?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6
[linux-2.6/btrfs-unstable.git] / net / ipv6 / udp.c
blob83f4c77c79d86eda12ec76b572c4f64546f43ae4
1 /*
2 * UDP over IPv6
3 * Linux INET6 implementation
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 *
8 * Based on linux/ipv4/udp.c
9 *
10 * Fixes:
11 * Hideaki YOSHIFUJI : sin6_scope_id support
12 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
13 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
14 * a single port at the same time.
15 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data
16 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file.
17 *
18 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License
20 * as published by the Free Software Foundation; either version
21 * 2 of the License, or (at your option) any later version.
22 */
23
24 #include <linux/errno.h>
25 #include <linux/types.h>
26 #include <linux/socket.h>
27 #include <linux/sockios.h>
28 #include <linux/net.h>
29 #include <linux/in6.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_arp.h>
32 #include <linux/ipv6.h>
33 #include <linux/icmpv6.h>
34 #include <linux/init.h>
35 #include <linux/module.h>
36 #include <linux/skbuff.h>
37 #include <linux/slab.h>
38 #include <linux/uaccess.h>
39
40 #include <net/addrconf.h>
41 #include <net/ndisc.h>
42 #include <net/protocol.h>
43 #include <net/transp_v6.h>
44 #include <net/ip6_route.h>
45 #include <net/raw.h>
46 #include <net/tcp_states.h>
47 #include <net/ip6_checksum.h>
48 #include <net/xfrm.h>
49 #include <net/inet_hashtables.h>
50 #include <net/inet6_hashtables.h>
51 #include <net/busy_poll.h>
52 #include <net/sock_reuseport.h>
53
54 #include <linux/proc_fs.h>
55 #include <linux/seq_file.h>
56 #include <trace/events/skb.h>
57 #include "udp_impl.h"
58
59 static bool udp6_lib_exact_dif_match(struct net *net, struct sk_buff *skb)
60 {
61 #if defined(CONFIG_NET_L3_MASTER_DEV)
62 if (!net->ipv4.sysctl_udp_l3mdev_accept &&
63 skb && ipv6_l3mdev_skb(IP6CB(skb)->flags))
64 return true;
65 #endif
66 return false;
67 }
68
69 static u32 udp6_ehashfn(const struct net *net,
70 const struct in6_addr *laddr,
71 const u16 lport,
72 const struct in6_addr *faddr,
73 const __be16 fport)
74 {
75 static u32 udp6_ehash_secret __read_mostly;
76 static u32 udp_ipv6_hash_secret __read_mostly;
77
78 u32 lhash, fhash;
79
80 net_get_random_once(&udp6_ehash_secret,
81 sizeof(udp6_ehash_secret));
82 net_get_random_once(&udp_ipv6_hash_secret,
83 sizeof(udp_ipv6_hash_secret));
84
85 lhash = (__force u32)laddr->s6_addr32[3];
86 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
87
88 return __inet6_ehashfn(lhash, lport, fhash, fport,
89 udp_ipv6_hash_secret + net_hash_mix(net));
90 }
91
92 int udp_v6_get_port(struct sock *sk, unsigned short snum)
93 {
94 unsigned int hash2_nulladdr =
95 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
96 unsigned int hash2_partial =
97 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
98
99 /* precompute partial secondary hash */
100 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
101 return udp_lib_get_port(sk, snum, hash2_nulladdr);
102 }
103
104 static void udp_v6_rehash(struct sock *sk)
105 {
106 u16 new_hash = ipv6_portaddr_hash(sock_net(sk),
107 &sk->sk_v6_rcv_saddr,
108 inet_sk(sk)->inet_num);
109
110 udp_lib_rehash(sk, new_hash);
111 }
112
113 static int compute_score(struct sock *sk, struct net *net,
114 const struct in6_addr *saddr, __be16 sport,
115 const struct in6_addr *daddr, unsigned short hnum,
116 int dif, int sdif, bool exact_dif)
117 {
118 int score;
119 struct inet_sock *inet;
120
121 if (!net_eq(sock_net(sk), net) ||
122 udp_sk(sk)->udp_port_hash != hnum ||
123 sk->sk_family != PF_INET6)
124 return -1;
125
126 score = 0;
127 inet = inet_sk(sk);
128
129 if (inet->inet_dport) {
130 if (inet->inet_dport != sport)
131 return -1;
132 score++;
133 }
134
135 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) {
136 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
137 return -1;
138 score++;
139 }
140
141 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
142 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
143 return -1;
144 score++;
145 }
146
147 if (sk->sk_bound_dev_if || exact_dif) {
148 bool dev_match = (sk->sk_bound_dev_if == dif ||
149 sk->sk_bound_dev_if == sdif);
150
151 if (!dev_match)
152 return -1;
153 if (sk->sk_bound_dev_if)
154 score++;
155 }
156
157 if (sk->sk_incoming_cpu == raw_smp_processor_id())
158 score++;
159
160 return score;
161 }
162
163 /* called with rcu_read_lock() */
164 static struct sock *udp6_lib_lookup2(struct net *net,
165 const struct in6_addr *saddr, __be16 sport,
166 const struct in6_addr *daddr, unsigned int hnum,
167 int dif, int sdif, bool exact_dif,
168 struct udp_hslot *hslot2, struct sk_buff *skb)
169 {
170 struct sock *sk, *result;
171 int score, badness;
172 u32 hash = 0;
173
174 result = NULL;
175 badness = -1;
176 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
177 score = compute_score(sk, net, saddr, sport,
178 daddr, hnum, dif, sdif, exact_dif);
179 if (score > badness) {
180 if (sk->sk_reuseport) {
181 hash = udp6_ehashfn(net, daddr, hnum,
