1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * UDP over IPv6
4 * Linux INET6 implementation
5 *
6 * Authors:
7 * Pedro Roque <roque@di.fc.ul.pt>
8 *
9 * Based on linux/ipv4/udp.c
10 *
11 * Fixes:
12 * Hideaki YOSHIFUJI : sin6_scope_id support
13 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
14 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
15 * a single port at the same time.
16 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data
17 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file.
18 */
19
20 #include <linux/bpf-cgroup.h>
21 #include <linux/errno.h>
22 #include <linux/types.h>
23 #include <linux/socket.h>
24 #include <linux/sockios.h>
25 #include <linux/net.h>
26 #include <linux/in6.h>
27 #include <linux/netdevice.h>
28 #include <linux/if_arp.h>
29 #include <linux/ipv6.h>
30 #include <linux/icmpv6.h>
31 #include <linux/init.h>
32 #include <linux/module.h>
33 #include <linux/skbuff.h>
34 #include <linux/slab.h>
35 #include <linux/uaccess.h>
36 #include <linux/indirect_call_wrapper.h>
37 #include <trace/events/udp.h>
38
39 #include <net/addrconf.h>
40 #include <net/ndisc.h>
41 #include <net/protocol.h>
42 #include <net/transp_v6.h>
43 #include <net/ip6_route.h>
44 #include <net/raw.h>
45 #include <net/seg6.h>
46 #include <net/tcp_states.h>
47 #include <net/ip6_checksum.h>
48 #include <net/ip6_tunnel.h>
49 #include <net/xfrm.h>
50 #include <net/inet_hashtables.h>
51 #include <net/inet6_hashtables.h>
52 #include <net/busy_poll.h>
53 #include <net/sock_reuseport.h>
54 #include <net/gro.h>
55
56 #include <linux/proc_fs.h>
57 #include <linux/seq_file.h>
58 #include <trace/events/skb.h>
59 #include "udp_impl.h"
60
61 static void udpv6_destruct_sock(struct sock *sk)
62 {
63 udp_destruct_common(sk);
64 inet6_sock_destruct(sk);
65 }
66
67 int udpv6_init_sock(struct sock *sk)
68 {
69 udp_lib_init_sock(sk);
70 sk->sk_destruct = udpv6_destruct_sock;
71 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
72 return 0;
73 }
74
75 INDIRECT_CALLABLE_SCOPE
76 u32 udp6_ehashfn(const struct net *net,
77 const struct in6_addr *laddr,
78 const u16 lport,
79 const struct in6_addr *faddr,
80 const __be16 fport)
81 {
82 u32 lhash, fhash;
83
84 net_get_random_once(&udp6_ehash_secret,
85 sizeof(udp6_ehash_secret));
86 net_get_random_once(&udp_ipv6_hash_secret,
87 sizeof(udp_ipv6_hash_secret));
88
89 lhash = (__force u32)laddr->s6_addr32[3];
90 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
91
92 return __inet6_ehashfn(lhash, lport, fhash, fport,
93 udp6_ehash_secret + net_hash_mix(net));
94 }
95
96 int udp_v6_get_port(struct sock *sk, unsigned short snum)
97 {
98 unsigned int hash2_nulladdr =
99 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
100 unsigned int hash2_partial =
101 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
102
103 /* precompute partial secondary hash */
104 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
105 return udp_lib_get_port(sk, snum, hash2_nulladdr);
106 }
107
108 void udp_v6_rehash(struct sock *sk)
109 {
110 u16 new_hash = ipv6_portaddr_hash(sock_net(sk),
111 &sk->sk_v6_rcv_saddr,
112 inet_sk(sk)->inet_num);
113
114 udp_lib_rehash(sk, new_hash);
115 }
116
117 static int compute_score(struct sock *sk, struct net *net,
118 const struct in6_addr *saddr, __be16 sport,
119 const struct in6_addr *daddr, unsigned short hnum,
120 int dif, int sdif)
121 {
122 int bound_dev_if, score;
123 struct inet_sock *inet;
124 bool dev_match;
125
126 if (!net_eq(sock_net(sk), net) ||
127 udp_sk(sk)->udp_port_hash != hnum ||
128 sk->sk_family != PF_INET6)
129 return -1;
130
131 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
132 return -1;
133
134 score = 0;
135 inet = inet_sk(sk);
136
137 if (inet->inet_dport) {
138 if (inet->inet_dport != sport)
139 return -1;
140 score++;
141 }
142
143 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
144 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
145 return -1;
146 score++;
147 }
148
149 bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
150 dev_match = udp_sk_bound_dev_eq(net, bound_dev_if, dif, sdif);
151 if (!dev_match)
152 return -1;
153 if (bound_dev_if)
154 score++;
155
156 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
157 score++;
158
159 return score;
160 }
161
162 /* called with rcu_read_lock() */
163 static struct sock *udp6_lib_lookup2(struct net *net,
164 const struct in6_addr *saddr, __be16 sport,
165 const struct in6_addr *daddr, unsigned int hnum,
166 int dif, int sdif, struct udp_hslot *hslot2,
167 struct sk_buff *skb)
168 {
169 struct sock *sk, *result;
170 int score, badness;
171 bool need_rescore;
172
173 result = NULL;
174 badness = -1;
175 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
176 need_rescore = false;
177 rescore:
178 score = compute_score(need_rescore ? result : sk, net, saddr,
179 sport, daddr, hnum, dif, sdif);
180 if (score > badness) {
181 badness = score;
182
183 if (need_rescore)
184 continue;
185
186 if (sk->sk_state == TCP_ESTABLISHED) {
187 result = sk;
188 continue;
189 }
190
191 result = inet6_lookup_reuseport(net, sk, skb, sizeof(struct udphdr),
192 saddr, sport, daddr, hnum, udp6_ehashfn);
193 if (!result) {
194 result = sk;
195 continue;
196 }
197
198 /* Fall back to scoring if group has connections */
199 if (!reuseport_has_conns(sk))
200 return result;
201
202 /* Reuseport logic returned an error, keep original score. */
203 if (IS_ERR(result))
204 continue;
205
206 /* compute_score is too long of a function to be
207 * inlined, and calling it again here yields
208 * measureable overhead for some
209 * workloads. Work around it by jumping
210 * backwards to rescore 'result'.
