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/errno.h>
21 #include <linux/types.h>
22 #include <linux/socket.h>
23 #include <linux/sockios.h>
24 #include <linux/net.h>
25 #include <linux/in6.h>
26 #include <linux/netdevice.h>
27 #include <linux/if_arp.h>
28 #include <linux/ipv6.h>
29 #include <linux/icmpv6.h>
30 #include <linux/init.h>
31 #include <linux/module.h>
32 #include <linux/skbuff.h>
33 #include <linux/slab.h>
34 #include <linux/uaccess.h>
35 #include <linux/indirect_call_wrapper.h>
36
37 #include <net/addrconf.h>
38 #include <net/ndisc.h>
39 #include <net/protocol.h>
40 #include <net/transp_v6.h>
41 #include <net/ip6_route.h>
42 #include <net/raw.h>
43 #include <net/seg6.h>
44 #include <net/tcp_states.h>
45 #include <net/ip6_checksum.h>
46 #include <net/ip6_tunnel.h>
47 #include <net/xfrm.h>
48 #include <net/inet_hashtables.h>
49 #include <net/inet6_hashtables.h>
50 #include <net/busy_poll.h>
51 #include <net/sock_reuseport.h>
52
53 #include <linux/proc_fs.h>
54 #include <linux/seq_file.h>
55 #include <trace/events/skb.h>
56 #include "udp_impl.h"
57
58 static u32 udp6_ehashfn(const struct net *net,
59 const struct in6_addr *laddr,
60 const u16 lport,
61 const struct in6_addr *faddr,
62 const __be16 fport)
63 {
64 static u32 udp6_ehash_secret __read_mostly;
65 static u32 udp_ipv6_hash_secret __read_mostly;
66
67 u32 lhash, fhash;
68
69 net_get_random_once(&udp6_ehash_secret,
70 sizeof(udp6_ehash_secret));
71 net_get_random_once(&udp_ipv6_hash_secret,
72 sizeof(udp_ipv6_hash_secret));
73
74 lhash = (__force u32)laddr->s6_addr32[3];
75 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
76
77 return __inet6_ehashfn(lhash, lport, fhash, fport,
78 udp_ipv6_hash_secret + net_hash_mix(net));
79 }
80
81 int udp_v6_get_port(struct sock *sk, unsigned short snum)
82 {
83 unsigned int hash2_nulladdr =
84 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
85 unsigned int hash2_partial =
86 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
87
88 /* precompute partial secondary hash */
89 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
90 return udp_lib_get_port(sk, snum, hash2_nulladdr);
91 }
92
93 void udp_v6_rehash(struct sock *sk)
94 {
95 u16 new_hash = ipv6_portaddr_hash(sock_net(sk),
96 &sk->sk_v6_rcv_saddr,
97 inet_sk(sk)->inet_num);
98
99 udp_lib_rehash(sk, new_hash);
100 }
101
102 static int compute_score(struct sock *sk, struct net *net,
103 const struct in6_addr *saddr, __be16 sport,
104 const struct in6_addr *daddr, unsigned short hnum,
105 int dif, int sdif)
106 {
107 int score;
108 struct inet_sock *inet;
109 bool dev_match;
110
111 if (!net_eq(sock_net(sk), net) ||
112 udp_sk(sk)->udp_port_hash != hnum ||
113 sk->sk_family != PF_INET6)
114 return -1;
115
116 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
117 return -1;
118
119 score = 0;
120 inet = inet_sk(sk);
121
122 if (inet->inet_dport) {
123 if (inet->inet_dport != sport)
124 return -1;
125 score++;
126 }
127
128 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
129 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
130 return -1;
131 score++;
132 }
133
134 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif);
135 if (!dev_match)
136 return -1;
137 if (sk->sk_bound_dev_if)
138 score++;
139
140 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
141 score++;
142
143 return score;
144 }
145
146 static struct sock *lookup_reuseport(struct net *net, struct sock *sk,
147 struct sk_buff *skb,
148 const struct in6_addr *saddr,
149 __be16 sport,
150 const struct in6_addr *daddr,
151 unsigned int hnum)
152 {
153 struct sock *reuse_sk = NULL;
154 u32 hash;
155
156 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) {
157 hash = udp6_ehashfn(net, daddr, hnum, saddr, sport);
158 reuse_sk = reuseport_select_sock(sk, hash, skb,
159 sizeof(struct udphdr));
160 }
161 return reuse_sk;
162 }
163
164 /* called with rcu_read_lock() */
165 static struct sock *udp6_lib_lookup2(struct net *net,
166 const struct in6_addr *saddr, __be16 sport,
167 const struct in6_addr *daddr, unsigned int hnum,
168 int dif, int sdif, struct udp_hslot *hslot2,
169 struct sk_buff *skb)
170 {
171 struct sock *sk, *result;
172 int score, badness;
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);
179 if (score > badness) {
180 result = lookup_reuseport(net, sk, skb,
181 saddr, sport, daddr, hnum);
182 /* Fall back to scoring if group has connections */
183 if (result && !reuseport_has_conns(sk, false))
184 return result;
185
186 result = result ? : sk;
187 badness = score;
188 }
189 }
190 return result;
191 }
192
193 static inline struct sock *udp6_lookup_run_bpf(struct net *net,
194 struct udp_table *udptable,
195 struct sk_buff *skb,
196 const struct in6_addr *saddr,
197 __be16 sport,
198 const struct in6_addr *daddr,
199 u16 hnum)
200 {
201 struct sock *sk, *reuse_sk;
202 bool no_reuseport;
203
204 if (udptable != &udp_table)
205 return NULL; /* only UDP is supported */
206
207 no_reuseport = bpf_sk_lookup_run_v6(net, IPPROTO_UDP,
208 saddr, sport, daddr, hnum, &sk);
209 if (no_reuseport || IS_ERR_OR_NULL(sk))
210 return sk;
211
212 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
213 if (reuse_sk)
214 sk = reuse_sk;
215 return sk;
216 }
217
218 /* rcu_read_lock() must be held */
219 struct sock *__udp6_lib_lookup(struct net *net,
220 const struct in6_addr *saddr, __be16 sport,
221 const struct in6_addr *daddr, __be16 dport,
222 int dif, int sdif, struct udp_table *udptable,
223 struct sk_buff *skb)
224 {
225 unsigned short hnum = ntohs(dport);
226 unsigned int hash2, slot2;
227 struct udp_hslot *hslot2;
228 struct sock *result, *sk;
229
230 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
231 slot2 = hash2 & udptable->mask;
232 hslot2 = &udptable->hash2[slot2];
233
234 /* Lookup connected or non-wildcard sockets */
235 result = udp6_lib_lookup2(net, saddr, sport,
236 daddr, hnum, dif, sdif,
237 hslot2, skb);
238 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
239 goto done;
240
241 /* Lookup redirect from BPF */
242 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
243 sk = udp6_lookup_run_bpf(net, udptable, skb,
244 saddr, sport, daddr, hnum);
245 if (sk) {
246 result = sk;
247 goto done;
248 }
249 }
250
251 /* Got non-wildcard socket or error on first lookup */
252 if (result)
253 goto done;
254
255 /* Lookup wildcard sockets */
256 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
257 slot2 = hash2 & udptable->mask;
258 hslot2 = &udptable->hash2[slot2];
259
260 result = udp6_lib_lookup2(net, saddr, sport,
261 &in6addr_any, hnum, dif, sdif,
262 hslot2, skb);
263 done:
264 if (IS_ERR(result))
265 return NULL;
266 return result;
267 }
268 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
269
270 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
271 __be16 sport, __be16 dport,
272 struct udp_table *udptable)
273 {
274 const struct ipv6hdr *iph = ipv6_hdr(skb);
275
276 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
277 &iph->daddr, dport, inet6_iif(skb),
278 inet6_sdif(skb), udptable, skb);
279 }
280
281 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb,
282 __be16 sport, __be16 dport)
283 {
284 const struct ipv6hdr *iph = ipv6_hdr(skb);
285
286 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
287 &iph->daddr, dport, inet6_iif(skb),
288 inet6_sdif(skb), &udp_table, NULL);
289 }
290
291 /* Must be called under rcu_read_lock().
