xref: /linux/net/sctp/associola.c (revision 4359a011e259a4608afc7fb3635370c9d4ba5943)
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 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] = sp->autoclose * HZ;
141 
142 	/* Initializes the timers */
143 	for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i)
144 		timer_setup(&asoc->timers[i], sctp_timer_events[i], 0);
145 
146 	/* Pull default initialization values from the sock options.
147 	 * Note: This assumes that the values have already been
148 	 * validated in the sock.
149 	 */
150 	asoc->c.sinit_max_instreams = sp->initmsg.sinit_max_instreams;
151 	asoc->c.sinit_num_ostreams  = sp->initmsg.sinit_num_ostreams;
152 	asoc->max_init_attempts	= sp->initmsg.sinit_max_attempts;
153 
154 	asoc->max_init_timeo =
155 		 msecs_to_jiffies(sp->initmsg.sinit_max_init_timeo);
156 
157 	/* Set the local window size for receive.
158 	 * This is also the rcvbuf space per association.
159 	 * RFC 6 - A SCTP receiver MUST be able to receive a minimum of
160 	 * 1500 bytes in one SCTP packet.
161 	 */
162 	if ((sk->sk_rcvbuf/2) < SCTP_DEFAULT_MINWINDOW)
163 		asoc->rwnd = SCTP_DEFAULT_MINWINDOW;
164 	else
165 		asoc->rwnd = sk->sk_rcvbuf/2;
166 
167 	asoc->a_rwnd = asoc->rwnd;
168 
169 	/* Use my own max window until I learn something better.  */
170 	asoc->peer.rwnd = SCTP_DEFAULT_MAXWINDOW;
171 
172 	/* Initialize the receive memory counter */
173 	atomic_set(&asoc->rmem_alloc, 0);
174 
175 	init_waitqueue_head(&asoc->wait);
176 
177 	asoc->c.my_vtag = sctp_generate_tag(ep);
178 	asoc->c.my_port = ep->base.bind_addr.port;
179 
180 	asoc->c.initial_tsn = sctp_generate_tsn(ep);
181 
182 	asoc->next_tsn = asoc->c.initial_tsn;
183 
184 	asoc->ctsn_ack_point = asoc->next_tsn - 1;
185 	asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
186 	asoc->highest_sacked = asoc->ctsn_ack_point;
187 	asoc->last_cwr_tsn = asoc->ctsn_ack_point;
188 
189 	/* ADDIP Section 4.1 Asconf Chunk Procedures
190 	 *
191 	 * When an endpoint has an ASCONF signaled change to be sent to the
192 	 * remote endpoint it should do the following:
193 	 * ...
194 	 * A2) a serial number should be assigned to the chunk. The serial
195 	 * number SHOULD be a monotonically increasing number. The serial
196 	 * numbers SHOULD be initialized at the start of the
197 	 * association to the same value as the initial TSN.
198 	 */
199 	asoc->addip_serial = asoc->c.initial_tsn;
200 	asoc->strreset_outseq = asoc->c.initial_tsn;
201 
202 	INIT_LIST_HEAD(&asoc->addip_chunk_list);
203 	INIT_LIST_HEAD(&asoc->asconf_ack_list);
204 
205 	/* Make an empty list of remote transport addresses.  */
206 	INIT_LIST_HEAD(&asoc->peer.transport_addr_list);
207 
208 	/* RFC 2960 5.1 Normal Establishment of an Association
209 	 *
210 	 * After the reception of the first data chunk in an
211 	 * association the endpoint must immediately respond with a
212 	 * sack to acknowledge the data chunk.  Subsequent
213 	 * acknowledgements should be done as described in Section
214 	 * 6.2.
215 	 *
216 	 * [We implement this by telling a new association that it
217 	 * already received one packet.]
218 	 */
219 	asoc->peer.sack_needed = 1;
220 	asoc->peer.sack_generation = 1;
221 
222 	/* Create an input queue.  */
223 	sctp_inq_init(&asoc->base.inqueue);
224 	sctp_inq_set_th_handler(&asoc->base.inqueue, sctp_assoc_bh_rcv);
225 
226 	/* Create an output queue.  */
227 	sctp_outq_init(asoc, &asoc->outqueue);
228 
229 	if (!sctp_ulpq_init(&asoc->ulpq, asoc))
230 		goto fail_init;
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 
278 stream_free:
279 	sctp_stream_free(&asoc->stream);
280 fail_init:
281 	sock_put(asoc->base.sk);
282 	sctp_endpoint_put(asoc->ep);
283 	return NULL;
284 }
285 
286 /* Allocate and initialize a new association */
287 struct sctp_association *sctp_association_new(const struct sctp_endpoint *ep,
288 					      const struct sock *sk,
289 					      enum sctp_scope scope, gfp_t gfp)
290 {
291 	struct sctp_association *asoc;
292 
293 	asoc = kzalloc(sizeof(*asoc), gfp);
294 	if (!asoc)
295 		goto fail;
296 
297 	if (!sctp_association_init(asoc, ep, sk, scope, gfp))
298 		goto fail_init;
299 
300 	SCTP_DBG_OBJCNT_INC(assoc);
301 
302 	pr_debug("Created asoc %p\n", asoc);
303 
304 	return asoc;
305 
306 fail_init:
307 	kfree(asoc);
308 fail:
309 	return NULL;
310 }
311 
312 /* Free this association if possible.  There may still be users, so
313  * the actual deallocation may be delayed.
314  */
315 void sctp_association_free(struct sctp_association *asoc)
316 {
317 	struct sock *sk = asoc->base.sk;
318 	struct sctp_transport *transport;
319 	struct list_head *pos, *temp;
320 	int i;
321 
322 	/* Only real associations count against the endpoint, so
323 	 * don't bother for if this is a temporary association.
324 	 */
325 	if (!list_empty(&asoc->asocs)) {
326 		list_del(&asoc->asocs);
327 
328 		/* Decrement the backlog value for a TCP-style listening
329 		 * socket.
330 		 */
331 		if (sctp_style(sk, TCP) && sctp_sstate(sk, LISTENING))
332 			sk_acceptq_removed(sk);
333 	}
334 
335 	/* Mark as dead, so other users can know this structure is
336 	 * going away.
337 	 */
338 	asoc->base.dead = true;
339 
340 	/* Dispose of any data lying around in the outqueue. */
341 	sctp_outq_free(&asoc->outqueue);
342 
343 	/* Dispose of any pending messages for the upper layer. */
344 	sctp_ulpq_free(&asoc->ulpq);
345 
346 	/* Dispose of any pending chunks on the inqueue. */
347 	sctp_inq_free(&asoc->base.inqueue);
348 
349 	sctp_tsnmap_free(&asoc->peer.tsn_map);
350 
351 	/* Free stream information. */
352 	sctp_stream_free(&asoc->stream);
353 
354 	if (asoc->strreset_chunk)
355 		sctp_chunk_free(asoc->strreset_chunk);
356 
357 	/* Clean up the bound address list. */
358 	sctp_bind_addr_free(&asoc->base.bind_addr);
359 
360 	/* Do we need to go through all of our timers and
361 	 * delete them?   To be safe we will try to delete all, but we
362 	 * should be able to go through and make a guess based
363 	 * on our state.
