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.
6 * This file is part of the SCTP kernel implementation
8 * These functions handle output processing.
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)
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.
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.
27 * Please send any bug reports or fixes you make to the
29 * lksctp developers <lksctp-developers@lists.sourceforge.net>
31 * Or submit a bug report through the following website:
32 * http://www.sf.net/projects/lksctp
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>
40 * Any bugs reported given to us we will try to fix... any fixes shared will
41 * be incorporated into the next SCTP release.
44 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
46 #include <linux/types.h>
47 #include <linux/kernel.h>
48 #include <linux/wait.h>
49 #include <linux/time.h>
51 #include <linux/ipv6.h>
52 #include <linux/init.h>
53 #include <linux/slab.h>
54 #include <net/inet_ecn.h>
57 #include <net/net_namespace.h>
59 #include <linux/socket.h> /* for sa_family_t */
62 #include <net/sctp/sctp.h>
63 #include <net/sctp/sm.h>
64 #include <net/sctp/checksum.h>
66 /* Forward declarations for private helpers. */
67 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
68 struct sctp_chunk *chunk);
69 static void sctp_packet_append_data(struct sctp_packet *packet,
70 struct sctp_chunk *chunk);
71 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
72 struct sctp_chunk *chunk,
75 static void sctp_packet_reset(struct sctp_packet *packet)
77 packet->size = packet->overhead;
78 packet->has_cookie_echo = 0;
87 * This appears to be a followup set of initializations.
89 struct sctp_packet *sctp_packet_config(struct sctp_packet *packet,
90 __u32 vtag, int ecn_capable)
92 struct sctp_chunk *chunk = NULL;
94 SCTP_DEBUG_PRINTK("%s: packet:%p vtag:0x%x\n", __func__,
99 if (ecn_capable && sctp_packet_empty(packet)) {
100 chunk = sctp_get_ecne_prepend(packet->transport->asoc);
102 /* If there a is a prepend chunk stick it on the list before
103 * any other chunks get appended.
106 sctp_packet_append_chunk(packet, chunk);
112 /* Initialize the packet structure. */
113 struct sctp_packet *sctp_packet_init(struct sctp_packet *packet,
114 struct sctp_transport *transport,
115 __u16 sport, __u16 dport)
117 struct sctp_association *asoc = transport->asoc;
120 SCTP_DEBUG_PRINTK("%s: packet:%p transport:%p\n", __func__,
123 packet->transport = transport;
124 packet->source_port = sport;
125 packet->destination_port = dport;
126 INIT_LIST_HEAD(&packet->chunk_list);
128 struct sctp_sock *sp = sctp_sk(asoc->base.sk);
129 overhead = sp->pf->af->net_header_len;
131 overhead = sizeof(struct ipv6hdr);
133 overhead += sizeof(struct sctphdr);
134 packet->overhead = overhead;
135 sctp_packet_reset(packet);
137 packet->malloced = 0;
142 void sctp_packet_free(struct sctp_packet *packet)
144 struct sctp_chunk *chunk, *tmp;
146 SCTP_DEBUG_PRINTK("%s: packet:%p\n", __func__, packet);
148 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
149 list_del_init(&chunk->list);
150 sctp_chunk_free(chunk);
153 if (packet->malloced)
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.
164 sctp_xmit_t sctp_packet_transmit_chunk(struct sctp_packet *packet,
165 struct sctp_chunk *chunk,
171 SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __func__,
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);
179 chunk->skb->sk->sk_err = -error;
181 /* If we have an empty packet, then we can NOT ever
185 retval = sctp_packet_append_chunk(packet,
190 case SCTP_XMIT_RWND_FULL:
192 case SCTP_XMIT_NAGLE_DELAY:
199 /* Try to bundle an auth chunk into the packet. */
200 static sctp_xmit_t sctp_packet_bundle_auth(struct sctp_packet *pkt,
201 struct sctp_chunk *chunk)
203 struct sctp_association *asoc = pkt->transport->asoc;
204 struct sctp_chunk *auth;
205 sctp_xmit_t retval = SCTP_XMIT_OK;
207 /* if we don't have an association, we can't do authentication */
211 /* See if this is an auth chunk we are bundling or if
212 * auth is already bundled.
