xref: /linux/net/sctp/associola.c (revision 2dbc0838bcf24ca59cabc3130cf3b1d6809cdcd4)
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. */
43 static void sctp_select_active_and_retran_path(struct sctp_association *asoc);
44 static void sctp_assoc_bh_rcv(struct work_struct *work);
45 static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc);
46 static void sctp_assoc_free_asconf_queue(struct sctp_association *asoc);
47 
48 /* 1st Level Abstractions. */
49 
50 /* Initialize a new association from provided memory. */
51 static 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 net *net = sock_net(sk);
58 	struct sctp_sock *sp;
59 	struct sctp_paramhdr *p;
60 	int i;
61 
62 	/* Retrieve the SCTP per socket area.  */
63 	sp = sctp_sk((struct sock *)sk);
64 
65 	/* Discarding const is appropriate here.  */
66 	asoc->ep = (struct sctp_endpoint *)ep;
67 	asoc->base.sk = (struct sock *)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 
91 	asoc->rto_initial = msecs_to_jiffies(sp->rtoinfo.srto_initial);
92 	asoc->rto_max = msecs_to_jiffies(sp->rtoinfo.srto_max);
93 	asoc->rto_min = msecs_to_jiffies(sp->rtoinfo.srto_min);
94 
95 	/* Initialize the association's heartbeat interval based on the
96 	 * sock configured value.
97 	 */
98 	asoc->hbinterval = msecs_to_jiffies(sp->hbinterval);
99 
100 	/* Initialize path max retrans value. */
101 	asoc->pathmaxrxt = sp->pathmaxrxt;
102 
103 	asoc->flowlabel = sp->flowlabel;
104 	asoc->dscp = sp->dscp;
105 
106 	/* Set association default SACK delay */
107 	asoc->sackdelay = msecs_to_jiffies(sp->sackdelay);
108 	asoc->sackfreq = sp->sackfreq;
109 
110 	/* Set the association default flags controlling
111 	 * Heartbeat, SACK delay, and Path MTU Discovery.
112 	 */
113 	asoc->param_flags = sp->param_flags;
114 
115 	/* Initialize the maximum number of new data packets that can be sent
116 	 * in a burst.
117 	 */
118 	asoc->max_burst = sp->max_burst;
119 
120 	asoc->subscribe = sp->subscribe;
121 
122 	/* initialize association timers */
123 	asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_COOKIE] = asoc->rto_initial;
124 	asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_INIT] = asoc->rto_initial;
125 	asoc->timeouts[SCTP_EVENT_TIMEOUT_T2_SHUTDOWN] = asoc->rto_initial;
126 
127 	/* sctpimpguide Section 2.12.2
128 	 * If the 'T5-shutdown-guard' timer is used, it SHOULD be set to the
129 	 * recommended value of 5 times 'RTO.Max'.
130 	 */
131 	asoc->timeouts[SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]
132 		= 5 * asoc->rto_max;
133 
134 	asoc->timeouts[SCTP_EVENT_TIMEOUT_SACK] = asoc->sackdelay;
135 	asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE] = sp->autoclose * HZ;
136 
137 	/* Initializes the timers */
138 	for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i)
139 		timer_setup(&asoc->timers[i], sctp_timer_events[i], 0);
140 
141 	/* Pull default initialization values from the sock options.
142 	 * Note: This assumes that the values have already been
143 	 * validated in the sock.
144 	 */
145 	asoc->c.sinit_max_instreams = sp->initmsg.sinit_max_instreams;
146 	asoc->c.sinit_num_ostreams  = sp->initmsg.sinit_num_ostreams;
147 	asoc->max_init_attempts	= sp->initmsg.sinit_max_attempts;
148 
149 	asoc->max_init_timeo =
150 		 msecs_to_jiffies(sp->initmsg.sinit_max_init_timeo);
151 
152 	/* Set the local window size for receive.
153 	 * This is also the rcvbuf space per association.
154 	 * RFC 6 - A SCTP receiver MUST be able to receive a minimum of
155 	 * 1500 bytes in one SCTP packet.
156 	 */
157 	if ((sk->sk_rcvbuf/2) < SCTP_DEFAULT_MINWINDOW)
158 		asoc->rwnd = SCTP_DEFAULT_MINWINDOW;
159 	else
160 		asoc->rwnd = sk->sk_rcvbuf/2;
161 
162 	asoc->a_rwnd = asoc->rwnd;
163 
164 	/* Use my own max window until I learn something better.  */
165 	asoc->peer.rwnd = SCTP_DEFAULT_MAXWINDOW;
166 
167 	/* Initialize the receive memory counter */
168 	atomic_set(&asoc->rmem_alloc, 0);
169 
170 	init_waitqueue_head(&asoc->wait);
171 
172 	asoc->c.my_vtag = sctp_generate_tag(ep);
173 	asoc->c.my_port = ep->base.bind_addr.port;
174 
175 	asoc->c.initial_tsn = sctp_generate_tsn(ep);
176 
177 	asoc->next_tsn = asoc->c.initial_tsn;
178 
179 	asoc->ctsn_ack_point = asoc->next_tsn - 1;
180 	asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
181 	asoc->highest_sacked = asoc->ctsn_ack_point;
182 	asoc->last_cwr_tsn = asoc->ctsn_ack_point;
183 
184 	/* ADDIP Section 4.1 Asconf Chunk Procedures
185 	 *
186 	 * When an endpoint has an ASCONF signaled change to be sent to the
187 	 * remote endpoint it should do the following:
188 	 * ...
189 	 * A2) a serial number should be assigned to the chunk. The serial
190 	 * number SHOULD be a monotonically increasing number. The serial
191 	 * numbers SHOULD be initialized at the start of the
192 	 * association to the same value as the initial TSN.
193 	 */
194 	asoc->addip_serial = asoc->c.initial_tsn;
195 	asoc->strreset_outseq = asoc->c.initial_tsn;
196 
197 	INIT_LIST_HEAD(&asoc->addip_chunk_list);
198 	INIT_LIST_HEAD(&asoc->asconf_ack_list);
199 
200 	/* Make an empty list of remote transport addresses.  */
201 	INIT_LIST_HEAD(&asoc->peer.transport_addr_list);
202 
203 	/* RFC 2960 5.1 Normal Establishment of an Association
204 	 *
205 	 * After the reception of the first data chunk in an
206 	 * association the endpoint must immediately respond with a
207 	 * sack to acknowledge the data chunk.  Subsequent
208 	 * acknowledgements should be done as described in Section
209 	 * 6.2.
210 	 *
211 	 * [We implement this by telling a new association that it
212 	 * already received one packet.]
213 	 */
214 	asoc->peer.sack_needed = 1;
215 	asoc->peer.sack_generation = 1;
216 
217 	/* Assume that the peer will tell us if he recognizes ASCONF
218 	 * as part of INIT exchange.
219 	 * The sctp_addip_noauth option is there for backward compatibility
220 	 * and will revert old behavior.
221 	 */
222 	if (net->sctp.addip_noauth)
223 		asoc->peer.asconf_capable = 1;
224 
225 	/* Create an input queue.  */
226 	sctp_inq_init(&asoc->base.inqueue);
227 	sctp_inq_set_th_handler(&asoc->base.inqueue, sctp_assoc_bh_rcv);
228 
229 	/* Create an output queue.  */
230 	sctp_outq_init(asoc, &asoc->outqueue);
231 
232 	if (!sctp_ulpq_init(&asoc->ulpq, asoc))
233 		goto fail_init;
234 
235 	if (sctp_stream_init(&asoc->stream, asoc->c.sinit_num_ostreams,
236 			     0, gfp))
237 		goto fail_init;
238 
239 	/* Initialize default path MTU. */
240 	asoc->pathmtu = sp->pathmtu;
241 	sctp_assoc_update_frag_point(asoc);
242 
243 	/* Assume that peer would support both address types unless we are
244 	 * told otherwise.