364 	 */
365 	for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i) {
366 		if (del_timer(&asoc->timers[i]))
367 			sctp_association_put(asoc);
368 	}
369 
370 	/* Free peer's cached cookie. */
371 	kfree(asoc->peer.cookie);
372 	kfree(asoc->peer.peer_random);
373 	kfree(asoc->peer.peer_chunks);
374 	kfree(asoc->peer.peer_hmacs);
375 
376 	/* Release the transport structures. */
377 	list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
378 		transport = list_entry(pos, struct sctp_transport, transports);
379 		list_del_rcu(pos);
380 		sctp_unhash_transport(transport);
381 		sctp_transport_free(transport);
382 	}
383 
384 	asoc->peer.transport_count = 0;
385 
386 	sctp_asconf_queue_teardown(asoc);
387 
388 	/* Free pending address space being deleted */
389 	kfree(asoc->asconf_addr_del_pending);
390 
391 	/* AUTH - Free the endpoint shared keys */
392 	sctp_auth_destroy_keys(&asoc->endpoint_shared_keys);
393 
394 	/* AUTH - Free the association shared key */
395 	sctp_auth_key_put(asoc->asoc_shared_key);
396 
397 	sctp_association_put(asoc);
398 }
399 
400 /* Cleanup and free up an association. */
401 static void sctp_association_destroy(struct sctp_association *asoc)
402 {
403 	if (unlikely(!asoc->base.dead)) {
404 		WARN(1, "Attempt to destroy undead association %p!\n", asoc);
405 		return;
406 	}
407 
408 	sctp_endpoint_put(asoc->ep);
409 	sock_put(asoc->base.sk);
410 
411 	if (asoc->assoc_id != 0) {
412 		spin_lock_bh(&sctp_assocs_id_lock);
413 		idr_remove(&sctp_assocs_id, asoc->assoc_id);
414 		spin_unlock_bh(&sctp_assocs_id_lock);
415 	}
416 
417 	WARN_ON(atomic_read(&asoc->rmem_alloc));
418 
419 	kfree_rcu(asoc, rcu);
420 	SCTP_DBG_OBJCNT_DEC(assoc);
421 }
422 
423 /* Change the primary destination address for the peer. */
424 void sctp_assoc_set_primary(struct sctp_association *asoc,
425 			    struct sctp_transport *transport)
426 {
427 	int changeover = 0;
428 
429 	/* it's a changeover only if we already have a primary path
430 	 * that we are changing
431 	 */
432 	if (asoc->peer.primary_path != NULL &&
433 	    asoc->peer.primary_path != transport)
434 		changeover = 1 ;
435 
436 	asoc->peer.primary_path = transport;
437 	sctp_ulpevent_notify_peer_addr_change(transport,
438 					      SCTP_ADDR_MADE_PRIM, 0);
439 
440 	/* Set a default msg_name for events. */
441 	memcpy(&asoc->peer.primary_addr, &transport->ipaddr,
442 	       sizeof(union sctp_addr));
443 
444 	/* If the primary path is changing, assume that the
445 	 * user wants to use this new path.
446 	 */
447 	if ((transport->state == SCTP_ACTIVE) ||
448 	    (transport->state == SCTP_UNKNOWN))
449 		asoc->peer.active_path = transport;
450 
451 	/*
452 	 * SFR-CACC algorithm:
453 	 * Upon the receipt of a request to change the primary
454 	 * destination address, on the data structure for the new
455 	 * primary destination, the sender MUST do the following:
456 	 *
457 	 * 1) If CHANGEOVER_ACTIVE is set, then there was a switch
458 	 * to this destination address earlier. The sender MUST set
459 	 * CYCLING_CHANGEOVER to indicate that this switch is a
460 	 * double switch to the same destination address.
461 	 *
462 	 * Really, only bother is we have data queued or outstanding on
463 	 * the association.
464 	 */
465 	if (!asoc->outqueue.outstanding_bytes && !asoc->outqueue.out_qlen)
466 		return;
467 
468 	if (transport->cacc.changeover_active)
469 		transport->cacc.cycling_changeover = changeover;
470 
471 	/* 2) The sender MUST set CHANGEOVER_ACTIVE to indicate that
472 	 * a changeover has occurred.
473 	 */
474 	transport->cacc.changeover_active = changeover;
475 
476 	/* 3) The sender MUST store the next TSN to be sent in
477 	 * next_tsn_at_change.
478 	 */
479 	transport->cacc.next_tsn_at_change = asoc->next_tsn;
480 }
481 
482 /* Remove a transport from an association.  */
483 void sctp_assoc_rm_peer(struct sctp_association *asoc,
484 			struct sctp_transport *peer)
485 {
486 	struct sctp_transport *transport;
487 	struct list_head *pos;
488 	struct sctp_chunk *ch;
489 
490 	pr_debug("%s: association:%p addr:%pISpc\n",
491 		 __func__, asoc, &peer->ipaddr.sa);
492 
493 	/* If we are to remove the current retran_path, update it
494 	 * to the next peer before removing this peer from the list.
495 	 */
496 	if (asoc->peer.retran_path == peer)
497 		sctp_assoc_update_retran_path(asoc);
498 
499 	/* Remove this peer from the list. */
500 	list_del_rcu(&peer->transports);
501 	/* Remove this peer from the transport hashtable */
502 	sctp_unhash_transport(peer);
503 
504 	/* Get the first transport of asoc. */
505 	pos = asoc->peer.transport_addr_list.next;
506 	transport = list_entry(pos, struct sctp_transport, transports);
507 
508 	/* Update any entries that match the peer to be deleted. */
509 	if (asoc->peer.primary_path == peer)
510 		sctp_assoc_set_primary(asoc, transport);
511 	if (asoc->peer.active_path == peer)
512 		asoc->peer.active_path = transport;
513 	if (asoc->peer.retran_path == peer)
514 		asoc->peer.retran_path = transport;
515 	if (asoc->peer.last_data_from == peer)
516 		asoc->peer.last_data_from = transport;
517 
518 	if (asoc->strreset_chunk &&
519 	    asoc->strreset_chunk->transport == peer) {
520 		asoc->strreset_chunk->transport = transport;
521 		sctp_transport_reset_reconf_timer(transport);
522 	}
523 
524 	/* If we remove the transport an INIT was last sent to, set it to
525 	 * NULL. Combined with the update of the retran path above, this
526 	 * will cause the next INIT to be sent to the next available
527 	 * transport, maintaining the cycle.
528 	 */
529 	if (asoc->init_last_sent_to == peer)
530 		asoc->init_last_sent_to = NULL;
531 
532 	/* If we remove the transport an SHUTDOWN was last sent to, set it
533 	 * to NULL. Combined with the update of the retran path above, this
534 	 * will cause the next SHUTDOWN to be sent to the next available
535 	 * transport, maintaining the cycle.
