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
8 * This file is part of the SCTP kernel implementation
10 * This module provides the abstraction for an SCTP association.
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)
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.
24 * You should have received a copy of the GNU General Public License
25 * along with GNU CC; see the file COPYING. If not, write to
26 * the Free Software Foundation, 59 Temple Place - Suite 330,
27 * Boston, MA 02111-1307, USA.
29 * Please send any bug reports or fixes you make to the
31 * lksctp developers <lksctp-developers@lists.sourceforge.net>
33 * Or submit a bug report through the following website:
34 * http://www.sf.net/projects/lksctp
36 * Written or modified by:
37 * La Monte H.P. Yarroll <piggy@acm.org>
38 * Karl Knutson <karl@athena.chicago.il.us>
39 * Jon Grimm <jgrimm@us.ibm.com>
40 * Xingang Guo <xingang.guo@intel.com>
41 * Hui Huang <hui.huang@nokia.com>
42 * Sridhar Samudrala <sri@us.ibm.com>
43 * Daisy Chang <daisyc@us.ibm.com>
44 * Ryan Layer <rmlayer@us.ibm.com>
45 * Kevin Gao <kevin.gao@intel.com>
47 * Any bugs reported given to us we will try to fix... any fixes shared will
48 * be incorporated into the next SCTP release.
51 #include <linux/types.h>
52 #include <linux/fcntl.h>
53 #include <linux/poll.h>
54 #include <linux/init.h>
56 #include <linux/slab.h>
59 #include <net/sctp/sctp.h>
60 #include <net/sctp/sm.h>
62 /* Forward declarations for internal functions. */
63 static void sctp_assoc_bh_rcv(struct work_struct *work);
64 static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc);
67 /* 1st Level Abstractions. */
69 /* Initialize a new association from provided memory. */
70 static struct sctp_association *sctp_association_init(struct sctp_association *asoc,
71 const struct sctp_endpoint *ep,
72 const struct sock *sk,
81 /* Retrieve the SCTP per socket area. */
82 sp = sctp_sk((struct sock *)sk);
84 /* Init all variables to a known value. */
85 memset(asoc, 0, sizeof(struct sctp_association));
87 /* Discarding const is appropriate here. */
88 asoc->ep = (struct sctp_endpoint *)ep;
89 sctp_endpoint_hold(asoc->ep);
92 asoc->base.sk = (struct sock *)sk;
93 sock_hold(asoc->base.sk);
95 /* Initialize the common base substructure. */
96 asoc->base.type = SCTP_EP_TYPE_ASSOCIATION;
98 /* Initialize the object handling fields. */
99 atomic_set(&asoc->base.refcnt, 1);
101 asoc->base.malloced = 0;
103 /* Initialize the bind addr area. */
104 sctp_bind_addr_init(&asoc->base.bind_addr, ep->base.bind_addr.port);
106 asoc->state = SCTP_STATE_CLOSED;
108 /* Set these values from the socket values, a conversion between
109 * millsecons to seconds/microseconds must also be done.
111 asoc->cookie_life.tv_sec = sp->assocparams.sasoc_cookie_life / 1000;
112 asoc->cookie_life.tv_usec = (sp->assocparams.sasoc_cookie_life % 1000)
114 asoc->frag_point = 0;
116 /* Set the association max_retrans and RTO values from the
119 asoc->max_retrans = sp->assocparams.sasoc_asocmaxrxt;
120 asoc->rto_initial = msecs_to_jiffies(sp->rtoinfo.srto_initial);
121 asoc->rto_max = msecs_to_jiffies(sp->rtoinfo.srto_max);
122 asoc->rto_min = msecs_to_jiffies(sp->rtoinfo.srto_min);
124 asoc->overall_error_count = 0;
126 /* Initialize the association's heartbeat interval based on the
127 * sock configured value.
129 asoc->hbinterval = msecs_to_jiffies(sp->hbinterval);
131 /* Initialize path max retrans value. */
132 asoc->pathmaxrxt = sp->pathmaxrxt;
134 /* Initialize default path MTU. */
135 asoc->pathmtu = sp->pathmtu;
137 /* Set association default SACK delay */
138 asoc->sackdelay = msecs_to_jiffies(sp->sackdelay);
139 asoc->sackfreq = sp->sackfreq;
141 /* Set the association default flags controlling
142 * Heartbeat, SACK delay, and Path MTU Discovery.
144 asoc->param_flags = sp->param_flags;
146 /* Initialize the maximum mumber of new data packets that can be sent
149 asoc->max_burst = sp->max_burst;
151 /* initialize association timers */
152 asoc->timeouts[SCTP_EVENT_TIMEOUT_NONE] = 0;
153 asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_COOKIE] = asoc->rto_initial;
154 asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_INIT] = asoc->rto_initial;
155 asoc->timeouts[SCTP_EVENT_TIMEOUT_T2_SHUTDOWN] = asoc->rto_initial;
156 asoc->timeouts[SCTP_EVENT_TIMEOUT_T3_RTX] = 0;
157 asoc->timeouts[SCTP_EVENT_TIMEOUT_T4_RTO] = 0;
159 /* sctpimpguide Section 2.12.2
160 * If the 'T5-shutdown-guard' timer is used, it SHOULD be set to the
161 * recommended value of 5 times 'RTO.Max'.
163 asoc->timeouts[SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]
166 asoc->timeouts[SCTP_EVENT_TIMEOUT_HEARTBEAT] = 0;
167 asoc->timeouts[SCTP_EVENT_TIMEOUT_SACK] = asoc->sackdelay;
168 asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE] =
171 /* Initilizes the timers */
172 for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i)
173 setup_timer(&asoc->timers[i], sctp_timer_events[i],
174 (unsigned long)asoc);
176 /* Pull default initialization values from the sock options.
177 * Note: This assumes that the values have already been
178 * validated in the sock.
