blob: b764114445156dc69146856c118cbf1a725addca [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 *
Vlad Yasevich60c778b2008-01-11 09:57:09 -05006 * This file is part of the SCTP kernel implementation
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 *
8 * These functions handle output processing.
9 *
Vlad Yasevich60c778b2008-01-11 09:57:09 -050010 * This SCTP implementation is free software;
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * you can redistribute it and/or modify it under the terms of
12 * the GNU General Public License as published by
13 * the Free Software Foundation; either version 2, or (at your option)
14 * any later version.
15 *
Vlad Yasevich60c778b2008-01-11 09:57:09 -050016 * This SCTP implementation is distributed in the hope that it
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 * will be useful, but WITHOUT ANY WARRANTY; without even the implied
18 * ************************
19 * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
20 * See the GNU General Public License for more details.
21 *
22 * You should have received a copy of the GNU General Public License
23 * along with GNU CC; see the file COPYING. If not, write to
24 * the Free Software Foundation, 59 Temple Place - Suite 330,
25 * Boston, MA 02111-1307, USA.
26 *
27 * Please send any bug reports or fixes you make to the
28 * email address(es):
29 * lksctp developers <lksctp-developers@lists.sourceforge.net>
30 *
31 * Or submit a bug report through the following website:
32 * http://www.sf.net/projects/lksctp
33 *
34 * Written or modified by:
35 * La Monte H.P. Yarroll <piggy@acm.org>
36 * Karl Knutson <karl@athena.chicago.il.us>
37 * Jon Grimm <jgrimm@austin.ibm.com>
38 * Sridhar Samudrala <sri@us.ibm.com>
39 *
40 * Any bugs reported given to us we will try to fix... any fixes shared will
41 * be incorporated into the next SCTP release.
42 */
43
44#include <linux/types.h>
45#include <linux/kernel.h>
46#include <linux/wait.h>
47#include <linux/time.h>
48#include <linux/ip.h>
49#include <linux/ipv6.h>
50#include <linux/init.h>
51#include <net/inet_ecn.h>
Vlad Yasevich8d2f9e812009-03-21 13:41:09 -070052#include <net/ip.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#include <net/icmp.h>
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -070054#include <net/net_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Linus Torvalds1da177e2005-04-16 15:20:36 -070056#include <linux/socket.h> /* for sa_family_t */
57#include <net/sock.h>
58
59#include <net/sctp/sctp.h>
60#include <net/sctp/sm.h>
Vlad Yasevich9ad09772007-12-16 14:06:41 -080061#include <net/sctp/checksum.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
63/* Forward declarations for private helpers. */
64static sctp_xmit_t sctp_packet_append_data(struct sctp_packet *packet,
65 struct sctp_chunk *chunk);
66
67/* Config a packet.
68 * This appears to be a followup set of initializations.
69 */
70struct sctp_packet *sctp_packet_config(struct sctp_packet *packet,
71 __u32 vtag, int ecn_capable)
72{
73 struct sctp_chunk *chunk = NULL;
74
Harvey Harrison0dc47872008-03-05 20:47:47 -080075 SCTP_DEBUG_PRINTK("%s: packet:%p vtag:0x%x\n", __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 packet, vtag);
77
78 packet->vtag = vtag;
79 packet->has_cookie_echo = 0;
80 packet->has_sack = 0;
Vlad Yasevicha29a5bd2007-09-16 19:31:35 -070081 packet->has_auth = 0;
Vlad Yasevich4cd57c82007-09-16 19:32:45 -070082 packet->has_data = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 packet->ipfragok = 0;
Vlad Yasevicha29a5bd2007-09-16 19:31:35 -070084 packet->auth = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
86 if (ecn_capable && sctp_packet_empty(packet)) {
87 chunk = sctp_get_ecne_prepend(packet->transport->asoc);
88
89 /* If there a is a prepend chunk stick it on the list before
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +090090 * any other chunks get appended.
