xref: /openbmc/linux/net/sctp/output.c (revision 160b8e75)
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 enum sctp_xmit __sctp_packet_append_chunk(struct sctp_packet *packet,
61 						 struct sctp_chunk *chunk);
62 static enum sctp_xmit 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 enum sctp_xmit 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 void sctp_packet_config(struct sctp_packet *packet, __u32 vtag,
85 			int ecn_capable)
86 {
87 	struct sctp_transport *tp = packet->transport;
88 	struct sctp_association *asoc = tp->asoc;
89 	struct sock *sk;
90 
91 	pr_debug("%s: packet:%p vtag:0x%x\n", __func__, packet, vtag);
92 	packet->vtag = vtag;
93 
94 	/* do the following jobs only once for a flush schedule */
95 	if (!sctp_packet_empty(packet))
96 		return;
97 
98 	/* set packet max_size with pathmtu */
99 	packet->max_size = tp->pathmtu;
100 	if (!asoc)
101 		return;
102 
103 	/* update dst or transport pathmtu if in need */
104 	sk = asoc->base.sk;
105 	if (!sctp_transport_dst_check(tp)) {
106 		sctp_transport_route(tp, NULL, sctp_sk(sk));
107 		if (asoc->param_flags & SPP_PMTUD_ENABLE)
108 			sctp_assoc_sync_pmtu(asoc);
109 	} else if (!sctp_transport_pmtu_check(tp)) {
110 		if (asoc->param_flags & SPP_PMTUD_ENABLE)
111 			sctp_assoc_sync_pmtu(asoc);
112 	}
113 
114 	/* If there a is a prepend chunk stick it on the list before
115 	 * any other chunks get appended.
116 	 */
117 	if (ecn_capable) {
118 		struct sctp_chunk *chunk = sctp_get_ecne_prepend(asoc);
119 
120 		if (chunk)
121 			sctp_packet_append_chunk(packet, chunk);
122 	}
123 
124 	if (!tp->dst)
125 		return;
126 
127 	/* set packet max_size with gso_max_size if gso is enabled*/
128 	rcu_read_lock();
129 	if (__sk_dst_get(sk) != tp->dst) {
130 		dst_hold(tp->dst);
131 		sk_setup_caps(sk, tp->dst);
132 	}
133 	packet->max_size = sk_can_gso(sk) ? tp->dst->dev->gso_max_size
134 					  : asoc->pathmtu;
135 	rcu_read_unlock();
136 }
137 
138 /* Initialize the packet structure. */
139 void sctp_packet_init(struct sctp_packet *packet,
140 		      struct sctp_transport *transport,
141 		      __u16 sport, __u16 dport)
142 {
143 	struct sctp_association *asoc = transport->asoc;
144 	size_t overhead;
145 
146 	pr_debug("%s: packet:%p transport:%p\n", __func__, packet, transport);
147 
148 	packet->transport = transport;
149 	packet->source_port = sport;
150 	packet->destination_port = dport;
151 	INIT_LIST_HEAD(&packet->chunk_list);
152 	if (asoc) {
153 		struct sctp_sock *sp = sctp_sk(asoc->base.sk);
154 		overhead = sp->pf->af->net_header_len;
155 	} else {
156 		overhead = sizeof(struct ipv6hdr);
157 	}
158 	overhead += sizeof(struct sctphdr);
159 	packet->overhead = overhead;
160 	sctp_packet_reset(packet);
161 	packet->vtag = 0;
162 }
163 
164 /* Free a packet.  */
165 void sctp_packet_free(struct sctp_packet *packet)
166 {
167 	struct sctp_chunk *chunk, *tmp;
168 
169 	pr_debug("%s: packet:%p\n", __func__, packet);
170 
171 	list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
172 		list_del_init(&chunk->list);
173 		sctp_chunk_free(chunk);
174 	}
175 }
176 
177 /* This routine tries to append the chunk to the offered packet. If adding
178  * the chunk causes the packet to exceed the path MTU and COOKIE_ECHO chunk
179  * is not present in the packet, it transmits the input packet.
180  * Data can be bundled with a packet containing a COOKIE_ECHO chunk as long
181  * as it can fit in the packet, but any more data that does not fit in this
182  * packet can be sent only after receiving the COOKIE_ACK.