180 asoc->c.sinit_max_instreams = sp->initmsg.sinit_max_instreams;
181 asoc->c.sinit_num_ostreams = sp->initmsg.sinit_num_ostreams;
182 asoc->max_init_attempts = sp->initmsg.sinit_max_attempts;
184 asoc->max_init_timeo =
185 msecs_to_jiffies(sp->initmsg.sinit_max_init_timeo);
187 /* Allocate storage for the ssnmap after the inbound and outbound
188 * streams have been negotiated during Init.
192 /* Set the local window size for receive.
193 * This is also the rcvbuf space per association.
194 * RFC 6 - A SCTP receiver MUST be able to receive a minimum of
195 * 1500 bytes in one SCTP packet.
197 if ((sk->sk_rcvbuf/2) < SCTP_DEFAULT_MINWINDOW)
198 asoc->rwnd = SCTP_DEFAULT_MINWINDOW;
200 asoc->rwnd = sk->sk_rcvbuf/2;
202 asoc->a_rwnd = asoc->rwnd;
206 /* Use my own max window until I learn something better. */
207 asoc->peer.rwnd = SCTP_DEFAULT_MAXWINDOW;
209 /* Set the sndbuf size for transmit. */
210 asoc->sndbuf_used = 0;
212 /* Initialize the receive memory counter */
213 atomic_set(&asoc->rmem_alloc, 0);
215 init_waitqueue_head(&asoc->wait);
217 asoc->c.my_vtag = sctp_generate_tag(ep);
218 asoc->peer.i.init_tag = 0; /* INIT needs a vtag of 0. */
219 asoc->c.peer_vtag = 0;
221 asoc->c.peer_ttag = 0;
222 asoc->c.my_port = ep->base.bind_addr.port;
224 asoc->c.initial_tsn = sctp_generate_tsn(ep);
226 asoc->next_tsn = asoc->c.initial_tsn;
228 asoc->ctsn_ack_point = asoc->next_tsn - 1;
229 asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
230 asoc->highest_sacked = asoc->ctsn_ack_point;
231 asoc->last_cwr_tsn = asoc->ctsn_ack_point;
232 asoc->unack_data = 0;
234 /* ADDIP Section 4.1 Asconf Chunk Procedures
236 * When an endpoint has an ASCONF signaled change to be sent to the
237 * remote endpoint it should do the following:
239 * A2) a serial number should be assigned to the chunk. The serial
240 * number SHOULD be a monotonically increasing number. The serial
241 * numbers SHOULD be initialized at the start of the
242 * association to the same value as the initial TSN.
244 asoc->addip_serial = asoc->c.initial_tsn;
246 INIT_LIST_HEAD(&asoc->addip_chunk_list);
247 INIT_LIST_HEAD(&asoc->asconf_ack_list);
249 /* Make an empty list of remote transport addresses. */
250 INIT_LIST_HEAD(&asoc->peer.transport_addr_list);
251 asoc->peer.transport_count = 0;
253 /* RFC 2960 5.1 Normal Establishment of an Association
255 * After the reception of the first data chunk in an
256 * association the endpoint must immediately respond with a
257 * sack to acknowledge the data chunk. Subsequent
258 * acknowledgements should be done as described in Section
261 * [We implement this by telling a new association that it
262 * already received one packet.]
264 asoc->peer.sack_needed = 1;
265 asoc->peer.sack_cnt = 0;
267 /* Assume that the peer will tell us if he recognizes ASCONF
268 * as part of INIT exchange.
269 * The sctp_addip_noauth option is there for backward compatibilty
270 * and will revert old behavior.
272 asoc->peer.asconf_capable = 0;
273 if (sctp_addip_noauth)
274 asoc->peer.asconf_capable = 1;
276 /* Create an input queue. */
277 sctp_inq_init(&asoc->base.inqueue);
278 sctp_inq_set_th_handler(&asoc->base.inqueue, sctp_assoc_bh_rcv);
280 /* Create an output queue. */
281 sctp_outq_init(asoc, &asoc->outqueue);
283 if (!sctp_ulpq_init(&asoc->ulpq, asoc))
286 /* Set up the tsn tracking. */
287 sctp_tsnmap_init(&asoc->peer.tsn_map, SCTP_TSN_MAP_SIZE, 0);
293 /* Assume that peer would support both address types unless we are
296 asoc->peer.ipv4_address = 1;
297 asoc->peer.ipv6_address = 1;
298 INIT_LIST_HEAD(&asoc->asocs);
300 asoc->autoclose = sp->autoclose;
302 asoc->default_stream = sp->default_stream;
303 asoc->default_ppid = sp->default_ppid;
304 asoc->default_flags = sp->default_flags;
305 asoc->default_context = sp->default_context;
306 asoc->default_timetolive = sp->default_timetolive;
307 asoc->default_rcv_context = sp->default_rcv_context;
309 /* AUTH related initializations */
310 INIT_LIST_HEAD(&asoc->endpoint_shared_keys);
311 err = sctp_auth_asoc_copy_shkeys(ep, asoc, gfp);
315 asoc->active_key_id = ep->active_key_id;
316 asoc->asoc_shared_key = NULL;
318 asoc->default_hmac_id = 0;
319 /* Save the hmacs and chunks list into this association */
320 if (ep->auth_hmacs_list)
321 memcpy(asoc->c.auth_hmacs, ep->auth_hmacs_list,
322 ntohs(ep->auth_hmacs_list->param_hdr.length));
323 if (ep->auth_chunk_list)
324 memcpy(asoc->c.auth_chunks, ep->auth_chunk_list,
325 ntohs(ep->auth_chunk_list->param_hdr.length));
327 /* Get the AUTH random number for this association */
328 p = (sctp_paramhdr_t *)asoc->c.auth_random;
329 p->type = SCTP_PARAM_RANDOM;
330 p->length = htons(sizeof(sctp_paramhdr_t) + SCTP_AUTH_RANDOM_LENGTH);
331 get_random_bytes(p+1, SCTP_AUTH_RANDOM_LENGTH);
336 sctp_endpoint_put(asoc->ep);
337 sock_put(asoc->base.sk);
341 /* Allocate and initialize a new association */
342 struct sctp_association *sctp_association_new(const struct sctp_endpoint *ep,
343 const struct sock *sk,
347 struct sctp_association *asoc;
349 asoc = t_new(struct sctp_association, gfp);
353 if (!sctp_association_init(asoc, ep, sk, scope, gfp))
356 asoc->base.malloced = 1;
357 SCTP_DBG_OBJCNT_INC(assoc);
358 SCTP_DEBUG_PRINTK("Created asoc %p\n", asoc);
368 /* Free this association if possible. There may still be users, so
369 * the actual deallocation may be delayed.
