blob: 39579c3e0d14c12f165e420be064d7fbf248c0ad [file] [log] [blame]
Vlad Yasevich60c778b2008-01-11 09:57:09 -05001/* SCTP kernel implementation
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * (C) Copyright IBM Corp. 2001, 2004
3 * Copyright (c) 1999-2000 Cisco, Inc.
4 * Copyright (c) 1999-2001 Motorola, Inc.
5 * Copyright (c) 2001 Intel Corp.
6 * Copyright (c) 2001 La Monte H.P. Yarroll
7 *
Vlad Yasevich60c778b2008-01-11 09:57:09 -05008 * This file is part of the SCTP kernel implementation
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 *
10 * This module provides the abstraction for an SCTP association.
11 *
Vlad Yasevich60c778b2008-01-11 09:57:09 -050012 * This SCTP implementation is free software;
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 * you can redistribute it and/or modify it under the terms of
14 * the GNU General Public License as published by
15 * the Free Software Foundation; either version 2, or (at your option)
16 * any later version.
17 *
Vlad Yasevich60c778b2008-01-11 09:57:09 -050018 * This SCTP implementation is distributed in the hope that it
Linus Torvalds1da177e2005-04-16 15:20:36 -070019 * will be useful, but WITHOUT ANY WARRANTY; without even the implied
20 * ************************
21 * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
22 * See the GNU General Public License for more details.
23 *
24 * You should have received a copy of the GNU General Public License
Jeff Kirsher4b2f13a2013-12-06 06:28:48 -080025 * along with GNU CC; see the file COPYING. If not, see
26 * <http://www.gnu.org/licenses/>.
Linus Torvalds1da177e2005-04-16 15:20:36 -070027 *
28 * Please send any bug reports or fixes you make to the
29 * email address(es):
Daniel Borkmann91705c62013-07-23 14:51:47 +020030 * lksctp developers <linux-sctp@vger.kernel.org>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070032 * Written or modified by:
33 * La Monte H.P. Yarroll <piggy@acm.org>
34 * Karl Knutson <karl@athena.chicago.il.us>
35 * Jon Grimm <jgrimm@us.ibm.com>
36 * Xingang Guo <xingang.guo@intel.com>
37 * Hui Huang <hui.huang@nokia.com>
38 * Sridhar Samudrala <sri@us.ibm.com>
39 * Daisy Chang <daisyc@us.ibm.com>
40 * Ryan Layer <rmlayer@us.ibm.com>
41 * Kevin Gao <kevin.gao@intel.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 */
43
Joe Perches145ce502010-08-24 13:21:08 +000044#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
45
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include <linux/types.h>
47#include <linux/fcntl.h>
48#include <linux/poll.h>
49#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
51#include <linux/slab.h>
52#include <linux/in.h>
53#include <net/ipv6.h>
54#include <net/sctp/sctp.h>
55#include <net/sctp/sm.h>
56
57/* Forward declarations for internal functions. */
David Howellsc4028952006-11-22 14:57:56 +000058static void sctp_assoc_bh_rcv(struct work_struct *work);
Vlad Yasevicha08de642007-12-20 14:11:47 -080059static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc);
Wei Yongjun8b4472c2011-05-24 21:48:02 +000060static void sctp_assoc_free_asconf_queue(struct sctp_association *asoc);
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Linus Torvalds1da177e2005-04-16 15:20:36 -070062/* 1st Level Abstractions. */
63
64/* Initialize a new association from provided memory. */
65static struct sctp_association *sctp_association_init(struct sctp_association *asoc,
66 const struct sctp_endpoint *ep,
67 const struct sock *sk,
68 sctp_scope_t scope,
Al Virodd0fc662005-10-07 07:46:04 +010069 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -070070{
Eric W. Biedermane1fc3b12012-08-07 07:29:57 +000071 struct net *net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 struct sctp_sock *sp;
73 int i;
Vlad Yasevicha29a5bd2007-09-16 19:31:35 -070074 sctp_paramhdr_t *p;
75 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
77 /* Retrieve the SCTP per socket area. */
78 sp = sctp_sk((struct sock *)sk);
79
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 /* Discarding const is appropriate here. */
81 asoc->ep = (struct sctp_endpoint *)ep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 asoc->base.sk = (struct sock *)sk;
Daniel Borkmann2e0c9e72013-06-14 18:24:07 +020083
84 sctp_endpoint_hold(asoc->ep);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 sock_hold(asoc->base.sk);
86
87 /* Initialize the common base substructure. */
88 asoc->base.type = SCTP_EP_TYPE_ASSOCIATION;
89
90 /* Initialize the object handling fields. */
91 atomic_set(&asoc->base.refcnt, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
93 /* Initialize the bind addr area. */
94 sctp_bind_addr_init(&asoc->base.bind_addr, ep->base.bind_addr.port);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
96 asoc->state = SCTP_STATE_CLOSED;
Daniel Borkmann52db8822013-06-25 18:17:27 +020097 asoc->cookie_life = ms_to_ktime(sp->assocparams.sasoc_cookie_life);
Vlad Yasevichf68b2e02009-09-04 18:21:00 -040098 asoc->user_frag = sp->user_frag;
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
100 /* Set the association max_retrans and RTO values from the
101 * socket values.
102 */
103 asoc->max_retrans = sp->assocparams.sasoc_asocmaxrxt;
Eric W. Biedermane1fc3b12012-08-07 07:29:57 +0000104 asoc->pf_retrans = net->sctp.pf_retrans;
Neil Horman5aa93bc2012-07-21 07:56:07 +0000105
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 asoc->rto_initial = msecs_to_jiffies(sp->rtoinfo.srto_initial);
107 asoc->rto_max = msecs_to_jiffies(sp->rtoinfo.srto_max);
108 asoc->rto_min = msecs_to_jiffies(sp->rtoinfo.srto_min);
109
Frank Filz52ccb8e2005-12-22 11:36:46 -0800110 /* Initialize the association's heartbeat interval based on the
111 * sock configured value.
112 */
113 asoc->hbinterval = msecs_to_jiffies(sp->hbinterval);
114
115 /* Initialize path max retrans value. */
116 asoc->pathmaxrxt = sp->pathmaxrxt;
117
118 /* Initialize default path MTU. */
119 asoc->pathmtu = sp->pathmtu;
120
121 /* Set association default SACK delay */
122 asoc->sackdelay = msecs_to_jiffies(sp->sackdelay);
Wei Yongjund364d922008-05-09 15:13:26 -0700123 asoc->sackfreq = sp->sackfreq;
Frank Filz52ccb8e2005-12-22 11:36:46 -0800124
125 /* Set the association default flags controlling
126 * Heartbeat, SACK delay, and Path MTU Discovery.
127 */
128 asoc->param_flags = sp->param_flags;
129
wangweidong9d2c8812013-12-06 09:36:30 +0800130 /* Initialize the maximum number of new data packets that can be sent
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 * in a burst.
132 */
Vlad Yasevich70331572007-03-23 11:34:36 -0700133 asoc->max_burst = sp->max_burst;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134
Vladislav Yasevich1e7d3d92005-11-11 16:06:16 -0800135 /* initialize association timers */
Vladislav Yasevich1e7d3d92005-11-11 16:06:16 -0800136 asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_COOKIE] = asoc->rto_initial;
137 asoc->timeouts[SCTP_EVENT_TIMEOUT_T1_INIT] = asoc->rto_initial;
138 asoc->timeouts[SCTP_EVENT_TIMEOUT_T2_SHUTDOWN] = asoc->rto_initial;
Vladislav Yasevich1e7d3d92005-11-11 16:06:16 -0800139
140 /* sctpimpguide Section 2.12.2
141 * If the 'T5-shutdown-guard' timer is used, it SHOULD be set to the
142 * recommended value of 5 times 'RTO.Max'.
143 */
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900144 asoc->timeouts[SCTP_EVENT_TIMEOUT_T5_SHUTDOWN_GUARD]
Vladislav Yasevich1e7d3d92005-11-11 16:06:16 -0800145 = 5 * asoc->rto_max;
146
Frank Filz52ccb8e2005-12-22 11:36:46 -0800147 asoc->timeouts[SCTP_EVENT_TIMEOUT_SACK] = asoc->sackdelay;
Neil Horman9f70f462013-12-10 06:48:15 -0500148 asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE] = sp->autoclose * HZ;
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900149
Uwe Kleine-König421f91d2010-06-11 12:17:00 +0200150 /* Initializes the timers */
Pavel Emelyanovb24b8a22008-01-23 21:20:07 -0800151 for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i)
152 setup_timer(&asoc->timers[i], sctp_timer_events[i],
153 (unsigned long)asoc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
155 /* Pull default initialization values from the sock options.
156 * Note: This assumes that the values have already been
157 * validated in the sock.
158 */
159 asoc->c.sinit_max_instreams = sp->initmsg.sinit_max_instreams;
160 asoc->c.sinit_num_ostreams = sp->initmsg.sinit_num_ostreams;
161 asoc->max_init_attempts = sp->initmsg.sinit_max_attempts;
162
163 asoc->max_init_timeo =
164 msecs_to_jiffies(sp->initmsg.sinit_max_init_timeo);
165
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 /* Set the local window size for receive.
167 * This is also the rcvbuf space per association.
168 * RFC 6 - A SCTP receiver MUST be able to receive a minimum of
169 * 1500 bytes in one SCTP packet.