536 	 */
537 	if (asoc->shutdown_last_sent_to == peer)
538 		asoc->shutdown_last_sent_to = NULL;
539 
540 	/* If we remove the transport an ASCONF was last sent to, set it to
541 	 * NULL.
542 	 */
543 	if (asoc->addip_last_asconf &&
544 	    asoc->addip_last_asconf->transport == peer)
545 		asoc->addip_last_asconf->transport = NULL;
546 
547 	/* If we have something on the transmitted list, we have to
548 	 * save it off.  The best place is the active path.
549 	 */
550 	if (!list_empty(&peer->transmitted)) {
551 		struct sctp_transport *active = asoc->peer.active_path;
552 
553 		/* Reset the transport of each chunk on this list */
554 		list_for_each_entry(ch, &peer->transmitted,
555 					transmitted_list) {
556 			ch->transport = NULL;
557 			ch->rtt_in_progress = 0;
558 		}
559 
560 		list_splice_tail_init(&peer->transmitted,
561 					&active->transmitted);
562 
563 		/* Start a T3 timer here in case it wasn't running so
564 		 * that these migrated packets have a chance to get
565 		 * retransmitted.
566 		 */
567 		if (!timer_pending(&active->T3_rtx_timer))
568 			if (!mod_timer(&active->T3_rtx_timer,
569 					jiffies + active->rto))
570 				sctp_transport_hold(active);
571 	}
572 
573 	list_for_each_entry(ch, &asoc->outqueue.out_chunk_list, list)
574 		if (ch->transport == peer)
575 			ch->transport = NULL;
576 
577 	asoc->peer.transport_count--;
578 
579 	sctp_ulpevent_notify_peer_addr_change(peer, SCTP_ADDR_REMOVED, 0);
580 	sctp_transport_free(peer);
581 }
582 
583 /* Add a transport address to an association.  */
584 struct sctp_transport *sctp_assoc_add_peer(struct sctp_association *asoc,
585 					   const union sctp_addr *addr,
586 					   const gfp_t gfp,
587 					   const int peer_state)
588 {
589 	struct sctp_transport *peer;
590 	struct sctp_sock *sp;
591 	unsigned short port;
592 
593 	sp = sctp_sk(asoc->base.sk);
594 
595 	/* AF_INET and AF_INET6 share common port field. */
596 	port = ntohs(addr->v4.sin_port);
597 
598 	pr_debug("%s: association:%p addr:%pISpc state:%d\n", __func__,
599 		 asoc, &addr->sa, peer_state);
600 
601 	/* Set the port if it has not been set yet.  */
602 	if (0 == asoc->peer.port)
603 		asoc->peer.port = port;
604 
605 	/* Check to see if this is a duplicate. */
606 	peer = sctp_assoc_lookup_paddr(asoc, addr);
607 	if (peer) {
608 		/* An UNKNOWN state is only set on transports added by
609 		 * user in sctp_connectx() call.  Such transports should be
610 		 * considered CONFIRMED per RFC 4960, Section 5.4.
611 		 */
612 		if (peer->state == SCTP_UNKNOWN) {
613 			peer->state = SCTP_ACTIVE;
614 		}
615 		return peer;
616 	}
617 
618 	peer = sctp_transport_new(asoc->base.net, addr, gfp);
619 	if (!peer)
620 		return NULL;
621 
622 	sctp_transport_set_owner(peer, asoc);
623 
624 	/* Initialize the peer's heartbeat interval based on the
625 	 * association configured value.
626 	 */
627 	peer->hbinterval = asoc->hbinterval;
628 	peer->probe_interval = asoc->probe_interval;
629 
630 	peer->encap_port = asoc->encap_port;
631 
632 	/* Set the path max_retrans.  */
633 	peer->pathmaxrxt = asoc->pathmaxrxt;
634 
635 	/* And the partial failure retrans threshold */
636 	peer->pf_retrans = asoc->pf_retrans;
637 	/* And the primary path switchover retrans threshold */
638 	peer->ps_retrans = asoc->ps_retrans;
639 
640 	/* Initialize the peer's SACK delay timeout based on the
641 	 * association configured value.
642 	 */
643 	peer->sackdelay = asoc->sackdelay;
644 	peer->sackfreq = asoc->sackfreq;
645 
646 	if (addr->sa.sa_family == AF_INET6) {
647 		__be32 info = addr->v6.sin6_flowinfo;
648 
649 		if (info) {
650 			peer->flowlabel = ntohl(info & IPV6_FLOWLABEL_MASK);
651 			peer->flowlabel |= SCTP_FLOWLABEL_SET_MASK;
652 		} else {
653 			peer->flowlabel = asoc->flowlabel;
654 		}
655 	}
656 	peer->dscp = asoc->dscp;
657 
658 	/* Enable/disable heartbeat, SACK delay, and path MTU discovery
659 	 * based on association setting.
660 	 */
661 	peer->param_flags = asoc->param_flags;
662 
663 	/* Initialize the pmtu of the transport. */
664 	sctp_transport_route(peer, NULL, sp);
665 
666 	/* If this is the first transport addr on this association,
667 	 * initialize the association PMTU to the peer's PMTU.
668 	 * If not and the current association PMTU is higher than the new
669 	 * peer's PMTU, reset the association PMTU to the new peer's PMTU.
670 	 */
671 	sctp_assoc_set_pmtu(asoc, asoc->pathmtu ?
672 				  min_t(int, peer->pathmtu, asoc->pathmtu) :
673 				  peer->pathmtu);
674 
675 	peer->pmtu_pending = 0;
676 
677 	/* The asoc->peer.port might not be meaningful yet, but
678 	 * initialize the packet structure anyway.
679 	 */
680 	sctp_packet_init(&peer->packet, peer, asoc->base.bind_addr.port,
681 			 asoc->peer.port);
682 
683 	/* 7.2.1 Slow-Start
684 	 *
685 	 * o The initial cwnd before DATA transmission or after a sufficiently
686 	 *   long idle period MUST be set to
687 	 *      min(4*MTU, max(2*MTU, 4380 bytes))
688 	 *
689 	 * o The initial value of ssthresh MAY be arbitrarily high
690 	 *   (for example, implementations MAY use the size of the
691 	 *   receiver advertised window).
692 	 */
693 	peer->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
694 
695 	/* At this point, we may not have the receiver's advertised window,
696 	 * so initialize ssthresh to the default value and it will be set
697 	 * later when we process the INIT.
