xref: /openbmc/linux/net/sctp/associola.c (revision e3b9f1e8)
1 /* SCTP kernel implementation
2  * (C) Copyright IBM Corp. 2001, 2004
3  * Copyright (c) 1999-2000 Cisco, Inc.
4  * Copyright (c) 1999-2001 Motorola, Inc.
5  * Copyright (c) 2001 Intel Corp.
6  * Copyright (c) 2001 La Monte H.P. Yarroll
7  *
8  * This file is part of the SCTP kernel implementation
9  *
10  * This module provides the abstraction for an SCTP association.
11  *
12  * This SCTP implementation is free software;
13  * you can redistribute it and/or modify it under the terms of
14  * the GNU General Public License as published by
15  * the Free Software Foundation; either version 2, or (at your option)
16  * any later version.
17  *
18  * This SCTP implementation is distributed in the hope that it
19  * will be useful, but WITHOUT ANY WARRANTY; without even the implied
20  *                 ************************
21  * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
22  * See the GNU General Public License for more details.
23  *
24  * You should have received a copy of the GNU General Public License
25  * along with GNU CC; see the file COPYING.  If not, see
26  * <http://www.gnu.org/licenses/>.
27  *
28  * Please send any bug reports or fixes you make to the
29  * email address(es):
30  *    lksctp developers <linux-sctp@vger.kernel.org>
31  *
32  * Written or modified by:
33  *    La Monte H.P. Yarroll <piggy@acm.org>
34  *    Karl Knutson          <karl@athena.chicago.il.us>
35  *    Jon Grimm             <jgrimm@us.ibm.com>
36  *    Xingang Guo           <xingang.guo@intel.com>
37  *    Hui Huang             <hui.huang@nokia.com>
38  *    Sridhar Samudrala	    <sri@us.ibm.com>
39  *    Daisy Chang	    <daisyc@us.ibm.com>
40  *    Ryan Layer	    <rmlayer@us.ibm.com>
41  *    Kevin Gao             <kevin.gao@intel.com>
42  */
43 
44 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
45 
46 #include <linux/types.h>
47 #include <linux/fcntl.h>
48 #include <linux/poll.h>
49 #include <linux/init.h>
50 
51 #include <linux/slab.h>
52 #include <linux/in.h>
53 #include <net/ipv6.h>
54 #include <net/sctp/sctp.h>
55 #include <net/sctp/sm.h>
56 
57 /* Forward declarations for internal functions. */
58 static void sctp_select_active_and_retran_path(struct sctp_association *asoc);
59 static void sctp_assoc_bh_rcv(struct work_struct *work);
60 static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc);
61 static void sctp_assoc_free_asconf_queue(struct sctp_association *asoc);
62 
63 /* 1st Level Abstractions. */
64 
65 /* Initialize a new association from provided memory. */
66 static struct sctp_association *sctp_association_init(
67 					struct sctp_association *asoc,
68 					const struct sctp_endpoint *ep,
69 					const struct sock *sk,
70 					enum sctp_scope scope, gfp_t gfp)
71 {
72 	struct net *net = sock_net(sk);
73 	struct sctp_sock *sp;
74 	struct sctp_paramhdr *p;
75 	int i;
76 
77 	/* Retrieve the SCTP per socket area.  */
78 	sp = sctp_sk((struct sock *)sk);
79 
80 	/* Discarding const is appropriate here.  */
81 	asoc->ep = (struct sctp_endpoint *)ep;
82 	asoc->base.sk = (struct sock *)sk;
83 
84 	sctp_endpoint_hold(asoc->ep);
85 	sock_hold(asoc->base.sk);
86 
87 	/* Initialize the common base substructure.  */
88 	asoc->base.type = SCTP_EP_TYPE_ASSOCIATION;
89 
90 	/* Initialize the object handling fields.  */
91 	refcount_set(&asoc->base.refcnt, 1);
92 
93 	/* Initialize the bind addr area.  */
94 	sctp_bind_addr_init(&asoc->base.bind_addr, ep->base.bind_addr.port);
95 
96 	asoc->state = SCTP_STATE_CLOSED;
97 	asoc->cookie_life = ms_to_ktime(sp->assocparams.sasoc_cookie_life);
98 	asoc->user_frag = sp->user_frag;
99 
100 	/* Set the association max_retrans and RTO values from the
101 	 * socket values.
102 	 */
103 	asoc->max_retrans = sp->assocparams.sasoc_asocmaxrxt;
104 	asoc->pf_retrans  = net->sctp.pf_retrans;
105 
106 	asoc->rto_initial = msecs_to_jiffies(sp->rtoinfo.srto_initial);
107 	asoc->rto_max = msecs_to_jiffies(sp->rtoinfo.srto_max);
108 	asoc->rto_min = msecs_to_jiffies(sp->rtoinfo.srto_min);
109 
110 	/* Initialize the association's heartbeat interval based on the
111 	 * sock configured value.
112 	 */
113 	asoc->hbinterval = msecs_to_jiffies(sp->hbinterval);
114 
115 	/* Initialize path max retrans value. */
116 	asoc->pathmaxrxt = sp->pathmaxrxt;
117 
118 	/* Initialize default path MTU. */
119 	asoc->pathmtu = sp->pathmtu;
120 
121 	/* Set association default SACK delay */
122 	asoc->sackdelay = msecs_to_jiffies(sp->sackdelay);
123 	asoc->sackfreq = sp->sackfreq;
124 
125 	/* Set the association default flags controlling
126 	 * Heartbeat, SACK delay, and Path MTU Discovery.
127 	 */
128 	asoc->param_flags = sp->param_flags;
129 
130 	/* Initialize the maximum number of new data packets that can be sent
131 	 * in a burst.
132 	 */
133 	asoc->max_burst = sp->max_burst;
134 
135 	/* initialize association timers */
136 	asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_COOKIE] = asoc->rto_initial;
137 	asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_INIT] = asoc->rto_initial;
138 	asoc->timeouts[SCTP_EVENT_TIMEOUT_T2_SHUTDOWN] = asoc->rto_initial;
139 
140 	/* sctpimpguide Section 2.12.2
141 	 * If the 'T5-shutdown-guard' timer is used, it SHOULD be set to the
142 	 * recommended value of 5 times 'RTO.Max'.
143 	 */
144 	asoc->timeouts[SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]
145 		= 5 * asoc->rto_max;
146 
147 	asoc->timeouts[SCTP_EVENT_TIMEOUT_SACK] = asoc->sackdelay;
148 	asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE] = sp->autoclose * HZ;
149 
150 	/* Initializes the timers */
151 	for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i)
152 		timer_setup(&asoc->timers[i], sctp_timer_events[i], 0);
153 
154 	/* Pull default initialization values from the sock options.
155 	 * Note: This assumes that the values have already been
156 	 * validated in the sock.
157 	 */
158 	asoc->c.sinit_max_instreams = sp->initmsg.sinit_max_instreams;
159 	asoc->c.sinit_num_ostreams  = sp->initmsg.sinit_num_ostreams;
160 	asoc->max_init_attempts	= sp->initmsg.sinit_max_attempts;
161 
162 	asoc->max_init_timeo =
163 		 msecs_to_jiffies(sp->initmsg.sinit_max_init_timeo);
164 
165 	/* Set the local window size for receive.
166 	 * This is also the rcvbuf space per association.
167 	 * RFC 6 - A SCTP receiver MUST be able to receive a minimum of
168 	 * 1500 bytes in one SCTP packet.
169 	 */
170 	if ((sk->sk_rcvbuf/2) < SCTP_DEFAULT_MINWINDOW)
171 		asoc->rwnd = SCTP_DEFAULT_MINWINDOW;
172 	else
173 		asoc->rwnd = sk->sk_rcvbuf/2;
174 
175 	asoc->a_rwnd = asoc->rwnd;
176 
177 	/* Use my own max window until I learn something better.  */
178 	asoc->peer.rwnd = SCTP_DEFAULT_MAXWINDOW;
179 
180 	/* Initialize the receive memory counter */
181 	atomic_set(&asoc->rmem_alloc, 0);
182 
183 	init_waitqueue_head(&asoc->wait);
184 
185 	asoc->c.my_vtag = sctp_generate_tag(ep);
186 	asoc->c.my_port = ep->base.bind_addr.port;
187 
188 	asoc->c.initial_tsn = sctp_generate_tsn(ep);
189 
190 	asoc->next_tsn = asoc->c.initial_tsn;
191 
192 	asoc->ctsn_ack_point = asoc->next_tsn - 1;
193 	asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
194 	asoc->highest_sacked = asoc->ctsn_ack_point;
195 	asoc->last_cwr_tsn = asoc->ctsn_ack_point;
196 
197 	/* ADDIP Section 4.1 Asconf Chunk Procedures
198 	 *
199 	 * When an endpoint has an ASCONF signaled change to be sent to the
200 	 * remote endpoint it should do the following:
201 	 * ...