170 */
Neil Horman049b3ff2005-11-11 16:08:24 -0800171 if ((sk->sk_rcvbuf/2) < SCTP_DEFAULT_MINWINDOW)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 asoc->rwnd = SCTP_DEFAULT_MINWINDOW;
173 else
Neil Horman049b3ff2005-11-11 16:08:24 -0800174 asoc->rwnd = sk->sk_rcvbuf/2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175
176 asoc->a_rwnd = asoc->rwnd;
177
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 /* Use my own max window until I learn something better. */
179 asoc->peer.rwnd = SCTP_DEFAULT_MAXWINDOW;
180
Neil Horman049b3ff2005-11-11 16:08:24 -0800181 /* Initialize the receive memory counter */
182 atomic_set(&asoc->rmem_alloc, 0);
183
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184 init_waitqueue_head(&asoc->wait);
185
186 asoc->c.my_vtag = sctp_generate_tag(ep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 asoc->c.my_port = ep->base.bind_addr.port;
188
189 asoc->c.initial_tsn = sctp_generate_tsn(ep);
190
191 asoc->next_tsn = asoc->c.initial_tsn;
192
193 asoc->ctsn_ack_point = asoc->next_tsn - 1;
194 asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
195 asoc->highest_sacked = asoc->ctsn_ack_point;
196 asoc->last_cwr_tsn = asoc->ctsn_ack_point;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 /* ADDIP Section 4.1 Asconf Chunk Procedures
199 *
200 * When an endpoint has an ASCONF signaled change to be sent to the
201 * remote endpoint it should do the following:
202 * ...
203 * A2) a serial number should be assigned to the chunk. The serial
204 * number SHOULD be a monotonically increasing number. The serial
205 * numbers SHOULD be initialized at the start of the
206 * association to the same value as the initial TSN.
207 */
208 asoc->addip_serial = asoc->c.initial_tsn;
209
David S. Miller79af02c2005-07-08 21:47:49 -0700210 INIT_LIST_HEAD(&asoc->addip_chunk_list);
Vlad Yasevicha08de642007-12-20 14:11:47 -0800211 INIT_LIST_HEAD(&asoc->asconf_ack_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212
213 /* Make an empty list of remote transport addresses. */
214 INIT_LIST_HEAD(&asoc->peer.transport_addr_list);
215
216 /* RFC 2960 5.1 Normal Establishment of an Association
217 *
218 * After the reception of the first data chunk in an
219 * association the endpoint must immediately respond with a
220 * sack to acknowledge the data chunk. Subsequent
221 * acknowledgements should be done as described in Section
222 * 6.2.
223 *
224 * [We implement this by telling a new association that it
225 * already received one packet.]
226 */
227 asoc->peer.sack_needed = 1;
Neil Horman42448542012-06-30 03:04:26 +0000228 asoc->peer.sack_generation = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
Vlad Yasevich73d9c4f2007-10-24 17:24:26 -0400230 /* Assume that the peer will tell us if he recognizes ASCONF
231 * as part of INIT exchange.
wangweidong9d2c8812013-12-06 09:36:30 +0800232 * The sctp_addip_noauth option is there for backward compatibility
Vlad Yasevich73d9c4f2007-10-24 17:24:26 -0400233 * and will revert old behavior.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 */
Eric W. Biedermane1fc3b12012-08-07 07:29:57 +0000235 if (net->sctp.addip_noauth)
Vlad Yasevich73d9c4f2007-10-24 17:24:26 -0400236 asoc->peer.asconf_capable = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237
238 /* Create an input queue. */
239 sctp_inq_init(&asoc->base.inqueue);
David Howellsc4028952006-11-22 14:57:56 +0000240 sctp_inq_set_th_handler(&asoc->base.inqueue, sctp_assoc_bh_rcv);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
242 /* Create an output queue. */
243 sctp_outq_init(asoc, &asoc->outqueue);
244
245 if (!sctp_ulpq_init(&asoc->ulpq, asoc))
246 goto fail_init;
247
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 /* Assume that peer would support both address types unless we are
249 * told otherwise.
250 */
251 asoc->peer.ipv4_address = 1;
Wei Yongjuna2c39582009-04-07 16:35:11 +0800252 if (asoc->base.sk->sk_family == PF_INET6)
253 asoc->peer.ipv6_address = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 INIT_LIST_HEAD(&asoc->asocs);
255
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 asoc->default_stream = sp->default_stream;
257 asoc->default_ppid = sp->default_ppid;
258 asoc->default_flags = sp->default_flags;
259 asoc->default_context = sp->default_context;
260 asoc->default_timetolive = sp->default_timetolive;
Ivan Skytte Jorgensen6ab792f2006-12-13 16:34:22 -0800261 asoc->default_rcv_context = sp->default_rcv_context;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262
Vlad Yasevicha29a5bd2007-09-16 19:31:35 -0700263 /* AUTH related initializations */
264 INIT_LIST_HEAD(&asoc->endpoint_shared_keys);
265 err = sctp_auth_asoc_copy_shkeys(ep, asoc, gfp);
266 if (err)
267 goto fail_init;
268
269 asoc->active_key_id = ep->active_key_id;
Vlad Yasevicha29a5bd2007-09-16 19:31:35 -0700270
Vlad Yasevicha29a5bd2007-09-16 19:31:35 -0700271 /* Save the hmacs and chunks list into this association */
272 if (ep->auth_hmacs_list)
273 memcpy(asoc->c.auth_hmacs, ep->auth_hmacs_list,
274 ntohs(ep->auth_hmacs_list->param_hdr.length));
275 if (ep->auth_chunk_list)
276 memcpy(asoc->c.auth_chunks, ep->auth_chunk_list,
277 ntohs(ep->auth_chunk_list->param_hdr.length));
278
279 /* Get the AUTH random number for this association */
280 p = (sctp_paramhdr_t *)asoc->c.auth_random;
281 p->type = SCTP_PARAM_RANDOM;
282 p->length = htons(sizeof(sctp_paramhdr_t) + SCTP_AUTH_RANDOM_LENGTH);
283 get_random_bytes(p+1, SCTP_AUTH_RANDOM_LENGTH);
284
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 return asoc;
286
287fail_init:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 sock_put(asoc->base.sk);
Daniel Borkmann2e0c9e72013-06-14 18:24:07 +0200289 sctp_endpoint_put(asoc->ep);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 return NULL;
291}
292
293/* Allocate and initialize a new association */
294struct sctp_association *sctp_association_new(const struct sctp_endpoint *ep,
295 const struct sock *sk,
Alexey Dobriyan3182cd82005-07-11 20:57:47 -0700296 sctp_scope_t scope,
Al Virodd0fc662005-10-07 07:46:04 +0100297 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298{
299 struct sctp_association *asoc;
300
Daniel Borkmann939cfa72013-06-17 11:40:04 +0200301 asoc = kzalloc(sizeof(*asoc), gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 if (!asoc)
303 goto fail;
304
305 if (!sctp_association_init(asoc, ep, sk, scope, gfp))
306 goto fail_init;
307
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 SCTP_DBG_OBJCNT_INC(assoc);
Daniel Borkmannbb333812013-06-28 19:49:40 +0200309
310 pr_debug("Created asoc %p\n", asoc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311
312 return asoc;
313
314fail_init:
315 kfree(asoc);
316fail:
317 return NULL;
318}
319
320/* Free this association if possible. There may still be users, so
321 * the actual deallocation may be delayed.
322 */
323void sctp_association_free(struct sctp_association *asoc)
324{
325 struct sock *sk = asoc->base.sk;
326 struct sctp_transport *transport;
327 struct list_head *pos, *temp;
328 int i;
329
Vlad Yasevichde76e692006-10-30 18:55:11 -0800330 /* Only real associations count against the endpoint, so
331 * don't bother for if this is a temporary association.
332 */
333 if (!asoc->temp) {
334 list_del(&asoc->asocs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335
Vlad Yasevichde76e692006-10-30 18:55:11 -0800336 /* Decrement the backlog value for a TCP-style listening
337 * socket.
338 */
339 if (sctp_style(sk, TCP) && sctp_sstate(sk, LISTENING))
340 sk->sk_ack_backlog--;
341 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342
343 /* Mark as dead, so other users can know this structure is
344 * going away.
345 */
Daniel Borkmann0022d2d2013-04-15 03:27:18 +0000346 asoc->base.dead = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
348 /* Dispose of any data lying around in the outqueue. */
349 sctp_outq_free(&asoc->outqueue);
350
351 /* Dispose of any pending messages for the upper layer. */
352 sctp_ulpq_free(&asoc->ulpq);
353
354 /* Dispose of any pending chunks on the inqueue. */
355 sctp_inq_free(&asoc->base.inqueue);
356
Vlad Yasevich8e1ee182008-10-08 14:18:39 -0700357 sctp_tsnmap_free(&asoc->peer.tsn_map);
358
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 /* Free ssnmap storage. */
360 sctp_ssnmap_free(asoc->ssnmap);
361
362 /* Clean up the bound address list. */
363 sctp_bind_addr_free(&asoc->base.bind_addr);
364
365 /* Do we need to go through all of our timers and
366 * delete them? To be safe we will try to delete all, but we
367 * should be able to go through and make a guess based
368 * on our state.
