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,
54 int dif, int sdif);
55static struct sctp_endpoint *__sctp_rcv_lookup_endpoint(
56 struct net *net, struct sk_buff *skb,
57 const union sctp_addr *laddr,
58 const union sctp_addr *daddr,
59 int dif, int sdif);
60static struct sctp_association *__sctp_lookup_association(
61 struct net *net,
62 const union sctp_addr *local,
63 const union sctp_addr *peer,
64 struct sctp_transport **pt,
65 int dif, int sdif);
66
67static int sctp_add_backlog(struct sock *sk, struct sk_buff *skb);
68
69
70/* Calculate the SCTP checksum of an SCTP packet. */
71static inline int sctp_rcv_checksum(struct net *net, struct sk_buff *skb)
72{
73 struct sctphdr *sh = sctp_hdr(skb);
74 __le32 cmp = sh->checksum;
75 __le32 val = sctp_compute_cksum(skb, 0);
76
77 if (val != cmp) {
78 /* CRC failure, dump it. */
79 __SCTP_INC_STATS(net, SCTP_MIB_CHECKSUMERRORS);
80 return -1;
81 }
82 return 0;
83}
84
85/*
86 * This is the routine which IP calls when receiving an SCTP packet.
87 */
88int sctp_rcv(struct sk_buff *skb)
89{
90 struct sock *sk;
91 struct sctp_association *asoc;
92 struct sctp_endpoint *ep = NULL;
93 struct sctp_ep_common *rcvr;
94 struct sctp_transport *transport = NULL;
95 struct sctp_chunk *chunk;
96 union sctp_addr src;
97 union sctp_addr dest;
98 int family;
99 struct sctp_af *af;
100 struct net *net = dev_net(skb->dev);
101 bool is_gso = skb_is_gso(skb) && skb_is_gso_sctp(skb);
102 int dif, sdif;
103
104 if (skb->pkt_type != PACKET_HOST)
105 goto discard_it;
106
107 __SCTP_INC_STATS(net, SCTP_MIB_INSCTPPACKS);
108
109 /* If packet is too small to contain a single chunk, let's not
110 * waste time on it anymore.
111 */
112 if (skb->len < sizeof(struct sctphdr) + sizeof(struct sctp_chunkhdr) +
113 skb_transport_offset(skb))
114 goto discard_it;
115
116 /* If the packet is fragmented and we need to do crc checking,
117 * it's better to just linearize it otherwise crc computing
118 * takes longer.
119 */
120 if ((!is_gso && skb_linearize(skb)) ||
121 !pskb_may_pull(skb, sizeof(struct sctphdr)))
122 goto discard_it;
123
124 /* Pull up the IP header. */
125 __skb_pull(skb, skb_transport_offset(skb));
126
127 skb->csum_valid = 0; /* Previous value not applicable */
128 if (skb_csum_unnecessary(skb))
129 __skb_decr_checksum_unnecessary(skb);
130 else if (!sctp_checksum_disable &&
131 !is_gso &&
132 sctp_rcv_checksum(net, skb) < 0)
133 goto discard_it;
134 skb->csum_valid = 1;
135
136 __skb_pull(skb, sizeof(struct sctphdr));
137
138 family = ipver2af(ip_hdr(skb)->version);
139 af = sctp_get_af_specific(family);
140 if (unlikely(!af))
141 goto discard_it;
142 SCTP_INPUT_CB(skb)->af = af;
143
144 /* Initialize local addresses for lookups. */
145 af->from_skb(&src, skb, 1);
146 af->from_skb(&dest, skb, 0);
147 dif = af->skb_iif(skb);
148 sdif = af->skb_sdif(skb);
149
150 /* If the packet is to or from a non-unicast address,
151 * silently discard the packet.
152 *
153 * This is not clearly defined in the RFC except in section
154 * 8.4 - OOTB handling. However, based on the book "Stream Control
155 * Transmission Protocol" 2.1, "It is important to note that the
156 * IP address of an SCTP transport address must be a routable
157 * unicast address. In other words, IP multicast addresses and
158 * IP broadcast addresses cannot be used in an SCTP transport
159 * address."
160 */
161 if (!af->addr_valid(&src, NULL, skb) ||
162 !af->addr_valid(&dest, NULL, skb))
163 goto discard_it;
164
165 asoc = __sctp_rcv_lookup(net, skb, &src, &dest, &transport, dif, sdif);
166
167 if (!asoc)
168 ep = __sctp_rcv_lookup_endpoint(net, skb, &dest, &src, dif, sdif);
169
170 /* Retrieve the common input handling substructure. */
171 rcvr = asoc ? &asoc->base : &ep->base;
172 sk = rcvr->sk;
173
174 /*
175 * RFC 2960, 8.4 - Handle "Out of the blue" Packets.
176 * An SCTP packet is called an "out of the blue" (OOTB)
177 * packet if it is correctly formed, i.e., passed the
178 * receiver's checksum check, but the receiver is not
179 * able to identify the association to which this
180 * packet belongs.
181 */
182 if (!asoc) {
183 if (sctp_rcv_ootb(skb)) {
184 __SCTP_INC_STATS(net, SCTP_MIB_OUTOFBLUES);
185 goto discard_release;
186 }
187 }
188
189 if (!xfrm_policy_check(sk, XFRM_POLICY_IN, skb, family))
190 goto discard_release;
191 nf_reset_ct(skb);
192
193 if (sk_filter(sk, skb))
194 goto discard_release;
195
196 /* Create an SCTP packet structure. */
197 chunk = sctp_chunkify(skb, asoc, sk, GFP_ATOMIC);
198 if (!chunk)
199 goto discard_release;
200 SCTP_INPUT_CB(skb)->chunk = chunk;
201
202 /* Remember what endpoint is to handle this packet. */
203 chunk->rcvr = rcvr;
204
205 /* Remember the SCTP header. */
206 chunk->sctp_hdr = sctp_hdr(skb);
207
208 /* Set the source and destination addresses of the incoming chunk. */
209 sctp_init_addrs(chunk, &src, &dest);
210
211 /* Remember where we came from. */
212 chunk->transport = transport;
213
214 /* Acquire access to the sock lock. Note: We are safe from other
215 * bottom halves on this lock, but a user may be in the lock too,
216 * so check if it is busy.
217 */
218 bh_lock_sock(sk);
219
220 if (sk != rcvr->sk) {
221 /* Our cached sk is different from the rcvr->sk. This is
222 * because migrate()/accept() may have moved the association
223 * to a new socket and released all the sockets. So now we
224 * are holding a lock on the old socket while the user may
225 * be doing something with the new socket. Switch our veiw
226 * of the current sk.
227 */
228 bh_unlock_sock(sk);
229 sk = rcvr->sk;
230 bh_lock_sock(sk);
231 }
232
233 if (sock_owned_by_user(sk) || !sctp_newsk_ready(sk)) {
234 if (sctp_add_backlog(sk, skb)) {
235 bh_unlock_sock(sk);
236 sctp_chunk_free(chunk);
237 skb = NULL; /* sctp_chunk_free already freed the skb */
238 goto discard_release;
239 }
240 __SCTP_INC_STATS(net, SCTP_MIB_IN_PKT_BACKLOG);
241 } else {
242 __SCTP_INC_STATS(net, SCTP_MIB_IN_PKT_SOFTIRQ);
243 sctp_inq_push(&chunk->rcvr->inqueue, chunk);
244 }
245
246 bh_unlock_sock(sk);
247
248 /* Release the asoc/ep ref we took in the lookup calls. */
249 if (transport)
250 sctp_transport_put(transport);
251 else
252 sctp_endpoint_put(ep);
253
254 return 0;
255
256discard_it:
257 __SCTP_INC_STATS(net, SCTP_MIB_IN_PKT_DISCARDS);
258 kfree_skb(skb);
259 return 0;
260
261discard_release:
262 /* Release the asoc/ep ref we took in the lookup calls. */
263 if (transport)
264 sctp_transport_put(transport);
265 else
266 sctp_endpoint_put(ep);
267
268 goto discard_it;
269}
270
271/* Process the backlog queue of the socket. Every skb on
272 * the backlog holds a ref on an association or endpoint.
273 * We hold this ref throughout the state machine to make
274 * sure that the structure we need is still around.
275 */
276int sctp_backlog_rcv(struct sock *sk, struct sk_buff *skb)
277{
278 struct sctp_chunk *chunk = SCTP_INPUT_CB(skb)->chunk;
279 struct sctp_inq *inqueue = &chunk->rcvr->inqueue;
280 struct sctp_transport *t = chunk->transport;
281 struct sctp_ep_common *rcvr = NULL;
282 int backloged = 0;
283
284 rcvr = chunk->rcvr;
285
286 /* If the rcvr is dead then the association or endpoint
287 * has been deleted and we can safely drop the chunk
288 * and refs that we are holding.
289 */
290 if (rcvr->dead) {
291 sctp_chunk_free(chunk);
292 goto done;
293 }
294
295 if (unlikely(rcvr->sk != sk)) {
296 /* In this case, the association moved from one socket to
297 * another. We are currently sitting on the backlog of the
298 * old socket, so we need to move.
299 * However, since we are here in the process context we
300 * need to take make sure that the user doesn't own
301 * the new socket when we process the packet.
302 * If the new socket is user-owned, queue the chunk to the
303 * backlog of the new socket without dropping any refs.
304 * Otherwise, we can safely push the chunk on the inqueue.