202 	 * A2) a serial number should be assigned to the chunk. The serial
203 	 * number SHOULD be a monotonically increasing number. The serial
204 	 * numbers SHOULD be initialized at the start of the
205 	 * association to the same value as the initial TSN.
206 	 */
207 	asoc->addip_serial = asoc->c.initial_tsn;
208 	asoc->strreset_outseq = asoc->c.initial_tsn;
209 
210 	INIT_LIST_HEAD(&asoc->addip_chunk_list);
211 	INIT_LIST_HEAD(&asoc->asconf_ack_list);
212 
213 	/* Make an empty list of remote transport addresses.  */
214 	INIT_LIST_HEAD(&asoc->peer.transport_addr_list);
215 
216 	/* RFC 2960 5.1 Normal Establishment of an Association
217 	 *
218 	 * After the reception of the first data chunk in an
219 	 * association the endpoint must immediately respond with a
220 	 * sack to acknowledge the data chunk.  Subsequent
221 	 * acknowledgements should be done as described in Section
222 	 * 6.2.
223 	 *
224 	 * [We implement this by telling a new association that it
225 	 * already received one packet.]
226 	 */
227 	asoc->peer.sack_needed = 1;
228 	asoc->peer.sack_generation = 1;
229 
230 	/* Assume that the peer will tell us if he recognizes ASCONF
231 	 * as part of INIT exchange.
232 	 * The sctp_addip_noauth option is there for backward compatibility
233 	 * and will revert old behavior.
234 	 */
235 	if (net->sctp.addip_noauth)
236 		asoc->peer.asconf_capable = 1;
237 
238 	/* Create an input queue.  */
239 	sctp_inq_init(&asoc->base.inqueue);
240 	sctp_inq_set_th_handler(&asoc->base.inqueue, sctp_assoc_bh_rcv);
241 
242 	/* Create an output queue.  */
243 	sctp_outq_init(asoc, &asoc->outqueue);
244 
245 	if (!sctp_ulpq_init(&asoc->ulpq, asoc))
246 		goto fail_init;
247 
248 	if (sctp_stream_init(&asoc->stream, asoc->c.sinit_num_ostreams,
249 			     0, gfp))
250 		goto fail_init;
251 
252 	/* Assume that peer would support both address types unless we are
253 	 * told otherwise.
254 	 */
255 	asoc->peer.ipv4_address = 1;
256 	if (asoc->base.sk->sk_family == PF_INET6)
257 		asoc->peer.ipv6_address = 1;
258 	INIT_LIST_HEAD(&asoc->asocs);
259 
260 	asoc->default_stream = sp->default_stream;
261 	asoc->default_ppid = sp->default_ppid;
262 	asoc->default_flags = sp->default_flags;
263 	asoc->default_context = sp->default_context;
264 	asoc->default_timetolive = sp->default_timetolive;
265 	asoc->default_rcv_context = sp->default_rcv_context;
266 
267 	/* AUTH related initializations */
268 	INIT_LIST_HEAD(&asoc->endpoint_shared_keys);
269 	if (sctp_auth_asoc_copy_shkeys(ep, asoc, gfp))
270 		goto stream_free;
271 
272 	asoc->active_key_id = ep->active_key_id;
273 	asoc->prsctp_enable = ep->prsctp_enable;
274 	asoc->reconf_enable = ep->reconf_enable;
275 	asoc->strreset_enable = ep->strreset_enable;
276 
277 	/* Save the hmacs and chunks list into this association */
278 	if (ep->auth_hmacs_list)
279 		memcpy(asoc->c.auth_hmacs, ep->auth_hmacs_list,
280 			ntohs(ep->auth_hmacs_list->param_hdr.length));
281 	if (ep->auth_chunk_list)
282 		memcpy(asoc->c.auth_chunks, ep->auth_chunk_list,
283 			ntohs(ep->auth_chunk_list->param_hdr.length));
284 
285 	/* Get the AUTH random number for this association */
286 	p = (struct sctp_paramhdr *)asoc->c.auth_random;
287 	p->type = SCTP_PARAM_RANDOM;
288 	p->length = htons(sizeof(*p) + SCTP_AUTH_RANDOM_LENGTH);
289 	get_random_bytes(p+1, SCTP_AUTH_RANDOM_LENGTH);
290 
291 	return asoc;
292 
293 stream_free:
294 	sctp_stream_free(&asoc->stream);
295 fail_init:
296 	sock_put(asoc->base.sk);
297 	sctp_endpoint_put(asoc->ep);
298 	return NULL;
299 }
300 
301 /* Allocate and initialize a new association */
302 struct sctp_association *sctp_association_new(const struct sctp_endpoint *ep,
303 					      const struct sock *sk,
304 					      enum sctp_scope scope, gfp_t gfp)
305 {
306 	struct sctp_association *asoc;
307 
308 	asoc = kzalloc(sizeof(*asoc), gfp);
309 	if (!asoc)
310 		goto fail;
311 
312 	if (!sctp_association_init(asoc, ep, sk, scope, gfp))
313 		goto fail_init;
314 
315 	SCTP_DBG_OBJCNT_INC(assoc);
316 
317 	pr_debug("Created asoc %p\n", asoc);
318 
319 	return asoc;
320 
321 fail_init:
322 	kfree(asoc);
323 fail:
324 	return NULL;
325 }
326 
327 /* Free this association if possible.  There may still be users, so
328  * the actual deallocation may be delayed.
329  */
330 void sctp_association_free(struct sctp_association *asoc)
331 {
332 	struct sock *sk = asoc->base.sk;
333 	struct sctp_transport *transport;
334 	struct list_head *pos, *temp;
335 	int i;
336 
337 	/* Only real associations count against the endpoint, so
338 	 * don't bother for if this is a temporary association.
339 	 */
340 	if (!list_empty(&asoc->asocs)) {
341 		list_del(&asoc->asocs);
342 
343 		/* Decrement the backlog value for a TCP-style listening
344 		 * socket.
345 		 */
346 		if (sctp_style(sk, TCP) && sctp_sstate(sk, LISTENING))
347 			sk->sk_ack_backlog--;
348 	}
349 
350 	/* Mark as dead, so other users can know this structure is
351 	 * going away.
352 	 */
353 	asoc->base.dead = true;
354 
355 	/* Dispose of any data lying around in the outqueue. */
356 	sctp_outq_free(&asoc->outqueue);
357 
358 	/* Dispose of any pending messages for the upper layer. */
359 	sctp_ulpq_free(&asoc->ulpq);
360 
361 	/* Dispose of any pending chunks on the inqueue. */
362 	sctp_inq_free(&asoc->base.inqueue);
363 
364 	sctp_tsnmap_free(&asoc->peer.tsn_map);
365 
366 	/* Free stream information. */
367 	sctp_stream_free(&asoc->stream);
368 
369 	if (asoc->strreset_chunk)
370 		sctp_chunk_free(asoc->strreset_chunk);
371 
372 	/* Clean up the bound address list. */
373 	sctp_bind_addr_free(&asoc->base.bind_addr);
374 
375 	/* Do we need to go through all of our timers and
376 	 * delete them?   To be safe we will try to delete all, but we
377 	 * should be able to go through and make a guess based
378 	 * on our state.