371 void sctp_association_free(struct sctp_association *asoc)
373 struct sock *sk = asoc->base.sk;
374 struct sctp_transport *transport;
375 struct list_head *pos, *temp;
378 /* Only real associations count against the endpoint, so
379 * don't bother for if this is a temporary association.
382 list_del(&asoc->asocs);
384 /* Decrement the backlog value for a TCP-style listening
387 if (sctp_style(sk, TCP) && sctp_sstate(sk, LISTENING))
388 sk->sk_ack_backlog--;
391 /* Mark as dead, so other users can know this structure is
396 /* Dispose of any data lying around in the outqueue. */
397 sctp_outq_free(&asoc->outqueue);
399 /* Dispose of any pending messages for the upper layer. */
400 sctp_ulpq_free(&asoc->ulpq);
402 /* Dispose of any pending chunks on the inqueue. */
403 sctp_inq_free(&asoc->base.inqueue);
405 /* Free ssnmap storage. */
406 sctp_ssnmap_free(asoc->ssnmap);
408 /* Clean up the bound address list. */
409 sctp_bind_addr_free(&asoc->base.bind_addr);
411 /* Do we need to go through all of our timers and
412 * delete them? To be safe we will try to delete all, but we
413 * should be able to go through and make a guess based
416 for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i) {
417 if (timer_pending(&asoc->timers[i]) &&
418 del_timer(&asoc->timers[i]))
419 sctp_association_put(asoc);
422 /* Free peer's cached cookie. */
423 kfree(asoc->peer.cookie);
424 kfree(asoc->peer.peer_random);
425 kfree(asoc->peer.peer_chunks);
426 kfree(asoc->peer.peer_hmacs);
428 /* Release the transport structures. */
429 list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
430 transport = list_entry(pos, struct sctp_transport, transports);
432 sctp_transport_free(transport);
435 asoc->peer.transport_count = 0;
437 /* Free any cached ASCONF_ACK chunk. */
438 sctp_assoc_free_asconf_acks(asoc);
440 /* Free any cached ASCONF chunk. */
441 if (asoc->addip_last_asconf)
442 sctp_chunk_free(asoc->addip_last_asconf);
444 /* AUTH - Free the endpoint shared keys */
445 sctp_auth_destroy_keys(&asoc->endpoint_shared_keys);
447 /* AUTH - Free the association shared key */
448 sctp_auth_key_put(asoc->asoc_shared_key);
450 sctp_association_put(asoc);
453 /* Cleanup and free up an association. */
454 static void sctp_association_destroy(struct sctp_association *asoc)
456 SCTP_ASSERT(asoc->base.dead, "Assoc is not dead", return);
458 sctp_endpoint_put(asoc->ep);
459 sock_put(asoc->base.sk);
461 if (asoc->assoc_id != 0) {
462 spin_lock_bh(&sctp_assocs_id_lock);
463 idr_remove(&sctp_assocs_id, asoc->assoc_id);
464 spin_unlock_bh(&sctp_assocs_id_lock);
467 WARN_ON(atomic_read(&asoc->rmem_alloc));
469 if (asoc->base.malloced) {
471 SCTP_DBG_OBJCNT_DEC(assoc);
475 /* Change the primary destination address for the peer. */
476 void sctp_assoc_set_primary(struct sctp_association *asoc,
477 struct sctp_transport *transport)
481 /* it's a changeover only if we already have a primary path
482 * that we are changing
484 if (asoc->peer.primary_path != NULL &&
485 asoc->peer.primary_path != transport)
488 asoc->peer.primary_path = transport;
490 /* Set a default msg_name for events. */
491 memcpy(&asoc->peer.primary_addr, &transport->ipaddr,
492 sizeof(union sctp_addr));
494 /* If the primary path is changing, assume that the
495 * user wants to use this new path.
497 if ((transport->state == SCTP_ACTIVE) ||
498 (transport->state == SCTP_UNKNOWN))
499 asoc->peer.active_path = transport;
502 * SFR-CACC algorithm:
503 * Upon the receipt of a request to change the primary
504 * destination address, on the data structure for the new
505 * primary destination, the sender MUST do the following:
507 * 1) If CHANGEOVER_ACTIVE is set, then there was a switch
508 * to this destination address earlier. The sender MUST set
509 * CYCLING_CHANGEOVER to indicate that this switch is a
510 * double switch to the same destination address.
512 if (transport->cacc.changeover_active)
513 transport->cacc.cycling_changeover = changeover;
515 /* 2) The sender MUST set CHANGEOVER_ACTIVE to indicate that
516 * a changeover has occurred.
518 transport->cacc.changeover_active = changeover;
520 /* 3) The sender MUST store the next TSN to be sent in
521 * next_tsn_at_change.
523 transport->cacc.next_tsn_at_change = asoc->next_tsn;
526 /* Remove a transport from an association. */
527 void sctp_assoc_rm_peer(struct sctp_association *asoc,
528 struct sctp_transport *peer)
530 struct list_head *pos;
531 struct sctp_transport *transport;
533 SCTP_DEBUG_PRINTK_IPADDR("sctp_assoc_rm_peer:association %p addr: ",
537 ntohs(peer->ipaddr.v4.sin_port));
539 /* If we are to remove the current retran_path, update it
540 * to the next peer before removing this peer from the list.
