xref: /linux/net/sctp/output.c (revision 606b2f490fb80e55d05cf0e6cec0b6c0ff0fc18f)
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 #include <linux/types.h>
45 #include <linux/kernel.h>
46 #include <linux/wait.h>
47 #include <linux/time.h>
48 #include <linux/ip.h>
49 #include <linux/ipv6.h>
50 #include <linux/init.h>
51 #include <linux/slab.h>
52 #include <net/inet_ecn.h>
53 #include <net/ip.h>
54 #include <net/icmp.h>
55 #include <net/net_namespace.h>
56 
57 #include <linux/socket.h> /* for sa_family_t */
58 #include <net/sock.h>
59 
60 #include <net/sctp/sctp.h>
61 #include <net/sctp/sm.h>
62 #include <net/sctp/checksum.h>
63 
64 /* Forward declarations for private helpers. */
65 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
66 					   struct sctp_chunk *chunk);
67 static void sctp_packet_append_data(struct sctp_packet *packet,
68 					   struct sctp_chunk *chunk);
69 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
70 					struct sctp_chunk *chunk,
71 					u16 chunk_len);
72 
73 static void sctp_packet_reset(struct sctp_packet *packet)
74 {
75 	packet->size = packet->overhead;
76 	packet->has_cookie_echo = 0;
77 	packet->has_sack = 0;
78 	packet->has_data = 0;
79 	packet->has_auth = 0;
80 	packet->ipfragok = 0;
81 	packet->auth = NULL;
82 }
83 
84 /* Config a packet.
85  * This appears to be a followup set of initializations.
86  */
87 struct sctp_packet *sctp_packet_config(struct sctp_packet *packet,
88 				       __u32 vtag, int ecn_capable)
89 {
90 	struct sctp_chunk *chunk = NULL;
91 
92 	SCTP_DEBUG_PRINTK("%s: packet:%p vtag:0x%x\n", __func__,
93 			  packet, vtag);
94 
95 	sctp_packet_reset(packet);
96 	packet->vtag = vtag;
97 
98 	if (ecn_capable && sctp_packet_empty(packet)) {
99 		chunk = sctp_get_ecne_prepend(packet->transport->asoc);
100 
101 		/* If there a is a prepend chunk stick it on the list before
102 		 * any other chunks get appended.
103 		 */
104 		if (chunk)
105 			sctp_packet_append_chunk(packet, chunk);
106 	}
107 
108 	return packet;
109 }
110 
111 /* Initialize the packet structure. */
112 struct sctp_packet *sctp_packet_init(struct sctp_packet *packet,
113 				     struct sctp_transport *transport,
114 				     __u16 sport, __u16 dport)
115 {
116 	struct sctp_association *asoc = transport->asoc;
117 	size_t overhead;
118 
119 	SCTP_DEBUG_PRINTK("%s: packet:%p transport:%p\n", __func__,
120 			  packet, transport);
121 
122 	packet->transport = transport;
123 	packet->source_port = sport;
124 	packet->destination_port = dport;
125 	INIT_LIST_HEAD(&packet->chunk_list);
126 	if (asoc) {
127 		struct sctp_sock *sp = sctp_sk(asoc->base.sk);
128 		overhead = sp->pf->af->net_header_len;
129 	} else {
130 		overhead = sizeof(struct ipv6hdr);
131 	}
132 	overhead += sizeof(struct sctphdr);
133 	packet->overhead = overhead;
134 	sctp_packet_reset(packet);
135 	packet->vtag = 0;
136 	packet->malloced = 0;
137 	return packet;
138 }
139 
140 /* Free a packet.  */
141 void sctp_packet_free(struct sctp_packet *packet)
142 {
143 	struct sctp_chunk *chunk, *tmp;
144 
145 	SCTP_DEBUG_PRINTK("%s: packet:%p\n", __func__, packet);
146 
147 	list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
148 		list_del_init(&chunk->list);
149 		sctp_chunk_free(chunk);
150 	}
151 
152 	if (packet->malloced)
153 		kfree(packet);
154 }
155 
156 /* This routine tries to append the chunk to the offered packet. If adding
157  * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk
158  * is not present in the packet, it transmits the input packet.
