;
1515
1516		sctp_outq_tail(&asoc->outqueue, sack, GFP_ATOMIC);
1517
1518		/* Stop the SACK timer.  */
1519		timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
1520		if (del_timer(timer))
1521			sctp_association_put(asoc);
1522	}
1523}
1524
1525/* Decrease asoc's rwnd by len. */
1526void sctp_assoc_rwnd_decrease(struct sctp_association *asoc, unsigned int len)
1527{
1528	int rx_count;
1529	int over = 0;
1530
1531	if (unlikely(!asoc->rwnd || asoc->rwnd_over))
1532		pr_debug("%s: association:%p has asoc->rwnd:%u, "
1533			 "asoc->rwnd_over:%u!\n", __func__, asoc,
1534			 asoc->rwnd, asoc->rwnd_over);
1535
1536	if (asoc->ep->rcvbuf_policy)
1537		rx_count = atomic_read(&asoc->rmem_alloc);
1538	else
1539		rx_count = atomic_read(&asoc->base.sk->sk_rmem_alloc);
1540
1541	/* If we've reached or overflowed our receive buffer, announce
1542	 * a 0 rwnd if rwnd would still be positive.  Store the
1543	 * the potential pressure overflow so that the window can be restored
1544	 * back to original value.
1545	 */
1546	if (rx_count >= asoc->base.sk->sk_rcvbuf)
1547		over = 1;
1548
1549	if (asoc->rwnd >= len) {
1550		asoc->rwnd -= len;
1551		if (over) {
1552			asoc->rwnd_press += asoc->rwnd;
1553			asoc->rwnd = 0;
1554		}
1555	} else {
1556		asoc->rwnd_over += len - asoc->rwnd;
1557		asoc->rwnd = 0;
1558	}
1559
1560	pr_debug("%s: asoc:%p rwnd decreased by %d to (%u, %u, %u)\n",
1561		 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
1562		 asoc->rwnd_press);
1563}
1564
1565/* Build the bind address list for the association based on info from the
1566 * local endpoint and the remote peer.
1567 */
1568int sctp_assoc_set_bind_addr_from_ep(struct sctp_association *asoc,
1569				     enum sctp_scope scope, gfp_t gfp)
1570{
 
1571	int flags;
1572
1573	/* Use scoping rules to determine the subset of addresses from
1574	 * the endpoint.
1575	 */
1576	flags = (PF_INET6 == asoc->base.sk->sk_family) ? SCTP_ADDR6_ALLOWED : 0;
 
 
1577	if (asoc->peer.ipv4_address)
1578		flags |= SCTP_ADDR4_PEERSUPP;
1579	if (asoc->peer.ipv6_address)
1580		flags |= SCTP_ADDR6_PEERSUPP;
1581
1582	return sctp_bind_addr_copy(sock_net(asoc->base.sk),
1583				   &asoc->base.bind_addr,
1584				   &asoc->ep->base.bind_addr,
1585				   scope, gfp, flags);
1586}
1587
1588/* Build the association's bind address list from the cookie.  */
1589int sctp_assoc_set_bind_addr_from_cookie(struct sctp_association *asoc,
1590					 struct sctp_cookie *cookie,
1591					 gfp_t gfp)
1592{
1593	int var_size2 = ntohs(cookie->peer_init->chunk_hdr.length);
 