91 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 if (chunk)
93 sctp_packet_append_chunk(packet, chunk);
94 }
95
96 return packet;
97}
98
99/* Initialize the packet structure. */
100struct sctp_packet *sctp_packet_init(struct sctp_packet *packet,
101 struct sctp_transport *transport,
102 __u16 sport, __u16 dport)
103{
104 struct sctp_association *asoc = transport->asoc;
105 size_t overhead;
106
Harvey Harrison0dc47872008-03-05 20:47:47 -0800107 SCTP_DEBUG_PRINTK("%s: packet:%p transport:%p\n", __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 packet, transport);
109
110 packet->transport = transport;
111 packet->source_port = sport;
112 packet->destination_port = dport;
David S. Miller79af02c2005-07-08 21:47:49 -0700113 INIT_LIST_HEAD(&packet->chunk_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 if (asoc) {
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900115 struct sctp_sock *sp = sctp_sk(asoc->base.sk);
116 overhead = sp->pf->af->net_header_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 } else {
118 overhead = sizeof(struct ipv6hdr);
119 }
120 overhead += sizeof(struct sctphdr);
121 packet->overhead = overhead;
122 packet->size = overhead;
123 packet->vtag = 0;
124 packet->has_cookie_echo = 0;
125 packet->has_sack = 0;
Vlad Yasevicha29a5bd2007-09-16 19:31:35 -0700126 packet->has_auth = 0;
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700127 packet->has_data = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 packet->ipfragok = 0;
129 packet->malloced = 0;
Vlad Yasevicha29a5bd2007-09-16 19:31:35 -0700130 packet->auth = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 return packet;
132}
133
134/* Free a packet. */
135void sctp_packet_free(struct sctp_packet *packet)
136{
David S. Miller79af02c2005-07-08 21:47:49 -0700137 struct sctp_chunk *chunk, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
Harvey Harrison0dc47872008-03-05 20:47:47 -0800139 SCTP_DEBUG_PRINTK("%s: packet:%p\n", __func__, packet);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
David S. Miller79af02c2005-07-08 21:47:49 -0700141 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
142 list_del_init(&chunk->list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 sctp_chunk_free(chunk);
David S. Miller79af02c2005-07-08 21:47:49 -0700144 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
146 if (packet->malloced)
147 kfree(packet);
148}
149
150/* This routine tries to append the chunk to the offered packet. If adding
151 * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk
152 * is not present in the packet, it transmits the input packet.
153 * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long
154 * as it can fit in the packet, but any more data that does not fit in this
155 * packet can be sent only after receiving the COOKIE_ACK.
156 */
157sctp_xmit_t sctp_packet_transmit_chunk(struct sctp_packet *packet,
Vlad Yasevich2e3216c2008-06-19 16:08:18 -0700158 struct sctp_chunk *chunk,
159 int one_packet)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160{
161 sctp_xmit_t retval;
162 int error = 0;
163
Harvey Harrison0dc47872008-03-05 20:47:47 -0800164 SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 packet, chunk);
166
167 switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) {
168 case SCTP_XMIT_PMTU_FULL:
169 if (!packet->has_cookie_echo) {
170 error = sctp_packet_transmit(packet);
171 if (error < 0)
172 chunk->skb->sk->sk_err = -error;
173
174 /* If we have an empty packet, then we can NOT ever
175 * return PMTU_FULL.