292 * Does increment socket refcount.
293 */
294 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
295 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
296 const struct in6_addr *daddr, __be16 dport, int dif)
297 {
298 struct sock *sk;
299
300 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport,
301 dif, 0, &udp_table, NULL);
302 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
303 sk = NULL;
304 return sk;
305 }
306 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
307 #endif
308
309 /* do not use the scratch area len for jumbogram: their length execeeds the
310 * scratch area space; note that the IP6CB flags is still in the first
311 * cacheline, so checking for jumbograms is cheap
312 */
313 static int udp6_skb_len(struct sk_buff *skb)
314 {
315 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
316 }
317
318 /*
319 * This should be easy, if there is something there we
320 * return it, otherwise we block.
321 */
322
323 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
324 int noblock, int flags, int *addr_len)
325 {
326 struct ipv6_pinfo *np = inet6_sk(sk);
327 struct inet_sock *inet = inet_sk(sk);
328 struct sk_buff *skb;
329 unsigned int ulen, copied;
330 int off, err, peeking = flags & MSG_PEEK;
331 int is_udplite = IS_UDPLITE(sk);
332 struct udp_mib __percpu *mib;
333 bool checksum_valid = false;
334 int is_udp4;
335
336 if (flags & MSG_ERRQUEUE)
337 return ipv6_recv_error(sk, msg, len, addr_len);
338
339 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
340 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
341
342 try_again:
343 off = sk_peek_offset(sk, flags);
344 skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
345 if (!skb)
346 return err;
347 if (ccs_socket_post_recvmsg_permission(sk, skb, flags))
348 return -EAGAIN; /* Hope less harmful than -EPERM. */
349
350 ulen = udp6_skb_len(skb);
351 copied = len;
352 if (copied > ulen - off)
353 copied = ulen - off;
354 else if (copied < ulen)
355 msg->msg_flags |= MSG_TRUNC;
356
357 is_udp4 = (skb->protocol == htons(ETH_P_IP));
358 mib = __UDPX_MIB(sk, is_udp4);
359
360 /*
361 * If checksum is needed at all, try to do it while copying the
362 * data. If the data is truncated, or if we only want a partial
363 * coverage checksum (UDP-Lite), do it before the copy.
364 */
365
366 if (copied < ulen || peeking ||
367 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
368 checksum_valid = udp_skb_csum_unnecessary(skb) ||
369 !__udp_lib_checksum_complete(skb);
370 if (!checksum_valid)
371 goto csum_copy_err;
372 }
373
374 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
375 if (udp_skb_is_linear(skb))
376 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
377 else
378 err = skb_copy_datagram_msg(skb, off, msg, copied);
379 } else {
380 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
381 if (err == -EINVAL)
382 goto csum_copy_err;
383 }
384 if (unlikely(err)) {
385 if (!peeking) {
386 atomic_inc(&sk->sk_drops);
387 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
388 }
389 kfree_skb(skb);
390 return err;
391 }
392 if (!peeking)
393 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS);
394
395 sock_recv_ts_and_drops(msg, sk, skb);
396
397 /* Copy the address. */
398 if (msg->msg_name) {
399 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
400 sin6->sin6_family = AF_INET6;
401 sin6->sin6_port = udp_hdr(skb)->source;
402 sin6->sin6_flowinfo = 0;
403
404 if (is_udp4) {
405 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
406 &sin6->sin6_addr);
407 sin6->sin6_scope_id = 0;
408 } else {
409 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
410 sin6->sin6_scope_id =
411 ipv6_iface_scope_id(&sin6->sin6_addr,
412 inet6_iif(skb));
413 }
414 *addr_len = sizeof(*sin6);
415
416 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk,
417 (struct sockaddr *)sin6);
418 }
419
420 if (udp_sk(sk)->gro_enabled)
421 udp_cmsg_recv(msg, sk, skb);
422
423 if (np->rxopt.all)
424 ip6_datagram_recv_common_ctl(sk, msg, skb);
425
426 if (is_udp4) {
427 if (inet->cmsg_flags)
428 ip_cmsg_recv_offset(msg, sk, skb,
429 sizeof(struct udphdr), off);
430 } else {
431 if (np->rxopt.all)
432 ip6_datagram_recv_specific_ctl(sk, msg, skb);
433 }
434
435 err = copied;
436 if (flags & MSG_TRUNC)
437 err = ulen;
438
439 skb_consume_udp(sk, skb, peeking ? -err : err);
440 return err;
441
442 csum_copy_err:
443 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
444 udp_skb_destructor)) {
445 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS);
446 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
447 }
448 kfree_skb(skb);
449
450 /* starting over for a new packet, but check if we need to yield */
451 cond_resched();
452 msg->msg_flags &= ~MSG_TRUNC;
453 goto try_again;
454 }
455
456 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
457 void udpv6_encap_enable(void)
458 {
459 static_branch_inc(&udpv6_encap_needed_key);
460 }
461 EXPORT_SYMBOL(udpv6_encap_enable);
462
463 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
464 * through error handlers in encapsulations looking for a match.
