1// SPDX-License-Identifier: GPL-2.0-or-later
2/* SCTP kernel implementation
3 * (C) Copyright IBM Corp. 2001, 2004
4 * Copyright (c) 1999-2000 Cisco, Inc.
5 * Copyright (c) 1999-2001 Motorola, Inc.
6 * Copyright (c) 2001 Intel Corp.
7 * Copyright (c) 2001 La Monte H.P. Yarroll
8 *
9 * This file is part of the SCTP kernel implementation
10 *
11 * This module provides the abstraction for an SCTP association.
12 *
13 * Please send any bug reports or fixes you make to the
14 * email address(es):
15 *    lksctp developers <linux-sctp@vger.kernel.org>
16 *
17 * Written or modified by:
18 *    La Monte H.P. Yarroll <piggy@acm.org>
19 *    Karl Knutson          <karl@athena.chicago.il.us>
20 *    Jon Grimm             <jgrimm@us.ibm.com>
21 *    Xingang Guo           <xingang.guo@intel.com>
22 *    Hui Huang             <hui.huang@nokia.com>
23 *    Sridhar Samudrala	    <sri@us.ibm.com>
24 *    Daisy Chang	    <daisyc@us.ibm.com>
25 *    Ryan Layer	    <rmlayer@us.ibm.com>
26 *    Kevin Gao             <kevin.gao@intel.com>
27 */
28
29#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30
31#include <linux/types.h>
32#include <linux/fcntl.h>
33#include <linux/poll.h>
34#include <linux/init.h>
35
36#include <linux/slab.h>
37#include <linux/in.h>
38#include <net/ipv6.h>
39#include <net/sctp/sctp.h>
40#include <net/sctp/sm.h>
41
42/* Forward declarations for internal functions. */
43static void sctp_select_active_and_retran_path(struct sctp_association *asoc);
44static void sctp_assoc_bh_rcv(struct work_struct *work);
45static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc);
46static void sctp_assoc_free_asconf_queue(struct sctp_association *asoc);
47
48/* 1st Level Abstractions. */
49
50/* Initialize a new association from provided memory. */
51static struct sctp_association *sctp_association_init(
52					struct sctp_association *asoc,
53					const struct sctp_endpoint *ep,
54					const struct sock *sk,
55					enum sctp_scope scope, gfp_t gfp)
56{
57	struct sctp_sock *sp;
58	struct sctp_paramhdr *p;
59	int i;
60
61	/* Retrieve the SCTP per socket area.  */
62	sp = sctp_sk((struct sock *)sk);
63
64	/* Discarding const is appropriate here.  */
65	asoc->ep = (struct sctp_endpoint *)ep;
66	asoc->base.sk = (struct sock *)sk;
67	asoc->base.net = sock_net(sk);
68
69	sctp_endpoint_hold(asoc->ep);
70	sock_hold(asoc->base.sk);
71
72	/* Initialize the common base substructure.  */
73	asoc->base.type = SCTP_EP_TYPE_ASSOCIATION;
74
75	/* Initialize the object handling fields.  */
76	refcount_set(&asoc->base.refcnt, 1);
77
78	/* Initialize the bind addr area.  */
79	sctp_bind_addr_init(&asoc->base.bind_addr, ep->base.bind_addr.port);
80
81	asoc->state = SCTP_STATE_CLOSED;
82	asoc->cookie_life = ms_to_ktime(sp->assocparams.sasoc_cookie_life);
83	asoc->user_frag = sp->user_frag;
84
85	/* Set the association max_retrans and RTO values from the
86	 * socket values.
87	 */
88	asoc->max_retrans = sp->assocparams.sasoc_asocmaxrxt;
89	asoc->pf_retrans  = sp->pf_retrans;
90	asoc->ps_retrans  = sp->ps_retrans;
91	asoc->pf_expose   = sp->pf_expose;
92
93	asoc->rto_initial = msecs_to_jiffies(sp->rtoinfo.srto_initial);
94	asoc->rto_max = msecs_to_jiffies(sp->rtoinfo.srto_max);
95	asoc->rto_min = msecs_to_jiffies(sp->rtoinfo.srto_min);
96
97	/* Initialize the association's heartbeat interval based on the
98	 * sock configured value.
99	 */
100	asoc->hbinterval = msecs_to_jiffies(sp->hbinterval);
101	asoc->probe_interval = msecs_to_jiffies(sp->probe_interval);
102
103	asoc->encap_port = sp->encap_port;
104
105	/* Initialize path max retrans value. */
106	asoc->pathmaxrxt = sp->pathmaxrxt;
107
108	asoc->flowlabel = sp->flowlabel;
109	asoc->dscp = sp->dscp;
110
111	/* Set association default SACK delay */
112	asoc->sackdelay = msecs_to_jiffies(sp->sackdelay);
113	asoc->sackfreq = sp->sackfreq;
114
115	/* Set the association default flags controlling
116	 * Heartbeat, SACK delay, and Path MTU Discovery.
117	 */
118	asoc->param_flags = sp->param_flags;
119
120	/* Initialize the maximum number of new data packets that can be sent
121	 * in a burst.
122	 */
123	asoc->max_burst = sp->max_burst;
124
125	asoc->subscribe = sp->subscribe;
126
127	/* initialize association timers */
128	asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_COOKIE] = asoc->rto_initial;
129	asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_INIT] = asoc->rto_initial;
130	asoc->timeouts[SCTP_EVENT_TIMEOUT_T2_SHUTDOWN] = asoc->rto_initial;
131
132	/* sctpimpguide Section 2.12.2
133	 * If the 'T5-shutdown-guard' timer is used, it SHOULD be set to the
134	 * recommended value of 5 times 'RTO.Max'.
135	 */
136	asoc->timeouts[SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]
137		= 5 * asoc->rto_max;
138
139	asoc->timeouts[SCTP_EVENT_TIMEOUT_SACK] = asoc->sackdelay;
140	asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE] = 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	sctp_ulpq_init(&asoc->ulpq, asoc);
230
231	if (sctp_stream_init(&asoc->stream, asoc->c.sinit_num_ostreams, 0, gfp))
232		goto stream_free;
233
234	/* Initialize default path MTU. */
235	asoc->pathmtu = sp->pathmtu;
236	sctp_assoc_update_frag_point(asoc);
237
238	/* Assume that peer would support both address types unless we are
239	 * told otherwise.
240	 */
241	asoc->peer.ipv4_address = 1;
242	if (asoc->base.sk->sk_family == PF_INET6)
243		asoc->peer.ipv6_address = 1;
244	INIT_LIST_HEAD(&asoc->asocs);
245
246	asoc->default_stream = sp->default_stream;
247	asoc->default_ppid = sp->default_ppid;
248	asoc->default_flags = sp->default_flags;
249	asoc->default_context = sp->default_context;
250	asoc->default_timetolive = sp->default_timetolive;
251	asoc->default_rcv_context = sp->default_rcv_context;
252
253	/* AUTH related initializations */
254	INIT_LIST_HEAD(&asoc->endpoint_shared_keys);
255	if (sctp_auth_asoc_copy_shkeys(ep, asoc, gfp))
256		goto stream_free;
257
258	asoc->active_key_id = ep->active_key_id;
259	asoc->strreset_enable = ep->strreset_enable;
260
261	/* Save the hmacs and chunks list into this association */
262	if (ep->auth_hmacs_list)
263		memcpy(asoc->c.auth_hmacs, ep->auth_hmacs_list,
264			ntohs(ep->auth_hmacs_list->param_hdr.length));
265	if (ep->auth_chunk_list)
266		memcpy(asoc->c.auth_chunks, ep->auth_chunk_list,
267			ntohs(ep->auth_chunk_list->param_hdr.length));
268
269	/* Get the AUTH random number for this association */
270	p = (struct sctp_paramhdr *)asoc->c.auth_random;
271	p->type = SCTP_PARAM_RANDOM;
272	p->length = htons(sizeof(*p) + SCTP_AUTH_RANDOM_LENGTH);
273	get_random_bytes(p+1, SCTP_AUTH_RANDOM_LENGTH);
274
275	return asoc;
276
277stream_free:
278	sctp_stream_free(&asoc->stream);
279	sock_put(asoc->base.sk);
280	sctp_endpoint_put(asoc->ep);
281	return NULL;
282}
283
284/* Allocate and initialize a new association */
285struct sctp_association *sctp_association_new(const struct sctp_endpoint *ep,
286					      const struct sock *sk,
287					      enum sctp_scope scope, gfp_t gfp)
288{
289	struct sctp_association *asoc;
290
291	asoc = kzalloc(sizeof(*asoc), gfp);
292	if (!asoc)
293		goto fail;
294
295	if (!sctp_association_init(asoc, ep, sk, scope, gfp))
296		goto fail_init;
297
298	SCTP_DBG_OBJCNT_INC(assoc);
299
300	pr_debug("Created asoc %p\n", asoc);
301
302	return asoc;
303
304fail_init:
305	kfree(asoc);
306fail:
307	return NULL;
308}
309
310/* Free this association if possible.  There may still be users, so
311 * the actual deallocation may be delayed.