176 */
Vlad Yasevich2e3216c2008-06-19 16:08:18 -0700177 if (!one_packet)
178 retval = sctp_packet_append_chunk(packet,
179 chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 }
181 break;
182
183 case SCTP_XMIT_RWND_FULL:
184 case SCTP_XMIT_OK:
185 case SCTP_XMIT_NAGLE_DELAY:
186 break;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700187 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
189 return retval;
190}
191
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700192/* Try to bundle an auth chunk into the packet. */
193static sctp_xmit_t sctp_packet_bundle_auth(struct sctp_packet *pkt,
194 struct sctp_chunk *chunk)
195{
196 struct sctp_association *asoc = pkt->transport->asoc;
197 struct sctp_chunk *auth;
198 sctp_xmit_t retval = SCTP_XMIT_OK;
199
200 /* if we don't have an association, we can't do authentication */
201 if (!asoc)
202 return retval;
203
204 /* See if this is an auth chunk we are bundling or if
205 * auth is already bundled.
206 */
207 if (chunk->chunk_hdr->type == SCTP_CID_AUTH || pkt->auth)
208 return retval;
209
210 /* if the peer did not request this chunk to be authenticated,
211 * don't do it
212 */
213 if (!chunk->auth)
214 return retval;
215
216 auth = sctp_make_auth(asoc);
217 if (!auth)
218 return retval;
219
220 retval = sctp_packet_append_chunk(pkt, auth);
221
222 return retval;
223}
224
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225/* Try to bundle a SACK with the packet. */
226static sctp_xmit_t sctp_packet_bundle_sack(struct sctp_packet *pkt,
227 struct sctp_chunk *chunk)
228{
229 sctp_xmit_t retval = SCTP_XMIT_OK;
230
231 /* If sending DATA and haven't aleady bundled a SACK, try to
232 * bundle one in to the packet.
233 */
234 if (sctp_chunk_is_data(chunk) && !pkt->has_sack &&
235 !pkt->has_cookie_echo) {
236 struct sctp_association *asoc;
237 asoc = pkt->transport->asoc;
238
239 if (asoc->a_rwnd > asoc->rwnd) {
240 struct sctp_chunk *sack;
241 asoc->a_rwnd = asoc->rwnd;
242 sack = sctp_make_sack(asoc);
243 if (sack) {
244 struct timer_list *timer;
245 retval = sctp_packet_append_chunk(pkt, sack);
246 asoc->peer.sack_needed = 0;
247 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
248 if (timer_pending(timer) && del_timer(timer))
249 sctp_association_put(asoc);
250 }
251 }
252 }
253 return retval;
254}
255
256/* Append a chunk to the offered packet reporting back any inability to do
257 * so.
258 */
259sctp_xmit_t sctp_packet_append_chunk(struct sctp_packet *packet,
260 struct sctp_chunk *chunk)
261{
262 sctp_xmit_t retval = SCTP_XMIT_OK;
263 __u16 chunk_len = WORD_ROUND(ntohs(chunk->chunk_hdr->length));
264 size_t psize;
265 size_t pmtu;
266 int too_big;
267
Harvey Harrison0dc47872008-03-05 20:47:47 -0800268 SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __func__, packet,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 chunk);
270
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700271 /* Try to bundle AUTH chunk */
272 retval = sctp_packet_bundle_auth(packet, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 if (retval != SCTP_XMIT_OK)
274 goto finish;
275
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700276 /* Try to bundle SACK chunk */
277 retval = sctp_packet_bundle_sack(packet, chunk);
278 if (retval != SCTP_XMIT_OK)
279 goto finish;
280
281 psize = packet->size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 pmtu = ((packet->transport->asoc) ?
Frank Filz52ccb8e2005-12-22 11:36:46 -0800283 (packet->transport->asoc->pathmtu) :
284 (packet->transport->pathmtu));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285
286 too_big = (psize + chunk_len > pmtu);
287
288 /* Decide if we need to fragment or resubmit later. */
289 if (too_big) {
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700290 /* It's OK to fragmet at IP level if any one of the following
291 * is true:
292 * 1. The packet is empty (meaning this chunk is greater
293 * the MTU)
294 * 2. The chunk we are adding is a control chunk
295 * 3. The packet doesn't have any data in it yet and data
296 * requires authentication.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 */
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700298 if (sctp_packet_empty(packet) || !sctp_chunk_is_data(chunk) ||
299 (!packet->has_data && chunk->auth)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 /* We no longer do re-fragmentation.
