associola.c

   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	asoc->base.net = sock_net(sk);
  68
  69	sctp_endpoint_hold(asoc->ep);
  70	sock_hold(asoc->base.sk);
  71
  72	/* Initialize the common base substructure.  */
  73	asoc->base.type = SCTP_EP_TYPE_ASSOCIATION;
  74
  75	/* Initialize the object handling fields.  */
  76	refcount_set(&asoc->base.refcnt, 1);
  77
  78	/* Initialize the bind addr area.  */
  79	sctp_bind_addr_init(&asoc->base.bind_addr, ep->base.bind_addr.port);
  80
  81	asoc->state = SCTP_STATE_CLOSED;
  82	asoc->cookie_life = ms_to_ktime(sp->assocparams.sasoc_cookie_life);
  83	asoc->user_frag = sp->user_frag;
  84
  85	/* Set the association max_retrans and RTO values from the
  86	 * socket values.
  87	 */
  88	asoc->max_retrans = sp->assocparams.sasoc_asocmaxrxt;
  89	asoc->pf_retrans  = sp->pf_retrans;
  90	asoc->ps_retrans  = sp->ps_retrans;
  91	asoc->pf_expose   = sp->pf_expose;
  92
  93	asoc->rto_initial = msecs_to_jiffies(sp->rtoinfo.srto_initial);
  94	asoc->rto_max = msecs_to_jiffies(sp->rtoinfo.srto_max);
  95	asoc->rto_min = msecs_to_jiffies(sp->rtoinfo.srto_min);
  96
  97	/* Initialize the association's heartbeat interval based on the
  98	 * sock configured value.
  99	 */
 100	asoc->hbinterval = msecs_to_jiffies(sp->hbinterval);
 101	asoc->probe_interval = msecs_to_jiffies(sp->probe_interval);
 102
 103	asoc->encap_port = sp->encap_port;
 104
 105	/* Initialize path max retrans value. */
 106	asoc->pathmaxrxt = sp->pathmaxrxt;
 107
 108	asoc->flowlabel = sp->flowlabel;
 109	asoc->dscp = sp->dscp;
 110
 111	/* Set association default SACK delay */
 112	asoc->sackdelay = msecs_to_jiffies(sp->sackdelay);
 113	asoc->sackfreq = sp->sackfreq;
 114
 115	/* Set the association default flags controlling
 116	 * Heartbeat, SACK delay, and Path MTU Discovery.
 117	 */
 118	asoc->param_flags = sp->param_flags;
 119
 120	/* Initialize the maximum number of new data packets that can be sent
 121	 * in a burst.
 122	 */
 123	asoc->max_burst = sp->max_burst;
 124
 125	asoc->subscribe = sp->subscribe;
 126
 127	/* initialize association timers */
 128	asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_COOKIE] = asoc->rto_initial;
 129	asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_INIT] = asoc->rto_initial;
 130	asoc->timeouts[SCTP_EVENT_TIMEOUT_T2_SHUTDOWN] = asoc->rto_initial;
 131
 132	/* sctpimpguide Section 2.12.2
 133	 * If the 'T5-shutdown-guard' timer is used, it SHOULD be set to the
 134	 * recommended value of 5 times 'RTO.Max'.
 135	 */
 136	asoc->timeouts[SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]
 137		= 5 * asoc->rto_max;
 138
 139	asoc->timeouts[SCTP_EVENT_TIMEOUT_SACK] = asoc->sackdelay;
 140	asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE] =
 141		(unsigned long)sp->autoclose * HZ;
 142
 143	/* Initializes the timers */
 144	for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i)
 145		timer_setup(&asoc->timers[i], sctp_timer_events[i], 0);
 146
 147	/* Pull default initialization values from the sock options.
 148	 * Note: This assumes that the values have already been
 149	 * validated in the sock.
 150	 */
 151	asoc->c.sinit_max_instreams = sp->initmsg.sinit_max_instreams;
 152	asoc->c.sinit_num_ostreams  = sp->initmsg.sinit_num_ostreams;
 153	asoc->max_init_attempts	= sp->initmsg.sinit_max_attempts;
 154
 155	asoc->max_init_timeo =
 156		 msecs_to_jiffies(sp->initmsg.sinit_max_init_timeo);
 157
 158	/* Set the local window size for receive.
 159	 * This is also the rcvbuf space per association.
 160	 * RFC 6 - A SCTP receiver MUST be able to receive a minimum of
 161	 * 1500 bytes in one SCTP packet.
 162	 */
 163	if ((sk->sk_rcvbuf/2) < SCTP_DEFAULT_MINWINDOW)
 164		asoc->rwnd = SCTP_DEFAULT_MINWINDOW;
 165	else
 166		asoc->rwnd = sk->sk_rcvbuf/2;
 167
 168	asoc->a_rwnd = asoc->rwnd;
 169
 170	/* Use my own max window until I learn something better.  */
 171	asoc->peer.rwnd = SCTP_DEFAULT_MAXWINDOW;
 172
 173	/* Initialize the receive memory counter */
 174	atomic_set(&asoc->rmem_alloc, 0);
 175
 176	init_waitqueue_head(&asoc->wait);
 177
 178	asoc->c.my_vtag = sctp_generate_tag(ep);
 179	asoc->c.my_port = ep->base.bind_addr.port;
 180
 181	asoc->c.initial_tsn = sctp_generate_tsn(ep);
 182
 183	asoc->next_tsn = asoc->c.initial_tsn;
 184
 185	asoc->ctsn_ack_point = asoc->next_tsn - 1;
 186	asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
 187	asoc->highest_sacked = asoc->ctsn_ack_point;
 188	asoc->last_cwr_tsn = asoc->ctsn_ack_point;
 189
 190	/* ADDIP Section 4.1 Asconf Chunk Procedures
 191	 *
 192	 * When an endpoint has an ASCONF signaled change to be sent to the
 193	 * remote endpoint it should do the following:
 194	 * ...
 195	 * A2) a serial number should be assigned to the chunk. The serial
 196	 * number SHOULD be a monotonically increasing number. The serial
 197	 * numbers SHOULD be initialized at the start of the
 198	 * association to the same value as the initial TSN.
 199	 */
 200	asoc->addip_serial = asoc->c.initial_tsn;
 201	asoc->strreset_outseq = asoc->c.initial_tsn;
 202
 203	INIT_LIST_HEAD(&asoc->addip_chunk_list);
 204	INIT_LIST_HEAD(&asoc->asconf_ack_list);
 205
 206	/* Make an empty list of remote transport addresses.  */
 207	INIT_LIST_HEAD(&asoc->peer.transport_addr_list);
 208
 209	/* RFC 2960 5.1 Normal Establishment of an Association
 210	 *
 211	 * After the reception of the first data chunk in an
 212	 * association the endpoint must immediately respond with a
 213	 * sack to acknowledge the data chunk.  Subsequent
 214	 * acknowledgements should be done as described in Section
 215	 * 6.2.
 216	 *
 217	 * [We implement this by telling a new association that it
 218	 * already received one packet.]
 219	 */
 220	asoc->peer.sack_needed = 1;
 221	asoc->peer.sack_generation = 1;
 222
 223	/* Create an input queue.  */
 224	sctp_inq_init(&asoc->base.inqueue);
 225	sctp_inq_set_th_handler(&asoc->base.inqueue, sctp_assoc_bh_rcv);
 226
 227	/* Create an output queue.  */
 228	sctp_outq_init(asoc, &asoc->outqueue);
 229
 230	sctp_ulpq_init(&asoc->ulpq, asoc);
 231
 232	if (sctp_stream_init(&asoc->stream, asoc->c.sinit_num_ostreams, 0, gfp))
 233		goto stream_free;
 234
 235	/* Initialize default path MTU. */
 236	asoc->pathmtu = sp->pathmtu;
 237	sctp_assoc_update_frag_point(asoc);
 238
 239	/* Assume that peer would support both address types unless we are
 240	 * told otherwise.