245 	 */
246 	asoc->peer.ipv4_address = 1;
247 	if (asoc->base.sk->sk_family == PF_INET6)
248 		asoc->peer.ipv6_address = 1;
249 	INIT_LIST_HEAD(&asoc->asocs);
250 
251 	asoc->default_stream = sp->default_stream;
252 	asoc->default_ppid = sp->default_ppid;
253 	asoc->default_flags = sp->default_flags;
254 	asoc->default_context = sp->default_context;
255 	asoc->default_timetolive = sp->default_timetolive;
256 	asoc->default_rcv_context = sp->default_rcv_context;
257 
258 	/* AUTH related initializations */
259 	INIT_LIST_HEAD(&asoc->endpoint_shared_keys);
260 	if (sctp_auth_asoc_copy_shkeys(ep, asoc, gfp))
261 		goto stream_free;
262 
263 	asoc->active_key_id = ep->active_key_id;
264 	asoc->strreset_enable = ep->strreset_enable;
265 
266 	/* Save the hmacs and chunks list into this association */
267 	if (ep->auth_hmacs_list)
268 		memcpy(asoc->c.auth_hmacs, ep->auth_hmacs_list,
269 			ntohs(ep->auth_hmacs_list->param_hdr.length));
270 	if (ep->auth_chunk_list)
271 		memcpy(asoc->c.auth_chunks, ep->auth_chunk_list,
272 			ntohs(ep->auth_chunk_list->param_hdr.length));
273 
274 	/* Get the AUTH random number for this association */
275 	p = (struct sctp_paramhdr *)asoc->c.auth_random;
276 	p->type = SCTP_PARAM_RANDOM;
277 	p->length = htons(sizeof(*p) + SCTP_AUTH_RANDOM_LENGTH);
278 	get_random_bytes(p+1, SCTP_AUTH_RANDOM_LENGTH);
279 
280 	return asoc;
281 
282 stream_free:
283 	sctp_stream_free(&asoc->stream);
284 fail_init:
285 	sock_put(asoc->base.sk);
286 	sctp_endpoint_put(asoc->ep);
287 	return NULL;
288 }
289 
290 /* Allocate and initialize a new association */
291 struct sctp_association *sctp_association_new(const struct sctp_endpoint *ep,
292 					      const struct sock *sk,
293 					      enum sctp_scope scope, gfp_t gfp)
294 {
295 	struct sctp_association *asoc;
296 
297 	asoc = kzalloc(sizeof(*asoc), gfp);
298 	if (!asoc)
299 		goto fail;
300 
301 	if (!sctp_association_init(asoc, ep, sk, scope, gfp))
302 		goto fail_init;
303 
304 	SCTP_DBG_OBJCNT_INC(assoc);
305 
306 	pr_debug("Created asoc %p\n", asoc);
307 
308 	return asoc;
309 
310 fail_init:
311 	kfree(asoc);
312 fail:
313 	return NULL;
314 }
315 
316 /* Free this association if possible.  There may still be users, so
317  * the actual deallocation may be delayed.
318  */
319 void sctp_association_free(struct sctp_association *asoc)
320 {
321 	struct sock *sk = asoc->base.sk;
322 	struct sctp_transport *transport;
323 	struct list_head *pos, *temp;
324 	int i;
325 
326 	/* Only real associations count against the endpoint, so
327 	 * don't bother for if this is a temporary association.
328 	 */
329 	if (!list_empty(&asoc->asocs)) {
330 		list_del(&asoc->asocs);
331 
332 		/* Decrement the backlog value for a TCP-style listening
333 		 * socket.
334 		 */
335 		if (sctp_style(sk, TCP) && sctp_sstate(sk, LISTENING))
336 			sk->sk_ack_backlog--;
337 	}
338 
339 	/* Mark as dead, so other users can know this structure is
340 	 * going away.
341 	 */
342 	asoc->base.dead = true;
343 
344 	/* Dispose of any data lying around in the outqueue. */
345 	sctp_outq_free(&asoc->outqueue);
346 
347 	/* Dispose of any pending messages for the upper layer. */
348 	sctp_ulpq_free(&asoc->ulpq);
349 
350 	/* Dispose of any pending chunks on the inqueue. */
351 	sctp_inq_free(&asoc->base.inqueue);
352 
353 	sctp_tsnmap_free(&asoc->peer.tsn_map);
354 
355 	/* Free stream information. */
356 	sctp_stream_free(&asoc->stream);
357 
358 	if (asoc->strreset_chunk)
359 		sctp_chunk_free(asoc->strreset_chunk);
360 
361 	/* Clean up the bound address list. */
362 	sctp_bind_addr_free(&asoc->base.bind_addr);
363 
364 	/* Do we need to go through all of our timers and
365 	 * delete them?   To be safe we will try to delete all, but we
366 	 * should be able to go through and make a guess based
367 	 * on our state.
368 	 */
369 	for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i) {
370 		if (del_timer(&asoc->timers[i]))
371 			sctp_association_put(asoc);
372 	}
373 
374 	/* Free peer's cached cookie. */
375 	kfree(asoc->peer.cookie);
376 	kfree(asoc->peer.peer_random);
377 	kfree(asoc->peer.peer_chunks);
378 	kfree(asoc->peer.peer_hmacs);
379 
380 	/* Release the transport structures. */
381 	list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
382 		transport = list_entry(pos, struct sctp_transport, transports);
383 		list_del_rcu(pos);
384 		sctp_unhash_transport(transport);
385 		sctp_transport_free(transport);
386 	}
387 
388 	asoc->peer.transport_count = 0;
389 
390 	sctp_asconf_queue_teardown(asoc);
391 
392 	/* Free pending address space being deleted */
393 	kfree(asoc->asconf_addr_del_pending);
394 
395 	/* AUTH - Free the endpoint shared keys */
396 	sctp_auth_destroy_keys(&asoc->endpoint_shared_keys);
397 
398 	/* AUTH - Free the association shared key */
399 	sctp_auth_key_put(asoc->asoc_shared_key);
400 
401 	sctp_association_put(asoc);
402 }
403 
404 /* Cleanup and free up an association. */
405 static void sctp_association_destroy(struct sctp_association *asoc)
406 {
407 	if (unlikely(!asoc->base.dead)) {
408 		WARN(1, "Attempt to destroy undead association %p!\n", asoc);
409 		return;
410 	}
411 
412 	sctp_endpoint_put(asoc->ep);
413 	sock_put(asoc->base.sk);
414 
415 	if (asoc->assoc_id != 0) {
416 		spin_lock_bh(&sctp_assocs_id_lock);
417 		idr_remove(&sctp_assocs_id, asoc->assoc_id);
418 		spin_unlock_bh(&sctp_assocs_id_lock);
419 	}
420 
421 	WARN_ON(atomic_read(&asoc->rmem_alloc));
422 
423 	kfree_rcu(asoc, rcu);
424 	SCTP_DBG_OBJCNT_DEC(assoc);
425 }
426 
427 /* Change the primary destination address for the peer. */
428 void sctp_assoc_set_primary(struct sctp_association *asoc,
429 			    struct sctp_transport *transport)
430 {
431 	int changeover = 0;
432 
433 	/* it's a changeover only if we already have a primary path
434 	 * that we are changing
435 	 */
436 	if (asoc->peer.primary_path != NULL &&
437 	    asoc->peer.primary_path != transport)
438 		changeover = 1 ;
439 
440 	asoc->peer.primary_path = transport;
441 
442 	/* Set a default msg_name for events. */
443 	memcpy(&asoc->peer.primary_addr, &transport->ipaddr,
444 	       sizeof(union sctp_addr));
445 
446 	/* If the primary path is changing, assume that the
447 	 * user wants to use this new path.
448 	 */
449 	if ((transport->state == SCTP_ACTIVE) ||
450 	    (transport->state == SCTP_UNKNOWN))
451 		asoc->peer.active_path = transport;
452 
453 	/*
454 	 * SFR-CACC algorithm:
455 	 * Upon the receipt of a request to change the primary
456 	 * destination address, on the data structure for the new
457 	 * primary destination, the sender MUST do the following:
458 	 *
459 	 * 1) If CHANGEOVER_ACTIVE is set, then there was a switch
460 	 * to this destination address earlier. The sender MUST set
461 	 * CYCLING_CHANGEOVER to indicate that this switch is a
462 	 * double switch to the same destination address.
463 	 *
464 	 * Really, only bother is we have data queued or outstanding on
465 	 * the association.
466 	 */
467 	if (!asoc->outqueue.outstanding_bytes && !asoc->outqueue.out_qlen)
468 		return;
469 
470 	if (transport->cacc.changeover_active)
471 		transport->cacc.cycling_changeover = changeover;
472 
473 	/* 2) The sender MUST set CHANGEOVER_ACTIVE to indicate that
474 	 * a changeover has occurred.