698 	 */
699 	peer->ssthresh = SCTP_DEFAULT_MAXWINDOW;
700 
701 	peer->partial_bytes_acked = 0;
702 	peer->flight_size = 0;
703 	peer->burst_limited = 0;
704 
705 	/* Set the transport's RTO.initial value */
706 	peer->rto = asoc->rto_initial;
707 	sctp_max_rto(asoc, peer);
708 
709 	/* Set the peer's active state. */
710 	peer->state = peer_state;
711 
712 	/* Add this peer into the transport hashtable */
713 	if (sctp_hash_transport(peer)) {
714 		sctp_transport_free(peer);
715 		return NULL;
716 	}
717 
718 	sctp_transport_pl_reset(peer);
719 
720 	/* Attach the remote transport to our asoc.  */
721 	list_add_tail_rcu(&peer->transports, &asoc->peer.transport_addr_list);
722 	asoc->peer.transport_count++;
723 
724 	sctp_ulpevent_notify_peer_addr_change(peer, SCTP_ADDR_ADDED, 0);
725 
726 	/* If we do not yet have a primary path, set one.  */
727 	if (!asoc->peer.primary_path) {
728 		sctp_assoc_set_primary(asoc, peer);
729 		asoc->peer.retran_path = peer;
730 	}
731 
732 	if (asoc->peer.active_path == asoc->peer.retran_path &&
733 	    peer->state != SCTP_UNCONFIRMED) {
734 		asoc->peer.retran_path = peer;
735 	}
736 
737 	return peer;
738 }
739 
740 /* Delete a transport address from an association.  */
741 void sctp_assoc_del_peer(struct sctp_association *asoc,
742 			 const union sctp_addr *addr)
743 {
744 	struct list_head	*pos;
745 	struct list_head	*temp;
746 	struct sctp_transport	*transport;
747 
748 	list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
749 		transport = list_entry(pos, struct sctp_transport, transports);
750 		if (sctp_cmp_addr_exact(addr, &transport->ipaddr)) {
751 			/* Do book keeping for removing the peer and free it. */
752 			sctp_assoc_rm_peer(asoc, transport);
753 			break;
754 		}
755 	}
756 }
757 
758 /* Lookup a transport by address. */
759 struct sctp_transport *sctp_assoc_lookup_paddr(
760 					const struct sctp_association *asoc,
761 					const union sctp_addr *address)
762 {
763 	struct sctp_transport *t;
764 
765 	/* Cycle through all transports searching for a peer address. */
766 
767 	list_for_each_entry(t, &asoc->peer.transport_addr_list,
768 			transports) {
769 		if (sctp_cmp_addr_exact(address, &t->ipaddr))
770 			return t;
771 	}
772 
773 	return NULL;
774 }
775 
776 /* Remove all transports except a give one */
777 void sctp_assoc_del_nonprimary_peers(struct sctp_association *asoc,
778 				     struct sctp_transport *primary)
779 {
780 	struct sctp_transport	*temp;
781 	struct sctp_transport	*t;
782 
783 	list_for_each_entry_safe(t, temp, &asoc->peer.transport_addr_list,
784 				 transports) {
785 		/* if the current transport is not the primary one, delete it */
786 		if (t != primary)
787 			sctp_assoc_rm_peer(asoc, t);
788 	}
789 }
790 
791 /* Engage in transport control operations.
792  * Mark the transport up or down and send a notification to the user.
793  * Select and update the new active and retran paths.
794  */
795 void sctp_assoc_control_transport(struct sctp_association *asoc,
796 				  struct sctp_transport *transport,
797 				  enum sctp_transport_cmd command,
798 				  sctp_sn_error_t error)
799 {
800 	int spc_state = SCTP_ADDR_AVAILABLE;
801 	bool ulp_notify = true;
802 
803 	/* Record the transition on the transport.  */
804 	switch (command) {
805 	case SCTP_TRANSPORT_UP:
806 		/* If we are moving from UNCONFIRMED state due
807 		 * to heartbeat success, report the SCTP_ADDR_CONFIRMED
808 		 * state to the user, otherwise report SCTP_ADDR_AVAILABLE.
809 		 */
810 		if (transport->state == SCTP_PF &&
811 		    asoc->pf_expose != SCTP_PF_EXPOSE_ENABLE)
812 			ulp_notify = false;
813 		else if (transport->state == SCTP_UNCONFIRMED &&
814 			 error == SCTP_HEARTBEAT_SUCCESS)
815 			spc_state = SCTP_ADDR_CONFIRMED;
816 
817 		transport->state = SCTP_ACTIVE;
818 		sctp_transport_pl_reset(transport);
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 			sctp_transport_pl_reset(transport);
829 			spc_state = SCTP_ADDR_UNREACHABLE;
830 		} else {
831 			sctp_transport_dst_release(transport);
832 			ulp_notify = false;
833 		}
834 		break;
835 
836 	case SCTP_TRANSPORT_PF:
837 		transport->state = SCTP_PF;
838 		if (asoc->pf_expose != SCTP_PF_EXPOSE_ENABLE)
839 			ulp_notify = false;
840 		else
841 			spc_state = SCTP_ADDR_POTENTIALLY_FAILED;
842 		break;
843 
844 	default:
845 		return;
846 	}
847 
848 	/* Generate and send a SCTP_PEER_ADDR_CHANGE notification
849 	 * to the user.
850 	 */
851 	if (ulp_notify)
852 		sctp_ulpevent_notify_peer_addr_change(transport,
853 						      spc_state, error);
854 
855 	/* Select new active and retran paths. */
856 	sctp_select_active_and_retran_path(asoc);
857 }
858 
859 /* Hold a reference to an association. */
860 void sctp_association_hold(struct sctp_association *asoc)
861 {
862 	refcount_inc(&asoc->base.refcnt);
863 }
864 
865 /* Release a reference to an association and cleanup
866  * if there are no more references.
867  */
868 void sctp_association_put(struct sctp_association *asoc)
869 {
870 	if (refcount_dec_and_test(&asoc->base.refcnt))
871 		sctp_association_destroy(asoc);
872 }
873 
874 /* Allocate the next TSN, Transmission Sequence Number, for the given
875  * association.
876  */
877 __u32 sctp_association_get_next_tsn(struct sctp_association *asoc)
878 {
879 	/* From Section 1.6 Serial Number Arithmetic:
880 	 * Transmission Sequence Numbers wrap around when they reach
881 	 * 2**32 - 1.  That is, the next TSN a DATA chunk MUST use
882 	 * after transmitting TSN = 2*32 - 1 is TSN = 0.
883 	 */
884 	__u32 retval = asoc->next_tsn;
885 	asoc->next_tsn++;
886 	asoc->unack_data++;
887 
888 	return retval;
889 }
890 
891 /* Compare two addresses to see if they match.  Wildcard addresses
892  * only match themselves.
893  */
894 int sctp_cmp_addr_exact(const union sctp_addr *ss1,
895 			const union sctp_addr *ss2)
896 {
897 	struct sctp_af *af;
898 
899 	af = sctp_get_af_specific(ss1->sa.sa_family);
900 	if (unlikely(!af))
901 		return 0;
902 
903 	return af->cmp_addr(ss1, ss2);
904 }
905 
906 /* Return an ecne chunk to get prepended to a packet.
907  * Note:  We are sly and return a shared, prealloced chunk.  FIXME:
908  * No we don't, but we could/should.
