Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* SCTP kernel implementation
3 * Copyright (c) 1999-2000 Cisco, Inc.
4 * Copyright (c) 1999-2001 Motorola, Inc.
5 * Copyright (c) 2001-2003 International Business Machines, Corp.
6 * Copyright (c) 2001 Intel Corp.
7 * Copyright (c) 2001 Nokia, Inc.
8 * Copyright (c) 2001 La Monte H.P. Yarroll
9 *
10 * This file is part of the SCTP kernel implementation
11 *
12 * These functions handle all input from the IP layer into SCTP.
13 *
14 * Please send any bug reports or fixes you make to the
15 * email address(es):
16 * lksctp developers <linux-sctp@vger.kernel.org>
17 *
18 * Written or modified by:
19 * La Monte H.P. Yarroll <piggy@acm.org>
20 * Karl Knutson <karl@athena.chicago.il.us>
21 * Xingang Guo <xingang.guo@intel.com>
22 * Jon Grimm <jgrimm@us.ibm.com>
23 * Hui Huang <hui.huang@nokia.com>
24 * Daisy Chang <daisyc@us.ibm.com>
25 * Sridhar Samudrala <sri@us.ibm.com>
26 * Ardelle Fan <ardelle.fan@intel.com>
27 */
28
29#include <linux/types.h>
30#include <linux/list.h> /* For struct list_head */
31#include <linux/socket.h>
32#include <linux/ip.h>
33#include <linux/time.h> /* For struct timeval */
34#include <linux/slab.h>
35#include <net/ip.h>
36#include <net/icmp.h>
37#include <net/snmp.h>
38#include <net/sock.h>
39#include <net/xfrm.h>
40#include <net/sctp/sctp.h>
41#include <net/sctp/sm.h>
42#include <net/sctp/checksum.h>
43#include <net/net_namespace.h>
44#include <linux/rhashtable.h>
45#include <net/sock_reuseport.h>
46
47/* Forward declarations for internal helpers. */
48static int sctp_rcv_ootb(struct sk_buff *);
49static struct sctp_association *__sctp_rcv_lookup(struct net *net,
50 struct sk_buff *skb,
51 const union sctp_addr *paddr,
52 const union sctp_addr *laddr,
53 struct sctp_transport **transportp);
54static struct sctp_endpoint *__sctp_rcv_lookup_endpoint(
55 struct net *net, struct sk_buff *skb,
56 const union sctp_addr *laddr,
57 const union sctp_addr *daddr);
58static struct sctp_association *__sctp_lookup_association(
59 struct net *net,
60 const union sctp_addr *local,
61 const union sctp_addr *peer,
62 struct sctp_transport **pt);
63
64static int sctp_add_backlog(struct sock *sk, struct sk_buff *skb);
65
66
67/* Calculate the SCTP checksum of an SCTP packet. */
68static inline int sctp_rcv_checksum(struct net *net, struct sk_buff *skb)
69{
70 struct sctphdr *sh = sctp_hdr(skb);
71 __le32 cmp = sh->checksum;
72 __le32 val = sctp_compute_cksum(skb, 0);
73
74 if (val != cmp) {
75 /* CRC failure, dump it. */
76 __SCTP_INC_STATS(net, SCTP_MIB_CHECKSUMERRORS);
77 return -1;
78 }
79 return 0;
80}
81
82/*
83 * This is the routine which IP calls when receiving an SCTP packet.
84 */
85int sctp_rcv(struct sk_buff *skb)
86{
87 struct sock *sk;
88 struct sctp_association *asoc;
89 struct sctp_endpoint *ep = NULL;
90 struct sctp_ep_common *rcvr;
91 struct sctp_transport *transport = NULL;
92 struct sctp_chunk *chunk;
93 union sctp_addr src;
94 union sctp_addr dest;
95 int family;
96 struct sctp_af *af;
97 struct net *net = dev_net(skb->dev);
98 bool is_gso = skb_is_gso(skb) && skb_is_gso_sctp(skb);
99
100 if (skb->pkt_type != PACKET_HOST)
101 goto discard_it;
102
103 __SCTP_INC_STATS(net, SCTP_MIB_INSCTPPACKS);
104
105 /* If packet is too small to contain a single chunk, let's not
106 * waste time on it anymore.
107 */
108 if (skb->len < sizeof(struct sctphdr) + sizeof(struct sctp_chunkhdr) +
109 skb_transport_offset(skb))
110 goto discard_it;
111
112 /* If the packet is fragmented and we need to do crc checking,
113 * it's better to just linearize it otherwise crc computing
114 * takes longer.
115 */
116 if ((!is_gso && skb_linearize(skb)) ||
117 !pskb_may_pull(skb, sizeof(struct sctphdr)))
118 goto discard_it;
119
120 /* Pull up the IP header. */
121 __skb_pull(skb, skb_transport_offset(skb));
122
123 skb->csum_valid = 0; /* Previous value not applicable */
124 if (skb_csum_unnecessary(skb))
125 __skb_decr_checksum_unnecessary(skb);
126 else if (!sctp_checksum_disable &&
127 !is_gso &&
128 sctp_rcv_checksum(net, skb) < 0)
129 goto discard_it;
130 skb->csum_valid = 1;
131
132 __skb_pull(skb, sizeof(struct sctphdr));
133
134 family = ipver2af(ip_hdr(skb)->version);
135 af = sctp_get_af_specific(family);
136 if (unlikely(!af))
137 goto discard_it;
138 SCTP_INPUT_CB(skb)->af = af;
139
140 /* Initialize local addresses for lookups. */
141 af->from_skb(&src, skb, 1);
142 af->from_skb(&dest, skb, 0);
143
144 /* If the packet is to or from a non-unicast address,
145 * silently discard the packet.
146 *
147 * This is not clearly defined in the RFC except in section
148 * 8.4 - OOTB handling. However, based on the book "Stream Control
149 * Transmission Protocol" 2.1, "It is important to note that the
150 * IP address of an SCTP transport address must be a routable
151 * unicast address. In other words, IP multicast addresses and
152 * IP broadcast addresses cannot be used in an SCTP transport
153 * address."
154 */
155 if (!af->addr_valid(&src, NULL, skb) ||
156 !af->addr_valid(&dest, NULL, skb))
157 goto discard_it;
158
159 asoc = __sctp_rcv_lookup(net, skb, &src, &dest, &transport);
160
161 if (!asoc)
162 ep = __sctp_rcv_lookup_endpoint(net, skb, &dest, &src);
163
164 /* Retrieve the common input handling substructure. */
165 rcvr = asoc ? &asoc->base : &ep->base;
166 sk = rcvr->sk;
167
168 /*
169 * If a frame arrives on an interface and the receiving socket is
170 * bound to another interface, via SO_BINDTODEVICE, treat it as OOTB
171 */
172 if (sk->sk_bound_dev_if && (sk->sk_bound_dev_if != af->skb_iif(skb))) {
173 if (transport) {
174 sctp_transport_put(transport);
175 asoc = NULL;
176 transport = NULL;
177 } else {
178 sctp_endpoint_put(ep);
179 ep = NULL;
180 }
181 sk = net->sctp.ctl_sock;
182 ep = sctp_sk(sk)->ep;
183 sctp_endpoint_hold(ep);
184 rcvr = &ep->base;
185 }
186
187 /*
188 * RFC 2960, 8.4 - Handle "Out of the blue" Packets.
189 * An SCTP packet is called an "out of the blue" (OOTB)
190 * packet if it is correctly formed, i.e., passed the
191 * receiver's checksum check, but the receiver is not
192 * able to identify the association to which this
193 * packet belongs.
194 */
195 if (!asoc) {
196 if (sctp_rcv_ootb(skb)) {
197 __SCTP_INC_STATS(net, SCTP_MIB_OUTOFBLUES);
198 goto discard_release;
199 }
200 }
201
202 if (!xfrm_policy_check(sk, XFRM_POLICY_IN, skb, family))
203 goto discard_release;
204 nf_reset_ct(skb);
205
206 if (sk_filter(sk, skb))
207 goto discard_release;
208
209 /* Create an SCTP packet structure. */
210 chunk = sctp_chunkify(skb, asoc, sk, GFP_ATOMIC);
211 if (!chunk)
212 goto discard_release;
213 SCTP_INPUT_CB(skb)->chunk = chunk;
214
215 /* Remember what endpoint is to handle this packet. */
216 chunk->rcvr = rcvr;
217
218 /* Remember the SCTP header. */
219 chunk->sctp_hdr = sctp_hdr(skb);
220
221 /* Set the source and destination addresses of the incoming chunk. */
222 sctp_init_addrs(chunk, &src, &dest);
223
224 /* Remember where we came from. */
225 chunk->transport = transport;
226
227 /* Acquire access to the sock lock. Note: We are safe from other
228 * bottom halves on this lock, but a user may be in the lock too,
229 * so check if it is busy.
230 */
231 bh_lock_sock(sk);
232
233 if (sk != rcvr->sk) {
234 /* Our cached sk is different from the rcvr->sk. This is
235 * because migrate()/accept() may have moved the association
236 * to a new socket and released all the sockets. So now we
237 * are holding a lock on the old socket while the user may
238 * be doing something with the new socket. Switch our veiw
239 * of the current sk.
240 */
241 bh_unlock_sock(sk);
242 sk = rcvr->sk;
243 bh_lock_sock(sk);
244 }
245
246 if (sock_owned_by_user(sk) || !sctp_newsk_ready(sk)) {
247 if (sctp_add_backlog(sk, skb)) {
248 bh_unlock_sock(sk);
249 sctp_chunk_free(chunk);
250 skb = NULL; /* sctp_chunk_free already freed the skb */
251 goto discard_release;
252 }
253 __SCTP_INC_STATS(net, SCTP_MIB_IN_PKT_BACKLOG);
254 } else {
255 __SCTP_INC_STATS(net, SCTP_MIB_IN_PKT_SOFTIRQ);
256 sctp_inq_push(&chunk->rcvr->inqueue, chunk);
257 }
258
259 bh_unlock_sock(sk);
260
261 /* Release the asoc/ep ref we took in the lookup calls. */
262 if (transport)
263 sctp_transport_put(transport);
264 else
265 sctp_endpoint_put(ep);
266
267 return 0;
268
269discard_it:
270 __SCTP_INC_STATS(net, SCTP_MIB_IN_PKT_DISCARDS);
271 kfree_skb(skb);
272 return 0;
273
274discard_release:
275 /* Release the asoc/ep ref we took in the lookup calls. */
276 if (transport)
277 sctp_transport_put(transport);
278 else
279 sctp_endpoint_put(ep);
280
281 goto discard_it;
282}
283
284/* Process the backlog queue of the socket. Every skb on
285 * the backlog holds a ref on an association or endpoint.
286 * We hold this ref throughout the state machine to make
287 * sure that the structure we need is still around.