305 */
306
307 sk = rcvr->sk;
308 local_bh_disable();
309 bh_lock_sock(sk);
310
311 if (sock_owned_by_user(sk) || !sctp_newsk_ready(sk)) {
312 if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf)))
313 sctp_chunk_free(chunk);
314 else
315 backloged = 1;
316 } else
317 sctp_inq_push(inqueue, chunk);
318
319 bh_unlock_sock(sk);
320 local_bh_enable();
321
322 /* If the chunk was backloged again, don't drop refs */
323 if (backloged)
324 return 0;
325 } else {
326 if (!sctp_newsk_ready(sk)) {
327 if (!sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf)))
328 return 0;
329 sctp_chunk_free(chunk);
330 } else {
331 sctp_inq_push(inqueue, chunk);
332 }
333 }
334
335done:
336 /* Release the refs we took in sctp_add_backlog */
337 if (SCTP_EP_TYPE_ASSOCIATION == rcvr->type)
338 sctp_transport_put(t);
339 else if (SCTP_EP_TYPE_SOCKET == rcvr->type)
340 sctp_endpoint_put(sctp_ep(rcvr));
341 else
342 BUG();
343
344 return 0;
345}
346
347static int sctp_add_backlog(struct sock *sk, struct sk_buff *skb)
348{
349 struct sctp_chunk *chunk = SCTP_INPUT_CB(skb)->chunk;
350 struct sctp_transport *t = chunk->transport;
351 struct sctp_ep_common *rcvr = chunk->rcvr;
352 int ret;
353
354 ret = sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf));
355 if (!ret) {
356 /* Hold the assoc/ep while hanging on the backlog queue.
357 * This way, we know structures we need will not disappear
358 * from us
359 */
360 if (SCTP_EP_TYPE_ASSOCIATION == rcvr->type)
361 sctp_transport_hold(t);
362 else if (SCTP_EP_TYPE_SOCKET == rcvr->type)
363 sctp_endpoint_hold(sctp_ep(rcvr));
364 else
365 BUG();
366 }
367 return ret;
368
369}
370
371/* Handle icmp frag needed error. */
372void sctp_icmp_frag_needed(struct sock *sk, struct sctp_association *asoc,
373 struct sctp_transport *t, __u32 pmtu)
374{
375 if (!t ||
376 (t->pathmtu <= pmtu &&
377 t->pl.probe_size + sctp_transport_pl_hlen(t) <= pmtu))
378 return;
379
380 if (sock_owned_by_user(sk)) {
381 atomic_set(&t->mtu_info, pmtu);
382 asoc->pmtu_pending = 1;
383 t->pmtu_pending = 1;
384 return;
385 }
386
387 if (!(t->param_flags & SPP_PMTUD_ENABLE))
388 /* We can't allow retransmitting in such case, as the
389 * retransmission would be sized just as before, and thus we
390 * would get another icmp, and retransmit again.
391 */
392 return;
393
394 /* Update transports view of the MTU. Return if no update was needed.
395 * If an update wasn't needed/possible, it also doesn't make sense to
396 * try to retransmit now.
397 */
398 if (!sctp_transport_update_pmtu(t, pmtu))
399 return;
400
401 /* Update association pmtu. */
402 sctp_assoc_sync_pmtu(asoc);
403
404 /* Retransmit with the new pmtu setting. */
405 sctp_retransmit(&asoc->outqueue, t, SCTP_RTXR_PMTUD);
406}
407
408void sctp_icmp_redirect(struct sock *sk, struct sctp_transport *t,
409 struct sk_buff *skb)
410{
411 struct dst_entry *dst;
412
413 if (sock_owned_by_user(sk) || !t)
414 return;
415 dst = sctp_transport_dst_check(t);
416 if (dst)
417 dst->ops->redirect(dst, sk, skb);
418}
419
420/*
421 * SCTP Implementer's Guide, 2.37 ICMP handling procedures
422 *
423 * ICMP8) If the ICMP code is a "Unrecognized next header type encountered"
424 * or a "Protocol Unreachable" treat this message as an abort
425 * with the T bit set.
426 *
427 * This function sends an event to the state machine, which will abort the
428 * association.
429 *
430 */
431void sctp_icmp_proto_unreachable(struct sock *sk,
432 struct sctp_association *asoc,
433 struct sctp_transport *t)
434{
435 if (sock_owned_by_user(sk)) {
436 if (timer_pending(&t->proto_unreach_timer))
437 return;
438 else {
439 if (!mod_timer(&t->proto_unreach_timer,
440 jiffies + (HZ/20)))
441 sctp_transport_hold(t);
442 }
443 } else {
444 struct net *net = sock_net(sk);
445
446 pr_debug("%s: unrecognized next header type "
447 "encountered!\n", __func__);
448
449 if (del_timer(&t->proto_unreach_timer))
450 sctp_transport_put(t);
451
452 sctp_do_sm(net, SCTP_EVENT_T_OTHER,
453 SCTP_ST_OTHER(SCTP_EVENT_ICMP_PROTO_UNREACH),
454 asoc->state, asoc->ep, asoc, t,
455 GFP_ATOMIC);
456 }
457}
458
459/* Common lookup code for icmp/icmpv6 error handler. */
460struct sock *sctp_err_lookup(struct net *net, int family, struct sk_buff *skb,
461 struct sctphdr *sctphdr,
462 struct sctp_association **app,
463 struct sctp_transport **tpp)
464{
465 struct sctp_init_chunk *chunkhdr, _chunkhdr;
466 union sctp_addr saddr;
467 union sctp_addr daddr;
468 struct sctp_af *af;
469 struct sock *sk = NULL;
470 struct sctp_association *asoc;
471 struct sctp_transport *transport = NULL;
472 __u32 vtag = ntohl(sctphdr->vtag);
473 int sdif = inet_sdif(skb);
474 int dif = inet_iif(skb);
475
476 *app = NULL; *tpp = NULL;
477
478 af = sctp_get_af_specific(family);
479 if (unlikely(!af)) {
480 return NULL;
481 }
482
483 /* Initialize local addresses for lookups. */
484 af->from_skb(&saddr, skb, 1);
485 af->from_skb(&daddr, skb, 0);
486
487 /* Look for an association that matches the incoming ICMP error
488 * packet.
489 */
490 asoc = __sctp_lookup_association(net, &saddr, &daddr, &transport, dif, sdif);
491 if (!asoc)
492 return NULL;
493
494 sk = asoc->base.sk;
495
496 /* RFC 4960, Appendix C. ICMP Handling
497 *
498 * ICMP6) An implementation MUST validate that the Verification Tag
499 * contained in the ICMP message matches the Verification Tag of
500 * the peer. If the Verification Tag is not 0 and does NOT
501 * match, discard the ICMP message. If it is 0 and the ICMP
502 * message contains enough bytes to verify that the chunk type is
503 * an INIT chunk and that the Initiate Tag matches the tag of the
504 * peer, continue with ICMP7. If the ICMP message is too short
505 * or the chunk type or the Initiate Tag does not match, silently
506 * discard the packet.
507 */
508 if (vtag == 0) {
509 /* chunk header + first 4 octects of init header */
510 chunkhdr = skb_header_pointer(skb, skb_transport_offset(skb) +
511 sizeof(struct sctphdr),
512 sizeof(struct sctp_chunkhdr) +
513 sizeof(__be32), &_chunkhdr);
514 if (!chunkhdr ||
515 chunkhdr->chunk_hdr.type != SCTP_CID_INIT ||
516 ntohl(chunkhdr->init_hdr.init_tag) != asoc->c.my_vtag)
517 goto out;
518
519 } else if (vtag != asoc->c.peer_vtag) {
520 goto out;
521 }
522
523 bh_lock_sock(sk);
524
525 /* If too many ICMPs get dropped on busy
526 * servers this needs to be solved differently.
527 */
528 if (sock_owned_by_user(sk))
529 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
530
531 *app = asoc;
532 *tpp = transport;
533 return sk;
534
535out:
536 sctp_transport_put(transport);
537 return NULL;
538}
539
540/* Common cleanup code for icmp/icmpv6 error handler. */
541void sctp_err_finish(struct sock *sk, struct sctp_transport *t)
542 __releases(&((__sk)->sk_lock.slock))
543{
544 bh_unlock_sock(sk);
545 sctp_transport_put(t);
546}
547
548static void sctp_v4_err_handle(struct sctp_transport *t, struct sk_buff *skb,
549 __u8 type, __u8 code, __u32 info)
550{
551 struct sctp_association *asoc = t->asoc;
552 struct sock *sk = asoc->base.sk;
553 int err = 0;
554
555 switch (type) {
556 case ICMP_PARAMETERPROB:
557 err = EPROTO;
558 break;
559 case ICMP_DEST_UNREACH:
560 if (code > NR_ICMP_UNREACH)
561 return;
562 if (code == ICMP_FRAG_NEEDED) {
563 sctp_icmp_frag_needed(sk, asoc, t, SCTP_TRUNC4(info));
564 return;
565 }
566 if (code == ICMP_PROT_UNREACH) {
567 sctp_icmp_proto_unreachable(sk, asoc, t);
568 return;
569 }
570 err = icmp_err_convert[code].errno;
571 break;
572 case ICMP_TIME_EXCEEDED:
573 if (code == ICMP_EXC_FRAGTIME)
574 return;
575
576 err = EHOSTUNREACH;
577 break;
578 case ICMP_REDIRECT:
579 sctp_icmp_redirect(sk, t, skb);
580 return;
581 default:
582 return;
583 }
584 if (!sock_owned_by_user(sk) && inet_test_bit(RECVERR, sk)) {
585 sk->sk_err = err;
586 sk_error_report(sk);
587 } else { /* Only an error on timeout */
588 WRITE_ONCE(sk->sk_err_soft, err);
589 }
590}
591
592/*
593 * This routine is called by the ICMP module when it gets some
594 * sort of error condition. If err < 0 then the socket should
595 * be closed and the error returned to the user. If err > 0
596 * it's just the icmp type << 8 | icmp code. After adjustment
597 * header points to the first 8 bytes of the sctp header. We need
598 * to find the appropriate port.
599 *
600 * The locking strategy used here is very "optimistic". When
601 * someone else accesses the socket the ICMP is just dropped
602 * and for some paths there is no check at all.