159  * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long
160  * as it can fit in the packet, but any more data that does not fit in this
161  * packet can be sent only after receiving the COOKIE_ACK.
162  */
163 sctp_xmit_t sctp_packet_transmit_chunk(struct sctp_packet *packet,
164 				       struct sctp_chunk *chunk,
165 				       int one_packet)
166 {
167 	sctp_xmit_t retval;
168 	int error = 0;
169 
170 	SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __func__,
171 			  packet, chunk);
172 
173 	switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) {
174 	case SCTP_XMIT_PMTU_FULL:
175 		if (!packet->has_cookie_echo) {
176 			error = sctp_packet_transmit(packet);
177 			if (error < 0)
178 				chunk->skb->sk->sk_err = -error;
179 
180 			/* If we have an empty packet, then we can NOT ever
181 			 * return PMTU_FULL.
182 			 */
183 			if (!one_packet)
184 				retval = sctp_packet_append_chunk(packet,
185 								  chunk);
186 		}
187 		break;
188 
189 	case SCTP_XMIT_RWND_FULL:
190 	case SCTP_XMIT_OK:
191 	case SCTP_XMIT_NAGLE_DELAY:
192 		break;
193 	}
194 
195 	return retval;
196 }
197 
198 /* Try to bundle an auth chunk into the packet. */
199 static sctp_xmit_t sctp_packet_bundle_auth(struct sctp_packet *pkt,
200 					   struct sctp_chunk *chunk)
201 {
202 	struct sctp_association *asoc = pkt->transport->asoc;
203 	struct sctp_chunk *auth;
204 	sctp_xmit_t retval = SCTP_XMIT_OK;
205 
206 	/* if we don't have an association, we can't do authentication */
207 	if (!asoc)
208 		return retval;
209 
210 	/* See if this is an auth chunk we are bundling or if
211 	 * auth is already bundled.
212 	 */
213 	if (chunk->chunk_hdr->type == SCTP_CID_AUTH || pkt->has_auth)
214 		return retval;
215 
216 	/* if the peer did not request this chunk to be authenticated,
217 	 * don't do it
218 	 */
219 	if (!chunk->auth)
220 		return retval;
221 
222 	auth = sctp_make_auth(asoc);
223 	if (!auth)
224 		return retval;
225 
226 	retval = sctp_packet_append_chunk(pkt, auth);
227 
228 	return retval;
229 }
230 
231 /* Try to bundle a SACK with the packet. */
232 static sctp_xmit_t sctp_packet_bundle_sack(struct sctp_packet *pkt,
233 					   struct sctp_chunk *chunk)
234 {
235 	sctp_xmit_t retval = SCTP_XMIT_OK;
236 
237 	/* If sending DATA and haven't aleady bundled a SACK, try to
238 	 * bundle one in to the packet.
239 	 */
240 	if (sctp_chunk_is_data(chunk) && !pkt->has_sack &&
241 	    !pkt->has_cookie_echo) {
242 		struct sctp_association *asoc;
243 		struct timer_list *timer;
244 		asoc = pkt->transport->asoc;
245 		timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
246 
247 		/* If the SACK timer is running, we have a pending SACK */
248 		if (timer_pending(timer)) {
249 			struct sctp_chunk *sack;
250 			asoc->a_rwnd = asoc->rwnd;
251 			sack = sctp_make_sack(asoc);
252 			if (sack) {
253 				retval = sctp_packet_append_chunk(pkt, sack);
254 				asoc->peer.sack_needed = 0;
255 				if (del_timer(timer))
256 					sctp_association_put(asoc);
257 			}
258 		}
259 	}
260 	return retval;
261 }
262 
263 /* Append a chunk to the offered packet reporting back any inability to do
264  * so.