465 */
466 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb,
467 struct inet6_skb_parm *opt,
468 u8 type, u8 code, int offset, __be32 info)
469 {
470 int i;
471
472 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
473 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt,
474 u8 type, u8 code, int offset, __be32 info);
475 const struct ip6_tnl_encap_ops *encap;
476
477 encap = rcu_dereference(ip6tun_encaps[i]);
478 if (!encap)
479 continue;
480 handler = encap->err_handler;
481 if (handler && !handler(skb, opt, type, code, offset, info))
482 return 0;
483 }
484
485 return -ENOENT;
486 }
487
488 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
489 * reversing source and destination port: this will match tunnels that force the
490 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
491 * lwtunnels might actually break this assumption by being configured with
492 * different destination ports on endpoints, in this case we won't be able to
493 * trace ICMP messages back to them.
494 *
495 * If this doesn't match any socket, probe tunnels with arbitrary destination
496 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
497 * we've sent packets to won't necessarily match the local destination port.
498 *
499 * Then ask the tunnel implementation to match the error against a valid
500 * association.
501 *
502 * Return an error if we can't find a match, the socket if we need further
503 * processing, zero otherwise.
504 */
505 static struct sock *__udp6_lib_err_encap(struct net *net,
506 const struct ipv6hdr *hdr, int offset,
507 struct udphdr *uh,
508 struct udp_table *udptable,
509 struct sock *sk,
510 struct sk_buff *skb,
511 struct inet6_skb_parm *opt,
512 u8 type, u8 code, __be32 info)
513 {
514 int (*lookup)(struct sock *sk, struct sk_buff *skb);
515 int network_offset, transport_offset;
516 struct udp_sock *up;
517
518 network_offset = skb_network_offset(skb);
519 transport_offset = skb_transport_offset(skb);
520
521 /* Network header needs to point to the outer IPv6 header inside ICMP */
522 skb_reset_network_header(skb);
523
524 /* Transport header needs to point to the UDP header */
525 skb_set_transport_header(skb, offset);
526
527 if (sk) {
528 up = udp_sk(sk);
529
530 lookup = READ_ONCE(up->encap_err_lookup);
531 if (lookup && lookup(sk, skb))
532 sk = NULL;
533
534 goto out;
535 }
536
537 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source,
538 &hdr->saddr, uh->dest,
539 inet6_iif(skb), 0, udptable, skb);
540 if (sk) {
541 up = udp_sk(sk);
542
543 lookup = READ_ONCE(up->encap_err_lookup);
544 if (!lookup || lookup(sk, skb))
545 sk = NULL;
546 }
547
548 out:
549 if (!sk) {
550 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code,
551 offset, info));
552 }
553
554 skb_set_transport_header(skb, transport_offset);
555 skb_set_network_header(skb, network_offset);
556
557 return sk;
558 }
559
560 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
561 u8 type, u8 code, int offset, __be32 info,
562 struct udp_table *udptable)
563 {
564 struct ipv6_pinfo *np;
565 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
566 const struct in6_addr *saddr = &hdr->saddr;
567 const struct in6_addr *daddr = seg6_get_daddr(skb, opt) ? : &hdr->daddr;
568 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
569 bool tunnel = false;
570 struct sock *sk;
571 int harderr;
572 int err;
573 struct net *net = dev_net(skb->dev);
574
575 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
576 inet6_iif(skb), inet6_sdif(skb), udptable, NULL);
577
578 if (!sk || udp_sk(sk)->encap_type) {
579 /* No socket for error: try tunnels before discarding */
580 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
581 sk = __udp6_lib_err_encap(net, hdr, offset, uh,
582 udptable, sk, skb,
583 opt, type, code, info);
584 if (!sk)
585 return 0;
586 } else
587 sk = ERR_PTR(-ENOENT);
588
589 if (IS_ERR(sk)) {
590 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
591 ICMP6_MIB_INERRORS);
592 return PTR_ERR(sk);
593 }
594
595 tunnel = true;
596 }
597
598 harderr = icmpv6_err_convert(type, code, &err);
599 np = inet6_sk(sk);
600
601 if (type == ICMPV6_PKT_TOOBIG) {
602 if (!ip6_sk_accept_pmtu(sk))
603 goto out;
604 ip6_sk_update_pmtu(skb, sk, info);
605 if (np->pmtudisc != IPV6_PMTUDISC_DONT)
606 harderr = 1;
607 }
608 if (type == NDISC_REDIRECT) {
609 if (tunnel) {
610 ip6_redirect(skb, sock_net(sk), inet6_iif(skb),
611 sk->sk_mark, sk->sk_uid);
612 } else {
613 ip6_sk_redirect(skb, sk);
614 }
615 goto out;
616 }
617
618 /* Tunnels don't have an application socket: don't pass errors back */
619 if (tunnel)
620 goto out;
621
622 if (!np->recverr) {
623 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
624 goto out;
625 } else {
626 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
627 }
628
629 sk->sk_err = err;
630 sk_error_report(sk);
631 out:
632 return 0;
633 }
634
635 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
636 {
637 int rc;
638
639 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
640 sock_rps_save_rxhash(sk, skb);
641 sk_mark_napi_id(sk, skb);
642 sk_incoming_cpu_update(sk);
643 } else {
644 sk_mark_napi_id_once(sk, skb);
645 }
646
647 rc = __udp_enqueue_schedule_skb(sk, skb);
648 if (rc < 0) {
649 int is_udplite = IS_UDPLITE(sk);
650
651 /* Note that an ENOMEM error is charged twice */
652 if (rc == -ENOMEM)
653 UDP6_INC_STATS(sock_net(sk),
654 UDP_MIB_RCVBUFERRORS, is_udplite);
655 else
656 UDP6_INC_STATS(sock_net(sk),
657 UDP_MIB_MEMERRORS, is_udplite);
658 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
659 kfree_skb(skb);
660 return -1;
661 }
662
663 return 0;
664 }
665
666 static __inline__ int udpv6_err(struct sk_buff *skb,
667 struct inet6_skb_parm *opt, u8 type,
668 u8 code, int offset, __be32 info)
669 {
670 return __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
671 }
672
673 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
674 {
675 struct udp_sock *up = udp_sk(sk);
676 int is_udplite = IS_UDPLITE(sk);
677
678 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
679 goto drop;
680
681 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) {
682 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
683
684 /*
685 * This is an encapsulation socket so pass the skb to
686 * the socket's udp_encap_rcv() hook. Otherwise, just
687 * fall through and pass this up the UDP socket.
688 * up->encap_rcv() returns the following value:
689 * =0 if skb was successfully passed to the encap
690 * handler or was discarded by it.
691 * >0 if skb should be passed on to UDP.