183  */
184 enum sctp_xmit sctp_packet_transmit_chunk(struct sctp_packet *packet,
185 					  struct sctp_chunk *chunk,
186 					  int one_packet, gfp_t gfp)
187 {
188 	enum sctp_xmit retval;
189 
190 	pr_debug("%s: packet:%p size:%zu chunk:%p size:%d\n", __func__,
191 		 packet, packet->size, chunk, chunk->skb ? chunk->skb->len : -1);
192 
193 	switch ((retval = (sctp_packet_append_chunk(packet, chunk)))) {
194 	case SCTP_XMIT_PMTU_FULL:
195 		if (!packet->has_cookie_echo) {
196 			int error = 0;
197 
198 			error = sctp_packet_transmit(packet, gfp);
199 			if (error < 0)
200 				chunk->skb->sk->sk_err = -error;
201 
202 			/* If we have an empty packet, then we can NOT ever
203 			 * return PMTU_FULL.
204 			 */
205 			if (!one_packet)
206 				retval = sctp_packet_append_chunk(packet,
207 								  chunk);
208 		}
209 		break;
210 
211 	case SCTP_XMIT_RWND_FULL:
212 	case SCTP_XMIT_OK:
213 	case SCTP_XMIT_DELAY:
214 		break;
215 	}
216 
217 	return retval;
218 }
219 
220 /* Try to bundle an auth chunk into the packet. */
221 static enum sctp_xmit sctp_packet_bundle_auth(struct sctp_packet *pkt,
222 					      struct sctp_chunk *chunk)
223 {
224 	struct sctp_association *asoc = pkt->transport->asoc;
225 	enum sctp_xmit retval = SCTP_XMIT_OK;
226 	struct sctp_chunk *auth;
227 
228 	/* if we don't have an association, we can't do authentication */
229 	if (!asoc)
230 		return retval;
231 
232 	/* See if this is an auth chunk we are bundling or if
233 	 * auth is already bundled.
234 	 */
235 	if (chunk->chunk_hdr->type == SCTP_CID_AUTH || pkt->has_auth)
236 		return retval;
237 
238 	/* if the peer did not request this chunk to be authenticated,
239 	 * don't do it
240 	 */
241 	if (!chunk->auth)
242 		return retval;
243 
244 	auth = sctp_make_auth(asoc);
245 	if (!auth)
246 		return retval;
247 
248 	retval = __sctp_packet_append_chunk(pkt, auth);
249 
250 	if (retval != SCTP_XMIT_OK)
251 		sctp_chunk_free(auth);
252 
253 	return retval;
254 }
255 
256 /* Try to bundle a SACK with the packet. */
257 static enum sctp_xmit sctp_packet_bundle_sack(struct sctp_packet *pkt,
258 					      struct sctp_chunk *chunk)
259 {
260 	enum sctp_xmit retval = SCTP_XMIT_OK;
261 
262 	/* If sending DATA and haven't aleady bundled a SACK, try to
263 	 * bundle one in to the packet.
264 	 */
265 	if (sctp_chunk_is_data(chunk) && !pkt->has_sack &&
266 	    !pkt->has_cookie_echo) {
267 		struct sctp_association *asoc;
268 		struct timer_list *timer;
269 		asoc = pkt->transport->asoc;
270 		timer = &asoc->timers[SCTP_EVENT_TIMEOUT_SACK];
271 
272 		/* If the SACK timer is running, we have a pending SACK */
273 		if (timer_pending(timer)) {
274 			struct sctp_chunk *sack;
275 
276 			if (pkt->transport->sack_generation !=
277 			    pkt->transport->asoc->peer.sack_generation)
278 				return retval;
279 
280 			asoc->a_rwnd = asoc->rwnd;
281 			sack = sctp_make_sack(asoc);
282 			if (sack) {
283 				retval = __sctp_packet_append_chunk(pkt, sack);
284 				if (retval != SCTP_XMIT_OK) {
285 					sctp_chunk_free(sack);
286 					goto out;
287 				}
288 				asoc->peer.sack_needed = 0;
289 				if (del_timer(timer))
290 					sctp_association_put(asoc);
291 			}
292 		}
293 	}
294 out:
295 	return retval;
296 }
297 
298 
299 /* Append a chunk to the offered packet reporting back any inability to do
300  * so.
