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_append_chunk(struct sctp_packet *packet,
68 struct sctp_chunk *chunk);
69 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
70 struct sctp_chunk *chunk);
71 static void sctp_packet_append_data(struct sctp_packet *packet,
72 struct sctp_chunk *chunk);
73 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
74 struct sctp_chunk *chunk,
77 static void sctp_packet_reset(struct sctp_packet *packet)
79 packet->size = packet->overhead;
80 packet->has_cookie_echo = 0;
89 * This appears to be a followup set of initializations.
91 struct sctp_packet *sctp_packet_config(struct sctp_packet *packet,
92 __u32 vtag, int ecn_capable)
94 struct sctp_chunk *chunk = NULL;
96 SCTP_DEBUG_PRINTK("%s: packet:%p vtag:0x%x\n", __func__,
101 if (ecn_capable && sctp_packet_empty(packet)) {
102 chunk = sctp_get_ecne_prepend(packet->transport->asoc);
104 /* If there a is a prepend chunk stick it on the list before
105 * any other chunks get appended.
108 sctp_packet_append_chunk(packet, chunk);
114 /* Initialize the packet structure. */
115 struct sctp_packet *sctp_packet_init(struct sctp_packet *packet,
116 struct sctp_transport *transport,
117 __u16 sport, __u16 dport)
119 struct sctp_association *asoc = transport->asoc;
122 SCTP_DEBUG_PRINTK("%s: packet:%p transport:%p\n", __func__,
125 packet->transport = transport;
126 packet->source_port = sport;
127 packet->destination_port = dport;
128 INIT_LIST_HEAD(&packet->chunk_list);
130 struct sctp_sock *sp = sctp_sk(asoc->base.sk);
131 overhead = sp->pf->af->net_header_len;
133 overhead = sizeof(struct ipv6hdr);
135 overhead += sizeof(struct sctphdr);
136 packet->overhead = overhead;
137 sctp_packet_reset(packet);
139 packet->malloced = 0;
144 void sctp_packet_free(struct sctp_packet *packet)
146 struct sctp_chunk *chunk, *tmp;
148 SCTP_DEBUG_PRINTK("%s: packet:%p\n", __func__, packet);
150 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
151 list_del_init(&chunk->list);
152 sctp_chunk_free(chunk);
155 if (packet->malloced)
159 /* This routine tries to append the chunk to the offered packet. If adding
160 * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk
161 * is not present in the packet, it transmits the input packet.
162 * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long
163 * as it can fit in the packet, but any more data that does not fit in this
164 * packet can be sent only after receiving the COOKIE_ACK.
166 sctp_xmit_t sctp_packet_transmit_chunk(struct sctp_packet *packet,
167 struct sctp_chunk *chunk,
173 SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __func__,
176 switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) {
177 case SCTP_XMIT_PMTU_FULL:
178 if (!packet->has_cookie_echo) {
179 error = sctp_packet_transmit(packet);
181 chunk->skb->sk->sk_err = -error;
183 /* If we have an empty packet, then we can NOT ever
187 retval = sctp_packet_append_chunk(packet,
192 case SCTP_XMIT_RWND_FULL:
194 case SCTP_XMIT_NAGLE_DELAY:
201 /* Try to bundle an auth chunk into the packet. */
202 static sctp_xmit_t sctp_packet_bundle_auth(struct sctp_packet *pkt,
203 struct sctp_chunk *chunk)
205 struct sctp_association *asoc = pkt->transport->asoc;
206 struct sctp_chunk *auth;
207 sctp_xmit_t retval = SCTP_XMIT_OK;
209 /* if we don't have an association, we can't do authentication */
213 /* See if this is an auth chunk we are bundling or if
214 * auth is already bundled.
216 if (chunk->chunk_hdr->type == SCTP_CID_AUTH || pkt->has_auth)
219 /* if the peer did not request this chunk to be authenticated,
225 auth = sctp_make_auth(asoc);
229 retval = __sctp_packet_append_chunk(pkt, auth);
231 if (retval != SCTP_XMIT_OK)
232 sctp_chunk_free(auth);
237 /* Try to bundle a SACK with the packet. */
238 static sctp_xmit_t sctp_packet_bundle_sack(struct sctp_packet *pkt,
239 struct sctp_chunk *chunk)
241 sctp_xmit_t retval = SCTP_XMIT_OK;
243 /* If sending DATA and haven't aleady bundled a SACK, try to
244 * bundle one in to the packet.