182 saddr, sport);
183
184 result = reuseport_select_sock(sk, hash, skb,
185 sizeof(struct udphdr));
186 if (result)
187 return result;
188 }
189 result = sk;
190 badness = score;
191 }
192 }
193 return result;
194 }
195
196 /* rcu_read_lock() must be held */
197 struct sock *__udp6_lib_lookup(struct net *net,
198 const struct in6_addr *saddr, __be16 sport,
199 const struct in6_addr *daddr, __be16 dport,
200 int dif, int sdif, struct udp_table *udptable,
201 struct sk_buff *skb)
202 {
203 struct sock *sk, *result;
204 unsigned short hnum = ntohs(dport);
205 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
206 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
207 bool exact_dif = udp6_lib_exact_dif_match(net, skb);
208 int score, badness;
209 u32 hash = 0;
210
211 if (hslot->count > 10) {
212 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
213 slot2 = hash2 & udptable->mask;
214 hslot2 = &udptable->hash2[slot2];
215 if (hslot->count < hslot2->count)
216 goto begin;
217
218 result = udp6_lib_lookup2(net, saddr, sport,
219 daddr, hnum, dif, sdif, exact_dif,
220 hslot2, skb);
221 if (!result) {
222 unsigned int old_slot2 = slot2;
223 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
224 slot2 = hash2 & udptable->mask;
225 /* avoid searching the same slot again. */
226 if (unlikely(slot2 == old_slot2))
227 return result;
228
229 hslot2 = &udptable->hash2[slot2];
230 if (hslot->count < hslot2->count)
231 goto begin;
232
233 result = udp6_lib_lookup2(net, saddr, sport,
234 daddr, hnum, dif, sdif,
235 exact_dif, hslot2,
236 skb);
237 }
238 if (unlikely(IS_ERR(result)))
239 return NULL;
240 return result;
241 }
242 begin:
243 result = NULL;
244 badness = -1;
245 sk_for_each_rcu(sk, &hslot->head) {
246 score = compute_score(sk, net, saddr, sport, daddr, hnum, dif,
247 sdif, exact_dif);
248 if (score > badness) {
249 if (sk->sk_reuseport) {
250 hash = udp6_ehashfn(net, daddr, hnum,
251 saddr, sport);
252 result = reuseport_select_sock(sk, hash, skb,
253 sizeof(struct udphdr));
254 if (unlikely(IS_ERR(result)))
255 return NULL;
256 if (result)
257 return result;
258 }
259 result = sk;
260 badness = score;
261 }
262 }
263 return result;
264 }
265 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
266
267 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
268 __be16 sport, __be16 dport,
269 struct udp_table *udptable)
270 {
271 const struct ipv6hdr *iph = ipv6_hdr(skb);
272
273 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
274 &iph->daddr, dport, inet6_iif(skb),
275 inet6_sdif(skb), udptable, skb);
276 }
277
278 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb,
279 __be16 sport, __be16 dport)
280 {
281 const struct ipv6hdr *iph = ipv6_hdr(skb);
282
283 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
284 &iph->daddr, dport, inet6_iif(skb),
285 inet6_sdif(skb), &udp_table, skb);
286 }
287 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb);
288
289 /* Must be called under rcu_read_lock().
290 * Does increment socket refcount.
291 */
292 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
293 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
294 const struct in6_addr *daddr, __be16 dport, int dif)
295 {
296 struct sock *sk;
297
298 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport,
299 dif, 0, &udp_table, NULL);
300 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
301 sk = NULL;
302 return sk;
303 }
304 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
305 #endif
306
307 /* do not use the scratch area len for jumbogram: their length execeeds the
308 * scratch area space; note that the IP6CB flags is still in the first
309 * cacheline, so checking for jumbograms is cheap
310 */
311 static int udp6_skb_len(struct sk_buff *skb)
312 {
313 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
314 }
315
316 /*
317 * This should be easy, if there is something there we
318 * return it, otherwise we block.
319 */
320
321 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
322 int noblock, int flags, int *addr_len)
323 {
324 struct ipv6_pinfo *np = inet6_sk(sk);
325 struct inet_sock *inet = inet_sk(sk);
326 struct sk_buff *skb;
327 unsigned int ulen, copied;
328 int peeked, peeking, off;
329 int err;
330 int is_udplite = IS_UDPLITE(sk);
331 bool checksum_valid = false;
332 int is_udp4;
333
334 if (flags & MSG_ERRQUEUE)
335 return ipv6_recv_error(sk, msg, len, addr_len);
336
337 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
338 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
339
340 try_again:
341 peeking = flags & MSG_PEEK;
342 off = sk_peek_offset(sk, flags);
343 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
344 if (!skb)
345 return err;
346
347 ulen = udp6_skb_len(skb);
348 copied = len;
349 if (copied > ulen - off)
350 copied = ulen - off;
351 else if (copied < ulen)
352 msg->msg_flags |= MSG_TRUNC;
353
354 is_udp4 = (skb->protocol == htons(ETH_P_IP));
355
356 /*
357 * If checksum is needed at all, try to do it while copying the
358 * data. If the data is truncated, or if we only want a partial
359 * coverage checksum (UDP-Lite), do it before the copy.