301  */
302 static enum sctp_xmit __sctp_packet_append_chunk(struct sctp_packet *packet,
303 						 struct sctp_chunk *chunk)
304 {
305 	__u16 chunk_len = SCTP_PAD4(ntohs(chunk->chunk_hdr->length));
306 	enum sctp_xmit retval = SCTP_XMIT_OK;
307 
308 	/* Check to see if this chunk will fit into the packet */
309 	retval = sctp_packet_will_fit(packet, chunk, chunk_len);
310 	if (retval != SCTP_XMIT_OK)
311 		goto finish;
312 
313 	/* We believe that this chunk is OK to add to the packet */
314 	switch (chunk->chunk_hdr->type) {
315 	case SCTP_CID_DATA:
316 	case SCTP_CID_I_DATA:
317 		/* Account for the data being in the packet */
318 		sctp_packet_append_data(packet, chunk);
319 		/* Disallow SACK bundling after DATA. */
320 		packet->has_sack = 1;
321 		/* Disallow AUTH bundling after DATA */
322 		packet->has_auth = 1;
323 		/* Let it be knows that packet has DATA in it */
324 		packet->has_data = 1;
325 		/* timestamp the chunk for rtx purposes */
326 		chunk->sent_at = jiffies;
327 		/* Mainly used for prsctp RTX policy */
328 		chunk->sent_count++;
329 		break;
330 	case SCTP_CID_COOKIE_ECHO:
331 		packet->has_cookie_echo = 1;
332 		break;
333 
334 	case SCTP_CID_SACK:
335 		packet->has_sack = 1;
336 		if (chunk->asoc)
337 			chunk->asoc->stats.osacks++;
338 		break;
339 
340 	case SCTP_CID_AUTH:
341 		packet->has_auth = 1;
342 		packet->auth = chunk;
343 		break;
344 	}
345 
346 	/* It is OK to send this chunk.  */
347 	list_add_tail(&chunk->list, &packet->chunk_list);
348 	packet->size += chunk_len;
349 	chunk->transport = packet->transport;
350 finish:
351 	return retval;
352 }
353 
354 /* Append a chunk to the offered packet reporting back any inability to do
355  * so.
356  */
357 enum sctp_xmit sctp_packet_append_chunk(struct sctp_packet *packet,
358 					struct sctp_chunk *chunk)
359 {
360 	enum sctp_xmit retval = SCTP_XMIT_OK;
361 
362 	pr_debug("%s: packet:%p chunk:%p\n", __func__, packet, chunk);
363 
364 	/* Data chunks are special.  Before seeing what else we can
365 	 * bundle into this packet, check to see if we are allowed to
366 	 * send this DATA.
367 	 */
368 	if (sctp_chunk_is_data(chunk)) {
369 		retval = sctp_packet_can_append_data(packet, chunk);
370 		if (retval != SCTP_XMIT_OK)
371 			goto finish;
372 	}
373 
374 	/* Try to bundle AUTH chunk */
375 	retval = sctp_packet_bundle_auth(packet, chunk);
376 	if (retval != SCTP_XMIT_OK)
377 		goto finish;
378 
379 	/* Try to bundle SACK chunk */
380 	retval = sctp_packet_bundle_sack(packet, chunk);
381 	if (retval != SCTP_XMIT_OK)
382 		goto finish;
383 
384 	retval = __sctp_packet_append_chunk(packet, chunk);
385 
386 finish:
387 	return retval;
388 }
389 
390 static void sctp_packet_release_owner(struct sk_buff *skb)
391 {
392 	sk_free(skb->sk);
393 }
394 
395 static void sctp_packet_set_owner_w(struct sk_buff *skb, struct sock *sk)
396 {
397 	skb_orphan(skb);
398 	skb->sk = sk;
399 	skb->destructor = sctp_packet_release_owner;
400 
401 	/*
402 	 * The data chunks have already been accounted for in sctp_sendmsg(),
403 	 * therefore only reserve a single byte to keep socket around until
404 	 * the packet has been transmitted.
