blob: 07d58903a74625dd2ec25982a617a6cad2d59ae3 [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>
52#include <net/icmp.h>
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -070053#include <net/net_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
55#ifndef TEST_FRAME
56#include <net/tcp.h>
57#endif /* TEST_FRAME (not defined) */
58
59#include <linux/socket.h> /* for sa_family_t */
60#include <net/sock.h>
61
62#include <net/sctp/sctp.h>
63#include <net/sctp/sm.h>
Vlad Yasevich9ad09772007-12-16 14:06:41 -080064#include <net/sctp/checksum.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
66/* Forward declarations for private helpers. */
67static sctp_xmit_t sctp_packet_append_data(struct sctp_packet *packet,
68 struct sctp_chunk *chunk);
69
70/* Config a packet.
71 * This appears to be a followup set of initializations.
72 */
73struct sctp_packet *sctp_packet_config(struct sctp_packet *packet,
74 __u32 vtag, int ecn_capable)
75{
76 struct sctp_chunk *chunk = NULL;
77
Harvey Harrison0dc47872008-03-05 20:47:47 -080078 SCTP_DEBUG_PRINTK("%s: packet:%p vtag:0x%x\n", __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 packet, vtag);
80
81 packet->vtag = vtag;
82 packet->has_cookie_echo = 0;
83 packet->has_sack = 0;
Vlad Yasevicha29a5bd2007-09-16 19:31:35 -070084 packet->has_auth = 0;
Vlad Yasevich4cd57c82007-09-16 19:32:45 -070085 packet->has_data = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 packet->ipfragok = 0;
Vlad Yasevicha29a5bd2007-09-16 19:31:35 -070087 packet->auth = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
89 if (ecn_capable && sctp_packet_empty(packet)) {
90 chunk = sctp_get_ecne_prepend(packet->transport->asoc);
91
92 /* If there a is a prepend chunk stick it on the list before
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +090093 * any other chunks get appended.
94 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 if (chunk)
96 sctp_packet_append_chunk(packet, chunk);
97 }
98
99 return packet;
100}
101
102/* Initialize the packet structure. */
103struct sctp_packet *sctp_packet_init(struct sctp_packet *packet,
104 struct sctp_transport *transport,
105 __u16 sport, __u16 dport)
106{
107 struct sctp_association *asoc = transport->asoc;
108 size_t overhead;
109
Harvey Harrison0dc47872008-03-05 20:47:47 -0800110 SCTP_DEBUG_PRINTK("%s: packet:%p transport:%p\n", __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 packet, transport);
112
113 packet->transport = transport;
114 packet->source_port = sport;
115 packet->destination_port = dport;
David S. Miller79af02c2005-07-08 21:47:49 -0700116 INIT_LIST_HEAD(&packet->chunk_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 if (asoc) {
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900118 struct sctp_sock *sp = sctp_sk(asoc->base.sk);
119 overhead = sp->pf->af->net_header_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 } else {
121 overhead = sizeof(struct ipv6hdr);
122 }
123 overhead += sizeof(struct sctphdr);
124 packet->overhead = overhead;
125 packet->size = overhead;
126 packet->vtag = 0;
127 packet->has_cookie_echo = 0;
128 packet->has_sack = 0;
Vlad Yasevicha29a5bd2007-09-16 19:31:35 -0700129 packet->has_auth = 0;
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700130 packet->has_data = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 packet->ipfragok = 0;
132 packet->malloced = 0;
Vlad Yasevicha29a5bd2007-09-16 19:31:35 -0700133 packet->auth = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 return packet;
135}
136
137/* Free a packet. */
138void sctp_packet_free(struct sctp_packet *packet)
139{
David S. Miller79af02c2005-07-08 21:47:49 -0700140 struct sctp_chunk *chunk, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
Harvey Harrison0dc47872008-03-05 20:47:47 -0800142 SCTP_DEBUG_PRINTK("%s: packet:%p\n", __func__, packet);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
David S. Miller79af02c2005-07-08 21:47:49 -0700144 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
145 list_del_init(&chunk->list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 sctp_chunk_free(chunk);
David S. Miller79af02c2005-07-08 21:47:49 -0700147 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148
149 if (packet->malloced)
150 kfree(packet);
151}
152
153/* This routine tries to append the chunk to the offered packet. If adding
154 * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk
155 * is not present in the packet, it transmits the input packet.