214 if (chunk->chunk_hdr->type == SCTP_CID_AUTH || pkt->has_auth)
217 /* if the peer did not request this chunk to be authenticated,
223 auth = sctp_make_auth(asoc);
227 retval = sctp_packet_append_chunk(pkt, auth);
232 /* Try to bundle a SACK with the packet. */
233 static sctp_xmit_t sctp_packet_bundle_sack(struct sctp_packet *pkt,
234 struct sctp_chunk *chunk)
236 sctp_xmit_t retval = SCTP_XMIT_OK;
238 /* If sending DATA and haven't aleady bundled a SACK, try to
239 * bundle one in to the packet.
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];
248 /* If the SACK timer is running, we have a pending SACK */
249 if (timer_pending(timer)) {
250 struct sctp_chunk *sack;
252 if (pkt->transport->sack_generation !=
253 pkt->transport->asoc->peer.sack_generation)
256 asoc->a_rwnd = asoc->rwnd;
257 sack = sctp_make_sack(asoc);
259 retval = sctp_packet_append_chunk(pkt, sack);
260 asoc->peer.sack_needed = 0;
261 if (del_timer(timer))
262 sctp_association_put(asoc);
269 /* Append a chunk to the offered packet reporting back any inability to do
272 sctp_xmit_t sctp_packet_append_chunk(struct sctp_packet *packet,
273 struct sctp_chunk *chunk)
275 sctp_xmit_t retval = SCTP_XMIT_OK;
276 __u16 chunk_len = WORD_ROUND(ntohs(chunk->chunk_hdr->length));
278 SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __func__, packet,
281 /* Data chunks are special. Before seeing what else we can
282 * bundle into this packet, check to see if we are allowed to
285 if (sctp_chunk_is_data(chunk)) {
286 retval = sctp_packet_can_append_data(packet, chunk);
287 if (retval != SCTP_XMIT_OK)
291 /* Try to bundle AUTH chunk */
292 retval = sctp_packet_bundle_auth(packet, chunk);
293 if (retval != SCTP_XMIT_OK)
296 /* Try to bundle SACK chunk */
297 retval = sctp_packet_bundle_sack(packet, chunk);
298 if (retval != SCTP_XMIT_OK)
301 /* Check to see if this chunk will fit into the packet */
302 retval = sctp_packet_will_fit(packet, chunk, chunk_len);
303 if (retval != SCTP_XMIT_OK)
306 /* We believe that this chunk is OK to add to the packet */
307 switch (chunk->chunk_hdr->type) {
309 /* Account for the data being in the packet */
310 sctp_packet_append_data(packet, chunk);
311 /* Disallow SACK bundling after DATA. */
312 packet->has_sack = 1;
313 /* Disallow AUTH bundling after DATA */
314 packet->has_auth = 1;
315 /* Let it be knows that packet has DATA in it */
316 packet->has_data = 1;
317 /* timestamp the chunk for rtx purposes */
318 chunk->sent_at = jiffies;
320 case SCTP_CID_COOKIE_ECHO:
321 packet->has_cookie_echo = 1;
325 packet->has_sack = 1;
329 packet->has_auth = 1;
330 packet->auth = chunk;
334 /* It is OK to send this chunk. */
335 list_add_tail(&chunk->list, &packet->chunk_list);
336 packet->size += chunk_len;
337 chunk->transport = packet->transport;
342 /* All packets are sent to the network through this function from
345 * The return value is a normal kernel error return value.
347 int sctp_packet_transmit(struct sctp_packet *packet)
349 struct sctp_transport *tp = packet->transport;
350 struct sctp_association *asoc = tp->asoc;
352 struct sk_buff *nskb;
353 struct sctp_chunk *chunk, *tmp;
356 int padding; /* How much padding do we need? */
358 struct dst_entry *dst = tp->dst;
359 unsigned char *auth = NULL; /* pointer to auth in skb data */
360 __u32 cksum_buf_len = sizeof(struct sctphdr);
362 SCTP_DEBUG_PRINTK("%s: packet:%p\n", __func__, packet);
364 /* Do NOT generate a chunkless packet. */
365 if (list_empty(&packet->chunk_list))
368 /* Set up convenience variables... */
369 chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
372 /* Allocate the new skb. */
373 nskb = alloc_skb(packet->size + LL_MAX_HEADER, GFP_ATOMIC);
377 /* Make sure the outbound skb has enough header room reserved. */
378 skb_reserve(nskb, packet->overhead + LL_MAX_HEADER);
380 /* Set the owning socket so that we know where to get the
381 * destination IP address.