211 */
212 need_rescore = true;
213 goto rescore;
214 }
215 }
216 return result;
217 }
218
219 /* rcu_read_lock() must be held */
220 struct sock *__udp6_lib_lookup(struct net *net,
221 const struct in6_addr *saddr, __be16 sport,
222 const struct in6_addr *daddr, __be16 dport,
223 int dif, int sdif, struct udp_table *udptable,
224 struct sk_buff *skb)
225 {
226 unsigned short hnum = ntohs(dport);
227 unsigned int hash2, slot2;
228 struct udp_hslot *hslot2;
229 struct sock *result, *sk;
230
231 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
232 slot2 = hash2 & udptable->mask;
233 hslot2 = &udptable->hash2[slot2];
234
235 /* Lookup connected or non-wildcard sockets */
236 result = udp6_lib_lookup2(net, saddr, sport,
237 daddr, hnum, dif, sdif,
238 hslot2, skb);
239 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
240 goto done;
241
242 /* Lookup redirect from BPF */
243 if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
244 udptable == net->ipv4.udp_table) {
245 sk = inet6_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr),
246 saddr, sport, daddr, hnum, dif,
247 udp6_ehashfn);
248 if (sk) {
249 result = sk;
250 goto done;
251 }
252 }
253
254 /* Got non-wildcard socket or error on first lookup */
255 if (result)
256 goto done;
257
258 /* Lookup wildcard sockets */
259 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
260 slot2 = hash2 & udptable->mask;
261 hslot2 = &udptable->hash2[slot2];
262
263 result = udp6_lib_lookup2(net, saddr, sport,
264 &in6addr_any, hnum, dif, sdif,
265 hslot2, skb);
266 done:
267 if (IS_ERR(result))
268 return NULL;
269 return result;
270 }
271 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
272
273 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
274 __be16 sport, __be16 dport,
275 struct udp_table *udptable)
276 {
277 const struct ipv6hdr *iph = ipv6_hdr(skb);
278
279 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
280 &iph->daddr, dport, inet6_iif(skb),
281 inet6_sdif(skb), udptable, skb);
282 }
283
284 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb,
285 __be16 sport, __be16 dport)
286 {
287 const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
288 const struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + offset);
289 struct net *net = dev_net(skb->dev);
290 int iif, sdif;
291
292 inet6_get_iif_sdif(skb, &iif, &sdif);
293
294 return __udp6_lib_lookup(net, &iph->saddr, sport,
295 &iph->daddr, dport, iif,
296 sdif, net->ipv4.udp_table, NULL);
297 }
298
299 /* Must be called under rcu_read_lock().
300 * Does increment socket refcount.
301 */
302 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
303 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
304 const struct in6_addr *daddr, __be16 dport, int dif)
305 {
306 struct sock *sk;
307
308 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport,
309 dif, 0, net->ipv4.udp_table, NULL);
310 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
311 sk = NULL;
312 return sk;
313 }
314 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
315 #endif
316
317 /* do not use the scratch area len for jumbogram: their length execeeds the
318 * scratch area space; note that the IP6CB flags is still in the first
319 * cacheline, so checking for jumbograms is cheap
320 */
321 static int udp6_skb_len(struct sk_buff *skb)
322 {
323 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
324 }
325
326 /*
327 * This should be easy, if there is something there we
328 * return it, otherwise we block.
329 */
330
331 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
332 int flags, int *addr_len)
333 {
334 struct ipv6_pinfo *np = inet6_sk(sk);
335 struct inet_sock *inet = inet_sk(sk);
336 struct sk_buff *skb;
337 unsigned int ulen, copied;
338 int off, err, peeking = flags & MSG_PEEK;
339 int is_udplite = IS_UDPLITE(sk);
340 struct udp_mib __percpu *mib;
341 bool checksum_valid = false;
342 int is_udp4;
343
344 if (flags & MSG_ERRQUEUE)
345 return ipv6_recv_error(sk, msg, len, addr_len);
346
347 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
348 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
349
350 try_again:
351 off = sk_peek_offset(sk, flags);
352 skb = __skb_recv_udp(sk, flags, &off, &err);
353 if (!skb)
354 return err;
355 if (ccs_socket_post_recvmsg_permission(sk, skb, flags))
356 return -EAGAIN; /* Hope less harmful than -EPERM. */
357
358 ulen = udp6_skb_len(skb);
359 copied = len;
360 if (copied > ulen - off)
361 copied = ulen - off;
362 else if (copied < ulen)
363 msg->msg_flags |= MSG_TRUNC;
364
365 is_udp4 = (skb->protocol == htons(ETH_P_IP));
366 mib = __UDPX_MIB(sk, is_udp4);
367
368 /*
369 * If checksum is needed at all, try to do it while copying the
370 * data. If the data is truncated, or if we only want a partial
371 * coverage checksum (UDP-Lite), do it before the copy.
372 */
373
374 if (copied < ulen || peeking ||
375 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
376 checksum_valid = udp_skb_csum_unnecessary(skb) ||
377 !__udp_lib_checksum_complete(skb);
378 if (!checksum_valid)
379 goto csum_copy_err;
380 }
381
382 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
383 if (udp_skb_is_linear(skb))
384 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
385 else
386 err = skb_copy_datagram_msg(skb, off, msg, copied);
387 } else {
388 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
389 if (err == -EINVAL)
390 goto csum_copy_err;
391 }
392 if (unlikely(err)) {
393 if (!peeking) {
394 atomic_inc(&sk->sk_drops);
395 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
396 }
397 kfree_skb(skb);
398 return err;
399 }
400 if (!peeking)
401 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS);
402
403 sock_recv_cmsgs(msg, sk, skb);
404
405 /* Copy the address. */
406 if (msg->msg_name) {
407 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
408 sin6->sin6_family = AF_INET6;
409 sin6->sin6_port = udp_hdr(skb)->source;
410 sin6->sin6_flowinfo = 0;
411
412 if (is_udp4) {
413 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
414 &sin6->sin6_addr);
415 sin6->sin6_scope_id = 0;
416 } else {
417 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
418 sin6->sin6_scope_id =
419 ipv6_iface_scope_id(&sin6->sin6_addr,
420 inet6_iif(skb));
421 }
422 *addr_len = sizeof(*sin6);
423
424 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk,
425 (struct sockaddr *)sin6,
426 addr_len);
427 }
428
429 if (udp_test_bit(GRO_ENABLED, sk))
430 udp_cmsg_recv(msg, sk, skb);
431
432 if (np->rxopt.all)
433 ip6_datagram_recv_common_ctl(sk, msg, skb);
434
435 if (is_udp4) {
436 if (inet_cmsg_flags(inet))
437 ip_cmsg_recv_offset(msg, sk, skb,
438 sizeof(struct udphdr), off);
439 } else {
440 if (np->rxopt.all)
441 ip6_datagram_recv_specific_ctl(sk, msg, skb);
442 }
443
444 err = copied;
445 if (flags & MSG_TRUNC)
446 err = ulen;
447
448 skb_consume_udp(sk, skb, peeking ? -err : err);
449 return err;
450
451 csum_copy_err:
452 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
453 udp_skb_destructor)) {
454 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS);
455 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
456 }
457 kfree_skb(skb);
458
459 /* starting over for a new packet, but check if we need to yield */
460 cond_resched();
461 msg->msg_flags &= ~MSG_TRUNC;
462 goto try_again;
463 }
464
465 DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
466 void udpv6_encap_enable(void)
467 {
468 static_branch_inc(&udpv6_encap_needed_key);
469 }
470 EXPORT_SYMBOL(udpv6_encap_enable);
471
472 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
473 * through error handlers in encapsulations looking for a match.
474 */
475 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb,
476 struct inet6_skb_parm *opt,
477 u8 type, u8 code, int offset, __be32 info)
478 {
479 int i;
480
481 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
482 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt,
483 u8 type, u8 code, int offset, __be32 info);
484 const struct ip6_tnl_encap_ops *encap;
485
486 encap = rcu_dereference(ip6tun_encaps[i]);
487 if (!encap)
488 continue;
489 handler = encap->err_handler;
490 if (handler && !handler(skb, opt, type, code, offset, info))
491 return 0;
492 }
493
494 return -ENOENT;
495 }
496
497 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
498 * reversing source and destination port: this will match tunnels that force the
499 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
500 * lwtunnels might actually break this assumption by being configured with
501 * different destination ports on endpoints, in this case we won't be able to
502 * trace ICMP messages back to them.