265  */
266 sctp_xmit_t sctp_packet_append_chunk(struct sctp_packet *packet,
267 				     struct sctp_chunk *chunk)
268 {
269 	sctp_xmit_t retval = SCTP_XMIT_OK;
270 	__u16 chunk_len = WORD_ROUND(ntohs(chunk->chunk_hdr->length));
271 
272 	SCTP_DEBUG_PRINTK("%s: packet:%p chunk:%p\n", __func__, packet,
273 			  chunk);
274 
275 	/* Data chunks are special.  Before seeing what else we can
276 	 * bundle into this packet, check to see if we are allowed to
277 	 * send this DATA.
278 	 */
279 	if (sctp_chunk_is_data(chunk)) {
280 		retval = sctp_packet_can_append_data(packet, chunk);
281 		if (retval != SCTP_XMIT_OK)
282 			goto finish;
283 	}
284 
285 	/* Try to bundle AUTH chunk */
286 	retval = sctp_packet_bundle_auth(packet, chunk);
287 	if (retval != SCTP_XMIT_OK)
288 		goto finish;
289 
290 	/* Try to bundle SACK chunk */
291 	retval = sctp_packet_bundle_sack(packet, chunk);
292 	if (retval != SCTP_XMIT_OK)
293 		goto finish;
294 
295 	/* Check to see if this chunk will fit into the packet */
296 	retval = sctp_packet_will_fit(packet, chunk, chunk_len);
297 	if (retval != SCTP_XMIT_OK)
298 		goto finish;
299 
300 	/* We believe that this chunk is OK to add to the packet */
301 	switch (chunk->chunk_hdr->type) {
302 	    case SCTP_CID_DATA:
303 		/* Account for the data being in the packet */
304 		sctp_packet_append_data(packet, chunk);
305 		/* Disallow SACK bundling after DATA. */
306 		packet->has_sack = 1;
307 		/* Disallow AUTH bundling after DATA */
308 		packet->has_auth = 1;
309 		/* Let it be knows that packet has DATA in it */
310 		packet->has_data = 1;
311 		/* timestamp the chunk for rtx purposes */
312 		chunk->sent_at = jiffies;
313 		break;
314 	    case SCTP_CID_COOKIE_ECHO:
315 		packet->has_cookie_echo = 1;
316 		break;
317 
318 	    case SCTP_CID_SACK:
319 		packet->has_sack = 1;
320 		break;
321 
322 	    case SCTP_CID_AUTH:
323 		packet->has_auth = 1;
324 		packet->auth = chunk;
325 		break;
326 	}
327 
328 	/* It is OK to send this chunk.  */
329 	list_add_tail(&chunk->list, &packet->chunk_list);
330 	packet->size += chunk_len;
331 	chunk->transport = packet->transport;
332 finish:
333 	return retval;
334 }
335 
336 /* All packets are sent to the network through this function from
337  * sctp_outq_tail().
338  *
339  * The return value is a normal kernel error return value.
340  */
341 int sctp_packet_transmit(struct sctp_packet *packet)
342 {
343 	struct sctp_transport *tp = packet->transport;
344 	struct sctp_association *asoc = tp->asoc;
345 	struct sctphdr *sh;
346 	struct sk_buff *nskb;
347 	struct sctp_chunk *chunk, *tmp;
348 	struct sock *sk;
349 	int err = 0;
350 	int padding;		/* How much padding do we need?  */
351 	__u8 has_data = 0;
352 	struct dst_entry *dst = tp->dst;
353 	unsigned char *auth = NULL;	/* pointer to auth in skb data */
354 	__u32 cksum_buf_len = sizeof(struct sctphdr);
355 
356 	SCTP_DEBUG_PRINTK("%s: packet:%p\n", __func__, packet);
357 
358 	/* Do NOT generate a chunkless packet. */
359 	if (list_empty(&packet->chunk_list))
360 		return err;
361 
362 	/* Set up convenience variables... */
363 	chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
364 	sk = chunk->skb->sk;
365 
366 	/* Allocate the new skb.  */
367 	nskb = alloc_skb(packet->size + LL_MAX_HEADER, GFP_ATOMIC);
368 	if (!nskb)
369 		goto nomem;
370 
371 	/* Make sure the outbound skb has enough header room reserved. */
372 	skb_reserve(nskb, packet->overhead + LL_MAX_HEADER);
373 
374 	/* Set the owning socket so that we know where to get the
375 	 * destination IP address.