156 * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long
157 * as it can fit in the packet, but any more data that does not fit in this
158 * packet can be sent only after receiving the COOKIE_ACK.
159 */
160sctp_xmit_t sctp_packet_transmit_chunk(struct sctp_packet *packet,
Vlad Yasevich2e3216c2008-06-19 16:08:18 -0700161 struct sctp_chunk *chunk,
162 int one_packet)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163{
164 sctp_xmit_t retval;
165 int error = 0;
166
Harvey Harrison0dc47872008-03-05 20:47:47 -0800167 SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 packet, chunk);
169
170 switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) {
171 case SCTP_XMIT_PMTU_FULL:
172 if (!packet->has_cookie_echo) {
173 error = sctp_packet_transmit(packet);
174 if (error < 0)
175 chunk->skb->sk->sk_err = -error;
176
177 /* If we have an empty packet, then we can NOT ever
178 * return PMTU_FULL.
179 */
Vlad Yasevich2e3216c2008-06-19 16:08:18 -0700180 if (!one_packet)
181 retval = sctp_packet_append_chunk(packet,
182 chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 }
184 break;
185
186 case SCTP_XMIT_RWND_FULL:
187 case SCTP_XMIT_OK:
188 case SCTP_XMIT_NAGLE_DELAY:
189 break;
Stephen Hemminger3ff50b72007-04-20 17:09:22 -0700190 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191
192 return retval;
193}
194
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700195/* Try to bundle an auth chunk into the packet. */
196static sctp_xmit_t sctp_packet_bundle_auth(struct sctp_packet *pkt,
197 struct sctp_chunk *chunk)
198{
199 struct sctp_association *asoc = pkt->transport->asoc;
200 struct sctp_chunk *auth;
201 sctp_xmit_t retval = SCTP_XMIT_OK;
202
203 /* if we don't have an association, we can't do authentication */
204 if (!asoc)
205 return retval;
206
207 /* See if this is an auth chunk we are bundling or if
208 * auth is already bundled.
209 */
210 if (chunk->chunk_hdr->type == SCTP_CID_AUTH || pkt->auth)
211 return retval;
212
213 /* if the peer did not request this chunk to be authenticated,
214 * don't do it
215 */
216 if (!chunk->auth)
217 return retval;
218
219 auth = sctp_make_auth(asoc);
220 if (!auth)
221 return retval;
222
223 retval = sctp_packet_append_chunk(pkt, auth);
224
225 return retval;
226}
227
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228/* Try to bundle a SACK with the packet. */
229static sctp_xmit_t sctp_packet_bundle_sack(struct sctp_packet *pkt,
230 struct sctp_chunk *chunk)
231{
232 sctp_xmit_t retval = SCTP_XMIT_OK;
233
234 /* If sending DATA and haven't aleady bundled a SACK, try to
235 * bundle one in to the packet.
236 */
237 if (sctp_chunk_is_data(chunk) && !pkt->has_sack &&
238 !pkt->has_cookie_echo) {
239 struct sctp_association *asoc;
240 asoc = pkt->transport->asoc;
241
242 if (asoc->a_rwnd > asoc->rwnd) {
243 struct sctp_chunk *sack;
244 asoc->a_rwnd = asoc->rwnd;
245 sack = sctp_make_sack(asoc);
246 if (sack) {
247 struct timer_list *timer;
248 retval = sctp_packet_append_chunk(pkt, sack);
249 asoc->peer.sack_needed = 0;
250 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
251 if (timer_pending(timer) && del_timer(timer))
252 sctp_association_put(asoc);
253 }
254 }
255 }
256 return retval;
257}
258
259/* Append a chunk to the offered packet reporting back any inability to do
260 * so.