 241	 */
 242	asoc->peer.ipv4_address = 1;
 243	if (asoc->base.sk->sk_family == PF_INET6)
 244		asoc->peer.ipv6_address = 1;
 245	INIT_LIST_HEAD(&asoc->asocs);
 246
 247	asoc->default_stream = sp->default_stream;
 248	asoc->default_ppid = sp->default_ppid;
 249	asoc->default_flags = sp->default_flags;
 250	asoc->default_context = sp->default_context;
 251	asoc->default_timetolive = sp->default_timetolive;
 252	asoc->default_rcv_context = sp->default_rcv_context;
 253
 254	/* AUTH related initializations */
 255	INIT_LIST_HEAD(&asoc->endpoint_shared_keys);
 256	if (sctp_auth_asoc_copy_shkeys(ep, asoc, gfp))
 257		goto stream_free;
 258
 259	asoc->active_key_id = ep->active_key_id;
 260	asoc->strreset_enable = ep->strreset_enable;
 261
 262	/* Save the hmacs and chunks list into this association */
 263	if (ep->auth_hmacs_list)
 264		memcpy(asoc->c.auth_hmacs, ep->auth_hmacs_list,
 265			ntohs(ep->auth_hmacs_list->param_hdr.length));
 266	if (ep->auth_chunk_list)
 267		memcpy(asoc->c.auth_chunks, ep->auth_chunk_list,
 268			ntohs(ep->auth_chunk_list->param_hdr.length));
 269
 270	/* Get the AUTH random number for this association */
 271	p = (struct sctp_paramhdr *)asoc->c.auth_random;
 272	p->type = SCTP_PARAM_RANDOM;
 273	p->length = htons(sizeof(*p) + SCTP_AUTH_RANDOM_LENGTH);
 274	get_random_bytes(p+1, SCTP_AUTH_RANDOM_LENGTH);
 275
 276	return asoc;
 277
 278stream_free:
 279	sctp_stream_free(&asoc->stream);
 280	sock_put(asoc->base.sk);
 281	sctp_endpoint_put(asoc->ep);
 282	return NULL;
 283}
 284
 285/* Allocate and initialize a new association */
 286struct sctp_association *sctp_association_new(const struct sctp_endpoint *ep,
 287					      const struct sock *sk,
 288					      enum sctp_scope scope, gfp_t gfp)
 289{
 290	struct sctp_association *asoc;
 291
 292	asoc = kzalloc(sizeof(*asoc), gfp);
 293	if (!asoc)
 294		goto fail;
 295
 296	if (!sctp_association_init(asoc, ep, sk, scope, gfp))
 297		goto fail_init;
 298
 299	SCTP_DBG_OBJCNT_INC(assoc);
 300
 301	pr_debug("Created asoc %p\n", asoc);
 302
 303	return asoc;
 304
 305fail_init:
 306	kfree(asoc);
 307fail:
 308	return NULL;
 309}
 310
 311/* Free this association if possible.  There may still be users, so
 312 * the actual deallocation may be delayed.
 313 */
 314void sctp_association_free(struct sctp_association *asoc)
 315{
 316	struct sock *sk = asoc->base.sk;
 317	struct sctp_transport *transport;
 318	struct list_head *pos, *temp;
 319	int i;
 320
 321	/* Only real associations count against the endpoint, so
 322	 * don't bother for if this is a temporary association.
 323	 */
 324	if (!list_empty(&asoc->asocs)) {
 325		list_del(&asoc->asocs);
 326
 327		/* Decrement the backlog value for a TCP-style listening
 328		 * socket.
 329		 */
 330		if (sctp_style(sk, TCP) && sctp_sstate(sk, LISTENING))
 331			sk_acceptq_removed(sk);
 332	}
 333
 334	/* Mark as dead, so other users can know this structure is
 335	 * going away.
 336	 */
 337	asoc->base.dead = true;
 338
 339	/* Dispose of any data lying around in the outqueue. */
 340	sctp_outq_free(&asoc->outqueue);
 341
 342	/* Dispose of any pending messages for the upper layer. */
 343	sctp_ulpq_free(&asoc->ulpq);
 344
 345	/* Dispose of any pending chunks on the inqueue. */
 346	sctp_inq_free(&asoc->base.inqueue);
 347
 348	sctp_tsnmap_free(&asoc->peer.tsn_map);
 349
 350	/* Free stream information. */
 351	sctp_stream_free(&asoc->stream);
 352
 353	if (asoc->strreset_chunk)
 354		sctp_chunk_free(asoc->strreset_chunk);
 355
 356	/* Clean up the bound address list. */
 357	sctp_bind_addr_free(&asoc->base.bind_addr);
 358
 359	/* Do we need to go through all of our timers and
 360	 * delete them?   To be safe we will try to delete all, but we
 361	 * should be able to go through and make a guess based
 362	 * on our state.
 363	 */
 364	for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i) {
 365		if (timer_delete(&asoc->timers[i]))
 366			sctp_association_put(asoc);
 367	}
 368
 369	/* Free peer's cached cookie. */
 370	kfree(asoc->peer.cookie);
 371	kfree(asoc->peer.peer_random);
 372	kfree(asoc->peer.peer_chunks);
 373	kfree(asoc->peer.peer_hmacs);
 374
 375	/* Release the transport structures. */
 376	list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
 377		transport = list_entry(pos, struct sctp_transport, transports);
 378		list_del_rcu(pos);
 379		sctp_unhash_transport(transport);
 380		sctp_transport_free(transport);
 381	}
 382
 383	asoc->peer.transport_count = 0;
 384
 385	sctp_asconf_queue_teardown(asoc);
 386
 387	/* Free pending address space being deleted */
 388	kfree(asoc->asconf_addr_del_pending);
 389
 390	/* AUTH - Free the endpoint shared keys */
 391	sctp_auth_destroy_keys(&asoc->endpoint_shared_keys);
 392
 393	/* AUTH - Free the association shared key */
 394	sctp_auth_key_put(asoc->asoc_shared_key);
 395
 396	sctp_association_put(asoc);
 397}
 398
 399/* Cleanup and free up an association. */
 400static void sctp_association_destroy(struct sctp_association *asoc)
 401{
 402	if (unlikely(!asoc->base.dead)) {
 403		WARN(1, "Attempt to destroy undead association %p!\n", asoc);
 404		return;
 405	}
 406
 407	sctp_endpoint_put(asoc->ep);
 408	sock_put(asoc->base.sk);
 409
 410	if (asoc->assoc_id != 0) {
 411		spin_lock_bh(&sctp_assocs_id_lock);
 412		idr_remove(&sctp_assocs_id, asoc->assoc_id);
 413		spin_unlock_bh(&sctp_assocs_id_lock);
 414	}
 415
 416	WARN_ON(atomic_read(&asoc->rmem_alloc));
 417
 418	kfree_rcu(asoc, rcu);
 419	SCTP_DBG_OBJCNT_DEC(assoc);
 420}
 421
 422/* Change the primary destination address for the peer. */
 423void sctp_assoc_set_primary(struct sctp_association *asoc,
 424			    struct sctp_transport *transport)
 425{
 426	int changeover = 0;
 427
 428	/* it's a changeover only if we already have a primary path
 429	 * that we are changing
 430	 */
 431	if (asoc->peer.primary_path != NULL &&
 432	    asoc->peer.primary_path != transport)
 433		changeover = 1 ;
 434
 435	asoc->peer.primary_path = transport;
 436	sctp_ulpevent_notify_peer_addr_change(transport,
 437					      SCTP_ADDR_MADE_PRIM, 0);
 438
 439	/* Set a default msg_name for events. */
 440	memcpy(&asoc->peer.primary_addr, &transport->ipaddr,
 441	       sizeof(union sctp_addr));
 442
 443	/* If the primary path is changing, assume that the
 444	 * user wants to use this new path.
 445	 */
 446	if ((transport->state == SCTP_ACTIVE) ||
 447	    (transport->state == SCTP_UNKNOWN))
 448		asoc->peer.active_path = transport;
 449
 450	/*
 451	 * SFR-CACC algorithm:
 452	 * Upon the receipt of a request to change the primary
 453	 * destination address, on the data structure for the new
 454	 * primary destination, the sender MUST do the following:
 455	 *
 456	 * 1) If CHANGEOVER_ACTIVE is set, then there was a switch
 457	 * to this destination address earlier. The sender MUST set
 458	 * CYCLING_CHANGEOVER to indicate that this switch is a
 459	 * double switch to the same destination address.
