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/tcp_states.h>
44 #include <net/ip6_checksum.h>
45 #include <net/ip6_tunnel.h>
46 #include <net/xfrm.h>
47 #include <net/inet_hashtables.h>
48 #include <net/inet6_hashtables.h>
49 #include <net/busy_poll.h>
50 #include <net/sock_reuseport.h>
51
52 #include <linux/proc_fs.h>
53 #include <linux/seq_file.h>
54 #include <trace/events/skb.h>
55 #include "udp_impl.h"
56
57 static u32 udp6_ehashfn(const struct net *net,
58 const struct in6_addr *laddr,
59 const u16 lport,
60 const struct in6_addr *faddr,
61 const __be16 fport)
62 {
63 static u32 udp6_ehash_secret __read_mostly;
64 static u32 udp_ipv6_hash_secret __read_mostly;
65
66 u32 lhash, fhash;
67
68 net_get_random_once(&udp6_ehash_secret,
69 sizeof(udp6_ehash_secret));
70 net_get_random_once(&udp_ipv6_hash_secret,
71 sizeof(udp_ipv6_hash_secret));
72
73 lhash = (__force u32)laddr->s6_addr32[3];
74 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
75
76 return __inet6_ehashfn(lhash, lport, fhash, fport,
77 udp_ipv6_hash_secret + net_hash_mix(net));
78 }
79
80 int udp_v6_get_port(struct sock *sk, unsigned short snum)
81 {
82 unsigned int hash2_nulladdr =
83 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
84 unsigned int hash2_partial =
85 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
86
87 /* precompute partial secondary hash */
88 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
89 return udp_lib_get_port(sk, snum, hash2_nulladdr);
90 }
91
92 void udp_v6_rehash(struct sock *sk)
93 {
94 u16 new_hash = ipv6_portaddr_hash(sock_net(sk),
95 &sk->sk_v6_rcv_saddr,
96 inet_sk(sk)->inet_num);
97
98 udp_lib_rehash(sk, new_hash);
99 }
100
101 static int compute_score(struct sock *sk, struct net *net,
102 const struct in6_addr *saddr, __be16 sport,
103 const struct in6_addr *daddr, unsigned short hnum,
104 int dif, int sdif)
105 {
106 int score;
107 struct inet_sock *inet;
108 bool dev_match;
109
110 if (!net_eq(sock_net(sk), net) ||
111 udp_sk(sk)->udp_port_hash != hnum ||
112 sk->sk_family != PF_INET6)
113 return -1;
114
115 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
116 return -1;
117
118 score = 0;
119 inet = inet_sk(sk);
120
121 if (inet->inet_dport) {
122 if (inet->inet_dport != sport)
123 return -1;
124 score++;
125 }
126
127 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
128 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
129 return -1;
130 score++;
131 }
132
133 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif);
134 if (!dev_match)
135 return -1;
136 score++;
137
138 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
139 score++;
140
141 return score;
142 }
143
144 /* called with rcu_read_lock() */
145 static struct sock *udp6_lib_lookup2(struct net *net,
146 const struct in6_addr *saddr, __be16 sport,
147 const struct in6_addr *daddr, unsigned int hnum,
148 int dif, int sdif, struct udp_hslot *hslot2,
149 struct sk_buff *skb)
150 {
151 struct sock *sk, *result;
152 int score, badness;
153 u32 hash = 0;
154
155 result = NULL;
156 badness = -1;
157 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
158 score = compute_score(sk, net, saddr, sport,
159 daddr, hnum, dif, sdif);
160 if (score > badness) {
161 if (sk->sk_reuseport &&
162 sk->sk_state != TCP_ESTABLISHED) {
163 hash = udp6_ehashfn(net, daddr, hnum,
164 saddr, sport);
165
166 result = reuseport_select_sock(sk, hash, skb,
167 sizeof(struct udphdr));
168 if (result && !reuseport_has_conns(sk, false))
169 return result;
170 }
171 result = sk;
172 badness = score;
173 }
174 }
175 return result;
176 }
177
178 /* rcu_read_lock() must be held */
179 struct sock *__udp6_lib_lookup(struct net *net,
180 const struct in6_addr *saddr, __be16 sport,
181 const struct in6_addr *daddr, __be16 dport,
182 int dif, int sdif, struct udp_table *udptable,