288 */
289int sctp_backlog_rcv(struct sock *sk, struct sk_buff *skb)
290{
291 struct sctp_chunk *chunk = SCTP_INPUT_CB(skb)->chunk;
292 struct sctp_inq *inqueue = &chunk->rcvr->inqueue;
293 struct sctp_transport *t = chunk->transport;
294 struct sctp_ep_common *rcvr = NULL;
295 int backloged = 0;
296
297 rcvr = chunk->rcvr;
298
299 /* If the rcvr is dead then the association or endpoint
300 * has been deleted and we can safely drop the chunk
301 * and refs that we are holding.
302 */
303 if (rcvr->dead) {
304 sctp_chunk_free(chunk);
305 goto done;
306 }
307
308 if (unlikely(rcvr->sk != sk)) {
309 /* In this case, the association moved from one socket to
310 * another. We are currently sitting on the backlog of the
311 * old socket, so we need to move.
312 * However, since we are here in the process context we
313 * need to take make sure that the user doesn't own
314 * the new socket when we process the packet.
315 * If the new socket is user-owned, queue the chunk to the
316 * backlog of the new socket without dropping any refs.
317 * Otherwise, we can safely push the chunk on the inqueue.
318 */
319
320 sk = rcvr->sk;
321 local_bh_disable();
322 bh_lock_sock(sk);
323
324 if (sock_owned_by_user(sk) || !sctp_newsk_ready(sk)) {
325 if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf)))
326 sctp_chunk_free(chunk);
327 else
328 backloged = 1;
329 } else
330 sctp_inq_push(inqueue, chunk);
331
332 bh_unlock_sock(sk);
333 local_bh_enable();
334
335 /* If the chunk was backloged again, don't drop refs */
336 if (backloged)
337 return 0;
338 } else {
339 if (!sctp_newsk_ready(sk)) {
340 if (!sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf)))
341 return 0;
342 sctp_chunk_free(chunk);
343 } else {
344 sctp_inq_push(inqueue, chunk);
345 }
346 }
347
348done:
349 /* Release the refs we took in sctp_add_backlog */
350 if (SCTP_EP_TYPE_ASSOCIATION == rcvr->type)
351 sctp_transport_put(t);
352 else if (SCTP_EP_TYPE_SOCKET == rcvr->type)
353 sctp_endpoint_put(sctp_ep(rcvr));
354 else
355 BUG();
356
357 return 0;
358}
359
360static int sctp_add_backlog(struct sock *sk, struct sk_buff *skb)
361{
362 struct sctp_chunk *chunk = SCTP_INPUT_CB(skb)->chunk;
363 struct sctp_transport *t = chunk->transport;
364 struct sctp_ep_common *rcvr = chunk->rcvr;
365 int ret;
366
367 ret = sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf));
368 if (!ret) {
369 /* Hold the assoc/ep while hanging on the backlog queue.
370 * This way, we know structures we need will not disappear
371 * from us
372 */
373 if (SCTP_EP_TYPE_ASSOCIATION == rcvr->type)
374 sctp_transport_hold(t);
375 else if (SCTP_EP_TYPE_SOCKET == rcvr->type)
376 sctp_endpoint_hold(sctp_ep(rcvr));
377 else
378 BUG();
379 }
380 return ret;
381
382}
383
384/* Handle icmp frag needed error. */
385void sctp_icmp_frag_needed(struct sock *sk, struct sctp_association *asoc,
386 struct sctp_transport *t, __u32 pmtu)
387{
388 if (!t ||
389 (t->pathmtu <= pmtu &&
390 t->pl.probe_size + sctp_transport_pl_hlen(t) <= pmtu))
391 return;
392
393 if (sock_owned_by_user(sk)) {
394 atomic_set(&t->mtu_info, pmtu);
395 asoc->pmtu_pending = 1;
396 t->pmtu_pending = 1;
397 return;
398 }
399
400 if (!(t->param_flags & SPP_PMTUD_ENABLE))
401 /* We can't allow retransmitting in such case, as the
402 * retransmission would be sized just as before, and thus we
403 * would get another icmp, and retransmit again.
404 */
405 return;
406
407 /* Update transports view of the MTU. Return if no update was needed.
408 * If an update wasn't needed/possible, it also doesn't make sense to
409 * try to retransmit now.
410 */
411 if (!sctp_transport_update_pmtu(t, pmtu))
412 return;
413
414 /* Update association pmtu. */
415 sctp_assoc_sync_pmtu(asoc);
416
417 /* Retransmit with the new pmtu setting. */
418 sctp_retransmit(&asoc->outqueue, t, SCTP_RTXR_PMTUD);
419}
420
421void sctp_icmp_redirect(struct sock *sk, struct sctp_transport *t,
422 struct sk_buff *skb)
423{
424 struct dst_entry *dst;
425
426 if (sock_owned_by_user(sk) || !t)
427 return;
428 dst = sctp_transport_dst_check(t);
429 if (dst)
430 dst->ops->redirect(dst, sk, skb);
431}
432
433/*
434 * SCTP Implementer's Guide, 2.37 ICMP handling procedures
435 *
436 * ICMP8) If the ICMP code is a "Unrecognized next header type encountered"
437 * or a "Protocol Unreachable" treat this message as an abort
438 * with the T bit set.
439 *
440 * This function sends an event to the state machine, which will abort the
441 * association.
442 *
443 */
444void sctp_icmp_proto_unreachable(struct sock *sk,
445 struct sctp_association *asoc,
446 struct sctp_transport *t)
447{
448 if (sock_owned_by_user(sk)) {
449 if (timer_pending(&t->proto_unreach_timer))
450 return;
451 else {
452 if (!mod_timer(&t->proto_unreach_timer,
453 jiffies + (HZ/20)))
454 sctp_transport_hold(t);
455 }
456 } else {
457 struct net *net = sock_net(sk);
458
459 pr_debug("%s: unrecognized next header type "
460 "encountered!\n", __func__);
461
462 if (del_timer(&t->proto_unreach_timer))
463 sctp_transport_put(t);
464
465 sctp_do_sm(net, SCTP_EVENT_T_OTHER,
466 SCTP_ST_OTHER(SCTP_EVENT_ICMP_PROTO_UNREACH),
467 asoc->state, asoc->ep, asoc, t,
468 GFP_ATOMIC);
469 }
470}
471
472/* Common lookup code for icmp/icmpv6 error handler. */
473struct sock *sctp_err_lookup(struct net *net, int family, struct sk_buff *skb,
474 struct sctphdr *sctphdr,
475 struct sctp_association **app,
476 struct sctp_transport **tpp)
477{
478 struct sctp_init_chunk *chunkhdr, _chunkhdr;
479 union sctp_addr saddr;
480 union sctp_addr daddr;
481 struct sctp_af *af;
482 struct sock *sk = NULL;
483 struct sctp_association *asoc;
484 struct sctp_transport *transport = NULL;
485 __u32 vtag = ntohl(sctphdr->vtag);
486
487 *app = NULL; *tpp = NULL;
488
489 af = sctp_get_af_specific(family);
490 if (unlikely(!af)) {
491 return NULL;
492 }
493
494 /* Initialize local addresses for lookups. */
495 af->from_skb(&saddr, skb, 1);
496 af->from_skb(&daddr, skb, 0);
497
498 /* Look for an association that matches the incoming ICMP error
499 * packet.
500 */
501 asoc = __sctp_lookup_association(net, &saddr, &daddr, &transport);
502 if (!asoc)
503 return NULL;
504
505 sk = asoc->base.sk;
506
507 /* RFC 4960, Appendix C. ICMP Handling
508 *
509 * ICMP6) An implementation MUST validate that the Verification Tag
510 * contained in the ICMP message matches the Verification Tag of
511 * the peer. If the Verification Tag is not 0 and does NOT
512 * match, discard the ICMP message. If it is 0 and the ICMP
513 * message contains enough bytes to verify that the chunk type is
514 * an INIT chunk and that the Initiate Tag matches the tag of the
515 * peer, continue with ICMP7. If the ICMP message is too short
516 * or the chunk type or the Initiate Tag does not match, silently
517 * discard the packet.
518 */
519 if (vtag == 0) {
520 /* chunk header + first 4 octects of init header */
521 chunkhdr = skb_header_pointer(skb, skb_transport_offset(skb) +
522 sizeof(struct sctphdr),
523 sizeof(struct sctp_chunkhdr) +
524 sizeof(__be32), &_chunkhdr);
525 if (!chunkhdr ||
526 chunkhdr->chunk_hdr.type != SCTP_CID_INIT ||
527 ntohl(chunkhdr->init_hdr.init_tag) != asoc->c.my_vtag)
528 goto out;
529
530 } else if (vtag != asoc->c.peer_vtag) {
531 goto out;
532 }
533
534 bh_lock_sock(sk);
535
536 /* If too many ICMPs get dropped on busy
537 * servers this needs to be solved differently.
538 */
539 if (sock_owned_by_user(sk))
540 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
541
542 *app = asoc;
543 *tpp = transport;
544 return sk;
545
546out:
547 sctp_transport_put(transport);
548 return NULL;
549}
550
551/* Common cleanup code for icmp/icmpv6 error handler. */
552void sctp_err_finish(struct sock *sk, struct sctp_transport *t)
553 __releases(&((__sk)->sk_lock.slock))
554{
555 bh_unlock_sock(sk);
556 sctp_transport_put(t);
557}
558
559static void sctp_v4_err_handle(struct sctp_transport *t, struct sk_buff *skb,
560 __u8 type, __u8 code, __u32 info)
561{
562 struct sctp_association *asoc = t->asoc;
563 struct sock *sk = asoc->base.sk;
564 int err = 0;
565
566 switch (type) {
567 case ICMP_PARAMETERPROB:
568 err = EPROTO;
569 break;
570 case ICMP_DEST_UNREACH:
571 if (code > NR_ICMP_UNREACH)
572 return;
573 if (code == ICMP_FRAG_NEEDED) {
574 sctp_icmp_frag_needed(sk, asoc, t, SCTP_TRUNC4(info));
575 return;
576 }
577 if (code == ICMP_PROT_UNREACH) {
578 sctp_icmp_proto_unreachable(sk, asoc, t);
579 return;
580 }
581 err = icmp_err_convert[code].errno;
582 break;
583 case ICMP_TIME_EXCEEDED:
584 if (code == ICMP_EXC_FRAGTIME)
585 return;
586
587 err = EHOSTUNREACH;
588 break;
589 case ICMP_REDIRECT:
590 sctp_icmp_redirect(sk, t, skb);
591 return;
592 default:
593 return;
594 }
595 if (!sock_owned_by_user(sk) && inet_sk(sk)->recverr) {
596 sk->sk_err = err;
597 sk_error_report(sk);
598 } else { /* Only an error on timeout */
599 sk->sk_err_soft = err;
600 }
601}
602
603/*
604 * This routine is called by the ICMP module when it gets some
605 * sort of error condition. If err < 0 then the socket should
606 * be closed and the error returned to the user. If err > 0
607 * it's just the icmp type << 8 | icmp code. After adjustment
608 * header points to the first 8 bytes of the sctp header. We need
609 * to find the appropriate port.
610 *
611 * The locking strategy used here is very "optimistic". When
612 * someone else accesses the socket the ICMP is just dropped
613 * and for some paths there is no check at all.
614 * A more general error queue to queue errors for later handling
615 * is probably better.