405 	 */
406 	refcount_inc(&sk->sk_wmem_alloc);
407 }
408 
409 static int sctp_packet_pack(struct sctp_packet *packet,
410 			    struct sk_buff *head, int gso, gfp_t gfp)
411 {
412 	struct sctp_transport *tp = packet->transport;
413 	struct sctp_auth_chunk *auth = NULL;
414 	struct sctp_chunk *chunk, *tmp;
415 	int pkt_count = 0, pkt_size;
416 	struct sock *sk = head->sk;
417 	struct sk_buff *nskb;
418 	int auth_len = 0;
419 
420 	if (gso) {
421 		skb_shinfo(head)->gso_type = sk->sk_gso_type;
422 		NAPI_GRO_CB(head)->last = head;
423 	} else {
424 		nskb = head;
425 		pkt_size = packet->size;
426 		goto merge;
427 	}
428 
429 	do {
430 		/* calculate the pkt_size and alloc nskb */
431 		pkt_size = packet->overhead;
432 		list_for_each_entry_safe(chunk, tmp, &packet->chunk_list,
433 					 list) {
434 			int padded = SCTP_PAD4(chunk->skb->len);
435 
436 			if (chunk == packet->auth)
437 				auth_len = padded;
438 			else if (auth_len + padded + packet->overhead >
439 				 tp->pathmtu)
440 				return 0;
441 			else if (pkt_size + padded > tp->pathmtu)
442 				break;
443 			pkt_size += padded;
444 		}
445 		nskb = alloc_skb(pkt_size + MAX_HEADER, gfp);
446 		if (!nskb)
447 			return 0;
448 		skb_reserve(nskb, packet->overhead + MAX_HEADER);
449 
450 merge:
451 		/* merge chunks into nskb and append nskb into head list */
452 		pkt_size -= packet->overhead;
453 		list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
454 			int padding;
455 
456 			list_del_init(&chunk->list);
457 			if (sctp_chunk_is_data(chunk)) {
458 				if (!sctp_chunk_retransmitted(chunk) &&
459 				    !tp->rto_pending) {
460 					chunk->rtt_in_progress = 1;
461 					tp->rto_pending = 1;
462 				}
463 			}
464 
465 			padding = SCTP_PAD4(chunk->skb->len) - chunk->skb->len;
466 			if (padding)
467 				skb_put_zero(chunk->skb, padding);
468 
469 			if (chunk == packet->auth)
470 				auth = (struct sctp_auth_chunk *)
471 							skb_tail_pointer(nskb);
472 
473 			skb_put_data(nskb, chunk->skb->data, chunk->skb->len);
474 
475 			pr_debug("*** Chunk:%p[%s] %s 0x%x, length:%d, chunk->skb->len:%d, rtt_in_progress:%d\n",
476 				 chunk,
477 				 sctp_cname(SCTP_ST_CHUNK(chunk->chunk_hdr->type)),
478 				 chunk->has_tsn ? "TSN" : "No TSN",
479 				 chunk->has_tsn ? ntohl(chunk->subh.data_hdr->tsn) : 0,
480 				 ntohs(chunk->chunk_hdr->length), chunk->skb->len,
481 				 chunk->rtt_in_progress);
482 
483 			pkt_size -= SCTP_PAD4(chunk->skb->len);
484 
485 			if (!sctp_chunk_is_data(chunk) && chunk != packet->auth)
486 				sctp_chunk_free(chunk);
487 
488 			if (!pkt_size)
489 				break;
490 		}
491 
492 		if (auth) {
493 			sctp_auth_calculate_hmac(tp->asoc, nskb, auth, gfp);
494 			/* free auth if no more chunks, or add it back */
495 			if (list_empty(&packet->chunk_list))
496 				sctp_chunk_free(packet->auth);
497 			else
498 				list_add(&packet->auth->list,
499 					 &packet->chunk_list);
500 		}
501 
502 		if (gso) {
503 			if (skb_gro_receive(&head, nskb)) {
504 				kfree_skb(nskb);
505 				return 0;
506 			}
507 			if (WARN_ON_ONCE(skb_shinfo(head)->gso_segs >=
508 					 sk->sk_gso_max_segs))
509 				return 0;
510 		}
511 
512 		pkt_count++;
513 	} while (!list_empty(&packet->chunk_list));
514 
515 	if (gso) {
516 		memset(head->cb, 0, max(sizeof(struct inet_skb_parm),
517 					sizeof(struct inet6_skb_parm)));
518 		skb_shinfo(head)->gso_segs = pkt_count;
519 		skb_shinfo(head)->gso_size = GSO_BY_FRAGS;
520 		rcu_read_lock();
521 		if (skb_dst(head) != tp->dst) {
522 			dst_hold(tp->dst);
523 			sk_setup_caps(sk, tp->dst);
524 		}
525 		rcu_read_unlock();
526 		goto chksum;
527 	}
528 
529 	if (sctp_checksum_disable)
530 		return 1;
531 
532 	if (!(skb_dst(head)->dev->features & NETIF_F_SCTP_CRC) ||
533 	    dst_xfrm(skb_dst(head)) || packet->ipfragok) {
534 		struct sctphdr *sh =
535 			(struct sctphdr *)skb_transport_header(head);
536 
537 		sh->checksum = sctp_compute_cksum(head, 0);
538 	} else {
539 chksum:
540 		head->ip_summed = CHECKSUM_PARTIAL;
541 		head->csum_not_inet = 1;
542 		head->csum_start = skb_transport_header(head) - head->head;
543 		head->csum_offset = offsetof(struct sctphdr, checksum);
544 	}
545 
546 	return pkt_count;
547 }
548 
549 /* All packets are sent to the network through this function from
550  * sctp_outq_tail().
