Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* SCTP kernel reference Implementation |
2 | * (C) Copyright IBM Corp. 2001, 2004 | |
3 | * Copyright (c) 1999-2000 Cisco, Inc. | |
4 | * Copyright (c) 1999-2001 Motorola, Inc. | |
5 | * | |
6 | * This file is part of the SCTP kernel reference Implementation | |
7 | * | |
8 | * These functions handle output processing. | |
9 | * | |
10 | * The SCTP reference implementation is free software; | |
11 | * 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 | * | |
16 | * The SCTP reference implementation is distributed in the hope that it | |
17 | * 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> | |
53 | ||
54 | #ifndef TEST_FRAME | |
55 | #include <net/tcp.h> | |
56 | #endif /* TEST_FRAME (not defined) */ | |
57 | ||
58 | #include <linux/socket.h> /* for sa_family_t */ | |
59 | #include <net/sock.h> | |
60 | ||
61 | #include <net/sctp/sctp.h> | |
62 | #include <net/sctp/sm.h> | |
63 | ||
64 | /* Forward declarations for private helpers. */ | |
65 | static sctp_xmit_t sctp_packet_append_data(struct sctp_packet *packet, | |
66 | struct sctp_chunk *chunk); | |
67 | ||
68 | /* Config a packet. | |
69 | * This appears to be a followup set of initializations. | |
70 | */ | |
71 | struct sctp_packet *sctp_packet_config(struct sctp_packet *packet, | |
72 | __u32 vtag, int ecn_capable) | |
73 | { | |
74 | struct sctp_chunk *chunk = NULL; | |
75 | ||
76 | SCTP_DEBUG_PRINTK("%s: packet:%p vtag:0x%x\n", __FUNCTION__, | |
77 | packet, vtag); | |
78 | ||
79 | packet->vtag = vtag; | |
80 | packet->has_cookie_echo = 0; | |
81 | packet->has_sack = 0; | |
a29a5bd4 | 82 | packet->has_auth = 0; |
1da177e4 | 83 | packet->ipfragok = 0; |
a29a5bd4 | 84 | packet->auth = NULL; |
1da177e4 LT |
85 | |
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 | |
d808ad9a YH |
90 | * any other chunks get appended. |
91 | */ | |
1da177e4 LT |
92 | if (chunk) |
93 | sctp_packet_append_chunk(packet, chunk); | |
94 | } | |
95 | ||
96 | return packet; | |
97 | } | |
98 | ||
99 | /* Initialize the packet structure. */ | |
100 | struct 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 | ||
107 | SCTP_DEBUG_PRINTK("%s: packet:%p transport:%p\n", __FUNCTION__, | |
108 | packet, transport); | |
109 | ||
110 | packet->transport = transport; | |
111 | packet->source_port = sport; | |
112 | packet->destination_port = dport; | |
79af02c2 | 113 | INIT_LIST_HEAD(&packet->chunk_list); |
1da177e4 | 114 | if (asoc) { |
d808ad9a YH |
115 | struct sctp_sock *sp = sctp_sk(asoc->base.sk); |
116 | overhead = sp->pf->af->net_header_len; | |
1da177e4 LT |
117 | } 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; | |
a29a5bd4 | 126 | packet->has_auth = 0; |
1da177e4 LT |
127 | packet->ipfragok = 0; |
128 | packet->malloced = 0; | |
a29a5bd4 | 129 | packet->auth = NULL; |
1da177e4 LT |
130 | return packet; |
131 | } | |
132 | ||
133 | /* Free a packet. */ | |
134 | void sctp_packet_free(struct sctp_packet *packet) | |
135 | { | |
79af02c2 | 136 | struct sctp_chunk *chunk, *tmp; |
1da177e4 LT |
137 | |
138 | SCTP_DEBUG_PRINTK("%s: packet:%p\n", __FUNCTION__, packet); | |
139 | ||
79af02c2 DM |
140 | list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) { |
141 | list_del_init(&chunk->list); | |
1da177e4 | 142 | sctp_chunk_free(chunk); |
79af02c2 | 143 | } |
1da177e4 LT |
144 | |
145 | if (packet->malloced) | |