692 * <0 if skb should be resubmitted as proto -N
693 */
694
695 /* if we're overly short, let UDP handle it */
696 encap_rcv = READ_ONCE(up->encap_rcv);
697 if (encap_rcv) {
698 int ret;
699
700 /* Verify checksum before giving to encap */
701 if (udp_lib_checksum_complete(skb))
702 goto csum_error;
703
704 ret = encap_rcv(sk, skb);
705 if (ret <= 0) {
706 __UDP6_INC_STATS(sock_net(sk),
707 UDP_MIB_INDATAGRAMS,
708 is_udplite);
709 return -ret;
710 }
711 }
712
713 /* FALLTHROUGH -- it's a UDP Packet */
714 }
715
716 /*
717 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
718 */
719 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
720
721 if (up->pcrlen == 0) { /* full coverage was set */
722 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
723 UDP_SKB_CB(skb)->cscov, skb->len);
724 goto drop;
725 }
726 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
727 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
728 UDP_SKB_CB(skb)->cscov, up->pcrlen);
729 goto drop;
730 }
731 }
732
733 prefetch(&sk->sk_rmem_alloc);
734 if (rcu_access_pointer(sk->sk_filter) &&
735 udp_lib_checksum_complete(skb))
736 goto csum_error;
737
738 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
739 goto drop;
740
741 udp_csum_pull_header(skb);
742
743 skb_dst_drop(skb);
744
745 return __udpv6_queue_rcv_skb(sk, skb);
746
747 csum_error:
748 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
749 drop:
750 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
751 atomic_inc(&sk->sk_drops);
752 kfree_skb(skb);
753 return -1;
754 }
755
756 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
757 {
758 struct sk_buff *next, *segs;
759 int ret;
760
761 if (likely(!udp_unexpected_gso(sk, skb)))
762 return udpv6_queue_rcv_one_skb(sk, skb);
763
764 __skb_push(skb, -skb_mac_offset(skb));
765 segs = udp_rcv_segment(sk, skb, false);
766 skb_list_walk_safe(segs, skb, next) {
767 __skb_pull(skb, skb_transport_offset(skb));
768
769 udp_post_segment_fix_csum(skb);
770 ret = udpv6_queue_rcv_one_skb(sk, skb);
771 if (ret > 0)
772 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret,
773 true);
774 }
775 return 0;
776 }
777
778 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
779 __be16 loc_port, const struct in6_addr *loc_addr,
780 __be16 rmt_port, const struct in6_addr *rmt_addr,
781 int dif, int sdif, unsigned short hnum)
782 {
783 struct inet_sock *inet = inet_sk(sk);
784
785 if (!net_eq(sock_net(sk), net))
786 return false;
787
788 if (udp_sk(sk)->udp_port_hash != hnum ||
789 sk->sk_family != PF_INET6 ||
790 (inet->inet_dport && inet->inet_dport != rmt_port) ||
791 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
792 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
793 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif) ||
794 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
795 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
796 return false;
797 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
798 return false;
799 return true;
800 }
801
802 static void udp6_csum_zero_error(struct sk_buff *skb)
803 {
804 /* RFC 2460 section 8.1 says that we SHOULD log
805 * this error. Well, it is reasonable.
806 */
807 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
808 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
809 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
810 }
811
812 /*
813 * Note: called only from the BH handler context,
814 * so we don't need to lock the hashes.
815 */
816 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
817 const struct in6_addr *saddr, const struct in6_addr *daddr,
818 struct udp_table *udptable, int proto)
819 {
820 struct sock *sk, *first = NULL;
821 const struct udphdr *uh = udp_hdr(skb);
822 unsigned short hnum = ntohs(uh->dest);
823 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
824 unsigned int offset = offsetof(typeof(*sk), sk_node);
825 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
826 int dif = inet6_iif(skb);
827 int sdif = inet6_sdif(skb);
828 struct hlist_node *node;
829 struct sk_buff *nskb;
830
831 if (use_hash2) {
832 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) &
833 udptable->mask;
834 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask;
835 start_lookup:
836 hslot = &udptable->hash2[hash2];
837 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
838 }
839
840 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
841 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
842 uh->source, saddr, dif, sdif,
843 hnum))
844 continue;
845 /* If zero checksum and no_check is not on for
846 * the socket then skip it.
847 */
848 if (!uh->check && !udp_sk(sk)->no_check6_rx)
849 continue;
850 if (!first) {
851 first = sk;
852 continue;
853 }
854 nskb = skb_clone(skb, GFP_ATOMIC);
855 if (unlikely(!nskb)) {
856 atomic_inc(&sk->sk_drops);
857 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
858 IS_UDPLITE(sk));
859 __UDP6_INC_STATS(net, UDP_MIB_INERRORS,
860 IS_UDPLITE(sk));
861 continue;
862 }
863
864 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
865 consume_skb(nskb);
866 }
867
868 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
869 if (use_hash2 && hash2 != hash2_any) {
870 hash2 = hash2_any;
871 goto start_lookup;
872 }
873
874 if (first) {
875 if (udpv6_queue_rcv_skb(first, skb) > 0)
876 consume_skb(skb);
877 } else {
878 kfree_skb(skb);
879 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
880 proto == IPPROTO_UDPLITE);
881 }
882 return 0;
883 }
884
885 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
886 {
887 if (udp_sk_rx_dst_set(sk, dst)) {
888 const struct rt6_info *rt = (const struct rt6_info *)dst;
889
890 sk->sk_rx_dst_cookie = rt6_get_cookie(rt);
891 }
892 }
893
894 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
895 * return code conversion for ip layer consumption
896 */