369 */
370 for (i = SCTP_EVENT_TIMEOUT_NONE; i < SCTP_NUM_TIMEOUT_TYPES; ++i) {
Ying Xue25cc4ae2013-02-03 20:32:57 +0000371 if (del_timer(&asoc->timers[i]))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 sctp_association_put(asoc);
373 }
374
375 /* Free peer's cached cookie. */
Jesper Juhla51482b2005-11-08 09:41:34 -0800376 kfree(asoc->peer.cookie);
Vlad Yasevich730fc3d2007-09-16 19:32:11 -0700377 kfree(asoc->peer.peer_random);
378 kfree(asoc->peer.peer_chunks);
379 kfree(asoc->peer.peer_hmacs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
381 /* Release the transport structures. */
382 list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
383 transport = list_entry(pos, struct sctp_transport, transports);
Thomas Graf45122ca262012-12-06 09:25:05 +0000384 list_del_rcu(pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 sctp_transport_free(transport);
386 }
387
Frank Filz3f7a87d2005-06-20 13:14:57 -0700388 asoc->peer.transport_count = 0;
389
Wei Yongjuna000c012011-05-29 23:23:36 +0000390 sctp_asconf_queue_teardown(asoc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
Michio Honda8a07eb02011-04-26 20:19:36 +0900392 /* Free pending address space being deleted */
393 if (asoc->asconf_addr_del_pending != NULL)
394 kfree(asoc->asconf_addr_del_pending);
395
Vlad Yasevicha29a5bd2007-09-16 19:31:35 -0700396 /* AUTH - Free the endpoint shared keys */
397 sctp_auth_destroy_keys(&asoc->endpoint_shared_keys);
398
399 /* AUTH - Free the association shared key */
400 sctp_auth_key_put(asoc->asoc_shared_key);
401
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 sctp_association_put(asoc);
403}
404
405/* Cleanup and free up an association. */
406static void sctp_association_destroy(struct sctp_association *asoc)
407{
Daniel Borkmannbb333812013-06-28 19:49:40 +0200408 if (unlikely(!asoc->base.dead)) {
409 WARN(1, "Attempt to destroy undead association %p!\n", asoc);
410 return;
411 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
413 sctp_endpoint_put(asoc->ep);
414 sock_put(asoc->base.sk);
415
416 if (asoc->assoc_id != 0) {
417 spin_lock_bh(&sctp_assocs_id_lock);
418 idr_remove(&sctp_assocs_id, asoc->assoc_id);
419 spin_unlock_bh(&sctp_assocs_id_lock);
420 }
421
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700422 WARN_ON(atomic_read(&asoc->rmem_alloc));
Neil Horman049b3ff2005-11-11 16:08:24 -0800423
Daniel Borkmannff2266c2013-04-15 03:27:17 +0000424 kfree(asoc);
425 SCTP_DBG_OBJCNT_DEC(assoc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426}
427
428/* Change the primary destination address for the peer. */
429void sctp_assoc_set_primary(struct sctp_association *asoc,
430 struct sctp_transport *transport)
431{
Vlad Yasevich319fa2a2008-06-16 17:00:29 -0700432 int changeover = 0;
433
434 /* it's a changeover only if we already have a primary path
435 * that we are changing
436 */
437 if (asoc->peer.primary_path != NULL &&
438 asoc->peer.primary_path != transport)
439 changeover = 1 ;
440
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 asoc->peer.primary_path = transport;
442
443 /* Set a default msg_name for events. */
444 memcpy(&asoc->peer.primary_addr, &transport->ipaddr,
445 sizeof(union sctp_addr));
446
447 /* If the primary path is changing, assume that the
448 * user wants to use this new path.
449 */
Sridhar Samudralaad8fec12006-07-21 14:48:50 -0700450 if ((transport->state == SCTP_ACTIVE) ||
451 (transport->state == SCTP_UNKNOWN))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 asoc->peer.active_path = transport;
453
454 /*
455 * SFR-CACC algorithm:
456 * Upon the receipt of a request to change the primary
457 * destination address, on the data structure for the new
458 * primary destination, the sender MUST do the following:
459 *
460 * 1) If CHANGEOVER_ACTIVE is set, then there was a switch
461 * to this destination address earlier. The sender MUST set
462 * CYCLING_CHANGEOVER to indicate that this switch is a
463 * double switch to the same destination address.
Vlad Yaseviche0e9db12009-11-23 15:53:57 -0500464 *
465 * Really, only bother is we have data queued or outstanding on
466 * the association.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 */
Vlad Yaseviche0e9db12009-11-23 15:53:57 -0500468 if (!asoc->outqueue.outstanding_bytes && !asoc->outqueue.out_qlen)
469 return;
470
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 if (transport->cacc.changeover_active)
Vlad Yasevich319fa2a2008-06-16 17:00:29 -0700472 transport->cacc.cycling_changeover = changeover;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
474 /* 2) The sender MUST set CHANGEOVER_ACTIVE to indicate that
475 * a changeover has occurred.
476 */
Vlad Yasevich319fa2a2008-06-16 17:00:29 -0700477 transport->cacc.changeover_active = changeover;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
479 /* 3) The sender MUST store the next TSN to be sent in
480 * next_tsn_at_change.
481 */
482 transport->cacc.next_tsn_at_change = asoc->next_tsn;
483}
484
Frank Filz3f7a87d2005-06-20 13:14:57 -0700485/* Remove a transport from an association. */
486void sctp_assoc_rm_peer(struct sctp_association *asoc,
487 struct sctp_transport *peer)
488{
489 struct list_head *pos;
490 struct sctp_transport *transport;
491
Daniel Borkmannbb333812013-06-28 19:49:40 +0200492 pr_debug("%s: association:%p addr:%pISpc\n",
493 __func__, asoc, &peer->ipaddr.sa);
Frank Filz3f7a87d2005-06-20 13:14:57 -0700494
495 /* If we are to remove the current retran_path, update it
496 * to the next peer before removing this peer from the list.
497 */
498 if (asoc->peer.retran_path == peer)
499 sctp_assoc_update_retran_path(asoc);
500
501 /* Remove this peer from the list. */
Thomas Graf45122ca262012-12-06 09:25:05 +0000502 list_del_rcu(&peer->transports);
Frank Filz3f7a87d2005-06-20 13:14:57 -0700503
504 /* Get the first transport of asoc. */
505 pos = asoc->peer.transport_addr_list.next;
506 transport = list_entry(pos, struct sctp_transport, transports);
507
508 /* Update any entries that match the peer to be deleted. */
509 if (asoc->peer.primary_path == peer)
510 sctp_assoc_set_primary(asoc, transport);
511 if (asoc->peer.active_path == peer)
512 asoc->peer.active_path = transport;
Wei Yongjun9494c7c2011-04-12 15:22:22 +0000513 if (asoc->peer.retran_path == peer)
514 asoc->peer.retran_path = transport;
Frank Filz3f7a87d2005-06-20 13:14:57 -0700515 if (asoc->peer.last_data_from == peer)
516 asoc->peer.last_data_from = transport;
517
518 /* If we remove the transport an INIT was last sent to, set it to
519 * NULL. Combined with the update of the retran path above, this
520 * will cause the next INIT to be sent to the next available
521 * transport, maintaining the cycle.
522 */
523 if (asoc->init_last_sent_to == peer)
524 asoc->init_last_sent_to = NULL;
525
Wei Yongjun6345b192009-04-26 23:13:35 +0800526 /* If we remove the transport an SHUTDOWN was last sent to, set it
527 * to NULL. Combined with the update of the retran path above, this
528 * will cause the next SHUTDOWN to be sent to the next available
529 * transport, maintaining the cycle.
530 */
531 if (asoc->shutdown_last_sent_to == peer)
532 asoc->shutdown_last_sent_to = NULL;
533
Wei Yongjun10a43ce2009-04-26 23:14:42 +0800534 /* If we remove the transport an ASCONF was last sent to, set it to
535 * NULL.
536 */
537 if (asoc->addip_last_asconf &&
538 asoc->addip_last_asconf->transport == peer)
539 asoc->addip_last_asconf->transport = NULL;
540
Vlad Yasevich31b02e12009-09-04 18:21:00 -0400541 /* If we have something on the transmitted list, we have to
542 * save it off. The best place is the active path.
543 */
544 if (!list_empty(&peer->transmitted)) {
545 struct sctp_transport *active = asoc->peer.active_path;
546 struct sctp_chunk *ch;
547
548 /* Reset the transport of each chunk on this list */
549 list_for_each_entry(ch, &peer->transmitted,
550 transmitted_list) {
551 ch->transport = NULL;
552 ch->rtt_in_progress = 0;
553 }
554
555 list_splice_tail_init(&peer->transmitted,
556 &active->transmitted);
557
558 /* Start a T3 timer here in case it wasn't running so
559 * that these migrated packets have a chance to get
wangweidong2bccbad2013-10-26 16:06:30 +0800560 * retransmitted.
Vlad Yasevich31b02e12009-09-04 18:21:00 -0400561 */
562 if (!timer_pending(&active->T3_rtx_timer))
563 if (!mod_timer(&active->T3_rtx_timer,
564 jiffies + active->rto))
565 sctp_transport_hold(active);
566 }
567
Frank Filz3f7a87d2005-06-20 13:14:57 -0700568 asoc->peer.transport_count--;
569
570 sctp_transport_free(peer);
571}
572
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573/* Add a transport address to an association. */
574struct sctp_transport *sctp_assoc_add_peer(struct sctp_association *asoc,
575 const union sctp_addr *addr,
Al Virodd0fc662005-10-07 07:46:04 +0100576 const gfp_t gfp,
Frank Filz3f7a87d2005-06-20 13:14:57 -0700577 const int peer_state)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578{
Eric W. Biederman89bf3452012-08-07 07:26:14 +0000579 struct net *net = sock_net(asoc->base.sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 struct sctp_transport *peer;
581 struct sctp_sock *sp;
582 unsigned short port;
583
584 sp = sctp_sk(asoc->base.sk);
585
586 /* AF_INET and AF_INET6 share common port field. */
Al Viro4bdf4b52006-11-20 17:10:20 -0800587 port = ntohs(addr->v4.sin_port);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
Daniel Borkmannbb333812013-06-28 19:49:40 +0200589 pr_debug("%s: association:%p addr:%pISpc state:%d\n", __func__,
590 asoc, &addr->sa, peer_state);
Frank Filz3f7a87d2005-06-20 13:14:57 -0700591
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 /* Set the port if it has not been set yet. */
593 if (0 == asoc->peer.port)
594 asoc->peer.port = port;
595
596 /* Check to see if this is a duplicate. */
597 peer = sctp_assoc_lookup_paddr(asoc, addr);
Frank Filz3f7a87d2005-06-20 13:14:57 -0700598 if (peer) {
Vlad Yasevichadd52372008-09-18 16:28:27 -0700599 /* An UNKNOWN state is only set on transports added by
600 * user in sctp_connectx() call. Such transports should be
601 * considered CONFIRMED per RFC 4960, Section 5.4.