360 */
361
362 if (copied < ulen || peeking ||
363 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
364 checksum_valid = udp_skb_csum_unnecessary(skb) ||
365 !__udp_lib_checksum_complete(skb);
366 if (!checksum_valid)
367 goto csum_copy_err;
368 }
369
370 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
371 if (udp_skb_is_linear(skb))
372 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
373 else
374 err = skb_copy_datagram_msg(skb, off, msg, copied);
375 } else {
376 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
377 if (err == -EINVAL)
378 goto csum_copy_err;
379 }
380 if (unlikely(err)) {
381 if (!peeked) {
382 atomic_inc(&sk->sk_drops);
383 if (is_udp4)
384 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
385 is_udplite);
386 else
387 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
388 is_udplite);
389 }
390 kfree_skb(skb);
391 return err;
392 }
393 if (!peeked) {
394 if (is_udp4)
395 UDP_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
396 is_udplite);
397 else
398 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS,
399 is_udplite);
400 }
401
402 sock_recv_ts_and_drops(msg, sk, skb);
403
404 /* Copy the address. */
405 if (msg->msg_name) {
406 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
407 sin6->sin6_family = AF_INET6;
408 sin6->sin6_port = udp_hdr(skb)->source;
409 sin6->sin6_flowinfo = 0;
410
411 if (is_udp4) {
412 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
413 &sin6->sin6_addr);
414 sin6->sin6_scope_id = 0;
415 } else {
416 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
417 sin6->sin6_scope_id =
418 ipv6_iface_scope_id(&sin6->sin6_addr,
419 inet6_iif(skb));
420 }
421 *addr_len = sizeof(*sin6);
422 }
423
424 if (np->rxopt.all)
425 ip6_datagram_recv_common_ctl(sk, msg, skb);
426
427 if (is_udp4) {
428 if (inet->cmsg_flags)
429 ip_cmsg_recv_offset(msg, sk, skb,
430 sizeof(struct udphdr), off);
431 } else {
432 if (np->rxopt.all)
433 ip6_datagram_recv_specific_ctl(sk, msg, skb);
434 }
435
436 err = copied;
437 if (flags & MSG_TRUNC)
438 err = ulen;
439
440 skb_consume_udp(sk, skb, peeking ? -err : err);
441 return err;
442
443 csum_copy_err:
444 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
445 udp_skb_destructor)) {
446 if (is_udp4) {
447 UDP_INC_STATS(sock_net(sk),
448 UDP_MIB_CSUMERRORS, is_udplite);
449 UDP_INC_STATS(sock_net(sk),
450 UDP_MIB_INERRORS, is_udplite);
451 } else {
452 UDP6_INC_STATS(sock_net(sk),
453 UDP_MIB_CSUMERRORS, is_udplite);
454 UDP6_INC_STATS(sock_net(sk),
455 UDP_MIB_INERRORS, is_udplite);
456 }
457 }
458 kfree_skb(skb);
459
460 /* starting over for a new packet, but check if we need to yield */
461 cond_resched();
462 msg->msg_flags &= ~MSG_TRUNC;
463 goto try_again;
464 }
465
466 void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
467 u8 type, u8 code, int offset, __be32 info,
468 struct udp_table *udptable)
469 {
470 struct ipv6_pinfo *np;
471 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
472 const struct in6_addr *saddr = &hdr->saddr;
473 const struct in6_addr *daddr = &hdr->daddr;
474 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
475 struct sock *sk;
476 int harderr;
477 int err;
478 struct net *net = dev_net(skb->dev);
479
480 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
481 inet6_iif(skb), 0, udptable, skb);
482 if (!sk) {
483 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
484 ICMP6_MIB_INERRORS);
485 return;
486 }
487
488 harderr = icmpv6_err_convert(type, code, &err);
489 np = inet6_sk(sk);
490
491 if (type == ICMPV6_PKT_TOOBIG) {
492 if (!ip6_sk_accept_pmtu(sk))
493 goto out;
494 ip6_sk_update_pmtu(skb, sk, info);
495 if (np->pmtudisc != IPV6_PMTUDISC_DONT)
496 harderr = 1;
497 }
498 if (type == NDISC_REDIRECT) {
499 ip6_sk_redirect(skb, sk);
500 goto out;
501 }
502
503 if (!np->recverr) {
504 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
505 goto out;
506 } else {
507 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
508 }
509
510 sk->sk_err = err;
511 sk->sk_error_report(sk);
512 out:
513 return;
514 }
515
516 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
517 {
518 int rc;
519
520 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
521 sock_rps_save_rxhash(sk, skb);
522 sk_mark_napi_id(sk, skb);
523 sk_incoming_cpu_update(sk);
524 } else {
525 sk_mark_napi_id_once(sk, skb);
526 }
527
528 rc = __udp_enqueue_schedule_skb(sk, skb);
529 if (rc < 0) {
530 int is_udplite = IS_UDPLITE(sk);
531
532 /* Note that an ENOMEM error is charged twice */
533 if (rc == -ENOMEM)
534 UDP6_INC_STATS(sock_net(sk),
535 UDP_MIB_RCVBUFERRORS, is_udplite);
536 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
537 kfree_skb(skb);
538 return -1;
539 }
540
541 return 0;
542 }
543
544 static __inline__ void udpv6_err(struct sk_buff *skb,
545 struct inet6_skb_parm *opt, u8 type,
546 u8 code, int offset, __be32 info)
547 {
548 __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
549 }
550
551 static DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
552 void udpv6_encap_enable(void)
553 {
554 static_branch_enable(&udpv6_encap_needed_key);
555 }
556 EXPORT_SYMBOL(udpv6_encap_enable);
557
558 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
559 {
560 struct udp_sock *up = udp_sk(sk);
561 int is_udplite = IS_UDPLITE(sk);
562
563 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
564 goto drop;
565
566 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) {
567 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
568
569 /*
570 * This is an encapsulation socket so pass the skb to
571 * the socket's udp_encap_rcv() hook. Otherwise, just
572 * fall through and pass this up the UDP socket.
573 * up->encap_rcv() returns the following value:
574 * =0 if skb was successfully passed to the encap
575 * handler or was discarded by it.
576 * >0 if skb should be passed on to UDP.