376 	 */
377 	skb_set_owner_w(nskb, sk);
378 
379 	/* The 'obsolete' field of dst is set to 2 when a dst is freed. */
380 	if (!dst || (dst->obsolete > 1)) {
381 		dst_release(dst);
382 		sctp_transport_route(tp, NULL, sctp_sk(sk));
383 		if (asoc && (asoc->param_flags & SPP_PMTUD_ENABLE)) {
384 			sctp_assoc_sync_pmtu(asoc);
385 		}
386 	}
387 	dst = dst_clone(tp->dst);
388 	skb_dst_set(nskb, dst);
389 	if (!dst)
390 		goto no_route;
391 
392 	/* Build the SCTP header.  */
393 	sh = (struct sctphdr *)skb_push(nskb, sizeof(struct sctphdr));
394 	skb_reset_transport_header(nskb);
395 	sh->source = htons(packet->source_port);
396 	sh->dest   = htons(packet->destination_port);
397 
398 	/* From 6.8 Adler-32 Checksum Calculation:
399 	 * After the packet is constructed (containing the SCTP common
400 	 * header and one or more control or DATA chunks), the
401 	 * transmitter shall:
402 	 *
403 	 * 1) Fill in the proper Verification Tag in the SCTP common
404 	 *    header and initialize the checksum field to 0's.
405 	 */
406 	sh->vtag     = htonl(packet->vtag);
407 	sh->checksum = 0;
408 
409 	/**
410 	 * 6.10 Bundling
411 	 *
412 	 *    An endpoint bundles chunks by simply including multiple
413 	 *    chunks in one outbound SCTP packet.  ...
414 	 */
415 
416 	/**
417 	 * 3.2  Chunk Field Descriptions
418 	 *
419 	 * The total length of a chunk (including Type, Length and
420 	 * Value fields) MUST be a multiple of 4 bytes.  If the length
421 	 * of the chunk is not a multiple of 4 bytes, the sender MUST
422 	 * pad the chunk with all zero bytes and this padding is not
423 	 * included in the chunk length field.  The sender should
424 	 * never pad with more than 3 bytes.
425 	 *
426 	 * [This whole comment explains WORD_ROUND() below.]
427 	 */
428 	SCTP_DEBUG_PRINTK("***sctp_transmit_packet***\n");
429 	list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
430 		list_del_init(&chunk->list);
431 		if (sctp_chunk_is_data(chunk)) {
432 			/* 6.3.1 C4) When data is in flight and when allowed
433 			 * by rule C5, a new RTT measurement MUST be made each
434 			 * round trip.  Furthermore, new RTT measurements
435 			 * SHOULD be made no more than once per round-trip
436 			 * for a given destination transport address.
437 			 */
438 
439 			if (!tp->rto_pending) {
440 				chunk->rtt_in_progress = 1;
441 				tp->rto_pending = 1;
442 			}
443 			has_data = 1;
444 		}
445 
446 		padding = WORD_ROUND(chunk->skb->len) - chunk->skb->len;
447 		if (padding)
448 			memset(skb_put(chunk->skb, padding), 0, padding);
449 
450 		/* if this is the auth chunk that we are adding,
451 		 * store pointer where it will be added and put
452 		 * the auth into the packet.
453 		 */
454 		if (chunk == packet->auth)
455 			auth = skb_tail_pointer(nskb);
456 
457 		cksum_buf_len += chunk->skb->len;
458 		memcpy(skb_put(nskb, chunk->skb->len),
459 			       chunk->skb->data, chunk->skb->len);
460 
461 		SCTP_DEBUG_PRINTK("%s %p[%s] %s 0x%x, %s %d, %s %d, %s %d\n",
462 				  "*** Chunk", chunk,
463 				  sctp_cname(SCTP_ST_CHUNK(
464 					  chunk->chunk_hdr->type)),
465 				  chunk->has_tsn ? "TSN" : "No TSN",
466 				  chunk->has_tsn ?