475 	 */
476 	transport->cacc.changeover_active = changeover;
477 
478 	/* 3) The sender MUST store the next TSN to be sent in
479 	 * next_tsn_at_change.
480 	 */
481 	transport->cacc.next_tsn_at_change = asoc->next_tsn;
482 }
483 
484 /* Remove a transport from an association.  */
485 void sctp_assoc_rm_peer(struct sctp_association *asoc,
486 			struct sctp_transport *peer)
487 {
488 	struct sctp_transport *transport;
489 	struct list_head *pos;
490 	struct sctp_chunk *ch;
491 
492 	pr_debug("%s: association:%p addr:%pISpc\n",
493 		 __func__, asoc, &peer->ipaddr.sa);
494 
495 	/* If we are to remove the current retran_path, update it
496 	 * to the next peer before removing this peer from the list.
497 	 */
498 	if (asoc->peer.retran_path == peer)
499 		sctp_assoc_update_retran_path(asoc);
500 
501 	/* Remove this peer from the list. */
502 	list_del_rcu(&peer->transports);
503 	/* Remove this peer from the transport hashtable */
504 	sctp_unhash_transport(peer);
505 
506 	/* Get the first transport of asoc. */
507 	pos = asoc->peer.transport_addr_list.next;
508 	transport = list_entry(pos, struct sctp_transport, transports);
509 
510 	/* Update any entries that match the peer to be deleted. */
511 	if (asoc->peer.primary_path == peer)
512 		sctp_assoc_set_primary(asoc, transport);
513 	if (asoc->peer.active_path == peer)
514 		asoc->peer.active_path = transport;
515 	if (asoc->peer.retran_path == peer)
516 		asoc->peer.retran_path = transport;
517 	if (asoc->peer.last_data_from == peer)
518 		asoc->peer.last_data_from = transport;
519 
520 	if (asoc->strreset_chunk &&
521 	    asoc->strreset_chunk->transport == peer) {
522 		asoc->strreset_chunk->transport = transport;
523 		sctp_transport_reset_reconf_timer(transport);
524 	}
525 
526 	/* If we remove the transport an INIT was last sent to, set it to
527 	 * NULL. Combined with the update of the retran path above, this
528 	 * will cause the next INIT to be sent to the next available
529 	 * transport, maintaining the cycle.
530 	 */
531 	if (asoc->init_last_sent_to == peer)
532 		asoc->init_last_sent_to = NULL;
533 
534 	/* If we remove the transport an SHUTDOWN was last sent to, set it
535 	 * to NULL. Combined with the update of the retran path above, this
536 	 * will cause the next SHUTDOWN to be sent to the next available
537 	 * transport, maintaining the cycle.
538 	 */
539 	if (asoc->shutdown_last_sent_to == peer)
540 		asoc->shutdown_last_sent_to = NULL;
541 
542 	/* If we remove the transport an ASCONF was last sent to, set it to
543 	 * NULL.
544 	 */
545 	if (asoc->addip_last_asconf &&
546 	    asoc->addip_last_asconf->transport == peer)
547 		asoc->addip_last_asconf->transport = NULL;
548 
549 	/* If we have something on the transmitted list, we have to
550 	 * save it off.  The best place is the active path.
551 	 */
552 	if (!list_empty(&peer->transmitted)) {
553 		struct sctp_transport *active = asoc->peer.active_path;
554 
555 		/* Reset the transport of each chunk on this list */
556 		list_for_each_entry(ch, &peer->transmitted,
557 					transmitted_list) {
558 			ch->transport = NULL;
559 			ch->rtt_in_progress = 0;
560 		}
561 
562 		list_splice_tail_init(&peer->transmitted,
563 					&active->transmitted);
564 
565 		/* Start a T3 timer here in case it wasn't running so
566 		 * that these migrated packets have a chance to get
567 		 * retransmitted.
568 		 */
569 		if (!timer_pending(&active->T3_rtx_timer))
570 			if (!mod_timer(&active->T3_rtx_timer,
571 					jiffies + active->rto))
572 				sctp_transport_hold(active);
573 	}
574 
575 	list_for_each_entry(ch, &asoc->outqueue.out_chunk_list, list)
576 		if (ch->transport == peer)
577 			ch->transport = NULL;
578 
579 	asoc->peer.transport_count--;
580 
581 	sctp_transport_free(peer);
582 }
583 
584 /* Add a transport address to an association.  */
585 struct sctp_transport *sctp_assoc_add_peer(struct sctp_association *asoc,
586 					   const union sctp_addr *addr,
587 					   const gfp_t gfp,
588 					   const int peer_state)
589 {
590 	struct net *net = sock_net(asoc->base.sk);
591 	struct sctp_transport *peer;
592 	struct sctp_sock *sp;
593 	unsigned short port;
594 
595 	sp = sctp_sk(asoc->base.sk);
596 
597 	/* AF_INET and AF_INET6 share common port field. */
598 	port = ntohs(addr->v4.sin_port);
599 
600 	pr_debug("%s: association:%p addr:%pISpc state:%d\n", __func__,
601 		 asoc, &addr->sa, peer_state);
602 
603 	/* Set the port if it has not been set yet.  */
604 	if (0 == asoc->peer.port)
605 		asoc->peer.port = port;
606 
607 	/* Check to see if this is a duplicate. */
608 	peer = sctp_assoc_lookup_paddr(asoc, addr);
609 	if (peer) {
610 		/* An UNKNOWN state is only set on transports added by
611 		 * user in sctp_connectx() call.  Such transports should be
612 		 * considered CONFIRMED per RFC 4960, Section 5.4.
613 		 */
614 		if (peer->state == SCTP_UNKNOWN) {
615 			peer->state = SCTP_ACTIVE;
616 		}
617 		return peer;
618 	}
619 
620 	peer = sctp_transport_new(net, addr, gfp);
621 	if (!peer)
622 		return NULL;
623 
624 	sctp_transport_set_owner(peer, asoc);
625 
626 	/* Initialize the peer's heartbeat interval based on the
627 	 * association configured value.
628 	 */
629 	peer->hbinterval = asoc->hbinterval;
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 
637 	/* Initialize the peer's SACK delay timeout based on the
638 	 * association configured value.
639 	 */
640 	peer->sackdelay = asoc->sackdelay;
641 	peer->sackfreq = asoc->sackfreq;
642 
643 	if (addr->sa.sa_family == AF_INET6) {
644 		__be32 info = addr->v6.sin6_flowinfo;
645 
646 		if (info) {
647 			peer->flowlabel = ntohl(info & IPV6_FLOWLABEL_MASK);
648 			peer->flowlabel |= SCTP_FLOWLABEL_SET_MASK;
649 		} else {
650 			peer->flowlabel = asoc->flowlabel;
651 		}
652 	}
653 	peer->dscp = asoc->dscp;
654 
655 	/* Enable/disable heartbeat, SACK delay, and path MTU discovery
656 	 * based on association setting.
657 	 */
658 	peer->param_flags = asoc->param_flags;
659 
660 	/* Initialize the pmtu of the transport. */
661 	sctp_transport_route(peer, NULL, sp);
662 
663 	/* If this is the first transport addr on this association,
664 	 * initialize the association PMTU to the peer's PMTU.
665 	 * If not and the current association PMTU is higher than the new
666 	 * peer's PMTU, reset the association PMTU to the new peer's PMTU.
667 	 */
668 	sctp_assoc_set_pmtu(asoc, asoc->pathmtu ?
669 				  min_t(int, peer->pathmtu, asoc->pathmtu) :
670 				  peer->pathmtu);
671 
672 	peer->pmtu_pending = 0;
673 
674 	/* The asoc->peer.port might not be meaningful yet, but
675 	 * initialize the packet structure anyway.
676 	 */
677 	sctp_packet_init(&peer->packet, peer, asoc->base.bind_addr.port,
678 			 asoc->peer.port);
679 
680 	/* 7.2.1 Slow-Start
681 	 *
682 	 * o The initial cwnd before DATA transmission or after a sufficiently
683 	 *   long idle period MUST be set to
684 	 *      min(4*MTU, max(2*MTU, 4380 bytes))
685 	 *
686 	 * o The initial value of ssthresh MAY be arbitrarily high
687 	 *   (for example, implementations MAY use the size of the
688 	 *   receiver advertised window).