312 */
313void sctp_association_free(struct sctp_association *asoc)
314{
315	struct sock *sk = asoc->base.sk;
316	struct sctp_transport *transport;
317	struct list_head *pos, *temp;
318	int i;
319
320	/* Only real associations count against the endpoint, so
321	 * don't bother for if this is a temporary association.
322	 */
323	if (!list_empty(&asoc->asocs)) {
324		list_del(&asoc->asocs);
325
326		/* Decrement the backlog value for a TCP-style listening
327		 * socket.
328		 */
329		if (sctp_style(sk, TCP) && sctp_sstate(sk, LISTENING))
330			sk_acceptq_removed(sk);
331	}
332
333	/* Mark as dead, so other users can know this structure is
334	 * going away.
335	 */
336	asoc->base.dead = true;
337
338	/* Dispose of any data lying around in the outqueue. */
339	sctp_outq_free(&asoc->outqueue);
340
341	/* Dispose of any pending messages for the upper layer. */
342	sctp_ulpq_free(&asoc->ulpq);
343
344	/* Dispose of any pending chunks on the inqueue. */
345	sctp_inq_free(&asoc->base.inqueue);
346
347	sctp_tsnmap_free(&asoc->peer.tsn_map);
348
349	/* Free stream information. */
350	sctp_stream_free(&asoc->stream);
351
352	if (asoc->strreset_chunk)
353		sctp_chunk_free(asoc->strreset_chunk);
354
355	/* Clean up the bound address list. */
356	sctp_bind_addr_free(&asoc->base.bind_addr);
357
358	/* Do we need to go through all of our timers and
359	 * delete them?   To be safe we will try to delete all, but we
360	 * should be able to go through and make a guess based
361	 * on our state.
362	 */
363	for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i) {
364		if (del_timer(&asoc->timers[i]))
365			sctp_association_put(asoc);
366	}
367
368	/* Free peer's cached cookie. */
369	kfree(asoc->peer.cookie);
370	kfree(asoc->peer.peer_random);
371	kfree(asoc->peer.peer_chunks);
372	kfree(asoc->peer.peer_hmacs);
373
374	/* Release the transport structures. */
375	list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
376		transport = list_entry(pos, struct sctp_transport, transports);
377		list_del_rcu(pos);
378		sctp_unhash_transport(transport);
379		sctp_transport_free(transport);
380	}
381
382	asoc->peer.transport_count = 0;
383
384	sctp_asconf_queue_teardown(asoc);
385
386	/* Free pending address space being deleted */
387	kfree(asoc->asconf_addr_del_pending);
388
389	/* AUTH - Free the endpoint shared keys */
390	sctp_auth_destroy_keys(&asoc->endpoint_shared_keys);
391
392	/* AUTH - Free the association shared key */
393	sctp_auth_key_put(asoc->asoc_shared_key);
394
395	sctp_association_put(asoc);
396}
397
398/* Cleanup and free up an association. */
399static void sctp_association_destroy(struct sctp_association *asoc)
400{
401	if (unlikely(!asoc->base.dead)) {
402		WARN(1, "Attempt to destroy undead association %p!\n", asoc);
403		return;
404	}
405
406	sctp_endpoint_put(asoc->ep);
407	sock_put(asoc->base.sk);
408
409	if (asoc->assoc_id != 0) {
410		spin_lock_bh(&sctp_assocs_id_lock);
411		idr_remove(&sctp_assocs_id, asoc->assoc_id);
412		spin_unlock_bh(&sctp_assocs_id_lock);
413	}
414
415	WARN_ON(atomic_read(&asoc->rmem_alloc));
416
417	kfree_rcu(asoc, rcu);
418	SCTP_DBG_OBJCNT_DEC(assoc);
419}
420
421/* Change the primary destination address for the peer. */
422void sctp_assoc_set_primary(struct sctp_association *asoc,
423			    struct sctp_transport *transport)
424{
425	int changeover = 0;
426
427	/* it's a changeover only if we already have a primary path
428	 * that we are changing
429	 */
430	if (asoc->peer.primary_path != NULL &&
431	    asoc->peer.primary_path != transport)
432		changeover = 1 ;
433
434	asoc->peer.primary_path = transport;
435	sctp_ulpevent_notify_peer_addr_change(transport,
436					      SCTP_ADDR_MADE_PRIM, 0);
437
438	/* Set a default msg_name for events. */
439	memcpy(&asoc->peer.primary_addr, &transport->ipaddr,
440	       sizeof(union sctp_addr));
441
442	/* If the primary path is changing, assume that the
443	 * user wants to use this new path.
444	 */
445	if ((transport->state == SCTP_ACTIVE) ||
446	    (transport->state == SCTP_UNKNOWN))
447		asoc->peer.active_path = transport;
448
449	/*
450	 * SFR-CACC algorithm:
451	 * Upon the receipt of a request to change the primary
452	 * destination address, on the data structure for the new
453	 * primary destination, the sender MUST do the following:
454	 *
455	 * 1) If CHANGEOVER_ACTIVE is set, then there was a switch
456	 * to this destination address earlier. The sender MUST set
457	 * CYCLING_CHANGEOVER to indicate that this switch is a
458	 * double switch to the same destination address.
459	 *
460	 * Really, only bother is we have data queued or outstanding on
461	 * the association.
462	 */
463	if (!asoc->outqueue.outstanding_bytes && !asoc->outqueue.out_qlen)
464		return;
465
466	if (transport->cacc.changeover_active)
467		transport->cacc.cycling_changeover = changeover;
468
469	/* 2) The sender MUST set CHANGEOVER_ACTIVE to indicate that
470	 * a changeover has occurred.
471	 */
472	transport->cacc.changeover_active = changeover;
473
474	/* 3) The sender MUST store the next TSN to be sent in
475	 * next_tsn_at_change.
476	 */
477	transport->cacc.next_tsn_at_change = asoc->next_tsn;
478}
479
480/* Remove a transport from an association.  */
481void sctp_assoc_rm_peer(struct sctp_association *asoc,
482			struct sctp_transport *peer)
483{
484	struct sctp_transport *transport;
485	struct list_head *pos;
486	struct sctp_chunk *ch;
487
488	pr_debug("%s: association:%p addr:%pISpc\n",
489		 __func__, asoc, &peer->ipaddr.sa);
490
491	/* If we are to remove the current retran_path, update it
492	 * to the next peer before removing this peer from the list.