542 if (asoc->peer.retran_path == peer)
543 sctp_assoc_update_retran_path(asoc);
545 /* Remove this peer from the list. */
546 list_del(&peer->transports);
548 /* Get the first transport of asoc. */
549 pos = asoc->peer.transport_addr_list.next;
550 transport = list_entry(pos, struct sctp_transport, transports);
552 /* Update any entries that match the peer to be deleted. */
553 if (asoc->peer.primary_path == peer)
554 sctp_assoc_set_primary(asoc, transport);
555 if (asoc->peer.active_path == peer)
556 asoc->peer.active_path = transport;
557 if (asoc->peer.last_data_from == peer)
558 asoc->peer.last_data_from = transport;
560 /* If we remove the transport an INIT was last sent to, set it to
561 * NULL. Combined with the update of the retran path above, this
562 * will cause the next INIT to be sent to the next available
563 * transport, maintaining the cycle.
565 if (asoc->init_last_sent_to == peer)
566 asoc->init_last_sent_to = NULL;
568 asoc->peer.transport_count--;
570 sctp_transport_free(peer);
573 /* Add a transport address to an association. */
574 struct sctp_transport *sctp_assoc_add_peer(struct sctp_association *asoc,
575 const union sctp_addr *addr,
577 const int peer_state)
579 struct sctp_transport *peer;
580 struct sctp_sock *sp;
583 sp = sctp_sk(asoc->base.sk);
585 /* AF_INET and AF_INET6 share common port field. */
586 port = ntohs(addr->v4.sin_port);
588 SCTP_DEBUG_PRINTK_IPADDR("sctp_assoc_add_peer:association %p addr: ",
589 " port: %d state:%d\n",
595 /* Set the port if it has not been set yet. */
596 if (0 == asoc->peer.port)
597 asoc->peer.port = port;
599 /* Check to see if this is a duplicate. */
600 peer = sctp_assoc_lookup_paddr(asoc, addr);
602 if (peer->state == SCTP_UNKNOWN) {
603 if (peer_state == SCTP_ACTIVE)
604 peer->state = SCTP_ACTIVE;
605 if (peer_state == SCTP_UNCONFIRMED)
606 peer->state = SCTP_UNCONFIRMED;
611 peer = sctp_transport_new(addr, gfp);
615 sctp_transport_set_owner(peer, asoc);
617 /* Initialize the peer's heartbeat interval based on the
618 * association configured value.
620 peer->hbinterval = asoc->hbinterval;
622 /* Set the path max_retrans. */
623 peer->pathmaxrxt = asoc->pathmaxrxt;
625 /* Initialize the peer's SACK delay timeout based on the
626 * association configured value.
628 peer->sackdelay = asoc->sackdelay;
629 peer->sackfreq = asoc->sackfreq;
631 /* Enable/disable heartbeat, SACK delay, and path MTU discovery
632 * based on association setting.
634 peer->param_flags = asoc->param_flags;
636 /* Initialize the pmtu of the transport. */
637 if (peer->param_flags & SPP_PMTUD_ENABLE)
638 sctp_transport_pmtu(peer);
639 else if (asoc->pathmtu)
640 peer->pathmtu = asoc->pathmtu;
642 peer->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
644 /* If this is the first transport addr on this association,
645 * initialize the association PMTU to the peer's PMTU.
646 * If not and the current association PMTU is higher than the new
647 * peer's PMTU, reset the association PMTU to the new peer's PMTU.
650 asoc->pathmtu = min_t(int, peer->pathmtu, asoc->pathmtu);
652 asoc->pathmtu = peer->pathmtu;
654 SCTP_DEBUG_PRINTK("sctp_assoc_add_peer:association %p PMTU set to "
655 "%d\n", asoc, asoc->pathmtu);
656 peer->pmtu_pending = 0;
658 asoc->frag_point = sctp_frag_point(sp, asoc->pathmtu);
660 /* The asoc->peer.port might not be meaningful yet, but
661 * initialize the packet structure anyway.
663 sctp_packet_init(&peer->packet, peer, asoc->base.bind_addr.port,
668 * o The initial cwnd before DATA transmission or after a sufficiently
669 * long idle period MUST be set to
670 * min(4*MTU, max(2*MTU, 4380 bytes))
672 * o The initial value of ssthresh MAY be arbitrarily high
673 * (for example, implementations MAY use the size of the
674 * receiver advertised window).
676 peer->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
678 /* At this point, we may not have the receiver's advertised window,
679 * so initialize ssthresh to the default value and it will be set
680 * later when we process the INIT.
682 peer->ssthresh = SCTP_DEFAULT_MAXWINDOW;
684 peer->partial_bytes_acked = 0;
685 peer->flight_size = 0;
687 /* Set the transport's RTO.initial value */
688 peer->rto = asoc->rto_initial;
690 /* Set the peer's active state. */
691 peer->state = peer_state;
693 /* Attach the remote transport to our asoc. */
694 list_add_tail(&peer->transports, &asoc->peer.transport_addr_list);
695 asoc->peer.transport_count++;
697 /* If we do not yet have a primary path, set one. */
698 if (!asoc->peer.primary_path) {
699 sctp_assoc_set_primary(asoc, peer);
700 asoc->peer.retran_path = peer;
703 if (asoc->peer.active_path == asoc->peer.retran_path) {
704 asoc->peer.retran_path = peer;
710 /* Delete a transport address from an association. */
711 void sctp_assoc_del_peer(struct sctp_association *asoc,
712 const union sctp_addr *addr)
714 struct list_head *pos;
715 struct list_head *temp;
716 struct sctp_transport *transport;
718 list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
719 transport = list_entry(pos, struct sctp_transport, transports);
720 if (sctp_cmp_addr_exact(addr, &transport->ipaddr)) {
721 /* Do book keeping for removing the peer and free it. */
722 sctp_assoc_rm_peer(asoc, transport);
728 /* Lookup a transport by address. */
729 struct sctp_transport *sctp_assoc_lookup_paddr(
730 const struct sctp_association *asoc,
731 const union sctp_addr *address)
733 struct sctp_transport *t;
735 /* Cycle through all transports searching for a peer address. */
737 list_for_each_entry(t, &asoc->peer.transport_addr_list,
739 if (sctp_cmp_addr_exact(address, &t->ipaddr))
746 /* Remove all transports except a give one */
747 void sctp_assoc_del_nonprimary_peers(struct sctp_association *asoc,
748 struct sctp_transport *primary)
750 struct sctp_transport *temp;
751 struct sctp_transport *t;
753 list_for_each_entry_safe(t, temp, &asoc->peer.transport_addr_list,
755 /* if the current transport is not the primary one, delete it */
757 sctp_assoc_rm_peer(asoc, t);
763 /* Engage in transport control operations.
