Loading...
1/* SCTP kernel implementation
2 * (C) Copyright IBM Corp. 2001, 2004
3 * Copyright (c) 1999-2000 Cisco, Inc.
4 * Copyright (c) 1999-2001 Motorola, Inc.
5 *
6 * This file is part of the SCTP kernel implementation
7 *
8 * These functions handle output processing.
9 *
10 * This SCTP 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 * This SCTP 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, see
24 * <http://www.gnu.org/licenses/>.
25 *
26 * Please send any bug reports or fixes you make to the
27 * email address(es):
28 * lksctp developers <linux-sctp@vger.kernel.org>
29 *
30 * Written or modified by:
31 * La Monte H.P. Yarroll <piggy@acm.org>
32 * Karl Knutson <karl@athena.chicago.il.us>
33 * Jon Grimm <jgrimm@austin.ibm.com>
34 * Sridhar Samudrala <sri@us.ibm.com>
35 */
36
37#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
38
39#include <linux/types.h>
40#include <linux/kernel.h>
41#include <linux/wait.h>
42#include <linux/time.h>
43#include <linux/ip.h>
44#include <linux/ipv6.h>
45#include <linux/init.h>
46#include <linux/slab.h>
47#include <net/inet_ecn.h>
48#include <net/ip.h>
49#include <net/icmp.h>
50#include <net/net_namespace.h>
51
52#include <linux/socket.h> /* for sa_family_t */
53#include <net/sock.h>
54
55#include <net/sctp/sctp.h>
56#include <net/sctp/sm.h>
57#include <net/sctp/checksum.h>
58
59/* Forward declarations for private helpers. */
60static enum sctp_xmit __sctp_packet_append_chunk(struct sctp_packet *packet,
61 struct sctp_chunk *chunk);
62static enum sctp_xmit sctp_packet_can_append_data(struct sctp_packet *packet,
63 struct sctp_chunk *chunk);
64static void sctp_packet_append_data(struct sctp_packet *packet,
65 struct sctp_chunk *chunk);
66static enum sctp_xmit sctp_packet_will_fit(struct sctp_packet *packet,
67 struct sctp_chunk *chunk,
68 u16 chunk_len);
69
70static void sctp_packet_reset(struct sctp_packet *packet)
71{
72 /* sctp_packet_transmit() relies on this to reset size to the
73 * current overhead after sending packets.
74 */
75 packet->size = packet->overhead;
76
77 packet->has_cookie_echo = 0;
78 packet->has_sack = 0;
79 packet->has_data = 0;
80 packet->has_auth = 0;
81 packet->ipfragok = 0;
82 packet->auth = NULL;
83}
84
85/* Config a packet.
86 * This appears to be a followup set of initializations.
87 */
88void sctp_packet_config(struct sctp_packet *packet, __u32 vtag,
89 int ecn_capable)
90{
91 struct sctp_transport *tp = packet->transport;
92 struct sctp_association *asoc = tp->asoc;
93 struct sock *sk;
94 size_t overhead = sizeof(struct ipv6hdr) + sizeof(struct sctphdr);
95
96 pr_debug("%s: packet:%p vtag:0x%x\n", __func__, packet, vtag);
97 packet->vtag = vtag;
98
99 /* do the following jobs only once for a flush schedule */
100 if (!sctp_packet_empty(packet))
101 return;
102
103 /* set packet max_size with pathmtu, then calculate overhead */
104 packet->max_size = tp->pathmtu;
105 if (asoc) {
106 struct sctp_sock *sp = sctp_sk(asoc->base.sk);
107 struct sctp_af *af = sp->pf->af;
108
109 overhead = af->net_header_len +
110 af->ip_options_len(asoc->base.sk);
111 overhead += sizeof(struct sctphdr);
112 packet->overhead = overhead;
113 packet->size = overhead;
114 } else {
115 packet->overhead = overhead;
116 packet->size = overhead;
117 return;
118 }
119
120 /* update dst or transport pathmtu if in need */
121 sk = asoc->base.sk;
122 if (!sctp_transport_dst_check(tp)) {
123 sctp_transport_route(tp, NULL, sctp_sk(sk));
124 if (asoc->param_flags & SPP_PMTUD_ENABLE)
125 sctp_assoc_sync_pmtu(asoc);
126 } else if (!sctp_transport_pmtu_check(tp)) {
127 if (asoc->param_flags & SPP_PMTUD_ENABLE)
128 sctp_assoc_sync_pmtu(asoc);
129 }
130
131 /* If there a is a prepend chunk stick it on the list before
132 * any other chunks get appended.
133 */
134 if (ecn_capable) {
135 struct sctp_chunk *chunk = sctp_get_ecne_prepend(asoc);
136
137 if (chunk)
138 sctp_packet_append_chunk(packet, chunk);
139 }
140
141 if (!tp->dst)
142 return;
143
144 /* set packet max_size with gso_max_size if gso is enabled*/
145 rcu_read_lock();
146 if (__sk_dst_get(sk) != tp->dst) {
147 dst_hold(tp->dst);
148 sk_setup_caps(sk, tp->dst);
149 }
150 packet->max_size = sk_can_gso(sk) ? tp->dst->dev->gso_max_size
151 : asoc->pathmtu;
152 rcu_read_unlock();
153}
154
155/* Initialize the packet structure. */
156void sctp_packet_init(struct sctp_packet *packet,
157 struct sctp_transport *transport,
158 __u16 sport, __u16 dport)
159{
160 pr_debug("%s: packet:%p transport:%p\n", __func__, packet, transport);
161
162 packet->transport = transport;
163 packet->source_port = sport;
164 packet->destination_port = dport;
165 INIT_LIST_HEAD(&packet->chunk_list);
166 /* The overhead will be calculated by sctp_packet_config() */
167 packet->overhead = 0;
168 sctp_packet_reset(packet);
169 packet->vtag = 0;
170}
171
172/* Free a packet. */
173void sctp_packet_free(struct sctp_packet *packet)
174{
175 struct sctp_chunk *chunk, *tmp;
176
177 pr_debug("%s: packet:%p\n", __func__, packet);
178
179 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
180 list_del_init(&chunk->list);
181 sctp_chunk_free(chunk);
182 }
183}
184
185/* This routine tries to append the chunk to the offered packet. If adding
186 * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk
187 * is not present in the packet, it transmits the input packet.
188 * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long
189 * as it can fit in the packet, but any more data that does not fit in this
190 * packet can be sent only after receiving the COOKIE_ACK.
191 */
192enum sctp_xmit sctp_packet_transmit_chunk(struct sctp_packet *packet,
193 struct sctp_chunk *chunk,
194 int one_packet, gfp_t gfp)
195{
196 enum sctp_xmit retval;
197
198 pr_debug("%s: packet:%p size:%zu chunk:%p size:%d\n", __func__,
199 packet, packet->size, chunk, chunk->skb ? chunk->skb->len : -1);
200
201 switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) {
202 case SCTP_XMIT_PMTU_FULL:
203 if (!packet->has_cookie_echo) {
204 int error = 0;
205
206 error = sctp_packet_transmit(packet, gfp);
207 if (error < 0)
208 chunk->skb->sk->sk_err = -error;
209
210 /* If we have an empty packet, then we can NOT ever
211 * return PMTU_FULL.