246 if (sctp_chunk_is_data(chunk) && !pkt->has_sack &&
247 !pkt->has_cookie_echo) {
248 struct sctp_association *asoc;
249 struct timer_list *timer;
250 asoc = pkt->transport->asoc;
251 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
253 /* If the SACK timer is running, we have a pending SACK */
254 if (timer_pending(timer)) {
255 struct sctp_chunk *sack;
257 if (pkt->transport->sack_generation !=
258 pkt->transport->asoc->peer.sack_generation)
261 asoc->a_rwnd = asoc->rwnd;
262 sack = sctp_make_sack(asoc);
264 retval = __sctp_packet_append_chunk(pkt, sack);
265 if (retval != SCTP_XMIT_OK) {
266 sctp_chunk_free(sack);
269 asoc->peer.sack_needed = 0;
270 if (del_timer(timer))
271 sctp_association_put(asoc);
280 /* Append a chunk to the offered packet reporting back any inability to do
283 static sctp_xmit_t __sctp_packet_append_chunk(struct sctp_packet *packet,
284 struct sctp_chunk *chunk)
286 sctp_xmit_t retval = SCTP_XMIT_OK;
287 __u16 chunk_len = WORD_ROUND(ntohs(chunk->chunk_hdr->length));
289 /* Check to see if this chunk will fit into the packet */
290 retval = sctp_packet_will_fit(packet, chunk, chunk_len);
291 if (retval != SCTP_XMIT_OK)
294 /* We believe that this chunk is OK to add to the packet */
295 switch (chunk->chunk_hdr->type) {
297 /* Account for the data being in the packet */
298 sctp_packet_append_data(packet, chunk);
299 /* Disallow SACK bundling after DATA. */
300 packet->has_sack = 1;
301 /* Disallow AUTH bundling after DATA */
302 packet->has_auth = 1;
303 /* Let it be knows that packet has DATA in it */
304 packet->has_data = 1;
305 /* timestamp the chunk for rtx purposes */
306 chunk->sent_at = jiffies;
308 case SCTP_CID_COOKIE_ECHO:
309 packet->has_cookie_echo = 1;
313 packet->has_sack = 1;
315 chunk->asoc->stats.osacks++;
319 packet->has_auth = 1;
320 packet->auth = chunk;
324 /* It is OK to send this chunk. */
325 list_add_tail(&chunk->list, &packet->chunk_list);
326 packet->size += chunk_len;
327 chunk->transport = packet->transport;
332 /* Append a chunk to the offered packet reporting back any inability to do
335 sctp_xmit_t sctp_packet_append_chunk(struct sctp_packet *packet,
336 struct sctp_chunk *chunk)
338 sctp_xmit_t retval = SCTP_XMIT_OK;
340 SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __func__, packet,
343 /* Data chunks are special. Before seeing what else we can
344 * bundle into this packet, check to see if we are allowed to
347 if (sctp_chunk_is_data(chunk)) {
348 retval = sctp_packet_can_append_data(packet, chunk);
349 if (retval != SCTP_XMIT_OK)
353 /* Try to bundle AUTH chunk */
354 retval = sctp_packet_bundle_auth(packet, chunk);
355 if (retval != SCTP_XMIT_OK)
358 /* Try to bundle SACK chunk */
359 retval = sctp_packet_bundle_sack(packet, chunk);
360 if (retval != SCTP_XMIT_OK)
363 retval = __sctp_packet_append_chunk(packet, chunk);
369 static void sctp_packet_release_owner(struct sk_buff *skb)
374 static void sctp_packet_set_owner_w(struct sk_buff *skb, struct sock *sk)
378 skb->destructor = sctp_packet_release_owner;
381 * The data chunks have already been accounted for in sctp_sendmsg(),
382 * therefore only reserve a single byte to keep socket around until
383 * the packet has been transmitted.
385 atomic_inc(&sk->sk_wmem_alloc);
388 /* All packets are sent to the network through this function from
391 * The return value is a normal kernel error return value.