261 */
262sctp_xmit_t sctp_packet_append_chunk(struct sctp_packet *packet,
263 struct sctp_chunk *chunk)
264{
265 sctp_xmit_t retval = SCTP_XMIT_OK;
266 __u16 chunk_len = WORD_ROUND(ntohs(chunk->chunk_hdr->length));
267 size_t psize;
268 size_t pmtu;
269 int too_big;
270
Harvey Harrison0dc47872008-03-05 20:47:47 -0800271 SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __func__, packet,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 chunk);
273
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700274 /* Try to bundle AUTH chunk */
275 retval = sctp_packet_bundle_auth(packet, chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 if (retval != SCTP_XMIT_OK)
277 goto finish;
278
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700279 /* Try to bundle SACK chunk */
280 retval = sctp_packet_bundle_sack(packet, chunk);
281 if (retval != SCTP_XMIT_OK)
282 goto finish;
283
284 psize = packet->size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 pmtu = ((packet->transport->asoc) ?
Frank Filz52ccb8e2005-12-22 11:36:46 -0800286 (packet->transport->asoc->pathmtu) :
287 (packet->transport->pathmtu));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288
289 too_big = (psize + chunk_len > pmtu);
290
291 /* Decide if we need to fragment or resubmit later. */
292 if (too_big) {
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700293 /* It's OK to fragmet at IP level if any one of the following
294 * is true:
295 * 1. The packet is empty (meaning this chunk is greater
296 * the MTU)
297 * 2. The chunk we are adding is a control chunk
298 * 3. The packet doesn't have any data in it yet and data
299 * requires authentication.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 */
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700301 if (sctp_packet_empty(packet) || !sctp_chunk_is_data(chunk) ||
302 (!packet->has_data && chunk->auth)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303 /* We no longer do re-fragmentation.
304 * Just fragment at the IP layer, if we
305 * actually hit this condition
306 */
307 packet->ipfragok = 1;
308 goto append;
309
310 } else {
311 retval = SCTP_XMIT_PMTU_FULL;
312 goto finish;
313 }
314 }
315
316append:
317 /* We believe that this chunk is OK to add to the packet (as
318 * long as we have the cwnd for it).
319 */
320
321 /* DATA is a special case since we must examine both rwnd and cwnd
322 * before we send DATA.
323 */
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700324 switch (chunk->chunk_hdr->type) {
325 case SCTP_CID_DATA:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 retval = sctp_packet_append_data(packet, chunk);
Vlad Yasevich759af002009-01-22 14:53:01 -0800327 if (SCTP_XMIT_OK != retval)
328 goto finish;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 /* Disallow SACK bundling after DATA. */
330 packet->has_sack = 1;
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700331 /* Disallow AUTH bundling after DATA */
332 packet->has_auth = 1;
333 /* Let it be knows that packet has DATA in it */
334 packet->has_data = 1;
Vlad Yasevich759af002009-01-22 14:53:01 -0800335 /* timestamp the chunk for rtx purposes */
336 chunk->sent_at = jiffies;
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700337 break;
338 case SCTP_CID_COOKIE_ECHO:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 packet->has_cookie_echo = 1;
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700340 break;
341
342 case SCTP_CID_SACK:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 packet->has_sack = 1;
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700344 break;
345
346 case SCTP_CID_AUTH:
347 packet->has_auth = 1;
348 packet->auth = chunk;
349 break;
350 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
352 /* It is OK to send this chunk. */
David S. Miller79af02c2005-07-08 21:47:49 -0700353 list_add_tail(&chunk->list, &packet->chunk_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 packet->size += chunk_len;
355 chunk->transport = packet->transport;
356finish:
357 return retval;
358}
359
360/* All packets are sent to the network through this function from
361 * sctp_outq_tail().
362 *
363 * The return value is a normal kernel error return value.