603 * A more general error queue to queue errors for later handling
604 * is probably better.
605 *
606 */
607int sctp_v4_err(struct sk_buff *skb, __u32 info)
608{
609 const struct iphdr *iph = (const struct iphdr *)skb->data;
610 const int type = icmp_hdr(skb)->type;
611 const int code = icmp_hdr(skb)->code;
612 struct net *net = dev_net(skb->dev);
613 struct sctp_transport *transport;
614 struct sctp_association *asoc;
615 __u16 saveip, savesctp;
616 struct sock *sk;
617
618 /* Fix up skb to look at the embedded net header. */
619 saveip = skb->network_header;
620 savesctp = skb->transport_header;
621 skb_reset_network_header(skb);
622 skb_set_transport_header(skb, iph->ihl * 4);
623 sk = sctp_err_lookup(net, AF_INET, skb, sctp_hdr(skb), &asoc, &transport);
624 /* Put back, the original values. */
625 skb->network_header = saveip;
626 skb->transport_header = savesctp;
627 if (!sk) {
628 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
629 return -ENOENT;
630 }
631
632 sctp_v4_err_handle(transport, skb, type, code, info);
633 sctp_err_finish(sk, transport);
634
635 return 0;
636}
637
638int sctp_udp_v4_err(struct sock *sk, struct sk_buff *skb)
639{
640 struct net *net = dev_net(skb->dev);
641 struct sctp_association *asoc;
642 struct sctp_transport *t;
643 struct icmphdr *hdr;
644 __u32 info = 0;
645
646 skb->transport_header += sizeof(struct udphdr);
647 sk = sctp_err_lookup(net, AF_INET, skb, sctp_hdr(skb), &asoc, &t);
648 if (!sk) {
649 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
650 return -ENOENT;
651 }
652
653 skb->transport_header -= sizeof(struct udphdr);
654 hdr = (struct icmphdr *)(skb_network_header(skb) - sizeof(struct icmphdr));
655 if (hdr->type == ICMP_REDIRECT) {
656 /* can't be handled without outer iphdr known, leave it to udp_err */
657 sctp_err_finish(sk, t);
658 return 0;
659 }
660 if (hdr->type == ICMP_DEST_UNREACH && hdr->code == ICMP_FRAG_NEEDED)
661 info = ntohs(hdr->un.frag.mtu);
662 sctp_v4_err_handle(t, skb, hdr->type, hdr->code, info);
663
664 sctp_err_finish(sk, t);
665 return 1;
666}
667
668/*
669 * RFC 2960, 8.4 - Handle "Out of the blue" Packets.
670 *
671 * This function scans all the chunks in the OOTB packet to determine if
672 * the packet should be discarded right away. If a response might be needed
673 * for this packet, or, if further processing is possible, the packet will
674 * be queued to a proper inqueue for the next phase of handling.
675 *
676 * Output:
677 * Return 0 - If further processing is needed.
678 * Return 1 - If the packet can be discarded right away.
679 */
680static int sctp_rcv_ootb(struct sk_buff *skb)
681{
682 struct sctp_chunkhdr *ch, _ch;
683 int ch_end, offset = 0;
684
685 /* Scan through all the chunks in the packet. */
686 do {
687 /* Make sure we have at least the header there */
688 if (offset + sizeof(_ch) > skb->len)
689 break;
690
691 ch = skb_header_pointer(skb, offset, sizeof(*ch), &_ch);
692
693 /* Break out if chunk length is less then minimal. */
694 if (!ch || ntohs(ch->length) < sizeof(_ch))
695 break;
696
697 ch_end = offset + SCTP_PAD4(ntohs(ch->length));
698 if (ch_end > skb->len)
699 break;
700
701 /* RFC 8.4, 2) If the OOTB packet contains an ABORT chunk, the
702 * receiver MUST silently discard the OOTB packet and take no
703 * further action.
704 */
705 if (SCTP_CID_ABORT == ch->type)
706 goto discard;
707
708 /* RFC 8.4, 6) If the packet contains a SHUTDOWN COMPLETE
709 * chunk, the receiver should silently discard the packet
710 * and take no further action.
711 */
712 if (SCTP_CID_SHUTDOWN_COMPLETE == ch->type)
713 goto discard;
714
715 /* RFC 4460, 2.11.2
716 * This will discard packets with INIT chunk bundled as
717 * subsequent chunks in the packet. When INIT is first,
718 * the normal INIT processing will discard the chunk.
719 */
720 if (SCTP_CID_INIT == ch->type && (void *)ch != skb->data)
721 goto discard;
722
723 offset = ch_end;
724 } while (ch_end < skb->len);
725
726 return 0;
727
728discard:
729 return 1;
730}
731
732/* Insert endpoint into the hash table. */
733static int __sctp_hash_endpoint(struct sctp_endpoint *ep)
734{
735 struct sock *sk = ep->base.sk;
736 struct net *net = sock_net(sk);
737 struct sctp_hashbucket *head;
738 int err = 0;
739
740 ep->hashent = sctp_ep_hashfn(net, ep->base.bind_addr.port);
741 head = &sctp_ep_hashtable[ep->hashent];
742
743 write_lock(&head->lock);
744 if (sk->sk_reuseport) {
745 bool any = sctp_is_ep_boundall(sk);
746 struct sctp_endpoint *ep2;
747 struct list_head *list;
748 int cnt = 0;
749
750 err = 1;
751
752 list_for_each(list, &ep->base.bind_addr.address_list)
753 cnt++;
754
755 sctp_for_each_hentry(ep2, &head->chain) {
756 struct sock *sk2 = ep2->base.sk;
757
758 if (!net_eq(sock_net(sk2), net) || sk2 == sk ||
759 !uid_eq(sock_i_uid(sk2), sock_i_uid(sk)) ||
760 !sk2->sk_reuseport)
761 continue;
762
763 err = sctp_bind_addrs_check(sctp_sk(sk2),
764 sctp_sk(sk), cnt);
765 if (!err) {
766 err = reuseport_add_sock(sk, sk2, any);
767 if (err)
768 goto out;
769 break;
770 } else if (err < 0) {
771 goto out;
772 }
773 }
774
775 if (err) {
776 err = reuseport_alloc(sk, any);
777 if (err)
778 goto out;
779 }
780 }
781
782 hlist_add_head(&ep->node, &head->chain);
783out:
784 write_unlock(&head->lock);
785 return err;
786}
787
788/* Add an endpoint to the hash. Local BH-safe. */
789int sctp_hash_endpoint(struct sctp_endpoint *ep)
790{
791 int err;
792
793 local_bh_disable();
794 err = __sctp_hash_endpoint(ep);
795 local_bh_enable();
796
797 return err;
798}
799
800/* Remove endpoint from the hash table. */
801static void __sctp_unhash_endpoint(struct sctp_endpoint *ep)
802{
803 struct sock *sk = ep->base.sk;
804 struct sctp_hashbucket *head;
805
806 ep->hashent = sctp_ep_hashfn(sock_net(sk), ep->base.bind_addr.port);
807
808 head = &sctp_ep_hashtable[ep->hashent];
809
810 write_lock(&head->lock);
811 if (rcu_access_pointer(sk->sk_reuseport_cb))
812 reuseport_detach_sock(sk);
813 hlist_del_init(&ep->node);
814 write_unlock(&head->lock);
815}
816
817/* Remove endpoint from the hash. Local BH-safe. */
818void sctp_unhash_endpoint(struct sctp_endpoint *ep)
819{
820 local_bh_disable();
821 __sctp_unhash_endpoint(ep);
822 local_bh_enable();
823}
824
825static inline __u32 sctp_hashfn(const struct net *net, __be16 lport,
826 const union sctp_addr *paddr, __u32 seed)
827{
828 __u32 addr;
829
830 if (paddr->sa.sa_family == AF_INET6)
831 addr = jhash(&paddr->v6.sin6_addr, 16, seed);
832 else
833 addr = (__force __u32)paddr->v4.sin_addr.s_addr;
834
835 return jhash_3words(addr, ((__force __u32)paddr->v4.sin_port) << 16 |
836 (__force __u32)lport, net_hash_mix(net), seed);
837}
838
839/* Look up an endpoint. */
840static struct sctp_endpoint *__sctp_rcv_lookup_endpoint(
841 struct net *net, struct sk_buff *skb,
842 const union sctp_addr *laddr,
843 const union sctp_addr *paddr,
844 int dif, int sdif)
845{
846 struct sctp_hashbucket *head;
847 struct sctp_endpoint *ep;
848 struct sock *sk;
849 __be16 lport;
850 int hash;
851
852 lport = laddr->v4.sin_port;
853 hash = sctp_ep_hashfn(net, ntohs(lport));
854 head = &sctp_ep_hashtable[hash];
855 read_lock(&head->lock);
856 sctp_for_each_hentry(ep, &head->chain) {
857 if (sctp_endpoint_is_match(ep, net, laddr, dif, sdif))
858 goto hit;
859 }
860
861 ep = sctp_sk(net->sctp.ctl_sock)->ep;
862
863hit:
864 sk = ep->base.sk;
865 if (sk->sk_reuseport) {
866 __u32 phash = sctp_hashfn(net, lport, paddr, 0);
867
868 sk = reuseport_select_sock(sk, phash, skb,
869 sizeof(struct sctphdr));
870 if (sk)
871 ep = sctp_sk(sk)->ep;
872 }
873 sctp_endpoint_hold(ep);
874 read_unlock(&head->lock);
875 return ep;
876}
877
878/* rhashtable for transport */
879struct sctp_hash_cmp_arg {
880 const union sctp_addr *paddr;
881 const struct net *net;
882 __be16 lport;
883};
884
885static inline int sctp_hash_cmp(struct rhashtable_compare_arg *arg,