183 struct sk_buff *skb)
184 {
185 unsigned short hnum = ntohs(dport);
186 unsigned int hash2, slot2;
187 struct udp_hslot *hslot2;
188 struct sock *result;
189
190 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
191 slot2 = hash2 & udptable->mask;
192 hslot2 = &udptable->hash2[slot2];
193
194 result = udp6_lib_lookup2(net, saddr, sport,
195 daddr, hnum, dif, sdif,
196 hslot2, skb);
197 if (!result) {
198 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
199 slot2 = hash2 & udptable->mask;
200
201 hslot2 = &udptable->hash2[slot2];
202
203 result = udp6_lib_lookup2(net, saddr, sport,
204 &in6addr_any, hnum, dif, sdif,
205 hslot2, skb);
206 }
207 if (IS_ERR(result))
208 return NULL;
209 return result;
210 }
211 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
212
213 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
214 __be16 sport, __be16 dport,
215 struct udp_table *udptable)
216 {
217 const struct ipv6hdr *iph = ipv6_hdr(skb);
218
219 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
220 &iph->daddr, dport, inet6_iif(skb),
221 inet6_sdif(skb), udptable, skb);
222 }
223
224 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb,
225 __be16 sport, __be16 dport)
226 {
227 const struct ipv6hdr *iph = ipv6_hdr(skb);
228
229 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
230 &iph->daddr, dport, inet6_iif(skb),
231 inet6_sdif(skb), &udp_table, NULL);
232 }
233 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb);
234
235 /* Must be called under rcu_read_lock().
236 * Does increment socket refcount.
237 */
238 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
239 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
240 const struct in6_addr *daddr, __be16 dport, int dif)
241 {
242 struct sock *sk;
243
244 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport,
245 dif, 0, &udp_table, NULL);
246 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
247 sk = NULL;
248 return sk;
249 }
250 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
251 #endif
252
253 /* do not use the scratch area len for jumbogram: their length execeeds the
254 * scratch area space; note that the IP6CB flags is still in the first
255 * cacheline, so checking for jumbograms is cheap
256 */
257 static int udp6_skb_len(struct sk_buff *skb)
258 {
259 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
260 }
261
262 /*
263 * This should be easy, if there is something there we
264 * return it, otherwise we block.
265 */
266
267 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
268 int noblock, int flags, int *addr_len)
269 {
270 struct ipv6_pinfo *np = inet6_sk(sk);
271 struct inet_sock *inet = inet_sk(sk);
272 struct sk_buff *skb;
273 unsigned int ulen, copied;
274 int off, err, peeking = flags & MSG_PEEK;
275 int is_udplite = IS_UDPLITE(sk);
276 struct udp_mib __percpu *mib;
277 bool checksum_valid = false;
278 int is_udp4;
279
280 if (flags & MSG_ERRQUEUE)
281 return ipv6_recv_error(sk, msg, len, addr_len);
282
283 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
284 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
285
286 try_again:
287 off = sk_peek_offset(sk, flags);
288 skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
289 if (!skb)
290 return err;
291 if (ccs_socket_post_recvmsg_permission(sk, skb, flags))
292 return -EAGAIN; /* Hope less harmful than -EPERM. */
293
294 ulen = udp6_skb_len(skb);
295 copied = len;
296 if (copied > ulen - off)
297 copied = ulen - off;
298 else if (copied < ulen)
299 msg->msg_flags |= MSG_TRUNC;
300
301 is_udp4 = (skb->protocol == htons(ETH_P_IP));
302 mib = __UDPX_MIB(sk, is_udp4);
303
304 /*
305 * If checksum is needed at all, try to do it while copying the
306 * data. If the data is truncated, or if we only want a partial
307 * coverage checksum (UDP-Lite), do it before the copy.