503 *
504 * If this doesn't match any socket, probe tunnels with arbitrary destination
505 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
506 * we've sent packets to won't necessarily match the local destination port.
507 *
508 * Then ask the tunnel implementation to match the error against a valid
509 * association.
510 *
511 * Return an error if we can't find a match, the socket if we need further
512 * processing, zero otherwise.
513 */
514 static struct sock *__udp6_lib_err_encap(struct net *net,
515 const struct ipv6hdr *hdr, int offset,
516 struct udphdr *uh,
517 struct udp_table *udptable,
518 struct sock *sk,
519 struct sk_buff *skb,
520 struct inet6_skb_parm *opt,
521 u8 type, u8 code, __be32 info)
522 {
523 int (*lookup)(struct sock *sk, struct sk_buff *skb);
524 int network_offset, transport_offset;
525 struct udp_sock *up;
526
527 network_offset = skb_network_offset(skb);
528 transport_offset = skb_transport_offset(skb);
529
530 /* Network header needs to point to the outer IPv6 header inside ICMP */
531 skb_reset_network_header(skb);
532
533 /* Transport header needs to point to the UDP header */
534 skb_set_transport_header(skb, offset);
535
536 if (sk) {
537 up = udp_sk(sk);
538
539 lookup = READ_ONCE(up->encap_err_lookup);
540 if (lookup && lookup(sk, skb))
541 sk = NULL;
542
543 goto out;
544 }
545
546 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source,
547 &hdr->saddr, uh->dest,
548 inet6_iif(skb), 0, udptable, skb);
549 if (sk) {
550 up = udp_sk(sk);
551
552 lookup = READ_ONCE(up->encap_err_lookup);
553 if (!lookup || lookup(sk, skb))
554 sk = NULL;
555 }
556
557 out:
558 if (!sk) {
559 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code,
560 offset, info));
561 }
562
563 skb_set_transport_header(skb, transport_offset);
564 skb_set_network_header(skb, network_offset);
565
566 return sk;
567 }
568
569 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
570 u8 type, u8 code, int offset, __be32 info,
571 struct udp_table *udptable)
572 {
573 struct ipv6_pinfo *np;
574 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
575 const struct in6_addr *saddr = &hdr->saddr;
576 const struct in6_addr *daddr = seg6_get_daddr(skb, opt) ? : &hdr->daddr;
577 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
578 bool tunnel = false;
579 struct sock *sk;
580 int harderr;
581 int err;
582 struct net *net = dev_net(skb->dev);
583
584 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
585 inet6_iif(skb), inet6_sdif(skb), udptable, NULL);
586
587 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) {
588 /* No socket for error: try tunnels before discarding */
589 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
590 sk = __udp6_lib_err_encap(net, hdr, offset, uh,
591 udptable, sk, skb,
592 opt, type, code, info);
593 if (!sk)
594 return 0;
595 } else
596 sk = ERR_PTR(-ENOENT);
597
598 if (IS_ERR(sk)) {
599 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
600 ICMP6_MIB_INERRORS);
601 return PTR_ERR(sk);
602 }
603
604 tunnel = true;
605 }
606
607 harderr = icmpv6_err_convert(type, code, &err);
608 np = inet6_sk(sk);
609
610 if (type == ICMPV6_PKT_TOOBIG) {
611 if (!ip6_sk_accept_pmtu(sk))
612 goto out;
613 ip6_sk_update_pmtu(skb, sk, info);
614 if (READ_ONCE(np->pmtudisc) != IPV6_PMTUDISC_DONT)
615 harderr = 1;
616 }
617 if (type == NDISC_REDIRECT) {
618 if (tunnel) {
619 ip6_redirect(skb, sock_net(sk), inet6_iif(skb),
620 READ_ONCE(sk->sk_mark), sk->sk_uid);
621 } else {
622 ip6_sk_redirect(skb, sk);
623 }
624 goto out;
625 }
626
627 /* Tunnels don't have an application socket: don't pass errors back */
628 if (tunnel) {
629 if (udp_sk(sk)->encap_err_rcv)
630 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest,
631 ntohl(info), (u8 *)(uh+1));
632 goto out;
633 }
634
635 if (!inet6_test_bit(RECVERR6, sk)) {
636 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
637 goto out;
638 } else {
639 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
640 }
641
642 sk->sk_err = err;
643 sk_error_report(sk);
644 out:
645 return 0;
646 }
647
648 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
649 {
650 int rc;
651
652 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
653 sock_rps_save_rxhash(sk, skb);
654 sk_mark_napi_id(sk, skb);
655 sk_incoming_cpu_update(sk);
656 } else {
657 sk_mark_napi_id_once(sk, skb);
658 }
659
660 rc = __udp_enqueue_schedule_skb(sk, skb);
661 if (rc < 0) {
662 int is_udplite = IS_UDPLITE(sk);
663 enum skb_drop_reason drop_reason;
664
665 /* Note that an ENOMEM error is charged twice */
666 if (rc == -ENOMEM) {
667 UDP6_INC_STATS(sock_net(sk),
668 UDP_MIB_RCVBUFERRORS, is_udplite);
669 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
670 } else {
671 UDP6_INC_STATS(sock_net(sk),
672 UDP_MIB_MEMERRORS, is_udplite);
673 drop_reason = SKB_DROP_REASON_PROTO_MEM;
674 }
675 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
676 trace_udp_fail_queue_rcv_skb(rc, sk, skb);
677 sk_skb_reason_drop(sk, skb, drop_reason);
678 return -1;
679 }
680
681 return 0;
682 }
683
684 static __inline__ int udpv6_err(struct sk_buff *skb,
685 struct inet6_skb_parm *opt, u8 type,
686 u8 code, int offset, __be32 info)
687 {
688 return __udp6_lib_err(skb, opt, type, code, offset, info,
689 dev_net(skb->dev)->ipv4.udp_table);
690 }
691
692 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
693 {
694 enum skb_drop_reason drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
695 struct udp_sock *up = udp_sk(sk);
696 int is_udplite = IS_UDPLITE(sk);
697
698 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) {
699 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
700 goto drop;
701 }
702 nf_reset_ct(skb);
703
704 if (static_branch_unlikely(&udpv6_encap_needed_key) &&
705 READ_ONCE(up->encap_type)) {
706 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
707
708 /*
709 * This is an encapsulation socket so pass the skb to
710 * the socket's udp_encap_rcv() hook. Otherwise, just
711 * fall through and pass this up the UDP socket.
712 * up->encap_rcv() returns the following value:
713 * =0 if skb was successfully passed to the encap
714 * handler or was discarded by it.
715 * >0 if skb should be passed on to UDP.