301 * Just fragment at the IP layer, if we
302 * actually hit this condition
303 */
304 packet->ipfragok = 1;
305 goto append;
306
307 } else {
308 retval = SCTP_XMIT_PMTU_FULL;
309 goto finish;
310 }
311 }
312
313append:
314 /* We believe that this chunk is OK to add to the packet (as
315 * long as we have the cwnd for it).
316 */
317
318 /* DATA is a special case since we must examine both rwnd and cwnd
319 * before we send DATA.
320 */
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700321 switch (chunk->chunk_hdr->type) {
322 case SCTP_CID_DATA:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 retval = sctp_packet_append_data(packet, chunk);
Vlad Yasevich759af002009-01-22 14:53:01 -0800324 if (SCTP_XMIT_OK != retval)
325 goto finish;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 /* Disallow SACK bundling after DATA. */
327 packet->has_sack = 1;
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700328 /* Disallow AUTH bundling after DATA */
329 packet->has_auth = 1;
330 /* Let it be knows that packet has DATA in it */
331 packet->has_data = 1;
Vlad Yasevich759af002009-01-22 14:53:01 -0800332 /* timestamp the chunk for rtx purposes */
333 chunk->sent_at = jiffies;
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700334 break;
335 case SCTP_CID_COOKIE_ECHO:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 packet->has_cookie_echo = 1;
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700337 break;
338
339 case SCTP_CID_SACK:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 packet->has_sack = 1;
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700341 break;
342
343 case SCTP_CID_AUTH:
344 packet->has_auth = 1;
345 packet->auth = chunk;
346 break;
347 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348
349 /* It is OK to send this chunk. */
David S. Miller79af02c2005-07-08 21:47:49 -0700350 list_add_tail(&chunk->list, &packet->chunk_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 packet->size += chunk_len;
352 chunk->transport = packet->transport;
353finish:
354 return retval;
355}
356
357/* All packets are sent to the network through this function from
358 * sctp_outq_tail().
359 *
360 * The return value is a normal kernel error return value.
361 */
362int sctp_packet_transmit(struct sctp_packet *packet)
363{
364 struct sctp_transport *tp = packet->transport;
365 struct sctp_association *asoc = tp->asoc;
366 struct sctphdr *sh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 struct sk_buff *nskb;
David S. Miller79af02c2005-07-08 21:47:49 -0700368 struct sctp_chunk *chunk, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 struct sock *sk;
370 int err = 0;
371 int padding; /* How much padding do we need? */
372 __u8 has_data = 0;
Sridhar Samudrala503b55f2006-06-17 22:57:28 -0700373 struct dst_entry *dst = tp->dst;
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700374 unsigned char *auth = NULL; /* pointer to auth in skb data */
375 __u32 cksum_buf_len = sizeof(struct sctphdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376
Harvey Harrison0dc47872008-03-05 20:47:47 -0800377 SCTP_DEBUG_PRINTK("%s: packet:%p\n", __func__, packet);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378
379 /* Do NOT generate a chunkless packet. */
David S. Miller79af02c2005-07-08 21:47:49 -0700380 if (list_empty(&packet->chunk_list))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 return err;
382
383 /* Set up convenience variables... */
David S. Miller79af02c2005-07-08 21:47:49 -0700384 chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 sk = chunk->skb->sk;
386
387 /* Allocate the new skb. */
Sridhar Samudrala594ccc12005-04-28 12:00:23 -0700388 nskb = alloc_skb(packet->size + LL_MAX_HEADER, GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 if (!nskb)
390 goto nomem;
391
392 /* Make sure the outbound skb has enough header room reserved. */
Sridhar Samudrala594ccc12005-04-28 12:00:23 -0700393 skb_reserve(nskb, packet->overhead + LL_MAX_HEADER);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394
395 /* Set the owning socket so that we know where to get the
396 * destination IP address.