379 	 */
380 	for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i) {
381 		if (del_timer(&asoc->timers[i]))
382 			sctp_association_put(asoc);
383 	}
384 
385 	/* Free peer's cached cookie. */
386 	kfree(asoc->peer.cookie);
387 	kfree(asoc->peer.peer_random);
388 	kfree(asoc->peer.peer_chunks);
389 	kfree(asoc->peer.peer_hmacs);
390 
391 	/* Release the transport structures. */
392 	list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
393 		transport = list_entry(pos, struct sctp_transport, transports);
394 		list_del_rcu(pos);
395 		sctp_unhash_transport(transport);
396 		sctp_transport_free(transport);
397 	}
398 
399 	asoc->peer.transport_count = 0;
400 
401 	sctp_asconf_queue_teardown(asoc);
402 
403 	/* Free pending address space being deleted */
404 	kfree(asoc->asconf_addr_del_pending);
405 
406 	/* AUTH - Free the endpoint shared keys */
407 	sctp_auth_destroy_keys(&asoc->endpoint_shared_keys);
408 
409 	/* AUTH - Free the association shared key */
410 	sctp_auth_key_put(asoc->asoc_shared_key);
411 
412 	sctp_association_put(asoc);
413 }
414 
415 /* Cleanup and free up an association. */
416 static void sctp_association_destroy(struct sctp_association *asoc)
417 {
418 	if (unlikely(!asoc->base.dead)) {
419 		WARN(1, "Attempt to destroy undead association %p!\n", asoc);
420 		return;
421 	}
422 
423 	sctp_endpoint_put(asoc->ep);
424 	sock_put(asoc->base.sk);
425 
426 	if (asoc->assoc_id != 0) {
427 		spin_lock_bh(&sctp_assocs_id_lock);
428 		idr_remove(&sctp_assocs_id, asoc->assoc_id);
429 		spin_unlock_bh(&sctp_assocs_id_lock);
430 	}
431 
432 	WARN_ON(atomic_read(&asoc->rmem_alloc));
433 
434 	kfree(asoc);
435 	SCTP_DBG_OBJCNT_DEC(assoc);
436 }
437 
438 /* Change the primary destination address for the peer. */
439 void sctp_assoc_set_primary(struct sctp_association *asoc,
440 			    struct sctp_transport *transport)
441 {
442 	int changeover = 0;
443 
444 	/* it's a changeover only if we already have a primary path
445 	 * that we are changing
446 	 */
447 	if (asoc->peer.primary_path != NULL &&
448 	    asoc->peer.primary_path != transport)
449 		changeover = 1 ;
450 
451 	asoc->peer.primary_path = transport;
452 
453 	/* Set a default msg_name for events. */
454 	memcpy(&asoc->peer.primary_addr, &transport->ipaddr,
455 	       sizeof(union sctp_addr));
456 
457 	/* If the primary path is changing, assume that the
458 	 * user wants to use this new path.
459 	 */
460 	if ((transport->state == SCTP_ACTIVE) ||
461 	    (transport->state == SCTP_UNKNOWN))
462 		asoc->peer.active_path = transport;
463 
464 	/*
465 	 * SFR-CACC algorithm:
466 	 * Upon the receipt of a request to change the primary
467 	 * destination address, on the data structure for the new
468 	 * primary destination, the sender MUST do the following:
469 	 *
470 	 * 1) If CHANGEOVER_ACTIVE is set, then there was a switch
471 	 * to this destination address earlier. The sender MUST set
472 	 * CYCLING_CHANGEOVER to indicate that this switch is a
473 	 * double switch to the same destination address.
474 	 *
475 	 * Really, only bother is we have data queued or outstanding on
476 	 * the association.
477 	 */
478 	if (!asoc->outqueue.outstanding_bytes && !asoc->outqueue.out_qlen)
479 		return;
480 
481 	if (transport->cacc.changeover_active)
482 		transport->cacc.cycling_changeover = changeover;
483 
484 	/* 2) The sender MUST set CHANGEOVER_ACTIVE to indicate that
485 	 * a changeover has occurred.
486 	 */
487 	transport->cacc.changeover_active = changeover;
488 
489 	/* 3) The sender MUST store the next TSN to be sent in
490 	 * next_tsn_at_change.
491 	 */
492 	transport->cacc.next_tsn_at_change = asoc->next_tsn;
493 }
494 
495 /* Remove a transport from an association.  */
496 void sctp_assoc_rm_peer(struct sctp_association *asoc,
497 			struct sctp_transport *peer)
498 {
499 	struct list_head	*pos;
500 	struct sctp_transport	*transport;
501 
502 	pr_debug("%s: association:%p addr:%pISpc\n",
503 		 __func__, asoc, &peer->ipaddr.sa);
504 
505 	/* If we are to remove the current retran_path, update it
506 	 * to the next peer before removing this peer from the list.
507 	 */
508 	if (asoc->peer.retran_path == peer)
509 		sctp_assoc_update_retran_path(asoc);
510 
511 	/* Remove this peer from the list. */
512 	list_del_rcu(&peer->transports);
513 	/* Remove this peer from the transport hashtable */
514 	sctp_unhash_transport(peer);
515 
516 	/* Get the first transport of asoc. */
517 	pos = asoc->peer.transport_addr_list.next;
518 	transport = list_entry(pos, struct sctp_transport, transports);
519 
520 	/* Update any entries that match the peer to be deleted. */
521 	if (asoc->peer.primary_path == peer)
522 		sctp_assoc_set_primary(asoc, transport);
523 	if (asoc->peer.active_path == peer)
524 		asoc->peer.active_path = transport;
525 	if (asoc->peer.retran_path == peer)
526 		asoc->peer.retran_path = transport;
527 	if (asoc->peer.last_data_from == peer)
528 		asoc->peer.last_data_from = transport;
529 
530 	if (asoc->strreset_chunk &&
531 	    asoc->strreset_chunk->transport == peer) {
532 		asoc->strreset_chunk->transport = transport;
533 		sctp_transport_reset_reconf_timer(transport);
534 	}
535 
536 	/* If we remove the transport an INIT was last sent to, set it to
537 	 * NULL. Combined with the update of the retran path above, this
538 	 * will cause the next INIT to be sent to the next available
539 	 * transport, maintaining the cycle.
540 	 */
541 	if (asoc->init_last_sent_to == peer)
542 		asoc->init_last_sent_to = NULL;
543 
544 	/* If we remove the transport an SHUTDOWN was last sent to, set it
545 	 * to NULL. Combined with the update of the retran path above, this
546 	 * will cause the next SHUTDOWN to be sent to the next available
547 	 * transport, maintaining the cycle.
548 	 */
549 	if (asoc->shutdown_last_sent_to == peer)
550 		asoc->shutdown_last_sent_to = NULL;
551 
552 	/* If we remove the transport an ASCONF was last sent to, set it to
553 	 * NULL.
554 	 */
555 	if (asoc->addip_last_asconf &&
556 	    asoc->addip_last_asconf->transport == peer)
557 		asoc->addip_last_asconf->transport = NULL;
558 
559 	/* If we have something on the transmitted list, we have to
560 	 * save it off.  The best place is the active path.
561 	 */
562 	if (!list_empty(&peer->transmitted)) {
563 		struct sctp_transport *active = asoc->peer.active_path;
564 		struct sctp_chunk *ch;
565 
566 		/* Reset the transport of each chunk on this list */
567 		list_for_each_entry(ch, &peer->transmitted,
568 					transmitted_list) {
569 			ch->transport = NULL;
570 			ch->rtt_in_progress = 0;
571 		}
572 
573 		list_splice_tail_init(&peer->transmitted,
574 					&active->transmitted);
575 
576 		/* Start a T3 timer here in case it wasn't running so
577 		 * that these migrated packets have a chance to get
578 		 * retransmitted.
579 		 */
580 		if (!timer_pending(&active->T3_rtx_timer))
581 			if (!mod_timer(&active->T3_rtx_timer,
582 					jiffies + active->rto))
583 				sctp_transport_hold(active);
584 	}
585 
586 	asoc->peer.transport_count--;
587 
588 	sctp_transport_free(peer);
589 }
590 
591 /* Add a transport address to an association.  */
592 struct sctp_transport *sctp_assoc_add_peer(struct sctp_association *asoc,
593 					   const union sctp_addr *addr,
594 					   const gfp_t gfp,
595 					   const int peer_state)
596 {
597 	struct net *net = sock_net(asoc->base.sk);
598 	struct sctp_transport *peer;
599 	struct sctp_sock *sp;
600 	unsigned short port;
601 
602 	sp = sctp_sk(asoc->base.sk);
603 
604 	/* AF_INET and AF_INET6 share common port field. */
605 	port = ntohs(addr->v4.sin_port);
606 
607 	pr_debug("%s: association:%p addr:%pISpc state:%d\n", __func__,
608 		 asoc, &addr->sa, peer_state);
609 
610 	/* Set the port if it has not been set yet.  */
611 	if (0 == asoc->peer.port)
612 		asoc->peer.port = port;
613 
614 	/* Check to see if this is a duplicate. */
615 	peer = sctp_assoc_lookup_paddr(asoc, addr);
616 	if (peer) {
617 		/* An UNKNOWN state is only set on transports added by
618 		 * user in sctp_connectx() call.  Such transports should be
619 		 * considered CONFIRMED per RFC 4960, Section 5.4.
620 		 */
621 		if (peer->state == SCTP_UNKNOWN) {
622 			peer->state = SCTP_ACTIVE;
623 		}
624 		return peer;
625 	}
626 
627 	peer = sctp_transport_new(net, addr, gfp);
628 	if (!peer)
629 		return NULL;
630 
631 	sctp_transport_set_owner(peer, asoc);
632 
633 	/* Initialize the peer's heartbeat interval based on the
634 	 * association configured value.