393 int sctp_packet_transmit(struct sctp_packet *packet)
395 struct sctp_transport *tp = packet->transport;
396 struct sctp_association *asoc = tp->asoc;
398 struct sk_buff *nskb;
399 struct sctp_chunk *chunk, *tmp;
402 int padding; /* How much padding do we need? */
404 struct dst_entry *dst = tp->dst;
405 unsigned char *auth = NULL; /* pointer to auth in skb data */
406 __u32 cksum_buf_len = sizeof(struct sctphdr);
408 SCTP_DEBUG_PRINTK("%s: packet:%p\n", __func__, packet);
410 /* Do NOT generate a chunkless packet. */
411 if (list_empty(&packet->chunk_list))
414 /* Set up convenience variables... */
415 chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
418 /* Allocate the new skb. */
419 nskb = alloc_skb(packet->size + LL_MAX_HEADER, GFP_ATOMIC);
423 /* Make sure the outbound skb has enough header room reserved. */
424 skb_reserve(nskb, packet->overhead + LL_MAX_HEADER);
426 /* Set the owning socket so that we know where to get the
427 * destination IP address.
429 sctp_packet_set_owner_w(nskb, sk);
431 if (!sctp_transport_dst_check(tp)) {
432 sctp_transport_route(tp, NULL, sctp_sk(sk));
433 if (asoc && (asoc->param_flags & SPP_PMTUD_ENABLE)) {
434 sctp_assoc_sync_pmtu(sk, asoc);
437 dst = dst_clone(tp->dst);
438 skb_dst_set(nskb, dst);
442 /* Build the SCTP header. */
443 sh = (struct sctphdr *)skb_push(nskb, sizeof(struct sctphdr));
444 skb_reset_transport_header(nskb);
445 sh->source = htons(packet->source_port);
446 sh->dest = htons(packet->destination_port);
448 /* From 6.8 Adler-32 Checksum Calculation:
449 * After the packet is constructed (containing the SCTP common
450 * header and one or more control or DATA chunks), the
453 * 1) Fill in the proper Verification Tag in the SCTP common
454 * header and initialize the checksum field to 0's.
456 sh->vtag = htonl(packet->vtag);
462 * An endpoint bundles chunks by simply including multiple
463 * chunks in one outbound SCTP packet. ...
467 * 3.2 Chunk Field Descriptions
469 * The total length of a chunk (including Type, Length and
470 * Value fields) MUST be a multiple of 4 bytes. If the length
471 * of the chunk is not a multiple of 4 bytes, the sender MUST
472 * pad the chunk with all zero bytes and this padding is not
473 * included in the chunk length field. The sender should
474 * never pad with more than 3 bytes.
476 * [This whole comment explains WORD_ROUND() below.]
478 SCTP_DEBUG_PRINTK("***sctp_transmit_packet***\n");
479 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
480 list_del_init(&chunk->list);
481 if (sctp_chunk_is_data(chunk)) {
482 /* 6.3.1 C4) When data is in flight and when allowed
483 * by rule C5, a new RTT measurement MUST be made each
484 * round trip. Furthermore, new RTT measurements
485 * SHOULD be made no more than once per round-trip
486 * for a given destination transport address.
489 if (!tp->rto_pending) {
490 chunk->rtt_in_progress = 1;
496 padding = WORD_ROUND(chunk->skb->len) - chunk->skb->len;
498 memset(skb_put(chunk->skb, padding), 0, padding);
500 /* if this is the auth chunk that we are adding,
501 * store pointer where it will be added and put
502 * the auth into the packet.
504 if (chunk == packet->auth)
505 auth = skb_tail_pointer(nskb);
507 cksum_buf_len += chunk->skb->len;
508 memcpy(skb_put(nskb, chunk->skb->len),
509 chunk->skb->data, chunk->skb->len);
511 SCTP_DEBUG_PRINTK("%s %p[%s] %s 0x%x, %s %d, %s %d, %s %d\n",
513 sctp_cname(SCTP_ST_CHUNK(
514 chunk->chunk_hdr->type)),
515 chunk->has_tsn ? "TSN" : "No TSN",
517 ntohl(chunk->subh.data_hdr->tsn) : 0,
518 "length", ntohs(chunk->chunk_hdr->length),
519 "chunk->skb->len", chunk->skb->len,
520 "rtt_in_progress", chunk->rtt_in_progress);
523 * If this is a control chunk, this is our last
524 * reference. Free data chunks after they've been
525 * acknowledged or have failed.