602 */
Sridhar Samudralaad8fec12006-07-21 14:48:50 -0700603 if (peer->state == SCTP_UNKNOWN) {
Vlad Yasevichadd52372008-09-18 16:28:27 -0700604 peer->state = SCTP_ACTIVE;
Sridhar Samudralaad8fec12006-07-21 14:48:50 -0700605 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 return peer;
Frank Filz3f7a87d2005-06-20 13:14:57 -0700607 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608
Eric W. Biederman89bf3452012-08-07 07:26:14 +0000609 peer = sctp_transport_new(net, addr, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 if (!peer)
611 return NULL;
612
613 sctp_transport_set_owner(peer, asoc);
614
Frank Filz52ccb8e2005-12-22 11:36:46 -0800615 /* Initialize the peer's heartbeat interval based on the
616 * association configured value.
617 */
618 peer->hbinterval = asoc->hbinterval;
619
620 /* Set the path max_retrans. */
621 peer->pathmaxrxt = asoc->pathmaxrxt;
622
wangweidong2bccbad2013-10-26 16:06:30 +0800623 /* And the partial failure retrans threshold */
Neil Horman5aa93bc2012-07-21 07:56:07 +0000624 peer->pf_retrans = asoc->pf_retrans;
625
Frank Filz52ccb8e2005-12-22 11:36:46 -0800626 /* Initialize the peer's SACK delay timeout based on the
627 * association configured value.
628 */
629 peer->sackdelay = asoc->sackdelay;
Wei Yongjund364d922008-05-09 15:13:26 -0700630 peer->sackfreq = asoc->sackfreq;
Frank Filz52ccb8e2005-12-22 11:36:46 -0800631
632 /* Enable/disable heartbeat, SACK delay, and path MTU discovery
633 * based on association setting.
634 */
635 peer->param_flags = asoc->param_flags;
636
Vlad Yasevich8da645e2009-09-04 18:21:01 -0400637 sctp_transport_route(peer, NULL, sp);
638
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 /* Initialize the pmtu of the transport. */
Vlad Yasevich8da645e2009-09-04 18:21:01 -0400640 if (peer->param_flags & SPP_PMTUD_DISABLE) {
641 if (asoc->pathmtu)
642 peer->pathmtu = asoc->pathmtu;
643 else
644 peer->pathmtu = SCTP_DEFAULT_MAXSEGMENT;
645 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
647 /* If this is the first transport addr on this association,
648 * initialize the association PMTU to the peer's PMTU.
649 * If not and the current association PMTU is higher than the new
650 * peer's PMTU, reset the association PMTU to the new peer's PMTU.
651 */
Frank Filz52ccb8e2005-12-22 11:36:46 -0800652 if (asoc->pathmtu)
653 asoc->pathmtu = min_t(int, peer->pathmtu, asoc->pathmtu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 else
Frank Filz52ccb8e2005-12-22 11:36:46 -0800655 asoc->pathmtu = peer->pathmtu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Daniel Borkmannbb333812013-06-28 19:49:40 +0200657 pr_debug("%s: association:%p PMTU set to %d\n", __func__, asoc,
658 asoc->pathmtu);
659
Florian Westphal6d0ccba2008-07-18 23:04:39 -0700660 peer->pmtu_pending = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661
Vlad Yasevichf68b2e02009-09-04 18:21:00 -0400662 asoc->frag_point = sctp_frag_point(asoc, asoc->pathmtu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
664 /* The asoc->peer.port might not be meaningful yet, but
665 * initialize the packet structure anyway.
666 */
667 sctp_packet_init(&peer->packet, peer, asoc->base.bind_addr.port,
668 asoc->peer.port);
669
670 /* 7.2.1 Slow-Start
671 *
672 * o The initial cwnd before DATA transmission or after a sufficiently
673 * long idle period MUST be set to
674 * min(4*MTU, max(2*MTU, 4380 bytes))
675 *
676 * o The initial value of ssthresh MAY be arbitrarily high
677 * (for example, implementations MAY use the size of the
678 * receiver advertised window).
679 */
Frank Filz52ccb8e2005-12-22 11:36:46 -0800680 peer->cwnd = min(4*asoc->pathmtu, max_t(__u32, 2*asoc->pathmtu, 4380));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
682 /* At this point, we may not have the receiver's advertised window,
683 * so initialize ssthresh to the default value and it will be set
684 * later when we process the INIT.
685 */
686 peer->ssthresh = SCTP_DEFAULT_MAXWINDOW;
687
688 peer->partial_bytes_acked = 0;
689 peer->flight_size = 0;
Vlad Yasevich46d5a802009-11-23 15:54:00 -0500690 peer->burst_limited = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 /* Set the transport's RTO.initial value */
693 peer->rto = asoc->rto_initial;
Michele Baldessari196d6752012-12-01 04:49:42 +0000694 sctp_max_rto(asoc, peer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695
Frank Filz3f7a87d2005-06-20 13:14:57 -0700696 /* Set the peer's active state. */
697 peer->state = peer_state;
698
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 /* Attach the remote transport to our asoc. */
Thomas Graf45122ca262012-12-06 09:25:05 +0000700 list_add_tail_rcu(&peer->transports, &asoc->peer.transport_addr_list);
Frank Filz3f7a87d2005-06-20 13:14:57 -0700701 asoc->peer.transport_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702
703 /* If we do not yet have a primary path, set one. */
704 if (!asoc->peer.primary_path) {
705 sctp_assoc_set_primary(asoc, peer);
706 asoc->peer.retran_path = peer;
707 }
708
Vlad Yasevichfbdf5012010-04-30 22:39:26 -0400709 if (asoc->peer.active_path == asoc->peer.retran_path &&
710 peer->state != SCTP_UNCONFIRMED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 asoc->peer.retran_path = peer;
Frank Filz3f7a87d2005-06-20 13:14:57 -0700712 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
714 return peer;
715}
716
717/* Delete a transport address from an association. */
718void sctp_assoc_del_peer(struct sctp_association *asoc,
719 const union sctp_addr *addr)
720{
721 struct list_head *pos;
722 struct list_head *temp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 struct sctp_transport *transport;
724
725 list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
726 transport = list_entry(pos, struct sctp_transport, transports);
727 if (sctp_cmp_addr_exact(addr, &transport->ipaddr)) {
Frank Filz3f7a87d2005-06-20 13:14:57 -0700728 /* Do book keeping for removing the peer and free it. */
729 sctp_assoc_rm_peer(asoc, transport);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 break;
731 }
732 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733}
734
735/* Lookup a transport by address. */
736struct sctp_transport *sctp_assoc_lookup_paddr(
737 const struct sctp_association *asoc,
738 const union sctp_addr *address)
739{
740 struct sctp_transport *t;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741
742 /* Cycle through all transports searching for a peer address. */
743
Robert P. J. Day9dbc15f2008-04-12 18:54:24 -0700744 list_for_each_entry(t, &asoc->peer.transport_addr_list,
745 transports) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 if (sctp_cmp_addr_exact(address, &t->ipaddr))
747 return t;
748 }
749
750 return NULL;
751}
752
Vlad Yasevich42e30bf2007-12-20 14:08:56 -0800753/* Remove all transports except a give one */
754void sctp_assoc_del_nonprimary_peers(struct sctp_association *asoc,
755 struct sctp_transport *primary)
756{
757 struct sctp_transport *temp;
758 struct sctp_transport *t;
759
760 list_for_each_entry_safe(t, temp, &asoc->peer.transport_addr_list,
761 transports) {
762 /* if the current transport is not the primary one, delete it */
763 if (t != primary)
764 sctp_assoc_rm_peer(asoc, t);
765 }
Vlad Yasevich42e30bf2007-12-20 14:08:56 -0800766}
767
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768/* Engage in transport control operations.
769 * Mark the transport up or down and send a notification to the user.
770 * Select and update the new active and retran paths.
771 */
772void sctp_assoc_control_transport(struct sctp_association *asoc,
773 struct sctp_transport *transport,
774 sctp_transport_cmd_t command,
775 sctp_sn_error_t error)
776{
777 struct sctp_transport *t = NULL;
778 struct sctp_transport *first;
779 struct sctp_transport *second;
780 struct sctp_ulpevent *event;
Al Viro0906e202006-11-20 17:03:01 -0800781 struct sockaddr_storage addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 int spc_state = 0;
Neil Horman5aa93bc2012-07-21 07:56:07 +0000783 bool ulp_notify = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784
785 /* Record the transition on the transport. */
786 switch (command) {
787 case SCTP_TRANSPORT_UP:
Vlad Yasevich1ae4114d2007-03-23 11:32:26 -0700788 /* If we are moving from UNCONFIRMED state due
789 * to heartbeat success, report the SCTP_ADDR_CONFIRMED
790 * state to the user, otherwise report SCTP_ADDR_AVAILABLE.
