xref: /openbmc/linux/net/sctp/output.c (revision 80483c3a)
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. */
60 static sctp_xmit_t __sctp_packet_append_chunk(struct sctp_packet *packet,
61 					      struct sctp_chunk *chunk);
62 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
63 					   struct sctp_chunk *chunk);
64 static void sctp_packet_append_data(struct sctp_packet *packet,
65 					   struct sctp_chunk *chunk);
66 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
67 					struct sctp_chunk *chunk,
68 					u16 chunk_len);
69 
70 static void sctp_packet_reset(struct sctp_packet *packet)
71 {
72 	packet->size = packet->overhead;
73 	packet->has_cookie_echo = 0;
74 	packet->has_sack = 0;
75 	packet->has_data = 0;
76 	packet->has_auth = 0;
77 	packet->ipfragok = 0;
78 	packet->auth = NULL;
79 }
80 
81 /* Config a packet.
82  * This appears to be a followup set of initializations.
83  */
84 struct sctp_packet *sctp_packet_config(struct sctp_packet *packet,
85 				       __u32 vtag, int ecn_capable)
86 {
87 	struct sctp_transport *tp = packet->transport;
88 	struct sctp_association *asoc = tp->asoc;
89 
90 	pr_debug("%s: packet:%p vtag:0x%x\n", __func__, packet, vtag);
91 
92 	packet->vtag = vtag;
93 
94 	if (asoc && tp->dst) {
95 		struct sock *sk = asoc->base.sk;
96 
97 		rcu_read_lock();
98 		if (__sk_dst_get(sk) != tp->dst) {
99 			dst_hold(tp->dst);
100 			sk_setup_caps(sk, tp->dst);
101 		}
102 
103 		if (sk_can_gso(sk)) {
104 			struct net_device *dev = tp->dst->dev;
105 
106 			packet->max_size = dev->gso_max_size;
107 		} else {
108 			packet->max_size = asoc->pathmtu;
109 		}
110 		rcu_read_unlock();
111 
112 	} else {
113 		packet->max_size = tp->pathmtu;
114 	}
115 
116 	if (ecn_capable && sctp_packet_empty(packet)) {
117 		struct sctp_chunk *chunk;
118 
119 		/* If there a is a prepend chunk stick it on the list before
120 		 * any other chunks get appended.
121 		 */
122 		chunk = sctp_get_ecne_prepend(asoc);
123 		if (chunk)
124 			sctp_packet_append_chunk(packet, chunk);
125 	}
126 
127 	return packet;
128 }
129 
130 /* Initialize the packet structure. */
131 struct sctp_packet *sctp_packet_init(struct sctp_packet *packet,
132 				     struct sctp_transport *transport,
133 				     __u16 sport, __u16 dport)
134 {
135 	struct sctp_association *asoc = transport->asoc;
136 	size_t overhead;
137 
138 	pr_debug("%s: packet:%p transport:%p\n", __func__, packet, transport);
139 
140 	packet->transport = transport;
141 	packet->source_port = sport;
142 	packet->destination_port = dport;
143 	INIT_LIST_HEAD(&packet->chunk_list);
144 	if (asoc) {
145 		struct sctp_sock *sp = sctp_sk(asoc->base.sk);
146 		overhead = sp->pf->af->net_header_len;
147 	} else {
148 		overhead = sizeof(struct ipv6hdr);
149 	}
150 	overhead += sizeof(struct sctphdr);
151 	packet->overhead = overhead;
152 	sctp_packet_reset(packet);
153 	packet->vtag = 0;
154 
155 	return packet;
156 }
157 
158 /* Free a packet.  */
159 void sctp_packet_free(struct sctp_packet *packet)
160 {
161 	struct sctp_chunk *chunk, *tmp;
162 
163 	pr_debug("%s: packet:%p\n", __func__, packet);
164 
165 	list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
166 		list_del_init(&chunk->list);
167 		sctp_chunk_free(chunk);
168 	}
169 }
170 
171 /* This routine tries to append the chunk to the offered packet. If adding
172  * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk
173  * is not present in the packet, it transmits the input packet.
174  * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long
175  * as it can fit in the packet, but any more data that does not fit in this
176  * packet can be sent only after receiving the COOKIE_ACK.