909  */
910 struct sctp_chunk *sctp_get_ecne_prepend(struct sctp_association *asoc)
911 {
912 	if (!asoc->need_ecne)
913 		return NULL;
914 
915 	/* Send ECNE if needed.
916 	 * Not being able to allocate a chunk here is not deadly.
917 	 */
918 	return sctp_make_ecne(asoc, asoc->last_ecne_tsn);
919 }
920 
921 /*
922  * Find which transport this TSN was sent on.
923  */
924 struct sctp_transport *sctp_assoc_lookup_tsn(struct sctp_association *asoc,
925 					     __u32 tsn)
926 {
927 	struct sctp_transport *active;
928 	struct sctp_transport *match;
929 	struct sctp_transport *transport;
930 	struct sctp_chunk *chunk;
931 	__be32 key = htonl(tsn);
932 
933 	match = NULL;
934 
935 	/*
936 	 * FIXME: In general, find a more efficient data structure for
937 	 * searching.
938 	 */
939 
940 	/*
941 	 * The general strategy is to search each transport's transmitted
942 	 * list.   Return which transport this TSN lives on.
943 	 *
944 	 * Let's be hopeful and check the active_path first.
945 	 * Another optimization would be to know if there is only one
946 	 * outbound path and not have to look for the TSN at all.
947 	 *
948 	 */
949 
950 	active = asoc->peer.active_path;
951 
952 	list_for_each_entry(chunk, &active->transmitted,
953 			transmitted_list) {
954 
955 		if (key == chunk->subh.data_hdr->tsn) {
956 			match = active;
957 			goto out;
958 		}
959 	}
960 
961 	/* If not found, go search all the other transports. */
962 	list_for_each_entry(transport, &asoc->peer.transport_addr_list,
963 			transports) {
964 
965 		if (transport == active)
966 			continue;
967 		list_for_each_entry(chunk, &transport->transmitted,
968 				transmitted_list) {
969 			if (key == chunk->subh.data_hdr->tsn) {
970 				match = transport;
971 				goto out;
972 			}
973 		}
974 	}
975 out:
976 	return match;
977 }
978 
979 /* Do delayed input processing.  This is scheduled by sctp_rcv(). */
980 static void sctp_assoc_bh_rcv(struct work_struct *work)
981 {
982 	struct sctp_association *asoc =
983 		container_of(work, struct sctp_association,
984 			     base.inqueue.immediate);
985 	struct net *net = asoc->base.net;
986 	union sctp_subtype subtype;
987 	struct sctp_endpoint *ep;
988 	struct sctp_chunk *chunk;
989 	struct sctp_inq *inqueue;
990 	int first_time = 1;	/* is this the first time through the loop */
991 	int error = 0;
992 	int state;
993 
994 	/* The association should be held so we should be safe. */
995 	ep = asoc->ep;
996 
997 	inqueue = &asoc->base.inqueue;
998 	sctp_association_hold(asoc);
999 	while (NULL != (chunk = sctp_inq_pop(inqueue))) {
1000 		state = asoc->state;
1001 		subtype = SCTP_ST_CHUNK(chunk->chunk_hdr->type);
1002 
1003 		/* If the first chunk in the packet is AUTH, do special
1004 		 * processing specified in Section 6.3 of SCTP-AUTH spec
1005 		 */
1006 		if (first_time && subtype.chunk == SCTP_CID_AUTH) {
1007 			struct sctp_chunkhdr *next_hdr;
1008 
1009 			next_hdr = sctp_inq_peek(inqueue);
1010 			if (!next_hdr)
1011 				goto normal;
1012 
1013 			/* If the next chunk is COOKIE-ECHO, skip the AUTH
1014 			 * chunk while saving a pointer to it so we can do
1015 			 * Authentication later (during cookie-echo
1016 			 * processing).
1017 			 */
1018 			if (next_hdr->type == SCTP_CID_COOKIE_ECHO) {
1019 				chunk->auth_chunk = skb_clone(chunk->skb,
1020 							      GFP_ATOMIC);
1021 				chunk->auth = 1;
1022 				continue;
1023 			}
1024 		}
1025 
1026 normal:
1027 		/* SCTP-AUTH, Section 6.3:
1028 		 *    The receiver has a list of chunk types which it expects
1029 		 *    to be received only after an AUTH-chunk.  This list has
1030 		 *    been sent to the peer during the association setup.  It
1031 		 *    MUST silently discard these chunks if they are not placed
1032 		 *    after an AUTH chunk in the packet.
1033 		 */
1034 		if (sctp_auth_recv_cid(subtype.chunk, asoc) && !chunk->auth)
1035 			continue;
1036 
1037 		/* Remember where the last DATA chunk came from so we
1038 		 * know where to send the SACK.
1039 		 */
1040 		if (sctp_chunk_is_data(chunk))
1041 			asoc->peer.last_data_from = chunk->transport;
1042 		else {
1043 			SCTP_INC_STATS(net, SCTP_MIB_INCTRLCHUNKS);
1044 			asoc->stats.ictrlchunks++;
1045 			if (chunk->chunk_hdr->type == SCTP_CID_SACK)
1046 				asoc->stats.isacks++;
1047 		}
1048 
1049 		if (chunk->transport)
1050 			chunk->transport->last_time_heard = ktime_get();
1051 
1052 		/* Run through the state machine. */
1053 		error = sctp_do_sm(net, SCTP_EVENT_T_CHUNK, subtype,
1054 				   state, ep, asoc, chunk, GFP_ATOMIC);
1055 
1056 		/* Check to see if the association is freed in response to
1057 		 * the incoming chunk.  If so, get out of the while loop.
1058 		 */
1059 		if (asoc->base.dead)
1060 			break;
1061 
1062 		/* If there is an error on chunk, discard this packet. */
1063 		if (error && chunk)
1064 			chunk->pdiscard = 1;
1065 
1066 		if (first_time)
1067 			first_time = 0;
1068 	}
1069 	sctp_association_put(asoc);
1070 }
1071 
1072 /* This routine moves an association from its old sk to a new sk.  */
1073 void sctp_assoc_migrate(struct sctp_association *assoc, struct sock *newsk)
1074 {
1075 	struct sctp_sock *newsp = sctp_sk(newsk);
1076 	struct sock *oldsk = assoc->base.sk;
1077 
1078 	/* Delete the association from the old endpoint's list of
1079 	 * associations.