791 */
792 if (SCTP_UNCONFIRMED == transport->state &&
793 SCTP_HEARTBEAT_SUCCESS == error)
794 spc_state = SCTP_ADDR_CONFIRMED;
795 else
796 spc_state = SCTP_ADDR_AVAILABLE;
Neil Horman5aa93bc2012-07-21 07:56:07 +0000797 /* Don't inform ULP about transition from PF to
Daniel Borkmannac4f9592013-08-09 15:09:08 +0200798 * active state and set cwnd to 1 MTU, see SCTP
Neil Horman5aa93bc2012-07-21 07:56:07 +0000799 * Quick failover draft section 5.1, point 5
800 */
801 if (transport->state == SCTP_PF) {
802 ulp_notify = false;
Daniel Borkmannac4f9592013-08-09 15:09:08 +0200803 transport->cwnd = asoc->pathmtu;
Neil Horman5aa93bc2012-07-21 07:56:07 +0000804 }
Frank Filz3f7a87d2005-06-20 13:14:57 -0700805 transport->state = SCTP_ACTIVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 break;
807
808 case SCTP_TRANSPORT_DOWN:
Vlad Yasevich40187882009-06-23 11:28:05 -0400809 /* If the transport was never confirmed, do not transition it
810 * to inactive state. Also, release the cached route since
811 * there may be a better route next time.
Vlad Yasevichcc756892007-08-24 19:30:25 +0900812 */
813 if (transport->state != SCTP_UNCONFIRMED)
814 transport->state = SCTP_INACTIVE;
Vlad Yasevich40187882009-06-23 11:28:05 -0400815 else {
816 dst_release(transport->dst);
817 transport->dst = NULL;
818 }
Vlad Yasevichcc756892007-08-24 19:30:25 +0900819
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 spc_state = SCTP_ADDR_UNREACHABLE;
821 break;
822
Neil Horman5aa93bc2012-07-21 07:56:07 +0000823 case SCTP_TRANSPORT_PF:
824 transport->state = SCTP_PF;
825 ulp_notify = false;
826 break;
827
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 default:
829 return;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700830 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831
832 /* Generate and send a SCTP_PEER_ADDR_CHANGE notification to the
833 * user.
834 */
Neil Horman5aa93bc2012-07-21 07:56:07 +0000835 if (ulp_notify) {
836 memset(&addr, 0, sizeof(struct sockaddr_storage));
837 memcpy(&addr, &transport->ipaddr,
838 transport->af_specific->sockaddr_len);
839 event = sctp_ulpevent_make_peer_addr_change(asoc, &addr,
840 0, spc_state, error, GFP_ATOMIC);
841 if (event)
842 sctp_ulpq_tail_event(&asoc->ulpq, event);
843 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844
845 /* Select new active and retran paths. */
846
847 /* Look for the two most recently used active transports.
848 *
849 * This code produces the wrong ordering whenever jiffies
850 * rolls over, but we still get usable transports, so we don't
851 * worry about it.
852 */
853 first = NULL; second = NULL;
854
Robert P. J. Day9dbc15f2008-04-12 18:54:24 -0700855 list_for_each_entry(t, &asoc->peer.transport_addr_list,
856 transports) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
Sridhar Samudralaad8fec12006-07-21 14:48:50 -0700858 if ((t->state == SCTP_INACTIVE) ||
Neil Horman5aa93bc2012-07-21 07:56:07 +0000859 (t->state == SCTP_UNCONFIRMED) ||
860 (t->state == SCTP_PF))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 continue;
862 if (!first || t->last_time_heard > first->last_time_heard) {
863 second = first;
864 first = t;
Chang Xiangzhongd30a58b2013-11-14 00:58:26 +0100865 } else if (!second ||
866 t->last_time_heard > second->last_time_heard)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 second = t;
868 }
869
870 /* RFC 2960 6.4 Multi-Homed SCTP Endpoints
871 *
872 * By default, an endpoint should always transmit to the
873 * primary path, unless the SCTP user explicitly specifies the
874 * destination transport address (and possibly source
875 * transport address) to use.
876 *
877 * [If the primary is active but not most recent, bump the most
878 * recently used transport.]
879 */
Sridhar Samudralaad8fec12006-07-21 14:48:50 -0700880 if (((asoc->peer.primary_path->state == SCTP_ACTIVE) ||
881 (asoc->peer.primary_path->state == SCTP_UNKNOWN)) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 first != asoc->peer.primary_path) {
883 second = first;
884 first = asoc->peer.primary_path;
885 }
886
Chang Xiangzhongd30a58b2013-11-14 00:58:26 +0100887 if (!second)
888 second = first;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 /* If we failed to find a usable transport, just camp on the
890 * primary, even if it is inactive.
891 */
892 if (!first) {
893 first = asoc->peer.primary_path;
894 second = asoc->peer.primary_path;
895 }
896
897 /* Set the active and retran transports. */
898 asoc->peer.active_path = first;
899 asoc->peer.retran_path = second;
900}
901
902/* Hold a reference to an association. */
903void sctp_association_hold(struct sctp_association *asoc)
904{
905 atomic_inc(&asoc->base.refcnt);
906}
907
908/* Release a reference to an association and cleanup
909 * if there are no more references.
910 */
911void sctp_association_put(struct sctp_association *asoc)
912{
913 if (atomic_dec_and_test(&asoc->base.refcnt))
914 sctp_association_destroy(asoc);
915}
916
917/* Allocate the next TSN, Transmission Sequence Number, for the given
918 * association.
919 */
920__u32 sctp_association_get_next_tsn(struct sctp_association *asoc)
921{
922 /* From Section 1.6 Serial Number Arithmetic:
923 * Transmission Sequence Numbers wrap around when they reach
924 * 2**32 - 1. That is, the next TSN a DATA chunk MUST use
925 * after transmitting TSN = 2*32 - 1 is TSN = 0.
926 */
927 __u32 retval = asoc->next_tsn;
928 asoc->next_tsn++;
929 asoc->unack_data++;
930
931 return retval;
932}
933
934/* Compare two addresses to see if they match. Wildcard addresses
935 * only match themselves.
936 */
937int sctp_cmp_addr_exact(const union sctp_addr *ss1,
938 const union sctp_addr *ss2)
939{
940 struct sctp_af *af;
941
942 af = sctp_get_af_specific(ss1->sa.sa_family);
943 if (unlikely(!af))
944 return 0;
945
946 return af->cmp_addr(ss1, ss2);
947}
948
949/* Return an ecne chunk to get prepended to a packet.
950 * Note: We are sly and return a shared, prealloced chunk. FIXME:
951 * No we don't, but we could/should.
952 */
953struct sctp_chunk *sctp_get_ecne_prepend(struct sctp_association *asoc)
954{
wangweidong8b7318d2013-12-06 09:36:28 +0800955 if (!asoc->need_ecne)
956 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
958 /* Send ECNE if needed.
959 * Not being able to allocate a chunk here is not deadly.
960 */
wangweidong8b7318d2013-12-06 09:36:28 +0800961 return sctp_make_ecne(asoc, asoc->last_ecne_tsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962}
963
964/*
965 * Find which transport this TSN was sent on.
966 */
967struct sctp_transport *sctp_assoc_lookup_tsn(struct sctp_association *asoc,
968 __u32 tsn)
969{
970 struct sctp_transport *active;
971 struct sctp_transport *match;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 struct sctp_transport *transport;
973 struct sctp_chunk *chunk;
Al Virodbc16db2006-11-20 17:01:42 -0800974 __be32 key = htonl(tsn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
976 match = NULL;
977
978 /*
979 * FIXME: In general, find a more efficient data structure for
980 * searching.
981 */
982
983 /*
984 * The general strategy is to search each transport's transmitted
985 * list. Return which transport this TSN lives on.
986 *
987 * Let's be hopeful and check the active_path first.
988 * Another optimization would be to know if there is only one
989 * outbound path and not have to look for the TSN at all.
990 *
991 */
992
993 active = asoc->peer.active_path;
994
Robert P. J. Day9dbc15f2008-04-12 18:54:24 -0700995 list_for_each_entry(chunk, &active->transmitted,
996 transmitted_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
998 if (key == chunk->subh.data_hdr->tsn) {
999 match = active;
1000 goto out;
1001 }
1002 }
1003
1004 /* If not found, go search all the other transports. */
Robert P. J. Day9dbc15f2008-04-12 18:54:24 -07001005 list_for_each_entry(transport, &asoc->peer.transport_addr_list,
1006 transports) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
1008 if (transport == active)
Xufeng Zhang2317f442013-03-07 21:39:37 +00001009 continue;
Robert P. J. Day9dbc15f2008-04-12 18:54:24 -07001010 list_for_each_entry(chunk, &transport->transmitted,
1011 transmitted_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 if (key == chunk->subh.data_hdr->tsn) {
1013 match = transport;
1014 goto out;
1015 }
1016 }
1017 }
1018out:
1019 return match;
1020}
1021
1022/* Is this the association we are looking for? */
1023struct sctp_transport *sctp_assoc_is_match(struct sctp_association *asoc,
Eric W. Biederman4110cc22012-08-06 08:41:13 +00001024 struct net *net,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 const union sctp_addr *laddr,
1026 const union sctp_addr *paddr)
1027{
1028 struct sctp_transport *transport;
1029
Al Viroe2fcced2006-11-20 17:08:41 -08001030 if ((htons(asoc->base.bind_addr.port) == laddr->v4.sin_port) &&
Eric W. Biederman4110cc22012-08-06 08:41:13 +00001031 (htons(asoc->peer.port) == paddr->v4.sin_port) &&
1032 net_eq(sock_net(asoc->base.sk), net)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 transport = sctp_assoc_lookup_paddr(asoc, paddr);
1034 if (!transport)
1035 goto out;
1036
1037 if (sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
1038 sctp_sk(asoc->base.sk)))
1039 goto out;
1040 }
1041 transport = NULL;
1042
1043out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 return transport;
1045}
1046
1047/* Do delayed input processing. This is scheduled by sctp_rcv(). */
David Howellsc4028952006-11-22 14:57:56 +00001048static void sctp_assoc_bh_rcv(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049{
David Howellsc4028952006-11-22 14:57:56 +00001050 struct sctp_association *asoc =
1051 container_of(work, struct sctp_association,
1052 base.inqueue.immediate);
Eric W. Biederman55e26eb2012-08-07 07:25:24 +00001053 struct net *net = sock_net(asoc->base.sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 struct sctp_endpoint *ep;
1055 struct sctp_chunk *chunk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 struct sctp_inq *inqueue;
1057 int state;
1058 sctp_subtype_t subtype;
1059 int error = 0;
1060
1061 /* The association should be held so we should be safe. */
1062 ep = asoc->ep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
1064 inqueue = &asoc->base.inqueue;
1065 sctp_association_hold(asoc);
1066 while (NULL != (chunk = sctp_inq_pop(inqueue))) {
1067 state = asoc->state;
1068 subtype = SCTP_ST_CHUNK(chunk->chunk_hdr->type);
1069
Vlad Yasevichbbd0d592007-10-03 17:51:34 -07001070 /* SCTP-AUTH, Section 6.3:
1071 * The receiver has a list of chunk types which it expects
1072 * to be received only after an AUTH-chunk. This list has
1073 * been sent to the peer during the association setup. It
1074 * MUST silently discard these chunks if they are not placed
1075 * after an AUTH chunk in the packet.