689 	 */
690 	peer->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
691 
692 	/* At this point, we may not have the receiver's advertised window,
693 	 * so initialize ssthresh to the default value and it will be set
694 	 * later when we process the INIT.
695 	 */
696 	peer->ssthresh = SCTP_DEFAULT_MAXWINDOW;
697 
698 	peer->partial_bytes_acked = 0;
699 	peer->flight_size = 0;
700 	peer->burst_limited = 0;
701 
702 	/* Set the transport's RTO.initial value */
703 	peer->rto = asoc->rto_initial;
704 	sctp_max_rto(asoc, peer);
705 
706 	/* Set the peer's active state. */
707 	peer->state = peer_state;
708 
709 	/* Add this peer into the transport hashtable */
710 	if (sctp_hash_transport(peer)) {
711 		sctp_transport_free(peer);
712 		return NULL;
713 	}
714 
715 	/* Attach the remote transport to our asoc.  */
716 	list_add_tail_rcu(&peer->transports, &asoc->peer.transport_addr_list);
717 	asoc->peer.transport_count++;
718 
719 	/* If we do not yet have a primary path, set one.  */
720 	if (!asoc->peer.primary_path) {
721 		sctp_assoc_set_primary(asoc, peer);
722 		asoc->peer.retran_path = peer;
723 	}
724 
725 	if (asoc->peer.active_path == asoc->peer.retran_path &&
726 	    peer->state != SCTP_UNCONFIRMED) {
727 		asoc->peer.retran_path = peer;
728 	}
729 
730 	return peer;
731 }
732 
733 /* Delete a transport address from an association.  */
734 void sctp_assoc_del_peer(struct sctp_association *asoc,
735 			 const union sctp_addr *addr)
736 {
737 	struct list_head	*pos;
738 	struct list_head	*temp;
739 	struct sctp_transport	*transport;
740 
741 	list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
742 		transport = list_entry(pos, struct sctp_transport, transports);
743 		if (sctp_cmp_addr_exact(addr, &transport->ipaddr)) {
744 			/* Do book keeping for removing the peer and free it. */
745 			sctp_assoc_rm_peer(asoc, transport);
746 			break;
747 		}
748 	}
749 }
750 
751 /* Lookup a transport by address. */
752 struct sctp_transport *sctp_assoc_lookup_paddr(
753 					const struct sctp_association *asoc,
754 					const union sctp_addr *address)
755 {
756 	struct sctp_transport *t;
757 
758 	/* Cycle through all transports searching for a peer address. */
759 
760 	list_for_each_entry(t, &asoc->peer.transport_addr_list,
761 			transports) {
762 		if (sctp_cmp_addr_exact(address, &t->ipaddr))
763 			return t;
764 	}
765 
766 	return NULL;
767 }
768 
769 /* Remove all transports except a give one */
770 void sctp_assoc_del_nonprimary_peers(struct sctp_association *asoc,
771 				     struct sctp_transport *primary)
772 {
773 	struct sctp_transport	*temp;
774 	struct sctp_transport	*t;
775 
776 	list_for_each_entry_safe(t, temp, &asoc->peer.transport_addr_list,
777 				 transports) {
778 		/* if the current transport is not the primary one, delete it */
779 		if (t != primary)
780 			sctp_assoc_rm_peer(asoc, t);
781 	}
782 }
783 
784 /* Engage in transport control operations.
785  * Mark the transport up or down and send a notification to the user.
786  * Select and update the new active and retran paths.
787  */
788 void sctp_assoc_control_transport(struct sctp_association *asoc,
789 				  struct sctp_transport *transport,
790 				  enum sctp_transport_cmd command,
791 				  sctp_sn_error_t error)
792 {
793 	struct sctp_ulpevent *event;
794 	struct sockaddr_storage addr;
795 	int spc_state = 0;
796 	bool ulp_notify = true;
797 
798 	/* Record the transition on the transport.  */
799 	switch (command) {
800 	case SCTP_TRANSPORT_UP:
801 		/* If we are moving from UNCONFIRMED state due
802 		 * to heartbeat success, report the SCTP_ADDR_CONFIRMED
803 		 * state to the user, otherwise report SCTP_ADDR_AVAILABLE.
804 		 */
805 		if (SCTP_UNCONFIRMED == transport->state &&
806 		    SCTP_HEARTBEAT_SUCCESS == error)
807 			spc_state = SCTP_ADDR_CONFIRMED;
808 		else
809 			spc_state = SCTP_ADDR_AVAILABLE;
810 		/* Don't inform ULP about transition from PF to
811 		 * active state and set cwnd to 1 MTU, see SCTP
812 		 * Quick failover draft section 5.1, point 5
813 		 */
814 		if (transport->state == SCTP_PF) {
815 			ulp_notify = false;
816 			transport->cwnd = asoc->pathmtu;
817 		}
818 		transport->state = SCTP_ACTIVE;
819 		break;
820 
821 	case SCTP_TRANSPORT_DOWN:
822 		/* If the transport was never confirmed, do not transition it
823 		 * to inactive state.  Also, release the cached route since
824 		 * there may be a better route next time.
825 		 */
826 		if (transport->state != SCTP_UNCONFIRMED)
827 			transport->state = SCTP_INACTIVE;
828 		else {
829 			sctp_transport_dst_release(transport);
830 			ulp_notify = false;
831 		}
832 
833 		spc_state = SCTP_ADDR_UNREACHABLE;
834 		break;
835 
836 	case SCTP_TRANSPORT_PF:
837 		transport->state = SCTP_PF;
838 		ulp_notify = false;
839 		break;
840 
841 	default:
842 		return;
843 	}
844 
845 	/* Generate and send a SCTP_PEER_ADDR_CHANGE notification
846 	 * to the user.
847 	 */
848 	if (ulp_notify) {
849 		memset(&addr, 0, sizeof(struct sockaddr_storage));
850 		memcpy(&addr, &transport->ipaddr,
851 		       transport->af_specific->sockaddr_len);
852 
853 		event = sctp_ulpevent_make_peer_addr_change(asoc, &addr,
854 					0, spc_state, error, GFP_ATOMIC);
855 		if (event)
856 			asoc->stream.si->enqueue_event(&asoc->ulpq, event);
857 	}
858 
859 	/* Select new active and retran paths. */
860 	sctp_select_active_and_retran_path(asoc);
861 }
862 
863 /* Hold a reference to an association. */
864 void sctp_association_hold(struct sctp_association *asoc)
865 {
866 	refcount_inc(&asoc->base.refcnt);
867 }
868 
869 /* Release a reference to an association and cleanup
870  * if there are no more references.
871  */
872 void sctp_association_put(struct sctp_association *asoc)
873 {
874 	if (refcount_dec_and_test(&asoc->base.refcnt))
875 		sctp_association_destroy(asoc);
876 }
877 
878 /* Allocate the next TSN, Transmission Sequence Number, for the given
879  * association.
880  */
881 __u32 sctp_association_get_next_tsn(struct sctp_association *asoc)
882 {
883 	/* From Section 1.6 Serial Number Arithmetic:
884 	 * Transmission Sequence Numbers wrap around when they reach
885 	 * 2**32 - 1.  That is, the next TSN a DATA chunk MUST use
886 	 * after transmitting TSN = 2*32 - 1 is TSN = 0.
887 	 */
888 	__u32 retval = asoc->next_tsn;
889 	asoc->next_tsn++;
890 	asoc->unack_data++;
891 
892 	return retval;
893 }
894 
895 /* Compare two addresses to see if they match.  Wildcard addresses
896  * only match themselves.
897  */
898 int sctp_cmp_addr_exact(const union sctp_addr *ss1,
899 			const union sctp_addr *ss2)
900 {
901 	struct sctp_af *af;
902 
903 	af = sctp_get_af_specific(ss1->sa.sa_family);
904 	if (unlikely(!af))
905 		return 0;
906 
907 	return af->cmp_addr(ss1, ss2);
908 }
909 
910 /* Return an ecne chunk to get prepended to a packet.
911  * Note:  We are sly and return a shared, prealloced chunk.  FIXME:
912  * No we don't, but we could/should.
913  */
914 struct sctp_chunk *sctp_get_ecne_prepend(struct sctp_association *asoc)
915 {
916 	if (!asoc->need_ecne)
917 		return NULL;
918 
919 	/* Send ECNE if needed.
920 	 * Not being able to allocate a chunk here is not deadly.