177  */
178 sctp_xmit_t sctp_packet_transmit_chunk(struct sctp_packet *packet,
179 				       struct sctp_chunk *chunk,
180 				       int one_packet, gfp_t gfp)
181 {
182 	sctp_xmit_t retval;
183 	int error = 0;
184 
185 	pr_debug("%s: packet:%p size:%Zu chunk:%p size:%d\n", __func__,
186 		 packet, packet->size, chunk, chunk->skb ? chunk->skb->len : -1);
187 
188 	switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) {
189 	case SCTP_XMIT_PMTU_FULL:
190 		if (!packet->has_cookie_echo) {
191 			error = sctp_packet_transmit(packet, gfp);
192 			if (error < 0)
193 				chunk->skb->sk->sk_err = -error;
194 
195 			/* If we have an empty packet, then we can NOT ever
196 			 * return PMTU_FULL.
197 			 */
198 			if (!one_packet)
199 				retval = sctp_packet_append_chunk(packet,
200 								  chunk);
201 		}
202 		break;
203 
204 	case SCTP_XMIT_RWND_FULL:
205 	case SCTP_XMIT_OK:
206 	case SCTP_XMIT_DELAY:
207 		break;
208 	}
209 
210 	return retval;
211 }
212 
213 /* Try to bundle an auth chunk into the packet. */
214 static sctp_xmit_t sctp_packet_bundle_auth(struct sctp_packet *pkt,
215 					   struct sctp_chunk *chunk)
216 {
217 	struct sctp_association *asoc = pkt->transport->asoc;
218 	struct sctp_chunk *auth;
219 	sctp_xmit_t retval = SCTP_XMIT_OK;
220 
221 	/* if we don't have an association, we can't do authentication */
222 	if (!asoc)
223 		return retval;
224 
225 	/* See if this is an auth chunk we are bundling or if
226 	 * auth is already bundled.
227 	 */
228 	if (chunk->chunk_hdr->type == SCTP_CID_AUTH || pkt->has_auth)
229 		return retval;
230 
231 	/* if the peer did not request this chunk to be authenticated,
232 	 * don't do it
233 	 */
234 	if (!chunk->auth)
235 		return retval;
236 
237 	auth = sctp_make_auth(asoc);
238 	if (!auth)
239 		return retval;
240 
241 	retval = __sctp_packet_append_chunk(pkt, auth);
242 
243 	if (retval != SCTP_XMIT_OK)
244 		sctp_chunk_free(auth);
245 
246 	return retval;
247 }
248 
249 /* Try to bundle a SACK with the packet. */
250 static sctp_xmit_t sctp_packet_bundle_sack(struct sctp_packet *pkt,
251 					   struct sctp_chunk *chunk)
252 {
253 	sctp_xmit_t retval = SCTP_XMIT_OK;
254 
255 	/* If sending DATA and haven't aleady bundled a SACK, try to
256 	 * bundle one in to the packet.
257 	 */
258 	if (sctp_chunk_is_data(chunk) && !pkt->has_sack &&
259 	    !pkt->has_cookie_echo) {
260 		struct sctp_association *asoc;
261 		struct timer_list *timer;
262 		asoc = pkt->transport->asoc;
263 		timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
264 
265 		/* If the SACK timer is running, we have a pending SACK */
266 		if (timer_pending(timer)) {
267 			struct sctp_chunk *sack;
268 
269 			if (pkt->transport->sack_generation !=
270 			    pkt->transport->asoc->peer.sack_generation)
271 				return retval;
272 
273 			asoc->a_rwnd = asoc->rwnd;
274 			sack = sctp_make_sack(asoc);
275 			if (sack) {
276 				retval = __sctp_packet_append_chunk(pkt, sack);
277 				if (retval != SCTP_XMIT_OK) {
278 					sctp_chunk_free(sack);
279 					goto out;
280 				}
281 				asoc->peer.sack_needed = 0;
282 				if (del_timer(timer))
283 					sctp_association_put(asoc);
284 			}
285 		}
286 	}
287 out:
288 	return retval;
289 }
290 
291 
292 /* Append a chunk to the offered packet reporting back any inability to do
293  * so.