635 	 */
636 	peer->hbinterval = asoc->hbinterval;
637 
638 	/* Set the path max_retrans.  */
639 	peer->pathmaxrxt = asoc->pathmaxrxt;
640 
641 	/* And the partial failure retrans threshold */
642 	peer->pf_retrans = asoc->pf_retrans;
643 
644 	/* Initialize the peer's SACK delay timeout based on the
645 	 * association configured value.
646 	 */
647 	peer->sackdelay = asoc->sackdelay;
648 	peer->sackfreq = asoc->sackfreq;
649 
650 	/* Enable/disable heartbeat, SACK delay, and path MTU discovery
651 	 * based on association setting.
652 	 */
653 	peer->param_flags = asoc->param_flags;
654 
655 	sctp_transport_route(peer, NULL, sp);
656 
657 	/* Initialize the pmtu of the transport. */
658 	if (peer->param_flags & SPP_PMTUD_DISABLE) {
659 		if (asoc->pathmtu)
660 			peer->pathmtu = asoc->pathmtu;
661 		else
662 			peer->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
663 	}
664 
665 	/* If this is the first transport addr on this association,
666 	 * initialize the association PMTU to the peer's PMTU.
667 	 * If not and the current association PMTU is higher than the new
668 	 * peer's PMTU, reset the association PMTU to the new peer's PMTU.
669 	 */
670 	if (asoc->pathmtu)
671 		asoc->pathmtu = min_t(int, peer->pathmtu, asoc->pathmtu);
672 	else
673 		asoc->pathmtu = peer->pathmtu;
674 
675 	pr_debug("%s: association:%p PMTU set to %d\n", __func__, asoc,
676 		 asoc->pathmtu);
677 
678 	peer->pmtu_pending = 0;
679 
680 	asoc->frag_point = sctp_frag_point(asoc, asoc->pathmtu);
681 
682 	/* The asoc->peer.port might not be meaningful yet, but
683 	 * initialize the packet structure anyway.
684 	 */
685 	sctp_packet_init(&peer->packet, peer, asoc->base.bind_addr.port,
686 			 asoc->peer.port);
687 
688 	/* 7.2.1 Slow-Start
689 	 *
690 	 * o The initial cwnd before DATA transmission or after a sufficiently
691 	 *   long idle period MUST be set to
692 	 *      min(4*MTU, max(2*MTU, 4380 bytes))
693 	 *
694 	 * o The initial value of ssthresh MAY be arbitrarily high
695 	 *   (for example, implementations MAY use the size of the
696 	 *   receiver advertised window).
697 	 */
698 	peer->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
699 
700 	/* At this point, we may not have the receiver's advertised window,
701 	 * so initialize ssthresh to the default value and it will be set
702 	 * later when we process the INIT.
703 	 */
704 	peer->ssthresh = SCTP_DEFAULT_MAXWINDOW;
705 
706 	peer->partial_bytes_acked = 0;
707 	peer->flight_size = 0;
708 	peer->burst_limited = 0;
709 
710 	/* Set the transport's RTO.initial value */
711 	peer->rto = asoc->rto_initial;
712 	sctp_max_rto(asoc, peer);
713 
714 	/* Set the peer's active state. */
715 	peer->state = peer_state;
716 
717 	/* Add this peer into the transport hashtable */
718 	if (sctp_hash_transport(peer)) {
719 		sctp_transport_free(peer);
720 		return NULL;
721 	}
722 
723 	/* Attach the remote transport to our asoc.  */
724 	list_add_tail_rcu(&peer->transports, &asoc->peer.transport_addr_list);
725 	asoc->peer.transport_count++;
726 
727 	/* If we do not yet have a primary path, set one.  */
728 	if (!asoc->peer.primary_path) {
729 		sctp_assoc_set_primary(asoc, peer);
730 		asoc->peer.retran_path = peer;
731 	}
732 
733 	if (asoc->peer.active_path == asoc->peer.retran_path &&
734 	    peer->state != SCTP_UNCONFIRMED) {
735 		asoc->peer.retran_path = peer;
736 	}
737 
738 	return peer;
739 }
740 
741 /* Delete a transport address from an association.  */
742 void sctp_assoc_del_peer(struct sctp_association *asoc,
743 			 const union sctp_addr *addr)
744 {
745 	struct list_head	*pos;
746 	struct list_head	*temp;
747 	struct sctp_transport	*transport;
748 
749 	list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
750 		transport = list_entry(pos, struct sctp_transport, transports);
751 		if (sctp_cmp_addr_exact(addr, &transport->ipaddr)) {
752 			/* Do book keeping for removing the peer and free it. */
753 			sctp_assoc_rm_peer(asoc, transport);
754 			break;
755 		}
756 	}
757 }
758 
759 /* Lookup a transport by address. */
760 struct sctp_transport *sctp_assoc_lookup_paddr(
761 					const struct sctp_association *asoc,
762 					const union sctp_addr *address)
763 {
764 	struct sctp_transport *t;
765 
766 	/* Cycle through all transports searching for a peer address. */
767 
768 	list_for_each_entry(t, &asoc->peer.transport_addr_list,
769 			transports) {
770 		if (sctp_cmp_addr_exact(address, &t->ipaddr))
771 			return t;
772 	}
773 
774 	return NULL;
775 }
776 
777 /* Remove all transports except a give one */
778 void sctp_assoc_del_nonprimary_peers(struct sctp_association *asoc,
779 				     struct sctp_transport *primary)
780 {
781 	struct sctp_transport	*temp;
782 	struct sctp_transport	*t;
783 
784 	list_for_each_entry_safe(t, temp, &asoc->peer.transport_addr_list,
785 				 transports) {
786 		/* if the current transport is not the primary one, delete it */
787 		if (t != primary)
788 			sctp_assoc_rm_peer(asoc, t);
789 	}
790 }
791 
792 /* Engage in transport control operations.
793  * Mark the transport up or down and send a notification to the user.
794  * Select and update the new active and retran paths.
795  */
796 void sctp_assoc_control_transport(struct sctp_association *asoc,
797 				  struct sctp_transport *transport,
798 				  enum sctp_transport_cmd command,
799 				  sctp_sn_error_t error)
800 {
801 	struct sctp_ulpevent *event;
802 	struct sockaddr_storage addr;
803 	int spc_state = 0;
804 	bool ulp_notify = true;
805 
806 	/* Record the transition on the transport.  */
807 	switch (command) {
808 	case SCTP_TRANSPORT_UP:
809 		/* If we are moving from UNCONFIRMED state due
810 		 * to heartbeat success, report the SCTP_ADDR_CONFIRMED
811 		 * state to the user, otherwise report SCTP_ADDR_AVAILABLE.
812 		 */
813 		if (SCTP_UNCONFIRMED == transport->state &&
814 		    SCTP_HEARTBEAT_SUCCESS == error)
815 			spc_state = SCTP_ADDR_CONFIRMED;
816 		else
817 			spc_state = SCTP_ADDR_AVAILABLE;
818 		/* Don't inform ULP about transition from PF to
819 		 * active state and set cwnd to 1 MTU, see SCTP
820 		 * Quick failover draft section 5.1, point 5
821 		 */
822 		if (transport->state == SCTP_PF) {
823 			ulp_notify = false;
824 			transport->cwnd = asoc->pathmtu;
825 		}
826 		transport->state = SCTP_ACTIVE;
827 		break;
828 
829 	case SCTP_TRANSPORT_DOWN:
830 		/* If the transport was never confirmed, do not transition it
831 		 * to inactive state.  Also, release the cached route since
832 		 * there may be a better route next time.
833 		 */
834 		if (transport->state != SCTP_UNCONFIRMED)
835 			transport->state = SCTP_INACTIVE;
836 		else {
837 			sctp_transport_dst_release(transport);
838 			ulp_notify = false;
839 		}
840 
841 		spc_state = SCTP_ADDR_UNREACHABLE;
842 		break;
843 
844 	case SCTP_TRANSPORT_PF:
845 		transport->state = SCTP_PF;
846 		ulp_notify = false;
847 		break;
848 
849 	default:
850 		return;
851 	}
852 
853 	/* Generate and send a SCTP_PEER_ADDR_CHANGE notification
854 	 * to the user.