146 | kfree(packet); | |
147 | } | |
148 | ||
149 | /* This routine tries to append the chunk to the offered packet. If adding | |
150 | * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk | |
151 | * is not present in the packet, it transmits the input packet. | |
152 | * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long | |
153 | * as it can fit in the packet, but any more data that does not fit in this | |
154 | * packet can be sent only after receiving the COOKIE_ACK. | |
155 | */ | |
156 | sctp_xmit_t sctp_packet_transmit_chunk(struct sctp_packet *packet, | |
157 | struct sctp_chunk *chunk) | |
158 | { | |
159 | sctp_xmit_t retval; | |
160 | int error = 0; | |
161 | ||
162 | SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __FUNCTION__, | |
163 | packet, chunk); | |
164 | ||
165 | switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) { | |
166 | case SCTP_XMIT_PMTU_FULL: | |
167 | if (!packet->has_cookie_echo) { | |
168 | error = sctp_packet_transmit(packet); | |
169 | if (error < 0) | |
170 | chunk->skb->sk->sk_err = -error; | |
171 | ||
172 | /* If we have an empty packet, then we can NOT ever | |
173 | * return PMTU_FULL. | |
174 | */ | |
175 | retval = sctp_packet_append_chunk(packet, chunk); | |
176 | } | |
177 | break; | |
178 | ||
179 | case SCTP_XMIT_RWND_FULL: | |
180 | case SCTP_XMIT_OK: | |
181 | case SCTP_XMIT_NAGLE_DELAY: | |
182 | break; | |
3ff50b79 | 183 | } |
1da177e4 LT |
184 | |
185 | return retval; | |
186 | } | |
187 | ||
188 | /* Try to bundle a SACK with the packet. */ | |
189 | static sctp_xmit_t sctp_packet_bundle_sack(struct sctp_packet *pkt, | |
190 | struct sctp_chunk *chunk) | |
191 | { | |
192 | sctp_xmit_t retval = SCTP_XMIT_OK; | |
193 | ||
194 | /* If sending DATA and haven't aleady bundled a SACK, try to | |
195 | * bundle one in to the packet. | |
196 | */ | |
197 | if (sctp_chunk_is_data(chunk) && !pkt->has_sack && | |
198 | !pkt->has_cookie_echo) { | |
199 | struct sctp_association *asoc; | |
200 | asoc = pkt->transport->asoc; | |
201 | ||
202 | if (asoc->a_rwnd > asoc->rwnd) { | |
203 | struct sctp_chunk *sack; | |
204 | asoc->a_rwnd = asoc->rwnd; | |
205 | sack = sctp_make_sack(asoc); | |
206 | if (sack) { | |
207 | struct timer_list *timer; | |
208 | retval = sctp_packet_append_chunk(pkt, sack); | |
209 | asoc->peer.sack_needed = 0; | |
210 | timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK]; | |
211 | if (timer_pending(timer) && del_timer(timer)) | |
212 | sctp_association_put(asoc); | |
213 | } | |
214 | } | |
215 | } | |
216 | return retval; | |
217 | } | |
218 | ||
219 | /* Append a chunk to the offered packet reporting back any inability to do | |
220 | * so. | |
221 | */ | |
222 | sctp_xmit_t sctp_packet_append_chunk(struct sctp_packet *packet, | |
223 | struct sctp_chunk *chunk) | |
224 | { | |
225 | sctp_xmit_t retval = SCTP_XMIT_OK; | |
226 | __u16 chunk_len = WORD_ROUND(ntohs(chunk->chunk_hdr->length)); | |
227 | size_t psize; | |
228 | size_t pmtu; | |
229 | int too_big; | |
230 | ||
231 | SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __FUNCTION__, packet, | |
232 | chunk); | |
233 | ||
234 | retval = sctp_packet_bundle_sack(packet, chunk); | |
235 | psize = packet->size; | |
236 | ||
237 | if (retval != SCTP_XMIT_OK) | |
238 | goto finish; | |
239 | ||
240 | pmtu = ((packet->transport->asoc) ? | |
52ccb8e9 FF |
241 | (packet->transport->asoc->pathmtu) : |