616 *
617 */
618int sctp_v4_err(struct sk_buff *skb, __u32 info)
619{
620 const struct iphdr *iph = (const struct iphdr *)skb->data;
621 const int type = icmp_hdr(skb)->type;
622 const int code = icmp_hdr(skb)->code;
623 struct net *net = dev_net(skb->dev);
624 struct sctp_transport *transport;
625 struct sctp_association *asoc;
626 __u16 saveip, savesctp;
627 struct sock *sk;
628
629 /* Fix up skb to look at the embedded net header. */
630 saveip = skb->network_header;
631 savesctp = skb->transport_header;
632 skb_reset_network_header(skb);
633 skb_set_transport_header(skb, iph->ihl * 4);
634 sk = sctp_err_lookup(net, AF_INET, skb, sctp_hdr(skb), &asoc, &transport);
635 /* Put back, the original values. */
636 skb->network_header = saveip;
637 skb->transport_header = savesctp;
638 if (!sk) {
639 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
640 return -ENOENT;
641 }
642
643 sctp_v4_err_handle(transport, skb, type, code, info);
644 sctp_err_finish(sk, transport);
645
646 return 0;
647}
648
649int sctp_udp_v4_err(struct sock *sk, struct sk_buff *skb)
650{
651 struct net *net = dev_net(skb->dev);
652 struct sctp_association *asoc;
653 struct sctp_transport *t;
654 struct icmphdr *hdr;
655 __u32 info = 0;
656
657 skb->transport_header += sizeof(struct udphdr);
658 sk = sctp_err_lookup(net, AF_INET, skb, sctp_hdr(skb), &asoc, &t);
659 if (!sk) {
660 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
661 return -ENOENT;
662 }
663
664 skb->transport_header -= sizeof(struct udphdr);
665 hdr = (struct icmphdr *)(skb_network_header(skb) - sizeof(struct icmphdr));
666 if (hdr->type == ICMP_REDIRECT) {
667 /* can't be handled without outer iphdr known, leave it to udp_err */
668 sctp_err_finish(sk, t);
669 return 0;
670 }
671 if (hdr->type == ICMP_DEST_UNREACH && hdr->code == ICMP_FRAG_NEEDED)
672 info = ntohs(hdr->un.frag.mtu);
673 sctp_v4_err_handle(t, skb, hdr->type, hdr->code, info);
674
675 sctp_err_finish(sk, t);
676 return 1;
677}
678
679/*
680 * RFC 2960, 8.4 - Handle "Out of the blue" Packets.
681 *
682 * This function scans all the chunks in the OOTB packet to determine if
683 * the packet should be discarded right away. If a response might be needed
684 * for this packet, or, if further processing is possible, the packet will
685 * be queued to a proper inqueue for the next phase of handling.
686 *
687 * Output:
688 * Return 0 - If further processing is needed.
689 * Return 1 - If the packet can be discarded right away.
690 */
691static int sctp_rcv_ootb(struct sk_buff *skb)
692{
693 struct sctp_chunkhdr *ch, _ch;
694 int ch_end, offset = 0;
695
696 /* Scan through all the chunks in the packet. */
697 do {
698 /* Make sure we have at least the header there */
699 if (offset + sizeof(_ch) > skb->len)
700 break;
701
702 ch = skb_header_pointer(skb, offset, sizeof(*ch), &_ch);
703
704 /* Break out if chunk length is less then minimal. */
705 if (!ch || ntohs(ch->length) < sizeof(_ch))
706 break;
707
708 ch_end = offset + SCTP_PAD4(ntohs(ch->length));
709 if (ch_end > skb->len)
710 break;
711
712 /* RFC 8.4, 2) If the OOTB packet contains an ABORT chunk, the
713 * receiver MUST silently discard the OOTB packet and take no
714 * further action.
715 */
716 if (SCTP_CID_ABORT == ch->type)
717 goto discard;
718
719 /* RFC 8.4, 6) If the packet contains a SHUTDOWN COMPLETE
720 * chunk, the receiver should silently discard the packet
721 * and take no further action.
722 */
723 if (SCTP_CID_SHUTDOWN_COMPLETE == ch->type)
724 goto discard;
725
726 /* RFC 4460, 2.11.2
727 * This will discard packets with INIT chunk bundled as
728 * subsequent chunks in the packet. When INIT is first,
729 * the normal INIT processing will discard the chunk.
730 */
731 if (SCTP_CID_INIT == ch->type && (void *)ch != skb->data)
732 goto discard;
733
734 offset = ch_end;
735 } while (ch_end < skb->len);
736
737 return 0;
738
739discard:
740 return 1;
741}
742
743/* Insert endpoint into the hash table. */
744static int __sctp_hash_endpoint(struct sctp_endpoint *ep)
745{
746 struct sock *sk = ep->base.sk;
747 struct net *net = sock_net(sk);
748 struct sctp_hashbucket *head;
749 struct sctp_ep_common *epb;
750
751 epb = &ep->base;
752 epb->hashent = sctp_ep_hashfn(net, epb->bind_addr.port);
753 head = &sctp_ep_hashtable[epb->hashent];
754
755 if (sk->sk_reuseport) {
756 bool any = sctp_is_ep_boundall(sk);
757 struct sctp_ep_common *epb2;
758 struct list_head *list;
759 int cnt = 0, err = 1;
760
761 list_for_each(list, &ep->base.bind_addr.address_list)
762 cnt++;
763
764 sctp_for_each_hentry(epb2, &head->chain) {
765 struct sock *sk2 = epb2->sk;
766
767 if (!net_eq(sock_net(sk2), net) || sk2 == sk ||
768 !uid_eq(sock_i_uid(sk2), sock_i_uid(sk)) ||
769 !sk2->sk_reuseport)
770 continue;
771
772 err = sctp_bind_addrs_check(sctp_sk(sk2),
773 sctp_sk(sk), cnt);
774 if (!err) {
775 err = reuseport_add_sock(sk, sk2, any);
776 if (err)
777 return err;
778 break;
779 } else if (err < 0) {
780 return err;
781 }
782 }
783
784 if (err) {
785 err = reuseport_alloc(sk, any);
786 if (err)
787 return err;
788 }
789 }
790
791 write_lock(&head->lock);
792 hlist_add_head(&epb->node, &head->chain);
793 write_unlock(&head->lock);
794 return 0;
795}
796
797/* Add an endpoint to the hash. Local BH-safe. */
798int sctp_hash_endpoint(struct sctp_endpoint *ep)
799{
800 int err;
801
802 local_bh_disable();
803 err = __sctp_hash_endpoint(ep);
804 local_bh_enable();
805
806 return err;
807}
808
809/* Remove endpoint from the hash table. */
810static void __sctp_unhash_endpoint(struct sctp_endpoint *ep)
811{
812 struct sock *sk = ep->base.sk;
813 struct sctp_hashbucket *head;
814 struct sctp_ep_common *epb;
815
816 epb = &ep->base;
817
818 epb->hashent = sctp_ep_hashfn(sock_net(sk), epb->bind_addr.port);
819
820 head = &sctp_ep_hashtable[epb->hashent];
821
822 if (rcu_access_pointer(sk->sk_reuseport_cb))
823 reuseport_detach_sock(sk);
824
825 write_lock(&head->lock);
826 hlist_del_init(&epb->node);
827 write_unlock(&head->lock);
828}
829
830/* Remove endpoint from the hash. Local BH-safe. */
831void sctp_unhash_endpoint(struct sctp_endpoint *ep)
832{
833 local_bh_disable();
834 __sctp_unhash_endpoint(ep);
835 local_bh_enable();
836}
837
838static inline __u32 sctp_hashfn(const struct net *net, __be16 lport,
839 const union sctp_addr *paddr, __u32 seed)
840{
841 __u32 addr;
842
843 if (paddr->sa.sa_family == AF_INET6)
844 addr = jhash(&paddr->v6.sin6_addr, 16, seed);
845 else
846 addr = (__force __u32)paddr->v4.sin_addr.s_addr;
847
848 return jhash_3words(addr, ((__force __u32)paddr->v4.sin_port) << 16 |
849 (__force __u32)lport, net_hash_mix(net), seed);
850}
851
852/* Look up an endpoint. */
853static struct sctp_endpoint *__sctp_rcv_lookup_endpoint(
854 struct net *net, struct sk_buff *skb,
855 const union sctp_addr *laddr,
856 const union sctp_addr *paddr)
857{
858 struct sctp_hashbucket *head;
859 struct sctp_ep_common *epb;
860 struct sctp_endpoint *ep;
861 struct sock *sk;
862 __be16 lport;
863 int hash;
864
865 lport = laddr->v4.sin_port;
866 hash = sctp_ep_hashfn(net, ntohs(lport));
867 head = &sctp_ep_hashtable[hash];
868 read_lock(&head->lock);
869 sctp_for_each_hentry(epb, &head->chain) {
870 ep = sctp_ep(epb);
871 if (sctp_endpoint_is_match(ep, net, laddr))
872 goto hit;
873 }
874
875 ep = sctp_sk(net->sctp.ctl_sock)->ep;
876
877hit:
878 sk = ep->base.sk;
879 if (sk->sk_reuseport) {
880 __u32 phash = sctp_hashfn(net, lport, paddr, 0);
881
882 sk = reuseport_select_sock(sk, phash, skb,
883 sizeof(struct sctphdr));
884 if (sk)
885 ep = sctp_sk(sk)->ep;
886 }
887 sctp_endpoint_hold(ep);
888 read_unlock(&head->lock);
889 return ep;
890}
891
892/* rhashtable for transport */
893struct sctp_hash_cmp_arg {
894 const union sctp_addr *paddr;
895 const struct net *net;
896 __be16 lport;
897};
898
899static inline int sctp_hash_cmp(struct rhashtable_compare_arg *arg,
900 const void *ptr)
901{
902 struct sctp_transport *t = (struct sctp_transport *)ptr;
903 const struct sctp_hash_cmp_arg *x = arg->key;
904 int err = 1;
905
906 if (!sctp_cmp_addr_exact(&t->ipaddr, x->paddr))
907 return err;
908 if (!sctp_transport_hold(t))
909 return err;
910
911 if (!net_eq(t->asoc->base.net, x->net))
912 goto out;
913 if (x->lport != htons(t->asoc->base.bind_addr.port))
914 goto out;
915
916 err = 0;
917out:
918 sctp_transport_put(t);
919 return err;
920}
921
922static inline __u32 sctp_hash_obj(const void *data, u32 len, u32 seed)
923{
924 const struct sctp_transport *t = data;
925
926 return sctp_hashfn(t->asoc->base.net,
927 htons(t->asoc->base.bind_addr.port),
928 &t->ipaddr, seed);
929}
930
931static inline __u32 sctp_hash_key(const void *data, u32 len, u32 seed)
932{
933 const struct sctp_hash_cmp_arg *x = data;
934
935 return sctp_hashfn(x->net, x->lport, x->paddr, seed);
936}
937
938static const struct rhashtable_params sctp_hash_params = {
939 .head_offset = offsetof(struct sctp_transport, node),
940 .hashfn = sctp_hash_key,
941 .obj_hashfn = sctp_hash_obj,
942 .obj_cmpfn = sctp_hash_cmp,
943 .automatic_shrinking = true,
944};
945
946int sctp_transport_hashtable_init(void)
947{
948 return rhltable_init(&sctp_transport_hashtable, &sctp_hash_params);
949}
950
951void sctp_transport_hashtable_destroy(void)
952{
953 rhltable_destroy(&sctp_transport_hashtable);