493	 */
494	if (asoc->peer.retran_path == peer)
495		sctp_assoc_update_retran_path(asoc);
496
497	/* Remove this peer from the list. */
498	list_del_rcu(&peer->transports);
499	/* Remove this peer from the transport hashtable */
500	sctp_unhash_transport(peer);
501
502	/* Get the first transport of asoc. */
503	pos = asoc->peer.transport_addr_list.next;
504	transport = list_entry(pos, struct sctp_transport, transports);
505
506	/* Update any entries that match the peer to be deleted. */
507	if (asoc->peer.primary_path == peer)
508		sctp_assoc_set_primary(asoc, transport);
509	if (asoc->peer.active_path == peer)
510		asoc->peer.active_path = transport;
511	if (asoc->peer.retran_path == peer)
512		asoc->peer.retran_path = transport;
513	if (asoc->peer.last_data_from == peer)
514		asoc->peer.last_data_from = transport;
515
516	if (asoc->strreset_chunk &&
517	    asoc->strreset_chunk->transport == peer) {
518		asoc->strreset_chunk->transport = transport;
519		sctp_transport_reset_reconf_timer(transport);
520	}
521
522	/* If we remove the transport an INIT was last sent to, set it to
523	 * NULL. Combined with the update of the retran path above, this
524	 * will cause the next INIT to be sent to the next available
525	 * transport, maintaining the cycle.
526	 */
527	if (asoc->init_last_sent_to == peer)
528		asoc->init_last_sent_to = NULL;
529
530	/* If we remove the transport an SHUTDOWN was last sent to, set it
531	 * to NULL. Combined with the update of the retran path above, this
532	 * will cause the next SHUTDOWN to be sent to the next available
533	 * transport, maintaining the cycle.
534	 */
535	if (asoc->shutdown_last_sent_to == peer)
536		asoc->shutdown_last_sent_to = NULL;
537
538	/* If we remove the transport an ASCONF was last sent to, set it to
539	 * NULL.
540	 */
541	if (asoc->addip_last_asconf &&
542	    asoc->addip_last_asconf->transport == peer)
543		asoc->addip_last_asconf->transport = NULL;
544
545	/* If we have something on the transmitted list, we have to
546	 * save it off.  The best place is the active path.
547	 */
548	if (!list_empty(&peer->transmitted)) {
549		struct sctp_transport *active = asoc->peer.active_path;
550
551		/* Reset the transport of each chunk on this list */
552		list_for_each_entry(ch, &peer->transmitted,
553					transmitted_list) {
554			ch->transport = NULL;
555			ch->rtt_in_progress = 0;
556		}
557
558		list_splice_tail_init(&peer->transmitted,
559					&active->transmitted);
560
561		/* Start a T3 timer here in case it wasn't running so
562		 * that these migrated packets have a chance to get
563		 * retransmitted.
564		 */
565		if (!timer_pending(&active->T3_rtx_timer))
566			if (!mod_timer(&active->T3_rtx_timer,
567					jiffies + active->rto))
568				sctp_transport_hold(active);
569	}
570
571	list_for_each_entry(ch, &asoc->outqueue.out_chunk_list, list)
572		if (ch->transport == peer)
573			ch->transport = NULL;
574
575	asoc->peer.transport_count--;
576
577	sctp_ulpevent_notify_peer_addr_change(peer, SCTP_ADDR_REMOVED, 0);
578	sctp_transport_free(peer);
579}
580
581/* Add a transport address to an association.  */
582struct sctp_transport *sctp_assoc_add_peer(struct sctp_association *asoc,
583					   const union sctp_addr *addr,
584					   const gfp_t gfp,
585					   const int peer_state)
586{
587	struct sctp_transport *peer;
588	struct sctp_sock *sp;
589	unsigned short port;
590
591	sp = sctp_sk(asoc->base.sk);
592
593	/* AF_INET and AF_INET6 share common port field. */
594	port = ntohs(addr->v4.sin_port);
595
596	pr_debug("%s: association:%p addr:%pISpc state:%d\n", __func__,
597		 asoc, &addr->sa, peer_state);
598
599	/* Set the port if it has not been set yet.  */
600	if (0 == asoc->peer.port)
601		asoc->peer.port = port;
602
603	/* Check to see if this is a duplicate. */
604	peer = sctp_assoc_lookup_paddr(asoc, addr);
605	if (peer) {
606		/* An UNKNOWN state is only set on transports added by
607		 * user in sctp_connectx() call.  Such transports should be
608		 * considered CONFIRMED per RFC 4960, Section 5.4.
609		 */
610		if (peer->state == SCTP_UNKNOWN) {
611			peer->state = SCTP_ACTIVE;
612		}
613		return peer;
614	}
615
616	peer = sctp_transport_new(asoc->base.net, addr, gfp);
617	if (!peer)
618		return NULL;
619
620	sctp_transport_set_owner(peer, asoc);
621
622	/* Initialize the peer's heartbeat interval based on the
623	 * association configured value.
624	 */
625	peer->hbinterval = asoc->hbinterval;
626	peer->probe_interval = asoc->probe_interval;
627
628	peer->encap_port = asoc->encap_port;
629
630	/* Set the path max_retrans.  */
631	peer->pathmaxrxt = asoc->pathmaxrxt;
632
633	/* And the partial failure retrans threshold */
634	peer->pf_retrans = asoc->pf_retrans;
635	/* And the primary path switchover retrans threshold */
636	peer->ps_retrans = asoc->ps_retrans;
637
638	/* Initialize the peer's SACK delay timeout based on the
639	 * association configured value.
640	 */
641	peer->sackdelay = asoc->sackdelay;
642	peer->sackfreq = asoc->sackfreq;
643
644	if (addr->sa.sa_family == AF_INET6) {
645		__be32 info = addr->v6.sin6_flowinfo;
646
647		if (info) {
648			peer->flowlabel = ntohl(info & IPV6_FLOWLABEL_MASK);
649			peer->flowlabel |= SCTP_FLOWLABEL_SET_MASK;
650		} else {
651			peer->flowlabel = asoc->flowlabel;
652		}
653	}
654	peer->dscp = asoc->dscp;
655
656	/* Enable/disable heartbeat, SACK delay, and path MTU discovery
657	 * based on association setting.
658	 */
659	peer->param_flags = asoc->param_flags;
660
661	/* Initialize the pmtu of the transport. */
662	sctp_transport_route(peer, NULL, sp);
663
664	/* If this is the first transport addr on this association,
665	 * initialize the association PMTU to the peer's PMTU.
666	 * If not and the current association PMTU is higher than the new
667	 * peer's PMTU, reset the association PMTU to the new peer's PMTU.
668	 */
669	sctp_assoc_set_pmtu(asoc, asoc->pathmtu ?
670				  min_t(int, peer->pathmtu, asoc->pathmtu) :
671				  peer->pathmtu);
672
673	peer->pmtu_pending = 0;
674
675	/* The asoc->peer.port might not be meaningful yet, but
676	 * initialize the packet structure anyway.
677	 */
678	sctp_packet_init(&peer->packet, peer, asoc->base.bind_addr.port,
679			 asoc->peer.port);
680
681	/* 7.2.1 Slow-Start
682	 *
683	 * o The initial cwnd before DATA transmission or after a sufficiently
684	 *   long idle period MUST be set to
685	 *      min(4*MTU, max(2*MTU, 4380 bytes))
686	 *
687	 * o The initial value of ssthresh MAY be arbitrarily high
688	 *   (for example, implementations MAY use the size of the
689	 *   receiver advertised window).
690	 */
691	peer->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
692
693	/* At this point, we may not have the receiver's advertised window,
694	 * so initialize ssthresh to the default value and it will be set
695	 * later when we process the INIT.