764 * Mark the transport up or down and send a notification to the user.
765 * Select and update the new active and retran paths.
767 void sctp_assoc_control_transport(struct sctp_association *asoc,
768 struct sctp_transport *transport,
769 sctp_transport_cmd_t command,
770 sctp_sn_error_t error)
772 struct sctp_transport *t = NULL;
773 struct sctp_transport *first;
774 struct sctp_transport *second;
775 struct sctp_ulpevent *event;
776 struct sockaddr_storage addr;
779 /* Record the transition on the transport. */
781 case SCTP_TRANSPORT_UP:
782 /* If we are moving from UNCONFIRMED state due
783 * to heartbeat success, report the SCTP_ADDR_CONFIRMED
784 * state to the user, otherwise report SCTP_ADDR_AVAILABLE.
786 if (SCTP_UNCONFIRMED == transport->state &&
787 SCTP_HEARTBEAT_SUCCESS == error)
788 spc_state = SCTP_ADDR_CONFIRMED;
790 spc_state = SCTP_ADDR_AVAILABLE;
791 transport->state = SCTP_ACTIVE;
794 case SCTP_TRANSPORT_DOWN:
795 /* if the transort was never confirmed, do not transition it
798 if (transport->state != SCTP_UNCONFIRMED)
799 transport->state = SCTP_INACTIVE;
801 spc_state = SCTP_ADDR_UNREACHABLE;
808 /* Generate and send a SCTP_PEER_ADDR_CHANGE notification to the
811 memset(&addr, 0, sizeof(struct sockaddr_storage));
812 memcpy(&addr, &transport->ipaddr, transport->af_specific->sockaddr_len);
813 event = sctp_ulpevent_make_peer_addr_change(asoc, &addr,
814 0, spc_state, error, GFP_ATOMIC);
816 sctp_ulpq_tail_event(&asoc->ulpq, event);
818 /* Select new active and retran paths. */
820 /* Look for the two most recently used active transports.
822 * This code produces the wrong ordering whenever jiffies
823 * rolls over, but we still get usable transports, so we don't
826 first = NULL; second = NULL;
828 list_for_each_entry(t, &asoc->peer.transport_addr_list,
831 if ((t->state == SCTP_INACTIVE) ||
832 (t->state == SCTP_UNCONFIRMED))
834 if (!first || t->last_time_heard > first->last_time_heard) {
838 if (!second || t->last_time_heard > second->last_time_heard)
842 /* RFC 2960 6.4 Multi-Homed SCTP Endpoints
844 * By default, an endpoint should always transmit to the
845 * primary path, unless the SCTP user explicitly specifies the
846 * destination transport address (and possibly source
847 * transport address) to use.
849 * [If the primary is active but not most recent, bump the most
850 * recently used transport.]
852 if (((asoc->peer.primary_path->state == SCTP_ACTIVE) ||
853 (asoc->peer.primary_path->state == SCTP_UNKNOWN)) &&
854 first != asoc->peer.primary_path) {
856 first = asoc->peer.primary_path;
859 /* If we failed to find a usable transport, just camp on the
860 * primary, even if it is inactive.
863 first = asoc->peer.primary_path;
864 second = asoc->peer.primary_path;
867 /* Set the active and retran transports. */
868 asoc->peer.active_path = first;
869 asoc->peer.retran_path = second;
872 /* Hold a reference to an association. */
873 void sctp_association_hold(struct sctp_association *asoc)
875 atomic_inc(&asoc->base.refcnt);
878 /* Release a reference to an association and cleanup
879 * if there are no more references.
881 void sctp_association_put(struct sctp_association *asoc)
883 if (atomic_dec_and_test(&asoc->base.refcnt))
884 sctp_association_destroy(asoc);
887 /* Allocate the next TSN, Transmission Sequence Number, for the given
890 __u32 sctp_association_get_next_tsn(struct sctp_association *asoc)
892 /* From Section 1.6 Serial Number Arithmetic:
893 * Transmission Sequence Numbers wrap around when they reach
894 * 2**32 - 1. That is, the next TSN a DATA chunk MUST use
895 * after transmitting TSN = 2*32 - 1 is TSN = 0.
897 __u32 retval = asoc->next_tsn;
904 /* Compare two addresses to see if they match. Wildcard addresses
905 * only match themselves.
907 int sctp_cmp_addr_exact(const union sctp_addr *ss1,
908 const union sctp_addr *ss2)
912 af = sctp_get_af_specific(ss1->sa.sa_family);
916 return af->cmp_addr(ss1, ss2);
919 /* Return an ecne chunk to get prepended to a packet.
920 * Note: We are sly and return a shared, prealloced chunk. FIXME:
921 * No we don't, but we could/should.
923 struct sctp_chunk *sctp_get_ecne_prepend(struct sctp_association *asoc)
925 struct sctp_chunk *chunk;
927 /* Send ECNE if needed.
928 * Not being able to allocate a chunk here is not deadly.
931 chunk = sctp_make_ecne(asoc, asoc->last_ecne_tsn);
939 * Find which transport this TSN was sent on.
941 struct sctp_transport *sctp_assoc_lookup_tsn(struct sctp_association *asoc,
944 struct sctp_transport *active;
945 struct sctp_transport *match;
946 struct sctp_transport *transport;
947 struct sctp_chunk *chunk;
948 __be32 key = htonl(tsn);
953 * FIXME: In general, find a more efficient data structure for
958 * The general strategy is to search each transport's transmitted
959 * list. Return which transport this TSN lives on.
961 * Let's be hopeful and check the active_path first.
962 * Another optimization would be to know if there is only one
963 * outbound path and not have to look for the TSN at all.