1076 */
1077 if (sctp_auth_recv_cid(subtype.chunk, asoc) && !chunk->auth)
1078 continue;
1079
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 /* Remember where the last DATA chunk came from so we
1081 * know where to send the SACK.
1082 */
1083 if (sctp_chunk_is_data(chunk))
1084 asoc->peer.last_data_from = chunk->transport;
Michele Baldessari196d6752012-12-01 04:49:42 +00001085 else {
Eric W. Biederman55e26eb2012-08-07 07:25:24 +00001086 SCTP_INC_STATS(net, SCTP_MIB_INCTRLCHUNKS);
Michele Baldessari196d6752012-12-01 04:49:42 +00001087 asoc->stats.ictrlchunks++;
1088 if (chunk->chunk_hdr->type == SCTP_CID_SACK)
1089 asoc->stats.isacks++;
1090 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091
1092 if (chunk->transport)
1093 chunk->transport->last_time_heard = jiffies;
1094
1095 /* Run through the state machine. */
Eric W. Biederman55e26eb2012-08-07 07:25:24 +00001096 error = sctp_do_sm(net, SCTP_EVENT_T_CHUNK, subtype,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 state, ep, asoc, chunk, GFP_ATOMIC);
1098
1099 /* Check to see if the association is freed in response to
1100 * the incoming chunk. If so, get out of the while loop.
1101 */
1102 if (asoc->base.dead)
1103 break;
1104
1105 /* If there is an error on chunk, discard this packet. */
1106 if (error && chunk)
1107 chunk->pdiscard = 1;
1108 }
1109 sctp_association_put(asoc);
1110}
1111
1112/* This routine moves an association from its old sk to a new sk. */
1113void sctp_assoc_migrate(struct sctp_association *assoc, struct sock *newsk)
1114{
1115 struct sctp_sock *newsp = sctp_sk(newsk);
1116 struct sock *oldsk = assoc->base.sk;
1117
1118 /* Delete the association from the old endpoint's list of
1119 * associations.
1120 */
1121 list_del_init(&assoc->asocs);
1122
1123 /* Decrement the backlog value for a TCP-style socket. */
1124 if (sctp_style(oldsk, TCP))
1125 oldsk->sk_ack_backlog--;
1126
1127 /* Release references to the old endpoint and the sock. */
1128 sctp_endpoint_put(assoc->ep);
1129 sock_put(assoc->base.sk);
1130
1131 /* Get a reference to the new endpoint. */
1132 assoc->ep = newsp->ep;
1133 sctp_endpoint_hold(assoc->ep);
1134
1135 /* Get a reference to the new sock. */
1136 assoc->base.sk = newsk;
1137 sock_hold(assoc->base.sk);
1138
1139 /* Add the association to the new endpoint's list of associations. */
1140 sctp_endpoint_add_asoc(newsp->ep, assoc);
1141}
1142
1143/* Update an association (possibly from unexpected COOKIE-ECHO processing). */
1144void sctp_assoc_update(struct sctp_association *asoc,
1145 struct sctp_association *new)
1146{
1147 struct sctp_transport *trans;
1148 struct list_head *pos, *temp;
1149
1150 /* Copy in new parameters of peer. */
1151 asoc->c = new->c;
1152 asoc->peer.rwnd = new->peer.rwnd;
1153 asoc->peer.sack_needed = new->peer.sack_needed;
1154 asoc->peer.i = new->peer.i;
Vlad Yasevich8e1ee182008-10-08 14:18:39 -07001155 sctp_tsnmap_init(&asoc->peer.tsn_map, SCTP_TSN_MAP_INITIAL,
1156 asoc->peer.i.initial_tsn, GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157
1158 /* Remove any peer addresses not present in the new association. */
1159 list_for_each_safe(pos, temp, &asoc->peer.transport_addr_list) {
1160 trans = list_entry(pos, struct sctp_transport, transports);
Vlad Yasevich0c427492010-04-28 08:47:19 +00001161 if (!sctp_assoc_lookup_paddr(new, &trans->ipaddr)) {
1162 sctp_assoc_rm_peer(asoc, trans);
1163 continue;
1164 }
Vlad Yasevich749bf922007-03-19 17:02:30 -07001165
1166 if (asoc->state >= SCTP_STATE_ESTABLISHED)
1167 sctp_transport_reset(trans);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 }
1169
1170 /* If the case is A (association restart), use
1171 * initial_tsn as next_tsn. If the case is B, use
1172 * current next_tsn in case data sent to peer
1173 * has been discarded and needs retransmission.
1174 */
1175 if (asoc->state >= SCTP_STATE_ESTABLISHED) {
1176 asoc->next_tsn = new->next_tsn;
1177 asoc->ctsn_ack_point = new->ctsn_ack_point;
1178 asoc->adv_peer_ack_point = new->adv_peer_ack_point;
1179
1180 /* Reinitialize SSN for both local streams
1181 * and peer's streams.
1182 */
1183 sctp_ssnmap_clear(asoc->ssnmap);
1184
Vlad Yasevich0b58a812007-03-19 17:01:17 -07001185 /* Flush the ULP reassembly and ordered queue.
1186 * Any data there will now be stale and will
1187 * cause problems.
1188 */
1189 sctp_ulpq_flush(&asoc->ulpq);
1190
Vlad Yasevich749bf922007-03-19 17:02:30 -07001191 /* reset the overall association error count so
1192 * that the restarted association doesn't get torn
1193 * down on the next retransmission timer.
1194 */
1195 asoc->overall_error_count = 0;
1196
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 } else {
1198 /* Add any peer addresses from the new association. */
Robert P. J. Day9dbc15f2008-04-12 18:54:24 -07001199 list_for_each_entry(trans, &new->peer.transport_addr_list,
1200 transports) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 if (!sctp_assoc_lookup_paddr(asoc, &trans->ipaddr))
1202 sctp_assoc_add_peer(asoc, &trans->ipaddr,
Sridhar Samudralaad8fec12006-07-21 14:48:50 -07001203 GFP_ATOMIC, trans->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 }
1205
1206 asoc->ctsn_ack_point = asoc->next_tsn - 1;
1207 asoc->adv_peer_ack_point = asoc->ctsn_ack_point;
1208 if (!asoc->ssnmap) {
1209 /* Move the ssnmap. */
1210 asoc->ssnmap = new->ssnmap;
1211 new->ssnmap = NULL;
1212 }
Vlad Yasevich07d93962007-05-04 13:55:27 -07001213
1214 if (!asoc->assoc_id) {
1215 /* get a new association id since we don't have one
1216 * yet.
1217 */
1218 sctp_assoc_set_id(asoc, GFP_ATOMIC);
1219 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 }
Vlad Yasevicha29a5bd2007-09-16 19:31:35 -07001221
wangweidong9d2c8812013-12-06 09:36:30 +08001222 /* SCTP-AUTH: Save the peer parameters from the new associations
Vlad Yasevich730fc3d2007-09-16 19:32:11 -07001223 * and also move the association shared keys over
1224 */
1225 kfree(asoc->peer.peer_random);
1226 asoc->peer.peer_random = new->peer.peer_random;
1227 new->peer.peer_random = NULL;
1228
1229 kfree(asoc->peer.peer_chunks);
1230 asoc->peer.peer_chunks = new->peer.peer_chunks;
1231 new->peer.peer_chunks = NULL;
1232
1233 kfree(asoc->peer.peer_hmacs);
1234 asoc->peer.peer_hmacs = new->peer.peer_hmacs;
1235 new->peer.peer_hmacs = NULL;
1236
1237 sctp_auth_key_put(asoc->asoc_shared_key);
1238 sctp_auth_asoc_init_active_key(asoc, GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239}
1240
1241/* Update the retran path for sending a retransmitted packet.
Daniel Borkmann4c47af42014-02-20 20:51:06 +01001242 * See also RFC4960, 6.4. Multi-Homed SCTP Endpoints:
1243 *
1244 * When there is outbound data to send and the primary path
1245 * becomes inactive (e.g., due to failures), or where the
1246 * SCTP user explicitly requests to send data to an
1247 * inactive destination transport address, before reporting
1248 * an error to its ULP, the SCTP endpoint should try to send
1249 * the data to an alternate active destination transport
1250 * address if one exists.
1251 *
1252 * When retransmitting data that timed out, if the endpoint
1253 * is multihomed, it should consider each source-destination
1254 * address pair in its retransmission selection policy.
1255 * When retransmitting timed-out data, the endpoint should
1256 * attempt to pick the most divergent source-destination
1257 * pair from the original source-destination pair to which
1258 * the packet was transmitted.
1259 *
1260 * Note: Rules for picking the most divergent source-destination
1261 * pair are an implementation decision and are not specified
1262 * within this document.