855 	 */
856 	if (ulp_notify) {
857 		memset(&addr, 0, sizeof(struct sockaddr_storage));
858 		memcpy(&addr, &transport->ipaddr,
859 		       transport->af_specific->sockaddr_len);
860 
861 		event = sctp_ulpevent_make_peer_addr_change(asoc, &addr,
862 					0, spc_state, error, GFP_ATOMIC);
863 		if (event)
864 			asoc->stream.si->enqueue_event(&asoc->ulpq, event);
865 	}
866 
867 	/* Select new active and retran paths. */
868 	sctp_select_active_and_retran_path(asoc);
869 }
870 
871 /* Hold a reference to an association. */
872 void sctp_association_hold(struct sctp_association *asoc)
873 {
874 	refcount_inc(&asoc->base.refcnt);
875 }
876 
877 /* Release a reference to an association and cleanup
878  * if there are no more references.
879  */
880 void sctp_association_put(struct sctp_association *asoc)
881 {
882 	if (refcount_dec_and_test(&asoc->base.refcnt))
883 		sctp_association_destroy(asoc);
884 }
885 
886 /* Allocate the next TSN, Transmission Sequence Number, for the given
887  * association.
888  */
889 __u32 sctp_association_get_next_tsn(struct sctp_association *asoc)
890 {
891 	/* From Section 1.6 Serial Number Arithmetic:
892 	 * Transmission Sequence Numbers wrap around when they reach
893 	 * 2**32 - 1.  That is, the next TSN a DATA chunk MUST use
894 	 * after transmitting TSN = 2*32 - 1 is TSN = 0.
895 	 */
896 	__u32 retval = asoc->next_tsn;
897 	asoc->next_tsn++;
898 	asoc->unack_data++;
899 
900 	return retval;
901 }
902 
903 /* Compare two addresses to see if they match.  Wildcard addresses
904  * only match themselves.
905  */
906 int sctp_cmp_addr_exact(const union sctp_addr *ss1,
907 			const union sctp_addr *ss2)
908 {
909 	struct sctp_af *af;
910 
911 	af = sctp_get_af_specific(ss1->sa.sa_family);
912 	if (unlikely(!af))
913 		return 0;
914 
915 	return af->cmp_addr(ss1, ss2);
916 }
917 
918 /* Return an ecne chunk to get prepended to a packet.
919  * Note:  We are sly and return a shared, prealloced chunk.  FIXME:
920  * No we don't, but we could/should.
921  */
922 struct sctp_chunk *sctp_get_ecne_prepend(struct sctp_association *asoc)
923 {
924 	if (!asoc->need_ecne)
925 		return NULL;
926 
927 	/* Send ECNE if needed.
928 	 * Not being able to allocate a chunk here is not deadly.
929 	 */
930 	return sctp_make_ecne(asoc, asoc->last_ecne_tsn);
931 }
932 
933 /*
934  * Find which transport this TSN was sent on.
935  */
936 struct sctp_transport *sctp_assoc_lookup_tsn(struct sctp_association *asoc,
937 					     __u32 tsn)
938 {
939 	struct sctp_transport *active;
940 	struct sctp_transport *match;
941 	struct sctp_transport *transport;
942 	struct sctp_chunk *chunk;
943 	__be32 key = htonl(tsn);
944 
945 	match = NULL;
946 
947 	/*
948 	 * FIXME: In general, find a more efficient data structure for
949 	 * searching.
950 	 */
951 
952 	/*
953 	 * The general strategy is to search each transport's transmitted
954 	 * list.   Return which transport this TSN lives on.
955 	 *
956 	 * Let's be hopeful and check the active_path first.
957 	 * Another optimization would be to know if there is only one
958 	 * outbound path and not have to look for the TSN at all.
959 	 *
960 	 */
961 
962 	active = asoc->peer.active_path;
963 
964 	list_for_each_entry(chunk, &active->transmitted,
965 			transmitted_list) {
966 
967 		if (key == chunk->subh.data_hdr->tsn) {
968 			match = active;
969 			goto out;
970 		}
971 	}
972 
973 	/* If not found, go search all the other transports. */
974 	list_for_each_entry(transport, &asoc->peer.transport_addr_list,
975 			transports) {
976 
977 		if (transport == active)
978 			continue;
979 		list_for_each_entry(chunk, &transport->transmitted,
980 				transmitted_list) {
981 			if (key == chunk->subh.data_hdr->tsn) {
982 				match = transport;
983 				goto out;
984 			}
985 		}
986 	}
987 out:
988 	return match;
989 }
990 
991 /* Is this the association we are looking for? */
992 struct sctp_transport *sctp_assoc_is_match(struct sctp_association *asoc,
993 					   struct net *net,
994 					   const union sctp_addr *laddr,
995 					   const union sctp_addr *paddr)
996 {
997 	struct sctp_transport *transport;
998 
999 	if ((htons(asoc->base.bind_addr.port) == laddr->v4.sin_port) &&
1000 	    (htons(asoc->peer.port) == paddr->v4.sin_port) &&
1001 	    net_eq(sock_net(asoc->base.sk), net)) {
1002 		transport = sctp_assoc_lookup_paddr(asoc, paddr);
1003 		if (!transport)
1004 			goto out;
1005 
1006 		if (sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
1007 					 sctp_sk(asoc->base.sk)))
1008 			goto out;
1009 	}
1010 	transport = NULL;
1011 
1012 out:
1013 	return transport;
1014 }
1015 
1016 /* Do delayed input processing.  This is scheduled by sctp_rcv(). */
1017 static void sctp_assoc_bh_rcv(struct work_struct *work)
1018 {
1019 	struct sctp_association *asoc =
1020 		container_of(work, struct sctp_association,
1021 			     base.inqueue.immediate);
1022 	struct net *net = sock_net(asoc->base.sk);
1023 	union sctp_subtype subtype;
1024 	struct sctp_endpoint *ep;
1025 	struct sctp_chunk *chunk;
1026 	struct sctp_inq *inqueue;
1027 	int state;
1028 	int error = 0;
1029 
1030 	/* The association should be held so we should be safe. */
1031 	ep = asoc->ep;
1032 
1033 	inqueue = &asoc->base.inqueue;
1034 	sctp_association_hold(asoc);
1035 	while (NULL != (chunk = sctp_inq_pop(inqueue))) {
1036 		state = asoc->state;
1037 		subtype = SCTP_ST_CHUNK(chunk->chunk_hdr->type);
1038 
1039 		/* SCTP-AUTH, Section 6.3:
1040 		 *    The receiver has a list of chunk types which it expects
1041 		 *    to be received only after an AUTH-chunk.  This list has
1042 		 *    been sent to the peer during the association setup.  It
1043 		 *    MUST silently discard these chunks if they are not placed
1044 		 *    after an AUTH chunk in the packet.
1045 		 */
1046 		if (sctp_auth_recv_cid(subtype.chunk, asoc) && !chunk->auth)
1047 			continue;
1048 
1049 		/* Remember where the last DATA chunk came from so we
1050 		 * know where to send the SACK.
1051 		 */
1052 		if (sctp_chunk_is_data(chunk))
1053 			asoc->peer.last_data_from = chunk->transport;
1054 		else {
1055 			SCTP_INC_STATS(net, SCTP_MIB_INCTRLCHUNKS);
1056 			asoc->stats.ictrlchunks++;
1057 			if (chunk->chunk_hdr->type == SCTP_CID_SACK)
1058 				asoc->stats.isacks++;
1059 		}
1060 
1061 		if (chunk->transport)
1062 			chunk->transport->last_time_heard = ktime_get();
1063 
1064 		/* Run through the state machine. */
1065 		error = sctp_do_sm(net, SCTP_EVENT_T_CHUNK, subtype,
1066 				   state, ep, asoc, chunk, GFP_ATOMIC);
1067 
1068 		/* Check to see if the association is freed in response to
1069 		 * the incoming chunk.  If so, get out of the while loop.
1070 		 */
1071 		if (asoc->base.dead)
1072 			break;
1073 
1074 		/* If there is an error on chunk, discard this packet. */
1075 		if (error && chunk)
1076 			chunk->pdiscard = 1;
1077 	}
1078 	sctp_association_put(asoc);
1079 }
1080 
1081 /* This routine moves an association from its old sk to a new sk.  */
1082 void sctp_assoc_migrate(struct sctp_association *assoc, struct sock *newsk)
1083 {
1084 	struct sctp_sock *newsp = sctp_sk(newsk);
1085 	struct sock *oldsk = assoc->base.sk;
1086 
1087 	/* Delete the association from the old endpoint's list of
1088 	 * associations.