 460	 *
 461	 * Really, only bother is we have data queued or outstanding on
 462	 * the association.
 463	 */
 464	if (!asoc->outqueue.outstanding_bytes && !asoc->outqueue.out_qlen)
 465		return;
 466
 467	if (transport->cacc.changeover_active)
 468		transport->cacc.cycling_changeover = changeover;
 469
 470	/* 2) The sender MUST set CHANGEOVER_ACTIVE to indicate that
 471	 * a changeover has occurred.
 472	 */
 473	transport->cacc.changeover_active = changeover;
 474
 475	/* 3) The sender MUST store the next TSN to be sent in
 476	 * next_tsn_at_change.
 477	 */
 478	transport->cacc.next_tsn_at_change = asoc->next_tsn;
 479}
 480
 481/* Remove a transport from an association.  */
 482void sctp_assoc_rm_peer(struct sctp_association *asoc,
 483			struct sctp_transport *peer)
 484{
 485	struct sctp_transport *transport;
 486	struct list_head *pos;
 487	struct sctp_chunk *ch;
 488
 489	pr_debug("%s: association:%p addr:%pISpc\n",
 490		 __func__, asoc, &peer->ipaddr.sa);
 491
 492	/* If we are to remove the current retran_path, update it
 493	 * to the next peer before removing this peer from the list.
 494	 */
 495	if (asoc->peer.retran_path == peer)
 496		sctp_assoc_update_retran_path(asoc);
 497
 498	/* Remove this peer from the list. */
 499	list_del_rcu(&peer->transports);
 500	/* Remove this peer from the transport hashtable */
 501	sctp_unhash_transport(peer);
 502
 503	/* Get the first transport of asoc. */
 504	pos = asoc->peer.transport_addr_list.next;
 505	transport = list_entry(pos, struct sctp_transport, transports);
 506
 507	/* Update any entries that match the peer to be deleted. */
 508	if (asoc->peer.primary_path == peer)
 509		sctp_assoc_set_primary(asoc, transport);
 510	if (asoc->peer.active_path == peer)
 511		asoc->peer.active_path = transport;
 512	if (asoc->peer.retran_path == peer)
 513		asoc->peer.retran_path = transport;
 514	if (asoc->peer.last_data_from == peer)
 515		asoc->peer.last_data_from = transport;
 516
 517	if (asoc->strreset_chunk &&
 518	    asoc->strreset_chunk->transport == peer) {
 519		asoc->strreset_chunk->transport = transport;
 520		sctp_transport_reset_reconf_timer(transport);
 521	}
 522
 523	/* If we remove the transport an INIT was last sent to, set it to
 524	 * NULL. Combined with the update of the retran path above, this
 525	 * will cause the next INIT to be sent to the next available
 526	 * transport, maintaining the cycle.
 527	 */
 528	if (asoc->init_last_sent_to == peer)
 529		asoc->init_last_sent_to = NULL;
 530
 531	/* If we remove the transport an SHUTDOWN was last sent to, set it
 532	 * to NULL. Combined with the update of the retran path above, this
 533	 * will cause the next SHUTDOWN to be sent to the next available
 534	 * transport, maintaining the cycle.
 535	 */
 536	if (asoc->shutdown_last_sent_to == peer)
 537		asoc->shutdown_last_sent_to = NULL;
 538
 539	/* If we remove the transport an ASCONF was last sent to, set it to
 540	 * NULL.
 541	 */
 542	if (asoc->addip_last_asconf &&
 543	    asoc->addip_last_asconf->transport == peer)
 544		asoc->addip_last_asconf->transport = NULL;
 545
 546	/* If we have something on the transmitted list, we have to
 547	 * save it off.  The best place is the active path.
 548	 */
 549	if (!list_empty(&peer->transmitted)) {
 550		struct sctp_transport *active = asoc->peer.active_path;
 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	list_for_each_entry(ch, &asoc->outqueue.out_chunk_list, list)
 573		if (ch->transport == peer)
 574			ch->transport = NULL;
 575
 576	asoc->peer.transport_count--;
 577
 578	sctp_ulpevent_notify_peer_addr_change(peer, SCTP_ADDR_REMOVED, 0);
 579	sctp_transport_free(peer);
 580}
 581
 582/* Add a transport address to an association.  */
 583struct sctp_transport *sctp_assoc_add_peer(struct sctp_association *asoc,
 584					   const union sctp_addr *addr,
 585					   const gfp_t gfp,
 586					   const int peer_state)
 587{
 588	struct sctp_transport *peer;
 589	struct sctp_sock *sp;
 590	unsigned short port;
 591
 592	sp = sctp_sk(asoc->base.sk);
 593
 594	/* AF_INET and AF_INET6 share common port field. */
 595	port = ntohs(addr->v4.sin_port);
 596
 597	pr_debug("%s: association:%p addr:%pISpc state:%d\n", __func__,
 598		 asoc, &addr->sa, peer_state);
 599
 600	/* Set the port if it has not been set yet.  */
 601	if (0 == asoc->peer.port)
 602		asoc->peer.port = port;
 603
 604	/* Check to see if this is a duplicate. */
 605	peer = sctp_assoc_lookup_paddr(asoc, addr);
 606	if (peer) {
 607		/* An UNKNOWN state is only set on transports added by
 608		 * user in sctp_connectx() call.  Such transports should be
 609		 * considered CONFIRMED per RFC 4960, Section 5.4.
 610		 */
 611		if (peer->state == SCTP_UNKNOWN) {
 612			peer->state = SCTP_ACTIVE;
 613		}
 614		return peer;
 615	}
 616
 617	peer = sctp_transport_new(asoc->base.net, addr, gfp);
 618	if (!peer)
 619		return NULL;
 620
 621	sctp_transport_set_owner(peer, asoc);
 622
 623	/* Initialize the peer's heartbeat interval based on the
 624	 * association configured value.
 625	 */
 626	peer->hbinterval = asoc->hbinterval;
 627	peer->probe_interval = asoc->probe_interval;
 628
 629	peer->encap_port = asoc->encap_port;
 630
 631	/* Set the path max_retrans.  */
 632	peer->pathmaxrxt = asoc->pathmaxrxt;
 633
 634	/* And the partial failure retrans threshold */
 635	peer->pf_retrans = asoc->pf_retrans;
 636	/* And the primary path switchover retrans threshold */
 637	peer->ps_retrans = asoc->ps_retrans;
 638
 639	/* Initialize the peer's SACK delay timeout based on the
 640	 * association configured value.
 641	 */
 642	peer->sackdelay = asoc->sackdelay;
 643	peer->sackfreq = asoc->sackfreq;
 644
 645	if (addr->sa.sa_family == AF_INET6) {
 646		__be32 info = addr->v6.sin6_flowinfo;
 647
 648		if (info) {
 649			peer->flowlabel = ntohl(info & IPV6_FLOWLABEL_MASK);
 650			peer->flowlabel |= SCTP_FLOWLABEL_SET_MASK;
 651		} else {
 652			peer->flowlabel = asoc->flowlabel;
 653		}
 654	}
 655	peer->dscp = asoc->dscp;
 656
 657	/* Enable/disable heartbeat, SACK delay, and path MTU discovery
 658	 * based on association setting.
 659	 */
 660	peer->param_flags = asoc->param_flags;
 661
 662	/* Initialize the pmtu of the transport. */
 663	sctp_transport_route(peer, NULL, sp);
 664
 665	/* If this is the first transport addr on this association,
 666	 * initialize the association PMTU to the peer's PMTU.
 667	 * If not and the current association PMTU is higher than the new
 668	 * peer's PMTU, reset the association PMTU to the new peer's PMTU.
 669	 */
 670	sctp_assoc_set_pmtu(asoc, asoc->pathmtu ?