696	 */
697	peer->ssthresh = SCTP_DEFAULT_MAXWINDOW;
698
699	peer->partial_bytes_acked = 0;
700	peer->flight_size = 0;
701	peer->burst_limited = 0;
702
703	/* Set the transport's RTO.initial value */
704	peer->rto = asoc->rto_initial;
705	sctp_max_rto(asoc, peer);
706
707	/* Set the peer's active state. */
708	peer->state = peer_state;
709
710	/* Add this peer into the transport hashtable */
711	if (sctp_hash_transport(peer)) {
712		sctp_transport_free(peer);
713		return NULL;
714	}
715
716	sctp_transport_pl_reset(peer);
717
718	/* Attach the remote transport to our asoc.  */
719	list_add_tail_rcu(&peer->transports, &asoc->peer.transport_addr_list);
720	asoc->peer.transport_count++;
721
722	sctp_ulpevent_notify_peer_addr_change(peer, SCTP_ADDR_ADDED, 0);
723
724	/* If we do not yet have a primary path, set one.  */
725	if (!asoc->peer.primary_path) {
726		sctp_assoc_set_primary(asoc, peer);
727		asoc->peer.retran_path = peer;
728	}
729
730	if (asoc->peer.active_path == asoc->peer.retran_path &&
731	    peer->state != SCTP_UNCONFIRMED) {
732		asoc->peer.retran_path = peer;
733	}
734
735	return peer;
736}
737
738/* Delete a transport address from an association.  */
739void sctp_assoc_del_peer(struct sctp_association *asoc,
740			 const union sctp_addr *addr)
741{
742	struct list_head	*pos;
743	struct list_head	*temp;
744	struct sctp_transport	*transport;
745
746	list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
747		transport = list_entry(pos, struct sctp_transport, transports);
748		if (sctp_cmp_addr_exact(addr, &transport->ipaddr)) {
749			/* Do book keeping for removing the peer and free it. */
750			sctp_assoc_rm_peer(asoc, transport);
751			break;
752		}
753	}
754}
755
756/* Lookup a transport by address. */
757struct sctp_transport *sctp_assoc_lookup_paddr(
758					const struct sctp_association *asoc,
759					const union sctp_addr *address)
760{
761	struct sctp_transport *t;
762
763	/* Cycle through all transports searching for a peer address. */
764
765	list_for_each_entry(t, &asoc->peer.transport_addr_list,
766			transports) {
767		if (sctp_cmp_addr_exact(address, &t->ipaddr))
768			return t;
769	}
770
771	return NULL;
772}
773
774/* Remove all transports except a give one */
775void sctp_assoc_del_nonprimary_peers(struct sctp_association *asoc,
776				     struct sctp_transport *primary)
777{
778	struct sctp_transport	*temp;
779	struct sctp_transport	*t;
780
781	list_for_each_entry_safe(t, temp, &asoc->peer.transport_addr_list,
782				 transports) {
783		/* if the current transport is not the primary one, delete it */
784		if (t != primary)
785			sctp_assoc_rm_peer(asoc, t);
786	}
787}
788
789/* Engage in transport control operations.
790 * Mark the transport up or down and send a notification to the user.
791 * Select and update the new active and retran paths.
792 */
793void sctp_assoc_control_transport(struct sctp_association *asoc,
794				  struct sctp_transport *transport,
795				  enum sctp_transport_cmd command,
796				  sctp_sn_error_t error)
797{
798	int spc_state = SCTP_ADDR_AVAILABLE;
799	bool ulp_notify = true;
800
801	/* Record the transition on the transport.  */
802	switch (command) {
803	case SCTP_TRANSPORT_UP:
804		/* If we are moving from UNCONFIRMED state due
805		 * to heartbeat success, report the SCTP_ADDR_CONFIRMED
806		 * state to the user, otherwise report SCTP_ADDR_AVAILABLE.
807		 */
808		if (transport->state == SCTP_PF &&
809		    asoc->pf_expose != SCTP_PF_EXPOSE_ENABLE)
810			ulp_notify = false;
811		else if (transport->state == SCTP_UNCONFIRMED &&
812			 error == SCTP_HEARTBEAT_SUCCESS)
813			spc_state = SCTP_ADDR_CONFIRMED;
814
815		transport->state = SCTP_ACTIVE;
816		sctp_transport_pl_reset(transport);
817		break;
818
819	case SCTP_TRANSPORT_DOWN:
820		/* If the transport was never confirmed, do not transition it
821		 * to inactive state.  Also, release the cached route since
822		 * there may be a better route next time.
823		 */
824		if (transport->state != SCTP_UNCONFIRMED) {
825			transport->state = SCTP_INACTIVE;
826			sctp_transport_pl_reset(transport);
827			spc_state = SCTP_ADDR_UNREACHABLE;
828		} else {
829			sctp_transport_dst_release(transport);
830			ulp_notify = false;
831		}
832		break;
833
834	case SCTP_TRANSPORT_PF:
835		transport->state = SCTP_PF;
836		if (asoc->pf_expose != SCTP_PF_EXPOSE_ENABLE)
837			ulp_notify = false;
838		else
839			spc_state = SCTP_ADDR_POTENTIALLY_FAILED;
840		break;
841
842	default:
843		return;
844	}
845
846	/* Generate and send a SCTP_PEER_ADDR_CHANGE notification
847	 * to the user.
848	 */
849	if (ulp_notify)
850		sctp_ulpevent_notify_peer_addr_change(transport,
851						      spc_state, error);
852
853	/* Select new active and retran paths. */
854	sctp_select_active_and_retran_path(asoc);
855}
856
857/* Hold a reference to an association. */
858void sctp_association_hold(struct sctp_association *asoc)
859{
860	refcount_inc(&asoc->base.refcnt);
861}
862
863/* Release a reference to an association and cleanup
864 * if there are no more references.
865 */
866void sctp_association_put(struct sctp_association *asoc)
867{
868	if (refcount_dec_and_test(&asoc->base.refcnt))
869		sctp_association_destroy(asoc);
870}
871
872/* Allocate the next TSN, Transmission Sequence Number, for the given
873 * association.
874 */
875__u32 sctp_association_get_next_tsn(struct sctp_association *asoc)
876{
877	/* From Section 1.6 Serial Number Arithmetic:
878	 * Transmission Sequence Numbers wrap around when they reach
879	 * 2**32 - 1.  That is, the next TSN a DATA chunk MUST use
880	 * after transmitting TSN = 2*32 - 1 is TSN = 0.
881	 */
882	__u32 retval = asoc->next_tsn;
883	asoc->next_tsn++;
884	asoc->unack_data++;
885
886	return retval;
887}
888
889/* Compare two addresses to see if they match.  Wildcard addresses
890 * only match themselves.
891 */
892int sctp_cmp_addr_exact(const union sctp_addr *ss1,
893			const union sctp_addr *ss2)
894{
895	struct sctp_af *af;
896
897	af = sctp_get_af_specific(ss1->sa.sa_family);
898	if (unlikely(!af))
899		return 0;
900
901	return af->cmp_addr(ss1, ss2);
902}
903
904/* Return an ecne chunk to get prepended to a packet.
905 * Note:  We are sly and return a shared, prealloced chunk.  FIXME:
906 * No we don't, but we could/should.
907 */
908struct sctp_chunk *sctp_get_ecne_prepend(struct sctp_association *asoc)
909{
910	if (!asoc->need_ecne)
911		return NULL;
912
913	/* Send ECNE if needed.
914	 * Not being able to allocate a chunk here is not deadly.
915	 */
916	return sctp_make_ecne(asoc, asoc->last_ecne_tsn);
917}
918
919/*
920 * Find which transport this TSN was sent on.
921 */
922struct sctp_transport *sctp_assoc_lookup_tsn(struct sctp_association *asoc,
923					     __u32 tsn)
924{
925	struct sctp_transport *active;
926	struct sctp_transport *match;
927	struct sctp_transport *transport;
928	struct sctp_chunk *chunk;
929	__be32 key = htonl(tsn);
930
931	match = NULL;
932
933	/*
934	 * FIXME: In general, find a more efficient data structure for
935	 * searching.