921 	 */
922 	return sctp_make_ecne(asoc, asoc->last_ecne_tsn);
923 }
924 
925 /*
926  * Find which transport this TSN was sent on.
927  */
928 struct sctp_transport *sctp_assoc_lookup_tsn(struct sctp_association *asoc,
929 					     __u32 tsn)
930 {
931 	struct sctp_transport *active;
932 	struct sctp_transport *match;
933 	struct sctp_transport *transport;
934 	struct sctp_chunk *chunk;
935 	__be32 key = htonl(tsn);
936 
937 	match = NULL;
938 
939 	/*
940 	 * FIXME: In general, find a more efficient data structure for
941 	 * searching.
942 	 */
943 
944 	/*
945 	 * The general strategy is to search each transport's transmitted
946 	 * list.   Return which transport this TSN lives on.
947 	 *
948 	 * Let's be hopeful and check the active_path first.
949 	 * Another optimization would be to know if there is only one
950 	 * outbound path and not have to look for the TSN at all.
951 	 *
952 	 */
953 
954 	active = asoc->peer.active_path;
955 
956 	list_for_each_entry(chunk, &active->transmitted,
957 			transmitted_list) {
958 
959 		if (key == chunk->subh.data_hdr->tsn) {
960 			match = active;
961 			goto out;
962 		}
963 	}
964 
965 	/* If not found, go search all the other transports. */
966 	list_for_each_entry(transport, &asoc->peer.transport_addr_list,
967 			transports) {
968 
969 		if (transport == active)
970 			continue;
971 		list_for_each_entry(chunk, &transport->transmitted,
972 				transmitted_list) {
973 			if (key == chunk->subh.data_hdr->tsn) {
974 				match = transport;
975 				goto out;
976 			}
977 		}
978 	}
979 out:
980 	return match;
981 }
982 
983 /* Do delayed input processing.  This is scheduled by sctp_rcv(). */
984 static void sctp_assoc_bh_rcv(struct work_struct *work)
985 {
986 	struct sctp_association *asoc =
987 		container_of(work, struct sctp_association,
988 			     base.inqueue.immediate);
989 	struct net *net = sock_net(asoc->base.sk);
990 	union sctp_subtype subtype;
991 	struct sctp_endpoint *ep;
992 	struct sctp_chunk *chunk;
993 	struct sctp_inq *inqueue;
994 	int first_time = 1;	/* is this the first time through the loop */
995 	int error = 0;
996 	int state;
997 
998 	/* The association should be held so we should be safe. */
999 	ep = asoc->ep;
1000 
1001 	inqueue = &asoc->base.inqueue;
1002 	sctp_association_hold(asoc);
1003 	while (NULL != (chunk = sctp_inq_pop(inqueue))) {
1004 		state = asoc->state;
1005 		subtype = SCTP_ST_CHUNK(chunk->chunk_hdr->type);
1006 
1007 		/* If the first chunk in the packet is AUTH, do special
1008 		 * processing specified in Section 6.3 of SCTP-AUTH spec
1009 		 */
1010 		if (first_time && subtype.chunk == SCTP_CID_AUTH) {
1011 			struct sctp_chunkhdr *next_hdr;
1012 
1013 			next_hdr = sctp_inq_peek(inqueue);
1014 			if (!next_hdr)
1015 				goto normal;
1016 
1017 			/* If the next chunk is COOKIE-ECHO, skip the AUTH
1018 			 * chunk while saving a pointer to it so we can do
1019 			 * Authentication later (during cookie-echo
1020 			 * processing).
1021 			 */
1022 			if (next_hdr->type == SCTP_CID_COOKIE_ECHO) {
1023 				chunk->auth_chunk = skb_clone(chunk->skb,
1024 							      GFP_ATOMIC);
1025 				chunk->auth = 1;
1026 				continue;
1027 			}
1028 		}
1029 
1030 normal:
1031 		/* SCTP-AUTH, Section 6.3:
1032 		 *    The receiver has a list of chunk types which it expects
1033 		 *    to be received only after an AUTH-chunk.  This list has
1034 		 *    been sent to the peer during the association setup.  It
1035 		 *    MUST silently discard these chunks if they are not placed
1036 		 *    after an AUTH chunk in the packet.
1037 		 */
1038 		if (sctp_auth_recv_cid(subtype.chunk, asoc) && !chunk->auth)
1039 			continue;
1040 
1041 		/* Remember where the last DATA chunk came from so we
1042 		 * know where to send the SACK.
1043 		 */
1044 		if (sctp_chunk_is_data(chunk))
1045 			asoc->peer.last_data_from = chunk->transport;
1046 		else {
1047 			SCTP_INC_STATS(net, SCTP_MIB_INCTRLCHUNKS);
1048 			asoc->stats.ictrlchunks++;
1049 			if (chunk->chunk_hdr->type == SCTP_CID_SACK)
1050 				asoc->stats.isacks++;
1051 		}
1052 
1053 		if (chunk->transport)
1054 			chunk->transport->last_time_heard = ktime_get();
1055 
1056 		/* Run through the state machine. */
1057 		error = sctp_do_sm(net, SCTP_EVENT_T_CHUNK, subtype,
1058 				   state, ep, asoc, chunk, GFP_ATOMIC);
1059 
1060 		/* Check to see if the association is freed in response to
1061 		 * the incoming chunk.  If so, get out of the while loop.
1062 		 */
1063 		if (asoc->base.dead)
1064 			break;
1065 
1066 		/* If there is an error on chunk, discard this packet. */
1067 		if (error && chunk)
1068 			chunk->pdiscard = 1;
1069 
1070 		if (first_time)
1071 			first_time = 0;
1072 	}
1073 	sctp_association_put(asoc);
1074 }
1075 
1076 /* This routine moves an association from its old sk to a new sk.  */
1077 void sctp_assoc_migrate(struct sctp_association *assoc, struct sock *newsk)
1078 {
1079 	struct sctp_sock *newsp = sctp_sk(newsk);
1080 	struct sock *oldsk = assoc->base.sk;
1081 
1082 	/* Delete the association from the old endpoint's list of
1083 	 * associations.
1084 	 */
1085 	list_del_init(&assoc->asocs);
1086 
1087 	/* Decrement the backlog value for a TCP-style socket. */
1088 	if (sctp_style(oldsk, TCP))
1089 		oldsk->sk_ack_backlog--;
1090 
1091 	/* Release references to the old endpoint and the sock.  */
1092 	sctp_endpoint_put(assoc->ep);
1093 	sock_put(assoc->base.sk);
1094 
1095 	/* Get a reference to the new endpoint.  */
1096 	assoc->ep = newsp->ep;
1097 	sctp_endpoint_hold(assoc->ep);
1098 
1099 	/* Get a reference to the new sock.  */
1100 	assoc->base.sk = newsk;
1101 	sock_hold(assoc->base.sk);
1102 
1103 	/* Add the association to the new endpoint's list of associations.  */
1104 	sctp_endpoint_add_asoc(newsp->ep, assoc);
1105 }
1106 
1107 /* Update an association (possibly from unexpected COOKIE-ECHO processing).  */
1108 int sctp_assoc_update(struct sctp_association *asoc,
1109 		      struct sctp_association *new)
1110 {
1111 	struct sctp_transport *trans;
1112 	struct list_head *pos, *temp;
1113 
1114 	/* Copy in new parameters of peer. */
1115 	asoc->c = new->c;
1116 	asoc->peer.rwnd = new->peer.rwnd;
1117 	asoc->peer.sack_needed = new->peer.sack_needed;
1118 	asoc->peer.auth_capable = new->peer.auth_capable;
1119 	asoc->peer.i = new->peer.i;
1120 
1121 	if (!sctp_tsnmap_init(&asoc->peer.tsn_map, SCTP_TSN_MAP_INITIAL,
1122 			      asoc->peer.i.initial_tsn, GFP_ATOMIC))
1123 		return -ENOMEM;
1124 
1125 	/* Remove any peer addresses not present in the new association. */
1126 	list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
1127 		trans = list_entry(pos, struct sctp_transport, transports);
1128 		if (!sctp_assoc_lookup_paddr(new, &trans->ipaddr)) {
1129 			sctp_assoc_rm_peer(asoc, trans);
1130 			continue;
1131 		}
1132 
1133 		if (asoc->state >= SCTP_STATE_ESTABLISHED)
1134 			sctp_transport_reset(trans);
1135 	}
1136 
1137 	/* If the case is A (association restart), use
1138 	 * initial_tsn as next_tsn. If the case is B, use
1139 	 * current next_tsn in case data sent to peer
1140 	 * has been discarded and needs retransmission.