1089 	 */
1090 	list_del_init(&assoc->asocs);
1091 
1092 	/* Decrement the backlog value for a TCP-style socket. */
1093 	if (sctp_style(oldsk, TCP))
1094 		oldsk->sk_ack_backlog--;
1095 
1096 	/* Release references to the old endpoint and the sock.  */
1097 	sctp_endpoint_put(assoc->ep);
1098 	sock_put(assoc->base.sk);
1099 
1100 	/* Get a reference to the new endpoint.  */
1101 	assoc->ep = newsp->ep;
1102 	sctp_endpoint_hold(assoc->ep);
1103 
1104 	/* Get a reference to the new sock.  */
1105 	assoc->base.sk = newsk;
1106 	sock_hold(assoc->base.sk);
1107 
1108 	/* Add the association to the new endpoint's list of associations.  */
1109 	sctp_endpoint_add_asoc(newsp->ep, assoc);
1110 }
1111 
1112 /* Update an association (possibly from unexpected COOKIE-ECHO processing).  */
1113 int sctp_assoc_update(struct sctp_association *asoc,
1114 		      struct sctp_association *new)
1115 {
1116 	struct sctp_transport *trans;
1117 	struct list_head *pos, *temp;
1118 
1119 	/* Copy in new parameters of peer. */
1120 	asoc->c = new->c;
1121 	asoc->peer.rwnd = new->peer.rwnd;
1122 	asoc->peer.sack_needed = new->peer.sack_needed;
1123 	asoc->peer.auth_capable = new->peer.auth_capable;
1124 	asoc->peer.i = new->peer.i;
1125 
1126 	if (!sctp_tsnmap_init(&asoc->peer.tsn_map, SCTP_TSN_MAP_INITIAL,
1127 			      asoc->peer.i.initial_tsn, GFP_ATOMIC))
1128 		return -ENOMEM;
1129 
1130 	/* Remove any peer addresses not present in the new association. */
1131 	list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
1132 		trans = list_entry(pos, struct sctp_transport, transports);
1133 		if (!sctp_assoc_lookup_paddr(new, &trans->ipaddr)) {
1134 			sctp_assoc_rm_peer(asoc, trans);
1135 			continue;
1136 		}
1137 
1138 		if (asoc->state >= SCTP_STATE_ESTABLISHED)
1139 			sctp_transport_reset(trans);
1140 	}
1141 
1142 	/* If the case is A (association restart), use
1143 	 * initial_tsn as next_tsn. If the case is B, use
1144 	 * current next_tsn in case data sent to peer
1145 	 * has been discarded and needs retransmission.
1146 	 */
1147 	if (asoc->state >= SCTP_STATE_ESTABLISHED) {
1148 		asoc->next_tsn = new->next_tsn;
1149 		asoc->ctsn_ack_point = new->ctsn_ack_point;
1150 		asoc->adv_peer_ack_point = new->adv_peer_ack_point;
1151 
1152 		/* Reinitialize SSN for both local streams
1153 		 * and peer's streams.
1154 		 */
1155 		sctp_stream_clear(&asoc->stream);
1156 
1157 		/* Flush the ULP reassembly and ordered queue.
1158 		 * Any data there will now be stale and will
1159 		 * cause problems.
1160 		 */
1161 		sctp_ulpq_flush(&asoc->ulpq);
1162 
1163 		/* reset the overall association error count so
1164 		 * that the restarted association doesn't get torn
1165 		 * down on the next retransmission timer.
1166 		 */
1167 		asoc->overall_error_count = 0;
1168 
1169 	} else {
1170 		/* Add any peer addresses from the new association. */
1171 		list_for_each_entry(trans, &new->peer.transport_addr_list,
1172 				    transports)
1173 			if (!sctp_assoc_lookup_paddr(asoc, &trans->ipaddr) &&
1174 			    !sctp_assoc_add_peer(asoc, &trans->ipaddr,
1175 						 GFP_ATOMIC, trans->state))
1176 				return -ENOMEM;
1177 
1178 		asoc->ctsn_ack_point = asoc->next_tsn - 1;
1179 		asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
1180 
1181 		if (sctp_state(asoc, COOKIE_WAIT))
1182 			sctp_stream_update(&asoc->stream, &new->stream);
1183 
1184 		/* get a new assoc id if we don't have one yet. */
1185 		if (sctp_assoc_set_id(asoc, GFP_ATOMIC))
1186 			return -ENOMEM;
1187 	}
1188 
1189 	/* SCTP-AUTH: Save the peer parameters from the new associations
1190 	 * and also move the association shared keys over
1191 	 */
1192 	kfree(asoc->peer.peer_random);
1193 	asoc->peer.peer_random = new->peer.peer_random;
1194 	new->peer.peer_random = NULL;
1195 
1196 	kfree(asoc->peer.peer_chunks);
1197 	asoc->peer.peer_chunks = new->peer.peer_chunks;
1198 	new->peer.peer_chunks = NULL;
1199 
1200 	kfree(asoc->peer.peer_hmacs);
1201 	asoc->peer.peer_hmacs = new->peer.peer_hmacs;
1202 	new->peer.peer_hmacs = NULL;
1203 
1204 	return sctp_auth_asoc_init_active_key(asoc, GFP_ATOMIC);
1205 }
1206 
1207 /* Update the retran path for sending a retransmitted packet.
1208  * See also RFC4960, 6.4. Multi-Homed SCTP Endpoints:
1209  *
1210  *   When there is outbound data to send and the primary path
1211  *   becomes inactive (e.g., due to failures), or where the
1212  *   SCTP user explicitly requests to send data to an
1213  *   inactive destination transport address, before reporting
1214  *   an error to its ULP, the SCTP endpoint should try to send
1215  *   the data to an alternate active destination transport
1216  *   address if one exists.
1217  *
1218  *   When retransmitting data that timed out, if the endpoint
1219  *   is multihomed, it should consider each source-destination
1220  *   address pair in its retransmission selection policy.
1221  *   When retransmitting timed-out data, the endpoint should
1222  *   attempt to pick the most divergent source-destination
1223  *   pair from the original source-destination pair to which
1224  *   the packet was transmitted.
1225  *
1226  *   Note: Rules for picking the most divergent source-destination
1227  *   pair are an implementation decision and are not specified
1228  *   within this document.
1229  *
1230  * Our basic strategy is to round-robin transports in priorities
1231  * according to sctp_trans_score() e.g., if no such
1232  * transport with state SCTP_ACTIVE exists, round-robin through
1233  * SCTP_UNKNOWN, etc. You get the picture.
1234  */
1235 static u8 sctp_trans_score(const struct sctp_transport *trans)
1236 {
1237 	switch (trans->state) {
1238 	case SCTP_ACTIVE:
1239 		return 3;	/* best case */
1240 	case SCTP_UNKNOWN:
1241 		return 2;
1242 	case SCTP_PF:
1243 		return 1;
1244 	default: /* case SCTP_INACTIVE */
1245 		return 0;	/* worst case */
1246 	}
1247 }
1248 
1249 static struct sctp_transport *sctp_trans_elect_tie(struct sctp_transport *trans1,
1250 						   struct sctp_transport *trans2)
1251 {
1252 	if (trans1->error_count > trans2->error_count) {
1253 		return trans2;
1254 	} else if (trans1->error_count == trans2->error_count &&
1255 		   ktime_after(trans2->last_time_heard,
1256 			       trans1->last_time_heard)) {
1257 		return trans2;
1258 	} else {
1259 		return trans1;
1260 	}
1261 }
1262 
1263 static struct sctp_transport *sctp_trans_elect_best(struct sctp_transport *curr,
1264 						    struct sctp_transport *best)
1265 {
1266 	u8 score_curr, score_best;
1267 
1268 	if (best == NULL || curr == best)
1269 		return curr;
1270 
1271 	score_curr = sctp_trans_score(curr);
1272 	score_best = sctp_trans_score(best);
1273 
1274 	/* First, try a score-based selection if both transport states
1275 	 * differ. If we're in a tie, lets try to make a more clever
1276 	 * decision here based on error counts and last time heard.
1277 	 */
1278 	if (score_curr > score_best)
1279 		return curr;
1280 	else if (score_curr == score_best)
1281 		return sctp_trans_elect_tie(best, curr);
1282 	else
1283 		return best;
1284 }
1285 
1286 void sctp_assoc_update_retran_path(struct sctp_association *asoc)
1287 {
1288 	struct sctp_transport *trans = asoc->peer.retran_path;
1289 	struct sctp_transport *trans_next = NULL;
1290 
1291 	/* We're done as we only have the one and only path. */
1292 	if (asoc->peer.transport_count == 1)
1293 		return;
1294 	/* If active_path and retran_path are the same and active,
1295 	 * then this is the only active path. Use it.