294  */
295 static sctp_xmit_t __sctp_packet_append_chunk(struct sctp_packet *packet,
296 					      struct sctp_chunk *chunk)
297 {
298 	sctp_xmit_t retval = SCTP_XMIT_OK;
299 	__u16 chunk_len = WORD_ROUND(ntohs(chunk->chunk_hdr->length));
300 
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)
304 		goto finish;
305 
306 	/* We believe that this chunk is OK to add to the packet */
307 	switch (chunk->chunk_hdr->type) {
308 	case SCTP_CID_DATA:
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;
319 		/* Mainly used for prsctp RTX policy */
320 		chunk->sent_count++;
321 		break;
322 	case SCTP_CID_COOKIE_ECHO:
323 		packet->has_cookie_echo = 1;
324 		break;
325 
326 	case SCTP_CID_SACK:
327 		packet->has_sack = 1;
328 		if (chunk->asoc)
329 			chunk->asoc->stats.osacks++;
330 		break;
331 
332 	case SCTP_CID_AUTH:
333 		packet->has_auth = 1;
334 		packet->auth = chunk;
335 		break;
336 	}
337 
338 	/* It is OK to send this chunk.  */
339 	list_add_tail(&chunk->list, &packet->chunk_list);
340 	packet->size += chunk_len;
341 	chunk->transport = packet->transport;
342 finish:
343 	return retval;
344 }
345 
346 /* Append a chunk to the offered packet reporting back any inability to do
347  * so.
348  */
349 sctp_xmit_t sctp_packet_append_chunk(struct sctp_packet *packet,
350 				     struct sctp_chunk *chunk)
351 {
352 	sctp_xmit_t retval = SCTP_XMIT_OK;
353 
354 	pr_debug("%s: packet:%p chunk:%p\n", __func__, packet, chunk);
355 
356 	/* Data chunks are special.  Before seeing what else we can
357 	 * bundle into this packet, check to see if we are allowed to
358 	 * send this DATA.
359 	 */
360 	if (sctp_chunk_is_data(chunk)) {
361 		retval = sctp_packet_can_append_data(packet, chunk);
362 		if (retval != SCTP_XMIT_OK)
363 			goto finish;
364 	}
365 
366 	/* Try to bundle AUTH chunk */
367 	retval = sctp_packet_bundle_auth(packet, chunk);
368 	if (retval != SCTP_XMIT_OK)
369 		goto finish;
370 
371 	/* Try to bundle SACK chunk */
372 	retval = sctp_packet_bundle_sack(packet, chunk);
373 	if (retval != SCTP_XMIT_OK)
374 		goto finish;
375 
376 	retval = __sctp_packet_append_chunk(packet, chunk);
377 
378 finish:
379 	return retval;
380 }
381 
382 static void sctp_packet_release_owner(struct sk_buff *skb)
383 {
384 	sk_free(skb->sk);
385 }
386 
387 static void sctp_packet_set_owner_w(struct sk_buff *skb, struct sock *sk)
388 {
389 	skb_orphan(skb);
390 	skb->sk = sk;
391 	skb->destructor = sctp_packet_release_owner;
392 
393 	/*
394 	 * The data chunks have already been accounted for in sctp_sendmsg(),
395 	 * therefore only reserve a single byte to keep socket around until
396 	 * the packet has been transmitted.
397 	 */
398 	atomic_inc(&sk->sk_wmem_alloc);
399 }
400 
401 /* All packets are sent to the network through this function from
402  * sctp_outq_tail().
403  *
404  * The return value is a normal kernel error return value.