527 if (!sctp_chunk_is_data(chunk))
528 sctp_chunk_free(chunk);
531 /* SCTP-AUTH, Section 6.2
532 * The sender MUST calculate the MAC as described in RFC2104 [2]
533 * using the hash function H as described by the MAC Identifier and
534 * the shared association key K based on the endpoint pair shared key
535 * described by the shared key identifier. The 'data' used for the
536 * computation of the AUTH-chunk is given by the AUTH chunk with its
537 * HMAC field set to zero (as shown in Figure 6) followed by all
538 * chunks that are placed after the AUTH chunk in the SCTP packet.
541 sctp_auth_calculate_hmac(asoc, nskb,
542 (struct sctp_auth_chunk *)auth,
545 /* 2) Calculate the Adler-32 checksum of the whole packet,
546 * including the SCTP common header and all the
549 * Note: Adler-32 is no longer applicable, as has been replaced
550 * by CRC32-C as described in <draft-ietf-tsvwg-sctpcsum-02.txt>.
552 if (!sctp_checksum_disable) {
553 if (!(dst->dev->features & NETIF_F_SCTP_CSUM)) {
554 __u32 crc32 = sctp_start_cksum((__u8 *)sh, cksum_buf_len);
556 /* 3) Put the resultant value into the checksum field in the
557 * common header, and leave the rest of the bits unchanged.
559 sh->checksum = sctp_end_cksum(crc32);
561 /* no need to seed pseudo checksum for SCTP */
562 nskb->ip_summed = CHECKSUM_PARTIAL;
563 nskb->csum_start = (skb_transport_header(nskb) -
565 nskb->csum_offset = offsetof(struct sctphdr, checksum);
569 /* IP layer ECN support
571 * "The ECN-Capable Transport (ECT) bit would be set by the
572 * data sender to indicate that the end-points of the
573 * transport protocol are ECN-capable."
575 * Now setting the ECT bit all the time, as it should not cause
576 * any problems protocol-wise even if our peer ignores it.
578 * Note: The works for IPv6 layer checks this bit too later
579 * in transmission. See IP6_ECN_flow_xmit().
581 (*tp->af_specific->ecn_capable)(nskb->sk);
583 /* Set up the IP options. */
584 /* BUG: not implemented
585 * For v4 this all lives somewhere in sk->sk_opt...
588 /* Dump that on IP! */
590 asoc->stats.opackets++;
591 if (asoc->peer.last_sent_to != tp)
592 /* Considering the multiple CPU scenario, this is a
593 * "correcter" place for last_sent_to. --xguo
595 asoc->peer.last_sent_to = tp;
599 struct timer_list *timer;
600 unsigned long timeout;
602 /* Restart the AUTOCLOSE timer when sending data. */
603 if (sctp_state(asoc, ESTABLISHED) && asoc->autoclose) {
604 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
605 timeout = asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
607 if (!mod_timer(timer, jiffies + timeout))
608 sctp_association_hold(asoc);
612 SCTP_DEBUG_PRINTK("***sctp_transmit_packet*** skb len %d\n",
615 nskb->local_df = packet->ipfragok;
616 (*tp->af_specific->sctp_xmit)(nskb, tp);
619 sctp_packet_reset(packet);
623 IP_INC_STATS_BH(sock_net(asoc->base.sk), IPSTATS_MIB_OUTNOROUTES);
625 /* FIXME: Returning the 'err' will effect all the associations
626 * associated with a socket, although only one of the paths of the
627 * association is unreachable.
628 * The real failure of a transport or association can be passed on
629 * to the user via notifications. So setting this error may not be
632 /* err = -EHOSTUNREACH; */
634 /* Control chunks are unreliable so just drop them. DATA chunks
635 * will get resent or dropped later.
638 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
639 list_del_init(&chunk->list);
640 if (!sctp_chunk_is_data(chunk))
641 sctp_chunk_free(chunk);
649 /********************************************************************
650 * 2nd Level Abstractions
651 ********************************************************************/
653 /* This private function check to see if a chunk can be added */
654 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
655 struct sctp_chunk *chunk)
657 sctp_xmit_t retval = SCTP_XMIT_OK;
658 size_t datasize, rwnd, inflight, flight_size;
659 struct sctp_transport *transport = packet->transport;
660 struct sctp_association *asoc = transport->asoc;
661 struct sctp_outq *q = &asoc->outqueue;
663 /* RFC 2960 6.1 Transmission of DATA Chunks
665 * A) At any given time, the data sender MUST NOT transmit new data to
666 * any destination transport address if its peer's rwnd indicates
667 * that the peer has no buffer space (i.e. rwnd is 0, see Section
668 * 6.2.1). However, regardless of the value of rwnd (including if it
669 * is 0), the data sender can always have one DATA chunk in flight to
670 * the receiver if allowed by cwnd (see rule B below). This rule
671 * allows the sender to probe for a change in rwnd that the sender
672 * missed due to the SACK having been lost in transit from the data
673 * receiver to the data sender.