 671				  min_t(int, peer->pathmtu, asoc->pathmtu) :
 672				  peer->pathmtu);
 673
 674	peer->pmtu_pending = 0;
 675
 676	/* The asoc->peer.port might not be meaningful yet, but
 677	 * initialize the packet structure anyway.
 678	 */
 679	sctp_packet_init(&peer->packet, peer, asoc->base.bind_addr.port,
 680			 asoc->peer.port);
 681
 682	/* 7.2.1 Slow-Start
 683	 *
 684	 * o The initial cwnd before DATA transmission or after a sufficiently
 685	 *   long idle period MUST be set to
 686	 *      min(4*MTU, max(2*MTU, 4380 bytes))
 687	 *
 688	 * o The initial value of ssthresh MAY be arbitrarily high
 689	 *   (for example, implementations MAY use the size of the
 690	 *   receiver advertised window).
 691	 */
 692	peer->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
 693
 694	/* At this point, we may not have the receiver's advertised window,
 695	 * so initialize ssthresh to the default value and it will be set
 696	 * later when we process the INIT.
 697	 */
 698	peer->ssthresh = SCTP_DEFAULT_MAXWINDOW;
 699
 700	peer->partial_bytes_acked = 0;
 701	peer->flight_size = 0;
 702	peer->burst_limited = 0;
 703
 704	/* Set the transport's RTO.initial value */
 705	peer->rto = asoc->rto_initial;
 706	sctp_max_rto(asoc, peer);
 707
 708	/* Set the peer's active state. */
 709	peer->state = peer_state;
 710
 711	/* Add this peer into the transport hashtable */
 712	if (sctp_hash_transport(peer)) {
 713		sctp_transport_free(peer);
 714		return NULL;
 715	}
 716
 717	sctp_transport_pl_reset(peer);
 718
 719	/* Attach the remote transport to our asoc.  */
 720	list_add_tail_rcu(&peer->transports, &asoc->peer.transport_addr_list);
 721	asoc->peer.transport_count++;
 722
 723	sctp_ulpevent_notify_peer_addr_change(peer, SCTP_ADDR_ADDED, 0);
 724
 725	/* If we do not yet have a primary path, set one.  */
 726	if (!asoc->peer.primary_path) {
 727		sctp_assoc_set_primary(asoc, peer);
 728		asoc->peer.retran_path = peer;
 729	}
 730
 731	if (asoc->peer.active_path == asoc->peer.retran_path &&
 732	    peer->state != SCTP_UNCONFIRMED) {
 733		asoc->peer.retran_path = peer;
 734	}
 735
 736	return peer;
 737}
 738
 739/* Lookup a transport by address. */
 740struct sctp_transport *sctp_assoc_lookup_paddr(
 741					const struct sctp_association *asoc,
 742					const union sctp_addr *address)
 743{
 744	struct sctp_transport *t;
 745
 746	/* Cycle through all transports searching for a peer address. */
 747
 748	list_for_each_entry(t, &asoc->peer.transport_addr_list,
 749			transports) {
 750		if (sctp_cmp_addr_exact(address, &t->ipaddr))
 751			return t;
 752	}
 753
 754	return NULL;
 755}
 756
 757/* Remove all transports except a give one */
 758void sctp_assoc_del_nonprimary_peers(struct sctp_association *asoc,
 759				     struct sctp_transport *primary)
 760{
 761	struct sctp_transport	*temp;
 762	struct sctp_transport	*t;
 763
 764	list_for_each_entry_safe(t, temp, &asoc->peer.transport_addr_list,
 765				 transports) {
 766		/* if the current transport is not the primary one, delete it */
 767		if (t != primary)
 768			sctp_assoc_rm_peer(asoc, t);
 769	}
 770}
 771
 772/* Engage in transport control operations.
 773 * Mark the transport up or down and send a notification to the user.
 774 * Select and update the new active and retran paths.
 775 */
 776void sctp_assoc_control_transport(struct sctp_association *asoc,
 777				  struct sctp_transport *transport,
 778				  enum sctp_transport_cmd command,
 779				  sctp_sn_error_t error)
 780{
 781	int spc_state = SCTP_ADDR_AVAILABLE;
 782	bool ulp_notify = true;
 783
 784	/* Record the transition on the transport.  */
 785	switch (command) {
 786	case SCTP_TRANSPORT_UP:
 787		/* If we are moving from UNCONFIRMED state due
 788		 * to heartbeat success, report the SCTP_ADDR_CONFIRMED
 789		 * state to the user, otherwise report SCTP_ADDR_AVAILABLE.
 790		 */
 791		if (transport->state == SCTP_PF &&
 792		    asoc->pf_expose != SCTP_PF_EXPOSE_ENABLE)
 793			ulp_notify = false;
 794		else if (transport->state == SCTP_UNCONFIRMED &&
 795			 error == SCTP_HEARTBEAT_SUCCESS)
 796			spc_state = SCTP_ADDR_CONFIRMED;
 797
 798		transport->state = SCTP_ACTIVE;
 799		sctp_transport_pl_reset(transport);
 800		break;
 801
 802	case SCTP_TRANSPORT_DOWN:
 803		/* If the transport was never confirmed, do not transition it
 804		 * to inactive state.  Also, release the cached route since
 805		 * there may be a better route next time.
 806		 */
 807		if (transport->state != SCTP_UNCONFIRMED) {
 808			transport->state = SCTP_INACTIVE;
 809			sctp_transport_pl_reset(transport);
 810			spc_state = SCTP_ADDR_UNREACHABLE;
 811		} else {
 812			sctp_transport_dst_release(transport);
 813			ulp_notify = false;
 814		}
 815		break;
 816
 817	case SCTP_TRANSPORT_PF:
 818		transport->state = SCTP_PF;
 819		if (asoc->pf_expose != SCTP_PF_EXPOSE_ENABLE)
 820			ulp_notify = false;
 821		else
 822			spc_state = SCTP_ADDR_POTENTIALLY_FAILED;
 823		break;
 824
 825	default:
 826		return;
 827	}
 828
 829	/* Generate and send a SCTP_PEER_ADDR_CHANGE notification
 830	 * to the user.
 831	 */
 832	if (ulp_notify)
 833		sctp_ulpevent_notify_peer_addr_change(transport,
 834						      spc_state, error);
 835
 836	/* Select new active and retran paths. */
 837	sctp_select_active_and_retran_path(asoc);
 838}
 839
 840/* Hold a reference to an association. */
 841void sctp_association_hold(struct sctp_association *asoc)
 842{
 843	refcount_inc(&asoc->base.refcnt);
 844}
 845
 846/* Release a reference to an association and cleanup
 847 * if there are no more references.
 848 */
 849void sctp_association_put(struct sctp_association *asoc)
 850{
 851	if (refcount_dec_and_test(&asoc->base.refcnt))
 852		sctp_association_destroy(asoc);
 853}
 854
 855/* Allocate the next TSN, Transmission Sequence Number, for the given
 856 * association.
 857 */
 858__u32 sctp_association_get_next_tsn(struct sctp_association *asoc)
 859{
 860	/* From Section 1.6 Serial Number Arithmetic:
 861	 * Transmission Sequence Numbers wrap around when they reach
 862	 * 2**32 - 1.  That is, the next TSN a DATA chunk MUST use
 863	 * after transmitting TSN = 2*32 - 1 is TSN = 0.
 864	 */
 865	__u32 retval = asoc->next_tsn;
 866	asoc->next_tsn++;
 867	asoc->unack_data++;
 868
 869	return retval;
 870}
 871
 872/* Compare two addresses to see if they match.  Wildcard addresses
 873 * only match themselves.
 874 */
 875int sctp_cmp_addr_exact(const union sctp_addr *ss1,
 876			const union sctp_addr *ss2)
 877{
 878	struct sctp_af *af;
 879
 880	af = sctp_get_af_specific(ss1->sa.sa_family);
 881	if (unlikely(!af))
 882		return 0;
 883
 884	return af->cmp_addr(ss1, ss2);
 885}
 886
 887/* Return an ecne chunk to get prepended to a packet.