716 * <0 if skb should be resubmitted as proto -N
717 */
718
719 /* if we're overly short, let UDP handle it */
720 encap_rcv = READ_ONCE(up->encap_rcv);
721 if (encap_rcv) {
722 int ret;
723
724 /* Verify checksum before giving to encap */
725 if (udp_lib_checksum_complete(skb))
726 goto csum_error;
727
728 ret = encap_rcv(sk, skb);
729 if (ret <= 0) {
730 __UDP6_INC_STATS(sock_net(sk),
731 UDP_MIB_INDATAGRAMS,
732 is_udplite);
733 return -ret;
734 }
735 }
736
737 /* FALLTHROUGH -- it's a UDP Packet */
738 }
739
740 /*
741 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
742 */
743 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) {
744 u16 pcrlen = READ_ONCE(up->pcrlen);
745
746 if (pcrlen == 0) { /* full coverage was set */
747 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
748 UDP_SKB_CB(skb)->cscov, skb->len);
749 goto drop;
750 }
751 if (UDP_SKB_CB(skb)->cscov < pcrlen) {
752 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
753 UDP_SKB_CB(skb)->cscov, pcrlen);
754 goto drop;
755 }
756 }
757
758 prefetch(&sk->sk_rmem_alloc);
759 if (rcu_access_pointer(sk->sk_filter) &&
760 udp_lib_checksum_complete(skb))
761 goto csum_error;
762
763 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
764 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
765 goto drop;
766 }
767
768 udp_csum_pull_header(skb);
769
770 skb_dst_drop(skb);
771
772 return __udpv6_queue_rcv_skb(sk, skb);
773
774 csum_error:
775 drop_reason = SKB_DROP_REASON_UDP_CSUM;
776 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
777 drop:
778 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
779 atomic_inc(&sk->sk_drops);
780 sk_skb_reason_drop(sk, skb, drop_reason);
781 return -1;
782 }
783
784 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
785 {
786 struct sk_buff *next, *segs;
787 int ret;
788
789 if (likely(!udp_unexpected_gso(sk, skb)))
790 return udpv6_queue_rcv_one_skb(sk, skb);
791
792 __skb_push(skb, -skb_mac_offset(skb));
793 segs = udp_rcv_segment(sk, skb, false);
794 skb_list_walk_safe(segs, skb, next) {
795 __skb_pull(skb, skb_transport_offset(skb));
796
797 udp_post_segment_fix_csum(skb);
798 ret = udpv6_queue_rcv_one_skb(sk, skb);
799 if (ret > 0)
800 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret,
801 true);
802 }
803 return 0;
804 }
805
806 static bool __udp_v6_is_mcast_sock(struct net *net, const struct sock *sk,
807 __be16 loc_port, const struct in6_addr *loc_addr,
808 __be16 rmt_port, const struct in6_addr *rmt_addr,
809 int dif, int sdif, unsigned short hnum)
810 {
811 const struct inet_sock *inet = inet_sk(sk);
812
813 if (!net_eq(sock_net(sk), net))
814 return false;
815
816 if (udp_sk(sk)->udp_port_hash != hnum ||
817 sk->sk_family != PF_INET6 ||
818 (inet->inet_dport && inet->inet_dport != rmt_port) ||
819 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
820 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
821 !udp_sk_bound_dev_eq(net, READ_ONCE(sk->sk_bound_dev_if), dif, sdif) ||
822 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
823 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
824 return false;
825 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
826 return false;
827 return true;
828 }
829
830 static void udp6_csum_zero_error(struct sk_buff *skb)
831 {
832 /* RFC 2460 section 8.1 says that we SHOULD log
833 * this error. Well, it is reasonable.
834 */
835 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
836 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
837 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
838 }
839
840 /*
841 * Note: called only from the BH handler context,
842 * so we don't need to lock the hashes.
843 */
844 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
845 const struct in6_addr *saddr, const struct in6_addr *daddr,
846 struct udp_table *udptable, int proto)
847 {
848 struct sock *sk, *first = NULL;
849 const struct udphdr *uh = udp_hdr(skb);
850 unsigned short hnum = ntohs(uh->dest);
851 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
852 unsigned int offset = offsetof(typeof(*sk), sk_node);
853 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
854 int dif = inet6_iif(skb);
855 int sdif = inet6_sdif(skb);
856 struct hlist_node *node;
857 struct sk_buff *nskb;
858
859 if (use_hash2) {
860 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) &
861 udptable->mask;
862 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask;
863 start_lookup:
864 hslot = &udptable->hash2[hash2];
865 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
866 }
867
868 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
869 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
870 uh->source, saddr, dif, sdif,
871 hnum))
872 continue;
873 /* If zero checksum and no_check is not on for
874 * the socket then skip it.
875 */
876 if (!uh->check && !udp_get_no_check6_rx(sk))
877 continue;
878 if (!first) {
879 first = sk;
880 continue;
881 }
882 nskb = skb_clone(skb, GFP_ATOMIC);
883 if (unlikely(!nskb)) {
884 atomic_inc(&sk->sk_drops);
885 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
886 IS_UDPLITE(sk));
887 __UDP6_INC_STATS(net, UDP_MIB_INERRORS,
888 IS_UDPLITE(sk));
889 continue;
890 }
891
892 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
893 consume_skb(nskb);
894 }
895
896 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
897 if (use_hash2 && hash2 != hash2_any) {
898 hash2 = hash2_any;
899 goto start_lookup;
900 }
901
902 if (first) {
903 if (udpv6_queue_rcv_skb(first, skb) > 0)
904 consume_skb(skb);
905 } else {
906 kfree_skb(skb);
907 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
908 proto == IPPROTO_UDPLITE);
909 }
910 return 0;
911 }
912
913 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
914 {
915 if (udp_sk_rx_dst_set(sk, dst))
916 sk->sk_rx_dst_cookie = rt6_get_cookie(dst_rt6_info(dst));
917 }
918
919 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
920 * return code conversion for ip layer consumption
921 */
922 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
923 struct udphdr *uh)
924 {
925 int ret;
926
927 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
928 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo);
929
930 ret = udpv6_queue_rcv_skb(sk, skb);