551  *
552  * The return value is always 0 for now.
553  */
554 int sctp_packet_transmit(struct sctp_packet *packet, gfp_t gfp)
555 {
556 	struct sctp_transport *tp = packet->transport;
557 	struct sctp_association *asoc = tp->asoc;
558 	struct sctp_chunk *chunk, *tmp;
559 	int pkt_count, gso = 0;
560 	struct dst_entry *dst;
561 	struct sk_buff *head;
562 	struct sctphdr *sh;
563 	struct sock *sk;
564 
565 	pr_debug("%s: packet:%p\n", __func__, packet);
566 	if (list_empty(&packet->chunk_list))
567 		return 0;
568 	chunk = list_entry(packet->chunk_list.next, struct sctp_chunk, list);
569 	sk = chunk->skb->sk;
570 
571 	/* check gso */
572 	if (packet->size > tp->pathmtu && !packet->ipfragok) {
573 		if (!sk_can_gso(sk)) {
574 			pr_err_once("Trying to GSO but underlying device doesn't support it.");
575 			goto out;
576 		}
577 		gso = 1;
578 	}
579 
580 	/* alloc head skb */
581 	head = alloc_skb((gso ? packet->overhead : packet->size) +
582 			 MAX_HEADER, gfp);
583 	if (!head)
584 		goto out;
585 	skb_reserve(head, packet->overhead + MAX_HEADER);
586 	sctp_packet_set_owner_w(head, sk);
587 
588 	/* set sctp header */
589 	sh = skb_push(head, sizeof(struct sctphdr));
590 	skb_reset_transport_header(head);
591 	sh->source = htons(packet->source_port);
592 	sh->dest = htons(packet->destination_port);
593 	sh->vtag = htonl(packet->vtag);
594 	sh->checksum = 0;
595 
596 	/* drop packet if no dst */
597 	dst = dst_clone(tp->dst);
598 	if (!dst) {
599 		IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
600 		kfree_skb(head);
601 		goto out;
602 	}
603 	skb_dst_set(head, dst);
604 
605 	/* pack up chunks */
606 	pkt_count = sctp_packet_pack(packet, head, gso, gfp);
607 	if (!pkt_count) {
608 		kfree_skb(head);
609 		goto out;
610 	}
611 	pr_debug("***sctp_transmit_packet*** skb->len:%d\n", head->len);
612 
613 	/* start autoclose timer */
614 	if (packet->has_data && sctp_state(asoc, ESTABLISHED) &&
615 	    asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE]) {
616 		struct timer_list *timer =
617 			&asoc->timers[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
618 		unsigned long timeout =
619 			asoc->timeouts[SCTP_EVENT_TIMEOUT_AUTOCLOSE];
620 
621 		if (!mod_timer(timer, jiffies + timeout))
622 			sctp_association_hold(asoc);
623 	}
624 
625 	/* sctp xmit */
626 	tp->af_specific->ecn_capable(sk);
627 	if (asoc) {
628 		asoc->stats.opackets += pkt_count;
629 		if (asoc->peer.last_sent_to != tp)
630 			asoc->peer.last_sent_to = tp;
631 	}
632 	head->ignore_df = packet->ipfragok;
633 	if (tp->dst_pending_confirm)
634 		skb_set_dst_pending_confirm(head, 1);
635 	/* neighbour should be confirmed on successful transmission or
636 	 * positive error
637 	 */
638 	if (tp->af_specific->sctp_xmit(head, tp) >= 0 &&
639 	    tp->dst_pending_confirm)
640 		tp->dst_pending_confirm = 0;
641 
642 out:
643 	list_for_each_entry_safe(chunk, tmp, &packet->chunk_list, list) {
644 		list_del_init(&chunk->list);
645 		if (!sctp_chunk_is_data(chunk))
646 			sctp_chunk_free(chunk);