1263 *
1264 * Our basic strategy is to round-robin transports in priorities
1265 * according to sctp_state_prio_map[] e.g., if no such
1266 * transport with state SCTP_ACTIVE exists, round-robin through
1267 * SCTP_UNKNOWN, etc. You get the picture.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 */
Daniel Borkmann4c47af42014-02-20 20:51:06 +01001269static const u8 sctp_trans_state_to_prio_map[] = {
1270 [SCTP_ACTIVE] = 3, /* best case */
1271 [SCTP_UNKNOWN] = 2,
1272 [SCTP_PF] = 1,
1273 [SCTP_INACTIVE] = 0, /* worst case */
1274};
1275
1276static u8 sctp_trans_score(const struct sctp_transport *trans)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277{
Daniel Borkmann4c47af42014-02-20 20:51:06 +01001278 return sctp_trans_state_to_prio_map[trans->state];
Frank Filz3f7a87d2005-06-20 13:14:57 -07001279}
1280
Daniel Borkmann4c47af42014-02-20 20:51:06 +01001281static struct sctp_transport *sctp_trans_elect_best(struct sctp_transport *curr,
1282 struct sctp_transport *best)
1283{
1284 if (best == NULL)
1285 return curr;
1286
1287 return sctp_trans_score(curr) > sctp_trans_score(best) ? curr : best;
1288}
1289
1290void sctp_assoc_update_retran_path(struct sctp_association *asoc)
1291{
1292 struct sctp_transport *trans = asoc->peer.retran_path;
1293 struct sctp_transport *trans_next = NULL;
1294
1295 /* We're done as we only have the one and only path. */
1296 if (asoc->peer.transport_count == 1)
1297 return;
1298 /* If active_path and retran_path are the same and active,
1299 * then this is the only active path. Use it.
1300 */
1301 if (asoc->peer.active_path == asoc->peer.retran_path &&
1302 asoc->peer.active_path->state == SCTP_ACTIVE)
1303 return;
1304
1305 /* Iterate from retran_path's successor back to retran_path. */
1306 for (trans = list_next_entry(trans, transports); 1;
1307 trans = list_next_entry(trans, transports)) {
1308 /* Manually skip the head element. */
1309 if (&trans->transports == &asoc->peer.transport_addr_list)
1310 continue;
1311 if (trans->state == SCTP_UNCONFIRMED)
1312 continue;
1313 trans_next = sctp_trans_elect_best(trans, trans_next);
1314 /* Active is good enough for immediate return. */
1315 if (trans_next->state == SCTP_ACTIVE)
1316 break;
1317 /* We've reached the end, time to update path. */
1318 if (trans == asoc->peer.retran_path)
1319 break;
1320 }
1321
Daniel Borkmann433131b2014-03-13 14:45:26 +01001322 asoc->peer.retran_path = trans_next;
Daniel Borkmann4c47af42014-02-20 20:51:06 +01001323
1324 pr_debug("%s: association:%p updated new path to addr:%pISpc\n",
1325 __func__, asoc, &asoc->peer.retran_path->ipaddr.sa);
1326}
1327
1328struct sctp_transport *
1329sctp_assoc_choose_alter_transport(struct sctp_association *asoc,
1330 struct sctp_transport *last_sent_to)
Frank Filz3f7a87d2005-06-20 13:14:57 -07001331{
Wei Yongjun9919b452009-05-12 21:52:51 +08001332 /* If this is the first time packet is sent, use the active path,
1333 * else use the retran path. If the last packet was sent over the
Frank Filz3f7a87d2005-06-20 13:14:57 -07001334 * retran path, update the retran path and use it.
1335 */
Daniel Borkmann4c47af42014-02-20 20:51:06 +01001336 if (last_sent_to == NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 return asoc->peer.active_path;
Daniel Borkmann4c47af42014-02-20 20:51:06 +01001338 } else {
Wei Yongjun9919b452009-05-12 21:52:51 +08001339 if (last_sent_to == asoc->peer.retran_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 sctp_assoc_update_retran_path(asoc);
Daniel Borkmann4c47af42014-02-20 20:51:06 +01001341
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 return asoc->peer.retran_path;
1343 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344}
1345
1346/* Update the association's pmtu and frag_point by going through all the
1347 * transports. This routine is called when a transport's PMTU has changed.
1348 */
David S. Miller02f3d4c2012-07-16 03:57:14 -07001349void sctp_assoc_sync_pmtu(struct sock *sk, struct sctp_association *asoc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350{
1351 struct sctp_transport *t;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 __u32 pmtu = 0;
1353
1354 if (!asoc)
1355 return;
1356
1357 /* Get the lowest pmtu of all the transports. */
Robert P. J. Day9dbc15f2008-04-12 18:54:24 -07001358 list_for_each_entry(t, &asoc->peer.transport_addr_list,
1359 transports) {
Vlad Yasevich8a479492007-06-07 14:21:05 -04001360 if (t->pmtu_pending && t->dst) {
David S. Miller02f3d4c2012-07-16 03:57:14 -07001361 sctp_transport_update_pmtu(sk, t, dst_mtu(t->dst));
Vlad Yasevich8a479492007-06-07 14:21:05 -04001362 t->pmtu_pending = 0;
1363 }
Frank Filz52ccb8e2005-12-22 11:36:46 -08001364 if (!pmtu || (t->pathmtu < pmtu))
1365 pmtu = t->pathmtu;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 }
1367
1368 if (pmtu) {
Frank Filz52ccb8e2005-12-22 11:36:46 -08001369 asoc->pathmtu = pmtu;
Vlad Yasevichf68b2e02009-09-04 18:21:00 -04001370 asoc->frag_point = sctp_frag_point(asoc, pmtu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 }
1372
Daniel Borkmannbb333812013-06-28 19:49:40 +02001373 pr_debug("%s: asoc:%p, pmtu:%d, frag_point:%d\n", __func__, asoc,
1374 asoc->pathmtu, asoc->frag_point);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375}
1376
1377/* Should we send a SACK to update our peer? */
wangweidongce4a03d2013-12-06 09:36:29 +08001378static inline bool sctp_peer_needs_update(struct sctp_association *asoc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379{
Eric W. Biedermane1fc3b12012-08-07 07:29:57 +00001380 struct net *net = sock_net(asoc->base.sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 switch (asoc->state) {
1382 case SCTP_STATE_ESTABLISHED:
1383 case SCTP_STATE_SHUTDOWN_PENDING:
1384 case SCTP_STATE_SHUTDOWN_RECEIVED:
1385 case SCTP_STATE_SHUTDOWN_SENT:
1386 if ((asoc->rwnd > asoc->a_rwnd) &&
Vlad Yasevich90f2f532009-11-23 15:53:57 -05001387 ((asoc->rwnd - asoc->a_rwnd) >= max_t(__u32,
Eric W. Biedermane1fc3b12012-08-07 07:29:57 +00001388 (asoc->base.sk->sk_rcvbuf >> net->sctp.rwnd_upd_shift),
Vlad Yasevich90f2f532009-11-23 15:53:57 -05001389 asoc->pathmtu)))
wangweidongce4a03d2013-12-06 09:36:29 +08001390 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 break;
1392 default:
1393 break;
1394 }
wangweidongce4a03d2013-12-06 09:36:29 +08001395 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396}
1397
Daniel Borkmann362d5202014-04-14 21:45:17 +02001398/* Increase asoc's rwnd by len and send any window update SACK if needed. */
1399void sctp_assoc_rwnd_increase(struct sctp_association *asoc, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400{
1401 struct sctp_chunk *sack;
1402 struct timer_list *timer;
1403
Daniel Borkmann362d5202014-04-14 21:45:17 +02001404 if (asoc->rwnd_over) {
1405 if (asoc->rwnd_over >= len) {
1406 asoc->rwnd_over -= len;
1407 } else {
1408 asoc->rwnd += (len - asoc->rwnd_over);
1409 asoc->rwnd_over = 0;
1410 }
1411 } else {
1412 asoc->rwnd += len;
1413 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
Daniel Borkmann362d5202014-04-14 21:45:17 +02001415 /* If we had window pressure, start recovering it
1416 * once our rwnd had reached the accumulated pressure
1417 * threshold. The idea is to recover slowly, but up
1418 * to the initial advertised window.
1419 */
1420 if (asoc->rwnd_press && asoc->rwnd >= asoc->rwnd_press) {
1421 int change = min(asoc->pathmtu, asoc->rwnd_press);
1422 asoc->rwnd += change;
1423 asoc->rwnd_press -= change;
1424 }
Vlad Yasevich4d3c46e2009-09-04 18:20:59 -04001425
Daniel Borkmann362d5202014-04-14 21:45:17 +02001426 pr_debug("%s: asoc:%p rwnd increased by %d to (%u, %u) - %u\n",
1427 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
1428 asoc->a_rwnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429
1430 /* Send a window update SACK if the rwnd has increased by at least the
1431 * minimum of the association's PMTU and half of the receive buffer.
1432 * The algorithm used is similar to the one described in
1433 * Section 4.2.3.3 of RFC 1122.