676 rwnd = asoc->peer.rwnd;
677 inflight = q->outstanding_bytes;
678 flight_size = transport->flight_size;
680 datasize = sctp_data_size(chunk);
682 if (datasize > rwnd) {
684 /* We have (at least) one data chunk in flight,
685 * so we can't fall back to rule 6.1 B).
687 retval = SCTP_XMIT_RWND_FULL;
692 /* RFC 2960 6.1 Transmission of DATA Chunks
694 * B) At any given time, the sender MUST NOT transmit new data
695 * to a given transport address if it has cwnd or more bytes
696 * of data outstanding to that transport address.
698 /* RFC 7.2.4 & the Implementers Guide 2.8.
701 * When a Fast Retransmit is being performed the sender SHOULD
702 * ignore the value of cwnd and SHOULD NOT delay retransmission.
704 if (chunk->fast_retransmit != SCTP_NEED_FRTX)
705 if (flight_size >= transport->cwnd) {
706 retval = SCTP_XMIT_RWND_FULL;
710 /* Nagle's algorithm to solve small-packet problem:
711 * Inhibit the sending of new chunks when new outgoing data arrives
712 * if any previously transmitted data on the connection remains
715 if (!sctp_sk(asoc->base.sk)->nodelay && sctp_packet_empty(packet) &&
716 inflight && sctp_state(asoc, ESTABLISHED)) {
717 unsigned int max = transport->pathmtu - packet->overhead;
718 unsigned int len = chunk->skb->len + q->out_qlen;
720 /* Check whether this chunk and all the rest of pending
721 * data will fit or delay in hopes of bundling a full
723 * Don't delay large message writes that may have been
724 * fragmeneted into small peices.
726 if ((len < max) && chunk->msg->can_delay) {
727 retval = SCTP_XMIT_NAGLE_DELAY;
736 /* This private function does management things when adding DATA chunk */
737 static void sctp_packet_append_data(struct sctp_packet *packet,
738 struct sctp_chunk *chunk)
740 struct sctp_transport *transport = packet->transport;
741 size_t datasize = sctp_data_size(chunk);
742 struct sctp_association *asoc = transport->asoc;
743 u32 rwnd = asoc->peer.rwnd;
745 /* Keep track of how many bytes are in flight over this transport. */
746 transport->flight_size += datasize;
748 /* Keep track of how many bytes are in flight to the receiver. */
749 asoc->outqueue.outstanding_bytes += datasize;
751 /* Update our view of the receiver's rwnd. */
757 asoc->peer.rwnd = rwnd;
758 /* Has been accepted for transmission. */
759 if (!asoc->peer.prsctp_capable)
760 chunk->msg->can_abandon = 0;
761 sctp_chunk_assign_tsn(chunk);
762 sctp_chunk_assign_ssn(chunk);
765 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
766 struct sctp_chunk *chunk,
772 sctp_xmit_t retval = SCTP_XMIT_OK;
774 psize = packet->size;
775 pmtu = ((packet->transport->asoc) ?
776 (packet->transport->asoc->pathmtu) :
777 (packet->transport->pathmtu));
779 too_big = (psize + chunk_len > pmtu);
781 /* Decide if we need to fragment or resubmit later. */
783 /* It's OK to fragmet at IP level if any one of the following
785 * 1. The packet is empty (meaning this chunk is greater
787 * 2. The chunk we are adding is a control chunk
788 * 3. The packet doesn't have any data in it yet and data
789 * requires authentication.
791 if (sctp_packet_empty(packet) || !sctp_chunk_is_data(chunk) ||
792 (!packet->has_data && chunk->auth)) {
793 /* We no longer do re-fragmentation.
794 * Just fragment at the IP layer, if we
795 * actually hit this condition
797 packet->ipfragok = 1;
799 retval = SCTP_XMIT_PMTU_FULL;