936	 */
937
938	/*
939	 * The general strategy is to search each transport's transmitted
940	 * list.   Return which transport this TSN lives on.
941	 *
942	 * Let's be hopeful and check the active_path first.
943	 * Another optimization would be to know if there is only one
944	 * outbound path and not have to look for the TSN at all.
945	 *
946	 */
947
948	active = asoc->peer.active_path;
949
950	list_for_each_entry(chunk, &active->transmitted,
951			transmitted_list) {
952
953		if (key == chunk->subh.data_hdr->tsn) {
954			match = active;
955			goto out;
956		}
957	}
958
959	/* If not found, go search all the other transports. */
960	list_for_each_entry(transport, &asoc->peer.transport_addr_list,
961			transports) {
962
963		if (transport == active)
964			continue;
965		list_for_each_entry(chunk, &transport->transmitted,
966				transmitted_list) {
967			if (key == chunk->subh.data_hdr->tsn) {
968				match = transport;
969				goto out;
970			}
971		}
972	}
973out:
974	return match;
975}
976
977/* Do delayed input processing.  This is scheduled by sctp_rcv(). */
978static void sctp_assoc_bh_rcv(struct work_struct *work)
979{
980	struct sctp_association *asoc =
981		container_of(work, struct sctp_association,
982			     base.inqueue.immediate);
983	struct net *net = asoc->base.net;
984	union sctp_subtype subtype;
985	struct sctp_endpoint *ep;
986	struct sctp_chunk *chunk;
987	struct sctp_inq *inqueue;
988	int first_time = 1;	/* is this the first time through the loop */
989	int error = 0;
990	int state;
991
992	/* The association should be held so we should be safe. */
993	ep = asoc->ep;
994
995	inqueue = &asoc->base.inqueue;
996	sctp_association_hold(asoc);
997	while (NULL != (chunk = sctp_inq_pop(inqueue))) {
998		state = asoc->state;
999		subtype = SCTP_ST_CHUNK(chunk->chunk_hdr->type);
1000
1001		/* If the first chunk in the packet is AUTH, do special
1002		 * processing specified in Section 6.3 of SCTP-AUTH spec
1003		 */
1004		if (first_time && subtype.chunk == SCTP_CID_AUTH) {
1005			struct sctp_chunkhdr *next_hdr;
1006
1007			next_hdr = sctp_inq_peek(inqueue);
1008			if (!next_hdr)
1009				goto normal;
1010
1011			/* If the next chunk is COOKIE-ECHO, skip the AUTH
1012			 * chunk while saving a pointer to it so we can do
1013			 * Authentication later (during cookie-echo
1014			 * processing).
1015			 */
1016			if (next_hdr->type == SCTP_CID_COOKIE_ECHO) {
1017				chunk->auth_chunk = skb_clone(chunk->skb,
1018							      GFP_ATOMIC);
1019				chunk->auth = 1;
1020				continue;
1021			}
1022		}
1023
1024normal:
1025		/* SCTP-AUTH, Section 6.3:
1026		 *    The receiver has a list of chunk types which it expects
1027		 *    to be received only after an AUTH-chunk.  This list has
1028		 *    been sent to the peer during the association setup.  It
1029		 *    MUST silently discard these chunks if they are not placed
1030		 *    after an AUTH chunk in the packet.
1031		 */
1032		if (sctp_auth_recv_cid(subtype.chunk, asoc) && !chunk->auth)
1033			continue;
1034
1035		/* Remember where the last DATA chunk came from so we
1036		 * know where to send the SACK.
1037		 */
1038		if (sctp_chunk_is_data(chunk))
1039			asoc->peer.last_data_from = chunk->transport;
1040		else {
1041			SCTP_INC_STATS(net, SCTP_MIB_INCTRLCHUNKS);
1042			asoc->stats.ictrlchunks++;
1043			if (chunk->chunk_hdr->type == SCTP_CID_SACK)
1044				asoc->stats.isacks++;
1045		}
1046
1047		if (chunk->transport)
1048			chunk->transport->last_time_heard = ktime_get();
1049
1050		/* Run through the state machine. */
1051		error = sctp_do_sm(net, SCTP_EVENT_T_CHUNK, subtype,
1052				   state, ep, asoc, chunk, GFP_ATOMIC);
1053
1054		/* Check to see if the association is freed in response to
1055		 * the incoming chunk.  If so, get out of the while loop.
1056		 */
1057		if (asoc->base.dead)
1058			break;
1059
1060		/* If there is an error on chunk, discard this packet. */
1061		if (error && chunk)
1062			chunk->pdiscard = 1;
1063
1064		if (first_time)
1065			first_time = 0;
1066	}
1067	sctp_association_put(asoc);
1068}
1069
1070/* This routine moves an association from its old sk to a new sk.  */
1071void sctp_assoc_migrate(struct sctp_association *assoc, struct sock *newsk)
1072{
1073	struct sctp_sock *newsp = sctp_sk(newsk);
1074	struct sock *oldsk = assoc->base.sk;
1075
1076	/* Delete the association from the old endpoint's list of
1077	 * associations.
1078	 */
1079	list_del_init(&assoc->asocs);
1080
1081	/* Decrement the backlog value for a TCP-style socket. */
1082	if (sctp_style(oldsk, TCP))
1083		sk_acceptq_removed(oldsk);
1084
1085	/* Release references to the old endpoint and the sock.  */
1086	sctp_endpoint_put(assoc->ep);
1087	sock_put(assoc->base.sk);
1088
1089	/* Get a reference to the new endpoint.  */
1090	assoc->ep = newsp->ep;
1091	sctp_endpoint_hold(assoc->ep);
1092
1093	/* Get a reference to the new sock.  */
1094	assoc->base.sk = newsk;
1095	sock_hold(assoc->base.sk);
1096
1097	/* Add the association to the new endpoint's list of associations.  */
1098	sctp_endpoint_add_asoc(newsp->ep, assoc);
1099}
1100
1101/* Update an association (possibly from unexpected COOKIE-ECHO processing).  */
1102int sctp_assoc_update(struct sctp_association *asoc,
1103		      struct sctp_association *new)
1104{
1105	struct sctp_transport *trans;
1106	struct list_head *pos, *temp;
1107
1108	/* Copy in new parameters of peer. */
1109	asoc->c = new->c;
1110	asoc->peer.rwnd = new->peer.rwnd;
1111	asoc->peer.sack_needed = new->peer.sack_needed;
1112	asoc->peer.auth_capable = new->peer.auth_capable;
1113	asoc->peer.i = new->peer.i;
1114
1115	if (!sctp_tsnmap_init(&asoc->peer.tsn_map, SCTP_TSN_MAP_INITIAL,
1116			      asoc->peer.i.initial_tsn, GFP_ATOMIC))
1117		return -ENOMEM;
1118
1119	/* Remove any peer addresses not present in the new association. */
1120	list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
1121		trans = list_entry(pos, struct sctp_transport, transports);
1122		if (!sctp_assoc_lookup_paddr(new, &trans->ipaddr)) {
1123			sctp_assoc_rm_peer(asoc, trans);
1124			continue;
1125		}
1126
1127		if (asoc->state >= SCTP_STATE_ESTABLISHED)
1128			sctp_transport_reset(trans);
1129	}
1130
1131	/* If the case is A (association restart), use
1132	 * initial_tsn as next_tsn. If the case is B, use
1133	 * current next_tsn in case data sent to peer
1134	 * has been discarded and needs retransmission.