1080 	 */
1081 	list_del_init(&assoc->asocs);
1082 
1083 	/* Decrement the backlog value for a TCP-style socket. */
1084 	if (sctp_style(oldsk, TCP))
1085 		sk_acceptq_removed(oldsk);
1086 
1087 	/* Release references to the old endpoint and the sock.  */
1088 	sctp_endpoint_put(assoc->ep);
1089 	sock_put(assoc->base.sk);
1090 
1091 	/* Get a reference to the new endpoint.  */
1092 	assoc->ep = newsp->ep;
1093 	sctp_endpoint_hold(assoc->ep);
1094 
1095 	/* Get a reference to the new sock.  */
1096 	assoc->base.sk = newsk;
1097 	sock_hold(assoc->base.sk);
1098 
1099 	/* Add the association to the new endpoint's list of associations.  */
1100 	sctp_endpoint_add_asoc(newsp->ep, assoc);
1101 }
1102 
1103 /* Update an association (possibly from unexpected COOKIE-ECHO processing).  */
1104 int sctp_assoc_update(struct sctp_association *asoc,
1105 		      struct sctp_association *new)
1106 {
1107 	struct sctp_transport *trans;
1108 	struct list_head *pos, *temp;
1109 
1110 	/* Copy in new parameters of peer. */
1111 	asoc->c = new->c;
1112 	asoc->peer.rwnd = new->peer.rwnd;
1113 	asoc->peer.sack_needed = new->peer.sack_needed;
1114 	asoc->peer.auth_capable = new->peer.auth_capable;
1115 	asoc->peer.i = new->peer.i;
1116 
1117 	if (!sctp_tsnmap_init(&asoc->peer.tsn_map, SCTP_TSN_MAP_INITIAL,
1118 			      asoc->peer.i.initial_tsn, GFP_ATOMIC))
1119 		return -ENOMEM;
1120 
1121 	/* Remove any peer addresses not present in the new association. */
1122 	list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
1123 		trans = list_entry(pos, struct sctp_transport, transports);
1124 		if (!sctp_assoc_lookup_paddr(new, &trans->ipaddr)) {
1125 			sctp_assoc_rm_peer(asoc, trans);
1126 			continue;
1127 		}
1128 
1129 		if (asoc->state >= SCTP_STATE_ESTABLISHED)
1130 			sctp_transport_reset(trans);
1131 	}
1132 
1133 	/* If the case is A (association restart), use
1134 	 * initial_tsn as next_tsn. If the case is B, use
1135 	 * current next_tsn in case data sent to peer
1136 	 * has been discarded and needs retransmission.
1137 	 */
1138 	if (asoc->state >= SCTP_STATE_ESTABLISHED) {
1139 		asoc->next_tsn = new->next_tsn;
1140 		asoc->ctsn_ack_point = new->ctsn_ack_point;
1141 		asoc->adv_peer_ack_point = new->adv_peer_ack_point;
1142 
1143 		/* Reinitialize SSN for both local streams
1144 		 * and peer's streams.
1145 		 */
1146 		sctp_stream_clear(&asoc->stream);
1147 
1148 		/* Flush the ULP reassembly and ordered queue.
1149 		 * Any data there will now be stale and will
1150 		 * cause problems.
1151 		 */
1152 		sctp_ulpq_flush(&asoc->ulpq);
1153 
1154 		/* reset the overall association error count so
1155 		 * that the restarted association doesn't get torn
1156 		 * down on the next retransmission timer.
1157 		 */
1158 		asoc->overall_error_count = 0;
1159 
1160 	} else {
1161 		/* Add any peer addresses from the new association. */
1162 		list_for_each_entry(trans, &new->peer.transport_addr_list,
1163 				    transports)
1164 			if (!sctp_assoc_lookup_paddr(asoc, &trans->ipaddr) &&
1165 			    !sctp_assoc_add_peer(asoc, &trans->ipaddr,
1166 						 GFP_ATOMIC, trans->state))
1167 				return -ENOMEM;
1168 
1169 		asoc->ctsn_ack_point = asoc->next_tsn - 1;
1170 		asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
1171 
1172 		if (sctp_state(asoc, COOKIE_WAIT))
1173 			sctp_stream_update(&asoc->stream, &new->stream);
1174 
1175 		/* get a new assoc id if we don't have one yet. */
1176 		if (sctp_assoc_set_id(asoc, GFP_ATOMIC))
1177 			return -ENOMEM;
1178 	}
1179 
1180 	/* SCTP-AUTH: Save the peer parameters from the new associations
1181 	 * and also move the association shared keys over
1182 	 */
1183 	kfree(asoc->peer.peer_random);
1184 	asoc->peer.peer_random = new->peer.peer_random;
1185 	new->peer.peer_random = NULL;
1186 
1187 	kfree(asoc->peer.peer_chunks);
1188 	asoc->peer.peer_chunks = new->peer.peer_chunks;
1189 	new->peer.peer_chunks = NULL;
1190 
1191 	kfree(asoc->peer.peer_hmacs);
1192 	asoc->peer.peer_hmacs = new->peer.peer_hmacs;
1193 	new->peer.peer_hmacs = NULL;
1194 
1195 	return sctp_auth_asoc_init_active_key(asoc, GFP_ATOMIC);
1196 }
1197 
1198 /* Update the retran path for sending a retransmitted packet.
1199  * See also RFC4960, 6.4. Multi-Homed SCTP Endpoints:
1200  *
1201  *   When there is outbound data to send and the primary path
1202  *   becomes inactive (e.g., due to failures), or where the
1203  *   SCTP user explicitly requests to send data to an
1204  *   inactive destination transport address, before reporting
1205  *   an error to its ULP, the SCTP endpoint should try to send
1206  *   the data to an alternate active destination transport
1207  *   address if one exists.
1208  *
1209  *   When retransmitting data that timed out, if the endpoint
1210  *   is multihomed, it should consider each source-destination
1211  *   address pair in its retransmission selection policy.
1212  *   When retransmitting timed-out data, the endpoint should
1213  *   attempt to pick the most divergent source-destination
1214  *   pair from the original source-destination pair to which
1215  *   the packet was transmitted.
1216  *
1217  *   Note: Rules for picking the most divergent source-destination
1218  *   pair are an implementation decision and are not specified
1219  *   within this document.
1220  *
1221  * Our basic strategy is to round-robin transports in priorities
1222  * according to sctp_trans_score() e.g., if no such
1223  * transport with state SCTP_ACTIVE exists, round-robin through
1224  * SCTP_UNKNOWN, etc. You get the picture.
1225  */
1226 static u8 sctp_trans_score(const struct sctp_transport *trans)
1227 {
1228 	switch (trans->state) {
1229 	case SCTP_ACTIVE:
1230 		return 3;	/* best case */
1231 	case SCTP_UNKNOWN:
1232 		return 2;
1233 	case SCTP_PF:
1234 		return 1;
1235 	default: /* case SCTP_INACTIVE */
1236 		return 0;	/* worst case */
1237 	}
1238 }
1239 
1240 static struct sctp_transport *sctp_trans_elect_tie(struct sctp_transport *trans1,
1241 						   struct sctp_transport *trans2)
1242 {
1243 	if (trans1->error_count > trans2->error_count) {
1244 		return trans2;
1245 	} else if (trans1->error_count == trans2->error_count &&
1246 		   ktime_after(trans2->last_time_heard,
1247 			       trans1->last_time_heard)) {
1248 		return trans2;
1249 	} else {
1250 		return trans1;
1251 	}
1252 }
1253 
1254 static struct sctp_transport *sctp_trans_elect_best(struct sctp_transport *curr,
1255 						    struct sctp_transport *best)
1256 {
1257 	u8 score_curr, score_best;
1258 
1259 	if (best == NULL || curr == best)
1260 		return curr;
1261 
1262 	score_curr = sctp_trans_score(curr);
1263 	score_best = sctp_trans_score(best);
1264 
1265 	/* First, try a score-based selection if both transport states
1266 	 * differ. If we're in a tie, lets try to make a more clever
1267 	 * decision here based on error counts and last time heard.