242 | (packet->transport->pathmtu)); | |
1da177e4 LT |
243 | |
244 | too_big = (psize + chunk_len > pmtu); | |
245 | ||
246 | /* Decide if we need to fragment or resubmit later. */ | |
247 | if (too_big) { | |
248 | /* Both control chunks and data chunks with TSNs are | |
249 | * non-fragmentable. | |
250 | */ | |
251 | if (sctp_packet_empty(packet) || !sctp_chunk_is_data(chunk)) { | |
252 | /* We no longer do re-fragmentation. | |
253 | * Just fragment at the IP layer, if we | |
254 | * actually hit this condition | |
255 | */ | |
256 | packet->ipfragok = 1; | |
257 | goto append; | |
258 | ||
259 | } else { | |
260 | retval = SCTP_XMIT_PMTU_FULL; | |
261 | goto finish; | |
262 | } | |
263 | } | |
264 | ||
265 | append: | |
266 | /* We believe that this chunk is OK to add to the packet (as | |
267 | * long as we have the cwnd for it). | |
268 | */ | |
269 | ||
270 | /* DATA is a special case since we must examine both rwnd and cwnd | |
271 | * before we send DATA. | |
272 | */ | |
273 | if (sctp_chunk_is_data(chunk)) { | |
274 | retval = sctp_packet_append_data(packet, chunk); | |
275 | /* Disallow SACK bundling after DATA. */ | |
276 | packet->has_sack = 1; | |
277 | if (SCTP_XMIT_OK != retval) | |
278 | goto finish; | |
279 | } else if (SCTP_CID_COOKIE_ECHO == chunk->chunk_hdr->type) | |
280 | packet->has_cookie_echo = 1; | |
281 | else if (SCTP_CID_SACK == chunk->chunk_hdr->type) | |
282 | packet->has_sack = 1; | |
283 | ||
284 | /* It is OK to send this chunk. */ | |
79af02c2 | 285 | list_add_tail(&chunk->list, &packet->chunk_list); |
1da177e4 LT |
286 | packet->size += chunk_len; |
287 | chunk->transport = packet->transport; | |
288 | finish: | |
289 | return retval; | |
290 | } | |
291 | ||
292 | /* All packets are sent to the network through this function from | |
293 | * sctp_outq_tail(). | |
294 | * | |
295 | * The return value is a normal kernel error return value. | |
296 | */ | |
297 | int sctp_packet_transmit(struct sctp_packet *packet) | |
298 | { | |
299 | struct sctp_transport *tp = packet->transport; | |
300 | struct sctp_association *asoc = tp->asoc; | |
301 | struct sctphdr *sh; | |
503b55fd | 302 | __u32 crc32 = 0; |
1da177e4 | 303 | struct sk_buff *nskb; |
79af02c2 | 304 | struct sctp_chunk *chunk, *tmp; |
1da177e4 LT |
305 | struct sock *sk; |
306 | int err = 0; | |
307 | int padding; /* How much padding do we need? */ | |
308 | __u8 has_data = 0; | |
503b55fd | 309 | struct dst_entry *dst = tp->dst; |
1da177e4 LT |
310 | |
311 | SCTP_DEBUG_PRINTK("%s: packet:%p\n", __FUNCTION__, packet); | |
312 | ||
313 | /* Do NOT generate a chunkless packet. */ | |
79af02c2 | 314 | if (list_empty(&packet->chunk_list)) |
1da177e4 LT |
315 | return err; |
316 | ||
317 | /* Set up convenience variables... */ | |
79af02c2 | 318 | chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list); |
1da177e4 LT |
319 | sk = chunk->skb->sk; |
320 | ||
321 | /* Allocate the new skb. */ | |
594ccc14 | 322 | nskb = alloc_skb(packet->size + LL_MAX_HEADER, GFP_ATOMIC); |
1da177e4 LT |
323 | if (!nskb) |
324 | goto nomem; | |
325 | ||
326 | /* Make sure the outbound skb has enough header room reserved. */ | |
594ccc14 | 327 | skb_reserve(nskb, packet->overhead + LL_MAX_HEADER); |
1da177e4 LT |
328 | |
329 | /* Set the owning socket so that we know where to get the | |
330 | * destination IP address. | |
331 | */ | |