577 * <0 if skb should be resubmitted as proto -N
578 */
579
580 /* if we're overly short, let UDP handle it */
581 encap_rcv = READ_ONCE(up->encap_rcv);
582 if (encap_rcv) {
583 int ret;
584
585 /* Verify checksum before giving to encap */
586 if (udp_lib_checksum_complete(skb))
587 goto csum_error;
588
589 ret = encap_rcv(sk, skb);
590 if (ret <= 0) {
591 __UDP_INC_STATS(sock_net(sk),
592 UDP_MIB_INDATAGRAMS,
593 is_udplite);
594 return -ret;
595 }
596 }
597
598 /* FALLTHROUGH -- it's a UDP Packet */
599 }
600
601 /*
602 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
603 */
604 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
605
606 if (up->pcrlen == 0) { /* full coverage was set */
607 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
608 UDP_SKB_CB(skb)->cscov, skb->len);
609 goto drop;
610 }
611 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
612 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
613 UDP_SKB_CB(skb)->cscov, up->pcrlen);
614 goto drop;
615 }
616 }
617
618 prefetch(&sk->sk_rmem_alloc);
619 if (rcu_access_pointer(sk->sk_filter) &&
620 udp_lib_checksum_complete(skb))
621 goto csum_error;
622
623 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
624 goto drop;
625
626 udp_csum_pull_header(skb);
627
628 skb_dst_drop(skb);
629
630 return __udpv6_queue_rcv_skb(sk, skb);
631
632 csum_error:
633 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
634 drop:
635 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
636 atomic_inc(&sk->sk_drops);
637 kfree_skb(skb);
638 return -1;
639 }
640
641 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
642 __be16 loc_port, const struct in6_addr *loc_addr,
643 __be16 rmt_port, const struct in6_addr *rmt_addr,
644 int dif, unsigned short hnum)
645 {
646 struct inet_sock *inet = inet_sk(sk);
647
648 if (!net_eq(sock_net(sk), net))
649 return false;
650
651 if (udp_sk(sk)->udp_port_hash != hnum ||
652 sk->sk_family != PF_INET6 ||
653 (inet->inet_dport && inet->inet_dport != rmt_port) ||
654 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
655 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
656 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) ||
657 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
658 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
659 return false;
660 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
661 return false;
662 return true;
663 }
664
665 static void udp6_csum_zero_error(struct sk_buff *skb)
666 {
667 /* RFC 2460 section 8.1 says that we SHOULD log
668 * this error. Well, it is reasonable.
669 */
670 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
671 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
672 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
673 }
674
675 /*
676 * Note: called only from the BH handler context,
677 * so we don't need to lock the hashes.
678 */
679 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
680 const struct in6_addr *saddr, const struct in6_addr *daddr,
681 struct udp_table *udptable, int proto)
682 {
683 struct sock *sk, *first = NULL;
684 const struct udphdr *uh = udp_hdr(skb);
685 unsigned short hnum = ntohs(uh->dest);
686 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
687 unsigned int offset = offsetof(typeof(*sk), sk_node);
688 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
689 int dif = inet6_iif(skb);
690 struct hlist_node *node;
691 struct sk_buff *nskb;
692
693 if (use_hash2) {
694 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) &
695 udptable->mask;
696 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask;
697 start_lookup:
698 hslot = &udptable->hash2[hash2];
699 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
700 }
701
702 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
703 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
704 uh->source, saddr, dif, hnum))
705 continue;
706 /* If zero checksum and no_check is not on for
707 * the socket then skip it.
708 */
709 if (!uh->check && !udp_sk(sk)->no_check6_rx)
710 continue;
711 if (!first) {
712 first = sk;
713 continue;
714 }
715 nskb = skb_clone(skb, GFP_ATOMIC);
716 if (unlikely(!nskb)) {
717 atomic_inc(&sk->sk_drops);
718 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
719 IS_UDPLITE(sk));
720 __UDP6_INC_STATS(net, UDP_MIB_INERRORS,
721 IS_UDPLITE(sk));
722 continue;
723 }
724
725 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
726 consume_skb(nskb);
727 }
728
729 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
730 if (use_hash2 && hash2 != hash2_any) {
731 hash2 = hash2_any;
732 goto start_lookup;
733 }
734
735 if (first) {
736 if (udpv6_queue_rcv_skb(first, skb) > 0)
737 consume_skb(skb);
738 } else {
739 kfree_skb(skb);
740 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
741 proto == IPPROTO_UDPLITE);
742 }
743 return 0;
744 }
745
746 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
747 {
748 if (udp_sk_rx_dst_set(sk, dst)) {
749 const struct rt6_info *rt = (const struct rt6_info *)dst;
750
751 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
752 }
753 }
754
755 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
756 int proto)
757 {
758 const struct in6_addr *saddr, *daddr;
759 struct net *net = dev_net(skb->dev);
760 struct udphdr *uh;
761 struct sock *sk;
762 u32 ulen = 0;
763
764 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
765 goto discard;
766
767 saddr = &ipv6_hdr(skb)->saddr;
768 daddr = &ipv6_hdr(skb)->daddr;
769 uh = udp_hdr(skb);
770
771 ulen = ntohs(uh->len);
772 if (ulen > skb->len)
773 goto short_packet;
774
775 if (proto == IPPROTO_UDP) {
776 /* UDP validates ulen. */
777
778 /* Check for jumbo payload */
779 if (ulen == 0)
780 ulen = skb->len;
781
782 if (ulen < sizeof(*uh))
783 goto short_packet;
784
785 if (ulen < skb->len) {
786 if (pskb_trim_rcsum(skb, ulen))
787 goto short_packet;
788 saddr = &ipv6_hdr(skb)->saddr;
789 daddr = &ipv6_hdr(skb)->daddr;
790 uh = udp_hdr(skb);
791 }
792 }
793
794 if (udp6_csum_init(skb, uh, proto))
795 goto csum_error;
796
797 /* Check if the socket is already available, e.g. due to early demux */
798 sk = skb_steal_sock(skb);
799 if (sk) {
800 struct dst_entry *dst = skb_dst(skb);
801 int ret;
802
803 if (unlikely(sk->sk_rx_dst != dst))
804 udp6_sk_rx_dst_set(sk, dst);
805
806 ret = udpv6_queue_rcv_skb(sk, skb);
807 sock_put(sk);
808
809 /* a return value > 0 means to resubmit the input */
810 if (ret > 0)
811 return ret;
812 return 0;
813 }
814
815 /*
816 * Multicast receive code
817 */
818 if (ipv6_addr_is_multicast(daddr))
819 return __udp6_lib_mcast_deliver(net, skb,
820 saddr, daddr, udptable, proto);
821
822 /* Unicast */
823 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
824 if (sk) {
825 int ret;
826
827 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
828 udp6_csum_zero_error(skb);
829 goto csum_error;
830 }
831
832 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
833 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
834 ip6_compute_pseudo);
835
836 ret = udpv6_queue_rcv_skb(sk, skb);
837
838 /* a return value > 0 means to resubmit the input */
839 if (ret > 0)
840 return ret;
841
842 return 0;
843 }
844
845 if (!uh->check) {
846 udp6_csum_zero_error(skb);
847 goto csum_error;