308 */
309
310 if (copied < ulen || peeking ||
311 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
312 checksum_valid = udp_skb_csum_unnecessary(skb) ||
313 !__udp_lib_checksum_complete(skb);
314 if (!checksum_valid)
315 goto csum_copy_err;
316 }
317
318 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
319 if (udp_skb_is_linear(skb))
320 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
321 else
322 err = skb_copy_datagram_msg(skb, off, msg, copied);
323 } else {
324 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
325 if (err == -EINVAL)
326 goto csum_copy_err;
327 }
328 if (unlikely(err)) {
329 if (!peeking) {
330 atomic_inc(&sk->sk_drops);
331 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
332 }
333 kfree_skb(skb);
334 return err;
335 }
336 if (!peeking)
337 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS);
338
339 sock_recv_ts_and_drops(msg, sk, skb);
340
341 /* Copy the address. */
342 if (msg->msg_name) {
343 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
344 sin6->sin6_family = AF_INET6;
345 sin6->sin6_port = udp_hdr(skb)->source;
346 sin6->sin6_flowinfo = 0;
347
348 if (is_udp4) {
349 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
350 &sin6->sin6_addr);
351 sin6->sin6_scope_id = 0;
352 } else {
353 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
354 sin6->sin6_scope_id =
355 ipv6_iface_scope_id(&sin6->sin6_addr,
356 inet6_iif(skb));
357 }
358 *addr_len = sizeof(*sin6);
359
360 if (cgroup_bpf_enabled)
361 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk,
362 (struct sockaddr *)sin6);
363 }
364
365 if (udp_sk(sk)->gro_enabled)
366 udp_cmsg_recv(msg, sk, skb);
367
368 if (np->rxopt.all)
369 ip6_datagram_recv_common_ctl(sk, msg, skb);
370
371 if (is_udp4) {
372 if (inet->cmsg_flags)
373 ip_cmsg_recv_offset(msg, sk, skb,
374 sizeof(struct udphdr), off);
375 } else {
376 if (np->rxopt.all)
377 ip6_datagram_recv_specific_ctl(sk, msg, skb);
378 }
379
380 err = copied;
381 if (flags & MSG_TRUNC)
382 err = ulen;
383
384 skb_consume_udp(sk, skb, peeking ? -err : err);
385 return err;
386
387 csum_copy_err:
388 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
389 udp_skb_destructor)) {
390 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS);
391 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
392 }
393 kfree_skb(skb);
394
395 /* starting over for a new packet, but check if we need to yield */
396 cond_resched();
397 msg->msg_flags &= ~MSG_TRUNC;
398 goto try_again;
399 }
400
401 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
402 void udpv6_encap_enable(void)
403 {
404 static_branch_inc(&udpv6_encap_needed_key);
405 }
406 EXPORT_SYMBOL(udpv6_encap_enable);
407
408 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
409 * through error handlers in encapsulations looking for a match.
410 */
411 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb,
412 struct inet6_skb_parm *opt,
413 u8 type, u8 code, int offset, __be32 info)
414 {
415 int i;
416
417 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
418 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt,
419 u8 type, u8 code, int offset, __be32 info);
420 const struct ip6_tnl_encap_ops *encap;
421
422 encap = rcu_dereference(ip6tun_encaps[i]);
423 if (!encap)
424 continue;
425 handler = encap->err_handler;
426 if (handler && !handler(skb, opt, type, code, offset, info))
427 return 0;
428 }
429
430 return -ENOENT;
431 }
432
433 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
434 * reversing source and destination port: this will match tunnels that force the
435 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
436 * lwtunnels might actually break this assumption by being configured with
437 * different destination ports on endpoints, in this case we won't be able to
438 * trace ICMP messages back to them.
439 *
440 * If this doesn't match any socket, probe tunnels with arbitrary destination
441 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
442 * we've sent packets to won't necessarily match the local destination port.
443 *
444 * Then ask the tunnel implementation to match the error against a valid
445 * association.
446 *
447 * Return an error if we can't find a match, the socket if we need further
448 * processing, zero otherwise.