383 skb_set_owner_w(nskb, sk);
385 if (!sctp_transport_dst_check(tp)) {
386 sctp_transport_route(tp, NULL, sctp_sk(sk));
387 if (asoc && (asoc->param_flags & SPP_PMTUD_ENABLE)) {
388 sctp_assoc_sync_pmtu(asoc);
391 dst = dst_clone(tp->dst);
392 skb_dst_set(nskb, dst);
396 /* Build the SCTP header. */
397 sh = (struct sctphdr *)skb_push(nskb, sizeof(struct sctphdr));
398 skb_reset_transport_header(nskb);
399 sh->source = htons(packet->source_port);
400 sh->dest = htons(packet->destination_port);
402 /* From 6.8 Adler-32 Checksum Calculation:
403 * After the packet is constructed (containing the SCTP common
404 * header and one or more control or DATA chunks), the
407 * 1) Fill in the proper Verification Tag in the SCTP common
408 * header and initialize the checksum field to 0's.
410 sh->vtag = htonl(packet->vtag);
416 * An endpoint bundles chunks by simply including multiple
417 * chunks in one outbound SCTP packet. ...
421 * 3.2 Chunk Field Descriptions
423 * The total length of a chunk (including Type, Length and
424 * Value fields) MUST be a multiple of 4 bytes. If the length
425 * of the chunk is not a multiple of 4 bytes, the sender MUST
426 * pad the chunk with all zero bytes and this padding is not
427 * included in the chunk length field. The sender should
428 * never pad with more than 3 bytes.
430 * [This whole comment explains WORD_ROUND() below.]
432 SCTP_DEBUG_PRINTK("***sctp_transmit_packet***\n");
433 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
434 list_del_init(&chunk->list);
435 if (sctp_chunk_is_data(chunk)) {
436 /* 6.3.1 C4) When data is in flight and when allowed
437 * by rule C5, a new RTT measurement MUST be made each
438 * round trip. Furthermore, new RTT measurements
439 * SHOULD be made no more than once per round-trip
440 * for a given destination transport address.
443 if (!tp->rto_pending) {
444 chunk->rtt_in_progress = 1;
450 padding = WORD_ROUND(chunk->skb->len) - chunk->skb->len;
452 memset(skb_put(chunk->skb, padding), 0, padding);
454 /* if this is the auth chunk that we are adding,
455 * store pointer where it will be added and put
456 * the auth into the packet.
458 if (chunk == packet->auth)
459 auth = skb_tail_pointer(nskb);
461 cksum_buf_len += chunk->skb->len;
462 memcpy(skb_put(nskb, chunk->skb->len),
463 chunk->skb->data, chunk->skb->len);
465 SCTP_DEBUG_PRINTK("%s %p[%s] %s 0x%x, %s %d, %s %d, %s %d\n",
467 sctp_cname(SCTP_ST_CHUNK(
468 chunk->chunk_hdr->type)),
469 chunk->has_tsn ? "TSN" : "No TSN",
471 ntohl(chunk->subh.data_hdr->tsn) : 0,
472 "length", ntohs(chunk->chunk_hdr->length),
473 "chunk->skb->len", chunk->skb->len,
474 "rtt_in_progress", chunk->rtt_in_progress);
477 * If this is a control chunk, this is our last
478 * reference. Free data chunks after they've been
479 * acknowledged or have failed.
481 if (!sctp_chunk_is_data(chunk))
482 sctp_chunk_free(chunk);
485 /* SCTP-AUTH, Section 6.2
486 * The sender MUST calculate the MAC as described in RFC2104 [2]
487 * using the hash function H as described by the MAC Identifier and
488 * the shared association key K based on the endpoint pair shared key
489 * described by the shared key identifier. The 'data' used for the
490 * computation of the AUTH-chunk is given by the AUTH chunk with its
491 * HMAC field set to zero (as shown in Figure 6) followed by all
492 * chunks that are placed after the AUTH chunk in the SCTP packet.