886 const void *ptr)
887{
888 struct sctp_transport *t = (struct sctp_transport *)ptr;
889 const struct sctp_hash_cmp_arg *x = arg->key;
890 int err = 1;
891
892 if (!sctp_cmp_addr_exact(&t->ipaddr, x->paddr))
893 return err;
894 if (!sctp_transport_hold(t))
895 return err;
896
897 if (!net_eq(t->asoc->base.net, x->net))
898 goto out;
899 if (x->lport != htons(t->asoc->base.bind_addr.port))
900 goto out;
901
902 err = 0;
903out:
904 sctp_transport_put(t);
905 return err;
906}
907
908static inline __u32 sctp_hash_obj(const void *data, u32 len, u32 seed)
909{
910 const struct sctp_transport *t = data;
911
912 return sctp_hashfn(t->asoc->base.net,
913 htons(t->asoc->base.bind_addr.port),
914 &t->ipaddr, seed);
915}
916
917static inline __u32 sctp_hash_key(const void *data, u32 len, u32 seed)
918{
919 const struct sctp_hash_cmp_arg *x = data;
920
921 return sctp_hashfn(x->net, x->lport, x->paddr, seed);
922}
923
924static const struct rhashtable_params sctp_hash_params = {
925 .head_offset = offsetof(struct sctp_transport, node),
926 .hashfn = sctp_hash_key,
927 .obj_hashfn = sctp_hash_obj,
928 .obj_cmpfn = sctp_hash_cmp,
929 .automatic_shrinking = true,
930};
931
932int sctp_transport_hashtable_init(void)
933{
934 return rhltable_init(&sctp_transport_hashtable, &sctp_hash_params);
935}
936
937void sctp_transport_hashtable_destroy(void)
938{
939 rhltable_destroy(&sctp_transport_hashtable);
940}
941
942int sctp_hash_transport(struct sctp_transport *t)
943{
944 struct sctp_transport *transport;
945 struct rhlist_head *tmp, *list;
946 struct sctp_hash_cmp_arg arg;
947 int err;
948
949 if (t->asoc->temp)
950 return 0;
951
952 arg.net = t->asoc->base.net;
953 arg.paddr = &t->ipaddr;
954 arg.lport = htons(t->asoc->base.bind_addr.port);
955
956 rcu_read_lock();
957 list = rhltable_lookup(&sctp_transport_hashtable, &arg,
958 sctp_hash_params);
959
960 rhl_for_each_entry_rcu(transport, tmp, list, node)
961 if (transport->asoc->ep == t->asoc->ep) {
962 rcu_read_unlock();
963 return -EEXIST;
964 }
965 rcu_read_unlock();
966
967 err = rhltable_insert_key(&sctp_transport_hashtable, &arg,
968 &t->node, sctp_hash_params);
969 if (err)
970 pr_err_once("insert transport fail, errno %d\n", err);
971
972 return err;
973}
974
975void sctp_unhash_transport(struct sctp_transport *t)
976{
977 if (t->asoc->temp)
978 return;
979
980 rhltable_remove(&sctp_transport_hashtable, &t->node,
981 sctp_hash_params);
982}
983
984bool sctp_sk_bound_dev_eq(struct net *net, int bound_dev_if, int dif, int sdif)
985{
986 bool l3mdev_accept = true;
987
988#if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
989 l3mdev_accept = !!READ_ONCE(net->sctp.l3mdev_accept);
990#endif
991 return inet_bound_dev_eq(l3mdev_accept, bound_dev_if, dif, sdif);
992}
993
994/* return a transport with holding it */
995struct sctp_transport *sctp_addrs_lookup_transport(
996 struct net *net,
997 const union sctp_addr *laddr,
998 const union sctp_addr *paddr,
999 int dif, int sdif)
1000{
1001 struct rhlist_head *tmp, *list;
1002 struct sctp_transport *t;
1003 int bound_dev_if;
1004 struct sctp_hash_cmp_arg arg = {
1005 .paddr = paddr,
1006 .net = net,
1007 .lport = laddr->v4.sin_port,
1008 };
1009
1010 list = rhltable_lookup(&sctp_transport_hashtable, &arg,
1011 sctp_hash_params);
1012
1013 rhl_for_each_entry_rcu(t, tmp, list, node) {
1014 if (!sctp_transport_hold(t))
1015 continue;
1016
1017 bound_dev_if = READ_ONCE(t->asoc->base.sk->sk_bound_dev_if);
1018 if (sctp_sk_bound_dev_eq(net, bound_dev_if, dif, sdif) &&
1019 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 int dif, int sdif)
1058{
1059 struct sctp_transport *t;
1060 struct sctp_association *asoc = NULL;
1061
1062 t = sctp_addrs_lookup_transport(net, local, peer, dif, sdif);
1063 if (!t)
1064 goto out;
1065
1066 asoc = t->asoc;
1067 *pt = t;
1068
1069out:
1070 return asoc;
1071}
1072
1073/* Look up an association. protected by RCU read lock */
1074static
1075struct sctp_association *sctp_lookup_association(struct net *net,
1076 const union sctp_addr *laddr,
1077 const union sctp_addr *paddr,
1078 struct sctp_transport **transportp,
1079 int dif, int sdif)
1080{
1081 struct sctp_association *asoc;
1082
1083 rcu_read_lock();
1084 asoc = __sctp_lookup_association(net, laddr, paddr, transportp, dif, sdif);
1085 rcu_read_unlock();
1086
1087 return asoc;
1088}
1089
1090/* Is there an association matching the given local and peer addresses? */
1091bool sctp_has_association(struct net *net,
1092 const union sctp_addr *laddr,
1093 const union sctp_addr *paddr,
1094 int dif, int sdif)
1095{
1096 struct sctp_transport *transport;
1097
1098 if (sctp_lookup_association(net, laddr, paddr, &transport, dif, sdif)) {
1099 sctp_transport_put(transport);
1100 return true;
1101 }
1102
1103 return false;
1104}
1105
1106/*
1107 * SCTP Implementors Guide, 2.18 Handling of address
1108 * parameters within the INIT or INIT-ACK.
1109 *
1110 * D) When searching for a matching TCB upon reception of an INIT
1111 * or INIT-ACK chunk the receiver SHOULD use not only the
1112 * source address of the packet (containing the INIT or
1113 * INIT-ACK) but the receiver SHOULD also use all valid
1114 * address parameters contained within the chunk.
1115 *
1116 * 2.18.3 Solution description
1117 *
1118 * This new text clearly specifies to an implementor the need
1119 * to look within the INIT or INIT-ACK. Any implementation that
1120 * does not do this, may not be able to establish associations
1121 * in certain circumstances.
1122 *
1123 */
1124static struct sctp_association *__sctp_rcv_init_lookup(struct net *net,
1125 struct sk_buff *skb,
1126 const union sctp_addr *laddr, struct sctp_transport **transportp,
1127 int dif, int sdif)
1128{
1129 struct sctp_association *asoc;
1130 union sctp_addr addr;
1131 union sctp_addr *paddr = &addr;
1132 struct sctphdr *sh = sctp_hdr(skb);
1133 union sctp_params params;
1134 struct sctp_init_chunk *init;
1135 struct sctp_af *af;
1136
1137 /*
1138 * This code will NOT touch anything inside the chunk--it is
1139 * strictly READ-ONLY.
1140 *
1141 * RFC 2960 3 SCTP packet Format
1142 *
1143 * Multiple chunks can be bundled into one SCTP packet up to
1144 * the MTU size, except for the INIT, INIT ACK, and SHUTDOWN
1145 * COMPLETE chunks. These chunks MUST NOT be bundled with any
1146 * other chunk in a packet. See Section 6.10 for more details
1147 * on chunk bundling.
1148 */
1149
1150 /* Find the start of the TLVs and the end of the chunk. This is
1151 * the region we search for address parameters.
1152 */
1153 init = (struct sctp_init_chunk *)skb->data;
1154
1155 /* Walk the parameters looking for embedded addresses. */
1156 sctp_walk_params(params, init) {
1157
1158 /* Note: Ignoring hostname addresses. */
1159 af = sctp_get_af_specific(param_type2af(params.p->type));
1160 if (!af)
1161 continue;
1162
1163 if (!af->from_addr_param(paddr, params.addr, sh->source, 0))
1164 continue;
1165
1166 asoc = __sctp_lookup_association(net, laddr, paddr, transportp, dif, sdif);
1167 if (asoc)
1168 return asoc;
1169 }
1170
1171 return NULL;
1172}
1173
1174/* ADD-IP, Section 5.2
1175 * When an endpoint receives an ASCONF Chunk from the remote peer
1176 * special procedures may be needed to identify the association the
1177 * ASCONF Chunk is associated with. To properly find the association
1178 * the following procedures SHOULD be followed:
1179 *
1180 * D2) If the association is not found, use the address found in the
1181 * Address Parameter TLV combined with the port number found in the
1182 * SCTP common header. If found proceed to rule D4.
1183 *
1184 * D2-ext) If more than one ASCONF Chunks are packed together, use the
1185 * address found in the ASCONF Address Parameter TLV of each of the
1186 * subsequent ASCONF Chunks. If found, proceed to rule D4.