449 */
450 static struct sock *__udp6_lib_err_encap(struct net *net,
451 const struct ipv6hdr *hdr, int offset,
452 struct udphdr *uh,
453 struct udp_table *udptable,
454 struct sk_buff *skb,
455 struct inet6_skb_parm *opt,
456 u8 type, u8 code, __be32 info)
457 {
458 int network_offset, transport_offset;
459 struct sock *sk;
460
461 network_offset = skb_network_offset(skb);
462 transport_offset = skb_transport_offset(skb);
463
464 /* Network header needs to point to the outer IPv6 header inside ICMP */
465 skb_reset_network_header(skb);
466
467 /* Transport header needs to point to the UDP header */
468 skb_set_transport_header(skb, offset);
469
470 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source,
471 &hdr->saddr, uh->dest,
472 inet6_iif(skb), 0, udptable, skb);
473 if (sk) {
474 int (*lookup)(struct sock *sk, struct sk_buff *skb);
475 struct udp_sock *up = udp_sk(sk);
476
477 lookup = READ_ONCE(up->encap_err_lookup);
478 if (!lookup || lookup(sk, skb))
479 sk = NULL;
480 }
481
482 if (!sk) {
483 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code,
484 offset, info));
485 }
486
487 skb_set_transport_header(skb, transport_offset);
488 skb_set_network_header(skb, network_offset);
489
490 return sk;
491 }
492
493 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
494 u8 type, u8 code, int offset, __be32 info,
495 struct udp_table *udptable)
496 {
497 struct ipv6_pinfo *np;
498 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
499 const struct in6_addr *saddr = &hdr->saddr;
500 const struct in6_addr *daddr = &hdr->daddr;
501 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
502 bool tunnel = false;
503 struct sock *sk;
504 int harderr;
505 int err;
506 struct net *net = dev_net(skb->dev);
507
508 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
509 inet6_iif(skb), inet6_sdif(skb), udptable, NULL);
510 if (!sk) {
511 /* No socket for error: try tunnels before discarding */
512 sk = ERR_PTR(-ENOENT);
513 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
514 sk = __udp6_lib_err_encap(net, hdr, offset, uh,
515 udptable, skb,
516 opt, type, code, info);
517 if (!sk)
518 return 0;
519 }
520
521 if (IS_ERR(sk)) {
522 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
523 ICMP6_MIB_INERRORS);
524 return PTR_ERR(sk);
525 }
526
527 tunnel = true;
528 }
529
530 harderr = icmpv6_err_convert(type, code, &err);
531 np = inet6_sk(sk);
532
533 if (type == ICMPV6_PKT_TOOBIG) {
534 if (!ip6_sk_accept_pmtu(sk))
535 goto out;
536 ip6_sk_update_pmtu(skb, sk, info);
537 if (np->pmtudisc != IPV6_PMTUDISC_DONT)
538 harderr = 1;
539 }
540 if (type == NDISC_REDIRECT) {
541 if (tunnel) {
542 ip6_redirect(skb, sock_net(sk), inet6_iif(skb),
543 sk->sk_mark, sk->sk_uid);
544 } else {
545 ip6_sk_redirect(skb, sk);
546 }
547 goto out;
548 }
549
550 /* Tunnels don't have an application socket: don't pass errors back */
551 if (tunnel)
552 goto out;
553
554 if (!np->recverr) {
555 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
556 goto out;
557 } else {
558 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
559 }
560
561 sk->sk_err = err;
562 sk->sk_error_report(sk);
563 out:
564 return 0;
565 }
566
567 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
568 {
569 int rc;
570
571 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
572 sock_rps_save_rxhash(sk, skb);
573 sk_mark_napi_id(sk, skb);
574 sk_incoming_cpu_update(sk);
575 } else {
576 sk_mark_napi_id_once(sk, skb);
577 }
578
579 rc = __udp_enqueue_schedule_skb(sk, skb);
580 if (rc < 0) {
581 int is_udplite = IS_UDPLITE(sk);
582
583 /* Note that an ENOMEM error is charged twice */
584 if (rc == -ENOMEM)
585 UDP6_INC_STATS(sock_net(sk),
586 UDP_MIB_RCVBUFERRORS, is_udplite);
587 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
588 kfree_skb(skb);
589 return -1;
590 }
591
592 return 0;
593 }
594
595 static __inline__ int udpv6_err(struct sk_buff *skb,
596 struct inet6_skb_parm *opt, u8 type,
597 u8 code, int offset, __be32 info)
598 {
599 return __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
600 }
601
602 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
603 {
604 struct udp_sock *up = udp_sk(sk);
605 int is_udplite = IS_UDPLITE(sk);
606
607 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
608 goto drop;
609
610 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) {
611 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
612
613 /*
614 * This is an encapsulation socket so pass the skb to
615 * the socket's udp_encap_rcv() hook. Otherwise, just
616 * fall through and pass this up the UDP socket.
617 * up->encap_rcv() returns the following value:
618 * =0 if skb was successfully passed to the encap
619 * handler or was discarded by it.
620 * >0 if skb should be passed on to UDP.