1434 */
Daniel Borkmann362d5202014-04-14 21:45:17 +02001435 if (sctp_peer_needs_update(asoc)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 asoc->a_rwnd = asoc->rwnd;
Daniel Borkmannbb333812013-06-28 19:49:40 +02001437
1438 pr_debug("%s: sending window update SACK- asoc:%p rwnd:%u "
1439 "a_rwnd:%u\n", __func__, asoc, asoc->rwnd,
1440 asoc->a_rwnd);
1441
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 sack = sctp_make_sack(asoc);
1443 if (!sack)
1444 return;
1445
1446 asoc->peer.sack_needed = 0;
1447
1448 sctp_outq_tail(&asoc->outqueue, sack);
1449
1450 /* Stop the SACK timer. */
1451 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
Ying Xue25cc4ae2013-02-03 20:32:57 +00001452 if (del_timer(timer))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 sctp_association_put(asoc);
1454 }
1455}
1456
Daniel Borkmann362d5202014-04-14 21:45:17 +02001457/* Decrease asoc's rwnd by len. */
1458void sctp_assoc_rwnd_decrease(struct sctp_association *asoc, unsigned int len)
1459{
1460 int rx_count;
1461 int over = 0;
1462
1463 if (unlikely(!asoc->rwnd || asoc->rwnd_over))
1464 pr_debug("%s: association:%p has asoc->rwnd:%u, "
1465 "asoc->rwnd_over:%u!\n", __func__, asoc,
1466 asoc->rwnd, asoc->rwnd_over);
1467
1468 if (asoc->ep->rcvbuf_policy)
1469 rx_count = atomic_read(&asoc->rmem_alloc);
1470 else
1471 rx_count = atomic_read(&asoc->base.sk->sk_rmem_alloc);
1472
1473 /* If we've reached or overflowed our receive buffer, announce
1474 * a 0 rwnd if rwnd would still be positive. Store the
1475 * the potential pressure overflow so that the window can be restored
1476 * back to original value.
1477 */
1478 if (rx_count >= asoc->base.sk->sk_rcvbuf)
1479 over = 1;
1480
1481 if (asoc->rwnd >= len) {
1482 asoc->rwnd -= len;
1483 if (over) {
1484 asoc->rwnd_press += asoc->rwnd;
1485 asoc->rwnd = 0;
1486 }
1487 } else {
1488 asoc->rwnd_over = len - asoc->rwnd;
1489 asoc->rwnd = 0;
1490 }
1491
1492 pr_debug("%s: asoc:%p rwnd decreased by %d to (%u, %u, %u)\n",
1493 __func__, asoc, len, asoc->rwnd, asoc->rwnd_over,
1494 asoc->rwnd_press);
1495}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496
1497/* Build the bind address list for the association based on info from the
1498 * local endpoint and the remote peer.
1499 */
Alexey Dobriyan3182cd82005-07-11 20:57:47 -07001500int sctp_assoc_set_bind_addr_from_ep(struct sctp_association *asoc,
Vlad Yasevich409b95a2009-11-10 08:57:34 +00001501 sctp_scope_t scope, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 int flags;
1504
1505 /* Use scoping rules to determine the subset of addresses from
1506 * the endpoint.
1507 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 flags = (PF_INET6 == asoc->base.sk->sk_family) ? SCTP_ADDR6_ALLOWED : 0;
1509 if (asoc->peer.ipv4_address)
1510 flags |= SCTP_ADDR4_PEERSUPP;
1511 if (asoc->peer.ipv6_address)
1512 flags |= SCTP_ADDR6_PEERSUPP;
1513
Eric W. Biederman4db67e82012-08-06 08:42:04 +00001514 return sctp_bind_addr_copy(sock_net(asoc->base.sk),
1515 &asoc->base.bind_addr,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 &asoc->ep->base.bind_addr,
1517 scope, gfp, flags);
1518}
1519
1520/* Build the association's bind address list from the cookie. */
1521int sctp_assoc_set_bind_addr_from_cookie(struct sctp_association *asoc,
Alexey Dobriyan3182cd82005-07-11 20:57:47 -07001522 struct sctp_cookie *cookie,
Al Virodd0fc662005-10-07 07:46:04 +01001523 gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524{
1525 int var_size2 = ntohs(cookie->peer_init->chunk_hdr.length);
1526 int var_size3 = cookie->raw_addr_list_len;
1527 __u8 *raw = (__u8 *)cookie->peer_init + var_size2;
1528
1529 return sctp_raw_to_bind_addrs(&asoc->base.bind_addr, raw, var_size3,
1530 asoc->ep->base.bind_addr.port, gfp);
1531}
1532
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +09001533/* Lookup laddr in the bind address list of an association. */
1534int sctp_assoc_lookup_laddr(struct sctp_association *asoc,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 const union sctp_addr *laddr)
1536{
Vlad Yasevich559cf712007-09-16 16:03:28 -07001537 int found = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 if ((asoc->base.bind_addr.port == ntohs(laddr->v4.sin_port)) &&
1540 sctp_bind_addr_match(&asoc->base.bind_addr, laddr,
Vlad Yasevich559cf712007-09-16 16:03:28 -07001541 sctp_sk(asoc->base.sk)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 found = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 return found;
1545}
Vlad Yasevich07d93962007-05-04 13:55:27 -07001546
1547/* Set an association id for a given association */
1548int sctp_assoc_set_id(struct sctp_association *asoc, gfp_t gfp)
1549{
Tejun Heo94960e82013-02-27 17:05:00 -08001550 bool preload = gfp & __GFP_WAIT;
1551 int ret;
Vlad Yasevichc6ba68a2009-06-01 12:41:15 -04001552
1553 /* If the id is already assigned, keep it. */
1554 if (asoc->assoc_id)
Tejun Heo94960e82013-02-27 17:05:00 -08001555 return 0;
Vlad Yasevich07d93962007-05-04 13:55:27 -07001556
Tejun Heo94960e82013-02-27 17:05:00 -08001557 if (preload)
1558 idr_preload(gfp);
Vlad Yasevich07d93962007-05-04 13:55:27 -07001559 spin_lock_bh(&sctp_assocs_id_lock);
Jeff Layton713e00a2013-04-29 16:21:22 -07001560 /* 0 is not a valid assoc_id, must be >= 1 */
1561 ret = idr_alloc_cyclic(&sctp_assocs_id, asoc, 1, 0, GFP_NOWAIT);
Vlad Yasevich07d93962007-05-04 13:55:27 -07001562 spin_unlock_bh(&sctp_assocs_id_lock);
Tejun Heo94960e82013-02-27 17:05:00 -08001563 if (preload)
1564 idr_preload_end();
1565 if (ret < 0)
1566 return ret;
Vlad Yasevich07d93962007-05-04 13:55:27 -07001567
Tejun Heo94960e82013-02-27 17:05:00 -08001568 asoc->assoc_id = (sctp_assoc_t)ret;
1569 return 0;
Vlad Yasevich07d93962007-05-04 13:55:27 -07001570}
Vlad Yasevicha08de642007-12-20 14:11:47 -08001571
Wei Yongjun8b4472c2011-05-24 21:48:02 +00001572/* Free the ASCONF queue */
1573static void sctp_assoc_free_asconf_queue(struct sctp_association *asoc)
1574{
1575 struct sctp_chunk *asconf;
1576 struct sctp_chunk *tmp;
1577
1578 list_for_each_entry_safe(asconf, tmp, &asoc->addip_chunk_list, list) {
1579 list_del_init(&asconf->list);
1580 sctp_chunk_free(asconf);
1581 }
1582}
1583
Vlad Yasevicha08de642007-12-20 14:11:47 -08001584/* Free asconf_ack cache */
1585static void sctp_assoc_free_asconf_acks(struct sctp_association *asoc)
1586{
1587 struct sctp_chunk *ack;
1588 struct sctp_chunk *tmp;
1589
1590 list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
1591 transmitted_list) {
1592 list_del_init(&ack->transmitted_list);
1593 sctp_chunk_free(ack);
1594 }
1595}
1596
1597/* Clean up the ASCONF_ACK queue */
1598void sctp_assoc_clean_asconf_ack_cache(const struct sctp_association *asoc)
1599{
1600 struct sctp_chunk *ack;
1601 struct sctp_chunk *tmp;
1602
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001603 /* We can remove all the entries from the queue up to
Vlad Yasevicha08de642007-12-20 14:11:47 -08001604 * the "Peer-Sequence-Number".
1605 */
1606 list_for_each_entry_safe(ack, tmp, &asoc->asconf_ack_list,
1607 transmitted_list) {
1608 if (ack->subh.addip_hdr->serial ==
1609 htonl(asoc->peer.addip_serial))
1610 break;
1611
1612 list_del_init(&ack->transmitted_list);
1613 sctp_chunk_free(ack);
1614 }
1615}
1616
1617/* Find the ASCONF_ACK whose serial number matches ASCONF */
1618struct sctp_chunk *sctp_assoc_lookup_asconf_ack(
1619 const struct sctp_association *asoc,
1620 __be32 serial)
1621{
Wei Yongjuna8699812008-02-05 23:35:04 +09001622 struct sctp_chunk *ack;
Vlad Yasevicha08de642007-12-20 14:11:47 -08001623
1624 /* Walk through the list of cached ASCONF-ACKs and find the
1625 * ack chunk whose serial number matches that of the request.
1626 */
1627 list_for_each_entry(ack, &asoc->asconf_ack_list, transmitted_list) {
1628 if (ack->subh.addip_hdr->serial == serial) {
1629 sctp_chunk_hold(ack);
Wei Yongjuna8699812008-02-05 23:35:04 +09001630 return ack;
Vlad Yasevicha08de642007-12-20 14:11:47 -08001631 }
1632 }
1633
Wei Yongjuna8699812008-02-05 23:35:04 +09001634 return NULL;
Vlad Yasevicha08de642007-12-20 14:11:47 -08001635}
Wei Yongjuna000c012011-05-29 23:23:36 +00001636
1637void sctp_asconf_queue_teardown(struct sctp_association *asoc)
1638{
1639 /* Free any cached ASCONF_ACK chunk. */
1640 sctp_assoc_free_asconf_acks(asoc);
1641
1642 /* Free the ASCONF queue. */
1643 sctp_assoc_free_asconf_queue(asoc);
1644
1645 /* Free any cached ASCONF chunk. */
1646 if (asoc->addip_last_asconf)
1647 sctp_chunk_free(asoc->addip_last_asconf);
1648}