 888 * Note:  We are sly and return a shared, prealloced chunk.  FIXME:
 889 * No we don't, but we could/should.
 890 */
 891struct sctp_chunk *sctp_get_ecne_prepend(struct sctp_association *asoc)
 892{
 893	if (!asoc->need_ecne)
 894		return NULL;
 895
 896	/* Send ECNE if needed.
 897	 * Not being able to allocate a chunk here is not deadly.
 898	 */
 899	return sctp_make_ecne(asoc, asoc->last_ecne_tsn);
 900}
 901
 902/*
 903 * Find which transport this TSN was sent on.
 904 */
 905struct sctp_transport *sctp_assoc_lookup_tsn(struct sctp_association *asoc,
 906					     __u32 tsn)
 907{
 908	struct sctp_transport *active;
 909	struct sctp_transport *match;
 910	struct sctp_transport *transport;
 911	struct sctp_chunk *chunk;
 912	__be32 key = htonl(tsn);
 913
 914	match = NULL;
 915
 916	/*
 917	 * FIXME: In general, find a more efficient data structure for
 918	 * searching.
 919	 */
 920
 921	/*
 922	 * The general strategy is to search each transport's transmitted
 923	 * list.   Return which transport this TSN lives on.
 924	 *
 925	 * Let's be hopeful and check the active_path first.
 926	 * Another optimization would be to know if there is only one
 927	 * outbound path and not have to look for the TSN at all.
 928	 *
 929	 */
 930
 931	active = asoc->peer.active_path;
 932
 933	list_for_each_entry(chunk, &active->transmitted,
 934			transmitted_list) {
 935
 936		if (key == chunk->subh.data_hdr->tsn) {
 937			match = active;
 938			goto out;
 939		}
 940	}
 941
 942	/* If not found, go search all the other transports. */
 943	list_for_each_entry(transport, &asoc->peer.transport_addr_list,
 944			transports) {
 945
 946		if (transport == active)
 947			continue;
 948		list_for_each_entry(chunk, &transport->transmitted,
 949				transmitted_list) {
 950			if (key == chunk->subh.data_hdr->tsn) {
 951				match = transport;
 952				goto out;
 953			}
 954		}
 955	}
 956out:
 957	return match;
 958}
 959
 960/* Do delayed input processing.  This is scheduled by sctp_rcv(). */
 961static void sctp_assoc_bh_rcv(struct work_struct *work)
 962{
 963	struct sctp_association *asoc =
 964		container_of(work, struct sctp_association,
 965			     base.inqueue.immediate);
 966	struct net *net = asoc->base.net;
 967	union sctp_subtype subtype;
 968	struct sctp_endpoint *ep;
 969	struct sctp_chunk *chunk;
 970	struct sctp_inq *inqueue;
 971	int first_time = 1;	/* is this the first time through the loop */
 972	int error = 0;
 973	int state;
 974
 975	/* The association should be held so we should be safe. */
 976	ep = asoc->ep;
 977
 978	inqueue = &asoc->base.inqueue;
 979	sctp_association_hold(asoc);
 980	while (NULL != (chunk = sctp_inq_pop(inqueue))) {
 981		state = asoc->state;
 982		subtype = SCTP_ST_CHUNK(chunk->chunk_hdr->type);
 983
 984		/* If the first chunk in the packet is AUTH, do special
 985		 * processing specified in Section 6.3 of SCTP-AUTH spec
 986		 */
 987		if (first_time && subtype.chunk == SCTP_CID_AUTH) {
 988			struct sctp_chunkhdr *next_hdr;
 989
 990			next_hdr = sctp_inq_peek(inqueue);
 991			if (!next_hdr)
 992				goto normal;
 993
 994			/* If the next chunk is COOKIE-ECHO, skip the AUTH
 995			 * chunk while saving a pointer to it so we can do
 996			 * Authentication later (during cookie-echo
 997			 * processing).
 998			 */
 999			if (next_hdr->type == SCTP_CID_COOKIE_ECHO) {
1000				chunk->auth_chunk = skb_clone(chunk->skb,
1001							      GFP_ATOMIC);
1002				chunk->auth = 1;
1003				continue;
1004			}
1005		}
1006
1007normal:
1008		/* SCTP-AUTH, Section 6.3:
1009		 *    The receiver has a list of chunk types which it expects
1010		 *    to be received only after an AUTH-chunk.  This list has
1011		 *    been sent to the peer during the association setup.  It
1012		 *    MUST silently discard these chunks if they are not placed
1013		 *    after an AUTH chunk in the packet.
1014		 */
1015		if (sctp_auth_recv_cid(subtype.chunk, asoc) && !chunk->auth)
1016			continue;
1017
1018		/* Remember where the last DATA chunk came from so we
1019		 * know where to send the SACK.
1020		 */
1021		if (sctp_chunk_is_data(chunk))
1022			asoc->peer.last_data_from = chunk->transport;
1023		else {
1024			SCTP_INC_STATS(net, SCTP_MIB_INCTRLCHUNKS);
1025			asoc->stats.ictrlchunks++;
1026			if (chunk->chunk_hdr->type == SCTP_CID_SACK)
1027				asoc->stats.isacks++;
1028		}
1029
1030		if (chunk->transport)
1031			chunk->transport->last_time_heard = ktime_get();
1032
1033		/* Run through the state machine. */
1034		error = sctp_do_sm(net, SCTP_EVENT_T_CHUNK, subtype,
1035				   state, ep, asoc, chunk, GFP_ATOMIC);
1036
1037		/* Check to see if the association is freed in response to
1038		 * the incoming chunk.  If so, get out of the while loop.
1039		 */
1040		if (asoc->base.dead)
1041			break;
1042
1043		/* If there is an error on chunk, discard this packet. */
1044		if (error && chunk)
1045			chunk->pdiscard = 1;
1046
1047		if (first_time)
1048			first_time = 0;
1049	}
1050	sctp_association_put(asoc);
1051}
1052
1053/* This routine moves an association from its old sk to a new sk.  */
1054void sctp_assoc_migrate(struct sctp_association *assoc, struct sock *newsk)
1055{
1056	struct sctp_sock *newsp = sctp_sk(newsk);
1057	struct sock *oldsk = assoc->base.sk;
1058
1059	/* Delete the association from the old endpoint's list of
1060	 * associations.
1061	 */
1062	list_del_init(&assoc->asocs);
1063
1064	/* Decrement the backlog value for a TCP-style socket. */
1065	if (sctp_style(oldsk, TCP))
1066		sk_acceptq_removed(oldsk);
1067
1068	/* Release references to the old endpoint and the sock.  */
1069	sctp_endpoint_put(assoc->ep);
1070	sock_put(assoc->base.sk);
1071
1072	/* Get a reference to the new endpoint.  */
1073	assoc->ep = newsp->ep;
1074	sctp_endpoint_hold(assoc->ep);
1075
1076	/* Get a reference to the new sock.  */
1077	assoc->base.sk = newsk;
1078	sock_hold(assoc->base.sk);
1079
1080	/* Add the association to the new endpoint's list of associations.  */
1081	sctp_endpoint_add_asoc(newsp->ep, assoc);
1082}
1083
1084/* Update an association (possibly from unexpected COOKIE-ECHO processing).  */
1085int sctp_assoc_update(struct sctp_association *asoc,
1086		      struct sctp_association *new)
1087{
1088	struct sctp_transport *trans;
1089	struct list_head *pos, *temp;
1090
1091	/* Copy in new parameters of peer. */
1092	asoc->c = new->c;
1093	asoc->peer.rwnd = new->peer.rwnd;
1094	asoc->peer.sack_needed = new->peer.sack_needed;
1095	asoc->peer.auth_capable = new->peer.auth_capable;
1096	asoc->peer.i = new->peer.i;
1097
1098	if (!sctp_tsnmap_init(&asoc->peer.tsn_map, SCTP_TSN_MAP_INITIAL,
1099			      asoc->peer.i.initial_tsn, GFP_ATOMIC))
1100		return -ENOMEM;
1101
1102	/* Remove any peer addresses not present in the new association. */
1103	list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
1104		trans = list_entry(pos, struct sctp_transport, transports);
1105		if (!sctp_assoc_lookup_paddr(new, &trans->ipaddr)) {
1106			sctp_assoc_rm_peer(asoc, trans);
1107			continue;
1108		}
1109
1110		if (asoc->state >= SCTP_STATE_ESTABLISHED)
1111			sctp_transport_reset(trans);
1112	}
1113
1114	/* If the case is A (association restart), use
1115	 * initial_tsn as next_tsn. If the case is B, use
1116	 * current next_tsn in case data sent to peer
1117	 * has been discarded and needs retransmission.