954}
955
956int sctp_hash_transport(struct sctp_transport *t)
957{
958 struct sctp_transport *transport;
959 struct rhlist_head *tmp, *list;
960 struct sctp_hash_cmp_arg arg;
961 int err;
962
963 if (t->asoc->temp)
964 return 0;
965
966 arg.net = t->asoc->base.net;
967 arg.paddr = &t->ipaddr;
968 arg.lport = htons(t->asoc->base.bind_addr.port);
969
970 rcu_read_lock();
971 list = rhltable_lookup(&sctp_transport_hashtable, &arg,
972 sctp_hash_params);
973
974 rhl_for_each_entry_rcu(transport, tmp, list, node)
975 if (transport->asoc->ep == t->asoc->ep) {
976 rcu_read_unlock();
977 return -EEXIST;
978 }
979 rcu_read_unlock();
980
981 err = rhltable_insert_key(&sctp_transport_hashtable, &arg,
982 &t->node, sctp_hash_params);
983 if (err)
984 pr_err_once("insert transport fail, errno %d\n", err);
985
986 return err;
987}
988
989void sctp_unhash_transport(struct sctp_transport *t)
990{
991 if (t->asoc->temp)
992 return;
993
994 rhltable_remove(&sctp_transport_hashtable, &t->node,
995 sctp_hash_params);
996}
997
998/* return a transport with holding it */
999struct sctp_transport *sctp_addrs_lookup_transport(
1000 struct net *net,
1001 const union sctp_addr *laddr,
1002 const union sctp_addr *paddr)
1003{
1004 struct rhlist_head *tmp, *list;
1005 struct sctp_transport *t;
1006 struct sctp_hash_cmp_arg arg = {
1007 .paddr = paddr,
1008 .net = net,
1009 .lport = laddr->v4.sin_port,
1010 };
1011
1012 list = rhltable_lookup(&sctp_transport_hashtable, &arg,
1013 sctp_hash_params);
1014
1015 rhl_for_each_entry_rcu(t, tmp, list, node) {
1016 if (!sctp_transport_hold(t))
1017 continue;
1018
1019 if (sctp_bind_addr_match(&t->asoc->base.bind_addr,
1020 laddr, sctp_sk(t->asoc->base.sk)))
1021 return t;
1022 sctp_transport_put(t);
1023 }
1024
1025 return NULL;
1026}
1027
1028/* return a transport without holding it, as it's only used under sock lock */
1029struct sctp_transport *sctp_epaddr_lookup_transport(
1030 const struct sctp_endpoint *ep,
1031 const union sctp_addr *paddr)
1032{
1033 struct rhlist_head *tmp, *list;
1034 struct sctp_transport *t;
1035 struct sctp_hash_cmp_arg arg = {
1036 .paddr = paddr,
1037 .net = ep->base.net,
1038 .lport = htons(ep->base.bind_addr.port),
1039 };
1040
1041 list = rhltable_lookup(&sctp_transport_hashtable, &arg,
1042 sctp_hash_params);
1043
1044 rhl_for_each_entry_rcu(t, tmp, list, node)
1045 if (ep == t->asoc->ep)
1046 return t;
1047
1048 return NULL;
1049}
1050
1051/* Look up an association. */
1052static struct sctp_association *__sctp_lookup_association(
1053 struct net *net,
1054 const union sctp_addr *local,
1055 const union sctp_addr *peer,
1056 struct sctp_transport **pt)
1057{
1058 struct sctp_transport *t;
1059 struct sctp_association *asoc = NULL;
1060
1061 t = sctp_addrs_lookup_transport(net, local, peer);
1062 if (!t)
1063 goto out;
1064
1065 asoc = t->asoc;
1066 *pt = t;
1067
1068out:
1069 return asoc;
1070}
1071
1072/* Look up an association. protected by RCU read lock */
1073static
1074struct sctp_association *sctp_lookup_association(struct net *net,
1075 const union sctp_addr *laddr,
1076 const union sctp_addr *paddr,
1077 struct sctp_transport **transportp)
1078{
1079 struct sctp_association *asoc;
1080
1081 rcu_read_lock();
1082 asoc = __sctp_lookup_association(net, laddr, paddr, transportp);
1083 rcu_read_unlock();
1084
1085 return asoc;
1086}
1087
1088/* Is there an association matching the given local and peer addresses? */
1089bool sctp_has_association(struct net *net,
1090 const union sctp_addr *laddr,
1091 const union sctp_addr *paddr)
1092{
1093 struct sctp_transport *transport;
1094
1095 if (sctp_lookup_association(net, laddr, paddr, &transport)) {
1096 sctp_transport_put(transport);
1097 return true;
1098 }
1099
1100 return false;
1101}
1102
1103/*
1104 * SCTP Implementors Guide, 2.18 Handling of address
1105 * parameters within the INIT or INIT-ACK.
1106 *
1107 * D) When searching for a matching TCB upon reception of an INIT
1108 * or INIT-ACK chunk the receiver SHOULD use not only the
1109 * source address of the packet (containing the INIT or
1110 * INIT-ACK) but the receiver SHOULD also use all valid
1111 * address parameters contained within the chunk.
1112 *
1113 * 2.18.3 Solution description
1114 *
1115 * This new text clearly specifies to an implementor the need
1116 * to look within the INIT or INIT-ACK. Any implementation that
1117 * does not do this, may not be able to establish associations
1118 * in certain circumstances.
1119 *
1120 */
1121static struct sctp_association *__sctp_rcv_init_lookup(struct net *net,
1122 struct sk_buff *skb,
1123 const union sctp_addr *laddr, struct sctp_transport **transportp)
1124{
1125 struct sctp_association *asoc;
1126 union sctp_addr addr;
1127 union sctp_addr *paddr = &addr;
1128 struct sctphdr *sh = sctp_hdr(skb);
1129 union sctp_params params;
1130 struct sctp_init_chunk *init;
1131 struct sctp_af *af;
1132
1133 /*
1134 * This code will NOT touch anything inside the chunk--it is
1135 * strictly READ-ONLY.
1136 *
1137 * RFC 2960 3 SCTP packet Format
1138 *
1139 * Multiple chunks can be bundled into one SCTP packet up to
1140 * the MTU size, except for the INIT, INIT ACK, and SHUTDOWN
1141 * COMPLETE chunks. These chunks MUST NOT be bundled with any
1142 * other chunk in a packet. See Section 6.10 for more details
1143 * on chunk bundling.
1144 */
1145
1146 /* Find the start of the TLVs and the end of the chunk. This is
1147 * the region we search for address parameters.
1148 */
1149 init = (struct sctp_init_chunk *)skb->data;
1150
1151 /* Walk the parameters looking for embedded addresses. */
1152 sctp_walk_params(params, init, init_hdr.params) {
1153
1154 /* Note: Ignoring hostname addresses. */
1155 af = sctp_get_af_specific(param_type2af(params.p->type));
1156 if (!af)
1157 continue;
1158
1159 if (!af->from_addr_param(paddr, params.addr, sh->source, 0))
1160 continue;
1161
1162 asoc = __sctp_lookup_association(net, laddr, paddr, transportp);
1163 if (asoc)
1164 return asoc;
1165 }
1166
1167 return NULL;
1168}
1169
1170/* ADD-IP, Section 5.2
1171 * When an endpoint receives an ASCONF Chunk from the remote peer
1172 * special procedures may be needed to identify the association the
1173 * ASCONF Chunk is associated with. To properly find the association
1174 * the following procedures SHOULD be followed:
1175 *
1176 * D2) If the association is not found, use the address found in the
1177 * Address Parameter TLV combined with the port number found in the
1178 * SCTP common header. If found proceed to rule D4.
1179 *
1180 * D2-ext) If more than one ASCONF Chunks are packed together, use the
1181 * address found in the ASCONF Address Parameter TLV of each of the
1182 * subsequent ASCONF Chunks. If found, proceed to rule D4.
1183 */
1184static struct sctp_association *__sctp_rcv_asconf_lookup(
1185 struct net *net,
1186 struct sctp_chunkhdr *ch,
1187 const union sctp_addr *laddr,
1188 __be16 peer_port,
1189 struct sctp_transport **transportp)
1190{
1191 struct sctp_addip_chunk *asconf = (struct sctp_addip_chunk *)ch;
1192 struct sctp_af *af;
1193 union sctp_addr_param *param;
1194 union sctp_addr paddr;
1195
1196 if (ntohs(ch->length) < sizeof(*asconf) + sizeof(struct sctp_paramhdr))
1197 return NULL;
1198
1199 /* Skip over the ADDIP header and find the Address parameter */
1200 param = (union sctp_addr_param *)(asconf + 1);
1201
1202 af = sctp_get_af_specific(param_type2af(param->p.type));
1203 if (unlikely(!af))
1204 return NULL;
1205
1206 if (!af->from_addr_param(&paddr, param, peer_port, 0))
1207 return NULL;
1208
1209 return __sctp_lookup_association(net, laddr, &paddr, transportp);
1210}
1211
1212
1213/* SCTP-AUTH, Section 6.3:
1214* If the receiver does not find a STCB for a packet containing an AUTH
1215* chunk as the first chunk and not a COOKIE-ECHO chunk as the second
1216* chunk, it MUST use the chunks after the AUTH chunk to look up an existing
1217* association.
1218*
1219* This means that any chunks that can help us identify the association need
1220* to be looked at to find this association.
1221*/
1222static struct sctp_association *__sctp_rcv_walk_lookup(struct net *net,
1223 struct sk_buff *skb,
1224 const union sctp_addr *laddr,
1225 struct sctp_transport **transportp)
1226{
1227 struct sctp_association *asoc = NULL;
1228 struct sctp_chunkhdr *ch;
1229 int have_auth = 0;
1230 unsigned int chunk_num = 1;
1231 __u8 *ch_end;
1232
1233 /* Walk through the chunks looking for AUTH or ASCONF chunks
1234 * to help us find the association.
1235 */
1236 ch = (struct sctp_chunkhdr *)skb->data;
1237 do {
1238 /* Break out if chunk length is less then minimal. */
1239 if (ntohs(ch->length) < sizeof(*ch))
1240 break;
1241
1242 ch_end = ((__u8 *)ch) + SCTP_PAD4(ntohs(ch->length));
1243 if (ch_end > skb_tail_pointer(skb))
1244 break;
1245
1246 switch (ch->type) {
1247 case SCTP_CID_AUTH:
1248 have_auth = chunk_num;
1249 break;
1250
1251 case SCTP_CID_COOKIE_ECHO:
1252 /* If a packet arrives containing an AUTH chunk as
1253 * a first chunk, a COOKIE-ECHO chunk as the second
1254 * chunk, and possibly more chunks after them, and
1255 * the receiver does not have an STCB for that
1256 * packet, then authentication is based on
1257 * the contents of the COOKIE- ECHO chunk.