1141 	 */
1142 	if (asoc->state >= SCTP_STATE_ESTABLISHED) {
1143 		asoc->next_tsn = new->next_tsn;
1144 		asoc->ctsn_ack_point = new->ctsn_ack_point;
1145 		asoc->adv_peer_ack_point = new->adv_peer_ack_point;
1146 
1147 		/* Reinitialize SSN for both local streams
1148 		 * and peer's streams.
1149 		 */
1150 		sctp_stream_clear(&asoc->stream);
1151 
1152 		/* Flush the ULP reassembly and ordered queue.
1153 		 * Any data there will now be stale and will
1154 		 * cause problems.
1155 		 */
1156 		sctp_ulpq_flush(&asoc->ulpq);
1157 
1158 		/* reset the overall association error count so
1159 		 * that the restarted association doesn't get torn
1160 		 * down on the next retransmission timer.
1161 		 */
1162 		asoc->overall_error_count = 0;
1163 
1164 	} else {
1165 		/* Add any peer addresses from the new association. */
1166 		list_for_each_entry(trans, &new->peer.transport_addr_list,
1167 				    transports)
1168 			if (!sctp_assoc_lookup_paddr(asoc, &trans->ipaddr) &&
1169 			    !sctp_assoc_add_peer(asoc, &trans->ipaddr,
1170 						 GFP_ATOMIC, trans->state))
1171 				return -ENOMEM;
1172 
1173 		asoc->ctsn_ack_point = asoc->next_tsn - 1;
1174 		asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
1175 
1176 		if (sctp_state(asoc, COOKIE_WAIT))
1177 			sctp_stream_update(&asoc->stream, &new->stream);
1178 
1179 		/* get a new assoc id if we don't have one yet. */
1180 		if (sctp_assoc_set_id(asoc, GFP_ATOMIC))
1181 			return -ENOMEM;
1182 	}
1183 
1184 	/* SCTP-AUTH: Save the peer parameters from the new associations
1185 	 * and also move the association shared keys over
1186 	 */
1187 	kfree(asoc->peer.peer_random);
1188 	asoc->peer.peer_random = new->peer.peer_random;
1189 	new->peer.peer_random = NULL;
1190 
1191 	kfree(asoc->peer.peer_chunks);
1192 	asoc->peer.peer_chunks = new->peer.peer_chunks;
1193 	new->peer.peer_chunks = NULL;
1194 
1195 	kfree(asoc->peer.peer_hmacs);
1196 	asoc->peer.peer_hmacs = new->peer.peer_hmacs;
1197 	new->peer.peer_hmacs = NULL;
1198 
1199 	return sctp_auth_asoc_init_active_key(asoc, GFP_ATOMIC);
1200 }
1201 
1202 /* Update the retran path for sending a retransmitted packet.
1203  * See also RFC4960, 6.4. Multi-Homed SCTP Endpoints:
1204  *
1205  *   When there is outbound data to send and the primary path
1206  *   becomes inactive (e.g., due to failures), or where the
1207  *   SCTP user explicitly requests to send data to an
1208  *   inactive destination transport address, before reporting
1209  *   an error to its ULP, the SCTP endpoint should try to send
1210  *   the data to an alternate active destination transport
1211  *   address if one exists.
1212  *
1213  *   When retransmitting data that timed out, if the endpoint
1214  *   is multihomed, it should consider each source-destination
1215  *   address pair in its retransmission selection policy.
1216  *   When retransmitting timed-out data, the endpoint should
1217  *   attempt to pick the most divergent source-destination
1218  *   pair from the original source-destination pair to which
1219  *   the packet was transmitted.
1220  *
1221  *   Note: Rules for picking the most divergent source-destination
1222  *   pair are an implementation decision and are not specified
1223  *   within this document.
1224  *
1225  * Our basic strategy is to round-robin transports in priorities
1226  * according to sctp_trans_score() e.g., if no such
1227  * transport with state SCTP_ACTIVE exists, round-robin through
1228  * SCTP_UNKNOWN, etc. You get the picture.
1229  */
1230 static u8 sctp_trans_score(const struct sctp_transport *trans)
1231 {
1232 	switch (trans->state) {
1233 	case SCTP_ACTIVE:
1234 		return 3;	/* best case */
1235 	case SCTP_UNKNOWN:
1236 		return 2;
1237 	case SCTP_PF:
1238 		return 1;
1239 	default: /* case SCTP_INACTIVE */
1240 		return 0;	/* worst case */
1241 	}
1242 }
1243 
1244 static struct sctp_transport *sctp_trans_elect_tie(struct sctp_transport *trans1,
1245 						   struct sctp_transport *trans2)
1246 {
1247 	if (trans1->error_count > trans2->error_count) {
1248 		return trans2;
1249 	} else if (trans1->error_count == trans2->error_count &&
1250 		   ktime_after(trans2->last_time_heard,
1251 			       trans1->last_time_heard)) {
1252 		return trans2;
1253 	} else {
1254 		return trans1;
1255 	}
1256 }
1257 
1258 static struct sctp_transport *sctp_trans_elect_best(struct sctp_transport *curr,
1259 						    struct sctp_transport *best)
1260 {
1261 	u8 score_curr, score_best;
1262 
1263 	if (best == NULL || curr == best)
1264 		return curr;
1265 
1266 	score_curr = sctp_trans_score(curr);
1267 	score_best = sctp_trans_score(best);
1268 
1269 	/* First, try a score-based selection if both transport states
1270 	 * differ. If we're in a tie, lets try to make a more clever
1271 	 * decision here based on error counts and last time heard.
1272 	 */
1273 	if (score_curr > score_best)
1274 		return curr;
1275 	else if (score_curr == score_best)
1276 		return sctp_trans_elect_tie(best, curr);
1277 	else
1278 		return best;
1279 }
1280 
1281 void sctp_assoc_update_retran_path(struct sctp_association *asoc)
1282 {
1283 	struct sctp_transport *trans = asoc->peer.retran_path;
1284 	struct sctp_transport *trans_next = NULL;
1285 
1286 	/* We're done as we only have the one and only path. */
1287 	if (asoc->peer.transport_count == 1)
1288 		return;
1289 	/* If active_path and retran_path are the same and active,
1290 	 * then this is the only active path. Use it.
1291 	 */
1292 	if (asoc->peer.active_path == asoc->peer.retran_path &&
1293 	    asoc->peer.active_path->state == SCTP_ACTIVE)
1294 		return;
1295 
1296 	/* Iterate from retran_path's successor back to retran_path. */
1297 	for (trans = list_next_entry(trans, transports); 1;
1298 	     trans = list_next_entry(trans, transports)) {
1299 		/* Manually skip the head element. */
1300 		if (&trans->transports == &asoc->peer.transport_addr_list)
1301 			continue;
1302 		if (trans->state == SCTP_UNCONFIRMED)
1303 			continue;
1304 		trans_next = sctp_trans_elect_best(trans, trans_next);
1305 		/* Active is good enough for immediate return. */
1306 		if (trans_next->state == SCTP_ACTIVE)
1307 			break;
1308 		/* We've reached the end, time to update path. */
1309 		if (trans == asoc->peer.retran_path)
1310 			break;
1311 	}
1312 
1313 	asoc->peer.retran_path = trans_next;
1314 
1315 	pr_debug("%s: association:%p updated new path to addr:%pISpc\n",
1316 		 __func__, asoc, &asoc->peer.retran_path->ipaddr.sa);
1317 }
1318 
1319 static void sctp_select_active_and_retran_path(struct sctp_association *asoc)
1320 {
1321 	struct sctp_transport *trans, *trans_pri = NULL, *trans_sec = NULL;
1322 	struct sctp_transport *trans_pf = NULL;
1323 
1324 	/* Look for the two most recently used active transports. */
1325 	list_for_each_entry(trans, &asoc->peer.transport_addr_list,
1326 			    transports) {
1327 		/* Skip uninteresting transports. */
1328 		if (trans->state == SCTP_INACTIVE ||
1329 		    trans->state == SCTP_UNCONFIRMED)
1330 			continue;
1331 		/* Keep track of the best PF transport from our
1332 		 * list in case we don't find an active one.