897 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
898 struct udphdr *uh)
899 {
900 int ret;
901
902 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
903 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo);
904
905 ret = udpv6_queue_rcv_skb(sk, skb);
906
907 /* a return value > 0 means to resubmit the input */
908 if (ret > 0)
909 return ret;
910 return 0;
911 }
912
913 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
914 int proto)
915 {
916 const struct in6_addr *saddr, *daddr;
917 struct net *net = dev_net(skb->dev);
918 struct udphdr *uh;
919 struct sock *sk;
920 bool refcounted;
921 u32 ulen = 0;
922
923 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
924 goto discard;
925
926 saddr = &ipv6_hdr(skb)->saddr;
927 daddr = &ipv6_hdr(skb)->daddr;
928 uh = udp_hdr(skb);
929
930 ulen = ntohs(uh->len);
931 if (ulen > skb->len)
932 goto short_packet;
933
934 if (proto == IPPROTO_UDP) {
935 /* UDP validates ulen. */
936
937 /* Check for jumbo payload */
938 if (ulen == 0)
939 ulen = skb->len;
940
941 if (ulen < sizeof(*uh))
942 goto short_packet;
943
944 if (ulen < skb->len) {
945 if (pskb_trim_rcsum(skb, ulen))
946 goto short_packet;
947 saddr = &ipv6_hdr(skb)->saddr;
948 daddr = &ipv6_hdr(skb)->daddr;
949 uh = udp_hdr(skb);
950 }
951 }
952
953 if (udp6_csum_init(skb, uh, proto))
954 goto csum_error;
955
956 /* Check if the socket is already available, e.g. due to early demux */
957 sk = skb_steal_sock(skb, &refcounted);
958 if (sk) {
959 struct dst_entry *dst = skb_dst(skb);
960 int ret;
961
962 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
963 udp6_sk_rx_dst_set(sk, dst);
964
965 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
966 if (refcounted)
967 sock_put(sk);
968 goto report_csum_error;
969 }
970
971 ret = udp6_unicast_rcv_skb(sk, skb, uh);
972 if (refcounted)
973 sock_put(sk);
974 return ret;
975 }
976
977 /*
978 * Multicast receive code
979 */
980 if (ipv6_addr_is_multicast(daddr))
981 return __udp6_lib_mcast_deliver(net, skb,
982 saddr, daddr, udptable, proto);
983
984 /* Unicast */
985 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
986 if (sk) {
987 if (!uh->check && !udp_sk(sk)->no_check6_rx)
988 goto report_csum_error;
989 return udp6_unicast_rcv_skb(sk, skb, uh);
990 }
991
992 if (!uh->check)
993 goto report_csum_error;
994
995 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
996 goto discard;
997
998 if (udp_lib_checksum_complete(skb))
999 goto csum_error;
1000
1001 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1002 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
1003
1004 kfree_skb(skb);
1005 return 0;
1006
1007 short_packet:
1008 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
1009 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1010 saddr, ntohs(uh->source),
1011 ulen, skb->len,
1012 daddr, ntohs(uh->dest));
1013 goto discard;
1014
1015 report_csum_error:
1016 udp6_csum_zero_error(skb);
1017 csum_error:
1018 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1019 discard:
1020 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1021 kfree_skb(skb);
1022 return 0;
1023 }
1024
1025
1026 static struct sock *__udp6_lib_demux_lookup(struct net *net,
1027 __be16 loc_port, const struct in6_addr *loc_addr,
1028 __be16 rmt_port, const struct in6_addr *rmt_addr,
1029 int dif, int sdif)
1030 {
1031 unsigned short hnum = ntohs(loc_port);
1032 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum);
1033 unsigned int slot2 = hash2 & udp_table.mask;
1034 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
1035 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
1036 struct sock *sk;
1037
1038 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
1039 if (sk->sk_state == TCP_ESTABLISHED &&
1040 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif))
1041 return sk;
1042 /* Only check first socket in chain */
1043 break;
1044 }
1045 return NULL;
1046 }
1047
1048 INDIRECT_CALLABLE_SCOPE void udp_v6_early_demux(struct sk_buff *skb)
1049 {
1050 struct net *net = dev_net(skb->dev);
1051 const struct udphdr *uh;
1052 struct sock *sk;
1053 struct dst_entry *dst;
1054 int dif = skb->dev->ifindex;
1055 int sdif = inet6_sdif(skb);
1056
1057 if (!pskb_may_pull(skb, skb_transport_offset(skb) +
1058 sizeof(struct udphdr)))
1059 return;
1060
1061 uh = udp_hdr(skb);
1062
1063 if (skb->pkt_type == PACKET_HOST)
1064 sk = __udp6_lib_demux_lookup(net, uh->dest,
1065 &ipv6_hdr(skb)->daddr,
1066 uh->source, &ipv6_hdr(skb)->saddr,
1067 dif, sdif);
1068 else
1069 return;
1070
1071 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
1072 return;
1073
1074 skb->sk = sk;
1075 skb->destructor = sock_efree;
1076 dst = rcu_dereference(sk->sk_rx_dst);
1077
1078 if (dst)
1079 dst = dst_check(dst, sk->sk_rx_dst_cookie);
1080 if (dst) {
1081 /* set noref for now.
1082 * any place which wants to hold dst has to call
1083 * dst_hold_safe()
1084 */
1085 skb_dst_set_noref(skb, dst);
1086 }
1087 }
1088
1089 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb)
1090 {
1091 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1092 }
1093
1094 /*
1095 * Throw away all pending data and cancel the corking. Socket is locked.
1096 */
1097 static void udp_v6_flush_pending_frames(struct sock *sk)
1098 {
1099 struct udp_sock *up = udp_sk(sk);
1100
1101 if (up->pending == AF_INET)
1102 udp_flush_pending_frames(sk);
1103 else if (up->pending) {
1104 up->len = 0;
1105 up->pending = 0;
1106 ip6_flush_pending_frames(sk);
1107 }
1108 }
1109
1110 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
1111 int addr_len)
1112 {
1113 if (addr_len < offsetofend(struct sockaddr, sa_family))
1114 return -EINVAL;
1115 /* The following checks are replicated from __ip6_datagram_connect()
1116 * and intended to prevent BPF program called below from accessing
1117 * bytes that are out of the bound specified by user in addr_len.