931
932 /* a return value > 0 means to resubmit the input */
933 if (ret > 0)
934 return ret;
935 return 0;
936 }
937
938 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
939 int proto)
940 {
941 enum skb_drop_reason reason = SKB_DROP_REASON_NOT_SPECIFIED;
942 const struct in6_addr *saddr, *daddr;
943 struct net *net = dev_net(skb->dev);
944 struct sock *sk = NULL;
945 struct udphdr *uh;
946 bool refcounted;
947 u32 ulen = 0;
948
949 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
950 goto discard;
951
952 saddr = &ipv6_hdr(skb)->saddr;
953 daddr = &ipv6_hdr(skb)->daddr;
954 uh = udp_hdr(skb);
955
956 ulen = ntohs(uh->len);
957 if (ulen > skb->len)
958 goto short_packet;
959
960 if (proto == IPPROTO_UDP) {
961 /* UDP validates ulen. */
962
963 /* Check for jumbo payload */
964 if (ulen == 0)
965 ulen = skb->len;
966
967 if (ulen < sizeof(*uh))
968 goto short_packet;
969
970 if (ulen < skb->len) {
971 if (pskb_trim_rcsum(skb, ulen))
972 goto short_packet;
973 saddr = &ipv6_hdr(skb)->saddr;
974 daddr = &ipv6_hdr(skb)->daddr;
975 uh = udp_hdr(skb);
976 }
977 }
978
979 if (udp6_csum_init(skb, uh, proto))
980 goto csum_error;
981
982 /* Check if the socket is already available, e.g. due to early demux */
983 sk = inet6_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest,
984 &refcounted, udp6_ehashfn);
985 if (IS_ERR(sk))
986 goto no_sk;
987
988 if (sk) {
989 struct dst_entry *dst = skb_dst(skb);
990 int ret;
991
992 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
993 udp6_sk_rx_dst_set(sk, dst);
994
995 if (!uh->check && !udp_get_no_check6_rx(sk)) {
996 if (refcounted)
997 sock_put(sk);
998 goto report_csum_error;
999 }
1000
1001 ret = udp6_unicast_rcv_skb(sk, skb, uh);
1002 if (refcounted)
1003 sock_put(sk);
1004 return ret;
1005 }
1006
1007 /*
1008 * Multicast receive code
1009 */
1010 if (ipv6_addr_is_multicast(daddr))
1011 return __udp6_lib_mcast_deliver(net, skb,
1012 saddr, daddr, udptable, proto);
1013
1014 /* Unicast */
1015 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1016 if (sk) {
1017 if (!uh->check && !udp_get_no_check6_rx(sk))
1018 goto report_csum_error;
1019 return udp6_unicast_rcv_skb(sk, skb, uh);
1020 }
1021 no_sk:
1022 reason = SKB_DROP_REASON_NO_SOCKET;
1023
1024 if (!uh->check)
1025 goto report_csum_error;
1026
1027 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1028 goto discard;
1029 nf_reset_ct(skb);
1030
1031 if (udp_lib_checksum_complete(skb))
1032 goto csum_error;
1033
1034 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1035 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
1036
1037 sk_skb_reason_drop(sk, skb, reason);
1038 return 0;
1039
1040 short_packet:
1041 if (reason == SKB_DROP_REASON_NOT_SPECIFIED)
1042 reason = SKB_DROP_REASON_PKT_TOO_SMALL;
1043 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
1044 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1045 saddr, ntohs(uh->source),
1046 ulen, skb->len,
1047 daddr, ntohs(uh->dest));
1048 goto discard;
1049
1050 report_csum_error:
1051 udp6_csum_zero_error(skb);
1052 csum_error:
1053 if (reason == SKB_DROP_REASON_NOT_SPECIFIED)
1054 reason = SKB_DROP_REASON_UDP_CSUM;
1055 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1056 discard:
1057 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1058 sk_skb_reason_drop(sk, skb, reason);
1059 return 0;
1060 }
1061
1062
1063 static struct sock *__udp6_lib_demux_lookup(struct net *net,
1064 __be16 loc_port, const struct in6_addr *loc_addr,
1065 __be16 rmt_port, const struct in6_addr *rmt_addr,
1066 int dif, int sdif)
1067 {
1068 struct udp_table *udptable = net->ipv4.udp_table;
1069 unsigned short hnum = ntohs(loc_port);
1070 unsigned int hash2, slot2;
1071 struct udp_hslot *hslot2;
1072 __portpair ports;
1073 struct sock *sk;
1074
1075 hash2 = ipv6_portaddr_hash(net, loc_addr, hnum);
1076 slot2 = hash2 & udptable->mask;
1077 hslot2 = &udptable->hash2[slot2];
1078 ports = INET_COMBINED_PORTS(rmt_port, hnum);
1079
1080 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
1081 if (sk->sk_state == TCP_ESTABLISHED &&
1082 inet6_match(net, sk, rmt_addr, loc_addr, ports, dif, sdif))
1083 return sk;
1084 /* Only check first socket in chain */
1085 break;
1086 }
1087 return NULL;
1088 }
1089
1090 void udp_v6_early_demux(struct sk_buff *skb)
1091 {
1092 struct net *net = dev_net(skb->dev);
1093 const struct udphdr *uh;
1094 struct sock *sk;
1095 struct dst_entry *dst;
1096 int dif = skb->dev->ifindex;
1097 int sdif = inet6_sdif(skb);
1098
1099 if (!pskb_may_pull(skb, skb_transport_offset(skb) +
1100 sizeof(struct udphdr)))
1101 return;
1102
1103 uh = udp_hdr(skb);
1104
1105 if (skb->pkt_type == PACKET_HOST)
1106 sk = __udp6_lib_demux_lookup(net, uh->dest,
1107 &ipv6_hdr(skb)->daddr,
1108 uh->source, &ipv6_hdr(skb)->saddr,
1109 dif, sdif);
1110 else
1111 return;
1112
1113 if (!sk)
1114 return;
1115
1116 skb->sk = sk;
1117 DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk));
1118 skb->destructor = sock_pfree;
1119 dst = rcu_dereference(sk->sk_rx_dst);
1120
1121 if (dst)
1122 dst = dst_check(dst, sk->sk_rx_dst_cookie);
1123 if (dst) {
1124 /* set noref for now.
1125 * any place which wants to hold dst has to call
1126 * dst_hold_safe()
1127 */
1128 skb_dst_set_noref(skb, dst);
1129 }
1130 }
1131
1132 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb)
1133 {
1134 return __udp6_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
1135 }
1136
1137 /*
1138 * Throw away all pending data and cancel the corking. Socket is locked.
1139 */
1140 static void udp_v6_flush_pending_frames(struct sock *sk)
1141 {
1142 struct udp_sock *up = udp_sk(sk);
1143
1144 if (up->pending == AF_INET)
1145 udp_flush_pending_frames(sk);
1146 else if (up->pending) {
1147 up->len = 0;
1148 WRITE_ONCE(up->pending, 0);
1149 ip6_flush_pending_frames(sk);
1150 }
1151 }
1152
1153 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
1154 int addr_len)
1155 {
1156 if (addr_len < offsetofend(struct sockaddr, sa_family))
1157 return -EINVAL;
1158 /* The following checks are replicated from __ip6_datagram_connect()
1159 * and intended to prevent BPF program called below from accessing
1160 * bytes that are out of the bound specified by user in addr_len.