1333 		 */
1334 		if (trans->state == SCTP_PF) {
1335 			trans_pf = sctp_trans_elect_best(trans, trans_pf);
1336 			continue;
1337 		}
1338 		/* For active transports, pick the most recent ones. */
1339 		if (trans_pri == NULL ||
1340 		    ktime_after(trans->last_time_heard,
1341 				trans_pri->last_time_heard)) {
1342 			trans_sec = trans_pri;
1343 			trans_pri = trans;
1344 		} else if (trans_sec == NULL ||
1345 			   ktime_after(trans->last_time_heard,
1346 				       trans_sec->last_time_heard)) {
1347 			trans_sec = trans;
1348 		}
1349 	}
1350 
1351 	/* RFC 2960 6.4 Multi-Homed SCTP Endpoints
1352 	 *
1353 	 * By default, an endpoint should always transmit to the primary
1354 	 * path, unless the SCTP user explicitly specifies the
1355 	 * destination transport address (and possibly source transport
1356 	 * address) to use. [If the primary is active but not most recent,
1357 	 * bump the most recently used transport.]
1358 	 */
1359 	if ((asoc->peer.primary_path->state == SCTP_ACTIVE ||
1360 	     asoc->peer.primary_path->state == SCTP_UNKNOWN) &&
1361 	     asoc->peer.primary_path != trans_pri) {
1362 		trans_sec = trans_pri;
1363 		trans_pri = asoc->peer.primary_path;
1364 	}
1365 
1366 	/* We did not find anything useful for a possible retransmission
1367 	 * path; either primary path that we found is the the same as
1368 	 * the current one, or we didn't generally find an active one.
1369 	 */
1370 	if (trans_sec == NULL)
1371 		trans_sec = trans_pri;
1372 
1373 	/* If we failed to find a usable transport, just camp on the
1374 	 * active or pick a PF iff it's the better choice.
1375 	 */
1376 	if (trans_pri == NULL) {
1377 		trans_pri = sctp_trans_elect_best(asoc->peer.active_path, trans_pf);
1378 		trans_sec = trans_pri;
1379 	}
1380 
1381 	/* Set the active and retran transports. */
1382 	asoc->peer.active_path = trans_pri;
1383 	asoc->peer.retran_path = trans_sec;
1384 }
1385 
1386 struct sctp_transport *
1387 sctp_assoc_choose_alter_transport(struct sctp_association *asoc,
1388 				  struct sctp_transport *last_sent_to)
1389 {
1390 	/* If this is the first time packet is sent, use the active path,
1391 	 * else use the retran path. If the last packet was sent over the
1392 	 * retran path, update the retran path and use it.
1393 	 */
1394 	if (last_sent_to == NULL) {
1395 		return asoc->peer.active_path;
1396 	} else {
1397 		if (last_sent_to == asoc->peer.retran_path)
1398 			sctp_assoc_update_retran_path(asoc);
1399 
1400 		return asoc->peer.retran_path;
1401 	}
1402 }
1403 
1404 void sctp_assoc_update_frag_point(struct sctp_association *asoc)
1405 {
1406 	int frag = sctp_mtu_payload(sctp_sk(asoc->base.sk), asoc->pathmtu,
1407 				    sctp_datachk_len(&asoc->stream));
1408 
1409 	if (asoc->user_frag)
1410 		frag = min_t(int, frag, asoc->user_frag);
1411 
1412 	frag = min_t(int, frag, SCTP_MAX_CHUNK_LEN -
1413 				sctp_datachk_len(&asoc->stream));
1414 
1415 	asoc->frag_point = SCTP_TRUNC4(frag);
1416 }
1417 
1418 void sctp_assoc_set_pmtu(struct sctp_association *asoc, __u32 pmtu)
1419 {
1420 	if (asoc->pathmtu != pmtu) {
1421 		asoc->pathmtu = pmtu;
1422 		sctp_assoc_update_frag_point(asoc);
1423 	}
1424 
1425 	pr_debug("%s: asoc:%p, pmtu:%d, frag_point:%d\n", __func__, asoc,
1426 		 asoc->pathmtu, asoc->frag_point);
1427 }
1428 
1429 /* Update the association's pmtu and frag_point by going through all the
1430  * transports. This routine is called when a transport's PMTU has changed.
1431  */
1432 void sctp_assoc_sync_pmtu(struct sctp_association *asoc)
1433 {
1434 	struct sctp_transport *t;
1435 	__u32 pmtu = 0;
1436 
1437 	if (!asoc)
1438 		return;
1439 
1440 	/* Get the lowest pmtu of all the transports. */
1441 	list_for_each_entry(t, &asoc->peer.transport_addr_list, transports) {
1442 		if (t->pmtu_pending && t->dst) {
1443 			sctp_transport_update_pmtu(t,
1444 						   atomic_read(&t->mtu_info));
1445 			t->pmtu_pending = 0;
1446 		}
1447 		if (!pmtu || (t->pathmtu < pmtu))
1448 			pmtu = t->pathmtu;
1449 	}
1450 
1451 	sctp_assoc_set_pmtu(asoc, pmtu);
1452 }
1453 
1454 /* Should we send a SACK to update our peer? */
1455 static inline bool sctp_peer_needs_update(struct sctp_association *asoc)
1456 {
1457 	struct net *net = sock_net(asoc->base.sk);
1458 	switch (asoc->state) {
1459 	case SCTP_STATE_ESTABLISHED:
1460 	case SCTP_STATE_SHUTDOWN_PENDING:
1461 	case SCTP_STATE_SHUTDOWN_RECEIVED:
1462 	case SCTP_STATE_SHUTDOWN_SENT:
1463 		if ((asoc->rwnd > asoc->a_rwnd) &&
1464 		    ((asoc->rwnd - asoc->a_rwnd) >= max_t(__u32,
1465 			   (asoc->base.sk->sk_rcvbuf >> net->sctp.rwnd_upd_shift),
1466 			   asoc->pathmtu)))
1467 			return true;
1468 		break;
1469 	default:
1470 		break;
1471 	}
1472 	return false;
1473 }
1474 
1475 /* Increase asoc's rwnd by len and send any window update SACK if needed. */
1476 void sctp_assoc_rwnd_increase(struct sctp_association *asoc, unsigned int len)
1477 {
1478 	struct sctp_chunk *sack;
1479 	struct timer_list *timer;
1480 
1481 	if (asoc->rwnd_over) {
1482 		if (asoc->rwnd_over >= len) {
1483 			asoc->rwnd_over -= len;
1484 		} else {
1485 			asoc->rwnd += (len - asoc->rwnd_over);
1486 			asoc->rwnd_over = 0;
1487 		}
1488 	} else {
1489 		asoc->rwnd += len;
1490 	}
1491 
1492 	/* If we had window pressure, start recovering it
1493 	 * once our rwnd had reached the accumulated pressure
1494 	 * threshold.  The idea is to recover slowly, but up
1495 	 * to the initial advertised window.
1496 	 */
1497 	if (asoc->rwnd_press) {
1498 		int change = min(asoc->pathmtu, asoc->rwnd_press);
1499 		asoc->rwnd += change;
1500 		asoc->rwnd_press -= change;
1501 	}
1502 
1503 	pr_debug("%s: asoc:%p rwnd increased by %d to (%u, %u) - %u\n",
1504 		 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
1505 		 asoc->a_rwnd);
1506 
1507 	/* Send a window update SACK if the rwnd has increased by at least the
1508 	 * minimum of the association's PMTU and half of the receive buffer.
1509 	 * The algorithm used is similar to the one described in
1510 	 * Section 4.2.3.3 of RFC 1122.