1268 	 */
1269 	if (score_curr > score_best)
1270 		return curr;
1271 	else if (score_curr == score_best)
1272 		return sctp_trans_elect_tie(best, curr);
1273 	else
1274 		return best;
1275 }
1276 
1277 void sctp_assoc_update_retran_path(struct sctp_association *asoc)
1278 {
1279 	struct sctp_transport *trans = asoc->peer.retran_path;
1280 	struct sctp_transport *trans_next = NULL;
1281 
1282 	/* We're done as we only have the one and only path. */
1283 	if (asoc->peer.transport_count == 1)
1284 		return;
1285 	/* If active_path and retran_path are the same and active,
1286 	 * then this is the only active path. Use it.
1287 	 */
1288 	if (asoc->peer.active_path == asoc->peer.retran_path &&
1289 	    asoc->peer.active_path->state == SCTP_ACTIVE)
1290 		return;
1291 
1292 	/* Iterate from retran_path's successor back to retran_path. */
1293 	for (trans = list_next_entry(trans, transports); 1;
1294 	     trans = list_next_entry(trans, transports)) {
1295 		/* Manually skip the head element. */
1296 		if (&trans->transports == &asoc->peer.transport_addr_list)
1297 			continue;
1298 		if (trans->state == SCTP_UNCONFIRMED)
1299 			continue;
1300 		trans_next = sctp_trans_elect_best(trans, trans_next);
1301 		/* Active is good enough for immediate return. */
1302 		if (trans_next->state == SCTP_ACTIVE)
1303 			break;
1304 		/* We've reached the end, time to update path. */
1305 		if (trans == asoc->peer.retran_path)
1306 			break;
1307 	}
1308 
1309 	asoc->peer.retran_path = trans_next;
1310 
1311 	pr_debug("%s: association:%p updated new path to addr:%pISpc\n",
1312 		 __func__, asoc, &asoc->peer.retran_path->ipaddr.sa);
1313 }
1314 
1315 static void sctp_select_active_and_retran_path(struct sctp_association *asoc)
1316 {
1317 	struct sctp_transport *trans, *trans_pri = NULL, *trans_sec = NULL;
1318 	struct sctp_transport *trans_pf = NULL;
1319 
1320 	/* Look for the two most recently used active transports. */
1321 	list_for_each_entry(trans, &asoc->peer.transport_addr_list,
1322 			    transports) {
1323 		/* Skip uninteresting transports. */
1324 		if (trans->state == SCTP_INACTIVE ||
1325 		    trans->state == SCTP_UNCONFIRMED)
1326 			continue;
1327 		/* Keep track of the best PF transport from our
1328 		 * list in case we don't find an active one.
1329 		 */
1330 		if (trans->state == SCTP_PF) {
1331 			trans_pf = sctp_trans_elect_best(trans, trans_pf);
1332 			continue;
1333 		}
1334 		/* For active transports, pick the most recent ones. */
1335 		if (trans_pri == NULL ||
1336 		    ktime_after(trans->last_time_heard,
1337 				trans_pri->last_time_heard)) {
1338 			trans_sec = trans_pri;
1339 			trans_pri = trans;
1340 		} else if (trans_sec == NULL ||
1341 			   ktime_after(trans->last_time_heard,
1342 				       trans_sec->last_time_heard)) {
1343 			trans_sec = trans;
1344 		}
1345 	}
1346 
1347 	/* RFC 2960 6.4 Multi-Homed SCTP Endpoints
1348 	 *
1349 	 * By default, an endpoint should always transmit to the primary
1350 	 * path, unless the SCTP user explicitly specifies the
1351 	 * destination transport address (and possibly source transport
1352 	 * address) to use. [If the primary is active but not most recent,
1353 	 * bump the most recently used transport.]
1354 	 */
1355 	if ((asoc->peer.primary_path->state == SCTP_ACTIVE ||
1356 	     asoc->peer.primary_path->state == SCTP_UNKNOWN) &&
1357 	     asoc->peer.primary_path != trans_pri) {
1358 		trans_sec = trans_pri;
1359 		trans_pri = asoc->peer.primary_path;
1360 	}
1361 
1362 	/* We did not find anything useful for a possible retransmission
1363 	 * path; either primary path that we found is the same as
1364 	 * the current one, or we didn't generally find an active one.
1365 	 */
1366 	if (trans_sec == NULL)
1367 		trans_sec = trans_pri;
1368 
1369 	/* If we failed to find a usable transport, just camp on the
1370 	 * active or pick a PF iff it's the better choice.
1371 	 */
1372 	if (trans_pri == NULL) {
1373 		trans_pri = sctp_trans_elect_best(asoc->peer.active_path, trans_pf);
1374 		trans_sec = trans_pri;
1375 	}
1376 
1377 	/* Set the active and retran transports. */
1378 	asoc->peer.active_path = trans_pri;
1379 	asoc->peer.retran_path = trans_sec;
1380 }
1381 
1382 struct sctp_transport *
1383 sctp_assoc_choose_alter_transport(struct sctp_association *asoc,
1384 				  struct sctp_transport *last_sent_to)
1385 {
1386 	/* If this is the first time packet is sent, use the active path,
1387 	 * else use the retran path. If the last packet was sent over the
1388 	 * retran path, update the retran path and use it.
1389 	 */
1390 	if (last_sent_to == NULL) {
1391 		return asoc->peer.active_path;
1392 	} else {
1393 		if (last_sent_to == asoc->peer.retran_path)
1394 			sctp_assoc_update_retran_path(asoc);
1395 
1396 		return asoc->peer.retran_path;
1397 	}
1398 }
1399 
1400 void sctp_assoc_update_frag_point(struct sctp_association *asoc)
1401 {
1402 	int frag = sctp_mtu_payload(sctp_sk(asoc->base.sk), asoc->pathmtu,
1403 				    sctp_datachk_len(&asoc->stream));
1404 
1405 	if (asoc->user_frag)
1406 		frag = min_t(int, frag, asoc->user_frag);
1407 
1408 	frag = min_t(int, frag, SCTP_MAX_CHUNK_LEN -
1409 				sctp_datachk_len(&asoc->stream));
1410 
1411 	asoc->frag_point = SCTP_TRUNC4(frag);
1412 }
1413 
1414 void sctp_assoc_set_pmtu(struct sctp_association *asoc, __u32 pmtu)
1415 {
1416 	if (asoc->pathmtu != pmtu) {
1417 		asoc->pathmtu = pmtu;
1418 		sctp_assoc_update_frag_point(asoc);
1419 	}
1420 
1421 	pr_debug("%s: asoc:%p, pmtu:%d, frag_point:%d\n", __func__, asoc,
1422 		 asoc->pathmtu, asoc->frag_point);
1423 }
1424 
1425 /* Update the association's pmtu and frag_point by going through all the
1426  * transports. This routine is called when a transport's PMTU has changed.