1135	 */
1136	if (asoc->state >= SCTP_STATE_ESTABLISHED) {
1137		asoc->next_tsn = new->next_tsn;
1138		asoc->ctsn_ack_point = new->ctsn_ack_point;
1139		asoc->adv_peer_ack_point = new->adv_peer_ack_point;
1140
1141		/* Reinitialize SSN for both local streams
1142		 * and peer's streams.
1143		 */
1144		sctp_stream_clear(&asoc->stream);
1145
1146		/* Flush the ULP reassembly and ordered queue.
1147		 * Any data there will now be stale and will
1148		 * cause problems.
1149		 */
1150		sctp_ulpq_flush(&asoc->ulpq);
1151
1152		/* reset the overall association error count so
1153		 * that the restarted association doesn't get torn
1154		 * down on the next retransmission timer.
1155		 */
1156		asoc->overall_error_count = 0;
1157
1158	} else {
1159		/* Add any peer addresses from the new association. */
1160		list_for_each_entry(trans, &new->peer.transport_addr_list,
1161				    transports)
1162			if (!sctp_assoc_add_peer(asoc, &trans->ipaddr,
1163						 GFP_ATOMIC, trans->state))
1164				return -ENOMEM;
1165
1166		asoc->ctsn_ack_point = asoc->next_tsn - 1;
1167		asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
1168
1169		if (sctp_state(asoc, COOKIE_WAIT))
1170			sctp_stream_update(&asoc->stream, &new->stream);
1171
1172		/* get a new assoc id if we don't have one yet. */
1173		if (sctp_assoc_set_id(asoc, GFP_ATOMIC))
1174			return -ENOMEM;
1175	}
1176
1177	/* SCTP-AUTH: Save the peer parameters from the new associations
1178	 * and also move the association shared keys over
1179	 */
1180	kfree(asoc->peer.peer_random);
1181	asoc->peer.peer_random = new->peer.peer_random;
1182	new->peer.peer_random = NULL;
1183
1184	kfree(asoc->peer.peer_chunks);
1185	asoc->peer.peer_chunks = new->peer.peer_chunks;
1186	new->peer.peer_chunks = NULL;
1187
1188	kfree(asoc->peer.peer_hmacs);
1189	asoc->peer.peer_hmacs = new->peer.peer_hmacs;
1190	new->peer.peer_hmacs = NULL;
1191
1192	return sctp_auth_asoc_init_active_key(asoc, GFP_ATOMIC);
1193}
1194
1195/* Update the retran path for sending a retransmitted packet.
1196 * See also RFC4960, 6.4. Multi-Homed SCTP Endpoints:
1197 *
1198 *   When there is outbound data to send and the primary path
1199 *   becomes inactive (e.g., due to failures), or where the
1200 *   SCTP user explicitly requests to send data to an
1201 *   inactive destination transport address, before reporting
1202 *   an error to its ULP, the SCTP endpoint should try to send
1203 *   the data to an alternate active destination transport
1204 *   address if one exists.
1205 *
1206 *   When retransmitting data that timed out, if the endpoint
1207 *   is multihomed, it should consider each source-destination
1208 *   address pair in its retransmission selection policy.
1209 *   When retransmitting timed-out data, the endpoint should
1210 *   attempt to pick the most divergent source-destination
1211 *   pair from the original source-destination pair to which
1212 *   the packet was transmitted.
1213 *
1214 *   Note: Rules for picking the most divergent source-destination
1215 *   pair are an implementation decision and are not specified
1216 *   within this document.
1217 *
1218 * Our basic strategy is to round-robin transports in priorities
1219 * according to sctp_trans_score() e.g., if no such
1220 * transport with state SCTP_ACTIVE exists, round-robin through
1221 * SCTP_UNKNOWN, etc. You get the picture.
1222 */
1223static u8 sctp_trans_score(const struct sctp_transport *trans)
1224{
1225	switch (trans->state) {
1226	case SCTP_ACTIVE:
1227		return 3;	/* best case */
1228	case SCTP_UNKNOWN:
1229		return 2;
1230	case SCTP_PF:
1231		return 1;
1232	default: /* case SCTP_INACTIVE */
1233		return 0;	/* worst case */
1234	}
1235}
1236
1237static struct sctp_transport *sctp_trans_elect_tie(struct sctp_transport *trans1,
1238						   struct sctp_transport *trans2)
1239{
1240	if (trans1->error_count > trans2->error_count) {
1241		return trans2;
1242	} else if (trans1->error_count == trans2->error_count &&
1243		   ktime_after(trans2->last_time_heard,
1244			       trans1->last_time_heard)) {
1245		return trans2;
1246	} else {
1247		return trans1;
1248	}
1249}
1250
1251static struct sctp_transport *sctp_trans_elect_best(struct sctp_transport *curr,
1252						    struct sctp_transport *best)
1253{
1254	u8 score_curr, score_best;
1255
1256	if (best == NULL || curr == best)
1257		return curr;
1258
1259	score_curr = sctp_trans_score(curr);
1260	score_best = sctp_trans_score(best);
1261
1262	/* First, try a score-based selection if both transport states
1263	 * differ. If we're in a tie, lets try to make a more clever
1264	 * decision here based on error counts and last time heard.
1265	 */
1266	if (score_curr > score_best)
1267		return curr;
1268	else if (score_curr == score_best)
1269		return sctp_trans_elect_tie(best, curr);
1270	else
1271		return best;
1272}
1273
1274void sctp_assoc_update_retran_path(struct sctp_association *asoc)
1275{
1276	struct sctp_transport *trans = asoc->peer.retran_path;
1277	struct sctp_transport *trans_next = NULL;
1278
1279	/* We're done as we only have the one and only path. */
1280	if (asoc->peer.transport_count == 1)
1281		return;
1282	/* If active_path and retran_path are the same and active,
1283	 * then this is the only active path. Use it.
1284	 */
1285	if (asoc->peer.active_path == asoc->peer.retran_path &&
1286	    asoc->peer.active_path->state == SCTP_ACTIVE)
1287		return;
1288
1289	/* Iterate from retran_path's successor back to retran_path. */
1290	for (trans = list_next_entry(trans, transports); 1;
1291	     trans = list_next_entry(trans, transports)) {
1292		/* Manually skip the head element. */
1293		if (&trans->transports == &asoc->peer.transport_addr_list)
1294			continue;
1295		if (trans->state == SCTP_UNCONFIRMED)
1296			continue;
1297		trans_next = sctp_trans_elect_best(trans, trans_next);
1298		/* Active is good enough for immediate return. */
1299		if (trans_next->state == SCTP_ACTIVE)
1300			break;
1301		/* We've reached the end, time to update path. */
1302		if (trans == asoc->peer.retran_path)
1303			break;
1304	}
1305
1306	asoc->peer.retran_path = trans_next;
1307
1308	pr_debug("%s: association:%p updated new path to addr:%pISpc\n",
1309		 __func__, asoc, &asoc->peer.retran_path->ipaddr.sa);
1310}
1311
1312static void sctp_select_active_and_retran_path(struct sctp_association *asoc)
1313{
1314	struct sctp_transport *trans, *trans_pri = NULL, *trans_sec = NULL;
1315	struct sctp_transport *trans_pf = NULL;
1316
1317	/* Look for the two most recently used active transports. */
1318	list_for_each_entry(trans, &asoc->peer.transport_addr_list,
1319			    transports) {
1320		/* Skip uninteresting transports. */
1321		if (trans->state == SCTP_INACTIVE ||
1322		    trans->state == SCTP_UNCONFIRMED)
1323			continue;
1324		/* Keep track of the best PF transport from our
1325		 * list in case we don't find an active one.