647 	}
648 	sctp_packet_reset(packet);
649 	return 0;
650 }
651 
652 /********************************************************************
653  * 2nd Level Abstractions
654  ********************************************************************/
655 
656 /* This private function check to see if a chunk can be added */
657 static enum sctp_xmit sctp_packet_can_append_data(struct sctp_packet *packet,
658 						  struct sctp_chunk *chunk)
659 {
660 	size_t datasize, rwnd, inflight, flight_size;
661 	struct sctp_transport *transport = packet->transport;
662 	struct sctp_association *asoc = transport->asoc;
663 	struct sctp_outq *q = &asoc->outqueue;
664 
665 	/* RFC 2960 6.1  Transmission of DATA Chunks
666 	 *
667 	 * A) At any given time, the data sender MUST NOT transmit new data to
668 	 * any destination transport address if its peer's rwnd indicates
669 	 * that the peer has no buffer space (i.e. rwnd is 0, see Section
670 	 * 6.2.1).  However, regardless of the value of rwnd (including if it
671 	 * is 0), the data sender can always have one DATA chunk in flight to
672 	 * the receiver if allowed by cwnd (see rule B below).  This rule
673 	 * allows the sender to probe for a change in rwnd that the sender
674 	 * missed due to the SACK having been lost in transit from the data
675 	 * receiver to the data sender.
676 	 */
677 
678 	rwnd = asoc->peer.rwnd;
679 	inflight = q->outstanding_bytes;
680 	flight_size = transport->flight_size;
681 
682 	datasize = sctp_data_size(chunk);
683 
684 	if (datasize > rwnd && inflight > 0)
685 		/* We have (at least) one data chunk in flight,
686 		 * so we can't fall back to rule 6.1 B).
687 		 */
688 		return SCTP_XMIT_RWND_FULL;
689 
690 	/* RFC 2960 6.1  Transmission of DATA Chunks
691 	 *
692 	 * B) At any given time, the sender MUST NOT transmit new data
693 	 * to a given transport address if it has cwnd or more bytes
694 	 * of data outstanding to that transport address.
695 	 */
696 	/* RFC 7.2.4 & the Implementers Guide 2.8.
697 	 *
698 	 * 3) ...
699 	 *    When a Fast Retransmit is being performed the sender SHOULD
700 	 *    ignore the value of cwnd and SHOULD NOT delay retransmission.
701 	 */
702 	if (chunk->fast_retransmit != SCTP_NEED_FRTX &&
703 	    flight_size >= transport->cwnd)
704 		return SCTP_XMIT_RWND_FULL;
705 
706 	/* Nagle's algorithm to solve small-packet problem:
707 	 * Inhibit the sending of new chunks when new outgoing data arrives
708 	 * if any previously transmitted data on the connection remains
709 	 * unacknowledged.
710 	 */
711 
712 	if ((sctp_sk(asoc->base.sk)->nodelay || inflight == 0) &&
713 	    !asoc->force_delay)
714 		/* Nothing unacked */
715 		return SCTP_XMIT_OK;
716 
717 	if (!sctp_packet_empty(packet))
718 		/* Append to packet */
719 		return SCTP_XMIT_OK;
720 
721 	if (!sctp_state(asoc, ESTABLISHED))
722 		return SCTP_XMIT_OK;
723 
724 	/* Check whether this chunk and all the rest of pending data will fit
725 	 * or delay in hopes of bundling a full sized packet.