212 */
213 if (!one_packet)
214 retval = sctp_packet_append_chunk(packet,
215 chunk);
216 }
217 break;
218
219 case SCTP_XMIT_RWND_FULL:
220 case SCTP_XMIT_OK:
221 case SCTP_XMIT_DELAY:
222 break;
223 }
224
225 return retval;
226}
227
228/* Try to bundle an auth chunk into the packet. */
229static enum sctp_xmit sctp_packet_bundle_auth(struct sctp_packet *pkt,
230 struct sctp_chunk *chunk)
231{
232 struct sctp_association *asoc = pkt->transport->asoc;
233 enum sctp_xmit retval = SCTP_XMIT_OK;
234 struct sctp_chunk *auth;
235
236 /* if we don't have an association, we can't do authentication */
237 if (!asoc)
238 return retval;
239
240 /* See if this is an auth chunk we are bundling or if
241 * auth is already bundled.
242 */
243 if (chunk->chunk_hdr->type == SCTP_CID_AUTH || pkt->has_auth)
244 return retval;
245
246 /* if the peer did not request this chunk to be authenticated,
247 * don't do it
248 */
249 if (!chunk->auth)
250 return retval;
251
252 auth = sctp_make_auth(asoc, chunk->shkey->key_id);
253 if (!auth)
254 return retval;
255
256 auth->shkey = chunk->shkey;
257 sctp_auth_shkey_hold(auth->shkey);
258
259 retval = __sctp_packet_append_chunk(pkt, auth);
260
261 if (retval != SCTP_XMIT_OK)
262 sctp_chunk_free(auth);
263
264 return retval;
265}
266
267/* Try to bundle a SACK with the packet. */
268static enum sctp_xmit sctp_packet_bundle_sack(struct sctp_packet *pkt,
269 struct sctp_chunk *chunk)
270{
271 enum sctp_xmit retval = SCTP_XMIT_OK;
272
273 /* If sending DATA and haven't aleady bundled a SACK, try to
274 * bundle one in to the packet.
275 */
276 if (sctp_chunk_is_data(chunk) && !pkt->has_sack &&
277 !pkt->has_cookie_echo) {
278 struct sctp_association *asoc;
279 struct timer_list *timer;
280 asoc = pkt->transport->asoc;
281 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
282
283 /* If the SACK timer is running, we have a pending SACK */
284 if (timer_pending(timer)) {
285 struct sctp_chunk *sack;
286
287 if (pkt->transport->sack_generation !=
288 pkt->transport->asoc->peer.sack_generation)
289 return retval;
290
291 asoc->a_rwnd = asoc->rwnd;
292 sack = sctp_make_sack(asoc);
293 if (sack) {
294 retval = __sctp_packet_append_chunk(pkt, sack);
295 if (retval != SCTP_XMIT_OK) {
296 sctp_chunk_free(sack);
297 goto out;
298 }
299 asoc->peer.sack_needed = 0;
300 if (del_timer(timer))
301 sctp_association_put(asoc);
302 }
303 }
304 }
305out:
306 return retval;
307}
308
309
310/* Append a chunk to the offered packet reporting back any inability to do
311 * so.
312 */
313static enum sctp_xmit __sctp_packet_append_chunk(struct sctp_packet *packet,
314 struct sctp_chunk *chunk)
315{
316 __u16 chunk_len = SCTP_PAD4(ntohs(chunk->chunk_hdr->length));
317 enum sctp_xmit retval = SCTP_XMIT_OK;
318
319 /* Check to see if this chunk will fit into the packet */
320 retval = sctp_packet_will_fit(packet, chunk, chunk_len);
321 if (retval != SCTP_XMIT_OK)
322 goto finish;
323
324 /* We believe that this chunk is OK to add to the packet */
325 switch (chunk->chunk_hdr->type) {
326 case SCTP_CID_DATA:
327 case SCTP_CID_I_DATA:
328 /* Account for the data being in the packet */
329 sctp_packet_append_data(packet, chunk);
330 /* Disallow SACK bundling after DATA. */
331 packet->has_sack = 1;
332 /* Disallow AUTH bundling after DATA */
333 packet->has_auth = 1;
334 /* Let it be knows that packet has DATA in it */
335 packet->has_data = 1;
336 /* timestamp the chunk for rtx purposes */
337 chunk->sent_at = jiffies;
338 /* Mainly used for prsctp RTX policy */
339 chunk->sent_count++;
340 break;
341 case SCTP_CID_COOKIE_ECHO:
342 packet->has_cookie_echo = 1;
343 break;
344
345 case SCTP_CID_SACK:
346 packet->has_sack = 1;
347 if (chunk->asoc)
348 chunk->asoc->stats.osacks++;
349 break;
350
351 case SCTP_CID_AUTH:
352 packet->has_auth = 1;
353 packet->auth = chunk;
354 break;
355 }
356
357 /* It is OK to send this chunk. */
358 list_add_tail(&chunk->list, &packet->chunk_list);
359 packet->size += chunk_len;
360 chunk->transport = packet->transport;
361finish:
362 return retval;
363}
364
365/* Append a chunk to the offered packet reporting back any inability to do
366 * so.
367 */
368enum sctp_xmit sctp_packet_append_chunk(struct sctp_packet *packet,
369 struct sctp_chunk *chunk)
370{
371 enum sctp_xmit retval = SCTP_XMIT_OK;
372
373 pr_debug("%s: packet:%p chunk:%p\n", __func__, packet, chunk);
374
375 /* Data chunks are special. Before seeing what else we can
376 * bundle into this packet, check to see if we are allowed to
377 * send this DATA.
378 */
379 if (sctp_chunk_is_data(chunk)) {
380 retval = sctp_packet_can_append_data(packet, chunk);
381 if (retval != SCTP_XMIT_OK)
382 goto finish;
383 }
384
385 /* Try to bundle AUTH chunk */
386 retval = sctp_packet_bundle_auth(packet, chunk);
387 if (retval != SCTP_XMIT_OK)
388 goto finish;
389
390 /* Try to bundle SACK chunk */
391 retval = sctp_packet_bundle_sack(packet, chunk);
392 if (retval != SCTP_XMIT_OK)
393 goto finish;
394
395 retval = __sctp_packet_append_chunk(packet, chunk);
396
397finish:
398 return retval;
399}
400
401static void sctp_packet_release_owner(struct sk_buff *skb)
402{
403 sk_free(skb->sk);
404}
405
406static void sctp_packet_set_owner_w(struct sk_buff *skb, struct sock *sk)
407{
408 skb_orphan(skb);
409 skb->sk = sk;
410 skb->destructor = sctp_packet_release_owner;
411
412 /*
413 * The data chunks have already been accounted for in sctp_sendmsg(),
414 * therefore only reserve a single byte to keep socket around until
415 * the packet has been transmitted.