1187 */
1188static struct sctp_association *__sctp_rcv_asconf_lookup(
1189 struct net *net,
1190 struct sctp_chunkhdr *ch,
1191 const union sctp_addr *laddr,
1192 __be16 peer_port,
1193 struct sctp_transport **transportp,
1194 int dif, int sdif)
1195{
1196 struct sctp_addip_chunk *asconf = (struct sctp_addip_chunk *)ch;
1197 struct sctp_af *af;
1198 union sctp_addr_param *param;
1199 union sctp_addr paddr;
1200
1201 if (ntohs(ch->length) < sizeof(*asconf) + sizeof(struct sctp_paramhdr))
1202 return NULL;
1203
1204 /* Skip over the ADDIP header and find the Address parameter */
1205 param = (union sctp_addr_param *)(asconf + 1);
1206
1207 af = sctp_get_af_specific(param_type2af(param->p.type));
1208 if (unlikely(!af))
1209 return NULL;
1210
1211 if (!af->from_addr_param(&paddr, param, peer_port, 0))
1212 return NULL;
1213
1214 return __sctp_lookup_association(net, laddr, &paddr, transportp, dif, sdif);
1215}
1216
1217
1218/* SCTP-AUTH, Section 6.3:
1219* If the receiver does not find a STCB for a packet containing an AUTH
1220* chunk as the first chunk and not a COOKIE-ECHO chunk as the second
1221* chunk, it MUST use the chunks after the AUTH chunk to look up an existing
1222* association.
1223*
1224* This means that any chunks that can help us identify the association need
1225* to be looked at to find this association.
1226*/
1227static struct sctp_association *__sctp_rcv_walk_lookup(struct net *net,
1228 struct sk_buff *skb,
1229 const union sctp_addr *laddr,
1230 struct sctp_transport **transportp,
1231 int dif, int sdif)
1232{
1233 struct sctp_association *asoc = NULL;
1234 struct sctp_chunkhdr *ch;
1235 int have_auth = 0;
1236 unsigned int chunk_num = 1;
1237 __u8 *ch_end;
1238
1239 /* Walk through the chunks looking for AUTH or ASCONF chunks
1240 * to help us find the association.
1241 */
1242 ch = (struct sctp_chunkhdr *)skb->data;
1243 do {
1244 /* Break out if chunk length is less then minimal. */
1245 if (ntohs(ch->length) < sizeof(*ch))
1246 break;
1247
1248 ch_end = ((__u8 *)ch) + SCTP_PAD4(ntohs(ch->length));
1249 if (ch_end > skb_tail_pointer(skb))
1250 break;
1251
1252 switch (ch->type) {
1253 case SCTP_CID_AUTH:
1254 have_auth = chunk_num;
1255 break;
1256
1257 case SCTP_CID_COOKIE_ECHO:
1258 /* If a packet arrives containing an AUTH chunk as
1259 * a first chunk, a COOKIE-ECHO chunk as the second
1260 * chunk, and possibly more chunks after them, and
1261 * the receiver does not have an STCB for that
1262 * packet, then authentication is based on
1263 * the contents of the COOKIE- ECHO chunk.
1264 */
1265 if (have_auth == 1 && chunk_num == 2)
1266 return NULL;
1267 break;
1268
1269 case SCTP_CID_ASCONF:
1270 if (have_auth || net->sctp.addip_noauth)
1271 asoc = __sctp_rcv_asconf_lookup(
1272 net, ch, laddr,
1273 sctp_hdr(skb)->source,
1274 transportp, dif, sdif);
1275 break;
1276 default:
1277 break;
1278 }
1279
1280 if (asoc)
1281 break;
1282
1283 ch = (struct sctp_chunkhdr *)ch_end;
1284 chunk_num++;
1285 } while (ch_end + sizeof(*ch) < skb_tail_pointer(skb));
1286
1287 return asoc;
1288}
1289
1290/*
1291 * There are circumstances when we need to look inside the SCTP packet
1292 * for information to help us find the association. Examples
1293 * include looking inside of INIT/INIT-ACK chunks or after the AUTH
1294 * chunks.
1295 */
1296static struct sctp_association *__sctp_rcv_lookup_harder(struct net *net,
1297 struct sk_buff *skb,
1298 const union sctp_addr *laddr,
1299 struct sctp_transport **transportp,
1300 int dif, int sdif)
1301{
1302 struct sctp_chunkhdr *ch;
1303
1304 /* We do not allow GSO frames here as we need to linearize and
1305 * then cannot guarantee frame boundaries. This shouldn't be an
1306 * issue as packets hitting this are mostly INIT or INIT-ACK and
1307 * those cannot be on GSO-style anyway.
1308 */
1309 if (skb_is_gso(skb) && skb_is_gso_sctp(skb))
1310 return NULL;
1311
1312 ch = (struct sctp_chunkhdr *)skb->data;
1313
1314 /* The code below will attempt to walk the chunk and extract
1315 * parameter information. Before we do that, we need to verify
1316 * that the chunk length doesn't cause overflow. Otherwise, we'll
1317 * walk off the end.
1318 */
1319 if (SCTP_PAD4(ntohs(ch->length)) > skb->len)
1320 return NULL;
1321
1322 /* If this is INIT/INIT-ACK look inside the chunk too. */
1323 if (ch->type == SCTP_CID_INIT || ch->type == SCTP_CID_INIT_ACK)
1324 return __sctp_rcv_init_lookup(net, skb, laddr, transportp, dif, sdif);
1325
1326 return __sctp_rcv_walk_lookup(net, skb, laddr, transportp, dif, sdif);
1327}
1328
1329/* Lookup an association for an inbound skb. */
1330static struct sctp_association *__sctp_rcv_lookup(struct net *net,
1331 struct sk_buff *skb,
1332 const union sctp_addr *paddr,
1333 const union sctp_addr *laddr,
1334 struct sctp_transport **transportp,
1335 int dif, int sdif)
1336{
1337 struct sctp_association *asoc;
1338
1339 asoc = __sctp_lookup_association(net, laddr, paddr, transportp, dif, sdif);
1340 if (asoc)
1341 goto out;
1342
1343 /* Further lookup for INIT/INIT-ACK packets.
1344 * SCTP Implementors Guide, 2.18 Handling of address
1345 * parameters within the INIT or INIT-ACK.
1346 */
1347 asoc = __sctp_rcv_lookup_harder(net, skb, laddr, transportp, dif, sdif);
1348 if (asoc)
1349 goto out;
1350
1351 if (paddr->sa.sa_family == AF_INET)
1352 pr_debug("sctp: asoc not found for src:%pI4:%d dst:%pI4:%d\n",
1353 &laddr->v4.sin_addr, ntohs(laddr->v4.sin_port),
1354 &paddr->v4.sin_addr, ntohs(paddr->v4.sin_port));
1355 else
1356 pr_debug("sctp: asoc not found for src:%pI6:%d dst:%pI6:%d\n",
1357 &laddr->v6.sin6_addr, ntohs(laddr->v6.sin6_port),
1358 &paddr->v6.sin6_addr, ntohs(paddr->v6.sin6_port));
1359
1360out:
1361 return asoc;
1362}
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 || (t->pathmtu <= pmtu))
389 return;
390
391 if (sock_owned_by_user(sk)) {
392 atomic_set(&t->mtu_info, pmtu);
393 asoc->pmtu_pending = 1;
394 t->pmtu_pending = 1;
395 return;
396 }
397
398 if (!(t->param_flags & SPP_PMTUD_ENABLE))
399 /* We can't allow retransmitting in such case, as the
400 * retransmission would be sized just as before, and thus we
401 * would get another icmp, and retransmit again.
402 */
403 return;
404
405 /* Update transports view of the MTU. Return if no update was needed.
406 * If an update wasn't needed/possible, it also doesn't make sense to
407 * try to retransmit now.
408 */
409 if (!sctp_transport_update_pmtu(t, pmtu))
410 return;
411
412 /* Update association pmtu. */
413 sctp_assoc_sync_pmtu(asoc);
414
415 /* Retransmit with the new pmtu setting. */
416 sctp_retransmit(&asoc->outqueue, t, SCTP_RTXR_PMTUD);
417}
418
419void sctp_icmp_redirect(struct sock *sk, struct sctp_transport *t,
420 struct sk_buff *skb)
421{
422 struct dst_entry *dst;
423
424 if (sock_owned_by_user(sk) || !t)
425 return;
426 dst = sctp_transport_dst_check(t);
427 if (dst)
428 dst->ops->redirect(dst, sk, skb);
429}
430
431/*
432 * SCTP Implementer's Guide, 2.37 ICMP handling procedures
433 *
434 * ICMP8) If the ICMP code is a "Unrecognized next header type encountered"
435 * or a "Protocol Unreachable" treat this message as an abort
436 * with the T bit set.
437 *
438 * This function sends an event to the state machine, which will abort the
439 * association.
440 *
441 */
442void sctp_icmp_proto_unreachable(struct sock *sk,
443 struct sctp_association *asoc,
444 struct sctp_transport *t)
445{
446 if (sock_owned_by_user(sk)) {
447 if (timer_pending(&t->proto_unreach_timer))
448 return;
449 else {
450 if (!mod_timer(&t->proto_unreach_timer,
451 jiffies + (HZ/20)))
452 sctp_association_hold(asoc);
453 }
454 } else {
455 struct net *net = sock_net(sk);
456
457 pr_debug("%s: unrecognized next header type "
458 "encountered!\n", __func__);
459
460 if (del_timer(&t->proto_unreach_timer))
461 sctp_association_put(asoc);
462
463 sctp_do_sm(net, SCTP_EVENT_T_OTHER,
464 SCTP_ST_OTHER(SCTP_EVENT_ICMP_PROTO_UNREACH),
465 asoc->state, asoc->ep, asoc, t,
466 GFP_ATOMIC);
467 }
468}
469
470/* Common lookup code for icmp/icmpv6 error handler. */
471struct sock *sctp_err_lookup(struct net *net, int family, struct sk_buff *skb,
472 struct sctphdr *sctphdr,
473 struct sctp_association **app,
474 struct sctp_transport **tpp)
475{
476 struct sctp_init_chunk *chunkhdr, _chunkhdr;
477 union sctp_addr saddr;
478 union sctp_addr daddr;
479 struct sctp_af *af;
480 struct sock *sk = NULL;
481 struct sctp_association *asoc;
482 struct sctp_transport *transport = NULL;
483 __u32 vtag = ntohl(sctphdr->vtag);
484
485 *app = NULL; *tpp = NULL;
486
487 af = sctp_get_af_specific(family);
488 if (unlikely(!af)) {
489 return NULL;
490 }
491
492 /* Initialize local addresses for lookups. */
493 af->from_skb(&saddr, skb, 1);
494 af->from_skb(&daddr, skb, 0);
495
496 /* Look for an association that matches the incoming ICMP error
497 * packet.