364 */
365int sctp_packet_transmit(struct sctp_packet *packet)
366{
367 struct sctp_transport *tp = packet->transport;
368 struct sctp_association *asoc = tp->asoc;
369 struct sctphdr *sh;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 struct sk_buff *nskb;
David S. Miller79af02c2005-07-08 21:47:49 -0700371 struct sctp_chunk *chunk, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 struct sock *sk;
373 int err = 0;
374 int padding; /* How much padding do we need? */
375 __u8 has_data = 0;
Sridhar Samudrala503b55f2006-06-17 22:57:28 -0700376 struct dst_entry *dst = tp->dst;
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700377 unsigned char *auth = NULL; /* pointer to auth in skb data */
378 __u32 cksum_buf_len = sizeof(struct sctphdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379
Harvey Harrison0dc47872008-03-05 20:47:47 -0800380 SCTP_DEBUG_PRINTK("%s: packet:%p\n", __func__, packet);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381
382 /* Do NOT generate a chunkless packet. */
David S. Miller79af02c2005-07-08 21:47:49 -0700383 if (list_empty(&packet->chunk_list))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 return err;
385
386 /* Set up convenience variables... */
David S. Miller79af02c2005-07-08 21:47:49 -0700387 chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 sk = chunk->skb->sk;
389
390 /* Allocate the new skb. */
Sridhar Samudrala594ccc12005-04-28 12:00:23 -0700391 nskb = alloc_skb(packet->size + LL_MAX_HEADER, GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 if (!nskb)
393 goto nomem;
394
395 /* Make sure the outbound skb has enough header room reserved. */
Sridhar Samudrala594ccc12005-04-28 12:00:23 -0700396 skb_reserve(nskb, packet->overhead + LL_MAX_HEADER);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
398 /* Set the owning socket so that we know where to get the
399 * destination IP address.
400 */
401 skb_set_owner_w(nskb, sk);
402
Sridhar Samudrala503b55f2006-06-17 22:57:28 -0700403 /* The 'obsolete' field of dst is set to 2 when a dst is freed. */
404 if (!dst || (dst->obsolete > 1)) {
405 dst_release(dst);
406 sctp_transport_route(tp, NULL, sctp_sk(sk));
407 if (asoc && (asoc->param_flags & SPP_PMTUD_ENABLE)) {
408 sctp_assoc_sync_pmtu(asoc);
409 }
410 }
411 nskb->dst = dst_clone(tp->dst);
412 if (!nskb->dst)
413 goto no_route;
414 dst = nskb->dst;
415
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 /* Build the SCTP header. */
417 sh = (struct sctphdr *)skb_push(nskb, sizeof(struct sctphdr));
418 sh->source = htons(packet->source_port);
419 sh->dest = htons(packet->destination_port);
420
421 /* From 6.8 Adler-32 Checksum Calculation:
422 * After the packet is constructed (containing the SCTP common
423 * header and one or more control or DATA chunks), the
424 * transmitter shall:
425 *
426 * 1) Fill in the proper Verification Tag in the SCTP common
427 * header and initialize the checksum field to 0's.
428 */
429 sh->vtag = htonl(packet->vtag);
430 sh->checksum = 0;
431
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 /**
433 * 6.10 Bundling
434 *
435 * An endpoint bundles chunks by simply including multiple
436 * chunks in one outbound SCTP packet. ...
437 */
438
439 /**
440 * 3.2 Chunk Field Descriptions
441 *
442 * The total length of a chunk (including Type, Length and
443 * Value fields) MUST be a multiple of 4 bytes. If the length
444 * of the chunk is not a multiple of 4 bytes, the sender MUST
445 * pad the chunk with all zero bytes and this padding is not
446 * included in the chunk length field. The sender should
447 * never pad with more than 3 bytes.
448 *
449 * [This whole comment explains WORD_ROUND() below.]
450 */
451 SCTP_DEBUG_PRINTK("***sctp_transmit_packet***\n");
David S. Miller79af02c2005-07-08 21:47:49 -0700452 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
453 list_del_init(&chunk->list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 if (sctp_chunk_is_data(chunk)) {
455
456 if (!chunk->has_tsn) {
457 sctp_chunk_assign_ssn(chunk);
458 sctp_chunk_assign_tsn(chunk);
459
460 /* 6.3.1 C4) When data is in flight and when allowed
461 * by rule C5, a new RTT measurement MUST be made each
462 * round trip. Furthermore, new RTT measurements
463 * SHOULD be made no more than once per round-trip
464 * for a given destination transport address.