416 */
417 refcount_inc(&sk->sk_wmem_alloc);
418}
419
420static int sctp_packet_pack(struct sctp_packet *packet,
421 struct sk_buff *head, int gso, gfp_t gfp)
422{
423 struct sctp_transport *tp = packet->transport;
424 struct sctp_auth_chunk *auth = NULL;
425 struct sctp_chunk *chunk, *tmp;
426 int pkt_count = 0, pkt_size;
427 struct sock *sk = head->sk;
428 struct sk_buff *nskb;
429 int auth_len = 0;
430
431 if (gso) {
432 skb_shinfo(head)->gso_type = sk->sk_gso_type;
433 NAPI_GRO_CB(head)->last = head;
434 } else {
435 nskb = head;
436 pkt_size = packet->size;
437 goto merge;
438 }
439
440 do {
441 /* calculate the pkt_size and alloc nskb */
442 pkt_size = packet->overhead;
443 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list,
444 list) {
445 int padded = SCTP_PAD4(chunk->skb->len);
446
447 if (chunk == packet->auth)
448 auth_len = padded;
449 else if (auth_len + padded + packet->overhead >
450 tp->pathmtu)
451 return 0;
452 else if (pkt_size + padded > tp->pathmtu)
453 break;
454 pkt_size += padded;
455 }
456 nskb = alloc_skb(pkt_size + MAX_HEADER, gfp);
457 if (!nskb)
458 return 0;
459 skb_reserve(nskb, packet->overhead + MAX_HEADER);
460
461merge:
462 /* merge chunks into nskb and append nskb into head list */
463 pkt_size -= packet->overhead;
464 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
465 int padding;
466
467 list_del_init(&chunk->list);
468 if (sctp_chunk_is_data(chunk)) {
469 if (!sctp_chunk_retransmitted(chunk) &&
470 !tp->rto_pending) {
471 chunk->rtt_in_progress = 1;
472 tp->rto_pending = 1;
473 }
474 }
475
476 padding = SCTP_PAD4(chunk->skb->len) - chunk->skb->len;
477 if (padding)
478 skb_put_zero(chunk->skb, padding);
479
480 if (chunk == packet->auth)
481 auth = (struct sctp_auth_chunk *)
482 skb_tail_pointer(nskb);
483
484 skb_put_data(nskb, chunk->skb->data, chunk->skb->len);
485
486 pr_debug("*** Chunk:%p[%s] %s 0x%x, length:%d, chunk->skb->len:%d, rtt_in_progress:%d\n",
487 chunk,
488 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)),
489 chunk->has_tsn ? "TSN" : "No TSN",
490 chunk->has_tsn ? ntohl(chunk->subh.data_hdr->tsn) : 0,
491 ntohs(chunk->chunk_hdr->length), chunk->skb->len,
492 chunk->rtt_in_progress);
493
494 pkt_size -= SCTP_PAD4(chunk->skb->len);
495
496 if (!sctp_chunk_is_data(chunk) && chunk != packet->auth)
497 sctp_chunk_free(chunk);
498
499 if (!pkt_size)
500 break;
501 }
502
503 if (auth) {
504 sctp_auth_calculate_hmac(tp->asoc, nskb, auth,
505 packet->auth->shkey, gfp);
506 /* free auth if no more chunks, or add it back */
507 if (list_empty(&packet->chunk_list))
508 sctp_chunk_free(packet->auth);
509 else
510 list_add(&packet->auth->list,
511 &packet->chunk_list);
512 }
513
514 if (gso) {
515 if (skb_gro_receive(&head, nskb)) {
516 kfree_skb(nskb);
517 return 0;
518 }
519 if (WARN_ON_ONCE(skb_shinfo(head)->gso_segs >=
520 sk->sk_gso_max_segs))
521 return 0;
522 }
523
524 pkt_count++;
525 } while (!list_empty(&packet->chunk_list));
526
527 if (gso) {
528 memset(head->cb, 0, max(sizeof(struct inet_skb_parm),
529 sizeof(struct inet6_skb_parm)));
530 skb_shinfo(head)->gso_segs = pkt_count;
531 skb_shinfo(head)->gso_size = GSO_BY_FRAGS;
532 rcu_read_lock();
533 if (skb_dst(head) != tp->dst) {
534 dst_hold(tp->dst);
535 sk_setup_caps(sk, tp->dst);
536 }
537 rcu_read_unlock();
538 goto chksum;
539 }
540
541 if (sctp_checksum_disable)
542 return 1;
543
544 if (!(skb_dst(head)->dev->features & NETIF_F_SCTP_CRC) ||
545 dst_xfrm(skb_dst(head)) || packet->ipfragok) {
546 struct sctphdr *sh =
547 (struct sctphdr *)skb_transport_header(head);
548
549 sh->checksum = sctp_compute_cksum(head, 0);
550 } else {
551chksum:
552 head->ip_summed = CHECKSUM_PARTIAL;
553 head->csum_not_inet = 1;
554 head->csum_start = skb_transport_header(head) - head->head;
555 head->csum_offset = offsetof(struct sctphdr, checksum);
556 }
557
558 return pkt_count;
559}
560
561/* All packets are sent to the network through this function from
562 * sctp_outq_tail().
563 *
564 * The return value is always 0 for now.