621 * <0 if skb should be resubmitted as proto -N
622 */
623
624 /* if we're overly short, let UDP handle it */
625 encap_rcv = READ_ONCE(up->encap_rcv);
626 if (encap_rcv) {
627 int ret;
628
629 /* Verify checksum before giving to encap */
630 if (udp_lib_checksum_complete(skb))
631 goto csum_error;
632
633 ret = encap_rcv(sk, skb);
634 if (ret <= 0) {
635 __UDP_INC_STATS(sock_net(sk),
636 UDP_MIB_INDATAGRAMS,
637 is_udplite);
638 return -ret;
639 }
640 }
641
642 /* FALLTHROUGH -- it's a UDP Packet */
643 }
644
645 /*
646 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
647 */
648 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
649
650 if (up->pcrlen == 0) { /* full coverage was set */
651 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
652 UDP_SKB_CB(skb)->cscov, skb->len);
653 goto drop;
654 }
655 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
656 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
657 UDP_SKB_CB(skb)->cscov, up->pcrlen);
658 goto drop;
659 }
660 }
661
662 prefetch(&sk->sk_rmem_alloc);
663 if (rcu_access_pointer(sk->sk_filter) &&
664 udp_lib_checksum_complete(skb))
665 goto csum_error;
666
667 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
668 goto drop;
669
670 udp_csum_pull_header(skb);
671
672 skb_dst_drop(skb);
673
674 return __udpv6_queue_rcv_skb(sk, skb);
675
676 csum_error:
677 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
678 drop:
679 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
680 atomic_inc(&sk->sk_drops);
681 kfree_skb(skb);
682 return -1;
683 }
684
685 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
686 {
687 struct sk_buff *next, *segs;
688 int ret;
689
690 if (likely(!udp_unexpected_gso(sk, skb)))
691 return udpv6_queue_rcv_one_skb(sk, skb);
692
693 __skb_push(skb, -skb_mac_offset(skb));
694 segs = udp_rcv_segment(sk, skb, false);
695 skb_list_walk_safe(segs, skb, next) {
696 __skb_pull(skb, skb_transport_offset(skb));
697
698 ret = udpv6_queue_rcv_one_skb(sk, skb);
699 if (ret > 0)
700 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret,
701 true);
702 }
703 return 0;
704 }
705
706 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
707 __be16 loc_port, const struct in6_addr *loc_addr,
708 __be16 rmt_port, const struct in6_addr *rmt_addr,
709 int dif, int sdif, unsigned short hnum)
710 {
711 struct inet_sock *inet = inet_sk(sk);
712
713 if (!net_eq(sock_net(sk), net))
714 return false;
715
716 if (udp_sk(sk)->udp_port_hash != hnum ||
717 sk->sk_family != PF_INET6 ||
718 (inet->inet_dport && inet->inet_dport != rmt_port) ||
719 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
720 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
721 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif) ||
722 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
723 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
724 return false;
725 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
726 return false;
727 return true;
728 }
729
730 static void udp6_csum_zero_error(struct sk_buff *skb)
731 {
732 /* RFC 2460 section 8.1 says that we SHOULD log
733 * this error. Well, it is reasonable.
734 */
735 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
736 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
737 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
738 }
739
740 /*
741 * Note: called only from the BH handler context,
742 * so we don't need to lock the hashes.
743 */
744 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
745 const struct in6_addr *saddr, const struct in6_addr *daddr,
746 struct udp_table *udptable, int proto)
747 {
748 struct sock *sk, *first = NULL;
749 const struct udphdr *uh = udp_hdr(skb);
750 unsigned short hnum = ntohs(uh->dest);
751 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
752 unsigned int offset = offsetof(typeof(*sk), sk_node);
753 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
754 int dif = inet6_iif(skb);
755 int sdif = inet6_sdif(skb);
756 struct hlist_node *node;
757 struct sk_buff *nskb;
758
759 if (use_hash2) {
760 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) &
761 udptable->mask;
762 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask;
763 start_lookup:
764 hslot = &udptable->hash2[hash2];
765 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
766 }
767
768 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
769 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
770 uh->source, saddr, dif, sdif,
771 hnum))
772 continue;
773 /* If zero checksum and no_check is not on for
774 * the socket then skip it.