1258 */
1259 if (have_auth == 1 && chunk_num == 2)
1260 return NULL;
1261 break;
1262
1263 case SCTP_CID_ASCONF:
1264 if (have_auth || net->sctp.addip_noauth)
1265 asoc = __sctp_rcv_asconf_lookup(
1266 net, ch, laddr,
1267 sctp_hdr(skb)->source,
1268 transportp);
1269 break;
1270 default:
1271 break;
1272 }
1273
1274 if (asoc)
1275 break;
1276
1277 ch = (struct sctp_chunkhdr *)ch_end;
1278 chunk_num++;
1279 } while (ch_end + sizeof(*ch) < skb_tail_pointer(skb));
1280
1281 return asoc;
1282}
1283
1284/*
1285 * There are circumstances when we need to look inside the SCTP packet
1286 * for information to help us find the association. Examples
1287 * include looking inside of INIT/INIT-ACK chunks or after the AUTH
1288 * chunks.
1289 */
1290static struct sctp_association *__sctp_rcv_lookup_harder(struct net *net,
1291 struct sk_buff *skb,
1292 const union sctp_addr *laddr,
1293 struct sctp_transport **transportp)
1294{
1295 struct sctp_chunkhdr *ch;
1296
1297 /* We do not allow GSO frames here as we need to linearize and
1298 * then cannot guarantee frame boundaries. This shouldn't be an
1299 * issue as packets hitting this are mostly INIT or INIT-ACK and
1300 * those cannot be on GSO-style anyway.
1301 */
1302 if (skb_is_gso(skb) && skb_is_gso_sctp(skb))
1303 return NULL;
1304
1305 ch = (struct sctp_chunkhdr *)skb->data;
1306
1307 /* The code below will attempt to walk the chunk and extract
1308 * parameter information. Before we do that, we need to verify
1309 * that the chunk length doesn't cause overflow. Otherwise, we'll
1310 * walk off the end.
1311 */
1312 if (SCTP_PAD4(ntohs(ch->length)) > skb->len)
1313 return NULL;
1314
1315 /* If this is INIT/INIT-ACK look inside the chunk too. */
1316 if (ch->type == SCTP_CID_INIT || ch->type == SCTP_CID_INIT_ACK)
1317 return __sctp_rcv_init_lookup(net, skb, laddr, transportp);
1318
1319 return __sctp_rcv_walk_lookup(net, skb, laddr, transportp);
1320}
1321
1322/* Lookup an association for an inbound skb. */
1323static struct sctp_association *__sctp_rcv_lookup(struct net *net,
1324 struct sk_buff *skb,
1325 const union sctp_addr *paddr,
1326 const union sctp_addr *laddr,
1327 struct sctp_transport **transportp)
1328{
1329 struct sctp_association *asoc;
1330
1331 asoc = __sctp_lookup_association(net, laddr, paddr, transportp);
1332 if (asoc)
1333 goto out;
1334
1335 /* Further lookup for INIT/INIT-ACK packets.
1336 * SCTP Implementors Guide, 2.18 Handling of address
1337 * parameters within the INIT or INIT-ACK.
1338 */
1339 asoc = __sctp_rcv_lookup_harder(net, skb, laddr, transportp);
1340 if (asoc)
1341 goto out;
1342
1343 if (paddr->sa.sa_family == AF_INET)
1344 pr_debug("sctp: asoc not found for src:%pI4:%d dst:%pI4:%d\n",
1345 &laddr->v4.sin_addr, ntohs(laddr->v4.sin_port),
1346 &paddr->v4.sin_addr, ntohs(paddr->v4.sin_port));
1347 else
1348 pr_debug("sctp: asoc not found for src:%pI6:%d dst:%pI6:%d\n",
1349 &laddr->v6.sin6_addr, ntohs(laddr->v6.sin6_port),
1350 &paddr->v6.sin6_addr, ntohs(paddr->v6.sin6_port));
1351
1352out:
1353 return asoc;
1354}
1/* SCTP kernel implementation
2 * Copyright (c) 1999-2000 Cisco, Inc.
3 * Copyright (c) 1999-2001 Motorola, Inc.
4 * Copyright (c) 2001-2003 International Business Machines, Corp.
5 * Copyright (c) 2001 Intel Corp.
6 * Copyright (c) 2001 Nokia, Inc.
7 * Copyright (c) 2001 La Monte H.P. Yarroll
8 *
9 * This file is part of the SCTP kernel implementation
10 *
11 * These functions handle all input from the IP layer into SCTP.
12 *
13 * This SCTP implementation is free software;
14 * you can redistribute it and/or modify it under the terms of
15 * the GNU General Public License as published by
16 * the Free Software Foundation; either version 2, or (at your option)
17 * any later version.
18 *
19 * This SCTP implementation is distributed in the hope that it
20 * will be useful, but WITHOUT ANY WARRANTY; without even the implied
21 * ************************
22 * warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
23 * See the GNU General Public License for more details.
24 *
25 * You should have received a copy of the GNU General Public License
26 * along with GNU CC; see the file COPYING. If not, see
27 * <http://www.gnu.org/licenses/>.
28 *
29 * Please send any bug reports or fixes you make to the
30 * email address(es):
31 * lksctp developers <linux-sctp@vger.kernel.org>
32 *
33 * Written or modified by:
34 * La Monte H.P. Yarroll <piggy@acm.org>
35 * Karl Knutson <karl@athena.chicago.il.us>
36 * Xingang Guo <xingang.guo@intel.com>
37 * Jon Grimm <jgrimm@us.ibm.com>
38 * Hui Huang <hui.huang@nokia.com>
39 * Daisy Chang <daisyc@us.ibm.com>
40 * Sridhar Samudrala <sri@us.ibm.com>
41 * Ardelle Fan <ardelle.fan@intel.com>
42 */
43
44#include <linux/types.h>
45#include <linux/list.h> /* For struct list_head */
46#include <linux/socket.h>
47#include <linux/ip.h>
48#include <linux/time.h> /* For struct timeval */
49#include <linux/slab.h>
50#include <net/ip.h>
51#include <net/icmp.h>
52#include <net/snmp.h>
53#include <net/sock.h>
54#include <net/xfrm.h>
55#include <net/sctp/sctp.h>
56#include <net/sctp/sm.h>
57#include <net/sctp/checksum.h>
58#include <net/net_namespace.h>
59
60/* Forward declarations for internal helpers. */
61static int sctp_rcv_ootb(struct sk_buff *);
62static struct sctp_association *__sctp_rcv_lookup(struct net *net,
63 struct sk_buff *skb,
64 const union sctp_addr *paddr,
65 const union sctp_addr *laddr,
66 struct sctp_transport **transportp);
67static struct sctp_endpoint *__sctp_rcv_lookup_endpoint(struct net *net,
68 const union sctp_addr *laddr);
69static struct sctp_association *__sctp_lookup_association(
70 struct net *net,
71 const union sctp_addr *local,
72 const union sctp_addr *peer,
73 struct sctp_transport **pt);
74
75static int sctp_add_backlog(struct sock *sk, struct sk_buff *skb);
76
77
78/* Calculate the SCTP checksum of an SCTP packet. */
79static inline int sctp_rcv_checksum(struct net *net, struct sk_buff *skb)
80{
81 struct sctphdr *sh = sctp_hdr(skb);
82 __le32 cmp = sh->checksum;
83 __le32 val = sctp_compute_cksum(skb, 0);
84
85 if (val != cmp) {
86 /* CRC failure, dump it. */
87 SCTP_INC_STATS_BH(net, SCTP_MIB_CHECKSUMERRORS);
88 return -1;
89 }
90 return 0;
91}
92
93struct sctp_input_cb {
94 union {
95 struct inet_skb_parm h4;
96#if IS_ENABLED(CONFIG_IPV6)
97 struct inet6_skb_parm h6;
98#endif
99 } header;
100 struct sctp_chunk *chunk;
101};
102#define SCTP_INPUT_CB(__skb) ((struct sctp_input_cb *)&((__skb)->cb[0]))
103
104/*
105 * This is the routine which IP calls when receiving an SCTP packet.
106 */
107int sctp_rcv(struct sk_buff *skb)
108{
109 struct sock *sk;
110 struct sctp_association *asoc;
111 struct sctp_endpoint *ep = NULL;
112 struct sctp_ep_common *rcvr;
113 struct sctp_transport *transport = NULL;
114 struct sctp_chunk *chunk;
115 struct sctphdr *sh;
116 union sctp_addr src;
117 union sctp_addr dest;
118 int family;
119 struct sctp_af *af;
120 struct net *net = dev_net(skb->dev);
121
122 if (skb->pkt_type != PACKET_HOST)
123 goto discard_it;
124
125 SCTP_INC_STATS_BH(net, SCTP_MIB_INSCTPPACKS);
126
127 if (skb_linearize(skb))
128 goto discard_it;
129
130 sh = sctp_hdr(skb);
131
132 /* Pull up the IP and SCTP headers. */
133 __skb_pull(skb, skb_transport_offset(skb));
134 if (skb->len < sizeof(struct sctphdr))
135 goto discard_it;
136 if (!sctp_checksum_disable && !skb_csum_unnecessary(skb) &&
137 sctp_rcv_checksum(net, skb) < 0)
138 goto discard_it;
139
140 skb_pull(skb, sizeof(struct sctphdr));
141
142 /* Make sure we at least have chunk headers worth of data left. */
143 if (skb->len < sizeof(struct sctp_chunkhdr))
144 goto discard_it;
145
146 family = ipver2af(ip_hdr(skb)->version);
147 af = sctp_get_af_specific(family);
148 if (unlikely(!af))
149 goto discard_it;
150
151 /* Initialize local addresses for lookups. */
152 af->from_skb(&src, skb, 1);
153 af->from_skb(&dest, skb, 0);
154
155 /* If the packet is to or from a non-unicast address,
156 * silently discard the packet.
157 *
158 * This is not clearly defined in the RFC except in section
159 * 8.4 - OOTB handling. However, based on the book "Stream Control
160 * Transmission Protocol" 2.1, "It is important to note that the
161 * IP address of an SCTP transport address must be a routable
162 * unicast address. In other words, IP multicast addresses and
163 * IP broadcast addresses cannot be used in an SCTP transport
164 * address."
165 */
166 if (!af->addr_valid(&src, NULL, skb) ||
167 !af->addr_valid(&dest, NULL, skb))
168 goto discard_it;
169
170 asoc = __sctp_rcv_lookup(net, skb, &src, &dest, &transport);
171
172 if (!asoc)
173 ep = __sctp_rcv_lookup_endpoint(net, &dest);
174
175 /* Retrieve the common input handling substructure. */
176 rcvr = asoc ? &asoc->base : &ep->base;
177 sk = rcvr->sk;
178
179 /*
180 * If a frame arrives on an interface and the receiving socket is
181 * bound to another interface, via SO_BINDTODEVICE, treat it as OOTB
182 */
183 if (sk->sk_bound_dev_if && (sk->sk_bound_dev_if != af->skb_iif(skb))) {
184 if (asoc) {
185 sctp_association_put(asoc);
186 asoc = NULL;
187 } else {
188 sctp_endpoint_put(ep);
189 ep = NULL;
190 }
191 sk = net->sctp.ctl_sock;
192 ep = sctp_sk(sk)->ep;
193 sctp_endpoint_hold(ep);
194 rcvr = &ep->base;
195 }
196
197 /*
198 * RFC 2960, 8.4 - Handle "Out of the blue" Packets.