467 				  ntohl(chunk->subh.data_hdr->tsn) : 0,
468 				  "length", ntohs(chunk->chunk_hdr->length),
469 				  "chunk->skb->len", chunk->skb->len,
470 				  "rtt_in_progress", chunk->rtt_in_progress);
471 
472 		/*
473 		 * If this is a control chunk, this is our last
474 		 * reference. Free data chunks after they've been
475 		 * acknowledged or have failed.
476 		 */
477 		if (!sctp_chunk_is_data(chunk))
478 			sctp_chunk_free(chunk);
479 	}
480 
481 	/* SCTP-AUTH, Section 6.2
482 	 *    The sender MUST calculate the MAC as described in RFC2104 [2]
483 	 *    using the hash function H as described by the MAC Identifier and
484 	 *    the shared association key K based on the endpoint pair shared key
485 	 *    described by the shared key identifier.  The 'data' used for the
486 	 *    computation of the AUTH-chunk is given by the AUTH chunk with its
487 	 *    HMAC field set to zero (as shown in Figure 6) followed by all
488 	 *    chunks that are placed after the AUTH chunk in the SCTP packet.
489 	 */
490 	if (auth)
491 		sctp_auth_calculate_hmac(asoc, nskb,
492 					(struct sctp_auth_chunk *)auth,
493 					GFP_ATOMIC);
494 
495 	/* 2) Calculate the Adler-32 checksum of the whole packet,
496 	 *    including the SCTP common header and all the
497 	 *    chunks.
498 	 *
499 	 * Note: Adler-32 is no longer applicable, as has been replaced
500 	 * by CRC32-C as described in <draft-ietf-tsvwg-sctpcsum-02.txt>.
501 	 */
502 	if (!sctp_checksum_disable &&
503 	    !(dst->dev->features & (NETIF_F_NO_CSUM | NETIF_F_SCTP_CSUM))) {
504 		__u32 crc32 = sctp_start_cksum((__u8 *)sh, cksum_buf_len);
505 
506 		/* 3) Put the resultant value into the checksum field in the
507 		 *    common header, and leave the rest of the bits unchanged.
508 		 */
509 		sh->checksum = sctp_end_cksum(crc32);
510 	} else {
511 		if (dst->dev->features & NETIF_F_SCTP_CSUM) {
512 			/* no need to seed psuedo 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 		} else {
518 			nskb->ip_summed = CHECKSUM_UNNECESSARY;
519 		}
520 	}
521 
522 	/* IP layer ECN support
523 	 * From RFC 2481
524 	 *  "The ECN-Capable Transport (ECT) bit would be set by the
525 	 *   data sender to indicate that the end-points of the
526 	 *   transport protocol are ECN-capable."
527 	 *
528 	 * Now setting the ECT bit all the time, as it should not cause
529 	 * any problems protocol-wise even if our peer ignores it.
530 	 *
531 	 * Note: The works for IPv6 layer checks this bit too later
532 	 * in transmission.  See IP6_ECN_flow_xmit().
533 	 */
534 	(*tp->af_specific->ecn_capable)(nskb->sk);
535 
536 	/* Set up the IP options.  */
537 	/* BUG: not implemented
538 	 * For v4 this all lives somewhere in sk->sk_opt...