397 */
398 skb_set_owner_w(nskb, sk);
399
Sridhar Samudrala503b55f2006-06-17 22:57:28 -0700400 /* The 'obsolete' field of dst is set to 2 when a dst is freed. */
401 if (!dst || (dst->obsolete > 1)) {
402 dst_release(dst);
403 sctp_transport_route(tp, NULL, sctp_sk(sk));
404 if (asoc && (asoc->param_flags & SPP_PMTUD_ENABLE)) {
405 sctp_assoc_sync_pmtu(asoc);
406 }
407 }
Eric Dumazetadf30902009-06-02 05:19:30 +0000408 dst = dst_clone(tp->dst);
409 skb_dst_set(nskb, dst);
410 if (dst)
Sridhar Samudrala503b55f2006-06-17 22:57:28 -0700411 goto no_route;
Sridhar Samudrala503b55f2006-06-17 22:57:28 -0700412
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 /* Build the SCTP header. */
414 sh = (struct sctphdr *)skb_push(nskb, sizeof(struct sctphdr));
Jesse Brandeburg8dc92f72009-04-27 22:35:52 +0000415 skb_reset_transport_header(nskb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 sh->source = htons(packet->source_port);
417 sh->dest = htons(packet->destination_port);
418
419 /* From 6.8 Adler-32 Checksum Calculation:
420 * After the packet is constructed (containing the SCTP common
421 * header and one or more control or DATA chunks), the
422 * transmitter shall:
423 *
424 * 1) Fill in the proper Verification Tag in the SCTP common
425 * header and initialize the checksum field to 0's.
426 */
427 sh->vtag = htonl(packet->vtag);
428 sh->checksum = 0;
429
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 /**
431 * 6.10 Bundling
432 *
433 * An endpoint bundles chunks by simply including multiple
434 * chunks in one outbound SCTP packet. ...
435 */
436
437 /**
438 * 3.2 Chunk Field Descriptions
439 *
440 * The total length of a chunk (including Type, Length and
441 * Value fields) MUST be a multiple of 4 bytes. If the length
442 * of the chunk is not a multiple of 4 bytes, the sender MUST
443 * pad the chunk with all zero bytes and this padding is not
444 * included in the chunk length field. The sender should
445 * never pad with more than 3 bytes.
446 *
447 * [This whole comment explains WORD_ROUND() below.]
448 */
449 SCTP_DEBUG_PRINTK("***sctp_transmit_packet***\n");
David S. Miller79af02c2005-07-08 21:47:49 -0700450 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
451 list_del_init(&chunk->list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 if (sctp_chunk_is_data(chunk)) {
453
454 if (!chunk->has_tsn) {
455 sctp_chunk_assign_ssn(chunk);
456 sctp_chunk_assign_tsn(chunk);
457
458 /* 6.3.1 C4) When data is in flight and when allowed
459 * by rule C5, a new RTT measurement MUST be made each
460 * round trip. Furthermore, new RTT measurements
461 * SHOULD be made no more than once per round-trip
462 * for a given destination transport address.
463 */
464
465 if (!tp->rto_pending) {
466 chunk->rtt_in_progress = 1;
467 tp->rto_pending = 1;
468 }
469 } else
470 chunk->resent = 1;
471
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 has_data = 1;
473 }
474
475 padding = WORD_ROUND(chunk->skb->len) - chunk->skb->len;
476 if (padding)
477 memset(skb_put(chunk->skb, padding), 0, padding);
478
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700479 /* if this is the auth chunk that we are adding,
480 * store pointer where it will be added and put
481 * the auth into the packet.