332 | skb_set_owner_w(nskb, sk); | |
333 | ||
503b55fd SS |
334 | /* The 'obsolete' field of dst is set to 2 when a dst is freed. */ |
335 | if (!dst || (dst->obsolete > 1)) { | |
336 | dst_release(dst); | |
337 | sctp_transport_route(tp, NULL, sctp_sk(sk)); | |
338 | if (asoc && (asoc->param_flags & SPP_PMTUD_ENABLE)) { | |
339 | sctp_assoc_sync_pmtu(asoc); | |
340 | } | |
341 | } | |
342 | nskb->dst = dst_clone(tp->dst); | |
343 | if (!nskb->dst) | |
344 | goto no_route; | |
345 | dst = nskb->dst; | |
346 | ||
1da177e4 LT |
347 | /* Build the SCTP header. */ |
348 | sh = (struct sctphdr *)skb_push(nskb, sizeof(struct sctphdr)); | |
349 | sh->source = htons(packet->source_port); | |
350 | sh->dest = htons(packet->destination_port); | |
351 | ||
352 | /* From 6.8 Adler-32 Checksum Calculation: | |
353 | * After the packet is constructed (containing the SCTP common | |
354 | * header and one or more control or DATA chunks), the | |
355 | * transmitter shall: | |
356 | * | |
357 | * 1) Fill in the proper Verification Tag in the SCTP common | |
358 | * header and initialize the checksum field to 0's. | |
359 | */ | |
360 | sh->vtag = htonl(packet->vtag); | |
361 | sh->checksum = 0; | |
362 | ||
363 | /* 2) Calculate the Adler-32 checksum of the whole packet, | |
364 | * including the SCTP common header and all the | |
365 | * chunks. | |
366 | * | |
367 | * Note: Adler-32 is no longer applicable, as has been replaced | |
368 | * by CRC32-C as described in <draft-ietf-tsvwg-sctpcsum-02.txt>. | |
369 | */ | |
503b55fd SS |
370 | if (!(dst->dev->features & NETIF_F_NO_CSUM)) |
371 | crc32 = sctp_start_cksum((__u8 *)sh, sizeof(struct sctphdr)); | |
1da177e4 LT |
372 | |
373 | /** | |
374 | * 6.10 Bundling | |
375 | * | |
376 | * An endpoint bundles chunks by simply including multiple | |
377 | * chunks in one outbound SCTP packet. ... | |
378 | */ | |
379 | ||
380 | /** | |
381 | * 3.2 Chunk Field Descriptions | |
382 | * | |
383 | * The total length of a chunk (including Type, Length and | |
384 | * Value fields) MUST be a multiple of 4 bytes. If the length | |
385 | * of the chunk is not a multiple of 4 bytes, the sender MUST | |
386 | * pad the chunk with all zero bytes and this padding is not | |
387 | * included in the chunk length field. The sender should | |
388 | * never pad with more than 3 bytes. | |
389 | * | |
390 | * [This whole comment explains WORD_ROUND() below.] | |
391 | */ | |
392 | SCTP_DEBUG_PRINTK("***sctp_transmit_packet***\n"); | |
79af02c2 DM |
393 | list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) { |
394 | list_del_init(&chunk->list); | |
1da177e4 LT |
395 | if (sctp_chunk_is_data(chunk)) { |
396 | ||
397 | if (!chunk->has_tsn) { | |
398 | sctp_chunk_assign_ssn(chunk); | |
399 | sctp_chunk_assign_tsn(chunk); | |
400 | ||
401 | /* 6.3.1 C4) When data is in flight and when allowed | |
402 | * by rule C5, a new RTT measurement MUST be made each | |
403 | * round trip. Furthermore, new RTT measurements | |
404 | * SHOULD be made no more than once per round-trip | |
405 | * for a given destination transport address. | |
406 | */ | |
407 | ||
408 | if (!tp->rto_pending) { | |
409 | chunk->rtt_in_progress = 1; | |
410 | tp->rto_pending = 1; | |
411 | } | |
412 | } else | |
413 | chunk->resent = 1; | |
414 | ||
415 | chunk->sent_at = jiffies; | |
416 | has_data = 1; | |
417 | } | |
418 | ||