465 */
466
467 if (!tp->rto_pending) {
468 chunk->rtt_in_progress = 1;
469 tp->rto_pending = 1;
470 }
471 } else
472 chunk->resent = 1;
473
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 has_data = 1;
475 }
476
477 padding = WORD_ROUND(chunk->skb->len) - chunk->skb->len;
478 if (padding)
479 memset(skb_put(chunk->skb, padding), 0, padding);
480
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700481 /* if this is the auth chunk that we are adding,
482 * store pointer where it will be added and put
483 * the auth into the packet.
484 */
485 if (chunk == packet->auth)
486 auth = skb_tail_pointer(nskb);
487
488 cksum_buf_len += chunk->skb->len;
489 memcpy(skb_put(nskb, chunk->skb->len),
Sridhar Samudrala503b55f2006-06-17 22:57:28 -0700490 chunk->skb->data, chunk->skb->len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
492 SCTP_DEBUG_PRINTK("%s %p[%s] %s 0x%x, %s %d, %s %d, %s %d\n",
493 "*** Chunk", chunk,
494 sctp_cname(SCTP_ST_CHUNK(
495 chunk->chunk_hdr->type)),
496 chunk->has_tsn ? "TSN" : "No TSN",
497 chunk->has_tsn ?
498 ntohl(chunk->subh.data_hdr->tsn) : 0,
499 "length", ntohs(chunk->chunk_hdr->length),
500 "chunk->skb->len", chunk->skb->len,
501 "rtt_in_progress", chunk->rtt_in_progress);
502
503 /*
504 * If this is a control chunk, this is our last
505 * reference. Free data chunks after they've been
506 * acknowledged or have failed.
507 */
508 if (!sctp_chunk_is_data(chunk))
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900509 sctp_chunk_free(chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 }
511
Vlad Yasevich4cd57c82007-09-16 19:32:45 -0700512 /* SCTP-AUTH, Section 6.2
513 * The sender MUST calculate the MAC as described in RFC2104 [2]
514 * using the hash function H as described by the MAC Identifier and
515 * the shared association key K based on the endpoint pair shared key
516 * described by the shared key identifier. The 'data' used for the
517 * computation of the AUTH-chunk is given by the AUTH chunk with its
518 * HMAC field set to zero (as shown in Figure 6) followed by all
519 * chunks that are placed after the AUTH chunk in the SCTP packet.
520 */
521 if (auth)
522 sctp_auth_calculate_hmac(asoc, nskb,
523 (struct sctp_auth_chunk *)auth,
524 GFP_ATOMIC);
525
526 /* 2) Calculate the Adler-32 checksum of the whole packet,
527 * including the SCTP common header and all the
528 * chunks.
529 *
530 * Note: Adler-32 is no longer applicable, as has been replaced
531 * by CRC32-C as described in <draft-ietf-tsvwg-sctpcsum-02.txt>.
532 */
Lucas Nussbaum06e86802009-02-13 08:33:41 +0000533 if (!sctp_checksum_disable && !(dst->dev->features & NETIF_F_NO_CSUM)) {
Vlad Yasevich4458f042009-02-13 08:33:42 +0000534 __u32 crc32 = sctp_start_cksum((__u8 *)sh, cksum_buf_len);
535
536 /* 3) Put the resultant value into the checksum field in the
537 * common header, and leave the rest of the bits unchanged.
538 */
539 sh->checksum = sctp_end_cksum(crc32);
Vlad Yasevicha3028b82008-09-18 02:48:25 -0700540 } else
541 nskb->ip_summed = CHECKSUM_UNNECESSARY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 /* IP layer ECN support
544 * From RFC 2481
545 * "The ECN-Capable Transport (ECT) bit would be set by the
546 * data sender to indicate that the end-points of the
547 * transport protocol are ECN-capable."
548 *
549 * Now setting the ECT bit all the time, as it should not cause
550 * any problems protocol-wise even if our peer ignores it.
