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;
317 packet->has_auth = 1;
318 packet->auth = chunk;
322 /* It is OK to send this chunk. */
323 list_add_tail(&chunk->list, &packet->chunk_list);
324 packet->size += chunk_len;
325 chunk->transport = packet->transport;
330 /* Append a chunk to the offered packet reporting back any inability to do
333 sctp_xmit_t sctp_packet_append_chunk(struct sctp_packet *packet,
334 struct sctp_chunk *chunk)
336 sctp_xmit_t retval = SCTP_XMIT_OK;
338 SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __func__, packet,
341 /* Data chunks are special. Before seeing what else we can
342 * bundle into this packet, check to see if we are allowed to
345 if (sctp_chunk_is_data(chunk)) {
346 retval = sctp_packet_can_append_data(packet, chunk);
347 if (retval != SCTP_XMIT_OK)
351 /* Try to bundle AUTH chunk */
352 retval = sctp_packet_bundle_auth(packet, chunk);
353 if (retval != SCTP_XMIT_OK)
356 /* Try to bundle SACK chunk */
357 retval = sctp_packet_bundle_sack(packet, chunk);
358 if (retval != SCTP_XMIT_OK)
361 retval = __sctp_packet_append_chunk(packet, chunk);
367 /* All packets are sent to the network through this function from
370 * The return value is a normal kernel error return value.
372 int sctp_packet_transmit(struct sctp_packet *packet)
374 struct sctp_transport *tp = packet->transport;
375 struct sctp_association *asoc = tp->asoc;
377 struct sk_buff *nskb;
378 struct sctp_chunk *chunk, *tmp;
381 int padding; /* How much padding do we need? */
383 struct dst_entry *dst = tp->dst;
384 unsigned char *auth = NULL; /* pointer to auth in skb data */
385 __u32 cksum_buf_len = sizeof(struct sctphdr);
387 SCTP_DEBUG_PRINTK("%s: packet:%p\n", __func__, packet);
389 /* Do NOT generate a chunkless packet. */
390 if (list_empty(&packet->chunk_list))
393 /* Set up convenience variables... */
394 chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
397 /* Allocate the new skb. */
398 nskb = alloc_skb(packet->size + LL_MAX_HEADER, GFP_ATOMIC);
402 /* Make sure the outbound skb has enough header room reserved. */
403 skb_reserve(nskb, packet->overhead + LL_MAX_HEADER);
405 /* Set the owning socket so that we know where to get the
406 * destination IP address.
408 skb_set_owner_w(nskb, sk);
410 if (!sctp_transport_dst_check(tp)) {
411 sctp_transport_route(tp, NULL, sctp_sk(sk));
412 if (asoc && (asoc->param_flags & SPP_PMTUD_ENABLE)) {
413 sctp_assoc_sync_pmtu(sk, asoc);
416 dst = dst_clone(tp->dst);
417 skb_dst_set(nskb, dst);
421 /* Build the SCTP header. */
422 sh = (struct sctphdr *)skb_push(nskb, sizeof(struct sctphdr));
423 skb_reset_transport_header(nskb);
424 sh->source = htons(packet->source_port);
425 sh->dest = htons(packet->destination_port);
427 /* From 6.8 Adler-32 Checksum Calculation:
428 * After the packet is constructed (containing the SCTP common
429 * header and one or more control or DATA chunks), the
432 * 1) Fill in the proper Verification Tag in the SCTP common
433 * header and initialize the checksum field to 0's.
435 sh->vtag = htonl(packet->vtag);
441 * An endpoint bundles chunks by simply including multiple
442 * chunks in one outbound SCTP packet. ...
446 * 3.2 Chunk Field Descriptions
448 * The total length of a chunk (including Type, Length and
449 * Value fields) MUST be a multiple of 4 bytes. If the length
450 * of the chunk is not a multiple of 4 bytes, the sender MUST
451 * pad the chunk with all zero bytes and this padding is not
452 * included in the chunk length field. The sender should
453 * never pad with more than 3 bytes.
455 * [This whole comment explains WORD_ROUND() below.]
457 SCTP_DEBUG_PRINTK("***sctp_transmit_packet***\n");
458 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
459 list_del_init(&chunk->list);
460 if (sctp_chunk_is_data(chunk)) {
461 /* 6.3.1 C4) When data is in flight and when allowed
462 * by rule C5, a new RTT measurement MUST be made each
463 * round trip. Furthermore, new RTT measurements
464 * SHOULD be made no more than once per round-trip
465 * for a given destination transport address.
468 if (!tp->rto_pending) {
469 chunk->rtt_in_progress = 1;
475 padding = WORD_ROUND(chunk->skb->len) - chunk->skb->len;
477 memset(skb_put(chunk->skb, padding), 0, padding);
479 /* if this is the auth chunk that we are adding,
480 * store pointer where it will be added and put
481 * the auth into the packet.