495 sctp_auth_calculate_hmac(asoc, nskb,
496 (struct sctp_auth_chunk *)auth,
499 /* 2) Calculate the Adler-32 checksum of the whole packet,
500 * including the SCTP common header and all the
503 * Note: Adler-32 is no longer applicable, as has been replaced
504 * by CRC32-C as described in <draft-ietf-tsvwg-sctpcsum-02.txt>.
506 if (!sctp_checksum_disable) {
507 if (!(dst->dev->features & NETIF_F_SCTP_CSUM)) {
508 __u32 crc32 = sctp_start_cksum((__u8 *)sh, cksum_buf_len);
510 /* 3) Put the resultant value into the checksum field in the
511 * common header, and leave the rest of the bits unchanged.
513 sh->checksum = sctp_end_cksum(crc32);
515 /* no need to seed pseudo checksum for SCTP */
516 nskb->ip_summed = CHECKSUM_PARTIAL;
517 nskb->csum_start = (skb_transport_header(nskb) -
519 nskb->csum_offset = offsetof(struct sctphdr, checksum);
523 /* IP layer ECN support
525 * "The ECN-Capable Transport (ECT) bit would be set by the
526 * data sender to indicate that the end-points of the
527 * transport protocol are ECN-capable."
529 * Now setting the ECT bit all the time, as it should not cause
530 * any problems protocol-wise even if our peer ignores it.
532 * Note: The works for IPv6 layer checks this bit too later
533 * in transmission. See IP6_ECN_flow_xmit().
535 (*tp->af_specific->ecn_capable)(nskb->sk);
537 /* Set up the IP options. */
538 /* BUG: not implemented
539 * For v4 this all lives somewhere in sk->sk_opt...
542 /* Dump that on IP! */
543 if (asoc && asoc->peer.last_sent_to != tp) {
544 /* Considering the multiple CPU scenario, this is a
545 * "correcter" place for last_sent_to. --xguo
547 asoc->peer.last_sent_to = tp;
551 struct timer_list *timer;
552 unsigned long timeout;
554 /* Restart the AUTOCLOSE timer when sending data. */
555 if (sctp_state(asoc, ESTABLISHED) && asoc->autoclose) {
556 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
557 timeout = asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
559 if (!mod_timer(timer, jiffies + timeout))
560 sctp_association_hold(asoc);
564 SCTP_DEBUG_PRINTK("***sctp_transmit_packet*** skb len %d\n",
567 nskb->local_df = packet->ipfragok;
568 (*tp->af_specific->sctp_xmit)(nskb, tp);
571 sctp_packet_reset(packet);
575 IP_INC_STATS_BH(&init_net, IPSTATS_MIB_OUTNOROUTES);
577 /* FIXME: Returning the 'err' will effect all the associations
578 * associated with a socket, although only one of the paths of the
579 * association is unreachable.
580 * The real failure of a transport or association can be passed on
581 * to the user via notifications. So setting this error may not be
584 /* err = -EHOSTUNREACH; */
586 /* Control chunks are unreliable so just drop them. DATA chunks
587 * will get resent or dropped later.
590 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
591 list_del_init(&chunk->list);
592 if (!sctp_chunk_is_data(chunk))
593 sctp_chunk_free(chunk);
601 /********************************************************************
602 * 2nd Level Abstractions
603 ********************************************************************/
605 /* This private function check to see if a chunk can be added */
606 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
607 struct sctp_chunk *chunk)
609 sctp_xmit_t retval = SCTP_XMIT_OK;
610 size_t datasize, rwnd, inflight, flight_size;
611 struct sctp_transport *transport = packet->transport;
612 struct sctp_association *asoc = transport->asoc;
613 struct sctp_outq *q = &asoc->outqueue;
615 /* RFC 2960 6.1 Transmission of DATA Chunks
617 * A) At any given time, the data sender MUST NOT transmit new data to
618 * any destination transport address if its peer's rwnd indicates
619 * that the peer has no buffer space (i.e. rwnd is 0, see Section
620 * 6.2.1). However, regardless of the value of rwnd (including if it
621 * is 0), the data sender can always have one DATA chunk in flight to
622 * the receiver if allowed by cwnd (see rule B below). This rule
623 * allows the sender to probe for a change in rwnd that the sender
624 * missed due to the SACK having been lost in transit from the data
625 * receiver to the data sender.