967 active = asoc->peer.active_path;
969 list_for_each_entry(chunk, &active->transmitted,
972 if (key == chunk->subh.data_hdr->tsn) {
978 /* If not found, go search all the other transports. */
979 list_for_each_entry(transport, &asoc->peer.transport_addr_list,
982 if (transport == active)
984 list_for_each_entry(chunk, &transport->transmitted,
986 if (key == chunk->subh.data_hdr->tsn) {
996 /* Is this the association we are looking for? */
997 struct sctp_transport *sctp_assoc_is_match(struct sctp_association *asoc,
998 const union sctp_addr *laddr,
999 const union sctp_addr *paddr)
1001 struct sctp_transport *transport;
1003 if ((htons(asoc->base.bind_addr.port) == laddr->v4.sin_port) &&
1004 (htons(asoc->peer.port) == paddr->v4.sin_port)) {
1005 transport = sctp_assoc_lookup_paddr(asoc, paddr);
1009 if (sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
1010 sctp_sk(asoc->base.sk)))
1019 /* Do delayed input processing. This is scheduled by sctp_rcv(). */
1020 static void sctp_assoc_bh_rcv(struct work_struct *work)
1022 struct sctp_association *asoc =
1023 container_of(work, struct sctp_association,
1024 base.inqueue.immediate);
1025 struct sctp_endpoint *ep;
1026 struct sctp_chunk *chunk;
1028 struct sctp_inq *inqueue;
1030 sctp_subtype_t subtype;
1033 /* The association should be held so we should be safe. */
1037 inqueue = &asoc->base.inqueue;
1038 sctp_association_hold(asoc);
1039 while (NULL != (chunk = sctp_inq_pop(inqueue))) {
1040 state = asoc->state;
1041 subtype = SCTP_ST_CHUNK(chunk->chunk_hdr->type);
1043 /* SCTP-AUTH, Section 6.3:
1044 * The receiver has a list of chunk types which it expects
1045 * to be received only after an AUTH-chunk. This list has
1046 * been sent to the peer during the association setup. It
1047 * MUST silently discard these chunks if they are not placed
1048 * after an AUTH chunk in the packet.
1050 if (sctp_auth_recv_cid(subtype.chunk, asoc) && !chunk->auth)
1053 /* Remember where the last DATA chunk came from so we
1054 * know where to send the SACK.
1056 if (sctp_chunk_is_data(chunk))
1057 asoc->peer.last_data_from = chunk->transport;
1059 SCTP_INC_STATS(SCTP_MIB_INCTRLCHUNKS);
1061 if (chunk->transport)
1062 chunk->transport->last_time_heard = jiffies;
1064 /* Run through the state machine. */
1065 error = sctp_do_sm(SCTP_EVENT_T_CHUNK, subtype,
1066 state, ep, asoc, chunk, GFP_ATOMIC);
1068 /* Check to see if the association is freed in response to
1069 * the incoming chunk. If so, get out of the while loop.
1071 if (asoc->base.dead)
1074 /* If there is an error on chunk, discard this packet. */
1076 chunk->pdiscard = 1;
1078 sctp_association_put(asoc);
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)
1084 struct sctp_sock *newsp = sctp_sk(newsk);
1085 struct sock *oldsk = assoc->base.sk;
1087 /* Delete the association from the old endpoint's list of
1090 list_del_init(&assoc->asocs);
1092 /* Decrement the backlog value for a TCP-style socket. */
1093 if (sctp_style(oldsk, TCP))
1094 oldsk->sk_ack_backlog--;
1096 /* Release references to the old endpoint and the sock. */
1097 sctp_endpoint_put(assoc->ep);
1098 sock_put(assoc->base.sk);
1100 /* Get a reference to the new endpoint. */
1101 assoc->ep = newsp->ep;
1102 sctp_endpoint_hold(assoc->ep);
1104 /* Get a reference to the new sock. */
1105 assoc->base.sk = newsk;
1106 sock_hold(assoc->base.sk);
1108 /* Add the association to the new endpoint's list of associations. */
1109 sctp_endpoint_add_asoc(newsp->ep, assoc);
1112 /* Update an association (possibly from unexpected COOKIE-ECHO processing). */
1113 void sctp_assoc_update(struct sctp_association *asoc,
1114 struct sctp_association *new)
1116 struct sctp_transport *trans;
1117 struct list_head *pos, *temp;
1119 /* Copy in new parameters of peer. */
1121 asoc->peer.rwnd = new->peer.rwnd;
1122 asoc->peer.sack_needed = new->peer.sack_needed;
1123 asoc->peer.i = new->peer.i;
1124 sctp_tsnmap_init(&asoc->peer.tsn_map, SCTP_TSN_MAP_SIZE,
1125 asoc->peer.i.initial_tsn);
1127 /* Remove any peer addresses not present in the new association. */
1128 list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
1129 trans = list_entry(pos, struct sctp_transport, transports);
1130 if (!sctp_assoc_lookup_paddr(new, &trans->ipaddr))
1131 sctp_assoc_del_peer(asoc, &trans->ipaddr);
1133 if (asoc->state >= SCTP_STATE_ESTABLISHED)
1134 sctp_transport_reset(trans);
1137 /* If the case is A (association restart), use
1138 * initial_tsn as next_tsn. If the case is B, use
1139 * current next_tsn in case data sent to peer
1140 * has been discarded and needs retransmission.
1142 if (asoc->state >= SCTP_STATE_ESTABLISHED) {
1143 asoc->next_tsn = new->next_tsn;
1144 asoc->ctsn_ack_point = new->ctsn_ack_point;
1145 asoc->adv_peer_ack_point = new->adv_peer_ack_point;
1147 /* Reinitialize SSN for both local streams
1148 * and peer's streams.
1150 sctp_ssnmap_clear(asoc->ssnmap);
1152 /* Flush the ULP reassembly and ordered queue.
1153 * Any data there will now be stale and will
1156 sctp_ulpq_flush(&asoc->ulpq);
1158 /* reset the overall association error count so
1159 * that the restarted association doesn't get torn
1160 * down on the next retransmission timer.