1161 */
1162 if (uaddr->sa_family == AF_INET) {
1163 if (ipv6_only_sock(sk))
1164 return -EAFNOSUPPORT;
1165 return udp_pre_connect(sk, uaddr, addr_len);
1166 }
1167
1168 if (addr_len < SIN6_LEN_RFC2133)
1169 return -EINVAL;
1170
1171 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr, &addr_len);
1172 }
1173
1174 /**
1175 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
1176 * @sk: socket we are sending on
1177 * @skb: sk_buff containing the filled-in UDP header
1178 * (checksum field must be zeroed out)
1179 * @saddr: source address
1180 * @daddr: destination address
1181 * @len: length of packet
1182 */
1183 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
1184 const struct in6_addr *saddr,
1185 const struct in6_addr *daddr, int len)
1186 {
1187 unsigned int offset;
1188 struct udphdr *uh = udp_hdr(skb);
1189 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
1190 __wsum csum = 0;
1191
1192 if (!frags) {
1193 /* Only one fragment on the socket. */
1194 skb->csum_start = skb_transport_header(skb) - skb->head;
1195 skb->csum_offset = offsetof(struct udphdr, check);
1196 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1197 } else {
1198 /*
1199 * HW-checksum won't work as there are two or more
1200 * fragments on the socket so that all csums of sk_buffs
1201 * should be together
1202 */
1203 offset = skb_transport_offset(skb);
1204 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1205 csum = skb->csum;
1206
1207 skb->ip_summed = CHECKSUM_NONE;
1208
1209 do {
1210 csum = csum_add(csum, frags->csum);
1211 } while ((frags = frags->next));
1212
1213 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1214 csum);
1215 if (uh->check == 0)
1216 uh->check = CSUM_MANGLED_0;
1217 }
1218 }
1219
1220 /*
1221 * Sending
1222 */
1223
1224 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6,
1225 struct inet_cork *cork)
1226 {
1227 struct sock *sk = skb->sk;
1228 struct udphdr *uh;
1229 int err = 0;
1230 int is_udplite = IS_UDPLITE(sk);
1231 __wsum csum = 0;
1232 int offset = skb_transport_offset(skb);
1233 int len = skb->len - offset;
1234 int datalen = len - sizeof(*uh);
1235
1236 /*
1237 * Create a UDP header
1238 */
1239 uh = udp_hdr(skb);
1240 uh->source = fl6->fl6_sport;
1241 uh->dest = fl6->fl6_dport;
1242 uh->len = htons(len);
1243 uh->check = 0;
1244
1245 if (cork->gso_size) {
1246 const int hlen = skb_network_header_len(skb) +
1247 sizeof(struct udphdr);
1248
1249 if (hlen + cork->gso_size > cork->fragsize) {
1250 kfree_skb(skb);
1251 return -EINVAL;
1252 }
1253 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
1254 kfree_skb(skb);
1255 return -EINVAL;
1256 }
1257 if (udp_get_no_check6_tx(sk)) {
1258 kfree_skb(skb);
1259 return -EINVAL;
1260 }
1261 if (is_udplite || dst_xfrm(skb_dst(skb))) {
1262 kfree_skb(skb);
1263 return -EIO;
1264 }
1265
1266 if (datalen > cork->gso_size) {
1267 skb_shinfo(skb)->gso_size = cork->gso_size;
1268 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1269 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
1270 cork->gso_size);
1271
1272 /* Don't checksum the payload, skb will get segmented */
1273 goto csum_partial;
1274 }
1275 }
1276
1277 if (is_udplite)
1278 csum = udplite_csum(skb);
1279 else if (udp_get_no_check6_tx(sk)) { /* UDP csum disabled */
1280 skb->ip_summed = CHECKSUM_NONE;
1281 goto send;
1282 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1283 csum_partial:
1284 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1285 goto send;
1286 } else
1287 csum = udp_csum(skb);
1288
1289 /* add protocol-dependent pseudo-header */
1290 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1291 len, fl6->flowi6_proto, csum);
1292 if (uh->check == 0)
1293 uh->check = CSUM_MANGLED_0;
1294
1295 send:
1296 err = ip6_send_skb(skb);
1297 if (err) {
1298 if (err == -ENOBUFS && !inet6_test_bit(RECVERR6, sk)) {
1299 UDP6_INC_STATS(sock_net(sk),
1300 UDP_MIB_SNDBUFERRORS, is_udplite);
1301 err = 0;
1302 }
1303 } else {
1304 UDP6_INC_STATS(sock_net(sk),
1305 UDP_MIB_OUTDATAGRAMS, is_udplite);
1306 }
1307 return err;
1308 }
1309
1310 static int udp_v6_push_pending_frames(struct sock *sk)
1311 {
1312 struct sk_buff *skb;
1313 struct udp_sock *up = udp_sk(sk);
1314 int err = 0;
1315
1316 if (up->pending == AF_INET)
1317 return udp_push_pending_frames(sk);
1318
1319 skb = ip6_finish_skb(sk);
1320 if (!skb)
1321 goto out;
1322
1323 err = udp_v6_send_skb(skb, &inet_sk(sk)->cork.fl.u.ip6,
1324 &inet_sk(sk)->cork.base);
1325 out:
1326 up->len = 0;
1327 WRITE_ONCE(up->pending, 0);
1328 return err;
1329 }
1330
1331 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1332 {
1333 struct ipv6_txoptions opt_space;
1334 struct udp_sock *up = udp_sk(sk);
1335 struct inet_sock *inet = inet_sk(sk);
1336 struct ipv6_pinfo *np = inet6_sk(sk);
1337 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1338 struct in6_addr *daddr, *final_p, final;
1339 struct ipv6_txoptions *opt = NULL;
1340 struct ipv6_txoptions *opt_to_free = NULL;
1341 struct ip6_flowlabel *flowlabel = NULL;
1342 struct inet_cork_full cork;
1343 struct flowi6 *fl6 = &cork.fl.u.ip6;
1344 struct dst_entry *dst;
1345 struct ipcm6_cookie ipc6;
1346 int addr_len = msg->msg_namelen;
1347 bool connected = false;
1348 int ulen = len;
1349 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
1350 int err;
1351 int is_udplite = IS_UDPLITE(sk);
1352 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1353
1354 ipcm6_init(&ipc6);
1355 ipc6.gso_size = READ_ONCE(up->gso_size);
1356 ipc6.sockc.tsflags = READ_ONCE(sk->sk_tsflags);
1357 ipc6.sockc.mark = READ_ONCE(sk->sk_mark);
1358
1359 /* destination address check */
1360 if (sin6) {
1361 if (addr_len < offsetof(struct sockaddr, sa_data))
1362 return -EINVAL;
1363
1364 switch (sin6->sin6_family) {
1365 case AF_INET6:
1366 if (addr_len < SIN6_LEN_RFC2133)
1367 return -EINVAL;
1368 daddr = &sin6->sin6_addr;
1369 if (ipv6_addr_any(daddr) &&
1370 ipv6_addr_v4mapped(&np->saddr))
1371 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1372 daddr);
1373 break;
1374 case AF_INET:
1375 goto do_udp_sendmsg;
1376 case AF_UNSPEC:
1377 msg->msg_name = sin6 = NULL;
1378 msg->msg_namelen = addr_len = 0;
1379 daddr = NULL;
1380 break;
1381 default:
1382 return -EINVAL;
1383 }
1384 } else if (!READ_ONCE(up->pending)) {
1385 if (sk->sk_state != TCP_ESTABLISHED)
1386 return -EDESTADDRREQ;
1387 daddr = &sk->sk_v6_daddr;
1388 } else
1389 daddr = NULL;
1390
1391 if (daddr) {
1392 if (ipv6_addr_v4mapped(daddr)) {
1393 struct sockaddr_in sin;
1394 sin.sin_family = AF_INET;
1395 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1396 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1397 msg->msg_name = &sin;
1398 msg->msg_namelen = sizeof(sin);
1399 do_udp_sendmsg:
1400 err = ipv6_only_sock(sk) ?
1401 -ENETUNREACH : udp_sendmsg(sk, msg, len);
1402 msg->msg_name = sin6;
1403 msg->msg_namelen = addr_len;
1404 return err;
1405 }
1406 }
1407
1408 /* Rough check on arithmetic overflow,
1409 better check is made in ip6_append_data().
1410 */
1411 if (len > INT_MAX - sizeof(struct udphdr))
1412 return -EMSGSIZE;
1413
1414 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1415 if (READ_ONCE(up->pending)) {
1416 if (READ_ONCE(up->pending) == AF_INET)
1417 return udp_sendmsg(sk, msg, len);
1418 /*
1419 * There are pending frames.
1420 * The socket lock must be held while it's corked.