551 *
552 * Note: The works for IPv6 layer checks this bit too later
553 * in transmission. See IP6_ECN_flow_xmit().
554 */
Vlad Yasevichb9031d92008-06-04 12:40:15 -0700555 (*tp->af_specific->ecn_capable)(nskb->sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556
557 /* Set up the IP options. */
558 /* BUG: not implemented
559 * For v4 this all lives somewhere in sk->sk_opt...
560 */
561
562 /* Dump that on IP! */
563 if (asoc && asoc->peer.last_sent_to != tp) {
564 /* Considering the multiple CPU scenario, this is a
565 * "correcter" place for last_sent_to. --xguo
566 */
567 asoc->peer.last_sent_to = tp;
568 }
569
570 if (has_data) {
571 struct timer_list *timer;
572 unsigned long timeout;
573
574 tp->last_time_used = jiffies;
575
576 /* Restart the AUTOCLOSE timer when sending data. */
577 if (sctp_state(asoc, ESTABLISHED) && asoc->autoclose) {
578 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
579 timeout = asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
580
581 if (!mod_timer(timer, jiffies + timeout))
582 sctp_association_hold(asoc);
583 }
584 }
585
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 SCTP_DEBUG_PRINTK("***sctp_transmit_packet*** skb len %d\n",
587 nskb->len);
588
Herbert Xuf8803742008-08-03 21:15:08 -0700589 nskb->local_df = packet->ipfragok;
590 (*tp->af_specific->sctp_xmit)(nskb, tp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
592out:
593 packet->size = packet->overhead;
594 return err;
595no_route:
596 kfree_skb(nskb);
Pavel Emelyanov7c73a6f2008-07-16 20:20:11 -0700597 IP_INC_STATS_BH(&init_net, IPSTATS_MIB_OUTNOROUTES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598
599 /* FIXME: Returning the 'err' will effect all the associations
600 * associated with a socket, although only one of the paths of the
601 * association is unreachable.
602 * The real failure of a transport or association can be passed on
603 * to the user via notifications. So setting this error may not be
604 * required.
605 */
606 /* err = -EHOSTUNREACH; */
607err:
608 /* Control chunks are unreliable so just drop them. DATA chunks
609 * will get resent or dropped later.
610 */
611
David S. Miller79af02c2005-07-08 21:47:49 -0700612 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
613 list_del_init(&chunk->list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 if (!sctp_chunk_is_data(chunk))
YOSHIFUJI Hideakid808ad92007-02-09 23:25:18 +0900615 sctp_chunk_free(chunk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 }
617 goto out;
618nomem:
619 err = -ENOMEM;
620 goto err;
621}
622
623/********************************************************************
624 * 2nd Level Abstractions
625 ********************************************************************/
626
627/* This private function handles the specifics of appending DATA chunks. */
628static sctp_xmit_t sctp_packet_append_data(struct sctp_packet *packet,
629 struct sctp_chunk *chunk)
630{
631 sctp_xmit_t retval = SCTP_XMIT_OK;
632 size_t datasize, rwnd, inflight;
633 struct sctp_transport *transport = packet->transport;
634 __u32 max_burst_bytes;
635 struct sctp_association *asoc = transport->asoc;
636 struct sctp_sock *sp = sctp_sk(asoc->base.sk);
637 struct sctp_outq *q = &asoc->outqueue;
638
639 /* RFC 2960 6.1 Transmission of DATA Chunks
640 *
641 * A) At any given time, the data sender MUST NOT transmit new data to
642 * any destination transport address if its peer's rwnd indicates
643 * that the peer has no buffer space (i.e. rwnd is 0, see Section
644 * 6.2.1). However, regardless of the value of rwnd (including if it
645 * is 0), the data sender can always have one DATA chunk in flight to
646 * the receiver if allowed by cwnd (see rule B below). This rule
647 * allows the sender to probe for a change in rwnd that the sender
648 * missed due to the SACK having been lost in transit from the data
649 * receiver to the data sender.