1594	int var_size3 = cookie->raw_addr_list_len;
1595	__u8 *raw = (__u8 *)cookie->peer_init + var_size2;
1596
1597	return sctp_raw_to_bind_addrs(&asoc->base.bind_addr, raw, var_size3,
1598				      asoc->ep->base.bind_addr.port, gfp);
1599}
1600
1601/* Lookup laddr in the bind address list of an association. */
1602int sctp_assoc_lookup_laddr(struct sctp_association *asoc,
1603			    const union sctp_addr *laddr)
1604{
1605	int found = 0;
1606
1607	if ((asoc->base.bind_addr.port == ntohs(laddr->v4.sin_port)) &&
1608	    sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
1609				 sctp_sk(asoc->base.sk)))
1610		found = 1;
1611
1612	return found;
1613}
1614
1615/* Set an association id for a given association */
1616int sctp_assoc_set_id(struct sctp_association *asoc, gfp_t gfp)
1617{
1618	bool preload = gfpflags_allow_blocking(gfp);
1619	int ret;
1620
1621	/* If the id is already assigned, keep it. */
1622	if (asoc->assoc_id)
1623		return 0;
1624
1625	if (preload)
1626		idr_preload(gfp);
1627	spin_lock_bh(&sctp_assocs_id_lock);
1628	/* 0, 1, 2 are used as SCTP_FUTURE_ASSOC, SCTP_CURRENT_ASSOC and
1629	 * SCTP_ALL_ASSOC, so an available id must be > SCTP_ALL_ASSOC.
1630	 */
1631	ret = idr_alloc_cyclic(&sctp_assocs_id, asoc, SCTP_ALL_ASSOC + 1, 0,
1632			       GFP_NOWAIT);
1633	spin_unlock_bh(&sctp_assocs_id_lock);
1634	if (preload)
1635		idr_preload_end();
1636	if (ret < 0)
1637		return ret;
1638
1639	asoc->assoc_id = (sctp_assoc_t)ret;
1640	return 0;
1641}
1642
1643/* Free the ASCONF queue */
1644static void sctp_assoc_free_asconf_queue(struct sctp_association *asoc)
1645{
1646	struct sctp_chunk *asconf;
1647	struct sctp_chunk *tmp;
1648
1649	list_for_each_entry_safe(asconf, tmp, &asoc->addip_chunk_list, list) {
1650		list_del_init(&asconf->list);
1651		sctp_chunk_free(asconf);
1652	}
1653}
1654
1655/* Free asconf_ack cache */
1656static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc)
1657{
1658	struct sctp_chunk *ack;
1659	struct sctp_chunk *tmp;
1660
1661	list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
1662				transmitted_list) {
1663		list_del_init(&ack->transmitted_list);
1664		sctp_chunk_free(ack);
1665	}
1666}
1667
1668/* Clean up the ASCONF_ACK queue */
1669void sctp_assoc_clean_asconf_ack_cache(const struct sctp_association *asoc)
1670{
1671	struct sctp_chunk *ack;
1672	struct sctp_chunk *tmp;
1673
1674	/* We can remove all the entries from the queue up to
1675	 * the "Peer-Sequence-Number".
1676	 */
1677	list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
1678				transmitted_list) {
1679		if (ack->subh.addip_hdr->serial ==
1680				htonl(asoc->peer.addip_serial))
1681			break;
1682
1683		list_del_init(&ack->transmitted_list);
1684		sctp_chunk_free(ack);
1685	}
1686}
1687
1688/* Find the ASCONF_ACK whose serial number matches ASCONF */
1689struct sctp_chunk *sctp_assoc_lookup_asconf_ack(
1690					const struct sctp_association *asoc,
1691					__be32 serial)
1692{
1693	struct sctp_chunk *ack;
1694
1695	/* Walk through the list of cached ASCONF-ACKs and find the
1696	 * ack chunk whose serial number matches that of the request.
1697	 */
1698	list_for_each_entry(ack, &asoc->asconf_ack_list, transmitted_list) {
1699		if (sctp_chunk_pending(ack))
1700			continue;
1701		if (ack->subh.addip_hdr->serial == serial) {
1702			sctp_chunk_hold(ack);
1703			return ack;
1704		}
1705	}
1706
1707	return NULL;
1708}
1709
1710void sctp_asconf_queue_teardown(struct sctp_association *asoc)
1711{
1712	/* Free any cached ASCONF_ACK chunk. */
1713	sctp_assoc_free_asconf_acks(asoc);
1714
1715	/* Free the ASCONF queue. */
1716	sctp_assoc_free_asconf_queue(asoc);
1717
1718	/* Free any cached ASCONF chunk. */
1719	if (asoc->addip_last_asconf)
1720		sctp_chunk_free(asoc->addip_last_asconf);
1721}