482 */
483 if (chunk == packet->auth)
484 auth = skb_tail_pointer(nskb);
485
486 cksum_buf_len += chunk->skb->len;
487 memcpy(skb_put(nskb, chunk->skb->len),
Sridhar Samudrala503b55f2006-06-17 22:57:28 -0700488 chunk->skb->data, chunk->skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489
490 SCTP_DEBUG_PRINTK("%s %p[%s] %s 0x%x, %s %d, %s %d, %s %d\n",
491 "*** Chunk", chunk,
492 sctp_cname(SCTP_ST_CHUNK(
493 chunk->chunk_hdr->type)),
494 chunk->has_tsn ? "TSN" : "No TSN",
495 chunk->has_tsn ?
496 ntohl(chunk->subh.data_hdr->tsn) : 0,
497 "length", ntohs(chunk->chunk_hdr->length),
498 "chunk->skb->len", chunk->skb->len,
499 "rtt_in_progress", chunk->rtt_in_progress);
500
501 /*
502 * If this is a control chunk, this is our last
503 * reference. Free data chunks after they've been
504 * acknowledged or have failed.
505 */
506 if (!sctp_chunk_is_data(chunk))
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900507 sctp_chunk_free(chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 }
509
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700510 /* SCTP-AUTH, Section 6.2
511 * The sender MUST calculate the MAC as described in RFC2104 [2]
512 * using the hash function H as described by the MAC Identifier and
513 * the shared association key K based on the endpoint pair shared key
514 * described by the shared key identifier. The 'data' used for the
515 * computation of the AUTH-chunk is given by the AUTH chunk with its
516 * HMAC field set to zero (as shown in Figure 6) followed by all
517 * chunks that are placed after the AUTH chunk in the SCTP packet.
518 */
519 if (auth)
520 sctp_auth_calculate_hmac(asoc, nskb,
521 (struct sctp_auth_chunk *)auth,
522 GFP_ATOMIC);
523
524 /* 2) Calculate the Adler-32 checksum of the whole packet,
525 * including the SCTP common header and all the
526 * chunks.
527 *
528 * Note: Adler-32 is no longer applicable, as has been replaced
529 * by CRC32-C as described in <draft-ietf-tsvwg-sctpcsum-02.txt>.
530 */
Jesse Brandeburg8dc92f72009-04-27 22:35:52 +0000531 if (!sctp_checksum_disable &&
532 !(dst->dev->features & (NETIF_F_NO_CSUM | NETIF_F_SCTP_CSUM))) {
Vlad Yasevich4458f042009-02-13 08:33:42 +0000533 __u32 crc32 = sctp_start_cksum((__u8 *)sh, cksum_buf_len);
534
535 /* 3) Put the resultant value into the checksum field in the
536 * common header, and leave the rest of the bits unchanged.
537 */
538 sh->checksum = sctp_end_cksum(crc32);
Jesse Brandeburg8dc92f72009-04-27 22:35:52 +0000539 } else {
540 if (dst->dev->features & NETIF_F_SCTP_CSUM) {
541 /* no need to seed psuedo checksum for SCTP */
542 nskb->ip_summed = CHECKSUM_PARTIAL;
543 nskb->csum_start = (skb_transport_header(nskb) -
544 nskb->head);
545 nskb->csum_offset = offsetof(struct sctphdr, checksum);
546 } else {
547 nskb->ip_summed = CHECKSUM_UNNECESSARY;
548 }
549 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 /* IP layer ECN support
552 * From RFC 2481
553 * "The ECN-Capable Transport (ECT) bit would be set by the
554 * data sender to indicate that the end-points of the
555 * transport protocol are ECN-capable."
556 *
557 * Now setting the ECT bit all the time, as it should not cause
558 * any problems protocol-wise even if our peer ignores it.
559 *
560 * Note: The works for IPv6 layer checks this bit too later
561 * in transmission. See IP6_ECN_flow_xmit().
562 */
Vlad Yasevichb9031d92008-06-04 12:40:15 -0700563 (*tp->af_specific->ecn_capable)(nskb->sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
565 /* Set up the IP options. */
566 /* BUG: not implemented
567 * For v4 this all lives somewhere in sk->sk_opt...