1427  */
1428 void sctp_assoc_sync_pmtu(struct sctp_association *asoc)
1429 {
1430 	struct sctp_transport *t;
1431 	__u32 pmtu = 0;
1432 
1433 	if (!asoc)
1434 		return;
1435 
1436 	/* Get the lowest pmtu of all the transports. */
1437 	list_for_each_entry(t, &asoc->peer.transport_addr_list, transports) {
1438 		if (t->pmtu_pending && t->dst) {
1439 			sctp_transport_update_pmtu(t,
1440 						   atomic_read(&t->mtu_info));
1441 			t->pmtu_pending = 0;
1442 		}
1443 		if (!pmtu || (t->pathmtu < pmtu))
1444 			pmtu = t->pathmtu;
1445 	}
1446 
1447 	sctp_assoc_set_pmtu(asoc, pmtu);
1448 }
1449 
1450 /* Should we send a SACK to update our peer? */
1451 static inline bool sctp_peer_needs_update(struct sctp_association *asoc)
1452 {
1453 	struct net *net = asoc->base.net;
1454 
1455 	switch (asoc->state) {
1456 	case SCTP_STATE_ESTABLISHED:
1457 	case SCTP_STATE_SHUTDOWN_PENDING:
1458 	case SCTP_STATE_SHUTDOWN_RECEIVED:
1459 	case SCTP_STATE_SHUTDOWN_SENT:
1460 		if ((asoc->rwnd > asoc->a_rwnd) &&
1461 		    ((asoc->rwnd - asoc->a_rwnd) >= max_t(__u32,
1462 			   (asoc->base.sk->sk_rcvbuf >> net->sctp.rwnd_upd_shift),
1463 			   asoc->pathmtu)))
1464 			return true;
1465 		break;
1466 	default:
1467 		break;
1468 	}
1469 	return false;
1470 }
1471 
1472 /* Increase asoc's rwnd by len and send any window update SACK if needed. */
1473 void sctp_assoc_rwnd_increase(struct sctp_association *asoc, unsigned int len)
1474 {
1475 	struct sctp_chunk *sack;
1476 	struct timer_list *timer;
1477 
1478 	if (asoc->rwnd_over) {
1479 		if (asoc->rwnd_over >= len) {
1480 			asoc->rwnd_over -= len;
1481 		} else {
1482 			asoc->rwnd += (len - asoc->rwnd_over);
1483 			asoc->rwnd_over = 0;
1484 		}
1485 	} else {
1486 		asoc->rwnd += len;
1487 	}
1488 
1489 	/* If we had window pressure, start recovering it
1490 	 * once our rwnd had reached the accumulated pressure
1491 	 * threshold.  The idea is to recover slowly, but up
1492 	 * to the initial advertised window.
1493 	 */
1494 	if (asoc->rwnd_press) {
1495 		int change = min(asoc->pathmtu, asoc->rwnd_press);
1496 		asoc->rwnd += change;
1497 		asoc->rwnd_press -= change;
1498 	}
1499 
1500 	pr_debug("%s: asoc:%p rwnd increased by %d to (%u, %u) - %u\n",
1501 		 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
1502 		 asoc->a_rwnd);
1503 
1504 	/* Send a window update SACK if the rwnd has increased by at least the
1505 	 * minimum of the association's PMTU and half of the receive buffer.
1506 	 * The algorithm used is similar to the one described in
1507 	 * Section 4.2.3.3 of RFC 1122.
1508 	 */
1509 	if (sctp_peer_needs_update(asoc)) {
1510 		asoc->a_rwnd = asoc->rwnd;
1511 
1512 		pr_debug("%s: sending window update SACK- asoc:%p rwnd:%u "
1513 			 "a_rwnd:%u\n", __func__, asoc, asoc->rwnd,
1514 			 asoc->a_rwnd);
1515 
1516 		sack = sctp_make_sack(asoc);
1517 		if (!sack)
1518 			return;
1519 
1520 		asoc->peer.sack_needed = 0;
1521 
1522 		sctp_outq_tail(&asoc->outqueue, sack, GFP_ATOMIC);
1523 
1524 		/* Stop the SACK timer.  */
1525 		timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
1526 		if (del_timer(timer))
1527 			sctp_association_put(asoc);
1528 	}
1529 }
1530 
1531 /* Decrease asoc's rwnd by len. */
1532 void sctp_assoc_rwnd_decrease(struct sctp_association *asoc, unsigned int len)
1533 {
1534 	int rx_count;
1535 	int over = 0;
1536 
1537 	if (unlikely(!asoc->rwnd || asoc->rwnd_over))
1538 		pr_debug("%s: association:%p has asoc->rwnd:%u, "
1539 			 "asoc->rwnd_over:%u!\n", __func__, asoc,
1540 			 asoc->rwnd, asoc->rwnd_over);
1541 
1542 	if (asoc->ep->rcvbuf_policy)
1543 		rx_count = atomic_read(&asoc->rmem_alloc);
1544 	else
1545 		rx_count = atomic_read(&asoc->base.sk->sk_rmem_alloc);
1546 
1547 	/* If we've reached or overflowed our receive buffer, announce
1548 	 * a 0 rwnd if rwnd would still be positive.  Store the
1549 	 * potential pressure overflow so that the window can be restored
1550 	 * back to original value.
1551 	 */
1552 	if (rx_count >= asoc->base.sk->sk_rcvbuf)
1553 		over = 1;
1554 
1555 	if (asoc->rwnd >= len) {
1556 		asoc->rwnd -= len;
1557 		if (over) {
1558 			asoc->rwnd_press += asoc->rwnd;
1559 			asoc->rwnd = 0;
1560 		}
1561 	} else {
1562 		asoc->rwnd_over += len - asoc->rwnd;
1563 		asoc->rwnd = 0;
1564 	}
1565 
1566 	pr_debug("%s: asoc:%p rwnd decreased by %d to (%u, %u, %u)\n",
1567 		 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
1568 		 asoc->rwnd_press);
1569 }
1570 
1571 /* Build the bind address list for the association based on info from the
1572  * local endpoint and the remote peer.
1573  */
1574 int sctp_assoc_set_bind_addr_from_ep(struct sctp_association *asoc,
1575 				     enum sctp_scope scope, gfp_t gfp)
1576 {
1577 	struct sock *sk = asoc->base.sk;
1578 	int flags;
1579 
1580 	/* Use scoping rules to determine the subset of addresses from
1581 	 * the endpoint.
1582 	 */
1583 	flags = (PF_INET6 == sk->sk_family) ? SCTP_ADDR6_ALLOWED : 0;
1584 	if (!inet_v6_ipv6only(sk))
1585 		flags |= SCTP_ADDR4_ALLOWED;
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(asoc->base.net,
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