726 	 */
727 	if (chunk->skb->len + q->out_qlen > transport->pathmtu -
728 	    packet->overhead - sctp_datachk_len(&chunk->asoc->stream) - 4)
729 		/* Enough data queued to fill a packet */
730 		return SCTP_XMIT_OK;
731 
732 	/* Don't delay large message writes that may have been fragmented */
733 	if (!chunk->msg->can_delay)
734 		return SCTP_XMIT_OK;
735 
736 	/* Defer until all data acked or packet full */
737 	return SCTP_XMIT_DELAY;
738 }
739 
740 /* This private function does management things when adding DATA chunk */
741 static void sctp_packet_append_data(struct sctp_packet *packet,
742 				struct sctp_chunk *chunk)
743 {
744 	struct sctp_transport *transport = packet->transport;
745 	size_t datasize = sctp_data_size(chunk);
746 	struct sctp_association *asoc = transport->asoc;
747 	u32 rwnd = asoc->peer.rwnd;
748 
749 	/* Keep track of how many bytes are in flight over this transport. */
750 	transport->flight_size += datasize;
751 
752 	/* Keep track of how many bytes are in flight to the receiver. */
753 	asoc->outqueue.outstanding_bytes += datasize;
754 
755 	/* Update our view of the receiver's rwnd. */
756 	if (datasize < rwnd)
757 		rwnd -= datasize;
758 	else
759 		rwnd = 0;
760 
761 	asoc->peer.rwnd = rwnd;
762 	sctp_chunk_assign_tsn(chunk);
763 	asoc->stream.si->assign_number(chunk);
764 }
765 
766 static enum sctp_xmit sctp_packet_will_fit(struct sctp_packet *packet,
767 					   struct sctp_chunk *chunk,
768 					   u16 chunk_len)
769 {
770 	enum sctp_xmit retval = SCTP_XMIT_OK;
771 	size_t psize, pmtu, maxsize;
772 
773 	psize = packet->size;
774 	if (packet->transport->asoc)
775 		pmtu = packet->transport->asoc->pathmtu;
776 	else
777 		pmtu = packet->transport->pathmtu;
778 
779 	/* Decide if we need to fragment or resubmit later. */
780 	if (psize + chunk_len > pmtu) {
781 		/* It's OK to fragment at IP level if any one of the following
782 		 * is true:
783 		 *	1. The packet is empty (meaning this chunk is greater
784 		 *	   the MTU)
785 		 *	2. The packet doesn't have any data in it yet and data
786 		 *	   requires authentication.
787 		 */
788 		if (sctp_packet_empty(packet) ||
789 		    (!packet->has_data && chunk->auth)) {
790 			/* We no longer do re-fragmentation.
791 			 * Just fragment at the IP layer, if we
792 			 * actually hit this condition
793 			 */
794 			packet->ipfragok = 1;
795 			goto out;
796 		}
797 
798 		/* Similarly, if this chunk was built before a PMTU
799 		 * reduction, we have to fragment it at IP level now. So
800 		 * if the packet already contains something, we need to
801 		 * flush.
802 		 */
803 		maxsize = pmtu - packet->overhead;
804 		if (packet->auth)
805 			maxsize -= SCTP_PAD4(packet->auth->skb->len);
806 		if (chunk_len > maxsize)
807 			retval = SCTP_XMIT_PMTU_FULL;
808 
809 		/* It is also okay to fragment if the chunk we are
810 		 * adding is a control chunk, but only if current packet
811 		 * is not a GSO one otherwise it causes fragmentation of
812 		 * a large frame. So in this case we allow the
813 		 * fragmentation by forcing it to be in a new packet.
814 		 */
815 		if (!sctp_chunk_is_data(chunk) && packet->has_data)
816 			retval = SCTP_XMIT_PMTU_FULL;
817 
818 		if (psize + chunk_len > packet->max_size)
819 			/* Hit GSO/PMTU limit, gotta flush */
820 			retval = SCTP_XMIT_PMTU_FULL;
821 
822 		if (!packet->transport->burst_limited &&
823 		    psize + chunk_len > (packet->transport->cwnd >> 1))
824 			/* Do not allow a single GSO packet to use more
825 			 * than half of cwnd.
826 			 */
827 			retval = SCTP_XMIT_PMTU_FULL;
828 
829 		if (packet->transport->burst_limited &&
830 		    psize + chunk_len > (packet->transport->burst_limited >> 1))
831 			/* Do not allow a single GSO packet to use more
832 			 * than half of original cwnd.
833 			 */
834 			retval = SCTP_XMIT_PMTU_FULL;
835 		/* Otherwise it will fit in the GSO packet */
836 	}
837 
838 out:
839 	return retval;
840 }
841