419 | padding = WORD_ROUND(chunk->skb->len) - chunk->skb->len; | |
420 | if (padding) | |
421 | memset(skb_put(chunk->skb, padding), 0, padding); | |
422 | ||
503b55fd SS |
423 | if (dst->dev->features & NETIF_F_NO_CSUM) |
424 | memcpy(skb_put(nskb, chunk->skb->len), | |
425 | chunk->skb->data, chunk->skb->len); | |
426 | else | |
427 | crc32 = sctp_update_copy_cksum(skb_put(nskb, | |
428 | chunk->skb->len), | |
429 | chunk->skb->data, | |
430 | chunk->skb->len, crc32); | |
1da177e4 LT |
431 | |
432 | SCTP_DEBUG_PRINTK("%s %p[%s] %s 0x%x, %s %d, %s %d, %s %d\n", | |
433 | "*** Chunk", chunk, | |
434 | sctp_cname(SCTP_ST_CHUNK( | |
435 | chunk->chunk_hdr->type)), | |
436 | chunk->has_tsn ? "TSN" : "No TSN", | |
437 | chunk->has_tsn ? | |
438 | ntohl(chunk->subh.data_hdr->tsn) : 0, | |
439 | "length", ntohs(chunk->chunk_hdr->length), | |
440 | "chunk->skb->len", chunk->skb->len, | |
441 | "rtt_in_progress", chunk->rtt_in_progress); | |
442 | ||
443 | /* | |
444 | * If this is a control chunk, this is our last | |
445 | * reference. Free data chunks after they've been | |
446 | * acknowledged or have failed. | |
447 | */ | |
448 | if (!sctp_chunk_is_data(chunk)) | |
d808ad9a | 449 | sctp_chunk_free(chunk); |
1da177e4 LT |
450 | } |
451 | ||
452 | /* Perform final transformation on checksum. */ | |
503b55fd SS |
453 | if (!(dst->dev->features & NETIF_F_NO_CSUM)) |
454 | crc32 = sctp_end_cksum(crc32); | |
1da177e4 LT |
455 | |
456 | /* 3) Put the resultant value into the checksum field in the | |
457 | * common header, and leave the rest of the bits unchanged. | |
458 | */ | |
459 | sh->checksum = htonl(crc32); | |
460 | ||
461 | /* IP layer ECN support | |
462 | * From RFC 2481 | |
463 | * "The ECN-Capable Transport (ECT) bit would be set by the | |
464 | * data sender to indicate that the end-points of the | |
465 | * transport protocol are ECN-capable." | |
466 | * | |
467 | * Now setting the ECT bit all the time, as it should not cause | |
468 | * any problems protocol-wise even if our peer ignores it. | |
469 | * | |
470 | * Note: The works for IPv6 layer checks this bit too later | |
471 | * in transmission. See IP6_ECN_flow_xmit(). | |
472 | */ | |
473 | INET_ECN_xmit(nskb->sk); | |
474 | ||
475 | /* Set up the IP options. */ | |
476 | /* BUG: not implemented | |
477 | * For v4 this all lives somewhere in sk->sk_opt... | |
478 | */ | |
479 | ||
480 | /* Dump that on IP! */ | |
481 | if (asoc && asoc->peer.last_sent_to != tp) { | |
482 | /* Considering the multiple CPU scenario, this is a | |
483 | * "correcter" place for last_sent_to. --xguo | |
484 | */ | |
485 | asoc->peer.last_sent_to = tp; | |
486 | } | |
487 | ||
488 | if (has_data) { | |
489 | struct timer_list *timer; | |
490 | unsigned long timeout; | |
491 | ||
492 | tp->last_time_used = jiffies; | |
493 | ||
494 | /* Restart the AUTOCLOSE timer when sending data. */ | |
495 | if (sctp_state(asoc, ESTABLISHED) && asoc->autoclose) { | |
496 | timer = &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE]; | |
497 | timeout = asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE]; | |
498 | ||
499 | if (!mod_timer(timer, jiffies + timeout)) | |
500 | sctp_association_hold(asoc); | |
501 | } | |
502 | } | |
503 | ||
1da177e4 LT |
504 | SCTP_DEBUG_PRINTK("***sctp_transmit_packet*** skb len %d\n", |
505 | nskb->len); | |
506 | ||
52ccb8e9 FF |
507 | if (tp->param_flags & SPP_PMTUD_ENABLE) |