565 */
566int sctp_packet_transmit(struct sctp_packet *packet, gfp_t gfp)
567{
568 struct sctp_transport *tp = packet->transport;
569 struct sctp_association *asoc = tp->asoc;
570 struct sctp_chunk *chunk, *tmp;
571 int pkt_count, gso = 0;
572 struct dst_entry *dst;
573 struct sk_buff *head;
574 struct sctphdr *sh;
575 struct sock *sk;
576
577 pr_debug("%s: packet:%p\n", __func__, packet);
578 if (list_empty(&packet->chunk_list))
579 return 0;
580 chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
581 sk = chunk->skb->sk;
582
583 /* check gso */
584 if (packet->size > tp->pathmtu && !packet->ipfragok) {
585 if (!sk_can_gso(sk)) {
586 pr_err_once("Trying to GSO but underlying device doesn't support it.");
587 goto out;
588 }
589 gso = 1;
590 }
591
592 /* alloc head skb */
593 head = alloc_skb((gso ? packet->overhead : packet->size) +
594 MAX_HEADER, gfp);
595 if (!head)
596 goto out;
597 skb_reserve(head, packet->overhead + MAX_HEADER);
598 sctp_packet_set_owner_w(head, sk);
599
600 /* set sctp header */
601 sh = skb_push(head, sizeof(struct sctphdr));
602 skb_reset_transport_header(head);
603 sh->source = htons(packet->source_port);
604 sh->dest = htons(packet->destination_port);
605 sh->vtag = htonl(packet->vtag);
606 sh->checksum = 0;
607
608 /* drop packet if no dst */
609 dst = dst_clone(tp->dst);
610 if (!dst) {
611 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
612 kfree_skb(head);
613 goto out;
614 }
615 skb_dst_set(head, dst);
616
617 /* pack up chunks */
618 pkt_count = sctp_packet_pack(packet, head, gso, gfp);
619 if (!pkt_count) {
620 kfree_skb(head);
621 goto out;
622 }
623 pr_debug("***sctp_transmit_packet*** skb->len:%d\n", head->len);
624
625 /* start autoclose timer */
626 if (packet->has_data && sctp_state(asoc, ESTABLISHED) &&
627 asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE]) {
628 struct timer_list *timer =
629 &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
630 unsigned long timeout =
631 asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
632
633 if (!mod_timer(timer, jiffies + timeout))
634 sctp_association_hold(asoc);
635 }
636
637 /* sctp xmit */
638 tp->af_specific->ecn_capable(sk);
639 if (asoc) {
640 asoc->stats.opackets += pkt_count;
641 if (asoc->peer.last_sent_to != tp)
642 asoc->peer.last_sent_to = tp;
643 }
644 head->ignore_df = packet->ipfragok;
645 if (tp->dst_pending_confirm)
646 skb_set_dst_pending_confirm(head, 1);
647 /* neighbour should be confirmed on successful transmission or
648 * positive error
649 */
650 if (tp->af_specific->sctp_xmit(head, tp) >= 0 &&
651 tp->dst_pending_confirm)
652 tp->dst_pending_confirm = 0;
653
654out:
655 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
656 list_del_init(&chunk->list);
657 if (!sctp_chunk_is_data(chunk))
658 sctp_chunk_free(chunk);
659 }
660 sctp_packet_reset(packet);
661 return 0;
662}
663
664/********************************************************************
665 * 2nd Level Abstractions
666 ********************************************************************/
667
668/* This private function check to see if a chunk can be added */
669static enum sctp_xmit sctp_packet_can_append_data(struct sctp_packet *packet,
670 struct sctp_chunk *chunk)
671{
672 size_t datasize, rwnd, inflight, flight_size;
673 struct sctp_transport *transport = packet->transport;
674 struct sctp_association *asoc = transport->asoc;
675 struct sctp_outq *q = &asoc->outqueue;
676
677 /* RFC 2960 6.1 Transmission of DATA Chunks
678 *
679 * A) At any given time, the data sender MUST NOT transmit new data to
680 * any destination transport address if its peer's rwnd indicates
681 * that the peer has no buffer space (i.e. rwnd is 0, see Section
682 * 6.2.1). However, regardless of the value of rwnd (including if it
683 * is 0), the data sender can always have one DATA chunk in flight to
684 * the receiver if allowed by cwnd (see rule B below). This rule
685 * allows the sender to probe for a change in rwnd that the sender
686 * missed due to the SACK having been lost in transit from the data
687 * receiver to the data sender.
688 */
689
690 rwnd = asoc->peer.rwnd;
691 inflight = q->outstanding_bytes;
692 flight_size = transport->flight_size;
693
694 datasize = sctp_data_size(chunk);
695
696 if (datasize > rwnd && inflight > 0)
697 /* We have (at least) one data chunk in flight,
698 * so we can't fall back to rule 6.1 B).
699 */
700 return SCTP_XMIT_RWND_FULL;
701
702 /* RFC 2960 6.1 Transmission of DATA Chunks
703 *
704 * B) At any given time, the sender MUST NOT transmit new data
705 * to a given transport address if it has cwnd or more bytes
706 * of data outstanding to that transport address.
707 */
708 /* RFC 7.2.4 & the Implementers Guide 2.8.
709 *
710 * 3) ...
711 * When a Fast Retransmit is being performed the sender SHOULD
712 * ignore the value of cwnd and SHOULD NOT delay retransmission.
713 */
714 if (chunk->fast_retransmit != SCTP_NEED_FRTX &&
715 flight_size >= transport->cwnd)
716 return SCTP_XMIT_RWND_FULL;
717
718 /* Nagle's algorithm to solve small-packet problem:
719 * Inhibit the sending of new chunks when new outgoing data arrives
720 * if any previously transmitted data on the connection remains
721 * unacknowledged.
722 */
723
724 if ((sctp_sk(asoc->base.sk)->nodelay || inflight == 0) &&
725 !asoc->force_delay)
726 /* Nothing unacked */
727 return SCTP_XMIT_OK;
728
729 if (!sctp_packet_empty(packet))
730 /* Append to packet */
731 return SCTP_XMIT_OK;
732
733 if (!sctp_state(asoc, ESTABLISHED))
734 return SCTP_XMIT_OK;
735
736 /* Check whether this chunk and all the rest of pending data will fit
737 * or delay in hopes of bundling a full sized packet.
738 */
739 if (chunk->skb->len + q->out_qlen > transport->pathmtu -
740 packet->overhead - sctp_datachk_len(&chunk->asoc->stream) - 4)
741 /* Enough data queued to fill a packet */
742 return SCTP_XMIT_OK;
743
744 /* Don't delay large message writes that may have been fragmented */
745 if (!chunk->msg->can_delay)
746 return SCTP_XMIT_OK;
747
748 /* Defer until all data acked or packet full */
749 return SCTP_XMIT_DELAY;
750}
751
752/* This private function does management things when adding DATA chunk */
753static void sctp_packet_append_data(struct sctp_packet *packet,
754 struct sctp_chunk *chunk)
755{
756 struct sctp_transport *transport = packet->transport;
757 size_t datasize = sctp_data_size(chunk);
758 struct sctp_association *asoc = transport->asoc;
759 u32 rwnd = asoc->peer.rwnd;
760
761 /* Keep track of how many bytes are in flight over this transport. */
762 transport->flight_size += datasize;
763
764 /* Keep track of how many bytes are in flight to the receiver. */
765 asoc->outqueue.outstanding_bytes += datasize;
766
767 /* Update our view of the receiver's rwnd. */
768 if (datasize < rwnd)
769 rwnd -= datasize;
770 else
771 rwnd = 0;
772
773 asoc->peer.rwnd = rwnd;
774 sctp_chunk_assign_tsn(chunk);
775 asoc->stream.si->assign_number(chunk);
776}
777
778static enum sctp_xmit sctp_packet_will_fit(struct sctp_packet *packet,
779 struct sctp_chunk *chunk,
780 u16 chunk_len)
781{
782 enum sctp_xmit retval = SCTP_XMIT_OK;
783 size_t psize, pmtu, maxsize;
784
785 /* Don't bundle in this packet if this chunk's auth key doesn't
786 * match other chunks already enqueued on this packet. Also,
787 * don't bundle the chunk with auth key if other chunks in this
788 * packet don't have auth key.