199 * An SCTP packet is called an "out of the blue" (OOTB)
200 * packet if it is correctly formed, i.e., passed the
201 * receiver's checksum check, but the receiver is not
202 * able to identify the association to which this
203 * packet belongs.
204 */
205 if (!asoc) {
206 if (sctp_rcv_ootb(skb)) {
207 SCTP_INC_STATS_BH(net, SCTP_MIB_OUTOFBLUES);
208 goto discard_release;
209 }
210 }
211
212 if (!xfrm_policy_check(sk, XFRM_POLICY_IN, skb, family))
213 goto discard_release;
214 nf_reset(skb);
215
216 if (sk_filter(sk, skb))
217 goto discard_release;
218
219 /* Create an SCTP packet structure. */
220 chunk = sctp_chunkify(skb, asoc, sk);
221 if (!chunk)
222 goto discard_release;
223 SCTP_INPUT_CB(skb)->chunk = chunk;
224
225 /* Remember what endpoint is to handle this packet. */
226 chunk->rcvr = rcvr;
227
228 /* Remember the SCTP header. */
229 chunk->sctp_hdr = sh;
230
231 /* Set the source and destination addresses of the incoming chunk. */
232 sctp_init_addrs(chunk, &src, &dest);
233
234 /* Remember where we came from. */
235 chunk->transport = transport;
236
237 /* Acquire access to the sock lock. Note: We are safe from other
238 * bottom halves on this lock, but a user may be in the lock too,
239 * so check if it is busy.
240 */
241 bh_lock_sock(sk);
242
243 if (sk != rcvr->sk) {
244 /* Our cached sk is different from the rcvr->sk. This is
245 * because migrate()/accept() may have moved the association
246 * to a new socket and released all the sockets. So now we
247 * are holding a lock on the old socket while the user may
248 * be doing something with the new socket. Switch our veiw
249 * of the current sk.
250 */
251 bh_unlock_sock(sk);
252 sk = rcvr->sk;
253 bh_lock_sock(sk);
254 }
255
256 if (sock_owned_by_user(sk)) {
257 if (sctp_add_backlog(sk, skb)) {
258 bh_unlock_sock(sk);
259 sctp_chunk_free(chunk);
260 skb = NULL; /* sctp_chunk_free already freed the skb */
261 goto discard_release;
262 }
263 SCTP_INC_STATS_BH(net, SCTP_MIB_IN_PKT_BACKLOG);
264 } else {
265 SCTP_INC_STATS_BH(net, SCTP_MIB_IN_PKT_SOFTIRQ);
266 sctp_inq_push(&chunk->rcvr->inqueue, chunk);
267 }
268
269 bh_unlock_sock(sk);
270
271 /* Release the asoc/ep ref we took in the lookup calls. */
272 if (asoc)
273 sctp_association_put(asoc);
274 else
275 sctp_endpoint_put(ep);
276
277 return 0;
278
279discard_it:
280 SCTP_INC_STATS_BH(net, SCTP_MIB_IN_PKT_DISCARDS);
281 kfree_skb(skb);
282 return 0;
283
284discard_release:
285 /* Release the asoc/ep ref we took in the lookup calls. */
286 if (asoc)
287 sctp_association_put(asoc);
288 else
289 sctp_endpoint_put(ep);
290
291 goto discard_it;
292}
293
294/* Process the backlog queue of the socket. Every skb on
295 * the backlog holds a ref on an association or endpoint.
296 * We hold this ref throughout the state machine to make
297 * sure that the structure we need is still around.
298 */
299int sctp_backlog_rcv(struct sock *sk, struct sk_buff *skb)
300{
301 struct sctp_chunk *chunk = SCTP_INPUT_CB(skb)->chunk;
302 struct sctp_inq *inqueue = &chunk->rcvr->inqueue;
303 struct sctp_ep_common *rcvr = NULL;
304 int backloged = 0;
305
306 rcvr = chunk->rcvr;
307
308 /* If the rcvr is dead then the association or endpoint
309 * has been deleted and we can safely drop the chunk
310 * and refs that we are holding.
311 */
312 if (rcvr->dead) {
313 sctp_chunk_free(chunk);
314 goto done;
315 }
316
317 if (unlikely(rcvr->sk != sk)) {
318 /* In this case, the association moved from one socket to
319 * another. We are currently sitting on the backlog of the
320 * old socket, so we need to move.
321 * However, since we are here in the process context we
322 * need to take make sure that the user doesn't own
323 * the new socket when we process the packet.
324 * If the new socket is user-owned, queue the chunk to the
325 * backlog of the new socket without dropping any refs.
326 * Otherwise, we can safely push the chunk on the inqueue.
327 */
328
329 sk = rcvr->sk;
330 bh_lock_sock(sk);
331
332 if (sock_owned_by_user(sk)) {
333 if (sk_add_backlog(sk, skb, sk->sk_rcvbuf))
334 sctp_chunk_free(chunk);
335 else
336 backloged = 1;
337 } else
338 sctp_inq_push(inqueue, chunk);
339
340 bh_unlock_sock(sk);
341
342 /* If the chunk was backloged again, don't drop refs */
343 if (backloged)
344 return 0;
345 } else {
346 sctp_inq_push(inqueue, chunk);
347 }
348
349done:
350 /* Release the refs we took in sctp_add_backlog */
351 if (SCTP_EP_TYPE_ASSOCIATION == rcvr->type)
352 sctp_association_put(sctp_assoc(rcvr));
353 else if (SCTP_EP_TYPE_SOCKET == rcvr->type)
354 sctp_endpoint_put(sctp_ep(rcvr));
355 else
356 BUG();
357
358 return 0;
359}
360
361static int sctp_add_backlog(struct sock *sk, struct sk_buff *skb)
362{
363 struct sctp_chunk *chunk = SCTP_INPUT_CB(skb)->chunk;
364 struct sctp_ep_common *rcvr = chunk->rcvr;
365 int ret;
366
367 ret = sk_add_backlog(sk, skb, sk->sk_rcvbuf);
368 if (!ret) {
369 /* Hold the assoc/ep while hanging on the backlog queue.
370 * This way, we know structures we need will not disappear
371 * from us
372 */
373 if (SCTP_EP_TYPE_ASSOCIATION == rcvr->type)
374 sctp_association_hold(sctp_assoc(rcvr));
375 else if (SCTP_EP_TYPE_SOCKET == rcvr->type)
376 sctp_endpoint_hold(sctp_ep(rcvr));
377 else
378 BUG();
379 }
380 return ret;
381
382}
383
384/* Handle icmp frag needed error. */
385void sctp_icmp_frag_needed(struct sock *sk, struct sctp_association *asoc,
386 struct sctp_transport *t, __u32 pmtu)
387{
388 if (!t || (t->pathmtu <= pmtu))
389 return;
390
391 if (sock_owned_by_user(sk)) {
392 asoc->pmtu_pending = 1;
393 t->pmtu_pending = 1;
394 return;
395 }
396
397 if (t->param_flags & SPP_PMTUD_ENABLE) {
398 /* Update transports view of the MTU */
399 sctp_transport_update_pmtu(sk, t, pmtu);
400
401 /* Update association pmtu. */
402 sctp_assoc_sync_pmtu(sk, asoc);
403 }
404
405 /* Retransmit with the new pmtu setting.
406 * Normally, if PMTU discovery is disabled, an ICMP Fragmentation
407 * Needed will never be sent, but if a message was sent before
408 * PMTU discovery was disabled that was larger than the PMTU, it
409 * would not be fragmented, so it must be re-transmitted fragmented.
410 */
411 sctp_retransmit(&asoc->outqueue, t, SCTP_RTXR_PMTUD);
412}
413
414void sctp_icmp_redirect(struct sock *sk, struct sctp_transport *t,
415 struct sk_buff *skb)
416{
417 struct dst_entry *dst;
418
419 if (!t)
420 return;
421 dst = sctp_transport_dst_check(t);
422 if (dst)
423 dst->ops->redirect(dst, sk, skb);
424}
425
426/*
427 * SCTP Implementer's Guide, 2.37 ICMP handling procedures
428 *
429 * ICMP8) If the ICMP code is a "Unrecognized next header type encountered"
430 * or a "Protocol Unreachable" treat this message as an abort
431 * with the T bit set.
432 *
433 * This function sends an event to the state machine, which will abort the
434 * association.
435 *
436 */
437void sctp_icmp_proto_unreachable(struct sock *sk,
438 struct sctp_association *asoc,
439 struct sctp_transport *t)
440{
441 if (sock_owned_by_user(sk)) {
442 if (timer_pending(&t->proto_unreach_timer))
443 return;
444 else {
445 if (!mod_timer(&t->proto_unreach_timer,
446 jiffies + (HZ/20)))
447 sctp_association_hold(asoc);
448 }
449 } else {
450 struct net *net = sock_net(sk);
451
452 pr_debug("%s: unrecognized next header type "
453 "encountered!\n", __func__);
454
455 if (del_timer(&t->proto_unreach_timer))
456 sctp_association_put(asoc);
457
458 sctp_do_sm(net, SCTP_EVENT_T_OTHER,
459 SCTP_ST_OTHER(SCTP_EVENT_ICMP_PROTO_UNREACH),
460 asoc->state, asoc->ep, asoc, t,
461 GFP_ATOMIC);
462 }
463}
464
465/* Common lookup code for icmp/icmpv6 error handler. */
466struct sock *sctp_err_lookup(struct net *net, int family, struct sk_buff *skb,
467 struct sctphdr *sctphdr,
468 struct sctp_association **app,
469 struct sctp_transport **tpp)
470{
471 union sctp_addr saddr;
472 union sctp_addr daddr;
473 struct sctp_af *af;
474 struct sock *sk = NULL;
475 struct sctp_association *asoc;
476 struct sctp_transport *transport = NULL;
477 struct sctp_init_chunk *chunkhdr;
478 __u32 vtag = ntohl(sctphdr->vtag);
479 int len = skb->len - ((void *)sctphdr - (void *)skb->data);
480
481 *app = NULL; *tpp = NULL;
482
483 af = sctp_get_af_specific(family);
484 if (unlikely(!af)) {
485 return NULL;
486 }
487
488 /* Initialize local addresses for lookups. */
489 af->from_skb(&saddr, skb, 1);
490 af->from_skb(&daddr, skb, 0);
491
492 /* Look for an association that matches the incoming ICMP error
493 * packet.