848 }
849
850 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
851 goto discard;
852
853 if (udp_lib_checksum_complete(skb))
854 goto csum_error;
855
856 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
857 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
858
859 kfree_skb(skb);
860 return 0;
861
862 short_packet:
863 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
864 proto == IPPROTO_UDPLITE ? "-Lite" : "",
865 saddr, ntohs(uh->source),
866 ulen, skb->len,
867 daddr, ntohs(uh->dest));
868 goto discard;
869 csum_error:
870 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
871 discard:
872 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
873 kfree_skb(skb);
874 return 0;
875 }
876
877
878 static struct sock *__udp6_lib_demux_lookup(struct net *net,
879 __be16 loc_port, const struct in6_addr *loc_addr,
880 __be16 rmt_port, const struct in6_addr *rmt_addr,
881 int dif, int sdif)
882 {
883 unsigned short hnum = ntohs(loc_port);
884 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum);
885 unsigned int slot2 = hash2 & udp_table.mask;
886 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
887 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
888 struct sock *sk;
889
890 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
891 if (sk->sk_state == TCP_ESTABLISHED &&
892 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif))
893 return sk;
894 /* Only check first socket in chain */
895 break;
896 }
897 return NULL;
898 }
899
900 static void udp_v6_early_demux(struct sk_buff *skb)
901 {
902 struct net *net = dev_net(skb->dev);
903 const struct udphdr *uh;
904 struct sock *sk;
905 struct dst_entry *dst;
906 int dif = skb->dev->ifindex;
907 int sdif = inet6_sdif(skb);
908
909 if (!pskb_may_pull(skb, skb_transport_offset(skb) +
910 sizeof(struct udphdr)))
911 return;
912
913 uh = udp_hdr(skb);
914
915 if (skb->pkt_type == PACKET_HOST)
916 sk = __udp6_lib_demux_lookup(net, uh->dest,
917 &ipv6_hdr(skb)->daddr,
918 uh->source, &ipv6_hdr(skb)->saddr,
919 dif, sdif);
920 else
921 return;
922
923 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
924 return;
925
926 skb->sk = sk;
927 skb->destructor = sock_efree;
928 dst = READ_ONCE(sk->sk_rx_dst);
929
930 if (dst)
931 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
932 if (dst) {
933 /* set noref for now.
934 * any place which wants to hold dst has to call
935 * dst_hold_safe()
936 */
937 skb_dst_set_noref(skb, dst);
938 }
939 }
940
941 static __inline__ int udpv6_rcv(struct sk_buff *skb)
942 {
943 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
944 }
945
946 /*
947 * Throw away all pending data and cancel the corking. Socket is locked.
948 */
949 static void udp_v6_flush_pending_frames(struct sock *sk)
950 {
951 struct udp_sock *up = udp_sk(sk);
952
953 if (up->pending == AF_INET)
954 udp_flush_pending_frames(sk);
955 else if (up->pending) {
956 up->len = 0;
957 up->pending = 0;
958 ip6_flush_pending_frames(sk);
959 }
960 }
961
962 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
963 int addr_len)
964 {
965 /* The following checks are replicated from __ip6_datagram_connect()
966 * and intended to prevent BPF program called below from accessing
967 * bytes that are out of the bound specified by user in addr_len.
968 */
969 if (uaddr->sa_family == AF_INET) {
970 if (__ipv6_only_sock(sk))
971 return -EAFNOSUPPORT;
972 return udp_pre_connect(sk, uaddr, addr_len);
973 }
974
975 if (addr_len < SIN6_LEN_RFC2133)
976 return -EINVAL;
977
978 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr);
979 }
980
981 /**
982 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
983 * @sk: socket we are sending on
984 * @skb: sk_buff containing the filled-in UDP header
985 * (checksum field must be zeroed out)
986 */
987 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
988 const struct in6_addr *saddr,
989 const struct in6_addr *daddr, int len)
990 {
991 unsigned int offset;
992 struct udphdr *uh = udp_hdr(skb);
993 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
994 __wsum csum = 0;
995
996 if (!frags) {
997 /* Only one fragment on the socket. */
998 skb->csum_start = skb_transport_header(skb) - skb->head;
999 skb->csum_offset = offsetof(struct udphdr, check);
1000 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1001 } else {
1002 /*
1003 * HW-checksum won't work as there are two or more
1004 * fragments on the socket so that all csums of sk_buffs
1005 * should be together
1006 */
1007 offset = skb_transport_offset(skb);
1008 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1009 csum = skb->csum;
1010
1011 skb->ip_summed = CHECKSUM_NONE;
1012
1013 do {
1014 csum = csum_add(csum, frags->csum);
1015 } while ((frags = frags->next));
1016
1017 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1018 csum);
1019 if (uh->check == 0)
1020 uh->check = CSUM_MANGLED_0;
1021 }
1022 }
1023
1024 /*
1025 * Sending
1026 */
1027
1028 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6,
1029 struct inet_cork *cork)
1030 {
1031 struct sock *sk = skb->sk;
1032 struct udphdr *uh;
1033 int err = 0;
1034 int is_udplite = IS_UDPLITE(sk);
1035 __wsum csum = 0;
1036 int offset = skb_transport_offset(skb);
1037 int len = skb->len - offset;
1038
1039 /*
1040 * Create a UDP header
1041 */
1042 uh = udp_hdr(skb);
1043 uh->source = fl6->fl6_sport;
1044 uh->dest = fl6->fl6_dport;
1045 uh->len = htons(len);
1046 uh->check = 0;
1047
1048 if (cork->gso_size) {
1049 const int hlen = skb_network_header_len(skb) +
1050 sizeof(struct udphdr);
1051
1052 if (hlen + cork->gso_size > cork->fragsize)
1053 return -EINVAL;
1054 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS)
1055 return -EINVAL;
1056 if (udp_sk(sk)->no_check6_tx)
1057 return -EINVAL;
1058 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
1059 dst_xfrm(skb_dst(skb)))
1060 return -EIO;
1061
1062 skb_shinfo(skb)->gso_size = cork->gso_size;
1063 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1064 goto csum_partial;
1065 }
1066
1067 if (is_udplite)
1068 csum = udplite_csum(skb);
1069 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
1070 skb->ip_summed = CHECKSUM_NONE;
1071 goto send;
1072 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1073 csum_partial:
1074 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1075 goto send;
1076 } else
1077 csum = udp_csum(skb);
1078
1079 /* add protocol-dependent pseudo-header */
1080 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1081 len, fl6->flowi6_proto, csum);
1082 if (uh->check == 0)
1083 uh->check = CSUM_MANGLED_0;
1084
1085 send:
1086 err = ip6_send_skb(skb);
1087 if (err) {
1088 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1089 UDP6_INC_STATS(sock_net(sk),
1090 UDP_MIB_SNDBUFERRORS, is_udplite);
1091 err = 0;
1092 }
1093 } else {
1094 UDP6_INC_STATS(sock_net(sk),
1095 UDP_MIB_OUTDATAGRAMS, is_udplite);
1096 }
1097 return err;
1098 }
1099
1100 static int udp_v6_push_pending_frames(struct sock *sk)
1101 {
1102 struct sk_buff *skb;
1103 struct udp_sock *up = udp_sk(sk);
1104 struct flowi6 fl6;
1105 int err = 0;
1106
1107 if (up->pending == AF_INET)
1108 return udp_push_pending_frames(sk);
1109
1110 /* ip6_finish_skb will release the cork, so make a copy of
1111 * fl6 here.