483 if (chunk == packet->auth)
484 auth = skb_tail_pointer(nskb);
486 cksum_buf_len += chunk->skb->len;
487 memcpy(skb_put(nskb, chunk->skb->len),
488 chunk->skb->data, chunk->skb->len);
490 SCTP_DEBUG_PRINTK("%s %p[%s] %s 0x%x, %s %d, %s %d, %s %d\n",
492 sctp_cname(SCTP_ST_CHUNK(
493 chunk->chunk_hdr->type)),
494 chunk->has_tsn ? "TSN" : "No TSN",
496 ntohl(chunk->subh.data_hdr->tsn) : 0,
497 "length", ntohs(chunk->chunk_hdr->length),
498 "chunk->skb->len", chunk->skb->len,
499 "rtt_in_progress", chunk->rtt_in_progress);
502 * If this is a control chunk, this is our last
503 * reference. Free data chunks after they've been
504 * acknowledged or have failed.
506 if (!sctp_chunk_is_data(chunk))
507 sctp_chunk_free(chunk);
510 /* SCTP-AUTH, Section 6.2
511 * The sender MUST calculate the MAC as described in RFC2104 [2]
512 * using the hash function H as described by the MAC Identifier and
513 * the shared association key K based on the endpoint pair shared key
514 * described by the shared key identifier. The 'data' used for the
515 * computation of the AUTH-chunk is given by the AUTH chunk with its
516 * HMAC field set to zero (as shown in Figure 6) followed by all
517 * chunks that are placed after the AUTH chunk in the SCTP packet.
520 sctp_auth_calculate_hmac(asoc, nskb,
521 (struct sctp_auth_chunk *)auth,
524 /* 2) Calculate the Adler-32 checksum of the whole packet,
525 * including the SCTP common header and all the
528 * Note: Adler-32 is no longer applicable, as has been replaced
529 * by CRC32-C as described in <draft-ietf-tsvwg-sctpcsum-02.txt>.
531 if (!sctp_checksum_disable) {
532 if (!(dst->dev->features & NETIF_F_SCTP_CSUM)) {
533 __u32 crc32 = sctp_start_cksum((__u8 *)sh, cksum_buf_len);
535 /* 3) Put the resultant value into the checksum field in the
536 * common header, and leave the rest of the bits unchanged.
538 sh->checksum = sctp_end_cksum(crc32);
540 /* no need to seed pseudo checksum for SCTP */
541 nskb->ip_summed = CHECKSUM_PARTIAL;
542 nskb->csum_start = (skb_transport_header(nskb) -
544 nskb->csum_offset = offsetof(struct sctphdr, checksum);
548 /* IP layer ECN support
550 * "The ECN-Capable Transport (ECT) bit would be set by the
551 * data sender to indicate that the end-points of the
552 * transport protocol are ECN-capable."
554 * Now setting the ECT bit all the time, as it should not cause
555 * any problems protocol-wise even if our peer ignores it.
557 * Note: The works for IPv6 layer checks this bit too later
558 * in transmission. See IP6_ECN_flow_xmit().
560 (*tp->af_specific->ecn_capable)(nskb->sk);
562 /* Set up the IP options. */
563 /* BUG: not implemented
564 * For v4 this all lives somewhere in sk->sk_opt...
567 /* Dump that on IP! */
568 if (asoc && asoc->peer.last_sent_to != tp) {
569 /* Considering the multiple CPU scenario, this is a
570 * "correcter" place for last_sent_to. --xguo
572 asoc->peer.last_sent_to = tp;
576 struct timer_list *timer;
577 unsigned long timeout;
579 /* Restart the AUTOCLOSE timer when sending data. */
580 if (sctp_state(asoc, ESTABLISHED) && asoc->autoclose) {
581 timer = &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
582 timeout = asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
584 if (!mod_timer(timer, jiffies + timeout))
585 sctp_association_hold(asoc);
589 SCTP_DEBUG_PRINTK("***sctp_transmit_packet*** skb len %d\n",
592 nskb->local_df = packet->ipfragok;
593 (*tp->af_specific->sctp_xmit)(nskb, tp);
596 sctp_packet_reset(packet);
600 IP_INC_STATS_BH(&init_net, IPSTATS_MIB_OUTNOROUTES);
602 /* FIXME: Returning the 'err' will effect all the associations
603 * associated with a socket, although only one of the paths of the
604 * association is unreachable.
605 * The real failure of a transport or association can be passed on
606 * to the user via notifications. So setting this error may not be
609 /* err = -EHOSTUNREACH; */
611 /* Control chunks are unreliable so just drop them. DATA chunks
612 * will get resent or dropped later.