1118 */
1119 if (uaddr->sa_family == AF_INET) {
1120 if (__ipv6_only_sock(sk))
1121 return -EAFNOSUPPORT;
1122 return udp_pre_connect(sk, uaddr, addr_len);
1123 }
1124
1125 if (addr_len < SIN6_LEN_RFC2133)
1126 return -EINVAL;
1127
1128 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr);
1129 }
1130
1131 /**
1132 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
1133 * @sk: socket we are sending on
1134 * @skb: sk_buff containing the filled-in UDP header
1135 * (checksum field must be zeroed out)
1136 * @saddr: source address
1137 * @daddr: destination address
1138 * @len: length of packet
1139 */
1140 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
1141 const struct in6_addr *saddr,
1142 const struct in6_addr *daddr, int len)
1143 {
1144 unsigned int offset;
1145 struct udphdr *uh = udp_hdr(skb);
1146 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
1147 __wsum csum = 0;
1148
1149 if (!frags) {
1150 /* Only one fragment on the socket. */
1151 skb->csum_start = skb_transport_header(skb) - skb->head;
1152 skb->csum_offset = offsetof(struct udphdr, check);
1153 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1154 } else {
1155 /*
1156 * HW-checksum won't work as there are two or more
1157 * fragments on the socket so that all csums of sk_buffs
1158 * should be together
1159 */
1160 offset = skb_transport_offset(skb);
1161 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1162 csum = skb->csum;
1163
1164 skb->ip_summed = CHECKSUM_NONE;
1165
1166 do {
1167 csum = csum_add(csum, frags->csum);
1168 } while ((frags = frags->next));
1169
1170 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1171 csum);
1172 if (uh->check == 0)
1173 uh->check = CSUM_MANGLED_0;
1174 }
1175 }
1176
1177 /*
1178 * Sending
1179 */
1180
1181 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6,
1182 struct inet_cork *cork)
1183 {
1184 struct sock *sk = skb->sk;
1185 struct udphdr *uh;
1186 int err = 0;
1187 int is_udplite = IS_UDPLITE(sk);
1188 __wsum csum = 0;
1189 int offset = skb_transport_offset(skb);
1190 int len = skb->len - offset;
1191 int datalen = len - sizeof(*uh);
1192
1193 /*
1194 * Create a UDP header
1195 */
1196 uh = udp_hdr(skb);
1197 uh->source = fl6->fl6_sport;
1198 uh->dest = fl6->fl6_dport;
1199 uh->len = htons(len);
1200 uh->check = 0;
1201
1202 if (cork->gso_size) {
1203 const int hlen = skb_network_header_len(skb) +
1204 sizeof(struct udphdr);
1205
1206 if (hlen + cork->gso_size > cork->fragsize) {
1207 kfree_skb(skb);
1208 return -EINVAL;
1209 }
1210 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
1211 kfree_skb(skb);
1212 return -EINVAL;
1213 }
1214 if (udp_sk(sk)->no_check6_tx) {
1215 kfree_skb(skb);
1216 return -EINVAL;
1217 }
1218 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
1219 dst_xfrm(skb_dst(skb))) {
1220 kfree_skb(skb);
1221 return -EIO;
1222 }
1223
1224 if (datalen > cork->gso_size) {
1225 skb_shinfo(skb)->gso_size = cork->gso_size;
1226 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1227 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
1228 cork->gso_size);
1229 }
1230 goto csum_partial;
1231 }
1232
1233 if (is_udplite)
1234 csum = udplite_csum(skb);
1235 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
1236 skb->ip_summed = CHECKSUM_NONE;
1237 goto send;
1238 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1239 csum_partial:
1240 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1241 goto send;
1242 } else
1243 csum = udp_csum(skb);
1244
1245 /* add protocol-dependent pseudo-header */
1246 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1247 len, fl6->flowi6_proto, csum);
1248 if (uh->check == 0)
1249 uh->check = CSUM_MANGLED_0;
1250
1251 send:
1252 err = ip6_send_skb(skb);
1253 if (err) {
1254 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1255 UDP6_INC_STATS(sock_net(sk),
1256 UDP_MIB_SNDBUFERRORS, is_udplite);
1257 err = 0;
1258 }
1259 } else {
1260 UDP6_INC_STATS(sock_net(sk),
1261 UDP_MIB_OUTDATAGRAMS, is_udplite);
1262 }
1263 return err;
1264 }
1265
1266 static int udp_v6_push_pending_frames(struct sock *sk)
1267 {
1268 struct sk_buff *skb;
1269 struct udp_sock *up = udp_sk(sk);
1270 struct flowi6 fl6;
1271 int err = 0;
1272
1273 if (up->pending == AF_INET)
1274 return udp_push_pending_frames(sk);
1275
1276 /* ip6_finish_skb will release the cork, so make a copy of
1277 * fl6 here.
1278 */
1279 fl6 = inet_sk(sk)->cork.fl.u.ip6;
1280
1281 skb = ip6_finish_skb(sk);
1282 if (!skb)
1283 goto out;
1284
1285 err = udp_v6_send_skb(skb, &fl6, &inet_sk(sk)->cork.base);
1286
1287 out:
1288 up->len = 0;
1289 up->pending = 0;
1290 return err;
1291 }
1292
1293 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1294 {
1295 struct ipv6_txoptions opt_space;
1296 struct udp_sock *up = udp_sk(sk);
1297 struct inet_sock *inet = inet_sk(sk);
1298 struct ipv6_pinfo *np = inet6_sk(sk);
1299 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1300 struct in6_addr *daddr, *final_p, final;
1301 struct ipv6_txoptions *opt = NULL;
1302 struct ipv6_txoptions *opt_to_free = NULL;
1303 struct ip6_flowlabel *flowlabel = NULL;
1304 struct flowi6 fl6;
1305 struct dst_entry *dst;
1306 struct ipcm6_cookie ipc6;
1307 int addr_len = msg->msg_namelen;
1308 bool connected = false;
1309 int ulen = len;
1310 int corkreq = READ_ONCE(up->corkflag) || msg->msg_flags&MSG_MORE;
1311 int err;
1312 int is_udplite = IS_UDPLITE(sk);
1313 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1314
1315 ipcm6_init(&ipc6);
1316 ipc6.gso_size = READ_ONCE(up->gso_size);
1317 ipc6.sockc.tsflags = sk->sk_tsflags;
1318 ipc6.sockc.mark = sk->sk_mark;
1319
1320 /* destination address check */
1321 if (sin6) {
1322 if (addr_len < offsetof(struct sockaddr, sa_data))
1323 return -EINVAL;
1324
1325 switch (sin6->sin6_family) {
1326 case AF_INET6:
1327 if (addr_len < SIN6_LEN_RFC2133)
1328 return -EINVAL;
1329 daddr = &sin6->sin6_addr;
1330 if (ipv6_addr_any(daddr) &&
1331 ipv6_addr_v4mapped(&np->saddr))
1332 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1333 daddr);
1334 break;
1335 case AF_INET:
1336 goto do_udp_sendmsg;
1337 case AF_UNSPEC:
1338 msg->msg_name = sin6 = NULL;
1339 msg->msg_namelen = addr_len = 0;
1340 daddr = NULL;
1341 break;
1342 default:
1343 return -EINVAL;
1344 }
1345 } else if (!up->pending) {
1346 if (sk->sk_state != TCP_ESTABLISHED)
1347 return -EDESTADDRREQ;
1348 daddr = &sk->sk_v6_daddr;
1349 } else
1350 daddr = NULL;
1351
1352 if (daddr) {
1353 if (ipv6_addr_v4mapped(daddr)) {
1354 struct sockaddr_in sin;
1355 sin.sin_family = AF_INET;
1356 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1357 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1358 msg->msg_name = &sin;
1359 msg->msg_namelen = sizeof(sin);
1360 do_udp_sendmsg:
1361 if (__ipv6_only_sock(sk))
1362 return -ENETUNREACH;
1363 return udp_sendmsg(sk, msg, len);
1364 }
1365 }
1366
1367 if (up->pending == AF_INET)
1368 return udp_sendmsg(sk, msg, len);
1369
1370 /* Rough check on arithmetic overflow,
1371 better check is made in ip6_append_data().