775 */
776 if (!uh->check && !udp_sk(sk)->no_check6_rx)
777 continue;
778 if (!first) {
779 first = sk;
780 continue;
781 }
782 nskb = skb_clone(skb, GFP_ATOMIC);
783 if (unlikely(!nskb)) {
784 atomic_inc(&sk->sk_drops);
785 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
786 IS_UDPLITE(sk));
787 __UDP6_INC_STATS(net, UDP_MIB_INERRORS,
788 IS_UDPLITE(sk));
789 continue;
790 }
791
792 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
793 consume_skb(nskb);
794 }
795
796 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
797 if (use_hash2 && hash2 != hash2_any) {
798 hash2 = hash2_any;
799 goto start_lookup;
800 }
801
802 if (first) {
803 if (udpv6_queue_rcv_skb(first, skb) > 0)
804 consume_skb(skb);
805 } else {
806 kfree_skb(skb);
807 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
808 proto == IPPROTO_UDPLITE);
809 }
810 return 0;
811 }
812
813 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
814 {
815 if (udp_sk_rx_dst_set(sk, dst)) {
816 const struct rt6_info *rt = (const struct rt6_info *)dst;
817
818 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
819 }
820 }
821
822 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
823 * return code conversion for ip layer consumption
824 */
825 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
826 struct udphdr *uh)
827 {
828 int ret;
829
830 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
831 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo);
832
833 ret = udpv6_queue_rcv_skb(sk, skb);
834
835 /* a return value > 0 means to resubmit the input */
836 if (ret > 0)
837 return ret;
838 return 0;
839 }
840
841 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
842 int proto)
843 {
844 const struct in6_addr *saddr, *daddr;
845 struct net *net = dev_net(skb->dev);
846 struct udphdr *uh;
847 struct sock *sk;
848 u32 ulen = 0;
849
850 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
851 goto discard;
852
853 saddr = &ipv6_hdr(skb)->saddr;
854 daddr = &ipv6_hdr(skb)->daddr;
855 uh = udp_hdr(skb);
856
857 ulen = ntohs(uh->len);
858 if (ulen > skb->len)
859 goto short_packet;
860
861 if (proto == IPPROTO_UDP) {
862 /* UDP validates ulen. */
863
864 /* Check for jumbo payload */
865 if (ulen == 0)
866 ulen = skb->len;
867
868 if (ulen < sizeof(*uh))
869 goto short_packet;
870
871 if (ulen < skb->len) {
872 if (pskb_trim_rcsum(skb, ulen))
873 goto short_packet;
874 saddr = &ipv6_hdr(skb)->saddr;
875 daddr = &ipv6_hdr(skb)->daddr;
876 uh = udp_hdr(skb);
877 }
878 }
879
880 if (udp6_csum_init(skb, uh, proto))
881 goto csum_error;
882
883 /* Check if the socket is already available, e.g. due to early demux */
884 sk = skb_steal_sock(skb);
885 if (sk) {
886 struct dst_entry *dst = skb_dst(skb);
887 int ret;
888
889 if (unlikely(sk->sk_rx_dst != dst))
890 udp6_sk_rx_dst_set(sk, dst);
891
892 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
893 sock_put(sk);
894 goto report_csum_error;
895 }
896
897 ret = udp6_unicast_rcv_skb(sk, skb, uh);
898 sock_put(sk);
899 return ret;
900 }
901
902 /*
903 * Multicast receive code
904 */
905 if (ipv6_addr_is_multicast(daddr))
906 return __udp6_lib_mcast_deliver(net, skb,
907 saddr, daddr, udptable, proto);
908
909 /* Unicast */
910 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
911 if (sk) {
912 if (!uh->check && !udp_sk(sk)->no_check6_rx)
913 goto report_csum_error;
914 return udp6_unicast_rcv_skb(sk, skb, uh);
915 }
916
917 if (!uh->check)
918 goto report_csum_error;
919
920 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
921 goto discard;
922
923 if (udp_lib_checksum_complete(skb))
924 goto csum_error;
925
926 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
927 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
928
929 kfree_skb(skb);
930 return 0;
931
932 short_packet:
933 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
934 proto == IPPROTO_UDPLITE ? "-Lite" : "",
935 saddr, ntohs(uh->source),
936 ulen, skb->len,
937 daddr, ntohs(uh->dest));
938 goto discard;
939
940 report_csum_error:
941 udp6_csum_zero_error(skb);
942 csum_error:
943 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
944 discard:
945 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
946 kfree_skb(skb);
947 return 0;
948 }
949
950
951 static struct sock *__udp6_lib_demux_lookup(struct net *net,
952 __be16 loc_port, const struct in6_addr *loc_addr,
953 __be16 rmt_port, const struct in6_addr *rmt_addr,
954 int dif, int sdif)
955 {
956 unsigned short hnum = ntohs(loc_port);
957 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum);
958 unsigned int slot2 = hash2 & udp_table.mask;
959 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
960 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
961 struct sock *sk;
962
963 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
964 if (sk->sk_state == TCP_ESTABLISHED &&
965 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif))
966 return sk;
967 /* Only check first socket in chain */
968 break;
969 }
970 return NULL;
971 }
972
973 INDIRECT_CALLABLE_SCOPE void udp_v6_early_demux(struct sk_buff *skb)
974 {
975 struct net *net = dev_net(skb->dev);
976 const struct udphdr *uh;
977 struct sock *sk;
978 struct dst_entry *dst;
979 int dif = skb->dev->ifindex;
980 int sdif = inet6_sdif(skb);
981
982 if (!pskb_may_pull(skb, skb_transport_offset(skb) +
983 sizeof(struct udphdr)))
984 return;
985
986 uh = udp_hdr(skb);
987
988 if (skb->pkt_type == PACKET_HOST)
989 sk = __udp6_lib_demux_lookup(net, uh->dest,
990 &ipv6_hdr(skb)->daddr,
991 uh->source, &ipv6_hdr(skb)->saddr,
992 dif, sdif);
993 else
994 return;
995
996 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
997 return;
998
999 skb->sk = sk;
1000 skb->destructor = sock_efree;
1001 dst = READ_ONCE(sk->sk_rx_dst);
1002
1003 if (dst)
1004 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1005 if (dst) {
1006 /* set noref for now.
1007 * any place which wants to hold dst has to call
1008 * dst_hold_safe()
1009 */
1010 skb_dst_set_noref(skb, dst);
1011 }
1012 }
1013
1014 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb)
1015 {
1016 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1017 }
1018
1019 /*
1020 * Throw away all pending data and cancel the corking. Socket is locked.
1021 */
1022 static void udp_v6_flush_pending_frames(struct sock *sk)
1023 {
1024 struct udp_sock *up = udp_sk(sk);
1025
1026 if (up->pending == AF_INET)
1027 udp_flush_pending_frames(sk);
1028 else if (up->pending) {
1029 up->len = 0;
1030 up->pending = 0;
1031 ip6_flush_pending_frames(sk);
1032 }
1033 }
1034
1035 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
1036 int addr_len)
1037 {
1038 if (addr_len < offsetofend(struct sockaddr, sa_family))
1039 return -EINVAL;
1040 /* The following checks are replicated from __ip6_datagram_connect()
1041 * and intended to prevent BPF program called below from accessing
1042 * bytes that are out of the bound specified by user in addr_len.
1043 */
1044 if (uaddr->sa_family == AF_INET) {
1045 if (__ipv6_only_sock(sk))
1046 return -EAFNOSUPPORT;
1047 return udp_pre_connect(sk, uaddr, addr_len);
1048 }
1049
1050 if (addr_len < SIN6_LEN_RFC2133)
1051 return -EINVAL;
1052
1053 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr);
1054 }
1055
1056 /**
1057 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
1058 * @sk: socket we are sending on
1059 * @skb: sk_buff containing the filled-in UDP header
1060 * (checksum field must be zeroed out)
1061 */
1062 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
1063 const struct in6_addr *saddr,
1064 const struct in6_addr *daddr, int len)
1065 {
1066 unsigned int offset;
1067 struct udphdr *uh = udp_hdr(skb);
1068 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
1069 __wsum csum = 0;
1070
1071 if (!frags) {
1072 /* Only one fragment on the socket. */
1073 skb->csum_start = skb_transport_header(skb) - skb->head;
1074 skb->csum_offset = offsetof(struct udphdr, check);
1075 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1076 } else {
1077 /*
1078 * HW-checksum won't work as there are two or more
1079 * fragments on the socket so that all csums of sk_buffs
1080 * should be together