1296 	 */
1297 	if (asoc->peer.active_path == asoc->peer.retran_path &&
1298 	    asoc->peer.active_path->state == SCTP_ACTIVE)
1299 		return;
1300 
1301 	/* Iterate from retran_path's successor back to retran_path. */
1302 	for (trans = list_next_entry(trans, transports); 1;
1303 	     trans = list_next_entry(trans, transports)) {
1304 		/* Manually skip the head element. */
1305 		if (&trans->transports == &asoc->peer.transport_addr_list)
1306 			continue;
1307 		if (trans->state == SCTP_UNCONFIRMED)
1308 			continue;
1309 		trans_next = sctp_trans_elect_best(trans, trans_next);
1310 		/* Active is good enough for immediate return. */
1311 		if (trans_next->state == SCTP_ACTIVE)
1312 			break;
1313 		/* We've reached the end, time to update path. */
1314 		if (trans == asoc->peer.retran_path)
1315 			break;
1316 	}
1317 
1318 	asoc->peer.retran_path = trans_next;
1319 
1320 	pr_debug("%s: association:%p updated new path to addr:%pISpc\n",
1321 		 __func__, asoc, &asoc->peer.retran_path->ipaddr.sa);
1322 }
1323 
1324 static void sctp_select_active_and_retran_path(struct sctp_association *asoc)
1325 {
1326 	struct sctp_transport *trans, *trans_pri = NULL, *trans_sec = NULL;
1327 	struct sctp_transport *trans_pf = NULL;
1328 
1329 	/* Look for the two most recently used active transports. */
1330 	list_for_each_entry(trans, &asoc->peer.transport_addr_list,
1331 			    transports) {
1332 		/* Skip uninteresting transports. */
1333 		if (trans->state == SCTP_INACTIVE ||
1334 		    trans->state == SCTP_UNCONFIRMED)
1335 			continue;
1336 		/* Keep track of the best PF transport from our
1337 		 * list in case we don't find an active one.
1338 		 */
1339 		if (trans->state == SCTP_PF) {
1340 			trans_pf = sctp_trans_elect_best(trans, trans_pf);
1341 			continue;
1342 		}
1343 		/* For active transports, pick the most recent ones. */
1344 		if (trans_pri == NULL ||
1345 		    ktime_after(trans->last_time_heard,
1346 				trans_pri->last_time_heard)) {
1347 			trans_sec = trans_pri;
1348 			trans_pri = trans;
1349 		} else if (trans_sec == NULL ||
1350 			   ktime_after(trans->last_time_heard,
1351 				       trans_sec->last_time_heard)) {
1352 			trans_sec = trans;
1353 		}
1354 	}
1355 
1356 	/* RFC 2960 6.4 Multi-Homed SCTP Endpoints
1357 	 *
1358 	 * By default, an endpoint should always transmit to the primary
1359 	 * path, unless the SCTP user explicitly specifies the
1360 	 * destination transport address (and possibly source transport
1361 	 * address) to use. [If the primary is active but not most recent,
1362 	 * bump the most recently used transport.]
1363 	 */
1364 	if ((asoc->peer.primary_path->state == SCTP_ACTIVE ||
1365 	     asoc->peer.primary_path->state == SCTP_UNKNOWN) &&
1366 	     asoc->peer.primary_path != trans_pri) {
1367 		trans_sec = trans_pri;
1368 		trans_pri = asoc->peer.primary_path;
1369 	}
1370 
1371 	/* We did not find anything useful for a possible retransmission
1372 	 * path; either primary path that we found is the the same as
1373 	 * the current one, or we didn't generally find an active one.
1374 	 */
1375 	if (trans_sec == NULL)
1376 		trans_sec = trans_pri;
1377 
1378 	/* If we failed to find a usable transport, just camp on the
1379 	 * active or pick a PF iff it's the better choice.
1380 	 */
1381 	if (trans_pri == NULL) {
1382 		trans_pri = sctp_trans_elect_best(asoc->peer.active_path, trans_pf);
1383 		trans_sec = trans_pri;
1384 	}
1385 
1386 	/* Set the active and retran transports. */
1387 	asoc->peer.active_path = trans_pri;
1388 	asoc->peer.retran_path = trans_sec;
1389 }
1390 
1391 struct sctp_transport *
1392 sctp_assoc_choose_alter_transport(struct sctp_association *asoc,
1393 				  struct sctp_transport *last_sent_to)
1394 {
1395 	/* If this is the first time packet is sent, use the active path,
1396 	 * else use the retran path. If the last packet was sent over the
1397 	 * retran path, update the retran path and use it.
1398 	 */
1399 	if (last_sent_to == NULL) {
1400 		return asoc->peer.active_path;
1401 	} else {
1402 		if (last_sent_to == asoc->peer.retran_path)
1403 			sctp_assoc_update_retran_path(asoc);
1404 
1405 		return asoc->peer.retran_path;
1406 	}
1407 }
1408 
1409 /* Update the association's pmtu and frag_point by going through all the
1410  * transports. This routine is called when a transport's PMTU has changed.
1411  */
1412 void sctp_assoc_sync_pmtu(struct sctp_association *asoc)
1413 {
1414 	struct sctp_transport *t;
1415 	__u32 pmtu = 0;
1416 
1417 	if (!asoc)
1418 		return;
1419 
1420 	/* Get the lowest pmtu of all the transports. */
1421 	list_for_each_entry(t, &asoc->peer.transport_addr_list,
1422 				transports) {
1423 		if (t->pmtu_pending && t->dst) {
1424 			sctp_transport_update_pmtu(
1425 					t, SCTP_TRUNC4(dst_mtu(t->dst)));
1426 			t->pmtu_pending = 0;
1427 		}
1428 		if (!pmtu || (t->pathmtu < pmtu))
1429 			pmtu = t->pathmtu;
1430 	}
1431 
1432 	if (pmtu) {
1433 		asoc->pathmtu = pmtu;
1434 		asoc->frag_point = sctp_frag_point(asoc, pmtu);
1435 	}
1436 
1437 	pr_debug("%s: asoc:%p, pmtu:%d, frag_point:%d\n", __func__, asoc,
1438 		 asoc->pathmtu, asoc->frag_point);
1439 }
1440 
1441 /* Should we send a SACK to update our peer? */
1442 static inline bool sctp_peer_needs_update(struct sctp_association *asoc)
1443 {
1444 	struct net *net = sock_net(asoc->base.sk);
1445 	switch (asoc->state) {
1446 	case SCTP_STATE_ESTABLISHED:
1447 	case SCTP_STATE_SHUTDOWN_PENDING:
1448 	case SCTP_STATE_SHUTDOWN_RECEIVED:
1449 	case SCTP_STATE_SHUTDOWN_SENT:
1450 		if ((asoc->rwnd > asoc->a_rwnd) &&
1451 		    ((asoc->rwnd - asoc->a_rwnd) >= max_t(__u32,
1452 			   (asoc->base.sk->sk_rcvbuf >> net->sctp.rwnd_upd_shift),
1453 			   asoc->pathmtu)))
1454 			return true;
1455 		break;
1456 	default:
1457 		break;
1458 	}
1459 	return false;
1460 }
1461 
1462 /* Increase asoc's rwnd by len and send any window update SACK if needed. */
1463 void sctp_assoc_rwnd_increase(struct sctp_association *asoc, unsigned int len)
1464 {
1465 	struct sctp_chunk *sack;
1466 	struct timer_list *timer;
1467 
1468 	if (asoc->rwnd_over) {
1469 		if (asoc->rwnd_over >= len) {
1470 			asoc->rwnd_over -= len;
1471 		} else {
1472 			asoc->rwnd += (len - asoc->rwnd_over);
1473 			asoc->rwnd_over = 0;
1474 		}
1475 	} else {
1476 		asoc->rwnd += len;
1477 	}
1478 
1479 	/* If we had window pressure, start recovering it
1480 	 * once our rwnd had reached the accumulated pressure
1481 	 * threshold.  The idea is to recover slowly, but up
1482 	 * to the initial advertised window.
1483 	 */
1484 	if (asoc->rwnd_press) {
1485 		int change = min(asoc->pathmtu, asoc->rwnd_press);
1486 		asoc->rwnd += change;
1487 		asoc->rwnd_press -= change;
1488 	}
1489 
1490 	pr_debug("%s: asoc:%p rwnd increased by %d to (%u, %u) - %u\n",
1491 		 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
1492 		 asoc->a_rwnd);
1493 
1494 	/* Send a window update SACK if the rwnd has increased by at least the
1495 	 * minimum of the association's PMTU and half of the receive buffer.
1496 	 * The algorithm used is similar to the one described in
1497 	 * Section 4.2.3.3 of RFC 1122.