1162 asoc->overall_error_count = 0;
1165 /* Add any peer addresses from the new association. */
1166 list_for_each_entry(trans, &new->peer.transport_addr_list,
1168 if (!sctp_assoc_lookup_paddr(asoc, &trans->ipaddr))
1169 sctp_assoc_add_peer(asoc, &trans->ipaddr,
1170 GFP_ATOMIC, trans->state);
1173 asoc->ctsn_ack_point = asoc->next_tsn - 1;
1174 asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
1175 if (!asoc->ssnmap) {
1176 /* Move the ssnmap. */
1177 asoc->ssnmap = new->ssnmap;
1181 if (!asoc->assoc_id) {
1182 /* get a new association id since we don't have one
1185 sctp_assoc_set_id(asoc, GFP_ATOMIC);
1189 /* SCTP-AUTH: Save the peer parameters from the new assocaitions
1190 * and also move the association shared keys over
1192 kfree(asoc->peer.peer_random);
1193 asoc->peer.peer_random = new->peer.peer_random;
1194 new->peer.peer_random = NULL;
1196 kfree(asoc->peer.peer_chunks);
1197 asoc->peer.peer_chunks = new->peer.peer_chunks;
1198 new->peer.peer_chunks = NULL;
1200 kfree(asoc->peer.peer_hmacs);
1201 asoc->peer.peer_hmacs = new->peer.peer_hmacs;
1202 new->peer.peer_hmacs = NULL;
1204 sctp_auth_key_put(asoc->asoc_shared_key);
1205 sctp_auth_asoc_init_active_key(asoc, GFP_ATOMIC);
1208 /* Update the retran path for sending a retransmitted packet.
1209 * Round-robin through the active transports, else round-robin
1210 * through the inactive transports as this is the next best thing
1213 void sctp_assoc_update_retran_path(struct sctp_association *asoc)
1215 struct sctp_transport *t, *next;
1216 struct list_head *head = &asoc->peer.transport_addr_list;
1217 struct list_head *pos;
1219 if (asoc->peer.transport_count == 1)
1222 /* Find the next transport in a round-robin fashion. */
1223 t = asoc->peer.retran_path;
1224 pos = &t->transports;
1228 /* Skip the head. */
1229 if (pos->next == head)
1234 t = list_entry(pos, struct sctp_transport, transports);
1236 /* We have exhausted the list, but didn't find any
1237 * other active transports. If so, use the next
1240 if (t == asoc->peer.retran_path) {
1245 /* Try to find an active transport. */
1247 if ((t->state == SCTP_ACTIVE) ||
1248 (t->state == SCTP_UNKNOWN)) {
1251 /* Keep track of the next transport in case
1252 * we don't find any active transport.
1259 asoc->peer.retran_path = t;
1261 SCTP_DEBUG_PRINTK_IPADDR("sctp_assoc_update_retran_path:association"
1266 ntohs(t->ipaddr.v4.sin_port));
1269 /* Choose the transport for sending a INIT packet. */
1270 struct sctp_transport *sctp_assoc_choose_init_transport(
1271 struct sctp_association *asoc)
1273 struct sctp_transport *t;
1275 /* Use the retran path. If the last INIT was sent over the
1276 * retran path, update the retran path and use it.
1278 if (!asoc->init_last_sent_to) {
1279 t = asoc->peer.active_path;
1281 if (asoc->init_last_sent_to == asoc->peer.retran_path)
1282 sctp_assoc_update_retran_path(asoc);
1283 t = asoc->peer.retran_path;
1286 SCTP_DEBUG_PRINTK_IPADDR("sctp_assoc_update_retran_path:association"
1291 ntohs(t->ipaddr.v4.sin_port));
1296 /* Choose the transport for sending a SHUTDOWN packet. */
1297 struct sctp_transport *sctp_assoc_choose_shutdown_transport(
1298 struct sctp_association *asoc)
1300 /* If this is the first time SHUTDOWN is sent, use the active path,
1301 * else use the retran path. If the last SHUTDOWN was sent over the
1302 * retran path, update the retran path and use it.
1304 if (!asoc->shutdown_last_sent_to)
1305 return asoc->peer.active_path;
1307 if (asoc->shutdown_last_sent_to == asoc->peer.retran_path)
1308 sctp_assoc_update_retran_path(asoc);
1309 return asoc->peer.retran_path;
1314 /* Update the association's pmtu and frag_point by going through all the
1315 * transports. This routine is called when a transport's PMTU has changed.
1317 void sctp_assoc_sync_pmtu(struct sctp_association *asoc)
1319 struct sctp_transport *t;
1325 /* Get the lowest pmtu of all the transports. */
1326 list_for_each_entry(t, &asoc->peer.transport_addr_list,
1328 if (t->pmtu_pending && t->dst) {
1329 sctp_transport_update_pmtu(t, dst_mtu(t->dst));
1330 t->pmtu_pending = 0;
1332 if (!pmtu || (t->pathmtu < pmtu))
1337 struct sctp_sock *sp = sctp_sk(asoc->base.sk);
1338 asoc->pathmtu = pmtu;
1339 asoc->frag_point = sctp_frag_point(sp, pmtu);
1342 SCTP_DEBUG_PRINTK("%s: asoc:%p, pmtu:%d, frag_point:%d\n",
1343 __func__, asoc, asoc->pathmtu, asoc->frag_point);
1346 /* Should we send a SACK to update our peer? */
1347 static inline int sctp_peer_needs_update(struct sctp_association *asoc)
1349 switch (asoc->state) {
1350 case SCTP_STATE_ESTABLISHED:
1351 case SCTP_STATE_SHUTDOWN_PENDING:
1352 case SCTP_STATE_SHUTDOWN_RECEIVED:
1353 case SCTP_STATE_SHUTDOWN_SENT:
1354 if ((asoc->rwnd > asoc->a_rwnd) &&
1355 ((asoc->rwnd - asoc->a_rwnd) >=
1356 min_t(__u32, (asoc->base.sk->sk_rcvbuf >> 1), asoc->pathmtu)))
1365 /* Increase asoc's rwnd by len and send any window update SACK if needed. */
1366 void sctp_assoc_rwnd_increase(struct sctp_association *asoc, unsigned len)
1368 struct sctp_chunk *sack;
1369 struct timer_list *timer;
1371 if (asoc->rwnd_over) {
1372 if (asoc->rwnd_over >= len) {
1373 asoc->rwnd_over -= len;
1375 asoc->rwnd += (len - asoc->rwnd_over);
1376 asoc->rwnd_over = 0;
1382 SCTP_DEBUG_PRINTK("%s: asoc %p rwnd increased by %d to (%u, %u) "
1383 "- %u\n", __func__, asoc, len, asoc->rwnd,
1384 asoc->rwnd_over, asoc->a_rwnd);
1386 /* Send a window update SACK if the rwnd has increased by at least the
1387 * minimum of the association's PMTU and half of the receive buffer.