650 */
651
652 rwnd = asoc->peer.rwnd;
653 inflight = asoc->outqueue.outstanding_bytes;
654
655 datasize = sctp_data_size(chunk);
656
657 if (datasize > rwnd) {
658 if (inflight > 0) {
659 /* We have (at least) one data chunk in flight,
660 * so we can't fall back to rule 6.1 B).
661 */
662 retval = SCTP_XMIT_RWND_FULL;
663 goto finish;
664 }
665 }
666
667 /* sctpimpguide-05 2.14.2
668 * D) When the time comes for the sender to
669 * transmit new DATA chunks, the protocol parameter Max.Burst MUST
670 * first be applied to limit how many new DATA chunks may be sent.
671 * The limit is applied by adjusting cwnd as follows:
672 * if ((flightsize + Max.Burst * MTU) < cwnd)
673 * cwnd = flightsize + Max.Burst * MTU
674 */
Frank Filz52ccb8e2005-12-22 11:36:46 -0800675 max_burst_bytes = asoc->max_burst * asoc->pathmtu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 if ((transport->flight_size + max_burst_bytes) < transport->cwnd) {
677 transport->cwnd = transport->flight_size + max_burst_bytes;
678 SCTP_DEBUG_PRINTK("%s: cwnd limited by max_burst: "
679 "transport: %p, cwnd: %d, "
680 "ssthresh: %d, flight_size: %d, "
681 "pba: %d\n",
Harvey Harrison0dc47872008-03-05 20:47:47 -0800682 __func__, transport,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 transport->cwnd,
684 transport->ssthresh,
685 transport->flight_size,
686 transport->partial_bytes_acked);
687 }
688
689 /* RFC 2960 6.1 Transmission of DATA Chunks
690 *
691 * B) At any given time, the sender MUST NOT transmit new data
692 * to a given transport address if it has cwnd or more bytes
693 * of data outstanding to that transport address.
694 */
695 /* RFC 7.2.4 & the Implementers Guide 2.8.
696 *
697 * 3) ...
698 * When a Fast Retransmit is being performed the sender SHOULD
699 * ignore the value of cwnd and SHOULD NOT delay retransmission.
700 */
Neil Hormanc226ef92008-07-25 12:44:09 -0400701 if (chunk->fast_retransmit != SCTP_NEED_FRTX)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 if (transport->flight_size >= transport->cwnd) {
703 retval = SCTP_XMIT_RWND_FULL;
704 goto finish;
705 }
706
707 /* Nagle's algorithm to solve small-packet problem:
708 * Inhibit the sending of new chunks when new outgoing data arrives
709 * if any previously transmitted data on the connection remains
710 * unacknowledged.
711 */
712 if (!sp->nodelay && sctp_packet_empty(packet) &&
713 q->outstanding_bytes && sctp_state(asoc, ESTABLISHED)) {
714 unsigned len = datasize + q->out_qlen;
715
716 /* Check whether this chunk and all the rest of pending
717 * data will fit or delay in hopes of bundling a full
718 * sized packet.
719 */
Sridhar Samudralacd497882006-09-29 17:09:05 -0700720 if (len < asoc->frag_point) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 retval = SCTP_XMIT_NAGLE_DELAY;
722 goto finish;
723 }
724 }
725
726 /* Keep track of how many bytes are in flight over this transport. */
727 transport->flight_size += datasize;
728
729 /* Keep track of how many bytes are in flight to the receiver. */
730 asoc->outqueue.outstanding_bytes += datasize;
731
Sridhar Samudralacd497882006-09-29 17:09:05 -0700732 /* Update our view of the receiver's rwnd. Include sk_buff overhead
733 * while updating peer.rwnd so that it reduces the chances of a
734 * receiver running out of receive buffer space even when receive
735 * window is still open. This can happen when a sender is sending
736 * sending small messages.
737 */
738 datasize += sizeof(struct sk_buff);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 if (datasize < rwnd)
740 rwnd -= datasize;
741 else
742 rwnd = 0;
743
744 asoc->peer.rwnd = rwnd;
745 /* Has been accepted for transmission. */
746 if (!asoc->peer.prsctp_capable)
747 chunk->msg->can_abandon = 0;
748
749finish:
750 return retval;
751}