628 rwnd = asoc->peer.rwnd;
629 inflight = q->outstanding_bytes;
630 flight_size = transport->flight_size;
632 datasize = sctp_data_size(chunk);
634 if (datasize > rwnd) {
636 /* We have (at least) one data chunk in flight,
637 * so we can't fall back to rule 6.1 B).
639 retval = SCTP_XMIT_RWND_FULL;
644 /* RFC 2960 6.1 Transmission of DATA Chunks
646 * B) At any given time, the sender MUST NOT transmit new data
647 * to a given transport address if it has cwnd or more bytes
648 * of data outstanding to that transport address.
650 /* RFC 7.2.4 & the Implementers Guide 2.8.
653 * When a Fast Retransmit is being performed the sender SHOULD
654 * ignore the value of cwnd and SHOULD NOT delay retransmission.
656 if (chunk->fast_retransmit != SCTP_NEED_FRTX)
657 if (flight_size >= transport->cwnd) {
658 retval = SCTP_XMIT_RWND_FULL;
662 /* Nagle's algorithm to solve small-packet problem:
663 * Inhibit the sending of new chunks when new outgoing data arrives
664 * if any previously transmitted data on the connection remains
667 if (!sctp_sk(asoc->base.sk)->nodelay && sctp_packet_empty(packet) &&
668 inflight && sctp_state(asoc, ESTABLISHED)) {
669 unsigned int max = transport->pathmtu - packet->overhead;
670 unsigned int len = chunk->skb->len + q->out_qlen;
672 /* Check whether this chunk and all the rest of pending
673 * data will fit or delay in hopes of bundling a full
675 * Don't delay large message writes that may have been
676 * fragmeneted into small peices.
678 if ((len < max) && chunk->msg->can_delay) {
679 retval = SCTP_XMIT_NAGLE_DELAY;
688 /* This private function does management things when adding DATA chunk */
689 static void sctp_packet_append_data(struct sctp_packet *packet,
690 struct sctp_chunk *chunk)
692 struct sctp_transport *transport = packet->transport;
693 size_t datasize = sctp_data_size(chunk);
694 struct sctp_association *asoc = transport->asoc;
695 u32 rwnd = asoc->peer.rwnd;
697 /* Keep track of how many bytes are in flight over this transport. */
698 transport->flight_size += datasize;
700 /* Keep track of how many bytes are in flight to the receiver. */
701 asoc->outqueue.outstanding_bytes += datasize;
703 /* Update our view of the receiver's rwnd. */
709 asoc->peer.rwnd = rwnd;
710 /* Has been accepted for transmission. */
711 if (!asoc->peer.prsctp_capable)
712 chunk->msg->can_abandon = 0;
713 sctp_chunk_assign_tsn(chunk);
714 sctp_chunk_assign_ssn(chunk);
717 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
718 struct sctp_chunk *chunk,
724 sctp_xmit_t retval = SCTP_XMIT_OK;
726 psize = packet->size;
727 pmtu = ((packet->transport->asoc) ?
728 (packet->transport->asoc->pathmtu) :
729 (packet->transport->pathmtu));
731 too_big = (psize + chunk_len > pmtu);
733 /* Decide if we need to fragment or resubmit later. */
735 /* It's OK to fragmet at IP level if any one of the following
737 * 1. The packet is empty (meaning this chunk is greater
739 * 2. The chunk we are adding is a control chunk
740 * 3. The packet doesn't have any data in it yet and data
741 * requires authentication.
743 if (sctp_packet_empty(packet) || !sctp_chunk_is_data(chunk) ||
744 (!packet->has_data && chunk->auth)) {
745 /* We no longer do re-fragmentation.
746 * Just fragment at the IP layer, if we
747 * actually hit this condition
749 packet->ipfragok = 1;
751 retval = SCTP_XMIT_PMTU_FULL;