498 */
499 asoc = __sctp_lookup_association(net, &saddr, &daddr, &transport);
500 if (!asoc)
501 return NULL;
502
503 sk = asoc->base.sk;
504
505 /* RFC 4960, Appendix C. ICMP Handling
506 *
507 * ICMP6) An implementation MUST validate that the Verification Tag
508 * contained in the ICMP message matches the Verification Tag of
509 * the peer. If the Verification Tag is not 0 and does NOT
510 * match, discard the ICMP message. If it is 0 and the ICMP
511 * message contains enough bytes to verify that the chunk type is
512 * an INIT chunk and that the Initiate Tag matches the tag of the
513 * peer, continue with ICMP7. If the ICMP message is too short
514 * or the chunk type or the Initiate Tag does not match, silently
515 * discard the packet.
516 */
517 if (vtag == 0) {
518 /* chunk header + first 4 octects of init header */
519 chunkhdr = skb_header_pointer(skb, skb_transport_offset(skb) +
520 sizeof(struct sctphdr),
521 sizeof(struct sctp_chunkhdr) +
522 sizeof(__be32), &_chunkhdr);
523 if (!chunkhdr ||
524 chunkhdr->chunk_hdr.type != SCTP_CID_INIT ||
525 ntohl(chunkhdr->init_hdr.init_tag) != asoc->c.my_vtag)
526 goto out;
527
528 } else if (vtag != asoc->c.peer_vtag) {
529 goto out;
530 }
531
532 bh_lock_sock(sk);
533
534 /* If too many ICMPs get dropped on busy
535 * servers this needs to be solved differently.
536 */
537 if (sock_owned_by_user(sk))
538 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
539
540 *app = asoc;
541 *tpp = transport;
542 return sk;
543
544out:
545 sctp_transport_put(transport);
546 return NULL;
547}
548
549/* Common cleanup code for icmp/icmpv6 error handler. */
550void sctp_err_finish(struct sock *sk, struct sctp_transport *t)
551 __releases(&((__sk)->sk_lock.slock))
552{
553 bh_unlock_sock(sk);
554 sctp_transport_put(t);
555}
556
557/*
558 * This routine is called by the ICMP module when it gets some
559 * sort of error condition. If err < 0 then the socket should
560 * be closed and the error returned to the user. If err > 0
561 * it's just the icmp type << 8 | icmp code. After adjustment
562 * header points to the first 8 bytes of the sctp header. We need
563 * to find the appropriate port.
564 *
565 * The locking strategy used here is very "optimistic". When
566 * someone else accesses the socket the ICMP is just dropped
567 * and for some paths there is no check at all.
568 * A more general error queue to queue errors for later handling
569 * is probably better.
570 *
571 */
572int sctp_v4_err(struct sk_buff *skb, __u32 info)
573{
574 const struct iphdr *iph = (const struct iphdr *)skb->data;
575 const int ihlen = iph->ihl * 4;
576 const int type = icmp_hdr(skb)->type;
577 const int code = icmp_hdr(skb)->code;
578 struct sock *sk;
579 struct sctp_association *asoc = NULL;
580 struct sctp_transport *transport;
581 struct inet_sock *inet;
582 __u16 saveip, savesctp;
583 int err;
584 struct net *net = dev_net(skb->dev);
585
586 /* Fix up skb to look at the embedded net header. */
587 saveip = skb->network_header;
588 savesctp = skb->transport_header;
589 skb_reset_network_header(skb);
590 skb_set_transport_header(skb, ihlen);
591 sk = sctp_err_lookup(net, AF_INET, skb, sctp_hdr(skb), &asoc, &transport);
592 /* Put back, the original values. */
593 skb->network_header = saveip;
594 skb->transport_header = savesctp;
595 if (!sk) {
596 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
597 return -ENOENT;
598 }
599 /* Warning: The sock lock is held. Remember to call
600 * sctp_err_finish!
601 */
602
603 switch (type) {
604 case ICMP_PARAMETERPROB:
605 err = EPROTO;
606 break;
607 case ICMP_DEST_UNREACH:
608 if (code > NR_ICMP_UNREACH)
609 goto out_unlock;
610
611 /* PMTU discovery (RFC1191) */
612 if (ICMP_FRAG_NEEDED == code) {
613 sctp_icmp_frag_needed(sk, asoc, transport,
614 SCTP_TRUNC4(info));
615 goto out_unlock;
616 } else {
617 if (ICMP_PROT_UNREACH == code) {
618 sctp_icmp_proto_unreachable(sk, asoc,
619 transport);
620 goto out_unlock;
621 }
622 }
623 err = icmp_err_convert[code].errno;
624 break;
625 case ICMP_TIME_EXCEEDED:
626 /* Ignore any time exceeded errors due to fragment reassembly
627 * timeouts.
628 */
629 if (ICMP_EXC_FRAGTIME == code)
630 goto out_unlock;
631
632 err = EHOSTUNREACH;
633 break;
634 case ICMP_REDIRECT:
635 sctp_icmp_redirect(sk, transport, skb);
636 /* Fall through to out_unlock. */
637 default:
638 goto out_unlock;
639 }
640
641 inet = inet_sk(sk);
642 if (!sock_owned_by_user(sk) && inet->recverr) {
643 sk->sk_err = err;
644 sk->sk_error_report(sk);
645 } else { /* Only an error on timeout */
646 sk->sk_err_soft = err;
647 }
648
649out_unlock:
650 sctp_err_finish(sk, transport);
651 return 0;
652}
653
654/*
655 * RFC 2960, 8.4 - Handle "Out of the blue" Packets.
656 *
657 * This function scans all the chunks in the OOTB packet to determine if
658 * the packet should be discarded right away. If a response might be needed
659 * for this packet, or, if further processing is possible, the packet will
660 * be queued to a proper inqueue for the next phase of handling.
661 *
662 * Output:
663 * Return 0 - If further processing is needed.
664 * Return 1 - If the packet can be discarded right away.
665 */
666static int sctp_rcv_ootb(struct sk_buff *skb)
667{
668 struct sctp_chunkhdr *ch, _ch;
669 int ch_end, offset = 0;
670
671 /* Scan through all the chunks in the packet. */
672 do {
673 /* Make sure we have at least the header there */
674 if (offset + sizeof(_ch) > skb->len)
675 break;
676
677 ch = skb_header_pointer(skb, offset, sizeof(*ch), &_ch);
678
679 /* Break out if chunk length is less then minimal. */
680 if (ntohs(ch->length) < sizeof(_ch))
681 break;
682
683 ch_end = offset + SCTP_PAD4(ntohs(ch->length));
684 if (ch_end > skb->len)
685 break;
686
687 /* RFC 8.4, 2) If the OOTB packet contains an ABORT chunk, the
688 * receiver MUST silently discard the OOTB packet and take no
689 * further action.
690 */
691 if (SCTP_CID_ABORT == ch->type)
692 goto discard;
693
694 /* RFC 8.4, 6) If the packet contains a SHUTDOWN COMPLETE
695 * chunk, the receiver should silently discard the packet
696 * and take no further action.
697 */
698 if (SCTP_CID_SHUTDOWN_COMPLETE == ch->type)
699 goto discard;
700
701 /* RFC 4460, 2.11.2
702 * This will discard packets with INIT chunk bundled as
703 * subsequent chunks in the packet. When INIT is first,
704 * the normal INIT processing will discard the chunk.