1326		 */
1327		if (trans->state == SCTP_PF) {
1328			trans_pf = sctp_trans_elect_best(trans, trans_pf);
1329			continue;
1330		}
1331		/* For active transports, pick the most recent ones. */
1332		if (trans_pri == NULL ||
1333		    ktime_after(trans->last_time_heard,
1334				trans_pri->last_time_heard)) {
1335			trans_sec = trans_pri;
1336			trans_pri = trans;
1337		} else if (trans_sec == NULL ||
1338			   ktime_after(trans->last_time_heard,
1339				       trans_sec->last_time_heard)) {
1340			trans_sec = trans;
1341		}
1342	}
1343
1344	/* RFC 2960 6.4 Multi-Homed SCTP Endpoints
1345	 *
1346	 * By default, an endpoint should always transmit to the primary
1347	 * path, unless the SCTP user explicitly specifies the
1348	 * destination transport address (and possibly source transport
1349	 * address) to use. [If the primary is active but not most recent,
1350	 * bump the most recently used transport.]
1351	 */
1352	if ((asoc->peer.primary_path->state == SCTP_ACTIVE ||
1353	     asoc->peer.primary_path->state == SCTP_UNKNOWN) &&
1354	     asoc->peer.primary_path != trans_pri) {
1355		trans_sec = trans_pri;
1356		trans_pri = asoc->peer.primary_path;
1357	}
1358
1359	/* We did not find anything useful for a possible retransmission
1360	 * path; either primary path that we found is the same as
1361	 * the current one, or we didn't generally find an active one.
1362	 */
1363	if (trans_sec == NULL)
1364		trans_sec = trans_pri;
1365
1366	/* If we failed to find a usable transport, just camp on the
1367	 * active or pick a PF iff it's the better choice.
1368	 */
1369	if (trans_pri == NULL) {
1370		trans_pri = sctp_trans_elect_best(asoc->peer.active_path, trans_pf);
1371		trans_sec = trans_pri;
1372	}
1373
1374	/* Set the active and retran transports. */
1375	asoc->peer.active_path = trans_pri;
1376	asoc->peer.retran_path = trans_sec;
1377}
1378
1379struct sctp_transport *
1380sctp_assoc_choose_alter_transport(struct sctp_association *asoc,
1381				  struct sctp_transport *last_sent_to)
1382{
1383	/* If this is the first time packet is sent, use the active path,
1384	 * else use the retran path. If the last packet was sent over the
1385	 * retran path, update the retran path and use it.
1386	 */
1387	if (last_sent_to == NULL) {
1388		return asoc->peer.active_path;
1389	} else {
1390		if (last_sent_to == asoc->peer.retran_path)
1391			sctp_assoc_update_retran_path(asoc);
1392
1393		return asoc->peer.retran_path;
1394	}
1395}
1396
1397void sctp_assoc_update_frag_point(struct sctp_association *asoc)
1398{
1399	int frag = sctp_mtu_payload(sctp_sk(asoc->base.sk), asoc->pathmtu,
1400				    sctp_datachk_len(&asoc->stream));
1401
1402	if (asoc->user_frag)
1403		frag = min_t(int, frag, asoc->user_frag);
1404
1405	frag = min_t(int, frag, SCTP_MAX_CHUNK_LEN -
1406				sctp_datachk_len(&asoc->stream));
1407
1408	asoc->frag_point = SCTP_TRUNC4(frag);
1409}
1410
1411void sctp_assoc_set_pmtu(struct sctp_association *asoc, __u32 pmtu)
1412{
1413	if (asoc->pathmtu != pmtu) {
1414		asoc->pathmtu = pmtu;
1415		sctp_assoc_update_frag_point(asoc);
1416	}
1417
1418	pr_debug("%s: asoc:%p, pmtu:%d, frag_point:%d\n", __func__, asoc,
1419		 asoc->pathmtu, asoc->frag_point);
1420}
1421
1422/* Update the association's pmtu and frag_point by going through all the
1423 * transports. This routine is called when a transport's PMTU has changed.
1424 */
1425void sctp_assoc_sync_pmtu(struct sctp_association *asoc)
1426{
1427	struct sctp_transport *t;
1428	__u32 pmtu = 0;
1429
1430	if (!asoc)
1431		return;
1432
1433	/* Get the lowest pmtu of all the transports. */
1434	list_for_each_entry(t, &asoc->peer.transport_addr_list, transports) {
1435		if (t->pmtu_pending && t->dst) {
1436			sctp_transport_update_pmtu(t,
1437						   atomic_read(&t->mtu_info));
1438			t->pmtu_pending = 0;
1439		}
1440		if (!pmtu || (t->pathmtu < pmtu))
1441			pmtu = t->pathmtu;
1442	}
1443
1444	sctp_assoc_set_pmtu(asoc, pmtu);
1445}
1446
1447/* Should we send a SACK to update our peer? */
1448static inline bool sctp_peer_needs_update(struct sctp_association *asoc)
1449{
1450	struct net *net = asoc->base.net;
1451
1452	switch (asoc->state) {
1453	case SCTP_STATE_ESTABLISHED:
1454	case SCTP_STATE_SHUTDOWN_PENDING:
1455	case SCTP_STATE_SHUTDOWN_RECEIVED:
1456	case SCTP_STATE_SHUTDOWN_SENT:
1457		if ((asoc->rwnd > asoc->a_rwnd) &&
1458		    ((asoc->rwnd - asoc->a_rwnd) >= max_t(__u32,
1459			   (asoc->base.sk->sk_rcvbuf >> net->sctp.rwnd_upd_shift),
1460			   asoc->pathmtu)))
1461			return true;
1462		break;
1463	default:
1464		break;
1465	}
1466	return false;
1467}
1468
1469/* Increase asoc's rwnd by len and send any window update SACK if needed. */
1470void sctp_assoc_rwnd_increase(struct sctp_association *asoc, unsigned int len)
1471{
1472	struct sctp_chunk *sack;
1473	struct timer_list *timer;
1474
1475	if (asoc->rwnd_over) {
1476		if (asoc->rwnd_over >= len) {
1477			asoc->rwnd_over -= len;
1478		} else {
1479			asoc->rwnd += (len - asoc->rwnd_over);
1480			asoc->rwnd_over = 0;
1481		}
1482	} else {
1483		asoc->rwnd += len;
1484	}
1485
1486	/* If we had window pressure, start recovering it
1487	 * once our rwnd had reached the accumulated pressure
1488	 * threshold.  The idea is to recover slowly, but up
1489	 * to the initial advertised window.
1490	 */
1491	if (asoc->rwnd_press) {
1492		int change = min(asoc->pathmtu, asoc->rwnd_press);
1493		asoc->rwnd += change;
1494		asoc->rwnd_press -= change;
1495	}
1496
1497	pr_debug("%s: asoc:%p rwnd increased by %d to (%u, %u) - %u\n",
1498		 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
1499		 asoc->a_rwnd);
1500
1501	/* Send a window update SACK if the rwnd has increased by at least the
1502	 * minimum of the association's PMTU and half of the receive buffer.
1503	 * The algorithm used is similar to the one described in
1504	 * Section 4.2.3.3 of RFC 1122.
1505	 */
1506	if (sctp_peer_needs_update(asoc)) {
1507		asoc->a_rwnd = asoc->rwnd;
1508
1509		pr_debug("%s: sending window update SACK- asoc:%p rwnd:%u "
1510			 "a_rwnd:%u\n", __func__, asoc, asoc->rwnd,
1511			 asoc->a_rwnd);
1512
1513		sack = sctp_make_sack(asoc);
1514		if (!sack)
1515			return;
1516
1517		asoc->peer.sack_needed = 0;
1518
1519		sctp_outq_tail(&asoc->outqueue, sack, GFP_ATOMIC);
1520
1521		/* Stop the SACK timer.  */
1522		timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
1523		if (del_timer(timer))
1524			sctp_association_put(asoc);
1525	}
1526}
1527
1528/* Decrease asoc's rwnd by len. */
1529void sctp_assoc_rwnd_decrease(struct sctp_association *asoc, unsigned int len)
1530{
1531	int rx_count;
1532	int over = 0;
1533
1534	if (unlikely(!asoc->rwnd || asoc->rwnd_over))
1535		pr_debug("%s: association:%p has asoc->rwnd:%u, "
1536			 "asoc->rwnd_over:%u!\n", __func__, asoc,
1537			 asoc->rwnd, asoc->rwnd_over);
1538
1539	if (asoc->ep->rcvbuf_policy)
1540		rx_count = atomic_read(&asoc->rmem_alloc);
1541	else
1542		rx_count = atomic_read(&asoc->base.sk->sk_rmem_alloc);
1543
1544	/* If we've reached or overflowed our receive buffer, announce
1545	 * a 0 rwnd if rwnd would still be positive.  Store the
1546	 * potential pressure overflow so that the window can be restored
1547	 * back to original value.