405  */
406 int sctp_packet_transmit(struct sctp_packet *packet, gfp_t gfp)
407 {
408 	struct sctp_transport *tp = packet->transport;
409 	struct sctp_association *asoc = tp->asoc;
410 	struct sctphdr *sh;
411 	struct sk_buff *nskb = NULL, *head = NULL;
412 	struct sctp_chunk *chunk, *tmp;
413 	struct sock *sk;
414 	int err = 0;
415 	int padding;		/* How much padding do we need?  */
416 	int pkt_size;
417 	__u8 has_data = 0;
418 	int gso = 0;
419 	int pktcount = 0;
420 	struct dst_entry *dst;
421 	unsigned char *auth = NULL;	/* pointer to auth in skb data */
422 
423 	pr_debug("%s: packet:%p\n", __func__, packet);
424 
425 	/* Do NOT generate a chunkless packet. */
426 	if (list_empty(&packet->chunk_list))
427 		return err;
428 
429 	/* Set up convenience variables... */
430 	chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
431 	sk = chunk->skb->sk;
432 
433 	/* Allocate the head skb, or main one if not in GSO */
434 	if (packet->size > tp->pathmtu && !packet->ipfragok) {
435 		if (sk_can_gso(sk)) {
436 			gso = 1;
437 			pkt_size = packet->overhead;
438 		} else {
439 			/* If this happens, we trash this packet and try
440 			 * to build a new one, hopefully correct this
441 			 * time. Application may notice this error.
442 			 */
443 			pr_err_once("Trying to GSO but underlying device doesn't support it.");
444 			goto nomem;
445 		}
446 	} else {
447 		pkt_size = packet->size;
448 	}
449 	head = alloc_skb(pkt_size + MAX_HEADER, gfp);
450 	if (!head)
451 		goto nomem;
452 	if (gso) {
453 		NAPI_GRO_CB(head)->last = head;
454 		skb_shinfo(head)->gso_type = sk->sk_gso_type;
455 	}
456 
457 	/* Make sure the outbound skb has enough header room reserved. */
458 	skb_reserve(head, packet->overhead + MAX_HEADER);
459 
460 	/* Set the owning socket so that we know where to get the
461 	 * destination IP address.
462 	 */
463 	sctp_packet_set_owner_w(head, sk);
464 
465 	if (!sctp_transport_dst_check(tp)) {
466 		sctp_transport_route(tp, NULL, sctp_sk(sk));
467 		if (asoc && (asoc->param_flags & SPP_PMTUD_ENABLE)) {
468 			sctp_assoc_sync_pmtu(sk, asoc);
469 		}
470 	}
471 	dst = dst_clone(tp->dst);
472 	if (!dst)
473 		goto no_route;
474 	skb_dst_set(head, dst);
475 
476 	/* Build the SCTP header.  */
477 	sh = (struct sctphdr *)skb_push(head, sizeof(struct sctphdr));
478 	skb_reset_transport_header(head);
479 	sh->source = htons(packet->source_port);
480 	sh->dest   = htons(packet->destination_port);
481 
482 	/* From 6.8 Adler-32 Checksum Calculation:
483 	 * After the packet is constructed (containing the SCTP common
484 	 * header and one or more control or DATA chunks), the
485 	 * transmitter shall:
486 	 *
487 	 * 1) Fill in the proper Verification Tag in the SCTP common
488 	 *    header and initialize the checksum field to 0's.
489 	 */
490 	sh->vtag     = htonl(packet->vtag);
491 	sh->checksum = 0;
492 
493 	pr_debug("***sctp_transmit_packet***\n");
494 
495 	do {
496 		/* Set up convenience variables... */
497 		chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
498 		pktcount++;
499 
500 		/* Calculate packet size, so it fits in PMTU. Leave
501 		 * other chunks for the next packets.
502 		 */
503 		if (gso) {
504 			pkt_size = packet->overhead;
505 			list_for_each_entry(chunk, &packet->chunk_list, list) {
506 				int padded = WORD_ROUND(chunk->skb->len);
507 
508 				if (pkt_size + padded > tp->pathmtu)
509 					break;
510 				pkt_size += padded;
511 			}
512 
513 			/* Allocate a new skb. */
514 			nskb = alloc_skb(pkt_size + MAX_HEADER, gfp);
515 			if (!nskb)
516 				goto nomem;
517 
518 			/* Make sure the outbound skb has enough header
519 			 * room reserved.
520 			 */
521 			skb_reserve(nskb, packet->overhead + MAX_HEADER);
522 		} else {
523 			nskb = head;
524 		}
525 
526 		/**
527 		 * 3.2  Chunk Field Descriptions
528 		 *
529 		 * The total length of a chunk (including Type, Length and
530 		 * Value fields) MUST be a multiple of 4 bytes.  If the length
531 		 * of the chunk is not a multiple of 4 bytes, the sender MUST
532 		 * pad the chunk with all zero bytes and this padding is not
533 		 * included in the chunk length field.  The sender should
534 		 * never pad with more than 3 bytes.