494 */
495 asoc = __sctp_lookup_association(net, &saddr, &daddr, &transport);
496 if (!asoc)
497 return NULL;
498
499 sk = asoc->base.sk;
500
501 /* RFC 4960, Appendix C. ICMP Handling
502 *
503 * ICMP6) An implementation MUST validate that the Verification Tag
504 * contained in the ICMP message matches the Verification Tag of
505 * the peer. If the Verification Tag is not 0 and does NOT
506 * match, discard the ICMP message. If it is 0 and the ICMP
507 * message contains enough bytes to verify that the chunk type is
508 * an INIT chunk and that the Initiate Tag matches the tag of the
509 * peer, continue with ICMP7. If the ICMP message is too short
510 * or the chunk type or the Initiate Tag does not match, silently
511 * discard the packet.
512 */
513 if (vtag == 0) {
514 chunkhdr = (void *)sctphdr + sizeof(struct sctphdr);
515 if (len < sizeof(struct sctphdr) + sizeof(sctp_chunkhdr_t)
516 + sizeof(__be32) ||
517 chunkhdr->chunk_hdr.type != SCTP_CID_INIT ||
518 ntohl(chunkhdr->init_hdr.init_tag) != asoc->c.my_vtag) {
519 goto out;
520 }
521 } else if (vtag != asoc->c.peer_vtag) {
522 goto out;
523 }
524
525 bh_lock_sock(sk);
526
527 /* If too many ICMPs get dropped on busy
528 * servers this needs to be solved differently.
529 */
530 if (sock_owned_by_user(sk))
531 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
532
533 *app = asoc;
534 *tpp = transport;
535 return sk;
536
537out:
538 sctp_association_put(asoc);
539 return NULL;
540}
541
542/* Common cleanup code for icmp/icmpv6 error handler. */
543void sctp_err_finish(struct sock *sk, struct sctp_association *asoc)
544{
545 bh_unlock_sock(sk);
546 sctp_association_put(asoc);
547}
548
549/*
550 * This routine is called by the ICMP module when it gets some
551 * sort of error condition. If err < 0 then the socket should
552 * be closed and the error returned to the user. If err > 0
553 * it's just the icmp type << 8 | icmp code. After adjustment
554 * header points to the first 8 bytes of the sctp header. We need
555 * to find the appropriate port.
556 *
557 * The locking strategy used here is very "optimistic". When
558 * someone else accesses the socket the ICMP is just dropped
559 * and for some paths there is no check at all.
560 * A more general error queue to queue errors for later handling
561 * is probably better.
562 *
563 */
564void sctp_v4_err(struct sk_buff *skb, __u32 info)
565{
566 const struct iphdr *iph = (const struct iphdr *)skb->data;
567 const int ihlen = iph->ihl * 4;
568 const int type = icmp_hdr(skb)->type;
569 const int code = icmp_hdr(skb)->code;
570 struct sock *sk;
571 struct sctp_association *asoc = NULL;
572 struct sctp_transport *transport;
573 struct inet_sock *inet;
574 __u16 saveip, savesctp;
575 int err;
576 struct net *net = dev_net(skb->dev);
577
578 if (skb->len < ihlen + 8) {
579 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
580 return;
581 }
582
583 /* Fix up skb to look at the embedded net header. */
584 saveip = skb->network_header;
585 savesctp = skb->transport_header;
586 skb_reset_network_header(skb);
587 skb_set_transport_header(skb, ihlen);
588 sk = sctp_err_lookup(net, AF_INET, skb, sctp_hdr(skb), &asoc, &transport);
589 /* Put back, the original values. */
590 skb->network_header = saveip;
591 skb->transport_header = savesctp;
592 if (!sk) {
593 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
594 return;
595 }
596 /* Warning: The sock lock is held. Remember to call
597 * sctp_err_finish!
598 */
599
600 switch (type) {
601 case ICMP_PARAMETERPROB:
602 err = EPROTO;
603 break;
604 case ICMP_DEST_UNREACH:
605 if (code > NR_ICMP_UNREACH)
606 goto out_unlock;
607
608 /* PMTU discovery (RFC1191) */
609 if (ICMP_FRAG_NEEDED == code) {
610 sctp_icmp_frag_needed(sk, asoc, transport, info);
611 goto out_unlock;
612 } else {
613 if (ICMP_PROT_UNREACH == code) {
614 sctp_icmp_proto_unreachable(sk, asoc,
615 transport);
616 goto out_unlock;
617 }
618 }
619 err = icmp_err_convert[code].errno;
620 break;
621 case ICMP_TIME_EXCEEDED:
622 /* Ignore any time exceeded errors due to fragment reassembly
623 * timeouts.
624 */
625 if (ICMP_EXC_FRAGTIME == code)
626 goto out_unlock;
627
628 err = EHOSTUNREACH;
629 break;
630 case ICMP_REDIRECT:
631 sctp_icmp_redirect(sk, transport, skb);
632 /* Fall through to out_unlock. */
633 default:
634 goto out_unlock;
635 }
636
637 inet = inet_sk(sk);
638 if (!sock_owned_by_user(sk) && inet->recverr) {
639 sk->sk_err = err;
640 sk->sk_error_report(sk);
641 } else { /* Only an error on timeout */
642 sk->sk_err_soft = err;
643 }
644
645out_unlock:
646 sctp_err_finish(sk, asoc);
647}
648
649/*
650 * RFC 2960, 8.4 - Handle "Out of the blue" Packets.
651 *
652 * This function scans all the chunks in the OOTB packet to determine if
653 * the packet should be discarded right away. If a response might be needed
654 * for this packet, or, if further processing is possible, the packet will
655 * be queued to a proper inqueue for the next phase of handling.
656 *
657 * Output:
658 * Return 0 - If further processing is needed.
659 * Return 1 - If the packet can be discarded right away.
660 */
661static int sctp_rcv_ootb(struct sk_buff *skb)
662{
663 sctp_chunkhdr_t *ch;
664 __u8 *ch_end;
665
666 ch = (sctp_chunkhdr_t *) skb->data;
667
668 /* Scan through all the chunks in the packet. */
669 do {
670 /* Break out if chunk length is less then minimal. */
671 if (ntohs(ch->length) < sizeof(sctp_chunkhdr_t))
672 break;
673
674 ch_end = ((__u8 *)ch) + WORD_ROUND(ntohs(ch->length));
675 if (ch_end > skb_tail_pointer(skb))
676 break;
677
678 /* RFC 8.4, 2) If the OOTB packet contains an ABORT chunk, the
679 * receiver MUST silently discard the OOTB packet and take no
680 * further action.
681 */
682 if (SCTP_CID_ABORT == ch->type)
683 goto discard;
684
685 /* RFC 8.4, 6) If the packet contains a SHUTDOWN COMPLETE
686 * chunk, the receiver should silently discard the packet
687 * and take no further action.
688 */
689 if (SCTP_CID_SHUTDOWN_COMPLETE == ch->type)
690 goto discard;
691
692 /* RFC 4460, 2.11.2
693 * This will discard packets with INIT chunk bundled as
694 * subsequent chunks in the packet. When INIT is first,
695 * the normal INIT processing will discard the chunk.
696 */
697 if (SCTP_CID_INIT == ch->type && (void *)ch != skb->data)
698 goto discard;
699
700 ch = (sctp_chunkhdr_t *) ch_end;
701 } while (ch_end < skb_tail_pointer(skb));
702
703 return 0;
704
705discard:
706 return 1;
707}
708
709/* Insert endpoint into the hash table. */
710static void __sctp_hash_endpoint(struct sctp_endpoint *ep)
711{
712 struct net *net = sock_net(ep->base.sk);
713 struct sctp_ep_common *epb;
714 struct sctp_hashbucket *head;
715
716 epb = &ep->base;
717
718 epb->hashent = sctp_ep_hashfn(net, epb->bind_addr.port);
719 head = &sctp_ep_hashtable[epb->hashent];
720
721 write_lock(&head->lock);
722 hlist_add_head(&epb->node, &head->chain);
723 write_unlock(&head->lock);
724}
725
726/* Add an endpoint to the hash. Local BH-safe. */
727void sctp_hash_endpoint(struct sctp_endpoint *ep)
728{
729 local_bh_disable();
730 __sctp_hash_endpoint(ep);
731 local_bh_enable();
732}
733
734/* Remove endpoint from the hash table. */
735static void __sctp_unhash_endpoint(struct sctp_endpoint *ep)
736{
737 struct net *net = sock_net(ep->base.sk);
738 struct sctp_hashbucket *head;
739 struct sctp_ep_common *epb;
740
741 epb = &ep->base;
742
743 epb->hashent = sctp_ep_hashfn(net, epb->bind_addr.port);
744
745 head = &sctp_ep_hashtable[epb->hashent];
746
747 write_lock(&head->lock);
748 hlist_del_init(&epb->node);
749 write_unlock(&head->lock);
750}
751
752/* Remove endpoint from the hash. Local BH-safe. */
753void sctp_unhash_endpoint(struct sctp_endpoint *ep)
754{
755 local_bh_disable();
756 __sctp_unhash_endpoint(ep);
757 local_bh_enable();
758}
759
760/* Look up an endpoint. */
761static struct sctp_endpoint *__sctp_rcv_lookup_endpoint(struct net *net,
762 const union sctp_addr *laddr)
763{
764 struct sctp_hashbucket *head;
765 struct sctp_ep_common *epb;
766 struct sctp_endpoint *ep;
767 int hash;
768
769 hash = sctp_ep_hashfn(net, ntohs(laddr->v4.sin_port));
770 head = &sctp_ep_hashtable[hash];
771 read_lock(&head->lock);
772 sctp_for_each_hentry(epb, &head->chain) {
773 ep = sctp_ep(epb);
774 if (sctp_endpoint_is_match(ep, net, laddr))
775 goto hit;
776 }
777
778 ep = sctp_sk(net->sctp.ctl_sock)->ep;
779
780hit:
781 sctp_endpoint_hold(ep);
782 read_unlock(&head->lock);
783 return ep;
784}
785
786/* Insert association into the hash table. */
787static void __sctp_hash_established(struct sctp_association *asoc)
788{
789 struct net *net = sock_net(asoc->base.sk);
790 struct sctp_ep_common *epb;
791 struct sctp_hashbucket *head;
792
793 epb = &asoc->base;
794
795 /* Calculate which chain this entry will belong to. */
796 epb->hashent = sctp_assoc_hashfn(net, epb->bind_addr.port,
797 asoc->peer.port);
798
799 head = &sctp_assoc_hashtable[epb->hashent];
800
801 write_lock(&head->lock);
802 hlist_add_head(&epb->node, &head->chain);
803 write_unlock(&head->lock);
804}
805
806/* Add an association to the hash. Local BH-safe. */
807void sctp_hash_established(struct sctp_association *asoc)
808{
809 if (asoc->temp)
810 return;
811
812 local_bh_disable();
813 __sctp_hash_established(asoc);
814 local_bh_enable();
815}
816
817/* Remove association from the hash table. */
818static void __sctp_unhash_established(struct sctp_association *asoc)
819{
820 struct net *net = sock_net(asoc->base.sk);
821 struct sctp_hashbucket *head;
822 struct sctp_ep_common *epb;
823
824 epb = &asoc->base;
825
826 epb->hashent = sctp_assoc_hashfn(net, epb->bind_addr.port,
827 asoc->peer.port);
828
829 head = &sctp_assoc_hashtable[epb->hashent];
830
831 write_lock(&head->lock);
832 hlist_del_init(&epb->node);
833 write_unlock(&head->lock);
834}
835
836/* Remove association from the hash table. Local BH-safe. */
837void sctp_unhash_established(struct sctp_association *asoc)
838{
839 if (asoc->temp)
840 return;
841
842 local_bh_disable();
843 __sctp_unhash_established(asoc);
844 local_bh_enable();
845}
846
847/* Look up an association. */
848static struct sctp_association *__sctp_lookup_association(
849 struct net *net,
850 const union sctp_addr *local,
851 const union sctp_addr *peer,
852 struct sctp_transport **pt)
853{
854 struct sctp_hashbucket *head;
855 struct sctp_ep_common *epb;
856 struct sctp_association *asoc;
857 struct sctp_transport *transport;
858 int hash;
859
860 /* Optimize here for direct hit, only listening connections can
861 * have wildcards anyways.