508 | (*tp->af_specific->sctp_xmit)(nskb, tp, packet->ipfragok); | |
509 | else | |
510 | (*tp->af_specific->sctp_xmit)(nskb, tp, 1); | |
1da177e4 LT |
511 | |
512 | out: | |
513 | packet->size = packet->overhead; | |
514 | return err; | |
515 | no_route: | |
516 | kfree_skb(nskb); | |
517 | IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES); | |
518 | ||
519 | /* FIXME: Returning the 'err' will effect all the associations | |
520 | * associated with a socket, although only one of the paths of the | |
521 | * association is unreachable. | |
522 | * The real failure of a transport or association can be passed on | |
523 | * to the user via notifications. So setting this error may not be | |
524 | * required. | |
525 | */ | |
526 | /* err = -EHOSTUNREACH; */ | |
527 | err: | |
528 | /* Control chunks are unreliable so just drop them. DATA chunks | |
529 | * will get resent or dropped later. | |
530 | */ | |
531 | ||
79af02c2 DM |
532 | list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) { |
533 | list_del_init(&chunk->list); | |
1da177e4 | 534 | if (!sctp_chunk_is_data(chunk)) |
d808ad9a | 535 | sctp_chunk_free(chunk); |
1da177e4 LT |
536 | } |
537 | goto out; | |
538 | nomem: | |
539 | err = -ENOMEM; | |
540 | goto err; | |
541 | } | |
542 | ||
543 | /******************************************************************** | |
544 | * 2nd Level Abstractions | |
545 | ********************************************************************/ | |
546 | ||
547 | /* This private function handles the specifics of appending DATA chunks. */ | |
548 | static sctp_xmit_t sctp_packet_append_data(struct sctp_packet *packet, | |
549 | struct sctp_chunk *chunk) | |
550 | { | |
551 | sctp_xmit_t retval = SCTP_XMIT_OK; | |
552 | size_t datasize, rwnd, inflight; | |
553 | struct sctp_transport *transport = packet->transport; | |
554 | __u32 max_burst_bytes; | |
555 | struct sctp_association *asoc = transport->asoc; | |
556 | struct sctp_sock *sp = sctp_sk(asoc->base.sk); | |
557 | struct sctp_outq *q = &asoc->outqueue; | |
558 | ||
559 | /* RFC 2960 6.1 Transmission of DATA Chunks | |
560 | * | |
561 | * A) At any given time, the data sender MUST NOT transmit new data to | |
562 | * any destination transport address if its peer's rwnd indicates | |
563 | * that the peer has no buffer space (i.e. rwnd is 0, see Section | |
564 | * 6.2.1). However, regardless of the value of rwnd (including if it | |
565 | * is 0), the data sender can always have one DATA chunk in flight to | |
566 | * the receiver if allowed by cwnd (see rule B below). This rule | |
567 | * allows the sender to probe for a change in rwnd that the sender | |
568 | * missed due to the SACK having been lost in transit from the data | |
569 | * receiver to the data sender. | |
570 | */ | |
571 | ||
572 | rwnd = asoc->peer.rwnd; | |
573 | inflight = asoc->outqueue.outstanding_bytes; | |
574 | ||
575 | datasize = sctp_data_size(chunk); | |
576 | ||
577 | if (datasize > rwnd) { | |
578 | if (inflight > 0) { | |
579 | /* We have (at least) one data chunk in flight, | |
580 | * so we can't fall back to rule 6.1 B). | |
581 | */ | |
582 | retval = SCTP_XMIT_RWND_FULL; | |
583 | goto finish; | |
584 | } | |
585 | } | |
586 | ||
587 | /* sctpimpguide-05 2.14.2 | |
588 | * D) When the time comes for the sender to | |
589 | * transmit new DATA chunks, the protocol parameter Max.Burst MUST | |