568 */
569
570 /* Dump that on IP! */
571 if (asoc && asoc->peer.last_sent_to != tp) {
572 /* Considering the multiple CPU scenario, this is a
573 * "correcter" place for last_sent_to. --xguo
574 */
575 asoc->peer.last_sent_to = tp;
576 }
577
578 if (has_data) {
579 struct timer_list *timer;
580 unsigned long timeout;
581
582 tp->last_time_used = jiffies;
583
584 /* Restart the AUTOCLOSE timer when sending data. */
585 if (sctp_state(asoc, ESTABLISHED) && asoc->autoclose) {
586 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
587 timeout = asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
588
589 if (!mod_timer(timer, jiffies + timeout))
590 sctp_association_hold(asoc);
591 }
592 }
593
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 SCTP_DEBUG_PRINTK("***sctp_transmit_packet*** skb len %d\n",
595 nskb->len);
596
Herbert Xuf8803742008-08-03 21:15:08 -0700597 nskb->local_df = packet->ipfragok;
598 (*tp->af_specific->sctp_xmit)(nskb, tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599
600out:
601 packet->size = packet->overhead;
602 return err;
603no_route:
604 kfree_skb(nskb);
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700605 IP_INC_STATS_BH(&init_net, IPSTATS_MIB_OUTNOROUTES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606
607 /* FIXME: Returning the 'err' will effect all the associations
608 * associated with a socket, although only one of the paths of the
609 * association is unreachable.
610 * The real failure of a transport or association can be passed on
611 * to the user via notifications. So setting this error may not be
612 * required.
613 */
614 /* err = -EHOSTUNREACH; */
615err:
616 /* Control chunks are unreliable so just drop them. DATA chunks
617 * will get resent or dropped later.
618 */
619
David S. Miller79af02c2005-07-08 21:47:49 -0700620 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
621 list_del_init(&chunk->list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 if (!sctp_chunk_is_data(chunk))
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900623 sctp_chunk_free(chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 }
625 goto out;
626nomem:
627 err = -ENOMEM;
628 goto err;
629}
630
631/********************************************************************
632 * 2nd Level Abstractions
633 ********************************************************************/
634
635/* This private function handles the specifics of appending DATA chunks. */
636static sctp_xmit_t sctp_packet_append_data(struct sctp_packet *packet,
637 struct sctp_chunk *chunk)
638{
639 sctp_xmit_t retval = SCTP_XMIT_OK;
640 size_t datasize, rwnd, inflight;
641 struct sctp_transport *transport = packet->transport;
642 __u32 max_burst_bytes;
643 struct sctp_association *asoc = transport->asoc;
644 struct sctp_sock *sp = sctp_sk(asoc->base.sk);
645 struct sctp_outq *q = &asoc->outqueue;
646
647 /* RFC 2960 6.1 Transmission of DATA Chunks
648 *
649 * A) At any given time, the data sender MUST NOT transmit new data to
650 * any destination transport address if its peer's rwnd indicates
651 * that the peer has no buffer space (i.e. rwnd is 0, see Section
652 * 6.2.1). However, regardless of the value of rwnd (including if it
653 * is 0), the data sender can always have one DATA chunk in flight to
654 * the receiver if allowed by cwnd (see rule B below). This rule
655 * allows the sender to probe for a change in rwnd that the sender
656 * missed due to the SACK having been lost in transit from the data
657 * receiver to the data sender.
658 */
659
660 rwnd = asoc->peer.rwnd;
661 inflight = asoc->outqueue.outstanding_bytes;
662
663 datasize = sctp_data_size(chunk);
664
665 if (datasize > rwnd) {
666 if (inflight > 0) {
667 /* We have (at least) one data chunk in flight,
668 * so we can't fall back to rule 6.1 B).