789 */
790 if ((packet->auth && chunk->shkey != packet->auth->shkey) ||
791 (!packet->auth && chunk->shkey &&
792 chunk->chunk_hdr->type != SCTP_CID_AUTH))
793 return SCTP_XMIT_PMTU_FULL;
794
795 psize = packet->size;
796 if (packet->transport->asoc)
797 pmtu = packet->transport->asoc->pathmtu;
798 else
799 pmtu = packet->transport->pathmtu;
800
801 /* Decide if we need to fragment or resubmit later. */
802 if (psize + chunk_len > pmtu) {
803 /* It's OK to fragment at IP level if any one of the following
804 * is true:
805 * 1. The packet is empty (meaning this chunk is greater
806 * the MTU)
807 * 2. The packet doesn't have any data in it yet and data
808 * requires authentication.
809 */
810 if (sctp_packet_empty(packet) ||
811 (!packet->has_data && chunk->auth)) {
812 /* We no longer do re-fragmentation.
813 * Just fragment at the IP layer, if we
814 * actually hit this condition
815 */
816 packet->ipfragok = 1;
817 goto out;
818 }
819
820 /* Similarly, if this chunk was built before a PMTU
821 * reduction, we have to fragment it at IP level now. So
822 * if the packet already contains something, we need to
823 * flush.
824 */
825 maxsize = pmtu - packet->overhead;
826 if (packet->auth)
827 maxsize -= SCTP_PAD4(packet->auth->skb->len);
828 if (chunk_len > maxsize)
829 retval = SCTP_XMIT_PMTU_FULL;
830
831 /* It is also okay to fragment if the chunk we are
832 * adding is a control chunk, but only if current packet
833 * is not a GSO one otherwise it causes fragmentation of
834 * a large frame. So in this case we allow the
835 * fragmentation by forcing it to be in a new packet.
836 */
837 if (!sctp_chunk_is_data(chunk) && packet->has_data)
838 retval = SCTP_XMIT_PMTU_FULL;
839
840 if (psize + chunk_len > packet->max_size)
841 /* Hit GSO/PMTU limit, gotta flush */
842 retval = SCTP_XMIT_PMTU_FULL;
843
844 if (!packet->transport->burst_limited &&
845 psize + chunk_len > (packet->transport->cwnd >> 1))
846 /* Do not allow a single GSO packet to use more
847 * than half of cwnd.
848 */
849 retval = SCTP_XMIT_PMTU_FULL;
850
851 if (packet->transport->burst_limited &&
852 psize + chunk_len > (packet->transport->burst_limited >> 1))
853 /* Do not allow a single GSO packet to use more
854 * than half of original cwnd.
855 */
856 retval = SCTP_XMIT_PMTU_FULL;
857 /* Otherwise it will fit in the GSO packet */
858 }
859
860out:
861 return retval;
862}
1/* SCTP kernel implementation
2 * (C) Copyright IBM Corp. 2001, 2004
3 * Copyright (c) 1999-2000 Cisco, Inc.
4 * Copyright (c) 1999-2001 Motorola, Inc.
5 *
6 * This file is part of the SCTP kernel implementation
7 *
8 * These functions handle output processing.
9 *
10 * This SCTP 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 * This SCTP 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#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
45
46#include <linux/types.h>
47#include <linux/kernel.h>
48#include <linux/wait.h>
49#include <linux/time.h>
50#include <linux/ip.h>
51#include <linux/ipv6.h>
52#include <linux/init.h>
53#include <linux/slab.h>
54#include <net/inet_ecn.h>
55#include <net/ip.h>
56#include <net/icmp.h>
57#include <net/net_namespace.h>
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>
64#include <net/sctp/checksum.h>
65
66/* Forward declarations for private helpers. */
67static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
68 struct sctp_chunk *chunk);
69static void sctp_packet_append_data(struct sctp_packet *packet,
70 struct sctp_chunk *chunk);
71static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
72 struct sctp_chunk *chunk,
73 u16 chunk_len);
74
75static void sctp_packet_reset(struct sctp_packet *packet)
76{
77 packet->size = packet->overhead;
78 packet->has_cookie_echo = 0;
79 packet->has_sack = 0;
80 packet->has_data = 0;
81 packet->has_auth = 0;
82 packet->ipfragok = 0;
83 packet->auth = NULL;
84}
85
86/* Config a packet.
87 * This appears to be a followup set of initializations.
88 */
89struct sctp_packet *sctp_packet_config(struct sctp_packet *packet,
90 __u32 vtag, int ecn_capable)
91{
92 struct sctp_chunk *chunk = NULL;
93
94 SCTP_DEBUG_PRINTK("%s: packet:%p vtag:0x%x\n", __func__,
95 packet, vtag);
96
97 packet->vtag = vtag;
98
99 if (ecn_capable && sctp_packet_empty(packet)) {
100 chunk = sctp_get_ecne_prepend(packet->transport->asoc);
101
102 /* If there a is a prepend chunk stick it on the list before
103 * any other chunks get appended.
104 */
105 if (chunk)
106 sctp_packet_append_chunk(packet, chunk);
107 }
108
109 return packet;
110}
111
112/* Initialize the packet structure. */
113struct sctp_packet *sctp_packet_init(struct sctp_packet *packet,
114 struct sctp_transport *transport,
115 __u16 sport, __u16 dport)
116{
117 struct sctp_association *asoc = transport->asoc;
118 size_t overhead;
119
120 SCTP_DEBUG_PRINTK("%s: packet:%p transport:%p\n", __func__,
121 packet, transport);
122
123 packet->transport = transport;
124 packet->source_port = sport;
125 packet->destination_port = dport;
126 INIT_LIST_HEAD(&packet->chunk_list);
127 if (asoc) {
128 struct sctp_sock *sp = sctp_sk(asoc->base.sk);
129 overhead = sp->pf->af->net_header_len;
130 } else {
131 overhead = sizeof(struct ipv6hdr);
132 }
133 overhead += sizeof(struct sctphdr);
134 packet->overhead = overhead;
135 sctp_packet_reset(packet);
136 packet->vtag = 0;
137 packet->malloced = 0;
138 return packet;
139}
140
141/* Free a packet. */
142void sctp_packet_free(struct sctp_packet *packet)
143{
144 struct sctp_chunk *chunk, *tmp;
145
146 SCTP_DEBUG_PRINTK("%s: packet:%p\n", __func__, packet);
147
148 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
149 list_del_init(&chunk->list);
150 sctp_chunk_free(chunk);
151 }
152
153 if (packet->malloced)
154 kfree(packet);
155}
156
157/* This routine tries to append the chunk to the offered packet. If adding
158 * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk
159 * is not present in the packet, it transmits the input packet.
160 * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long
161 * as it can fit in the packet, but any more data that does not fit in this
162 * packet can be sent only after receiving the COOKIE_ACK.
163 */
164sctp_xmit_t sctp_packet_transmit_chunk(struct sctp_packet *packet,
165 struct sctp_chunk *chunk,
166 int one_packet)
167{
168 sctp_xmit_t retval;
169 int error = 0;
170
171 SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __func__,
172 packet, chunk);
173
174 switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) {
175 case SCTP_XMIT_PMTU_FULL:
176 if (!packet->has_cookie_echo) {
177 error = sctp_packet_transmit(packet);
178 if (error < 0)
179 chunk->skb->sk->sk_err = -error;
180
181 /* If we have an empty packet, then we can NOT ever
182 * return PMTU_FULL.