705 */
706 if (SCTP_CID_INIT == ch->type && (void *)ch != skb->data)
707 goto discard;
708
709 offset = ch_end;
710 } while (ch_end < skb->len);
711
712 return 0;
713
714discard:
715 return 1;
716}
717
718/* Insert endpoint into the hash table. */
719static int __sctp_hash_endpoint(struct sctp_endpoint *ep)
720{
721 struct sock *sk = ep->base.sk;
722 struct net *net = sock_net(sk);
723 struct sctp_hashbucket *head;
724 struct sctp_ep_common *epb;
725
726 epb = &ep->base;
727 epb->hashent = sctp_ep_hashfn(net, epb->bind_addr.port);
728 head = &sctp_ep_hashtable[epb->hashent];
729
730 if (sk->sk_reuseport) {
731 bool any = sctp_is_ep_boundall(sk);
732 struct sctp_ep_common *epb2;
733 struct list_head *list;
734 int cnt = 0, err = 1;
735
736 list_for_each(list, &ep->base.bind_addr.address_list)
737 cnt++;
738
739 sctp_for_each_hentry(epb2, &head->chain) {
740 struct sock *sk2 = epb2->sk;
741
742 if (!net_eq(sock_net(sk2), net) || sk2 == sk ||
743 !uid_eq(sock_i_uid(sk2), sock_i_uid(sk)) ||
744 !sk2->sk_reuseport)
745 continue;
746
747 err = sctp_bind_addrs_check(sctp_sk(sk2),
748 sctp_sk(sk), cnt);
749 if (!err) {
750 err = reuseport_add_sock(sk, sk2, any);
751 if (err)
752 return err;
753 break;
754 } else if (err < 0) {
755 return err;
756 }
757 }
758
759 if (err) {
760 err = reuseport_alloc(sk, any);
761 if (err)
762 return err;
763 }
764 }
765
766 write_lock(&head->lock);
767 hlist_add_head(&epb->node, &head->chain);
768 write_unlock(&head->lock);
769 return 0;
770}
771
772/* Add an endpoint to the hash. Local BH-safe. */
773int sctp_hash_endpoint(struct sctp_endpoint *ep)
774{
775 int err;
776
777 local_bh_disable();
778 err = __sctp_hash_endpoint(ep);
779 local_bh_enable();
780
781 return err;
782}
783
784/* Remove endpoint from the hash table. */
785static void __sctp_unhash_endpoint(struct sctp_endpoint *ep)
786{
787 struct sock *sk = ep->base.sk;
788 struct sctp_hashbucket *head;
789 struct sctp_ep_common *epb;
790
791 epb = &ep->base;
792
793 epb->hashent = sctp_ep_hashfn(sock_net(sk), epb->bind_addr.port);
794
795 head = &sctp_ep_hashtable[epb->hashent];
796
797 if (rcu_access_pointer(sk->sk_reuseport_cb))
798 reuseport_detach_sock(sk);
799
800 write_lock(&head->lock);
801 hlist_del_init(&epb->node);
802 write_unlock(&head->lock);
803}
804
805/* Remove endpoint from the hash. Local BH-safe. */
806void sctp_unhash_endpoint(struct sctp_endpoint *ep)
807{
808 local_bh_disable();
809 __sctp_unhash_endpoint(ep);
810 local_bh_enable();
811}
812
813static inline __u32 sctp_hashfn(const struct net *net, __be16 lport,
814 const union sctp_addr *paddr, __u32 seed)
815{
816 __u32 addr;
817
818 if (paddr->sa.sa_family == AF_INET6)
819 addr = jhash(&paddr->v6.sin6_addr, 16, seed);
820 else
821 addr = (__force __u32)paddr->v4.sin_addr.s_addr;
822
823 return jhash_3words(addr, ((__force __u32)paddr->v4.sin_port) << 16 |
824 (__force __u32)lport, net_hash_mix(net), seed);
825}
826
827/* Look up an endpoint. */
828static struct sctp_endpoint *__sctp_rcv_lookup_endpoint(
829 struct net *net, struct sk_buff *skb,
830 const union sctp_addr *laddr,
831 const union sctp_addr *paddr)
832{
833 struct sctp_hashbucket *head;
834 struct sctp_ep_common *epb;
835 struct sctp_endpoint *ep;
836 struct sock *sk;
837 __be16 lport;
838 int hash;
839
840 lport = laddr->v4.sin_port;
841 hash = sctp_ep_hashfn(net, ntohs(lport));
842 head = &sctp_ep_hashtable[hash];
843 read_lock(&head->lock);
844 sctp_for_each_hentry(epb, &head->chain) {
845 ep = sctp_ep(epb);
846 if (sctp_endpoint_is_match(ep, net, laddr))
847 goto hit;
848 }
849
850 ep = sctp_sk(net->sctp.ctl_sock)->ep;
851
852hit:
853 sk = ep->base.sk;
854 if (sk->sk_reuseport) {
855 __u32 phash = sctp_hashfn(net, lport, paddr, 0);
856
857 sk = reuseport_select_sock(sk, phash, skb,
858 sizeof(struct sctphdr));
859 if (sk)
860 ep = sctp_sk(sk)->ep;
861 }
862 sctp_endpoint_hold(ep);
863 read_unlock(&head->lock);
864 return ep;
865}
866
867/* rhashtable for transport */
868struct sctp_hash_cmp_arg {
869 const union sctp_addr *paddr;
870 const struct net *net;
871 __be16 lport;
872};
873
874static inline int sctp_hash_cmp(struct rhashtable_compare_arg *arg,
875 const void *ptr)
876{
877 struct sctp_transport *t = (struct sctp_transport *)ptr;
878 const struct sctp_hash_cmp_arg *x = arg->key;
879 int err = 1;
880
881 if (!sctp_cmp_addr_exact(&t->ipaddr, x->paddr))
882 return err;
883 if (!sctp_transport_hold(t))
884 return err;
885
886 if (!net_eq(t->asoc->base.net, x->net))
887 goto out;
888 if (x->lport != htons(t->asoc->base.bind_addr.port))
889 goto out;
890
891 err = 0;
892out:
893 sctp_transport_put(t);
894 return err;
895}
896
897static inline __u32 sctp_hash_obj(const void *data, u32 len, u32 seed)
898{
899 const struct sctp_transport *t = data;
900
901 return sctp_hashfn(t->asoc->base.net,
902 htons(t->asoc->base.bind_addr.port),
903 &t->ipaddr, seed);
904}
905
906static inline __u32 sctp_hash_key(const void *data, u32 len, u32 seed)
907{
908 const struct sctp_hash_cmp_arg *x = data;
909
910 return sctp_hashfn(x->net, x->lport, x->paddr, seed);
911}
912
913static const struct rhashtable_params sctp_hash_params = {
914 .head_offset = offsetof(struct sctp_transport, node),
915 .hashfn = sctp_hash_key,
916 .obj_hashfn = sctp_hash_obj,
917 .obj_cmpfn = sctp_hash_cmp,
918 .automatic_shrinking = true,
919};
920
921int sctp_transport_hashtable_init(void)
922{
923 return rhltable_init(&sctp_transport_hashtable, &sctp_hash_params);
924}
925
926void sctp_transport_hashtable_destroy(void)
927{
928 rhltable_destroy(&sctp_transport_hashtable);
929}
930
931int sctp_hash_transport(struct sctp_transport *t)
932{
933 struct sctp_transport *transport;
934 struct rhlist_head *tmp, *list;
935 struct sctp_hash_cmp_arg arg;
936 int err;
937
938 if (t->asoc->temp)
939 return 0;
940
941 arg.net = t->asoc->base.net;
942 arg.paddr = &t->ipaddr;
943 arg.lport = htons(t->asoc->base.bind_addr.port);
944
945 rcu_read_lock();
946 list = rhltable_lookup(&sctp_transport_hashtable, &arg,
947 sctp_hash_params);
948
949 rhl_for_each_entry_rcu(transport, tmp, list, node)
950 if (transport->asoc->ep == t->asoc->ep) {
951 rcu_read_unlock();
952 return -EEXIST;
953 }
954 rcu_read_unlock();
955
956 err = rhltable_insert_key(&sctp_transport_hashtable, &arg,
957 &t->node, sctp_hash_params);
958 if (err)
959 pr_err_once("insert transport fail, errno %d\n", err);
960
961 return err;
962}
963
964void sctp_unhash_transport(struct sctp_transport *t)
965{
966 if (t->asoc->temp)
967 return;
968
969 rhltable_remove(&sctp_transport_hashtable, &t->node,
970 sctp_hash_params);
971}
972
973/* return a transport with holding it */
974struct sctp_transport *sctp_addrs_lookup_transport(
975 struct net *net,
976 const union sctp_addr *laddr,
977 const union sctp_addr *paddr)
978{
979 struct rhlist_head *tmp, *list;
980 struct sctp_transport *t;
981 struct sctp_hash_cmp_arg arg = {
982 .paddr = paddr,
983 .net = net,
984 .lport = laddr->v4.sin_port,
985 };
986
987 list = rhltable_lookup(&sctp_transport_hashtable, &arg,
988 sctp_hash_params);
989
990 rhl_for_each_entry_rcu(t, tmp, list, node) {
991 if (!sctp_transport_hold(t))
992 continue;
993
994 if (sctp_bind_addr_match(&t->asoc->base.bind_addr,
995 laddr, sctp_sk(t->asoc->base.sk)))
996 return t;
997 sctp_transport_put(t);
998 }
999
1000 return NULL;
1001}
1002
1003/* return a transport without holding it, as it's only used under sock lock */
1004struct sctp_transport *sctp_epaddr_lookup_transport(
1005 const struct sctp_endpoint *ep,
1006 const union sctp_addr *paddr)
1007{
1008 struct rhlist_head *tmp, *list;
1009 struct sctp_transport *t;
1010 struct sctp_hash_cmp_arg arg = {
1011 .paddr = paddr,
1012 .net = ep->base.net,
1013 .lport = htons(ep->base.bind_addr.port),
1014 };
1015
1016 list = rhltable_lookup(&sctp_transport_hashtable, &arg,
1017 sctp_hash_params);
1018
1019 rhl_for_each_entry_rcu(t, tmp, list, node)
1020 if (ep == t->asoc->ep)
1021 return t;
1022
1023 return NULL;
1024}
1025
1026/* Look up an association. */
1027static struct sctp_association *__sctp_lookup_association(
1028 struct net *net,
1029 const union sctp_addr *local,
1030 const union sctp_addr *peer,
1031 struct sctp_transport **pt)
1032{
1033 struct sctp_transport *t;
1034 struct sctp_association *asoc = NULL;
1035
1036 t = sctp_addrs_lookup_transport(net, local, peer);
1037 if (!t)
1038 goto out;
1039
1040 asoc = t->asoc;
1041 *pt = t;
1042
1043out:
1044 return asoc;
1045}
1046
1047/* Look up an association. protected by RCU read lock */
1048static
1049struct sctp_association *sctp_lookup_association(struct net *net,
1050 const union sctp_addr *laddr,
1051 const union sctp_addr *paddr,
1052 struct sctp_transport **transportp)
1053{
1054 struct sctp_association *asoc;
1055
1056 rcu_read_lock();
1057 asoc = __sctp_lookup_association(net, laddr, paddr, transportp);
1058 rcu_read_unlock();
1059
1060 return asoc;
1061}
1062
1063/* Is there an association matching the given local and peer addresses? */
1064bool sctp_has_association(struct net *net,
1065 const union sctp_addr *laddr,
1066 const union sctp_addr *paddr)
1067{
1068 struct sctp_transport *transport;
1069
1070 if (sctp_lookup_association(net, laddr, paddr, &transport)) {
1071 sctp_transport_put(transport);
1072 return true;
1073 }
1074
1075 return false;
1076}
1077
1078/*
1079 * SCTP Implementors Guide, 2.18 Handling of address
1080 * parameters within the INIT or INIT-ACK.