590 | * first be applied to limit how many new DATA chunks may be sent. | |
591 | * The limit is applied by adjusting cwnd as follows: | |
592 | * if ((flightsize + Max.Burst * MTU) < cwnd) | |
593 | * cwnd = flightsize + Max.Burst * MTU | |
594 | */ | |
52ccb8e9 | 595 | max_burst_bytes = asoc->max_burst * asoc->pathmtu; |
1da177e4 LT |
596 | if ((transport->flight_size + max_burst_bytes) < transport->cwnd) { |
597 | transport->cwnd = transport->flight_size + max_burst_bytes; | |
598 | SCTP_DEBUG_PRINTK("%s: cwnd limited by max_burst: " | |
599 | "transport: %p, cwnd: %d, " | |
600 | "ssthresh: %d, flight_size: %d, " | |
601 | "pba: %d\n", | |
602 | __FUNCTION__, transport, | |
603 | transport->cwnd, | |
604 | transport->ssthresh, | |
605 | transport->flight_size, | |
606 | transport->partial_bytes_acked); | |
607 | } | |
608 | ||
609 | /* RFC 2960 6.1 Transmission of DATA Chunks | |
610 | * | |
611 | * B) At any given time, the sender MUST NOT transmit new data | |
612 | * to a given transport address if it has cwnd or more bytes | |
613 | * of data outstanding to that transport address. | |
614 | */ | |
615 | /* RFC 7.2.4 & the Implementers Guide 2.8. | |
616 | * | |
617 | * 3) ... | |
618 | * When a Fast Retransmit is being performed the sender SHOULD | |
619 | * ignore the value of cwnd and SHOULD NOT delay retransmission. | |
620 | */ | |
27852c26 | 621 | if (chunk->fast_retransmit <= 0) |
1da177e4 LT |
622 | if (transport->flight_size >= transport->cwnd) { |
623 | retval = SCTP_XMIT_RWND_FULL; | |
624 | goto finish; | |
625 | } | |
626 | ||
627 | /* Nagle's algorithm to solve small-packet problem: | |
628 | * Inhibit the sending of new chunks when new outgoing data arrives | |
629 | * if any previously transmitted data on the connection remains | |
630 | * unacknowledged. | |
631 | */ | |
632 | if (!sp->nodelay && sctp_packet_empty(packet) && | |
633 | q->outstanding_bytes && sctp_state(asoc, ESTABLISHED)) { | |
634 | unsigned len = datasize + q->out_qlen; | |
635 | ||
636 | /* Check whether this chunk and all the rest of pending | |
637 | * data will fit or delay in hopes of bundling a full | |
638 | * sized packet. | |
639 | */ | |
cd497885 | 640 | if (len < asoc->frag_point) { |
1da177e4 LT |
641 | retval = SCTP_XMIT_NAGLE_DELAY; |
642 | goto finish; | |
643 | } | |
644 | } | |
645 | ||
646 | /* Keep track of how many bytes are in flight over this transport. */ | |
647 | transport->flight_size += datasize; | |
648 | ||
649 | /* Keep track of how many bytes are in flight to the receiver. */ | |
650 | asoc->outqueue.outstanding_bytes += datasize; | |
651 | ||
cd497885 SS |
652 | /* Update our view of the receiver's rwnd. Include sk_buff overhead |
653 | * while updating peer.rwnd so that it reduces the chances of a | |
654 | * receiver running out of receive buffer space even when receive | |
655 | * window is still open. This can happen when a sender is sending | |
656 | * sending small messages. | |
657 | */ | |
658 | datasize += sizeof(struct sk_buff); | |
1da177e4 LT |
659 | if (datasize < rwnd) |
660 | rwnd -= datasize; | |
661 | else | |
662 | rwnd = 0; | |
663 | ||
664 | asoc->peer.rwnd = rwnd; | |
665 | /* Has been accepted for transmission. */ | |
666 | if (!asoc->peer.prsctp_capable) | |
667 | chunk->msg->can_abandon = 0; | |
668 | ||
669 | finish: | |
670 | return retval; | |
671 | } |