535 		 *
536 		 * [This whole comment explains WORD_ROUND() below.]
537 		 */
538 
539 		pkt_size -= packet->overhead;
540 		list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
541 			list_del_init(&chunk->list);
542 			if (sctp_chunk_is_data(chunk)) {
543 				/* 6.3.1 C4) When data is in flight and when allowed
544 				 * by rule C5, a new RTT measurement MUST be made each
545 				 * round trip.  Furthermore, new RTT measurements
546 				 * SHOULD be made no more than once per round-trip
547 				 * for a given destination transport address.
548 				 */
549 
550 				if (!chunk->resent && !tp->rto_pending) {
551 					chunk->rtt_in_progress = 1;
552 					tp->rto_pending = 1;
553 				}
554 
555 				has_data = 1;
556 			}
557 
558 			padding = WORD_ROUND(chunk->skb->len) - chunk->skb->len;
559 			if (padding)
560 				memset(skb_put(chunk->skb, padding), 0, padding);
561 
562 			/* if this is the auth chunk that we are adding,
563 			 * store pointer where it will be added and put
564 			 * the auth into the packet.
565 			 */
566 			if (chunk == packet->auth)
567 				auth = skb_tail_pointer(nskb);
568 
569 			memcpy(skb_put(nskb, chunk->skb->len),
570 			       chunk->skb->data, chunk->skb->len);
571 
572 			pr_debug("*** Chunk:%p[%s] %s 0x%x, length:%d, chunk->skb->len:%d, rtt_in_progress:%d\n",
573 				 chunk,
574 				 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)),
575 				 chunk->has_tsn ? "TSN" : "No TSN",
576 				 chunk->has_tsn ? ntohl(chunk->subh.data_hdr->tsn) : 0,
577 				 ntohs(chunk->chunk_hdr->length), chunk->skb->len,
578 				 chunk->rtt_in_progress);
579 
580 			/* If this is a control chunk, this is our last
581 			 * reference. Free data chunks after they've been
582 			 * acknowledged or have failed.
583 			 * Re-queue auth chunks if needed.
584 			 */
585 			pkt_size -= WORD_ROUND(chunk->skb->len);
586 
587 			if (!sctp_chunk_is_data(chunk) && chunk != packet->auth)
588 				sctp_chunk_free(chunk);
589 
590 			if (!pkt_size)
591 				break;
592 		}
593 
594 		/* SCTP-AUTH, Section 6.2
595 		 *    The sender MUST calculate the MAC as described in RFC2104 [2]
596 		 *    using the hash function H as described by the MAC Identifier and
597 		 *    the shared association key K based on the endpoint pair shared key
598 		 *    described by the shared key identifier.  The 'data' used for the
599 		 *    computation of the AUTH-chunk is given by the AUTH chunk with its
600 		 *    HMAC field set to zero (as shown in Figure 6) followed by all
601 		 *    chunks that are placed after the AUTH chunk in the SCTP packet.
602 		 */
603 		if (auth)
604 			sctp_auth_calculate_hmac(asoc, nskb,
605 						 (struct sctp_auth_chunk *)auth,
606 						 gfp);
607 
608 		if (packet->auth) {
609 			if (!list_empty(&packet->chunk_list)) {
610 				/* We will generate more packets, so re-queue
611 				 * auth chunk.
612 				 */
613 				list_add(&packet->auth->list,
614 					 &packet->chunk_list);
615 			} else {
616 				sctp_chunk_free(packet->auth);
617 				packet->auth = NULL;
618 			}
619 		}
620 
621 		if (!gso)
622 			break;
623 
624 		if (skb_gro_receive(&head, nskb))
625 			goto nomem;
626 		nskb = NULL;
627 		if (WARN_ON_ONCE(skb_shinfo(head)->gso_segs >=
628 				 sk->sk_gso_max_segs))
629 			goto nomem;
630 	} while (!list_empty(&packet->chunk_list));
631 
632 	/* 2) Calculate the Adler-32 checksum of the whole packet,
633 	 *    including the SCTP common header and all the
634 	 *    chunks.