1498 	 */
1499 	if (sctp_peer_needs_update(asoc)) {
1500 		asoc->a_rwnd = asoc->rwnd;
1501 
1502 		pr_debug("%s: sending window update SACK- asoc:%p rwnd:%u "
1503 			 "a_rwnd:%u\n", __func__, asoc, asoc->rwnd,
1504 			 asoc->a_rwnd);
1505 
1506 		sack = sctp_make_sack(asoc);
1507 		if (!sack)
1508 			return;
1509 
1510 		asoc->peer.sack_needed = 0;
1511 
1512 		sctp_outq_tail(&asoc->outqueue, sack, GFP_ATOMIC);
1513 
1514 		/* Stop the SACK timer.  */
1515 		timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
1516 		if (del_timer(timer))
1517 			sctp_association_put(asoc);
1518 	}
1519 }
1520 
1521 /* Decrease asoc's rwnd by len. */
1522 void sctp_assoc_rwnd_decrease(struct sctp_association *asoc, unsigned int len)
1523 {
1524 	int rx_count;
1525 	int over = 0;
1526 
1527 	if (unlikely(!asoc->rwnd || asoc->rwnd_over))
1528 		pr_debug("%s: association:%p has asoc->rwnd:%u, "
1529 			 "asoc->rwnd_over:%u!\n", __func__, asoc,
1530 			 asoc->rwnd, asoc->rwnd_over);
1531 
1532 	if (asoc->ep->rcvbuf_policy)
1533 		rx_count = atomic_read(&asoc->rmem_alloc);
1534 	else
1535 		rx_count = atomic_read(&asoc->base.sk->sk_rmem_alloc);
1536 
1537 	/* If we've reached or overflowed our receive buffer, announce
1538 	 * a 0 rwnd if rwnd would still be positive.  Store the
1539 	 * the potential pressure overflow so that the window can be restored
1540 	 * back to original value.
1541 	 */
1542 	if (rx_count >= asoc->base.sk->sk_rcvbuf)
1543 		over = 1;
1544 
1545 	if (asoc->rwnd >= len) {
1546 		asoc->rwnd -= len;
1547 		if (over) {
1548 			asoc->rwnd_press += asoc->rwnd;
1549 			asoc->rwnd = 0;
1550 		}
1551 	} else {
1552 		asoc->rwnd_over += len - asoc->rwnd;
1553 		asoc->rwnd = 0;
1554 	}
1555 
1556 	pr_debug("%s: asoc:%p rwnd decreased by %d to (%u, %u, %u)\n",
1557 		 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
1558 		 asoc->rwnd_press);
1559 }
1560 
1561 /* Build the bind address list for the association based on info from the
1562  * local endpoint and the remote peer.
1563  */
1564 int sctp_assoc_set_bind_addr_from_ep(struct sctp_association *asoc,
1565 				     enum sctp_scope scope, gfp_t gfp)
1566 {
1567 	int flags;
1568 
1569 	/* Use scoping rules to determine the subset of addresses from
1570 	 * the endpoint.
1571 	 */
1572 	flags = (PF_INET6 == asoc->base.sk->sk_family) ? SCTP_ADDR6_ALLOWED : 0;
1573 	if (asoc->peer.ipv4_address)
1574 		flags |= SCTP_ADDR4_PEERSUPP;
1575 	if (asoc->peer.ipv6_address)
1576 		flags |= SCTP_ADDR6_PEERSUPP;
1577 
1578 	return sctp_bind_addr_copy(sock_net(asoc->base.sk),
1579 				   &asoc->base.bind_addr,
1580 				   &asoc->ep->base.bind_addr,
1581 				   scope, gfp, flags);
1582 }
1583 
1584 /* Build the association's bind address list from the cookie.  */
1585 int sctp_assoc_set_bind_addr_from_cookie(struct sctp_association *asoc,
1586 					 struct sctp_cookie *cookie,
1587 					 gfp_t gfp)
1588 {
1589 	int var_size2 = ntohs(cookie->peer_init->chunk_hdr.length);
1590 	int var_size3 = cookie->raw_addr_list_len;
1591 	__u8 *raw = (__u8 *)cookie->peer_init + var_size2;
1592 
1593 	return sctp_raw_to_bind_addrs(&asoc->base.bind_addr, raw, var_size3,
1594 				      asoc->ep->base.bind_addr.port, gfp);
1595 }
1596 
1597 /* Lookup laddr in the bind address list of an association. */
1598 int sctp_assoc_lookup_laddr(struct sctp_association *asoc,
1599 			    const union sctp_addr *laddr)
1600 {
1601 	int found = 0;
1602 
1603 	if ((asoc->base.bind_addr.port == ntohs(laddr->v4.sin_port)) &&
1604 	    sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
1605 				 sctp_sk(asoc->base.sk)))
1606 		found = 1;
1607 
1608 	return found;
1609 }
1610 
1611 /* Set an association id for a given association */
1612 int sctp_assoc_set_id(struct sctp_association *asoc, gfp_t gfp)
1613 {
1614 	bool preload = gfpflags_allow_blocking(gfp);
1615 	int ret;
1616 
1617 	/* If the id is already assigned, keep it. */
1618 	if (asoc->assoc_id)
1619 		return 0;
1620 
1621 	if (preload)
1622 		idr_preload(gfp);
1623 	spin_lock_bh(&sctp_assocs_id_lock);
1624 	/* 0 is not a valid assoc_id, must be >= 1 */
1625 	ret = idr_alloc_cyclic(&sctp_assocs_id, asoc, 1, 0, GFP_NOWAIT);
1626 	spin_unlock_bh(&sctp_assocs_id_lock);
1627 	if (preload)
1628 		idr_preload_end();
1629 	if (ret < 0)
1630 		return ret;
1631 
1632 	asoc->assoc_id = (sctp_assoc_t)ret;
1633 	return 0;
1634 }
1635 
1636 /* Free the ASCONF queue */
1637 static void sctp_assoc_free_asconf_queue(struct sctp_association *asoc)
1638 {
1639 	struct sctp_chunk *asconf;
1640 	struct sctp_chunk *tmp;
1641 
1642 	list_for_each_entry_safe(asconf, tmp, &asoc->addip_chunk_list, list) {
1643 		list_del_init(&asconf->list);
1644 		sctp_chunk_free(asconf);
1645 	}
1646 }
1647 
1648 /* Free asconf_ack cache */
1649 static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc)
1650 {
1651 	struct sctp_chunk *ack;
1652 	struct sctp_chunk *tmp;
1653 
1654 	list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
1655 				transmitted_list) {
1656 		list_del_init(&ack->transmitted_list);
1657 		sctp_chunk_free(ack);
1658 	}
1659 }
1660 
1661 /* Clean up the ASCONF_ACK queue */
1662 void sctp_assoc_clean_asconf_ack_cache(const struct sctp_association *asoc)
1663 {
1664 	struct sctp_chunk *ack;
1665 	struct sctp_chunk *tmp;
1666 
1667 	/* We can remove all the entries from the queue up to
1668 	 * the "Peer-Sequence-Number".
1669 	 */
1670 	list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
1671 				transmitted_list) {
1672 		if (ack->subh.addip_hdr->serial ==
1673 				htonl(asoc->peer.addip_serial))
1674 			break;
1675 
1676 		list_del_init(&ack->transmitted_list);
1677 		sctp_chunk_free(ack);
1678 	}
1679 }
1680 
1681 /* Find the ASCONF_ACK whose serial number matches ASCONF */
1682 struct sctp_chunk *sctp_assoc_lookup_asconf_ack(
1683 					const struct sctp_association *asoc,
1684 					__be32 serial)
1685 {
1686 	struct sctp_chunk *ack;
1687 
1688 	/* Walk through the list of cached ASCONF-ACKs and find the
1689 	 * ack chunk whose serial number matches that of the request.
1690 	 */
1691 	list_for_each_entry(ack, &asoc->asconf_ack_list, transmitted_list) {
1692 		if (sctp_chunk_pending(ack))
1693 			continue;
1694 		if (ack->subh.addip_hdr->serial == serial) {
1695 			sctp_chunk_hold(ack);
1696 			return ack;
1697 		}
1698 	}
1699 
1700 	return NULL;
1701 }
1702 
1703 void sctp_asconf_queue_teardown(struct sctp_association *asoc)
1704 {
1705 	/* Free any cached ASCONF_ACK chunk. */
1706 	sctp_assoc_free_asconf_acks(asoc);
1707 
1708 	/* Free the ASCONF queue. */
1709 	sctp_assoc_free_asconf_queue(asoc);
1710 
1711 	/* Free any cached ASCONF chunk. */
1712 	if (asoc->addip_last_asconf)
1713 		sctp_chunk_free(asoc->addip_last_asconf);
1714 }
1715