539 	 */
540 
541 	/* Dump that on IP!  */
542 	if (asoc && asoc->peer.last_sent_to != tp) {
543 		/* Considering the multiple CPU scenario, this is a
544 		 * "correcter" place for last_sent_to.  --xguo
545 		 */
546 		asoc->peer.last_sent_to = tp;
547 	}
548 
549 	if (has_data) {
550 		struct timer_list *timer;
551 		unsigned long timeout;
552 
553 		/* Restart the AUTOCLOSE timer when sending data. */
554 		if (sctp_state(asoc, ESTABLISHED) && asoc->autoclose) {
555 			timer = &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
556 			timeout = asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
557 
558 			if (!mod_timer(timer, jiffies + timeout))
559 				sctp_association_hold(asoc);
560 		}
561 	}
562 
563 	SCTP_DEBUG_PRINTK("***sctp_transmit_packet*** skb len %d\n",
564 			  nskb->len);
565 
566 	nskb->local_df = packet->ipfragok;
567 	(*tp->af_specific->sctp_xmit)(nskb, tp);
568 
569 out:
570 	sctp_packet_reset(packet);
571 	return err;
572 no_route:
573 	kfree_skb(nskb);
574 	IP_INC_STATS_BH(&init_net, IPSTATS_MIB_OUTNOROUTES);
575 
576 	/* FIXME: Returning the 'err' will effect all the associations
577 	 * associated with a socket, although only one of the paths of the
578 	 * association is unreachable.
579 	 * The real failure of a transport or association can be passed on
580 	 * to the user via notifications. So setting this error may not be
581 	 * required.
582 	 */
583 	 /* err = -EHOSTUNREACH; */
584 err:
585 	/* Control chunks are unreliable so just drop them.  DATA chunks
586 	 * will get resent or dropped later.
587 	 */
588 
589 	list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
590 		list_del_init(&chunk->list);
591 		if (!sctp_chunk_is_data(chunk))
592 			sctp_chunk_free(chunk);
593 	}
594 	goto out;
595 nomem:
596 	err = -ENOMEM;
597 	goto err;
598 }
599 
600 /********************************************************************
601  * 2nd Level Abstractions
602  ********************************************************************/
603 
604 /* This private function check to see if a chunk can be added */
605 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
606 					   struct sctp_chunk *chunk)
607 {
608 	sctp_xmit_t retval = SCTP_XMIT_OK;
609 	size_t datasize, rwnd, inflight, flight_size;
610 	struct sctp_transport *transport = packet->transport;
611 	struct sctp_association *asoc = transport->asoc;
612 	struct sctp_outq *q = &asoc->outqueue;
613 
614 	/* RFC 2960 6.1  Transmission of DATA Chunks
615 	 *
616 	 * A) At any given time, the data sender MUST NOT transmit new data to
617 	 * any destination transport address if its peer's rwnd indicates
618 	 * that the peer has no buffer space (i.e. rwnd is 0, see Section
619 	 * 6.2.1).  However, regardless of the value of rwnd (including if it
620 	 * is 0), the data sender can always have one DATA chunk in flight to
621 	 * the receiver if allowed by cwnd (see rule B below).  This rule
622 	 * allows the sender to probe for a change in rwnd that the sender
623 	 * missed due to the SACK having been lost in transit from the data
624 	 * receiver to the data sender.
625 	 */
626 
627 	rwnd = asoc->peer.rwnd;
628 	inflight = q->outstanding_bytes;
629 	flight_size = transport->flight_size;
630 
631 	datasize = sctp_data_size(chunk);
632 
633 	if (datasize > rwnd) {
634 		if (inflight > 0) {
635 			/* We have (at least) one data chunk in flight,
636 			 * so we can't fall back to rule 6.1 B).
637 			 */
638 			retval = SCTP_XMIT_RWND_FULL;
639 			goto finish;
640 		}
641 	}
642 
643 	/* RFC 2960 6.1  Transmission of DATA Chunks
644 	 *
645 	 * B) At any given time, the sender MUST NOT transmit new data
646 	 * to a given transport address if it has cwnd or more bytes
647 	 * of data outstanding to that transport address.
648 	 */
649 	/* RFC 7.2.4 & the Implementers Guide 2.8.
650 	 *
651 	 * 3) ...
652 	 *    When a Fast Retransmit is being performed the sender SHOULD
653 	 *    ignore the value of cwnd and SHOULD NOT delay retransmission.