1118	 */
1119	if (asoc->state >= SCTP_STATE_ESTABLISHED) {
1120		asoc->next_tsn = new->next_tsn;
1121		asoc->ctsn_ack_point = new->ctsn_ack_point;
1122		asoc->adv_peer_ack_point = new->adv_peer_ack_point;
1123
1124		/* Reinitialize SSN for both local streams
1125		 * and peer's streams.
1126		 */
1127		sctp_stream_clear(&asoc->stream);
1128
1129		/* Flush the ULP reassembly and ordered queue.
1130		 * Any data there will now be stale and will
1131		 * cause problems.
1132		 */
1133		sctp_ulpq_flush(&asoc->ulpq);
1134
1135		/* reset the overall association error count so
1136		 * that the restarted association doesn't get torn
1137		 * down on the next retransmission timer.
1138		 */
1139		asoc->overall_error_count = 0;
1140
1141	} else {
1142		/* Add any peer addresses from the new association. */
1143		list_for_each_entry(trans, &new->peer.transport_addr_list,
1144				    transports)
1145			if (!sctp_assoc_add_peer(asoc, &trans->ipaddr,
1146						 GFP_ATOMIC, trans->state))
1147				return -ENOMEM;
1148
1149		asoc->ctsn_ack_point = asoc->next_tsn - 1;
1150		asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
1151
1152		if (sctp_state(asoc, COOKIE_WAIT))
1153			sctp_stream_update(&asoc->stream, &new->stream);
1154
1155		/* get a new assoc id if we don't have one yet. */
1156		if (sctp_assoc_set_id(asoc, GFP_ATOMIC))
1157			return -ENOMEM;
1158	}
1159
1160	/* SCTP-AUTH: Save the peer parameters from the new associations
1161	 * and also move the association shared keys over
1162	 */
1163	kfree(asoc->peer.peer_random);
1164	asoc->peer.peer_random = new->peer.peer_random;
1165	new->peer.peer_random = NULL;
1166
1167	kfree(asoc->peer.peer_chunks);
1168	asoc->peer.peer_chunks = new->peer.peer_chunks;
1169	new->peer.peer_chunks = NULL;
1170
1171	kfree(asoc->peer.peer_hmacs);
1172	asoc->peer.peer_hmacs = new->peer.peer_hmacs;
1173	new->peer.peer_hmacs = NULL;
1174
1175	return sctp_auth_asoc_init_active_key(asoc, GFP_ATOMIC);
1176}
1177
1178/* Update the retran path for sending a retransmitted packet.
1179 * See also RFC4960, 6.4. Multi-Homed SCTP Endpoints:
1180 *
1181 *   When there is outbound data to send and the primary path
1182 *   becomes inactive (e.g., due to failures), or where the
1183 *   SCTP user explicitly requests to send data to an
1184 *   inactive destination transport address, before reporting
1185 *   an error to its ULP, the SCTP endpoint should try to send
1186 *   the data to an alternate active destination transport
1187 *   address if one exists.
1188 *
1189 *   When retransmitting data that timed out, if the endpoint
1190 *   is multihomed, it should consider each source-destination
1191 *   address pair in its retransmission selection policy.
1192 *   When retransmitting timed-out data, the endpoint should
1193 *   attempt to pick the most divergent source-destination
1194 *   pair from the original source-destination pair to which
1195 *   the packet was transmitted.
1196 *
1197 *   Note: Rules for picking the most divergent source-destination
1198 *   pair are an implementation decision and are not specified
1199 *   within this document.
1200 *
1201 * Our basic strategy is to round-robin transports in priorities
1202 * according to sctp_trans_score() e.g., if no such
1203 * transport with state SCTP_ACTIVE exists, round-robin through
1204 * SCTP_UNKNOWN, etc. You get the picture.
1205 */
1206static u8 sctp_trans_score(const struct sctp_transport *trans)
1207{
1208	switch (trans->state) {
1209	case SCTP_ACTIVE:
1210		return 3;	/* best case */
1211	case SCTP_UNKNOWN:
1212		return 2;
1213	case SCTP_PF:
1214		return 1;
1215	default: /* case SCTP_INACTIVE */
1216		return 0;	/* worst case */
1217	}
1218}
1219
1220static struct sctp_transport *sctp_trans_elect_tie(struct sctp_transport *trans1,
1221						   struct sctp_transport *trans2)
1222{
1223	if (trans1->error_count > trans2->error_count) {
1224		return trans2;
1225	} else if (trans1->error_count == trans2->error_count &&
1226		   ktime_after(trans2->last_time_heard,
1227			       trans1->last_time_heard)) {
1228		return trans2;
1229	} else {
1230		return trans1;
1231	}
1232}
1233
1234static struct sctp_transport *sctp_trans_elect_best(struct sctp_transport *curr,
1235						    struct sctp_transport *best)
1236{
1237	u8 score_curr, score_best;
1238
1239	if (best == NULL || curr == best)
1240		return curr;
1241
1242	score_curr = sctp_trans_score(curr);
1243	score_best = sctp_trans_score(best);
1244
1245	/* First, try a score-based selection if both transport states
1246	 * differ. If we're in a tie, lets try to make a more clever
1247	 * decision here based on error counts and last time heard.
1248	 */
1249	if (score_curr > score_best)
1250		return curr;
1251	else if (score_curr == score_best)
1252		return sctp_trans_elect_tie(best, curr);
1253	else
1254		return best;
1255}
1256
1257void sctp_assoc_update_retran_path(struct sctp_association *asoc)
1258{
1259	struct sctp_transport *trans = asoc->peer.retran_path;
1260	struct sctp_transport *trans_next = NULL;
1261
1262	/* We're done as we only have the one and only path. */
1263	if (asoc->peer.transport_count == 1)
1264		return;
1265	/* If active_path and retran_path are the same and active,
1266	 * then this is the only active path. Use it.
1267	 */
1268	if (asoc->peer.active_path == asoc->peer.retran_path &&
1269	    asoc->peer.active_path->state == SCTP_ACTIVE)
1270		return;
1271
1272	/* Iterate from retran_path's successor back to retran_path. */
1273	for (trans = list_next_entry(trans, transports); 1;
1274	     trans = list_next_entry(trans, transports)) {
1275		/* Manually skip the head element. */
1276		if (&trans->transports == &asoc->peer.transport_addr_list)
1277			continue;
1278		if (trans->state == SCTP_UNCONFIRMED)
1279			continue;
1280		trans_next = sctp_trans_elect_best(trans, trans_next);
1281		/* Active is good enough for immediate return. */
1282		if (trans_next->state == SCTP_ACTIVE)
1283			break;
1284		/* We've reached the end, time to update path. */
1285		if (trans == asoc->peer.retran_path)
1286			break;
1287	}
1288
1289	asoc->peer.retran_path = trans_next;
1290
1291	pr_debug("%s: association:%p updated new path to addr:%pISpc\n",
1292		 __func__, asoc, &asoc->peer.retran_path->ipaddr.sa);
1293}
1294
1295static void sctp_select_active_and_retran_path(struct sctp_association *asoc)
1296{
1297	struct sctp_transport *trans, *trans_pri = NULL, *trans_sec = NULL;
1298	struct sctp_transport *trans_pf = NULL;
1299
1300	/* Look for the two most recently used active transports. */
1301	list_for_each_entry(trans, &asoc->peer.transport_addr_list,
1302			    transports) {
1303		/* Skip uninteresting transports. */
1304		if (trans->state == SCTP_INACTIVE ||
1305		    trans->state == SCTP_UNCONFIRMED)
1306			continue;
1307		/* Keep track of the best PF transport from our
1308		 * list in case we don't find an active one.