1081 *
1082 * D) When searching for a matching TCB upon reception of an INIT
1083 * or INIT-ACK chunk the receiver SHOULD use not only the
1084 * source address of the packet (containing the INIT or
1085 * INIT-ACK) but the receiver SHOULD also use all valid
1086 * address parameters contained within the chunk.
1087 *
1088 * 2.18.3 Solution description
1089 *
1090 * This new text clearly specifies to an implementor the need
1091 * to look within the INIT or INIT-ACK. Any implementation that
1092 * does not do this, may not be able to establish associations
1093 * in certain circumstances.
1094 *
1095 */
1096static struct sctp_association *__sctp_rcv_init_lookup(struct net *net,
1097 struct sk_buff *skb,
1098 const union sctp_addr *laddr, struct sctp_transport **transportp)
1099{
1100 struct sctp_association *asoc;
1101 union sctp_addr addr;
1102 union sctp_addr *paddr = &addr;
1103 struct sctphdr *sh = sctp_hdr(skb);
1104 union sctp_params params;
1105 struct sctp_init_chunk *init;
1106 struct sctp_af *af;
1107
1108 /*
1109 * This code will NOT touch anything inside the chunk--it is
1110 * strictly READ-ONLY.
1111 *
1112 * RFC 2960 3 SCTP packet Format
1113 *
1114 * Multiple chunks can be bundled into one SCTP packet up to
1115 * the MTU size, except for the INIT, INIT ACK, and SHUTDOWN
1116 * COMPLETE chunks. These chunks MUST NOT be bundled with any
1117 * other chunk in a packet. See Section 6.10 for more details
1118 * on chunk bundling.
1119 */
1120
1121 /* Find the start of the TLVs and the end of the chunk. This is
1122 * the region we search for address parameters.
1123 */
1124 init = (struct sctp_init_chunk *)skb->data;
1125
1126 /* Walk the parameters looking for embedded addresses. */
1127 sctp_walk_params(params, init, init_hdr.params) {
1128
1129 /* Note: Ignoring hostname addresses. */
1130 af = sctp_get_af_specific(param_type2af(params.p->type));
1131 if (!af)
1132 continue;
1133
1134 af->from_addr_param(paddr, params.addr, sh->source, 0);
1135
1136 asoc = __sctp_lookup_association(net, laddr, paddr, transportp);
1137 if (asoc)
1138 return asoc;
1139 }
1140
1141 return NULL;
1142}
1143
1144/* ADD-IP, Section 5.2
1145 * When an endpoint receives an ASCONF Chunk from the remote peer
1146 * special procedures may be needed to identify the association the
1147 * ASCONF Chunk is associated with. To properly find the association
1148 * the following procedures SHOULD be followed:
1149 *
1150 * D2) If the association is not found, use the address found in the
1151 * Address Parameter TLV combined with the port number found in the
1152 * SCTP common header. If found proceed to rule D4.
1153 *
1154 * D2-ext) If more than one ASCONF Chunks are packed together, use the
1155 * address found in the ASCONF Address Parameter TLV of each of the
1156 * subsequent ASCONF Chunks. If found, proceed to rule D4.
1157 */
1158static struct sctp_association *__sctp_rcv_asconf_lookup(
1159 struct net *net,
1160 struct sctp_chunkhdr *ch,
1161 const union sctp_addr *laddr,
1162 __be16 peer_port,
1163 struct sctp_transport **transportp)
1164{
1165 struct sctp_addip_chunk *asconf = (struct sctp_addip_chunk *)ch;
1166 struct sctp_af *af;
1167 union sctp_addr_param *param;
1168 union sctp_addr paddr;
1169
1170 /* Skip over the ADDIP header and find the Address parameter */
1171 param = (union sctp_addr_param *)(asconf + 1);
1172
1173 af = sctp_get_af_specific(param_type2af(param->p.type));
1174 if (unlikely(!af))
1175 return NULL;
1176
1177 af->from_addr_param(&paddr, param, peer_port, 0);
1178
1179 return __sctp_lookup_association(net, laddr, &paddr, transportp);
1180}
1181
1182
1183/* SCTP-AUTH, Section 6.3:
1184* If the receiver does not find a STCB for a packet containing an AUTH
1185* chunk as the first chunk and not a COOKIE-ECHO chunk as the second
1186* chunk, it MUST use the chunks after the AUTH chunk to look up an existing
1187* association.
1188*
1189* This means that any chunks that can help us identify the association need
1190* to be looked at to find this association.
1191*/
1192static struct sctp_association *__sctp_rcv_walk_lookup(struct net *net,
1193 struct sk_buff *skb,
1194 const union sctp_addr *laddr,
1195 struct sctp_transport **transportp)
1196{
1197 struct sctp_association *asoc = NULL;
1198 struct sctp_chunkhdr *ch;
1199 int have_auth = 0;
1200 unsigned int chunk_num = 1;
1201 __u8 *ch_end;
1202
1203 /* Walk through the chunks looking for AUTH or ASCONF chunks
1204 * to help us find the association.
1205 */
1206 ch = (struct sctp_chunkhdr *)skb->data;
1207 do {
1208 /* Break out if chunk length is less then minimal. */
1209 if (ntohs(ch->length) < sizeof(*ch))
1210 break;
1211
1212 ch_end = ((__u8 *)ch) + SCTP_PAD4(ntohs(ch->length));
1213 if (ch_end > skb_tail_pointer(skb))
1214 break;
1215
1216 switch (ch->type) {
1217 case SCTP_CID_AUTH:
1218 have_auth = chunk_num;
1219 break;
1220
1221 case SCTP_CID_COOKIE_ECHO:
1222 /* If a packet arrives containing an AUTH chunk as
1223 * a first chunk, a COOKIE-ECHO chunk as the second
1224 * chunk, and possibly more chunks after them, and
1225 * the receiver does not have an STCB for that
1226 * packet, then authentication is based on
1227 * the contents of the COOKIE- ECHO chunk.
1228 */
1229 if (have_auth == 1 && chunk_num == 2)
1230 return NULL;
1231 break;
1232
1233 case SCTP_CID_ASCONF:
1234 if (have_auth || net->sctp.addip_noauth)
1235 asoc = __sctp_rcv_asconf_lookup(
1236 net, ch, laddr,
1237 sctp_hdr(skb)->source,
1238 transportp);
1239 default:
1240 break;
1241 }
1242
1243 if (asoc)
1244 break;
1245
1246 ch = (struct sctp_chunkhdr *)ch_end;
1247 chunk_num++;
1248 } while (ch_end < skb_tail_pointer(skb));
1249
1250 return asoc;
1251}
1252
1253/*
1254 * There are circumstances when we need to look inside the SCTP packet
1255 * for information to help us find the association. Examples
1256 * include looking inside of INIT/INIT-ACK chunks or after the AUTH
1257 * chunks.
1258 */
1259static struct sctp_association *__sctp_rcv_lookup_harder(struct net *net,
1260 struct sk_buff *skb,
1261 const union sctp_addr *laddr,
1262 struct sctp_transport **transportp)
1263{
1264 struct sctp_chunkhdr *ch;
1265
1266 /* We do not allow GSO frames here as we need to linearize and
1267 * then cannot guarantee frame boundaries. This shouldn't be an
1268 * issue as packets hitting this are mostly INIT or INIT-ACK and
1269 * those cannot be on GSO-style anyway.
1270 */
1271 if (skb_is_gso(skb) && skb_is_gso_sctp(skb))
1272 return NULL;
1273
1274 ch = (struct sctp_chunkhdr *)skb->data;
1275
1276 /* The code below will attempt to walk the chunk and extract
1277 * parameter information. Before we do that, we need to verify
1278 * that the chunk length doesn't cause overflow. Otherwise, we'll
1279 * walk off the end.
1280 */
1281 if (SCTP_PAD4(ntohs(ch->length)) > skb->len)
1282 return NULL;
1283
1284 /* If this is INIT/INIT-ACK look inside the chunk too. */
1285 if (ch->type == SCTP_CID_INIT || ch->type == SCTP_CID_INIT_ACK)
1286 return __sctp_rcv_init_lookup(net, skb, laddr, transportp);
1287
1288 return __sctp_rcv_walk_lookup(net, skb, laddr, transportp);
1289}
1290
1291/* Lookup an association for an inbound skb. */
1292static struct sctp_association *__sctp_rcv_lookup(struct net *net,
1293 struct sk_buff *skb,
1294 const union sctp_addr *paddr,
1295 const union sctp_addr *laddr,
1296 struct sctp_transport **transportp)
1297{
1298 struct sctp_association *asoc;
1299
1300 asoc = __sctp_lookup_association(net, laddr, paddr, transportp);
1301 if (asoc)
1302 goto out;
1303
1304 /* Further lookup for INIT/INIT-ACK packets.
1305 * SCTP Implementors Guide, 2.18 Handling of address
1306 * parameters within the INIT or INIT-ACK.
1307 */
1308 asoc = __sctp_rcv_lookup_harder(net, skb, laddr, transportp);
1309 if (asoc)
1310 goto out;
1311
1312 if (paddr->sa.sa_family == AF_INET)
1313 pr_debug("sctp: asoc not found for src:%pI4:%d dst:%pI4:%d\n",
1314 &laddr->v4.sin_addr, ntohs(laddr->v4.sin_port),
1315 &paddr->v4.sin_addr, ntohs(paddr->v4.sin_port));
1316 else
1317 pr_debug("sctp: asoc not found for src:%pI6:%d dst:%pI6:%d\n",
1318 &laddr->v6.sin6_addr, ntohs(laddr->v6.sin6_port),
1319 &paddr->v6.sin6_addr, ntohs(paddr->v6.sin6_port));
1320
1321out:
1322 return asoc;
1323}