654 	 */
655 	if (chunk->fast_retransmit != SCTP_NEED_FRTX)
656 		if (flight_size >= transport->cwnd) {
657 			retval = SCTP_XMIT_RWND_FULL;
658 			goto finish;
659 		}
660 
661 	/* Nagle's algorithm to solve small-packet problem:
662 	 * Inhibit the sending of new chunks when new outgoing data arrives
663 	 * if any previously transmitted data on the connection remains
664 	 * unacknowledged.
665 	 */
666 	if (!sctp_sk(asoc->base.sk)->nodelay && sctp_packet_empty(packet) &&
667 	    inflight && sctp_state(asoc, ESTABLISHED)) {
668 		unsigned max = transport->pathmtu - packet->overhead;
669 		unsigned len = chunk->skb->len + q->out_qlen;
670 
671 		/* Check whether this chunk and all the rest of pending
672 		 * data will fit or delay in hopes of bundling a full
673 		 * sized packet.
674 		 * Don't delay large message writes that may have been
675 		 * fragmeneted into small peices.
676 		 */
677 		if ((len < max) && chunk->msg->can_delay) {
678 			retval = SCTP_XMIT_NAGLE_DELAY;
679 			goto finish;
680 		}
681 	}
682 
683 finish:
684 	return retval;
685 }
686 
687 /* This private function does management things when adding DATA chunk */
688 static void sctp_packet_append_data(struct sctp_packet *packet,
689 				struct sctp_chunk *chunk)
690 {
691 	struct sctp_transport *transport = packet->transport;
692 	size_t datasize = sctp_data_size(chunk);
693 	struct sctp_association *asoc = transport->asoc;
694 	u32 rwnd = asoc->peer.rwnd;
695 
696 	/* Keep track of how many bytes are in flight over this transport. */
697 	transport->flight_size += datasize;
698 
699 	/* Keep track of how many bytes are in flight to the receiver. */
700 	asoc->outqueue.outstanding_bytes += datasize;
701 
702 	/* Update our view of the receiver's rwnd. Include sk_buff overhead
703 	 * while updating peer.rwnd so that it reduces the chances of a
704 	 * receiver running out of receive buffer space even when receive
705 	 * window is still open. This can happen when a sender is sending
706 	 * sending small messages.
707 	 */
708 	datasize += sizeof(struct sk_buff);
709 	if (datasize < rwnd)
710 		rwnd -= datasize;
711 	else
712 		rwnd = 0;
713 
714 	asoc->peer.rwnd = rwnd;
715 	/* Has been accepted for transmission. */
716 	if (!asoc->peer.prsctp_capable)
717 		chunk->msg->can_abandon = 0;
718 	sctp_chunk_assign_tsn(chunk);
719 	sctp_chunk_assign_ssn(chunk);
720 }
721 
722 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
723 					struct sctp_chunk *chunk,
724 					u16 chunk_len)
725 {
726 	size_t psize;
727 	size_t pmtu;
728 	int too_big;
729 	sctp_xmit_t retval = SCTP_XMIT_OK;
730 
731 	psize = packet->size;
732 	pmtu  = ((packet->transport->asoc) ?
733 		(packet->transport->asoc->pathmtu) :
734 		(packet->transport->pathmtu));
735 
736 	too_big = (psize + chunk_len > pmtu);
737 
738 	/* Decide if we need to fragment or resubmit later. */
739 	if (too_big) {
740 		/* It's OK to fragmet at IP level if any one of the following
741 		 * is true:
742 		 * 	1. The packet is empty (meaning this chunk is greater
743 		 * 	   the MTU)
744 		 * 	2. The chunk we are adding is a control chunk
745 		 * 	3. The packet doesn't have any data in it yet and data
746 		 * 	requires authentication.
747 		 */
748 		if (sctp_packet_empty(packet) || !sctp_chunk_is_data(chunk) ||
749 		    (!packet->has_data && chunk->auth)) {
750 			/* We no longer do re-fragmentation.
751 			 * Just fragment at the IP layer, if we
752 			 * actually hit this condition
753 			 */
754 			packet->ipfragok = 1;
755 		} else {
756 			retval = SCTP_XMIT_PMTU_FULL;
757 		}
758 	}
759 
760 	return retval;
761 }
762