1421 */
1422 lock_sock(sk);
1423 if (likely(up->pending)) {
1424 if (unlikely(up->pending != AF_INET6)) {
1425 release_sock(sk);
1426 return -EAFNOSUPPORT;
1427 }
1428 dst = NULL;
1429 goto do_append_data;
1430 }
1431 release_sock(sk);
1432 }
1433 ulen += sizeof(struct udphdr);
1434
1435 memset(fl6, 0, sizeof(*fl6));
1436
1437 if (sin6) {
1438 if (sin6->sin6_port == 0)
1439 return -EINVAL;
1440
1441 fl6->fl6_dport = sin6->sin6_port;
1442 daddr = &sin6->sin6_addr;
1443
1444 if (inet6_test_bit(SNDFLOW, sk)) {
1445 fl6->flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1446 if (fl6->flowlabel & IPV6_FLOWLABEL_MASK) {
1447 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel);
1448 if (IS_ERR(flowlabel))
1449 return -EINVAL;
1450 }
1451 }
1452
1453 /*
1454 * Otherwise it will be difficult to maintain
1455 * sk->sk_dst_cache.
1456 */
1457 if (sk->sk_state == TCP_ESTABLISHED &&
1458 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1459 daddr = &sk->sk_v6_daddr;
1460
1461 if (addr_len >= sizeof(struct sockaddr_in6) &&
1462 sin6->sin6_scope_id &&
1463 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1464 fl6->flowi6_oif = sin6->sin6_scope_id;
1465 } else {
1466 if (sk->sk_state != TCP_ESTABLISHED)
1467 return -EDESTADDRREQ;
1468
1469 fl6->fl6_dport = inet->inet_dport;
1470 daddr = &sk->sk_v6_daddr;
1471 fl6->flowlabel = np->flow_label;
1472 connected = true;
1473 }
1474
1475 if (!fl6->flowi6_oif)
1476 fl6->flowi6_oif = READ_ONCE(sk->sk_bound_dev_if);
1477
1478 if (!fl6->flowi6_oif)
1479 fl6->flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1480
1481 fl6->flowi6_uid = sk->sk_uid;
1482
1483 if (msg->msg_controllen) {
1484 opt = &opt_space;
1485 memset(opt, 0, sizeof(struct ipv6_txoptions));
1486 opt->tot_len = sizeof(*opt);
1487 ipc6.opt = opt;
1488
1489 err = udp_cmsg_send(sk, msg, &ipc6.gso_size);
1490 if (err > 0) {
1491 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, fl6,
1492 &ipc6);
1493 connected = false;
1494 }
1495 if (err < 0) {
1496 fl6_sock_release(flowlabel);
1497 return err;
1498 }
1499 if ((fl6->flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1500 flowlabel = fl6_sock_lookup(sk, fl6->flowlabel);
1501 if (IS_ERR(flowlabel))
1502 return -EINVAL;
1503 }
1504 if (!(opt->opt_nflen|opt->opt_flen))
1505 opt = NULL;
1506 }
1507 if (!opt) {
1508 opt = txopt_get(np);
1509 opt_to_free = opt;
1510 }
1511 if (flowlabel)
1512 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1513 opt = ipv6_fixup_options(&opt_space, opt);
1514 ipc6.opt = opt;
1515
1516 fl6->flowi6_proto = sk->sk_protocol;
1517 fl6->flowi6_mark = ipc6.sockc.mark;
1518 fl6->daddr = *daddr;
1519 if (ipv6_addr_any(&fl6->saddr) && !ipv6_addr_any(&np->saddr))
1520 fl6->saddr = np->saddr;
1521 fl6->fl6_sport = inet->inet_sport;
1522
1523 if (cgroup_bpf_enabled(CGROUP_UDP6_SENDMSG) && !connected) {
1524 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk,
1525 (struct sockaddr *)sin6,
1526 &addr_len,
1527 &fl6->saddr);
1528 if (err)
1529 goto out_no_dst;
1530 if (sin6) {
1531 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
1532 /* BPF program rewrote IPv6-only by IPv4-mapped
1533 * IPv6. It's currently unsupported.
1534 */
1535 err = -ENOTSUPP;
1536 goto out_no_dst;
1537 }
1538 if (sin6->sin6_port == 0) {
1539 /* BPF program set invalid port. Reject it. */
1540 err = -EINVAL;
1541 goto out_no_dst;
1542 }
1543 fl6->fl6_dport = sin6->sin6_port;
1544 fl6->daddr = sin6->sin6_addr;
1545 }
1546 }
1547
1548 if (ipv6_addr_any(&fl6->daddr))
1549 fl6->daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1550
1551 final_p = fl6_update_dst(fl6, opt, &final);
1552 if (final_p)
1553 connected = false;
1554
1555 if (!fl6->flowi6_oif && ipv6_addr_is_multicast(&fl6->daddr)) {
1556 fl6->flowi6_oif = READ_ONCE(np->mcast_oif);
1557 connected = false;
1558 } else if (!fl6->flowi6_oif)
1559 fl6->flowi6_oif = READ_ONCE(np->ucast_oif);
1560
1561 security_sk_classify_flow(sk, flowi6_to_flowi_common(fl6));
1562
1563 if (ipc6.tclass < 0)
1564 ipc6.tclass = np->tclass;
1565
1566 fl6->flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6->flowlabel);
1567
1568 dst = ip6_sk_dst_lookup_flow(sk, fl6, final_p, connected);
1569 if (IS_ERR(dst)) {
1570 err = PTR_ERR(dst);
1571 dst = NULL;
1572 goto out;
1573 }
1574
1575 if (ipc6.hlimit < 0)
1576 ipc6.hlimit = ip6_sk_dst_hoplimit(np, fl6, dst);
1577
1578 if (msg->msg_flags&MSG_CONFIRM)
1579 goto do_confirm;
1580 back_from_confirm:
1581
1582 /* Lockless fast path for the non-corking case */
1583 if (!corkreq) {
1584 struct sk_buff *skb;
1585
1586 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1587 sizeof(struct udphdr), &ipc6,
1588 dst_rt6_info(dst),
1589 msg->msg_flags, &cork);
1590 err = PTR_ERR(skb);
1591 if (!IS_ERR_OR_NULL(skb))
1592 err = udp_v6_send_skb(skb, fl6, &cork.base);
1593 /* ip6_make_skb steals dst reference */
1594 goto out_no_dst;
1595 }
1596
1597 lock_sock(sk);
1598 if (unlikely(up->pending)) {
1599 /* The socket is already corked while preparing it. */
1600 /* ... which is an evident application bug. --ANK */
1601 release_sock(sk);
1602
1603 net_dbg_ratelimited("udp cork app bug 2\n");
1604 err = -EINVAL;
1605 goto out;
1606 }
1607
1608 WRITE_ONCE(up->pending, AF_INET6);
1609
1610 do_append_data:
1611 if (ipc6.dontfrag < 0)
1612 ipc6.dontfrag = inet6_test_bit(DONTFRAG, sk);
1613 up->len += ulen;
1614 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1615 &ipc6, fl6, dst_rt6_info(dst),
1616 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1617 if (err)
1618 udp_v6_flush_pending_frames(sk);
1619 else if (!corkreq)
1620 err = udp_v6_push_pending_frames(sk);
1621 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1622 WRITE_ONCE(up->pending, 0);
1623
1624 if (err > 0)
1625 err = inet6_test_bit(RECVERR6, sk) ? net_xmit_errno(err) : 0;
1626 release_sock(sk);
1627
1628 out:
1629 dst_release(dst);
1630 out_no_dst:
1631 fl6_sock_release(flowlabel);
1632 txopt_put(opt_to_free);
1633 if (!err)
1634 return len;
1635 /*
1636 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1637 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1638 * we don't have a good statistic (IpOutDiscards but it can be too many
1639 * things). We could add another new stat but at least for now that
1640 * seems like overkill.