669 */
670 retval = SCTP_XMIT_RWND_FULL;
671 goto finish;
672 }
673 }
674
675 /* sctpimpguide-05 2.14.2
676 * D) When the time comes for the sender to
677 * transmit new DATA chunks, the protocol parameter Max.Burst MUST
678 * first be applied to limit how many new DATA chunks may be sent.
679 * The limit is applied by adjusting cwnd as follows:
680 * if ((flightsize + Max.Burst * MTU) < cwnd)
681 * cwnd = flightsize + Max.Burst * MTU
682 */
Frank Filz52ccb8e2005-12-22 11:36:46 -0800683 max_burst_bytes = asoc->max_burst * asoc->pathmtu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 if ((transport->flight_size + max_burst_bytes) < transport->cwnd) {
685 transport->cwnd = transport->flight_size + max_burst_bytes;
686 SCTP_DEBUG_PRINTK("%s: cwnd limited by max_burst: "
687 "transport: %p, cwnd: %d, "
688 "ssthresh: %d, flight_size: %d, "
689 "pba: %d\n",
Harvey Harrison0dc47872008-03-05 20:47:47 -0800690 __func__, transport,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 transport->cwnd,
692 transport->ssthresh,
693 transport->flight_size,
694 transport->partial_bytes_acked);
695 }
696
697 /* RFC 2960 6.1 Transmission of DATA Chunks
698 *
699 * B) At any given time, the sender MUST NOT transmit new data
700 * to a given transport address if it has cwnd or more bytes
701 * of data outstanding to that transport address.
702 */
703 /* RFC 7.2.4 & the Implementers Guide 2.8.
704 *
705 * 3) ...
706 * When a Fast Retransmit is being performed the sender SHOULD
707 * ignore the value of cwnd and SHOULD NOT delay retransmission.
708 */
Neil Hormanc226ef92008-07-25 12:44:09 -0400709 if (chunk->fast_retransmit != SCTP_NEED_FRTX)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 if (transport->flight_size >= transport->cwnd) {
711 retval = SCTP_XMIT_RWND_FULL;
712 goto finish;
713 }
714
715 /* Nagle's algorithm to solve small-packet problem:
716 * Inhibit the sending of new chunks when new outgoing data arrives
717 * if any previously transmitted data on the connection remains
718 * unacknowledged.
719 */
720 if (!sp->nodelay && sctp_packet_empty(packet) &&
721 q->outstanding_bytes && sctp_state(asoc, ESTABLISHED)) {
722 unsigned len = datasize + q->out_qlen;
723
724 /* Check whether this chunk and all the rest of pending
725 * data will fit or delay in hopes of bundling a full
726 * sized packet.
727 */
Sridhar Samudralacd497882006-09-29 17:09:05 -0700728 if (len < asoc->frag_point) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 retval = SCTP_XMIT_NAGLE_DELAY;
730 goto finish;
731 }
732 }
733
734 /* Keep track of how many bytes are in flight over this transport. */
735 transport->flight_size += datasize;
736
737 /* Keep track of how many bytes are in flight to the receiver. */
738 asoc->outqueue.outstanding_bytes += datasize;
739
Sridhar Samudralacd497882006-09-29 17:09:05 -0700740 /* Update our view of the receiver's rwnd. Include sk_buff overhead
741 * while updating peer.rwnd so that it reduces the chances of a
742 * receiver running out of receive buffer space even when receive
743 * window is still open. This can happen when a sender is sending
744 * sending small messages.
745 */
746 datasize += sizeof(struct sk_buff);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 if (datasize < rwnd)
748 rwnd -= datasize;
749 else
750 rwnd = 0;
751
752 asoc->peer.rwnd = rwnd;
753 /* Has been accepted for transmission. */
754 if (!asoc->peer.prsctp_capable)
755 chunk->msg->can_abandon = 0;
756
757finish:
758 return retval;
759}