1388 * The algorithm used is similar to the one described in
1389 * Section 4.2.3.3 of RFC 1122.
1391 if (sctp_peer_needs_update(asoc)) {
1392 asoc->a_rwnd = asoc->rwnd;
1393 SCTP_DEBUG_PRINTK("%s: Sending window update SACK- asoc: %p "
1394 "rwnd: %u a_rwnd: %u\n", __func__,
1395 asoc, asoc->rwnd, asoc->a_rwnd);
1396 sack = sctp_make_sack(asoc);
1400 asoc->peer.sack_needed = 0;
1402 sctp_outq_tail(&asoc->outqueue, sack);
1404 /* Stop the SACK timer. */
1405 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
1406 if (timer_pending(timer) && del_timer(timer))
1407 sctp_association_put(asoc);
1411 /* Decrease asoc's rwnd by len. */
1412 void sctp_assoc_rwnd_decrease(struct sctp_association *asoc, unsigned len)
1414 SCTP_ASSERT(asoc->rwnd, "rwnd zero", return);
1415 SCTP_ASSERT(!asoc->rwnd_over, "rwnd_over not zero", return);
1416 if (asoc->rwnd >= len) {
1419 asoc->rwnd_over = len - asoc->rwnd;
1422 SCTP_DEBUG_PRINTK("%s: asoc %p rwnd decreased by %d to (%u, %u)\n",
1423 __func__, asoc, len, asoc->rwnd,
1427 /* Build the bind address list for the association based on info from the
1428 * local endpoint and the remote peer.
1430 int sctp_assoc_set_bind_addr_from_ep(struct sctp_association *asoc,
1436 /* Use scoping rules to determine the subset of addresses from
1439 scope = sctp_scope(&asoc->peer.active_path->ipaddr);
1440 flags = (PF_INET6 == asoc->base.sk->sk_family) ? SCTP_ADDR6_ALLOWED : 0;
1441 if (asoc->peer.ipv4_address)
1442 flags |= SCTP_ADDR4_PEERSUPP;
1443 if (asoc->peer.ipv6_address)
1444 flags |= SCTP_ADDR6_PEERSUPP;
1446 return sctp_bind_addr_copy(&asoc->base.bind_addr,
1447 &asoc->ep->base.bind_addr,
1451 /* Build the association's bind address list from the cookie. */
1452 int sctp_assoc_set_bind_addr_from_cookie(struct sctp_association *asoc,
1453 struct sctp_cookie *cookie,
1456 int var_size2 = ntohs(cookie->peer_init->chunk_hdr.length);
1457 int var_size3 = cookie->raw_addr_list_len;
1458 __u8 *raw = (__u8 *)cookie->peer_init + var_size2;
1460 return sctp_raw_to_bind_addrs(&asoc->base.bind_addr, raw, var_size3,
1461 asoc->ep->base.bind_addr.port, gfp);
1464 /* Lookup laddr in the bind address list of an association. */
1465 int sctp_assoc_lookup_laddr(struct sctp_association *asoc,
1466 const union sctp_addr *laddr)
1470 if ((asoc->base.bind_addr.port == ntohs(laddr->v4.sin_port)) &&
1471 sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
1472 sctp_sk(asoc->base.sk)))
1478 /* Set an association id for a given association */
1479 int sctp_assoc_set_id(struct sctp_association *asoc, gfp_t gfp)
1484 if (unlikely(!idr_pre_get(&sctp_assocs_id, gfp)))
1487 spin_lock_bh(&sctp_assocs_id_lock);
1488 error = idr_get_new_above(&sctp_assocs_id, (void *)asoc,
1490 spin_unlock_bh(&sctp_assocs_id_lock);
1491 if (error == -EAGAIN)
1496 asoc->assoc_id = (sctp_assoc_t) assoc_id;
1500 /* Free asconf_ack cache */
1501 static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc)
1503 struct sctp_chunk *ack;
1504 struct sctp_chunk *tmp;
1506 list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
1508 list_del_init(&ack->transmitted_list);
1509 sctp_chunk_free(ack);
1513 /* Clean up the ASCONF_ACK queue */
1514 void sctp_assoc_clean_asconf_ack_cache(const struct sctp_association *asoc)
1516 struct sctp_chunk *ack;
1517 struct sctp_chunk *tmp;
1519 /* We can remove all the entries from the queue upto
1520 * the "Peer-Sequence-Number".
1522 list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
1524 if (ack->subh.addip_hdr->serial ==
1525 htonl(asoc->peer.addip_serial))
1528 list_del_init(&ack->transmitted_list);
1529 sctp_chunk_free(ack);
1533 /* Find the ASCONF_ACK whose serial number matches ASCONF */
1534 struct sctp_chunk *sctp_assoc_lookup_asconf_ack(
1535 const struct sctp_association *asoc,
1538 struct sctp_chunk *ack;
1540 /* Walk through the list of cached ASCONF-ACKs and find the
1541 * ack chunk whose serial number matches that of the request.
1543 list_for_each_entry(ack, &asoc->asconf_ack_list, transmitted_list) {
1544 if (ack->subh.addip_hdr->serial == serial) {
1545 sctp_chunk_hold(ack);