183 */
184 if (!one_packet)
185 retval = sctp_packet_append_chunk(packet,
186 chunk);
187 }
188 break;
189
190 case SCTP_XMIT_RWND_FULL:
191 case SCTP_XMIT_OK:
192 case SCTP_XMIT_NAGLE_DELAY:
193 break;
194 }
195
196 return retval;
197}
198
199/* Try to bundle an auth chunk into the packet. */
200static sctp_xmit_t sctp_packet_bundle_auth(struct sctp_packet *pkt,
201 struct sctp_chunk *chunk)
202{
203 struct sctp_association *asoc = pkt->transport->asoc;
204 struct sctp_chunk *auth;
205 sctp_xmit_t retval = SCTP_XMIT_OK;
206
207 /* if we don't have an association, we can't do authentication */
208 if (!asoc)
209 return retval;
210
211 /* See if this is an auth chunk we are bundling or if
212 * auth is already bundled.
213 */
214 if (chunk->chunk_hdr->type == SCTP_CID_AUTH || pkt->has_auth)
215 return retval;
216
217 /* if the peer did not request this chunk to be authenticated,
218 * don't do it
219 */
220 if (!chunk->auth)
221 return retval;
222
223 auth = sctp_make_auth(asoc);
224 if (!auth)
225 return retval;
226
227 retval = sctp_packet_append_chunk(pkt, auth);
228
229 return retval;
230}
231
232/* Try to bundle a SACK with the packet. */
233static sctp_xmit_t sctp_packet_bundle_sack(struct sctp_packet *pkt,
234 struct sctp_chunk *chunk)
235{
236 sctp_xmit_t retval = SCTP_XMIT_OK;
237
238 /* If sending DATA and haven't aleady bundled a SACK, try to
239 * bundle one in to the packet.
240 */
241 if (sctp_chunk_is_data(chunk) && !pkt->has_sack &&
242 !pkt->has_cookie_echo) {
243 struct sctp_association *asoc;
244 struct timer_list *timer;
245 asoc = pkt->transport->asoc;
246 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
247
248 /* If the SACK timer is running, we have a pending SACK */
249 if (timer_pending(timer)) {
250 struct sctp_chunk *sack;
251 asoc->a_rwnd = asoc->rwnd;
252 sack = sctp_make_sack(asoc);
253 if (sack) {
254 retval = sctp_packet_append_chunk(pkt, sack);
255 asoc->peer.sack_needed = 0;
256 if (del_timer(timer))
257 sctp_association_put(asoc);
258 }
259 }
260 }
261 return retval;
262}
263
264/* Append a chunk to the offered packet reporting back any inability to do
265 * so.
266 */
267sctp_xmit_t sctp_packet_append_chunk(struct sctp_packet *packet,
268 struct sctp_chunk *chunk)
269{
270 sctp_xmit_t retval = SCTP_XMIT_OK;
271 __u16 chunk_len = WORD_ROUND(ntohs(chunk->chunk_hdr->length));
272
273 SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __func__, packet,
274 chunk);
275
276 /* Data chunks are special. Before seeing what else we can
277 * bundle into this packet, check to see if we are allowed to
278 * send this DATA.
279 */
280 if (sctp_chunk_is_data(chunk)) {
281 retval = sctp_packet_can_append_data(packet, chunk);
282 if (retval != SCTP_XMIT_OK)
283 goto finish;
284 }
285
286 /* Try to bundle AUTH chunk */
287 retval = sctp_packet_bundle_auth(packet, chunk);
288 if (retval != SCTP_XMIT_OK)
289 goto finish;
290
291 /* Try to bundle SACK chunk */
292 retval = sctp_packet_bundle_sack(packet, chunk);
293 if (retval != SCTP_XMIT_OK)
294 goto finish;
295
296 /* Check to see if this chunk will fit into the packet */
297 retval = sctp_packet_will_fit(packet, chunk, chunk_len);
298 if (retval != SCTP_XMIT_OK)
299 goto finish;
300
301 /* We believe that this chunk is OK to add to the packet */
302 switch (chunk->chunk_hdr->type) {
303 case SCTP_CID_DATA:
304 /* Account for the data being in the packet */
305 sctp_packet_append_data(packet, chunk);
306 /* Disallow SACK bundling after DATA. */
307 packet->has_sack = 1;
308 /* Disallow AUTH bundling after DATA */
309 packet->has_auth = 1;
310 /* Let it be knows that packet has DATA in it */
311 packet->has_data = 1;
312 /* timestamp the chunk for rtx purposes */
313 chunk->sent_at = jiffies;
314 break;
315 case SCTP_CID_COOKIE_ECHO:
316 packet->has_cookie_echo = 1;
317 break;
318
319 case SCTP_CID_SACK:
320 packet->has_sack = 1;
321 break;
322
323 case SCTP_CID_AUTH:
324 packet->has_auth = 1;
325 packet->auth = chunk;
326 break;
327 }
328
329 /* It is OK to send this chunk. */
330 list_add_tail(&chunk->list, &packet->chunk_list);
331 packet->size += chunk_len;
332 chunk->transport = packet->transport;
333finish:
334 return retval;
335}
336
337/* All packets are sent to the network through this function from
338 * sctp_outq_tail().
339 *
340 * The return value is a normal kernel error return value.
341 */
342int sctp_packet_transmit(struct sctp_packet *packet)
343{
344 struct sctp_transport *tp = packet->transport;
345 struct sctp_association *asoc = tp->asoc;
346 struct sctphdr *sh;
347 struct sk_buff *nskb;
348 struct sctp_chunk *chunk, *tmp;
349 struct sock *sk;
350 int err = 0;
351 int padding; /* How much padding do we need? */
352 __u8 has_data = 0;
353 struct dst_entry *dst = tp->dst;
354 unsigned char *auth = NULL; /* pointer to auth in skb data */
355 __u32 cksum_buf_len = sizeof(struct sctphdr);
356
357 SCTP_DEBUG_PRINTK("%s: packet:%p\n", __func__, packet);
358
359 /* Do NOT generate a chunkless packet. */
360 if (list_empty(&packet->chunk_list))
361 return err;
362
363 /* Set up convenience variables... */
364 chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
365 sk = chunk->skb->sk;
366
367 /* Allocate the new skb. */
368 nskb = alloc_skb(packet->size + LL_MAX_HEADER, GFP_ATOMIC);
369 if (!nskb)
370 goto nomem;
371
372 /* Make sure the outbound skb has enough header room reserved. */
373 skb_reserve(nskb, packet->overhead + LL_MAX_HEADER);
374
375 /* Set the owning socket so that we know where to get the
376 * destination IP address.