635 	 *
636 	 * Note: Adler-32 is no longer applicable, as has been replaced
637 	 * by CRC32-C as described in <draft-ietf-tsvwg-sctpcsum-02.txt>.
638 	 *
639 	 * If it's a GSO packet, it's postponed to sctp_skb_segment.
640 	 */
641 	if (!sctp_checksum_disable || gso) {
642 		if (!gso && (!(dst->dev->features & NETIF_F_SCTP_CRC) ||
643 			     dst_xfrm(dst) || packet->ipfragok)) {
644 			sh->checksum = sctp_compute_cksum(head, 0);
645 		} else {
646 			/* no need to seed pseudo checksum for SCTP */
647 			head->ip_summed = CHECKSUM_PARTIAL;
648 			head->csum_start = skb_transport_header(head) - head->head;
649 			head->csum_offset = offsetof(struct sctphdr, checksum);
650 		}
651 	}
652 
653 	/* IP layer ECN support
654 	 * From RFC 2481
655 	 *  "The ECN-Capable Transport (ECT) bit would be set by the
656 	 *   data sender to indicate that the end-points of the
657 	 *   transport protocol are ECN-capable."
658 	 *
659 	 * Now setting the ECT bit all the time, as it should not cause
660 	 * any problems protocol-wise even if our peer ignores it.
661 	 *
662 	 * Note: The works for IPv6 layer checks this bit too later
663 	 * in transmission.  See IP6_ECN_flow_xmit().
664 	 */
665 	tp->af_specific->ecn_capable(sk);
666 
667 	/* Set up the IP options.  */
668 	/* BUG: not implemented
669 	 * For v4 this all lives somewhere in sk->sk_opt...
670 	 */
671 
672 	/* Dump that on IP!  */
673 	if (asoc) {
674 		asoc->stats.opackets += pktcount;
675 		if (asoc->peer.last_sent_to != tp)
676 			/* Considering the multiple CPU scenario, this is a
677 			 * "correcter" place for last_sent_to.  --xguo
678 			 */
679 			asoc->peer.last_sent_to = tp;
680 	}
681 
682 	if (has_data) {
683 		struct timer_list *timer;
684 		unsigned long timeout;
685 
686 		/* Restart the AUTOCLOSE timer when sending data. */
687 		if (sctp_state(asoc, ESTABLISHED) &&
688 		    asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE]) {
689 			timer = &asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
690 			timeout = asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
691 
692 			if (!mod_timer(timer, jiffies + timeout))
693 				sctp_association_hold(asoc);
694 		}
695 	}
696 
697 	pr_debug("***sctp_transmit_packet*** skb->len:%d\n", head->len);
698 
699 	if (gso) {
700 		/* Cleanup our debris for IP stacks */
701 		memset(head->cb, 0, max(sizeof(struct inet_skb_parm),
702 					sizeof(struct inet6_skb_parm)));
703 
704 		skb_shinfo(head)->gso_segs = pktcount;
705 		skb_shinfo(head)->gso_size = GSO_BY_FRAGS;
706 
707 		/* We have to refresh this in case we are xmiting to
708 		 * more than one transport at a time
709 		 */
710 		rcu_read_lock();
711 		if (__sk_dst_get(sk) != tp->dst) {
712 			dst_hold(tp->dst);
713 			sk_setup_caps(sk, tp->dst);
714 		}
715 		rcu_read_unlock();
716 	}
717 	head->ignore_df = packet->ipfragok;
718 	tp->af_specific->sctp_xmit(head, tp);
719 
720 out:
721 	sctp_packet_reset(packet);
722 	return err;
723 no_route:
724 	kfree_skb(head);
725 	if (nskb != head)
726 		kfree_skb(nskb);
727 
728 	if (asoc)
729 		IP_INC_STATS(sock_net(asoc->base.sk), IPSTATS_MIB_OUTNOROUTES);
730 
731 	/* FIXME: Returning the 'err' will effect all the associations
732 	 * associated with a socket, although only one of the paths of the
733 	 * association is unreachable.
734 	 * The real failure of a transport or association can be passed on
735 	 * to the user via notifications. So setting this error may not be
736 	 * required.