1511 	 */
1512 	if (sctp_peer_needs_update(asoc)) {
1513 		asoc->a_rwnd = asoc->rwnd;
1514 
1515 		pr_debug("%s: sending window update SACK- asoc:%p rwnd:%u "
1516 			 "a_rwnd:%u\n", __func__, asoc, asoc->rwnd,
1517 			 asoc->a_rwnd);
1518 
1519 		sack = sctp_make_sack(asoc);
1520 		if (!sack)
1521 			return;
1522 
1523 		asoc->peer.sack_needed = 0;
1524 
1525 		sctp_outq_tail(&asoc->outqueue, sack, GFP_ATOMIC);
1526 
1527 		/* Stop the SACK timer.  */
1528 		timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
1529 		if (del_timer(timer))
1530 			sctp_association_put(asoc);
1531 	}
1532 }
1533 
1534 /* Decrease asoc's rwnd by len. */
1535 void sctp_assoc_rwnd_decrease(struct sctp_association *asoc, unsigned int len)
1536 {
1537 	int rx_count;
1538 	int over = 0;
1539 
1540 	if (unlikely(!asoc->rwnd || asoc->rwnd_over))
1541 		pr_debug("%s: association:%p has asoc->rwnd:%u, "
1542 			 "asoc->rwnd_over:%u!\n", __func__, asoc,
1543 			 asoc->rwnd, asoc->rwnd_over);
1544 
1545 	if (asoc->ep->rcvbuf_policy)
1546 		rx_count = atomic_read(&asoc->rmem_alloc);
1547 	else
1548 		rx_count = atomic_read(&asoc->base.sk->sk_rmem_alloc);
1549 
1550 	/* If we've reached or overflowed our receive buffer, announce
1551 	 * a 0 rwnd if rwnd would still be positive.  Store the
1552 	 * the potential pressure overflow so that the window can be restored
1553 	 * back to original value.
1554 	 */
1555 	if (rx_count >= asoc->base.sk->sk_rcvbuf)
1556 		over = 1;
1557 
1558 	if (asoc->rwnd >= len) {
1559 		asoc->rwnd -= len;
1560 		if (over) {
1561 			asoc->rwnd_press += asoc->rwnd;
1562 			asoc->rwnd = 0;
1563 		}
1564 	} else {
1565 		asoc->rwnd_over += len - asoc->rwnd;
1566 		asoc->rwnd = 0;
1567 	}
1568 
1569 	pr_debug("%s: asoc:%p rwnd decreased by %d to (%u, %u, %u)\n",
1570 		 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
1571 		 asoc->rwnd_press);
1572 }
1573 
1574 /* Build the bind address list for the association based on info from the
1575  * local endpoint and the remote peer.
1576  */
1577 int sctp_assoc_set_bind_addr_from_ep(struct sctp_association *asoc,
1578 				     enum sctp_scope scope, gfp_t gfp)
1579 {
1580 	int flags;
1581 
1582 	/* Use scoping rules to determine the subset of addresses from
1583 	 * the endpoint.
1584 	 */
1585 	flags = (PF_INET6 == asoc->base.sk->sk_family) ? SCTP_ADDR6_ALLOWED : 0;
1586 	if (asoc->peer.ipv4_address)
1587 		flags |= SCTP_ADDR4_PEERSUPP;
1588 	if (asoc->peer.ipv6_address)
1589 		flags |= SCTP_ADDR6_PEERSUPP;
1590 
1591 	return sctp_bind_addr_copy(sock_net(asoc->base.sk),
1592 				   &asoc->base.bind_addr,
1593 				   &asoc->ep->base.bind_addr,
1594 				   scope, gfp, flags);
1595 }
1596 
1597 /* Build the association's bind address list from the cookie.  */
1598 int sctp_assoc_set_bind_addr_from_cookie(struct sctp_association *asoc,
1599 					 struct sctp_cookie *cookie,
1600 					 gfp_t gfp)
1601 {
1602 	int var_size2 = ntohs(cookie->peer_init->chunk_hdr.length);
1603 	int var_size3 = cookie->raw_addr_list_len;
1604 	__u8 *raw = (__u8 *)cookie->peer_init + var_size2;
1605 
1606 	return sctp_raw_to_bind_addrs(&asoc->base.bind_addr, raw, var_size3,
1607 				      asoc->ep->base.bind_addr.port, gfp);
1608 }
1609 
1610 /* Lookup laddr in the bind address list of an association. */
1611 int sctp_assoc_lookup_laddr(struct sctp_association *asoc,
1612 			    const union sctp_addr *laddr)
1613 {
1614 	int found = 0;
1615 
1616 	if ((asoc->base.bind_addr.port == ntohs(laddr->v4.sin_port)) &&
1617 	    sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
1618 				 sctp_sk(asoc->base.sk)))
1619 		found = 1;
1620 
1621 	return found;
1622 }
1623 
1624 /* Set an association id for a given association */
1625 int sctp_assoc_set_id(struct sctp_association *asoc, gfp_t gfp)
1626 {
1627 	bool preload = gfpflags_allow_blocking(gfp);
1628 	int ret;
1629 
1630 	/* If the id is already assigned, keep it. */
1631 	if (asoc->assoc_id)
1632 		return 0;
1633 
1634 	if (preload)
1635 		idr_preload(gfp);
1636 	spin_lock_bh(&sctp_assocs_id_lock);
1637 	/* 0, 1, 2 are used as SCTP_FUTURE_ASSOC, SCTP_CURRENT_ASSOC and
1638 	 * SCTP_ALL_ASSOC, so an available id must be > SCTP_ALL_ASSOC.
1639 	 */
1640 	ret = idr_alloc_cyclic(&sctp_assocs_id, asoc, SCTP_ALL_ASSOC + 1, 0,
1641 			       GFP_NOWAIT);
1642 	spin_unlock_bh(&sctp_assocs_id_lock);
1643 	if (preload)
1644 		idr_preload_end();
1645 	if (ret < 0)
1646 		return ret;
1647 
1648 	asoc->assoc_id = (sctp_assoc_t)ret;
1649 	return 0;
1650 }
1651 
1652 /* Free the ASCONF queue */
1653 static void sctp_assoc_free_asconf_queue(struct sctp_association *asoc)
1654 {
1655 	struct sctp_chunk *asconf;
1656 	struct sctp_chunk *tmp;
1657 
1658 	list_for_each_entry_safe(asconf, tmp, &asoc->addip_chunk_list, list) {
1659 		list_del_init(&asconf->list);
1660 		sctp_chunk_free(asconf);
1661 	}
1662 }
1663 
1664 /* Free asconf_ack cache */
1665 static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc)
1666 {
1667 	struct sctp_chunk *ack;
1668 	struct sctp_chunk *tmp;
1669 
1670 	list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
1671 				transmitted_list) {
1672 		list_del_init(&ack->transmitted_list);
1673 		sctp_chunk_free(ack);
1674 	}
1675 }
1676 
1677 /* Clean up the ASCONF_ACK queue */
1678 void sctp_assoc_clean_asconf_ack_cache(const struct sctp_association *asoc)
1679 {
1680 	struct sctp_chunk *ack;
1681 	struct sctp_chunk *tmp;
1682 
1683 	/* We can remove all the entries from the queue up to
1684 	 * the "Peer-Sequence-Number".
1685 	 */
1686 	list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
1687 				transmitted_list) {
1688 		if (ack->subh.addip_hdr->serial ==
1689 				htonl(asoc->peer.addip_serial))
1690 			break;
1691 
1692 		list_del_init(&ack->transmitted_list);
1693 		sctp_chunk_free(ack);
1694 	}
1695 }
1696 
1697 /* Find the ASCONF_ACK whose serial number matches ASCONF */
1698 struct sctp_chunk *sctp_assoc_lookup_asconf_ack(
1699 					const struct sctp_association *asoc,
1700 					__be32 serial)
1701 {
1702 	struct sctp_chunk *ack;
1703 
1704 	/* Walk through the list of cached ASCONF-ACKs and find the
1705 	 * ack chunk whose serial number matches that of the request.
1706 	 */
1707 	list_for_each_entry(ack, &asoc->asconf_ack_list, transmitted_list) {
1708 		if (sctp_chunk_pending(ack))
1709 			continue;
1710 		if (ack->subh.addip_hdr->serial == serial) {
1711 			sctp_chunk_hold(ack);
1712 			return ack;
1713 		}
1714 	}
1715 
1716 	return NULL;
1717 }
1718 
1719 void sctp_asconf_queue_teardown(struct sctp_association *asoc)
1720 {
1721 	/* Free any cached ASCONF_ACK chunk. */
1722 	sctp_assoc_free_asconf_acks(asoc);
1723 
1724 	/* Free the ASCONF queue. */
1725 	sctp_assoc_free_asconf_queue(asoc);
1726 
1727 	/* Free any cached ASCONF chunk. */
1728 	if (asoc->addip_last_asconf)
1729 		sctp_chunk_free(asoc->addip_last_asconf);
1730 }
1731