1641 */
1642 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1643 UDP6_INC_STATS(sock_net(sk),
1644 UDP_MIB_SNDBUFERRORS, is_udplite);
1645 }
1646 return err;
1647
1648 do_confirm:
1649 if (msg->msg_flags & MSG_PROBE)
1650 dst_confirm_neigh(dst, &fl6->daddr);
1651 if (!(msg->msg_flags&MSG_PROBE) || len)
1652 goto back_from_confirm;
1653 err = 0;
1654 goto out;
1655 }
1656 EXPORT_SYMBOL(udpv6_sendmsg);
1657
1658 static void udpv6_splice_eof(struct socket *sock)
1659 {
1660 struct sock *sk = sock->sk;
1661 struct udp_sock *up = udp_sk(sk);
1662
1663 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1664 return;
1665
1666 lock_sock(sk);
1667 if (up->pending && !udp_test_bit(CORK, sk))
1668 udp_v6_push_pending_frames(sk);
1669 release_sock(sk);
1670 }
1671
1672 void udpv6_destroy_sock(struct sock *sk)
1673 {
1674 struct udp_sock *up = udp_sk(sk);
1675 lock_sock(sk);
1676
1677 /* protects from races with udp_abort() */
1678 sock_set_flag(sk, SOCK_DEAD);
1679 udp_v6_flush_pending_frames(sk);
1680 release_sock(sk);
1681
1682 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
1683 if (up->encap_type) {
1684 void (*encap_destroy)(struct sock *sk);
1685 encap_destroy = READ_ONCE(up->encap_destroy);
1686 if (encap_destroy)
1687 encap_destroy(sk);
1688 }
1689 if (udp_test_bit(ENCAP_ENABLED, sk)) {
1690 static_branch_dec(&udpv6_encap_needed_key);
1691 udp_encap_disable();
1692 }
1693 }
1694 }
1695
1696 /*
1697 * Socket option code for UDP
1698 */
1699 int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
1700 unsigned int optlen)
1701 {
1702 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
1703 return udp_lib_setsockopt(sk, level, optname,
1704 optval, optlen,
1705 udp_v6_push_pending_frames);
1706 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1707 }
1708
1709 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1710 char __user *optval, int __user *optlen)
1711 {
1712 if (level == SOL_UDP || level == SOL_UDPLITE)
1713 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1714 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1715 }
1716
1717
1718 /* ------------------------------------------------------------------------ */
1719 #ifdef CONFIG_PROC_FS
1720 int udp6_seq_show(struct seq_file *seq, void *v)
1721 {
1722 if (v == SEQ_START_TOKEN) {
1723 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1724 } else {
1725 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1726 const struct inet_sock *inet = inet_sk((const struct sock *)v);
1727 __u16 srcp = ntohs(inet->inet_sport);
1728 __u16 destp = ntohs(inet->inet_dport);
1729 __ip6_dgram_sock_seq_show(seq, v, srcp, destp,
1730 udp_rqueue_get(v), bucket);
1731 }
1732 return 0;
1733 }
1734
1735 const struct seq_operations udp6_seq_ops = {
1736 .start = udp_seq_start,
1737 .next = udp_seq_next,
1738 .stop = udp_seq_stop,
1739 .show = udp6_seq_show,
1740 };
1741 EXPORT_SYMBOL(udp6_seq_ops);
1742
1743 static struct udp_seq_afinfo udp6_seq_afinfo = {
1744 .family = AF_INET6,
1745 .udp_table = NULL,
1746 };
1747
1748 int __net_init udp6_proc_init(struct net *net)
1749 {
1750 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops,
1751 sizeof(struct udp_iter_state), &udp6_seq_afinfo))
1752 return -ENOMEM;
1753 return 0;
1754 }
1755
1756 void udp6_proc_exit(struct net *net)
1757 {
1758 remove_proc_entry("udp6", net->proc_net);
1759 }
1760 #endif /* CONFIG_PROC_FS */
1761
1762 /* ------------------------------------------------------------------------ */
1763
1764 struct proto udpv6_prot = {
1765 .name = "UDPv6",
1766 .owner = THIS_MODULE,
1767 .close = udp_lib_close,
1768 .pre_connect = udpv6_pre_connect,
1769 .connect = ip6_datagram_connect,
1770 .disconnect = udp_disconnect,
1771 .ioctl = udp_ioctl,
1772 .init = udpv6_init_sock,
1773 .destroy = udpv6_destroy_sock,
1774 .setsockopt = udpv6_setsockopt,
1775 .getsockopt = udpv6_getsockopt,
1776 .sendmsg = udpv6_sendmsg,
1777 .recvmsg = udpv6_recvmsg,
1778 .splice_eof = udpv6_splice_eof,
1779 .release_cb = ip6_datagram_release_cb,
1780 .hash = udp_lib_hash,
1781 .unhash = udp_lib_unhash,
1782 .rehash = udp_v6_rehash,
1783 .get_port = udp_v6_get_port,
1784 .put_port = udp_lib_unhash,
1785 #ifdef CONFIG_BPF_SYSCALL
1786 .psock_update_sk_prot = udp_bpf_update_proto,
1787 #endif
1788
1789 .memory_allocated = &udp_memory_allocated,
1790 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
1791
1792 .sysctl_mem = sysctl_udp_mem,
1793 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
1794 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
1795 .obj_size = sizeof(struct udp6_sock),
1796 .ipv6_pinfo_offset = offsetof(struct udp6_sock, inet6),
1797 .h.udp_table = NULL,
1798 .diag_destroy = udp_abort,
1799 };
1800
1801 static struct inet_protosw udpv6_protosw = {
1802 .type = SOCK_DGRAM,
1803 .protocol = IPPROTO_UDP,
1804 .prot = &udpv6_prot,
1805 .ops = &inet6_dgram_ops,
1806 .flags = INET_PROTOSW_PERMANENT,
1807 };
1808
1809 int __init udpv6_init(void)
1810 {
1811 int ret;
1812
1813 net_hotdata.udpv6_protocol = (struct inet6_protocol) {
1814 .handler = udpv6_rcv,
1815 .err_handler = udpv6_err,
1816 .flags = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL,
1817 };
1818 ret = inet6_add_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
1819 if (ret)
1820 goto out;
1821
1822 ret = inet6_register_protosw(&udpv6_protosw);
1823 if (ret)
1824 goto out_udpv6_protocol;
1825 out:
1826 return ret;
1827
1828 out_udpv6_protocol:
1829 inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
1830 goto out;
1831 }
1832
1833 void udpv6_exit(void)
1834 {
1835 inet6_unregister_protosw(&udpv6_protosw);
1836 inet6_del_protocol(&net_hotdata.udpv6_protocol, IPPROTO_UDP);
1837 }
1838
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.