1372 */
1373 if (len > INT_MAX - sizeof(struct udphdr))
1374 return -EMSGSIZE;
1375
1376 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1377 if (up->pending) {
1378 /*
1379 * There are pending frames.
1380 * The socket lock must be held while it's corked.
1381 */
1382 lock_sock(sk);
1383 if (likely(up->pending)) {
1384 if (unlikely(up->pending != AF_INET6)) {
1385 release_sock(sk);
1386 return -EAFNOSUPPORT;
1387 }
1388 dst = NULL;
1389 goto do_append_data;
1390 }
1391 release_sock(sk);
1392 }
1393 ulen += sizeof(struct udphdr);
1394
1395 memset(&fl6, 0, sizeof(fl6));
1396
1397 if (sin6) {
1398 if (sin6->sin6_port == 0)
1399 return -EINVAL;
1400
1401 fl6.fl6_dport = sin6->sin6_port;
1402 daddr = &sin6->sin6_addr;
1403
1404 if (np->sndflow) {
1405 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1406 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1407 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1408 if (IS_ERR(flowlabel))
1409 return -EINVAL;
1410 }
1411 }
1412
1413 /*
1414 * Otherwise it will be difficult to maintain
1415 * sk->sk_dst_cache.
1416 */
1417 if (sk->sk_state == TCP_ESTABLISHED &&
1418 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1419 daddr = &sk->sk_v6_daddr;
1420
1421 if (addr_len >= sizeof(struct sockaddr_in6) &&
1422 sin6->sin6_scope_id &&
1423 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1424 fl6.flowi6_oif = sin6->sin6_scope_id;
1425 } else {
1426 if (sk->sk_state != TCP_ESTABLISHED)
1427 return -EDESTADDRREQ;
1428
1429 fl6.fl6_dport = inet->inet_dport;
1430 daddr = &sk->sk_v6_daddr;
1431 fl6.flowlabel = np->flow_label;
1432 connected = true;
1433 }
1434
1435 if (!fl6.flowi6_oif)
1436 fl6.flowi6_oif = sk->sk_bound_dev_if;
1437
1438 if (!fl6.flowi6_oif)
1439 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1440
1441 fl6.flowi6_uid = sk->sk_uid;
1442
1443 if (msg->msg_controllen) {
1444 opt = &opt_space;
1445 memset(opt, 0, sizeof(struct ipv6_txoptions));
1446 opt->tot_len = sizeof(*opt);
1447 ipc6.opt = opt;
1448
1449 err = udp_cmsg_send(sk, msg, &ipc6.gso_size);
1450 if (err > 0)
1451 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6,
1452 &ipc6);
1453 if (err < 0) {
1454 fl6_sock_release(flowlabel);
1455 return err;
1456 }
1457 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1458 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1459 if (IS_ERR(flowlabel))
1460 return -EINVAL;
1461 }
1462 if (!(opt->opt_nflen|opt->opt_flen))
1463 opt = NULL;
1464 connected = false;
1465 }
1466 if (!opt) {
1467 opt = txopt_get(np);
1468 opt_to_free = opt;
1469 }
1470 if (flowlabel)
1471 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1472 opt = ipv6_fixup_options(&opt_space, opt);
1473 ipc6.opt = opt;
1474
1475 fl6.flowi6_proto = sk->sk_protocol;
1476 fl6.flowi6_mark = ipc6.sockc.mark;
1477 fl6.daddr = *daddr;
1478 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1479 fl6.saddr = np->saddr;
1480 fl6.fl6_sport = inet->inet_sport;
1481
1482 if (cgroup_bpf_enabled(CGROUP_UDP6_SENDMSG) && !connected) {
1483 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk,
1484 (struct sockaddr *)sin6, &fl6.saddr);
1485 if (err)
1486 goto out_no_dst;
1487 if (sin6) {
1488 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
1489 /* BPF program rewrote IPv6-only by IPv4-mapped
1490 * IPv6. It's currently unsupported.
1491 */
1492 err = -ENOTSUPP;
1493 goto out_no_dst;
1494 }
1495 if (sin6->sin6_port == 0) {
1496 /* BPF program set invalid port. Reject it. */
1497 err = -EINVAL;
1498 goto out_no_dst;
1499 }
1500 fl6.fl6_dport = sin6->sin6_port;
1501 fl6.daddr = sin6->sin6_addr;
1502 }
1503 }
1504
1505 if (ipv6_addr_any(&fl6.daddr))
1506 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1507
1508 final_p = fl6_update_dst(&fl6, opt, &final);
1509 if (final_p)
1510 connected = false;
1511
1512 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1513 fl6.flowi6_oif = np->mcast_oif;
1514 connected = false;
1515 } else if (!fl6.flowi6_oif)
1516 fl6.flowi6_oif = np->ucast_oif;
1517
1518 security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6));
1519
1520 if (ipc6.tclass < 0)
1521 ipc6.tclass = np->tclass;
1522
1523 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel);
1524
1525 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, connected);
1526 if (IS_ERR(dst)) {
1527 err = PTR_ERR(dst);
1528 dst = NULL;
1529 goto out;
1530 }
1531
1532 if (ipc6.hlimit < 0)
1533 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1534
1535 if (msg->msg_flags&MSG_CONFIRM)
1536 goto do_confirm;
1537 back_from_confirm:
1538
1539 /* Lockless fast path for the non-corking case */
1540 if (!corkreq) {
1541 struct inet_cork_full cork;
1542 struct sk_buff *skb;
1543
1544 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1545 sizeof(struct udphdr), &ipc6,
1546 &fl6, (struct rt6_info *)dst,
1547 msg->msg_flags, &cork);
1548 err = PTR_ERR(skb);
1549 if (!IS_ERR_OR_NULL(skb))
1550 err = udp_v6_send_skb(skb, &fl6, &cork.base);
1551 goto out;
1552 }
1553
1554 lock_sock(sk);
1555 if (unlikely(up->pending)) {
1556 /* The socket is already corked while preparing it. */
1557 /* ... which is an evident application bug. --ANK */
1558 release_sock(sk);
1559
1560 net_dbg_ratelimited("udp cork app bug 2\n");
1561 err = -EINVAL;
1562 goto out;
1563 }
1564
1565 up->pending = AF_INET6;
1566
1567 do_append_data:
1568 if (ipc6.dontfrag < 0)
1569 ipc6.dontfrag = np->dontfrag;
1570 up->len += ulen;
1571 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1572 &ipc6, &fl6, (struct rt6_info *)dst,
1573 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1574 if (err)
1575 udp_v6_flush_pending_frames(sk);
1576 else if (!corkreq)
1577 err = udp_v6_push_pending_frames(sk);
1578 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1579 up->pending = 0;
1580
1581 if (err > 0)
1582 err = np->recverr ? net_xmit_errno(err) : 0;
1583 release_sock(sk);
1584
1585 out:
1586 dst_release(dst);
1587 out_no_dst:
1588 fl6_sock_release(flowlabel);
1589 txopt_put(opt_to_free);
1590 if (!err)
1591 return len;
1592 /*
1593 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1594 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1595 * we don't have a good statistic (IpOutDiscards but it can be too many
1596 * things). We could add another new stat but at least for now that
1597 * seems like overkill.