1309		 */
1310		if (trans->state == SCTP_PF) {
1311			trans_pf = sctp_trans_elect_best(trans, trans_pf);
1312			continue;
1313		}
1314		/* For active transports, pick the most recent ones. */
1315		if (trans_pri == NULL ||
1316		    ktime_after(trans->last_time_heard,
1317				trans_pri->last_time_heard)) {
1318			trans_sec = trans_pri;
1319			trans_pri = trans;
1320		} else if (trans_sec == NULL ||
1321			   ktime_after(trans->last_time_heard,
1322				       trans_sec->last_time_heard)) {
1323			trans_sec = trans;
1324		}
1325	}
1326
1327	/* RFC 2960 6.4 Multi-Homed SCTP Endpoints
1328	 *
1329	 * By default, an endpoint should always transmit to the primary
1330	 * path, unless the SCTP user explicitly specifies the
1331	 * destination transport address (and possibly source transport
1332	 * address) to use. [If the primary is active but not most recent,
1333	 * bump the most recently used transport.]
1334	 */
1335	if ((asoc->peer.primary_path->state == SCTP_ACTIVE ||
1336	     asoc->peer.primary_path->state == SCTP_UNKNOWN) &&
1337	     asoc->peer.primary_path != trans_pri) {
1338		trans_sec = trans_pri;
1339		trans_pri = asoc->peer.primary_path;
1340	}
1341
1342	/* We did not find anything useful for a possible retransmission
1343	 * path; either primary path that we found is the same as
1344	 * the current one, or we didn't generally find an active one.
1345	 */
1346	if (trans_sec == NULL)
1347		trans_sec = trans_pri;
1348
1349	/* If we failed to find a usable transport, just camp on the
1350	 * active or pick a PF iff it's the better choice.
1351	 */
1352	if (trans_pri == NULL) {
1353		trans_pri = sctp_trans_elect_best(asoc->peer.active_path, trans_pf);
1354		trans_sec = trans_pri;
1355	}
1356
1357	/* Set the active and retran transports. */
1358	asoc->peer.active_path = trans_pri;
1359	asoc->peer.retran_path = trans_sec;
1360}
1361
1362struct sctp_transport *
1363sctp_assoc_choose_alter_transport(struct sctp_association *asoc,
1364				  struct sctp_transport *last_sent_to)
1365{
1366	/* If this is the first time packet is sent, use the active path,
1367	 * else use the retran path. If the last packet was sent over the
1368	 * retran path, update the retran path and use it.
1369	 */
1370	if (last_sent_to == NULL) {
1371		return asoc->peer.active_path;
1372	} else {
1373		if (last_sent_to == asoc->peer.retran_path)
1374			sctp_assoc_update_retran_path(asoc);
1375
1376		return asoc->peer.retran_path;
1377	}
1378}
1379
1380void sctp_assoc_update_frag_point(struct sctp_association *asoc)
1381{
1382	int frag = sctp_mtu_payload(sctp_sk(asoc->base.sk), asoc->pathmtu,
1383				    sctp_datachk_len(&asoc->stream));
1384
1385	if (asoc->user_frag)
1386		frag = min_t(int, frag, asoc->user_frag);
1387
1388	frag = min_t(int, frag, SCTP_MAX_CHUNK_LEN -
1389				sctp_datachk_len(&asoc->stream));
1390
1391	asoc->frag_point = SCTP_TRUNC4(frag);
1392}
1393
1394void sctp_assoc_set_pmtu(struct sctp_association *asoc, __u32 pmtu)
1395{
1396	if (asoc->pathmtu != pmtu) {
1397		asoc->pathmtu = pmtu;
1398		sctp_assoc_update_frag_point(asoc);
1399	}
1400
1401	pr_debug("%s: asoc:%p, pmtu:%d, frag_point:%d\n", __func__, asoc,
1402		 asoc->pathmtu, asoc->frag_point);
1403}
1404
1405/* Update the association's pmtu and frag_point by going through all the
1406 * transports. This routine is called when a transport's PMTU has changed.
1407 */
1408void sctp_assoc_sync_pmtu(struct sctp_association *asoc)
1409{
1410	struct sctp_transport *t;
1411	__u32 pmtu = 0;
1412
1413	if (!asoc)
1414		return;
1415
1416	/* Get the lowest pmtu of all the transports. */
1417	list_for_each_entry(t, &asoc->peer.transport_addr_list, transports) {
1418		if (t->pmtu_pending && t->dst) {
1419			sctp_transport_update_pmtu(t,
1420						   atomic_read(&t->mtu_info));
1421			t->pmtu_pending = 0;
1422		}
1423		if (!pmtu || (t->pathmtu < pmtu))
1424			pmtu = t->pathmtu;
1425	}
1426
1427	sctp_assoc_set_pmtu(asoc, pmtu);
1428}
1429
1430/* Should we send a SACK to update our peer? */
1431static inline bool sctp_peer_needs_update(struct sctp_association *asoc)
1432{
1433	struct net *net = asoc->base.net;
1434
1435	switch (asoc->state) {
1436	case SCTP_STATE_ESTABLISHED:
1437	case SCTP_STATE_SHUTDOWN_PENDING:
1438	case SCTP_STATE_SHUTDOWN_RECEIVED:
1439	case SCTP_STATE_SHUTDOWN_SENT:
1440		if ((asoc->rwnd > asoc->a_rwnd) &&
1441		    ((asoc->rwnd - asoc->a_rwnd) >= max_t(__u32,
1442			   (asoc->base.sk->sk_rcvbuf >> net->sctp.rwnd_upd_shift),
1443			   asoc->pathmtu)))
1444			return true;
1445		break;
1446	default:
1447		break;
1448	}
1449	return false;
1450}
1451
1452/* Increase asoc's rwnd by len and send any window update SACK if needed. */
1453void sctp_assoc_rwnd_increase(struct sctp_association *asoc, unsigned int len)
1454{
1455	struct sctp_chunk *sack;
1456	struct timer_list *timer;
1457
1458	if (asoc->rwnd_over) {
1459		if (asoc->rwnd_over >= len) {
1460			asoc->rwnd_over -= len;
1461		} else {
1462			asoc->rwnd += (len - asoc->rwnd_over);
1463			asoc->rwnd_over = 0;
1464		}
1465	} else {
1466		asoc->rwnd += len;
1467	}
1468
1469	/* If we had window pressure, start recovering it
1470	 * once our rwnd had reached the accumulated pressure
1471	 * threshold.  The idea is to recover slowly, but up
1472	 * to the initial advertised window.
1473	 */
1474	if (asoc->rwnd_press) {
1475		int change = min(asoc->pathmtu, asoc->rwnd_press);
1476		asoc->rwnd += change;
1477		asoc->rwnd_press -= change;
1478	}
1479
1480	pr_debug("%s: asoc:%p rwnd increased by %d to (%u, %u) - %u\n",
1481		 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
1482		 asoc->a_rwnd);
1483
1484	/* Send a window update SACK if the rwnd has increased by at least the
1485	 * minimum of the association's PMTU and half of the receive buffer.
1486	 * The algorithm used is similar to the one described in
1487	 * Section 4.2.3.3 of RFC 1122.