1112 */
1113 fl6 = inet_sk(sk)->cork.fl.u.ip6;
1114
1115 skb = ip6_finish_skb(sk);
1116 if (!skb)
1117 goto out;
1118
1119 err = udp_v6_send_skb(skb, &fl6, &inet_sk(sk)->cork.base);
1120
1121 out:
1122 up->len = 0;
1123 up->pending = 0;
1124 return err;
1125 }
1126
1127 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1128 {
1129 struct ipv6_txoptions opt_space;
1130 struct udp_sock *up = udp_sk(sk);
1131 struct inet_sock *inet = inet_sk(sk);
1132 struct ipv6_pinfo *np = inet6_sk(sk);
1133 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1134 struct in6_addr *daddr, *final_p, final;
1135 struct ipv6_txoptions *opt = NULL;
1136 struct ipv6_txoptions *opt_to_free = NULL;
1137 struct ip6_flowlabel *flowlabel = NULL;
1138 struct flowi6 fl6;
1139 struct dst_entry *dst;
1140 struct ipcm6_cookie ipc6;
1141 int addr_len = msg->msg_namelen;
1142 bool connected = false;
1143 int ulen = len;
1144 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1145 int err;
1146 int is_udplite = IS_UDPLITE(sk);
1147 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1148
1149 ipcm6_init(&ipc6);
1150 ipc6.gso_size = up->gso_size;
1151 ipc6.sockc.tsflags = sk->sk_tsflags;
1152
1153 /* destination address check */
1154 if (sin6) {
1155 if (addr_len < offsetof(struct sockaddr, sa_data))
1156 return -EINVAL;
1157
1158 switch (sin6->sin6_family) {
1159 case AF_INET6:
1160 if (addr_len < SIN6_LEN_RFC2133)
1161 return -EINVAL;
1162 daddr = &sin6->sin6_addr;
1163 if (ipv6_addr_any(daddr) &&
1164 ipv6_addr_v4mapped(&np->saddr))
1165 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1166 daddr);
1167 break;
1168 case AF_INET:
1169 goto do_udp_sendmsg;
1170 case AF_UNSPEC:
1171 msg->msg_name = sin6 = NULL;
1172 msg->msg_namelen = addr_len = 0;
1173 daddr = NULL;
1174 break;
1175 default:
1176 return -EINVAL;
1177 }
1178 } else if (!up->pending) {
1179 if (sk->sk_state != TCP_ESTABLISHED)
1180 return -EDESTADDRREQ;
1181 daddr = &sk->sk_v6_daddr;
1182 } else
1183 daddr = NULL;
1184
1185 if (daddr) {
1186 if (ipv6_addr_v4mapped(daddr)) {
1187 struct sockaddr_in sin;
1188 sin.sin_family = AF_INET;
1189 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1190 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1191 msg->msg_name = &sin;
1192 msg->msg_namelen = sizeof(sin);
1193 do_udp_sendmsg:
1194 if (__ipv6_only_sock(sk))
1195 return -ENETUNREACH;
1196 return udp_sendmsg(sk, msg, len);
1197 }
1198 }
1199
1200 if (up->pending == AF_INET)
1201 return udp_sendmsg(sk, msg, len);
1202
1203 /* Rough check on arithmetic overflow,
1204 better check is made in ip6_append_data().
1205 */
1206 if (len > INT_MAX - sizeof(struct udphdr))
1207 return -EMSGSIZE;
1208
1209 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1210 if (up->pending) {
1211 /*
1212 * There are pending frames.
1213 * The socket lock must be held while it's corked.
1214 */
1215 lock_sock(sk);
1216 if (likely(up->pending)) {
1217 if (unlikely(up->pending != AF_INET6)) {
1218 release_sock(sk);
1219 return -EAFNOSUPPORT;
1220 }
1221 dst = NULL;
1222 goto do_append_data;
1223 }
1224 release_sock(sk);
1225 }
1226 ulen += sizeof(struct udphdr);
1227
1228 memset(&fl6, 0, sizeof(fl6));
1229
1230 if (sin6) {
1231 if (sin6->sin6_port == 0)
1232 return -EINVAL;
1233
1234 fl6.fl6_dport = sin6->sin6_port;
1235 daddr = &sin6->sin6_addr;
1236
1237 if (np->sndflow) {
1238 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1239 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1240 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1241 if (!flowlabel)
1242 return -EINVAL;
1243 }
1244 }
1245
1246 /*
1247 * Otherwise it will be difficult to maintain
1248 * sk->sk_dst_cache.