1598 */
1599 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1600 UDP6_INC_STATS(sock_net(sk),
1601 UDP_MIB_SNDBUFERRORS, is_udplite);
1602 }
1603 return err;
1604
1605 do_confirm:
1606 if (msg->msg_flags & MSG_PROBE)
1607 dst_confirm_neigh(dst, &fl6.daddr);
1608 if (!(msg->msg_flags&MSG_PROBE) || len)
1609 goto back_from_confirm;
1610 err = 0;
1611 goto out;
1612 }
1613
1614 void udpv6_destroy_sock(struct sock *sk)
1615 {
1616 struct udp_sock *up = udp_sk(sk);
1617 lock_sock(sk);
1618
1619 /* protects from races with udp_abort() */
1620 sock_set_flag(sk, SOCK_DEAD);
1621 udp_v6_flush_pending_frames(sk);
1622 release_sock(sk);
1623
1624 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
1625 if (up->encap_type) {
1626 void (*encap_destroy)(struct sock *sk);
1627 encap_destroy = READ_ONCE(up->encap_destroy);
1628 if (encap_destroy)
1629 encap_destroy(sk);
1630 }
1631 if (up->encap_enabled) {
1632 static_branch_dec(&udpv6_encap_needed_key);
1633 udp_encap_disable();
1634 }
1635 }
1636
1637 inet6_destroy_sock(sk);
1638 }
1639
1640 /*
1641 * Socket option code for UDP
1642 */
1643 int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
1644 unsigned int optlen)
1645 {
1646 if (level == SOL_UDP || level == SOL_UDPLITE)
1647 return udp_lib_setsockopt(sk, level, optname,
1648 optval, optlen,
1649 udp_v6_push_pending_frames);
1650 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1651 }
1652
1653 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1654 char __user *optval, int __user *optlen)
1655 {
1656 if (level == SOL_UDP || level == SOL_UDPLITE)
1657 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1658 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1659 }
1660
1661 /* thinking of making this const? Don't.
1662 * early_demux can change based on sysctl.
1663 */
1664 static struct inet6_protocol udpv6_protocol = {
1665 .early_demux = udp_v6_early_demux,
1666 .early_demux_handler = udp_v6_early_demux,
1667 .handler = udpv6_rcv,
1668 .err_handler = udpv6_err,
1669 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1670 };
1671
1672 /* ------------------------------------------------------------------------ */
1673 #ifdef CONFIG_PROC_FS
1674 int udp6_seq_show(struct seq_file *seq, void *v)
1675 {
1676 if (v == SEQ_START_TOKEN) {
1677 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1678 } else {
1679 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1680 struct inet_sock *inet = inet_sk(v);
1681 __u16 srcp = ntohs(inet->inet_sport);
1682 __u16 destp = ntohs(inet->inet_dport);
1683 __ip6_dgram_sock_seq_show(seq, v, srcp, destp,
1684 udp_rqueue_get(v), bucket);
1685 }
1686 return 0;
1687 }
1688
1689 const struct seq_operations udp6_seq_ops = {
1690 .start = udp_seq_start,
1691 .next = udp_seq_next,
1692 .stop = udp_seq_stop,
1693 .show = udp6_seq_show,
1694 };
1695 EXPORT_SYMBOL(udp6_seq_ops);
1696
1697 static struct udp_seq_afinfo udp6_seq_afinfo = {
1698 .family = AF_INET6,
1699 .udp_table = &udp_table,
1700 };
1701
1702 int __net_init udp6_proc_init(struct net *net)
1703 {
1704 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops,
1705 sizeof(struct udp_iter_state), &udp6_seq_afinfo))
1706 return -ENOMEM;
1707 return 0;
1708 }
1709
1710 void udp6_proc_exit(struct net *net)
1711 {
1712 remove_proc_entry("udp6", net->proc_net);
1713 }
1714 #endif /* CONFIG_PROC_FS */
1715
1716 /* ------------------------------------------------------------------------ */
1717
1718 struct proto udpv6_prot = {
1719 .name = "UDPv6",
1720 .owner = THIS_MODULE,
1721 .close = udp_lib_close,
1722 .pre_connect = udpv6_pre_connect,
1723 .connect = ip6_datagram_connect,
1724 .disconnect = udp_disconnect,
1725 .ioctl = udp_ioctl,
1726 .init = udp_init_sock,
1727 .destroy = udpv6_destroy_sock,
1728 .setsockopt = udpv6_setsockopt,
1729 .getsockopt = udpv6_getsockopt,
1730 .sendmsg = udpv6_sendmsg,
1731 .recvmsg = udpv6_recvmsg,
1732 .release_cb = ip6_datagram_release_cb,
1733 .hash = udp_lib_hash,
1734 .unhash = udp_lib_unhash,
1735 .rehash = udp_v6_rehash,
1736 .get_port = udp_v6_get_port,
1737 #ifdef CONFIG_BPF_SYSCALL
1738 .psock_update_sk_prot = udp_bpf_update_proto,
1739 #endif
1740 .memory_allocated = &udp_memory_allocated,
1741 .sysctl_mem = sysctl_udp_mem,
1742 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
1743 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
1744 .obj_size = sizeof(struct udp6_sock),
1745 .h.udp_table = &udp_table,
1746 .diag_destroy = udp_abort,
1747 };
1748
1749 static struct inet_protosw udpv6_protosw = {
1750 .type = SOCK_DGRAM,
1751 .protocol = IPPROTO_UDP,
1752 .prot = &udpv6_prot,
1753 .ops = &inet6_dgram_ops,
1754 .flags = INET_PROTOSW_PERMANENT,
1755 };
1756
1757 int __init udpv6_init(void)
1758 {
1759 int ret;
1760
1761 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1762 if (ret)
1763 goto out;
1764
1765 ret = inet6_register_protosw(&udpv6_protosw);
1766 if (ret)
1767 goto out_udpv6_protocol;
1768 out:
1769 return ret;
1770
1771 out_udpv6_protocol:
1772 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1773 goto out;
1774 }
1775
1776 void udpv6_exit(void)
1777 {
1778 inet6_unregister_protosw(&udpv6_protosw);
1779 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1780 }
1781
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.