737 	 */
738 	 /* err = -EHOSTUNREACH; */
739 err:
740 	/* Control chunks are unreliable so just drop them.  DATA chunks
741 	 * will get resent or dropped later.
742 	 */
743 
744 	list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
745 		list_del_init(&chunk->list);
746 		if (!sctp_chunk_is_data(chunk))
747 			sctp_chunk_free(chunk);
748 	}
749 	goto out;
750 nomem:
751 	if (packet->auth && list_empty(&packet->auth->list))
752 		sctp_chunk_free(packet->auth);
753 	err = -ENOMEM;
754 	goto err;
755 }
756 
757 /********************************************************************
758  * 2nd Level Abstractions
759  ********************************************************************/
760 
761 /* This private function check to see if a chunk can be added */
762 static sctp_xmit_t sctp_packet_can_append_data(struct sctp_packet *packet,
763 					   struct sctp_chunk *chunk)
764 {
765 	size_t datasize, rwnd, inflight, flight_size;
766 	struct sctp_transport *transport = packet->transport;
767 	struct sctp_association *asoc = transport->asoc;
768 	struct sctp_outq *q = &asoc->outqueue;
769 
770 	/* RFC 2960 6.1  Transmission of DATA Chunks
771 	 *
772 	 * A) At any given time, the data sender MUST NOT transmit new data to
773 	 * any destination transport address if its peer's rwnd indicates
774 	 * that the peer has no buffer space (i.e. rwnd is 0, see Section
775 	 * 6.2.1).  However, regardless of the value of rwnd (including if it
776 	 * is 0), the data sender can always have one DATA chunk in flight to
777 	 * the receiver if allowed by cwnd (see rule B below).  This rule
778 	 * allows the sender to probe for a change in rwnd that the sender
779 	 * missed due to the SACK having been lost in transit from the data
780 	 * receiver to the data sender.
781 	 */
782 
783 	rwnd = asoc->peer.rwnd;
784 	inflight = q->outstanding_bytes;
785 	flight_size = transport->flight_size;
786 
787 	datasize = sctp_data_size(chunk);
788 
789 	if (datasize > rwnd && inflight > 0)
790 		/* We have (at least) one data chunk in flight,
791 		 * so we can't fall back to rule 6.1 B).
792 		 */
793 		return SCTP_XMIT_RWND_FULL;
794 
795 	/* RFC 2960 6.1  Transmission of DATA Chunks
796 	 *
797 	 * B) At any given time, the sender MUST NOT transmit new data
798 	 * to a given transport address if it has cwnd or more bytes
799 	 * of data outstanding to that transport address.
800 	 */
801 	/* RFC 7.2.4 & the Implementers Guide 2.8.
802 	 *
803 	 * 3) ...
804 	 *    When a Fast Retransmit is being performed the sender SHOULD
805 	 *    ignore the value of cwnd and SHOULD NOT delay retransmission.
806 	 */
807 	if (chunk->fast_retransmit != SCTP_NEED_FRTX &&
808 	    flight_size >= transport->cwnd)
809 		return SCTP_XMIT_RWND_FULL;
810 
811 	/* Nagle's algorithm to solve small-packet problem:
812 	 * Inhibit the sending of new chunks when new outgoing data arrives
813 	 * if any previously transmitted data on the connection remains
814 	 * unacknowledged.
815 	 */
816 
817 	if (sctp_sk(asoc->base.sk)->nodelay)
818 		/* Nagle disabled */
819 		return SCTP_XMIT_OK;
820 
821 	if (!sctp_packet_empty(packet))
822 		/* Append to packet */
823 		return SCTP_XMIT_OK;
824 
825 	if (inflight == 0)
826 		/* Nothing unacked */
827 		return SCTP_XMIT_OK;
828 
829 	if (!sctp_state(asoc, ESTABLISHED))
830 		return SCTP_XMIT_OK;
831 
832 	/* Check whether this chunk and all the rest of pending data will fit
833 	 * or delay in hopes of bundling a full sized packet.