377 */
378 skb_set_owner_w(nskb, sk);
379
380 /* The 'obsolete' field of dst is set to 2 when a dst is freed. */
381 if (!dst || (dst->obsolete > 1)) {
382 dst_release(dst);
383 sctp_transport_route(tp, NULL, sctp_sk(sk));
384 if (asoc && (asoc->param_flags & SPP_PMTUD_ENABLE)) {
385 sctp_assoc_sync_pmtu(asoc);
386 }
387 }
388 dst = dst_clone(tp->dst);
389 skb_dst_set(nskb, dst);
390 if (!dst)
391 goto no_route;
392
393 /* Build the SCTP header. */
394 sh = (struct sctphdr *)skb_push(nskb, sizeof(struct sctphdr));
395 skb_reset_transport_header(nskb);
396 sh->source = htons(packet->source_port);
397 sh->dest = htons(packet->destination_port);
398
399 /* From 6.8 Adler-32 Checksum Calculation:
400 * After the packet is constructed (containing the SCTP common
401 * header and one or more control or DATA chunks), the
402 * transmitter shall:
403 *
404 * 1) Fill in the proper Verification Tag in the SCTP common
405 * header and initialize the checksum field to 0's.
406 */
407 sh->vtag = htonl(packet->vtag);
408 sh->checksum = 0;
409
410 /**
411 * 6.10 Bundling
412 *
413 * An endpoint bundles chunks by simply including multiple
414 * chunks in one outbound SCTP packet. ...
415 */
416
417 /**
418 * 3.2 Chunk Field Descriptions
419 *
420 * The total length of a chunk (including Type, Length and
421 * Value fields) MUST be a multiple of 4 bytes. If the length
422 * of the chunk is not a multiple of 4 bytes, the sender MUST
423 * pad the chunk with all zero bytes and this padding is not
424 * included in the chunk length field. The sender should
425 * never pad with more than 3 bytes.
426 *
427 * [This whole comment explains WORD_ROUND() below.]
428 */
429 SCTP_DEBUG_PRINTK("***sctp_transmit_packet***\n");
430 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
431 list_del_init(&chunk->list);
432 if (sctp_chunk_is_data(chunk)) {
433 /* 6.3.1 C4) When data is in flight and when allowed
434 * by rule C5, a new RTT measurement MUST be made each
435 * round trip. Furthermore, new RTT measurements
436 * SHOULD be made no more than once per round-trip
437 * for a given destination transport address.
438 */
439
440 if (!tp->rto_pending) {
441 chunk->rtt_in_progress = 1;
442 tp->rto_pending = 1;
443 }
444 has_data = 1;
445 }
446
447 padding = WORD_ROUND(chunk->skb->len) - chunk->skb->len;
448 if (padding)
449 memset(skb_put(chunk->skb, padding), 0, padding);
450
451 /* if this is the auth chunk that we are adding,
452 * store pointer where it will be added and put
453 * the auth into the packet.
454 */
455 if (chunk == packet->auth)
456 auth = skb_tail_pointer(nskb);
457
458 cksum_buf_len += chunk->skb->len;
459 memcpy(skb_put(nskb, chunk->skb->len),
460 chunk->skb->data, chunk->skb->len);
461
462 SCTP_DEBUG_PRINTK("%s %p[%s] %s 0x%x, %s %d, %s %d, %s %d\n",
463 "*** Chunk", chunk,
464 sctp_cname(SCTP_ST_CHUNK(
465 chunk->chunk_hdr->type)),
466 chunk->has_tsn ? "TSN" : "No TSN",
467 chunk->has_tsn ?
468 ntohl(chunk->subh.data_hdr->tsn) : 0,
469 "length", ntohs(chunk->chunk_hdr->length),
470 "chunk->skb->len", chunk->skb->len,
471 "rtt_in_progress", chunk->rtt_in_progress);
472
473 /*
474 * If this is a control chunk, this is our last
475 * reference. Free data chunks after they've been
476 * acknowledged or have failed.
477 */
478 if (!sctp_chunk_is_data(chunk))
479 sctp_chunk_free(chunk);
480 }
481
482 /* SCTP-AUTH, Section 6.2
483 * The sender MUST calculate the MAC as described in RFC2104 [2]
484 * using the hash function H as described by the MAC Identifier and
485 * the shared association key K based on the endpoint pair shared key
486 * described by the shared key identifier. The 'data' used for the
487 * computation of the AUTH-chunk is given by the AUTH chunk with its
488 * HMAC field set to zero (as shown in Figure 6) followed by all
489 * chunks that are placed after the AUTH chunk in the SCTP packet.
490 */
491 if (auth)
492 sctp_auth_calculate_hmac(asoc, nskb,
493 (struct sctp_auth_chunk *)auth,
494 GFP_ATOMIC);
495
496 /* 2) Calculate the Adler-32 checksum of the whole packet,
497 * including the SCTP common header and all the
498 * chunks.
499 *
500 * Note: Adler-32 is no longer applicable, as has been replaced
501 * by CRC32-C as described in <draft-ietf-tsvwg-sctpcsum-02.txt>.
502 */
503 if (!sctp_checksum_disable) {
504 if (!(dst->dev->features & NETIF_F_SCTP_CSUM)) {
505 __u32 crc32 = sctp_start_cksum((__u8 *)sh, cksum_buf_len);
506
507 /* 3) Put the resultant value into the checksum field in the
508 * common header, and leave the rest of the bits unchanged.
509 */
510 sh->checksum = sctp_end_cksum(crc32);
511 } else {
512 /* no need to seed pseudo checksum for SCTP */
513 nskb->ip_summed = CHECKSUM_PARTIAL;
514 nskb->csum_start = (skb_transport_header(nskb) -
515 nskb->head);
516 nskb->csum_offset = offsetof(struct sctphdr, checksum);
517 }
518 }
519
520 /* IP layer ECN support
521 * From RFC 2481
522 * "The ECN-Capable Transport (ECT) bit would be set by the
523 * data sender to indicate that the end-points of the
524 * transport protocol are ECN-capable."
525 *
526 * Now setting the ECT bit all the time, as it should not cause
527 * any problems protocol-wise even if our peer ignores it.
528 *
529 * Note: The works for IPv6 layer checks this bit too later
530 * in transmission. See IP6_ECN_flow_xmit().
531 */
532 (*tp->af_specific->ecn_capable)(nskb->sk);
533
534 /* Set up the IP options. */
535 /* BUG: not implemented
536 * For v4 this all lives somewhere in sk->sk_opt...
537 */
538
539 /* Dump that on IP! */
540 if (asoc && asoc->peer.last_sent_to != tp) {
541 /* Considering the multiple CPU scenario, this is a
542 * "correcter" place for last_sent_to. --xguo
543 */
544 asoc->peer.last_sent_to = tp;
545 }
546
547 if (has_data) {
548 struct timer_list *timer;
549 unsigned long timeout;
550
551 /* Restart the AUTOCLOSE timer when sending data. */
552 if (sctp_state(asoc, ESTABLISHED) && asoc->autoclose) {
553 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
554 timeout = asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
555
556 if (!mod_timer(timer, jiffies + timeout))
557 sctp_association_hold(asoc);
558 }
559 }
560
561 SCTP_DEBUG_PRINTK("***sctp_transmit_packet*** skb len %d\n",
562 nskb->len);
563
564 nskb->local_df = packet->ipfragok;
565 (*tp->af_specific->sctp_xmit)(nskb, tp);
566
567out:
568 sctp_packet_reset(packet);
569 return err;
570no_route:
571 kfree_skb(nskb);
572 IP_INC_STATS_BH(&init_net, IPSTATS_MIB_OUTNOROUTES);
573
574 /* FIXME: Returning the 'err' will effect all the associations
575 * associated with a socket, although only one of the paths of the
576 * association is unreachable.