1249 */
1250 if (sk->sk_state == TCP_ESTABLISHED &&
1251 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1252 daddr = &sk->sk_v6_daddr;
1253
1254 if (addr_len >= sizeof(struct sockaddr_in6) &&
1255 sin6->sin6_scope_id &&
1256 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1257 fl6.flowi6_oif = sin6->sin6_scope_id;
1258 } else {
1259 if (sk->sk_state != TCP_ESTABLISHED)
1260 return -EDESTADDRREQ;
1261
1262 fl6.fl6_dport = inet->inet_dport;
1263 daddr = &sk->sk_v6_daddr;
1264 fl6.flowlabel = np->flow_label;
1265 connected = true;
1266 }
1267
1268 if (!fl6.flowi6_oif)
1269 fl6.flowi6_oif = sk->sk_bound_dev_if;
1270
1271 if (!fl6.flowi6_oif)
1272 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1273
1274 fl6.flowi6_mark = sk->sk_mark;
1275 fl6.flowi6_uid = sk->sk_uid;
1276
1277 if (msg->msg_controllen) {
1278 opt = &opt_space;
1279 memset(opt, 0, sizeof(struct ipv6_txoptions));
1280 opt->tot_len = sizeof(*opt);
1281 ipc6.opt = opt;
1282
1283 err = udp_cmsg_send(sk, msg, &ipc6.gso_size);
1284 if (err > 0)
1285 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6,
1286 &ipc6);
1287 if (err < 0) {
1288 fl6_sock_release(flowlabel);
1289 return err;
1290 }
1291 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1292 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1293 if (!flowlabel)
1294 return -EINVAL;
1295 }
1296 if (!(opt->opt_nflen|opt->opt_flen))
1297 opt = NULL;
1298 connected = false;
1299 }
1300 if (!opt) {
1301 opt = txopt_get(np);
1302 opt_to_free = opt;
1303 }
1304 if (flowlabel)
1305 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1306 opt = ipv6_fixup_options(&opt_space, opt);
1307 ipc6.opt = opt;
1308
1309 fl6.flowi6_proto = sk->sk_protocol;
1310 if (!ipv6_addr_any(daddr))
1311 fl6.daddr = *daddr;
1312 else
1313 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1314 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1315 fl6.saddr = np->saddr;
1316 fl6.fl6_sport = inet->inet_sport;
1317
1318 if (cgroup_bpf_enabled && !connected) {
1319 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk,
1320 (struct sockaddr *)sin6, &fl6.saddr);
1321 if (err)
1322 goto out_no_dst;
1323 if (sin6) {
1324 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
1325 /* BPF program rewrote IPv6-only by IPv4-mapped
1326 * IPv6. It's currently unsupported.
1327 */
1328 err = -ENOTSUPP;
1329 goto out_no_dst;
1330 }
1331 if (sin6->sin6_port == 0) {
1332 /* BPF program set invalid port. Reject it. */
1333 err = -EINVAL;
1334 goto out_no_dst;
1335 }
1336 fl6.fl6_dport = sin6->sin6_port;
1337 fl6.daddr = sin6->sin6_addr;
1338 }
1339 }
1340
1341 final_p = fl6_update_dst(&fl6, opt, &final);
1342 if (final_p)
1343 connected = false;
1344
1345 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1346 fl6.flowi6_oif = np->mcast_oif;
1347 connected = false;
1348 } else if (!fl6.flowi6_oif)
1349 fl6.flowi6_oif = np->ucast_oif;
1350
1351 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1352
1353 if (ipc6.tclass < 0)
1354 ipc6.tclass = np->tclass;
1355
1356 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel);
1357
1358 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, connected);
1359 if (IS_ERR(dst)) {
1360 err = PTR_ERR(dst);
1361 dst = NULL;
1362 goto out;
1363 }
1364
1365 if (ipc6.hlimit < 0)
1366 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1367
1368 if (msg->msg_flags&MSG_CONFIRM)
1369 goto do_confirm;
1370 back_from_confirm:
1371
1372 /* Lockless fast path for the non-corking case */
1373 if (!corkreq) {
1374 struct inet_cork_full cork;
1375 struct sk_buff *skb;
1376
1377 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1378 sizeof(struct udphdr), &ipc6,
1379 &fl6, (struct rt6_info *)dst,
1380 msg->msg_flags, &cork);
1381 err = PTR_ERR(skb);
1382 if (!IS_ERR_OR_NULL(skb))
1383 err = udp_v6_send_skb(skb, &fl6, &cork.base);
1384 goto out;
1385 }
1386
1387 lock_sock(sk);
1388 if (unlikely(up->pending)) {
1389 /* The socket is already corked while preparing it. */
1390 /* ... which is an evident application bug. --ANK */
1391 release_sock(sk);
1392
1393 net_dbg_ratelimited("udp cork app bug 2\n");
1394 err = -EINVAL;
1395 goto out;
1396 }
1397
1398 up->pending = AF_INET6;
1399
1400 do_append_data:
1401 if (ipc6.dontfrag < 0)
1402 ipc6.dontfrag = np->dontfrag;
1403 up->len += ulen;
1404 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1405 &ipc6, &fl6, (struct rt6_info *)dst,
1406 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1407 if (err)
1408 udp_v6_flush_pending_frames(sk);
1409 else if (!corkreq)
1410 err = udp_v6_push_pending_frames(sk);
1411 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1412 up->pending = 0;
1413
1414 if (err > 0)
1415 err = np->recverr ? net_xmit_errno(err) : 0;
1416 release_sock(sk);
1417
1418 out:
1419 dst_release(dst);
1420 out_no_dst:
1421 fl6_sock_release(flowlabel);
1422 txopt_put(opt_to_free);
1423 if (!err)
1424 return len;
1425 /*
1426 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1427 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1428 * we don't have a good statistic (IpOutDiscards but it can be too many
1429 * things). We could add another new stat but at least for now that
1430 * seems like overkill.