615 list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
616 list_del_init(&chunk->list);
617 if (!sctp_chunk_is_data(chunk))
618 sctp_chunk_free(chunk);
626 /********************************************************************
627 * 2nd Level Abstractions
628 ********************************************************************/
630 /* This private function check to see if a chunk can be added */
631 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
632 struct sctp_chunk *chunk)
634 sctp_xmit_t retval = SCTP_XMIT_OK;
635 size_t datasize, rwnd, inflight, flight_size;
636 struct sctp_transport *transport = packet->transport;
637 struct sctp_association *asoc = transport->asoc;
638 struct sctp_outq *q = &asoc->outqueue;
640 /* RFC 2960 6.1 Transmission of DATA Chunks
642 * A) At any given time, the data sender MUST NOT transmit new data to
643 * any destination transport address if its peer's rwnd indicates
644 * that the peer has no buffer space (i.e. rwnd is 0, see Section
645 * 6.2.1). However, regardless of the value of rwnd (including if it
646 * is 0), the data sender can always have one DATA chunk in flight to
647 * the receiver if allowed by cwnd (see rule B below). This rule
648 * allows the sender to probe for a change in rwnd that the sender
649 * missed due to the SACK having been lost in transit from the data
650 * receiver to the data sender.
653 rwnd = asoc->peer.rwnd;
654 inflight = q->outstanding_bytes;
655 flight_size = transport->flight_size;
657 datasize = sctp_data_size(chunk);
659 if (datasize > rwnd) {
661 /* We have (at least) one data chunk in flight,
662 * so we can't fall back to rule 6.1 B).
664 retval = SCTP_XMIT_RWND_FULL;
669 /* RFC 2960 6.1 Transmission of DATA Chunks
671 * B) At any given time, the sender MUST NOT transmit new data
672 * to a given transport address if it has cwnd or more bytes
673 * of data outstanding to that transport address.
675 /* RFC 7.2.4 & the Implementers Guide 2.8.
678 * When a Fast Retransmit is being performed the sender SHOULD
679 * ignore the value of cwnd and SHOULD NOT delay retransmission.
681 if (chunk->fast_retransmit != SCTP_NEED_FRTX)
682 if (flight_size >= transport->cwnd) {
683 retval = SCTP_XMIT_RWND_FULL;
687 /* Nagle's algorithm to solve small-packet problem:
688 * Inhibit the sending of new chunks when new outgoing data arrives
689 * if any previously transmitted data on the connection remains
692 if (!sctp_sk(asoc->base.sk)->nodelay && sctp_packet_empty(packet) &&
693 inflight && sctp_state(asoc, ESTABLISHED)) {
694 unsigned int max = transport->pathmtu - packet->overhead;
695 unsigned int len = chunk->skb->len + q->out_qlen;
697 /* Check whether this chunk and all the rest of pending
698 * data will fit or delay in hopes of bundling a full
700 * Don't delay large message writes that may have been
701 * fragmeneted into small peices.
703 if ((len < max) && chunk->msg->can_delay) {
704 retval = SCTP_XMIT_NAGLE_DELAY;
713 /* This private function does management things when adding DATA chunk */
714 static void sctp_packet_append_data(struct sctp_packet *packet,
715 struct sctp_chunk *chunk)
717 struct sctp_transport *transport = packet->transport;
718 size_t datasize = sctp_data_size(chunk);
719 struct sctp_association *asoc = transport->asoc;
720 u32 rwnd = asoc->peer.rwnd;
722 /* Keep track of how many bytes are in flight over this transport. */
723 transport->flight_size += datasize;
725 /* Keep track of how many bytes are in flight to the receiver. */
726 asoc->outqueue.outstanding_bytes += datasize;
728 /* Update our view of the receiver's rwnd. */
734 asoc->peer.rwnd = rwnd;
735 /* Has been accepted for transmission. */
736 if (!asoc->peer.prsctp_capable)
737 chunk->msg->can_abandon = 0;
738 sctp_chunk_assign_tsn(chunk);
739 sctp_chunk_assign_ssn(chunk);
742 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
743 struct sctp_chunk *chunk,
749 sctp_xmit_t retval = SCTP_XMIT_OK;
751 psize = packet->size;
752 pmtu = ((packet->transport->asoc) ?
753 (packet->transport->asoc->pathmtu) :
754 (packet->transport->pathmtu));
756 too_big = (psize + chunk_len > pmtu);
758 /* Decide if we need to fragment or resubmit later. */
760 /* It's OK to fragmet at IP level if any one of the following
762 * 1. The packet is empty (meaning this chunk is greater
764 * 2. The chunk we are adding is a control chunk
765 * 3. The packet doesn't have any data in it yet and data
766 * requires authentication.
768 if (sctp_packet_empty(packet) || !sctp_chunk_is_data(chunk) ||
769 (!packet->has_data && chunk->auth)) {
770 /* We no longer do re-fragmentation.
771 * Just fragment at the IP layer, if we
772 * actually hit this condition
774 packet->ipfragok = 1;
776 retval = SCTP_XMIT_PMTU_FULL;