834 	 */
835 	if (chunk->skb->len + q->out_qlen >
836 		transport->pathmtu - packet->overhead - sizeof(sctp_data_chunk_t) - 4)
837 		/* Enough data queued to fill a packet */
838 		return SCTP_XMIT_OK;
839 
840 	/* Don't delay large message writes that may have been fragmented */
841 	if (!chunk->msg->can_delay)
842 		return SCTP_XMIT_OK;
843 
844 	/* Defer until all data acked or packet full */
845 	return SCTP_XMIT_DELAY;
846 }
847 
848 /* This private function does management things when adding DATA chunk */
849 static void sctp_packet_append_data(struct sctp_packet *packet,
850 				struct sctp_chunk *chunk)
851 {
852 	struct sctp_transport *transport = packet->transport;
853 	size_t datasize = sctp_data_size(chunk);
854 	struct sctp_association *asoc = transport->asoc;
855 	u32 rwnd = asoc->peer.rwnd;
856 
857 	/* Keep track of how many bytes are in flight over this transport. */
858 	transport->flight_size += datasize;
859 
860 	/* Keep track of how many bytes are in flight to the receiver. */
861 	asoc->outqueue.outstanding_bytes += datasize;
862 
863 	/* Update our view of the receiver's rwnd. */
864 	if (datasize < rwnd)
865 		rwnd -= datasize;
866 	else
867 		rwnd = 0;
868 
869 	asoc->peer.rwnd = rwnd;
870 	/* Has been accepted for transmission. */
871 	if (!asoc->peer.prsctp_capable)
872 		chunk->msg->can_abandon = 0;
873 	sctp_chunk_assign_tsn(chunk);
874 	sctp_chunk_assign_ssn(chunk);
875 }
876 
877 static sctp_xmit_t sctp_packet_will_fit(struct sctp_packet *packet,
878 					struct sctp_chunk *chunk,
879 					u16 chunk_len)
880 {
881 	size_t psize, pmtu;
882 	sctp_xmit_t retval = SCTP_XMIT_OK;
883 
884 	psize = packet->size;
885 	if (packet->transport->asoc)
886 		pmtu = packet->transport->asoc->pathmtu;
887 	else
888 		pmtu = packet->transport->pathmtu;
889 
890 	/* Decide if we need to fragment or resubmit later. */
891 	if (psize + chunk_len > pmtu) {
892 		/* It's OK to fragment at IP level if any one of the following
893 		 * is true:
894 		 *	1. The packet is empty (meaning this chunk is greater
895 		 *	   the MTU)
896 		 *	2. The packet doesn't have any data in it yet and data
897 		 *	   requires authentication.
898 		 */
899 		if (sctp_packet_empty(packet) ||
900 		    (!packet->has_data && chunk->auth)) {
901 			/* We no longer do re-fragmentation.
902 			 * Just fragment at the IP layer, if we
903 			 * actually hit this condition
904 			 */
905 			packet->ipfragok = 1;
906 			goto out;
907 		}
908 
909 		/* It is also okay to fragment if the chunk we are
910 		 * adding is a control chunk, but only if current packet
911 		 * is not a GSO one otherwise it causes fragmentation of
912 		 * a large frame. So in this case we allow the
913 		 * fragmentation by forcing it to be in a new packet.
914 		 */
915 		if (!sctp_chunk_is_data(chunk) && packet->has_data)
916 			retval = SCTP_XMIT_PMTU_FULL;
917 
918 		if (psize + chunk_len > packet->max_size)
919 			/* Hit GSO/PMTU limit, gotta flush */
920 			retval = SCTP_XMIT_PMTU_FULL;
921 
922 		if (!packet->transport->burst_limited &&
923 		    psize + chunk_len > (packet->transport->cwnd >> 1))
924 			/* Do not allow a single GSO packet to use more
925 			 * than half of cwnd.
926 			 */
927 			retval = SCTP_XMIT_PMTU_FULL;
928 
929 		if (packet->transport->burst_limited &&
930 		    psize + chunk_len > (packet->transport->burst_limited >> 1))
931 			/* Do not allow a single GSO packet to use more
932 			 * than half of original cwnd.
933 			 */
934 			retval = SCTP_XMIT_PMTU_FULL;
935 		/* Otherwise it will fit in the GSO packet */
936 	}
937 
938 out:
939 	return retval;
940 }
941