577 * The real failure of a transport or association can be passed on
578 * to the user via notifications. So setting this error may not be
579 * required.
580 */
581 /* err = -EHOSTUNREACH; */
582err:
583 /* Control chunks are unreliable so just drop them. DATA chunks
584 * will get resent or dropped later.
585 */
586
587 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
588 list_del_init(&chunk->list);
589 if (!sctp_chunk_is_data(chunk))
590 sctp_chunk_free(chunk);
591 }
592 goto out;
593nomem:
594 err = -ENOMEM;
595 goto err;
596}
597
598/********************************************************************
599 * 2nd Level Abstractions
600 ********************************************************************/
601
602/* This private function check to see if a chunk can be added */
603static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
604 struct sctp_chunk *chunk)
605{
606 sctp_xmit_t retval = SCTP_XMIT_OK;
607 size_t datasize, rwnd, inflight, flight_size;
608 struct sctp_transport *transport = packet->transport;
609 struct sctp_association *asoc = transport->asoc;
610 struct sctp_outq *q = &asoc->outqueue;
611
612 /* RFC 2960 6.1 Transmission of DATA Chunks
613 *
614 * A) At any given time, the data sender MUST NOT transmit new data to
615 * any destination transport address if its peer's rwnd indicates
616 * that the peer has no buffer space (i.e. rwnd is 0, see Section
617 * 6.2.1). However, regardless of the value of rwnd (including if it
618 * is 0), the data sender can always have one DATA chunk in flight to
619 * the receiver if allowed by cwnd (see rule B below). This rule
620 * allows the sender to probe for a change in rwnd that the sender
621 * missed due to the SACK having been lost in transit from the data
622 * receiver to the data sender.
623 */
624
625 rwnd = asoc->peer.rwnd;
626 inflight = q->outstanding_bytes;
627 flight_size = transport->flight_size;
628
629 datasize = sctp_data_size(chunk);
630
631 if (datasize > rwnd) {
632 if (inflight > 0) {
633 /* We have (at least) one data chunk in flight,
634 * so we can't fall back to rule 6.1 B).
635 */
636 retval = SCTP_XMIT_RWND_FULL;
637 goto finish;
638 }
639 }
640
641 /* RFC 2960 6.1 Transmission of DATA Chunks
642 *
643 * B) At any given time, the sender MUST NOT transmit new data
644 * to a given transport address if it has cwnd or more bytes
645 * of data outstanding to that transport address.
646 */
647 /* RFC 7.2.4 & the Implementers Guide 2.8.
648 *
649 * 3) ...
650 * When a Fast Retransmit is being performed the sender SHOULD
651 * ignore the value of cwnd and SHOULD NOT delay retransmission.
652 */
653 if (chunk->fast_retransmit != SCTP_NEED_FRTX)
654 if (flight_size >= transport->cwnd) {
655 retval = SCTP_XMIT_RWND_FULL;
656 goto finish;
657 }
658
659 /* Nagle's algorithm to solve small-packet problem:
660 * Inhibit the sending of new chunks when new outgoing data arrives
661 * if any previously transmitted data on the connection remains
662 * unacknowledged.
663 */
664 if (!sctp_sk(asoc->base.sk)->nodelay && sctp_packet_empty(packet) &&
665 inflight && sctp_state(asoc, ESTABLISHED)) {
666 unsigned max = transport->pathmtu - packet->overhead;
667 unsigned len = chunk->skb->len + q->out_qlen;
668
669 /* Check whether this chunk and all the rest of pending
670 * data will fit or delay in hopes of bundling a full
671 * sized packet.
672 * Don't delay large message writes that may have been
673 * fragmeneted into small peices.
674 */
675 if ((len < max) && chunk->msg->can_delay) {
676 retval = SCTP_XMIT_NAGLE_DELAY;
677 goto finish;
678 }
679 }
680
681finish:
682 return retval;
683}
684
685/* This private function does management things when adding DATA chunk */
686static void sctp_packet_append_data(struct sctp_packet *packet,
687 struct sctp_chunk *chunk)
688{
689 struct sctp_transport *transport = packet->transport;
690 size_t datasize = sctp_data_size(chunk);
691 struct sctp_association *asoc = transport->asoc;
692 u32 rwnd = asoc->peer.rwnd;
693
694 /* Keep track of how many bytes are in flight over this transport. */
695 transport->flight_size += datasize;
696
697 /* Keep track of how many bytes are in flight to the receiver. */
698 asoc->outqueue.outstanding_bytes += datasize;
699
700 /* Update our view of the receiver's rwnd. Include sk_buff overhead
701 * while updating peer.rwnd so that it reduces the chances of a
702 * receiver running out of receive buffer space even when receive
703 * window is still open. This can happen when a sender is sending
704 * sending small messages.
705 */
706 datasize += sizeof(struct sk_buff);
707 if (datasize < rwnd)
708 rwnd -= datasize;
709 else
710 rwnd = 0;
711
712 asoc->peer.rwnd = rwnd;
713 /* Has been accepted for transmission. */
714 if (!asoc->peer.prsctp_capable)
715 chunk->msg->can_abandon = 0;
716 sctp_chunk_assign_tsn(chunk);
717 sctp_chunk_assign_ssn(chunk);
718}
719
720static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
721 struct sctp_chunk *chunk,
722 u16 chunk_len)
723{
724 size_t psize;
725 size_t pmtu;
726 int too_big;
727 sctp_xmit_t retval = SCTP_XMIT_OK;
728
729 psize = packet->size;
730 pmtu = ((packet->transport->asoc) ?
731 (packet->transport->asoc->pathmtu) :
732 (packet->transport->pathmtu));
733
734 too_big = (psize + chunk_len > pmtu);
735
736 /* Decide if we need to fragment or resubmit later. */
737 if (too_big) {
738 /* It's OK to fragmet at IP level if any one of the following
739 * is true:
740 * 1. The packet is empty (meaning this chunk is greater
741 * the MTU)
742 * 2. The chunk we are adding is a control chunk
743 * 3. The packet doesn't have any data in it yet and data
744 * requires authentication.
745 */
746 if (sctp_packet_empty(packet) || !sctp_chunk_is_data(chunk) ||
747 (!packet->has_data && chunk->auth)) {
748 /* We no longer do re-fragmentation.
749 * Just fragment at the IP layer, if we
750 * actually hit this condition
751 */
752 packet->ipfragok = 1;
753 } else {
754 retval = SCTP_XMIT_PMTU_FULL;
755 }
756 }
757
758 return retval;
759}