1488	 */
1489	if (sctp_peer_needs_update(asoc)) {
1490		asoc->a_rwnd = asoc->rwnd;
1491
1492		pr_debug("%s: sending window update SACK- asoc:%p rwnd:%u "
1493			 "a_rwnd:%u\n", __func__, asoc, asoc->rwnd,
1494			 asoc->a_rwnd);
1495
1496		sack = sctp_make_sack(asoc);
1497		if (!sack)
1498			return;
1499
1500		asoc->peer.sack_needed = 0;
1501
1502		sctp_outq_tail(&asoc->outqueue, sack, GFP_ATOMIC);
1503
1504		/* Stop the SACK timer.  */
1505		timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
1506		if (timer_delete(timer))
1507			sctp_association_put(asoc);
1508	}
1509}
1510
1511/* Decrease asoc's rwnd by len. */
1512void sctp_assoc_rwnd_decrease(struct sctp_association *asoc, unsigned int len)
1513{
1514	int rx_count;
1515	int over = 0;
1516
1517	if (unlikely(!asoc->rwnd || asoc->rwnd_over))
1518		pr_debug("%s: association:%p has asoc->rwnd:%u, "
1519			 "asoc->rwnd_over:%u!\n", __func__, asoc,
1520			 asoc->rwnd, asoc->rwnd_over);
1521
1522	if (asoc->ep->rcvbuf_policy)
1523		rx_count = atomic_read(&asoc->rmem_alloc);
1524	else
1525		rx_count = atomic_read(&asoc->base.sk->sk_rmem_alloc);
1526
1527	/* If we've reached or overflowed our receive buffer, announce
1528	 * a 0 rwnd if rwnd would still be positive.  Store the
1529	 * potential pressure overflow so that the window can be restored
1530	 * back to original value.
1531	 */
1532	if (rx_count >= asoc->base.sk->sk_rcvbuf)
1533		over = 1;
1534
1535	if (asoc->rwnd >= len) {
1536		asoc->rwnd -= len;
1537		if (over) {
1538			asoc->rwnd_press += asoc->rwnd;
1539			asoc->rwnd = 0;
1540		}
1541	} else {
1542		asoc->rwnd_over += len - asoc->rwnd;
1543		asoc->rwnd = 0;
1544	}
1545
1546	pr_debug("%s: asoc:%p rwnd decreased by %d to (%u, %u, %u)\n",
1547		 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
1548		 asoc->rwnd_press);
1549}
1550
1551/* Build the bind address list for the association based on info from the
1552 * local endpoint and the remote peer.
1553 */
1554int sctp_assoc_set_bind_addr_from_ep(struct sctp_association *asoc,
1555				     enum sctp_scope scope, gfp_t gfp)
1556{
1557	struct sock *sk = asoc->base.sk;
1558	int flags;
1559
1560	/* Use scoping rules to determine the subset of addresses from
1561	 * the endpoint.
1562	 */
1563	flags = (PF_INET6 == sk->sk_family) ? SCTP_ADDR6_ALLOWED : 0;
1564	if (!inet_v6_ipv6only(sk))
1565		flags |= SCTP_ADDR4_ALLOWED;
1566	if (asoc->peer.ipv4_address)
1567		flags |= SCTP_ADDR4_PEERSUPP;
1568	if (asoc->peer.ipv6_address)
1569		flags |= SCTP_ADDR6_PEERSUPP;
1570
1571	return sctp_bind_addr_copy(asoc->base.net,
1572				   &asoc->base.bind_addr,
1573				   &asoc->ep->base.bind_addr,
1574				   scope, gfp, flags);
1575}
1576
1577/* Build the association's bind address list from the cookie.  */
1578int sctp_assoc_set_bind_addr_from_cookie(struct sctp_association *asoc,
1579					 struct sctp_cookie *cookie,
1580					 gfp_t gfp)
1581{
1582	struct sctp_init_chunk *peer_init = (struct sctp_init_chunk *)(cookie + 1);
1583	int var_size2 = ntohs(peer_init->chunk_hdr.length);
1584	int var_size3 = cookie->raw_addr_list_len;
1585	__u8 *raw = (__u8 *)peer_init + var_size2;
1586
1587	return sctp_raw_to_bind_addrs(&asoc->base.bind_addr, raw, var_size3,
1588				      asoc->ep->base.bind_addr.port, gfp);
1589}
1590
1591/* Lookup laddr in the bind address list of an association. */
1592int sctp_assoc_lookup_laddr(struct sctp_association *asoc,
1593			    const union sctp_addr *laddr)
1594{
1595	int found = 0;
1596
1597	if ((asoc->base.bind_addr.port == ntohs(laddr->v4.sin_port)) &&
1598	    sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
1599				 sctp_sk(asoc->base.sk)))
1600		found = 1;
1601
1602	return found;
1603}
1604
1605/* Set an association id for a given association */
1606int sctp_assoc_set_id(struct sctp_association *asoc, gfp_t gfp)
1607{
1608	bool preload = gfpflags_allow_blocking(gfp);
1609	int ret;
1610
1611	/* If the id is already assigned, keep it. */
1612	if (asoc->assoc_id)
1613		return 0;
1614
1615	if (preload)
1616		idr_preload(gfp);
1617	spin_lock_bh(&sctp_assocs_id_lock);
1618	/* 0, 1, 2 are used as SCTP_FUTURE_ASSOC, SCTP_CURRENT_ASSOC and
1619	 * SCTP_ALL_ASSOC, so an available id must be > SCTP_ALL_ASSOC.
1620	 */
1621	ret = idr_alloc_cyclic(&sctp_assocs_id, asoc, SCTP_ALL_ASSOC + 1, 0,
1622			       GFP_NOWAIT);
1623	spin_unlock_bh(&sctp_assocs_id_lock);
1624	if (preload)
1625		idr_preload_end();
1626	if (ret < 0)
1627		return ret;
1628
1629	asoc->assoc_id = (sctp_assoc_t)ret;
1630	return 0;
1631}
1632
1633/* Free the ASCONF queue */
1634static void sctp_assoc_free_asconf_queue(struct sctp_association *asoc)
1635{
1636	struct sctp_chunk *asconf;
1637	struct sctp_chunk *tmp;
1638
1639	list_for_each_entry_safe(asconf, tmp, &asoc->addip_chunk_list, list) {
1640		list_del_init(&asconf->list);
1641		sctp_chunk_free(asconf);
1642	}
1643}
1644
1645/* Free asconf_ack cache */
1646static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc)
1647{
1648	struct sctp_chunk *ack;
1649	struct sctp_chunk *tmp;
1650
1651	list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
1652				transmitted_list) {
1653		list_del_init(&ack->transmitted_list);
1654		sctp_chunk_free(ack);
1655	}
1656}
1657
1658/* Clean up the ASCONF_ACK queue */
1659void sctp_assoc_clean_asconf_ack_cache(const struct sctp_association *asoc)
1660{
1661	struct sctp_chunk *ack;
1662	struct sctp_chunk *tmp;
1663
1664	/* We can remove all the entries from the queue up to
1665	 * the "Peer-Sequence-Number".
1666	 */
1667	list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
1668				transmitted_list) {
1669		if (ack->subh.addip_hdr->serial ==
1670				htonl(asoc->peer.addip_serial))
1671			break;
1672
1673		list_del_init(&ack->transmitted_list);
1674		sctp_chunk_free(ack);
1675	}
1676}
1677
1678/* Find the ASCONF_ACK whose serial number matches ASCONF */
1679struct sctp_chunk *sctp_assoc_lookup_asconf_ack(
1680					const struct sctp_association *asoc,
1681					__be32 serial)
1682{
1683	struct sctp_chunk *ack;
1684
1685	/* Walk through the list of cached ASCONF-ACKs and find the
1686	 * ack chunk whose serial number matches that of the request.
1687	 */
1688	list_for_each_entry(ack, &asoc->asconf_ack_list, transmitted_list) {
1689		if (sctp_chunk_pending(ack))
1690			continue;
1691		if (ack->subh.addip_hdr->serial == serial) {
1692			sctp_chunk_hold(ack);
1693			return ack;
1694		}
1695	}
1696
1697	return NULL;
1698}
1699
1700void sctp_asconf_queue_teardown(struct sctp_association *asoc)
1701{
1702	/* Free any cached ASCONF_ACK chunk. */
1703	sctp_assoc_free_asconf_acks(asoc);
1704
1705	/* Free the ASCONF queue. */
1706	sctp_assoc_free_asconf_queue(asoc);
1707
1708	/* Free any cached ASCONF chunk. */
1709	if (asoc->addip_last_asconf)
1710		sctp_chunk_free(asoc->addip_last_asconf);
1711}