1431 */
1432 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1433 UDP6_INC_STATS(sock_net(sk),
1434 UDP_MIB_SNDBUFERRORS, is_udplite);
1435 }
1436 return err;
1437
1438 do_confirm:
1439 if (msg->msg_flags & MSG_PROBE)
1440 dst_confirm_neigh(dst, &fl6.daddr);
1441 if (!(msg->msg_flags&MSG_PROBE) || len)
1442 goto back_from_confirm;
1443 err = 0;
1444 goto out;
1445 }
1446
1447 void udpv6_destroy_sock(struct sock *sk)
1448 {
1449 struct udp_sock *up = udp_sk(sk);
1450 lock_sock(sk);
1451 udp_v6_flush_pending_frames(sk);
1452 release_sock(sk);
1453
1454 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) {
1455 void (*encap_destroy)(struct sock *sk);
1456 encap_destroy = READ_ONCE(up->encap_destroy);
1457 if (encap_destroy)
1458 encap_destroy(sk);
1459 }
1460
1461 inet6_destroy_sock(sk);
1462 }
1463
1464 /*
1465 * Socket option code for UDP
1466 */
1467 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1468 char __user *optval, unsigned int optlen)
1469 {
1470 if (level == SOL_UDP || level == SOL_UDPLITE)
1471 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1472 udp_v6_push_pending_frames);
1473 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1474 }
1475
1476 #ifdef CONFIG_COMPAT
1477 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1478 char __user *optval, unsigned int optlen)
1479 {
1480 if (level == SOL_UDP || level == SOL_UDPLITE)
1481 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1482 udp_v6_push_pending_frames);
1483 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1484 }
1485 #endif
1486
1487 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1488 char __user *optval, int __user *optlen)
1489 {
1490 if (level == SOL_UDP || level == SOL_UDPLITE)
1491 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1492 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1493 }
1494
1495 #ifdef CONFIG_COMPAT
1496 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1497 char __user *optval, int __user *optlen)
1498 {
1499 if (level == SOL_UDP || level == SOL_UDPLITE)
1500 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1501 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1502 }
1503 #endif
1504
1505 /* thinking of making this const? Don't.
1506 * early_demux can change based on sysctl.
1507 */
1508 static struct inet6_protocol udpv6_protocol = {
1509 .early_demux = udp_v6_early_demux,
1510 .early_demux_handler = udp_v6_early_demux,
1511 .handler = udpv6_rcv,
1512 .err_handler = udpv6_err,
1513 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1514 };
1515
1516 /* ------------------------------------------------------------------------ */
1517 #ifdef CONFIG_PROC_FS
1518 int udp6_seq_show(struct seq_file *seq, void *v)
1519 {
1520 if (v == SEQ_START_TOKEN) {
1521 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1522 } else {
1523 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1524 struct inet_sock *inet = inet_sk(v);
1525 __u16 srcp = ntohs(inet->inet_sport);
1526 __u16 destp = ntohs(inet->inet_dport);
1527 __ip6_dgram_sock_seq_show(seq, v, srcp, destp,
1528 udp_rqueue_get(v), bucket);
1529 }
1530 return 0;
1531 }
1532
1533 const struct seq_operations udp6_seq_ops = {
1534 .start = udp_seq_start,
1535 .next = udp_seq_next,
1536 .stop = udp_seq_stop,
1537 .show = udp6_seq_show,
1538 };
1539 EXPORT_SYMBOL(udp6_seq_ops);
1540
1541 static struct udp_seq_afinfo udp6_seq_afinfo = {
1542 .family = AF_INET6,
1543 .udp_table = &udp_table,
1544 };
1545
1546 int __net_init udp6_proc_init(struct net *net)
1547 {
1548 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops,
1549 sizeof(struct udp_iter_state), &udp6_seq_afinfo))
1550 return -ENOMEM;
1551 return 0;
1552 }
1553
1554 void udp6_proc_exit(struct net *net)
1555 {
1556 remove_proc_entry("udp6", net->proc_net);
1557 }
1558 #endif /* CONFIG_PROC_FS */
1559
1560 /* ------------------------------------------------------------------------ */
1561
1562 struct proto udpv6_prot = {
1563 .name = "UDPv6",
1564 .owner = THIS_MODULE,
1565 .close = udp_lib_close,
1566 .pre_connect = udpv6_pre_connect,
1567 .connect = ip6_datagram_connect,
1568 .disconnect = udp_disconnect,
1569 .ioctl = udp_ioctl,
1570 .init = udp_init_sock,
1571 .destroy = udpv6_destroy_sock,
1572 .setsockopt = udpv6_setsockopt,
1573 .getsockopt = udpv6_getsockopt,
1574 .sendmsg = udpv6_sendmsg,
1575 .recvmsg = udpv6_recvmsg,
1576 .release_cb = ip6_datagram_release_cb,
1577 .hash = udp_lib_hash,
1578 .unhash = udp_lib_unhash,
1579 .rehash = udp_v6_rehash,
1580 .get_port = udp_v6_get_port,
1581 .memory_allocated = &udp_memory_allocated,
1582 .sysctl_mem = sysctl_udp_mem,
1583 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
1584 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
1585 .obj_size = sizeof(struct udp6_sock),
1586 .h.udp_table = &udp_table,
1587 #ifdef CONFIG_COMPAT
1588 .compat_setsockopt = compat_udpv6_setsockopt,
1589 .compat_getsockopt = compat_udpv6_getsockopt,
1590 #endif
1591 .diag_destroy = udp_abort,
1592 };
1593
1594 static struct inet_protosw udpv6_protosw = {
1595 .type = SOCK_DGRAM,
1596 .protocol = IPPROTO_UDP,
1597 .prot = &udpv6_prot,
1598 .ops = &inet6_dgram_ops,
1599 .flags = INET_PROTOSW_PERMANENT,
1600 };
1601
1602 int __init udpv6_init(void)
1603 {
1604 int ret;
1605
1606 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1607 if (ret)
1608 goto out;
1609
1610 ret = inet6_register_protosw(&udpv6_protosw);
1611 if (ret)
1612 goto out_udpv6_protocol;
1613 out:
1614 return ret;
1615
1616 out_udpv6_protocol:
1617 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1618 goto out;
1619 }
1620
1621 void udpv6_exit(void)
1622 {
1623 inet6_unregister_protosw(&udpv6_protosw);
1624 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1625 }
(deprecated) Btrfs devel tree