862 */
863 hash = sctp_assoc_hashfn(net, ntohs(local->v4.sin_port),
864 ntohs(peer->v4.sin_port));
865 head = &sctp_assoc_hashtable[hash];
866 read_lock(&head->lock);
867 sctp_for_each_hentry(epb, &head->chain) {
868 asoc = sctp_assoc(epb);
869 transport = sctp_assoc_is_match(asoc, net, local, peer);
870 if (transport)
871 goto hit;
872 }
873
874 read_unlock(&head->lock);
875
876 return NULL;
877
878hit:
879 *pt = transport;
880 sctp_association_hold(asoc);
881 read_unlock(&head->lock);
882 return asoc;
883}
884
885/* Look up an association. BH-safe. */
886static
887struct sctp_association *sctp_lookup_association(struct net *net,
888 const union sctp_addr *laddr,
889 const union sctp_addr *paddr,
890 struct sctp_transport **transportp)
891{
892 struct sctp_association *asoc;
893
894 local_bh_disable();
895 asoc = __sctp_lookup_association(net, laddr, paddr, transportp);
896 local_bh_enable();
897
898 return asoc;
899}
900
901/* Is there an association matching the given local and peer addresses? */
902int sctp_has_association(struct net *net,
903 const union sctp_addr *laddr,
904 const union sctp_addr *paddr)
905{
906 struct sctp_association *asoc;
907 struct sctp_transport *transport;
908
909 if ((asoc = sctp_lookup_association(net, laddr, paddr, &transport))) {
910 sctp_association_put(asoc);
911 return 1;
912 }
913
914 return 0;
915}
916
917/*
918 * SCTP Implementors Guide, 2.18 Handling of address
919 * parameters within the INIT or INIT-ACK.
920 *
921 * D) When searching for a matching TCB upon reception of an INIT
922 * or INIT-ACK chunk the receiver SHOULD use not only the
923 * source address of the packet (containing the INIT or
924 * INIT-ACK) but the receiver SHOULD also use all valid
925 * address parameters contained within the chunk.
926 *
927 * 2.18.3 Solution description
928 *
929 * This new text clearly specifies to an implementor the need
930 * to look within the INIT or INIT-ACK. Any implementation that
931 * does not do this, may not be able to establish associations
932 * in certain circumstances.
933 *
934 */
935static struct sctp_association *__sctp_rcv_init_lookup(struct net *net,
936 struct sk_buff *skb,
937 const union sctp_addr *laddr, struct sctp_transport **transportp)
938{
939 struct sctp_association *asoc;
940 union sctp_addr addr;
941 union sctp_addr *paddr = &addr;
942 struct sctphdr *sh = sctp_hdr(skb);
943 union sctp_params params;
944 sctp_init_chunk_t *init;
945 struct sctp_transport *transport;
946 struct sctp_af *af;
947
948 /*
949 * This code will NOT touch anything inside the chunk--it is
950 * strictly READ-ONLY.
951 *
952 * RFC 2960 3 SCTP packet Format
953 *
954 * Multiple chunks can be bundled into one SCTP packet up to
955 * the MTU size, except for the INIT, INIT ACK, and SHUTDOWN
956 * COMPLETE chunks. These chunks MUST NOT be bundled with any
957 * other chunk in a packet. See Section 6.10 for more details
958 * on chunk bundling.
959 */
960
961 /* Find the start of the TLVs and the end of the chunk. This is
962 * the region we search for address parameters.
963 */
964 init = (sctp_init_chunk_t *)skb->data;
965
966 /* Walk the parameters looking for embedded addresses. */
967 sctp_walk_params(params, init, init_hdr.params) {
968
969 /* Note: Ignoring hostname addresses. */
970 af = sctp_get_af_specific(param_type2af(params.p->type));
971 if (!af)
972 continue;
973
974 af->from_addr_param(paddr, params.addr, sh->source, 0);
975
976 asoc = __sctp_lookup_association(net, laddr, paddr, &transport);
977 if (asoc)
978 return asoc;
979 }
980
981 return NULL;
982}
983
984/* ADD-IP, Section 5.2
985 * When an endpoint receives an ASCONF Chunk from the remote peer
986 * special procedures may be needed to identify the association the
987 * ASCONF Chunk is associated with. To properly find the association
988 * the following procedures SHOULD be followed:
989 *
990 * D2) If the association is not found, use the address found in the
991 * Address Parameter TLV combined with the port number found in the
992 * SCTP common header. If found proceed to rule D4.
993 *
994 * D2-ext) If more than one ASCONF Chunks are packed together, use the
995 * address found in the ASCONF Address Parameter TLV of each of the
996 * subsequent ASCONF Chunks. If found, proceed to rule D4.
997 */
998static struct sctp_association *__sctp_rcv_asconf_lookup(
999 struct net *net,
1000 sctp_chunkhdr_t *ch,
1001 const union sctp_addr *laddr,
1002 __be16 peer_port,
1003 struct sctp_transport **transportp)
1004{
1005 sctp_addip_chunk_t *asconf = (struct sctp_addip_chunk *)ch;
1006 struct sctp_af *af;
1007 union sctp_addr_param *param;
1008 union sctp_addr paddr;
1009
1010 /* Skip over the ADDIP header and find the Address parameter */
1011 param = (union sctp_addr_param *)(asconf + 1);
1012
1013 af = sctp_get_af_specific(param_type2af(param->p.type));
1014 if (unlikely(!af))
1015 return NULL;
1016
1017 af->from_addr_param(&paddr, param, peer_port, 0);
1018
1019 return __sctp_lookup_association(net, laddr, &paddr, transportp);
1020}
1021
1022
1023/* SCTP-AUTH, Section 6.3:
1024* If the receiver does not find a STCB for a packet containing an AUTH
1025* chunk as the first chunk and not a COOKIE-ECHO chunk as the second
1026* chunk, it MUST use the chunks after the AUTH chunk to look up an existing
1027* association.
1028*
1029* This means that any chunks that can help us identify the association need
1030* to be looked at to find this association.
1031*/
1032static struct sctp_association *__sctp_rcv_walk_lookup(struct net *net,
1033 struct sk_buff *skb,
1034 const union sctp_addr *laddr,
1035 struct sctp_transport **transportp)
1036{
1037 struct sctp_association *asoc = NULL;
1038 sctp_chunkhdr_t *ch;
1039 int have_auth = 0;
1040 unsigned int chunk_num = 1;
1041 __u8 *ch_end;
1042
1043 /* Walk through the chunks looking for AUTH or ASCONF chunks
1044 * to help us find the association.
1045 */
1046 ch = (sctp_chunkhdr_t *) skb->data;
1047 do {
1048 /* Break out if chunk length is less then minimal. */
1049 if (ntohs(ch->length) < sizeof(sctp_chunkhdr_t))
1050 break;
1051
1052 ch_end = ((__u8 *)ch) + WORD_ROUND(ntohs(ch->length));
1053 if (ch_end > skb_tail_pointer(skb))
1054 break;
1055
1056 switch (ch->type) {
1057 case SCTP_CID_AUTH:
1058 have_auth = chunk_num;
1059 break;
1060
1061 case SCTP_CID_COOKIE_ECHO:
1062 /* If a packet arrives containing an AUTH chunk as
1063 * a first chunk, a COOKIE-ECHO chunk as the second
1064 * chunk, and possibly more chunks after them, and
1065 * the receiver does not have an STCB for that
1066 * packet, then authentication is based on
1067 * the contents of the COOKIE- ECHO chunk.
1068 */
1069 if (have_auth == 1 && chunk_num == 2)
1070 return NULL;
1071 break;
1072
1073 case SCTP_CID_ASCONF:
1074 if (have_auth || net->sctp.addip_noauth)
1075 asoc = __sctp_rcv_asconf_lookup(
1076 net, ch, laddr,
1077 sctp_hdr(skb)->source,
1078 transportp);
1079 default:
1080 break;
1081 }
1082
1083 if (asoc)
1084 break;
1085
1086 ch = (sctp_chunkhdr_t *) ch_end;
1087 chunk_num++;
1088 } while (ch_end < skb_tail_pointer(skb));
1089
1090 return asoc;
1091}
1092
1093/*
1094 * There are circumstances when we need to look inside the SCTP packet
1095 * for information to help us find the association. Examples
1096 * include looking inside of INIT/INIT-ACK chunks or after the AUTH
1097 * chunks.
1098 */
1099static struct sctp_association *__sctp_rcv_lookup_harder(struct net *net,
1100 struct sk_buff *skb,
1101 const union sctp_addr *laddr,
1102 struct sctp_transport **transportp)
1103{
1104 sctp_chunkhdr_t *ch;
1105
1106 ch = (sctp_chunkhdr_t *) skb->data;
1107
1108 /* The code below will attempt to walk the chunk and extract
1109 * parameter information. Before we do that, we need to verify
1110 * that the chunk length doesn't cause overflow. Otherwise, we'll
1111 * walk off the end.
1112 */
1113 if (WORD_ROUND(ntohs(ch->length)) > skb->len)
1114 return NULL;
1115
1116 /* If this is INIT/INIT-ACK look inside the chunk too. */
1117 if (ch->type == SCTP_CID_INIT || ch->type == SCTP_CID_INIT_ACK)
1118 return __sctp_rcv_init_lookup(net, skb, laddr, transportp);
1119
1120 return __sctp_rcv_walk_lookup(net, skb, laddr, transportp);
1121}
1122
1123/* Lookup an association for an inbound skb. */
1124static struct sctp_association *__sctp_rcv_lookup(struct net *net,
1125 struct sk_buff *skb,
1126 const union sctp_addr *paddr,
1127 const union sctp_addr *laddr,
1128 struct sctp_transport **transportp)
1129{
1130 struct sctp_association *asoc;
1131
1132 asoc = __sctp_lookup_association(net, laddr, paddr, transportp);
1133
1134 /* Further lookup for INIT/INIT-ACK packets.
1135 * SCTP Implementors Guide, 2.18 Handling of address
1136 * parameters within the INIT or INIT-ACK.
1137 */
1138 if (!asoc)
1139 asoc = __sctp_rcv_lookup_harder(net, skb, laddr, transportp);
1140
1141 return asoc;
1142}