1548	 */
1549	if (rx_count >= asoc->base.sk->sk_rcvbuf)
1550		over = 1;
1551
1552	if (asoc->rwnd >= len) {
1553		asoc->rwnd -= len;
1554		if (over) {
1555			asoc->rwnd_press += asoc->rwnd;
1556			asoc->rwnd = 0;
1557		}
1558	} else {
1559		asoc->rwnd_over += len - asoc->rwnd;
1560		asoc->rwnd = 0;
1561	}
1562
1563	pr_debug("%s: asoc:%p rwnd decreased by %d to (%u, %u, %u)\n",
1564		 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
1565		 asoc->rwnd_press);
1566}
1567
1568/* Build the bind address list for the association based on info from the
1569 * local endpoint and the remote peer.
1570 */
1571int sctp_assoc_set_bind_addr_from_ep(struct sctp_association *asoc,
1572				     enum sctp_scope scope, gfp_t gfp)
1573{
1574	struct sock *sk = asoc->base.sk;
1575	int flags;
1576
1577	/* Use scoping rules to determine the subset of addresses from
1578	 * the endpoint.
1579	 */
1580	flags = (PF_INET6 == sk->sk_family) ? SCTP_ADDR6_ALLOWED : 0;
1581	if (!inet_v6_ipv6only(sk))
1582		flags |= SCTP_ADDR4_ALLOWED;
1583	if (asoc->peer.ipv4_address)
1584		flags |= SCTP_ADDR4_PEERSUPP;
1585	if (asoc->peer.ipv6_address)
1586		flags |= SCTP_ADDR6_PEERSUPP;
1587
1588	return sctp_bind_addr_copy(asoc->base.net,
1589				   &asoc->base.bind_addr,
1590				   &asoc->ep->base.bind_addr,
1591				   scope, gfp, flags);
1592}
1593
1594/* Build the association's bind address list from the cookie.  */
1595int sctp_assoc_set_bind_addr_from_cookie(struct sctp_association *asoc,
1596					 struct sctp_cookie *cookie,
1597					 gfp_t gfp)
1598{
1599	struct sctp_init_chunk *peer_init = (struct sctp_init_chunk *)(cookie + 1);
1600	int var_size2 = ntohs(peer_init->chunk_hdr.length);
1601	int var_size3 = cookie->raw_addr_list_len;
1602	__u8 *raw = (__u8 *)peer_init + var_size2;
1603
1604	return sctp_raw_to_bind_addrs(&asoc->base.bind_addr, raw, var_size3,
1605				      asoc->ep->base.bind_addr.port, gfp);
1606}
1607
1608/* Lookup laddr in the bind address list of an association. */
1609int sctp_assoc_lookup_laddr(struct sctp_association *asoc,
1610			    const union sctp_addr *laddr)
1611{
1612	int found = 0;
1613
1614	if ((asoc->base.bind_addr.port == ntohs(laddr->v4.sin_port)) &&
1615	    sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
1616				 sctp_sk(asoc->base.sk)))
1617		found = 1;
1618
1619	return found;
1620}
1621
1622/* Set an association id for a given association */
1623int sctp_assoc_set_id(struct sctp_association *asoc, gfp_t gfp)
1624{
1625	bool preload = gfpflags_allow_blocking(gfp);
1626	int ret;
1627
1628	/* If the id is already assigned, keep it. */
1629	if (asoc->assoc_id)
1630		return 0;
1631
1632	if (preload)
1633		idr_preload(gfp);
1634	spin_lock_bh(&sctp_assocs_id_lock);
1635	/* 0, 1, 2 are used as SCTP_FUTURE_ASSOC, SCTP_CURRENT_ASSOC and
1636	 * SCTP_ALL_ASSOC, so an available id must be > SCTP_ALL_ASSOC.
1637	 */
1638	ret = idr_alloc_cyclic(&sctp_assocs_id, asoc, SCTP_ALL_ASSOC + 1, 0,
1639			       GFP_NOWAIT);
1640	spin_unlock_bh(&sctp_assocs_id_lock);
1641	if (preload)
1642		idr_preload_end();
1643	if (ret < 0)
1644		return ret;
1645
1646	asoc->assoc_id = (sctp_assoc_t)ret;
1647	return 0;
1648}
1649
1650/* Free the ASCONF queue */
1651static void sctp_assoc_free_asconf_queue(struct sctp_association *asoc)
1652{
1653	struct sctp_chunk *asconf;
1654	struct sctp_chunk *tmp;
1655
1656	list_for_each_entry_safe(asconf, tmp, &asoc->addip_chunk_list, list) {
1657		list_del_init(&asconf->list);
1658		sctp_chunk_free(asconf);
1659	}
1660}
1661
1662/* Free asconf_ack cache */
1663static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc)
1664{
1665	struct sctp_chunk *ack;
1666	struct sctp_chunk *tmp;
1667
1668	list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
1669				transmitted_list) {
1670		list_del_init(&ack->transmitted_list);
1671		sctp_chunk_free(ack);
1672	}
1673}
1674
1675/* Clean up the ASCONF_ACK queue */
1676void sctp_assoc_clean_asconf_ack_cache(const struct sctp_association *asoc)
1677{
1678	struct sctp_chunk *ack;
1679	struct sctp_chunk *tmp;
1680
1681	/* We can remove all the entries from the queue up to
1682	 * the "Peer-Sequence-Number".
1683	 */
1684	list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
1685				transmitted_list) {
1686		if (ack->subh.addip_hdr->serial ==
1687				htonl(asoc->peer.addip_serial))
1688			break;
1689
1690		list_del_init(&ack->transmitted_list);
1691		sctp_chunk_free(ack);
1692	}
1693}
1694
1695/* Find the ASCONF_ACK whose serial number matches ASCONF */
1696struct sctp_chunk *sctp_assoc_lookup_asconf_ack(
1697					const struct sctp_association *asoc,
1698					__be32 serial)
1699{
1700	struct sctp_chunk *ack;
1701
1702	/* Walk through the list of cached ASCONF-ACKs and find the
1703	 * ack chunk whose serial number matches that of the request.
1704	 */
1705	list_for_each_entry(ack, &asoc->asconf_ack_list, transmitted_list) {
1706		if (sctp_chunk_pending(ack))
1707			continue;
1708		if (ack->subh.addip_hdr->serial == serial) {
1709			sctp_chunk_hold(ack);
1710			return ack;
1711		}
1712	}
1713
1714	return NULL;
1715}
1716
1717void sctp_asconf_queue_teardown(struct sctp_association *asoc)
1718{
1719	/* Free any cached ASCONF_ACK chunk. */
1720	sctp_assoc_free_asconf_acks(asoc);
1721
1722	/* Free the ASCONF queue. */
1723	sctp_assoc_free_asconf_queue(asoc);
1724
1725	/* Free any cached ASCONF chunk. */
1726	if (asoc->addip_last_asconf)
1727		sctp_chunk_free(asoc->addip_last_asconf);
1728}
1729