1081 */
1082 offset = skb_transport_offset(skb);
1083 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1084 csum = skb->csum;
1085
1086 skb->ip_summed = CHECKSUM_NONE;
1087
1088 do {
1089 csum = csum_add(csum, frags->csum);
1090 } while ((frags = frags->next));
1091
1092 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1093 csum);
1094 if (uh->check == 0)
1095 uh->check = CSUM_MANGLED_0;
1096 }
1097 }
1098
1099 /*
1100 * Sending
1101 */
1102
1103 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6,
1104 struct inet_cork *cork)
1105 {
1106 struct sock *sk = skb->sk;
1107 struct udphdr *uh;
1108 int err = 0;
1109 int is_udplite = IS_UDPLITE(sk);
1110 __wsum csum = 0;
1111 int offset = skb_transport_offset(skb);
1112 int len = skb->len - offset;
1113 int datalen = len - sizeof(*uh);
1114
1115 /*
1116 * Create a UDP header
1117 */
1118 uh = udp_hdr(skb);
1119 uh->source = fl6->fl6_sport;
1120 uh->dest = fl6->fl6_dport;
1121 uh->len = htons(len);
1122 uh->check = 0;
1123
1124 if (cork->gso_size) {
1125 const int hlen = skb_network_header_len(skb) +
1126 sizeof(struct udphdr);
1127
1128 if (hlen + cork->gso_size > cork->fragsize) {
1129 kfree_skb(skb);
1130 return -EINVAL;
1131 }
1132 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
1133 kfree_skb(skb);
1134 return -EINVAL;
1135 }
1136 if (udp_sk(sk)->no_check6_tx) {
1137 kfree_skb(skb);
1138 return -EINVAL;
1139 }
1140 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
1141 dst_xfrm(skb_dst(skb))) {
1142 kfree_skb(skb);
1143 return -EIO;
1144 }
1145
1146 if (datalen > cork->gso_size) {
1147 skb_shinfo(skb)->gso_size = cork->gso_size;
1148 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1149 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
1150 cork->gso_size);
1151 }
1152 goto csum_partial;
1153 }
1154
1155 if (is_udplite)
1156 csum = udplite_csum(skb);
1157 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
1158 skb->ip_summed = CHECKSUM_NONE;
1159 goto send;
1160 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1161 csum_partial:
1162 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1163 goto send;
1164 } else
1165 csum = udp_csum(skb);
1166
1167 /* add protocol-dependent pseudo-header */
1168 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1169 len, fl6->flowi6_proto, csum);
1170 if (uh->check == 0)
1171 uh->check = CSUM_MANGLED_0;
1172
1173 send:
1174 err = ip6_send_skb(skb);
1175 if (err) {
1176 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1177 UDP6_INC_STATS(sock_net(sk),
1178 UDP_MIB_SNDBUFERRORS, is_udplite);
1179 err = 0;
1180 }
1181 } else {
1182 UDP6_INC_STATS(sock_net(sk),
1183 UDP_MIB_OUTDATAGRAMS, is_udplite);
1184 }
1185 return err;
1186 }
1187
1188 static int udp_v6_push_pending_frames(struct sock *sk)
1189 {
1190 struct sk_buff *skb;
1191 struct udp_sock *up = udp_sk(sk);
1192 struct flowi6 fl6;
1193 int err = 0;
1194
1195 if (up->pending == AF_INET)
1196 return udp_push_pending_frames(sk);
1197
1198 /* ip6_finish_skb will release the cork, so make a copy of
1199 * fl6 here.
1200 */
1201 fl6 = inet_sk(sk)->cork.fl.u.ip6;
1202
1203 skb = ip6_finish_skb(sk);
1204 if (!skb)
1205 goto out;
1206
1207 err = udp_v6_send_skb(skb, &fl6, &inet_sk(sk)->cork.base);
1208
1209 out:
1210 up->len = 0;
1211 up->pending = 0;
1212 return err;
1213 }
1214
1215 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1216 {
1217 struct ipv6_txoptions opt_space;
1218 struct udp_sock *up = udp_sk(sk);
1219 struct inet_sock *inet = inet_sk(sk);
1220 struct ipv6_pinfo *np = inet6_sk(sk);
1221 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1222 struct in6_addr *daddr, *final_p, final;
1223 struct ipv6_txoptions *opt = NULL;
1224 struct ipv6_txoptions *opt_to_free = NULL;
1225 struct ip6_flowlabel *flowlabel = NULL;
1226 struct flowi6 fl6;
1227 struct dst_entry *dst;
1228 struct ipcm6_cookie ipc6;
1229 int addr_len = msg->msg_namelen;
1230 bool connected = false;
1231 int ulen = len;
1232 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1233 int err;
1234 int is_udplite = IS_UDPLITE(sk);
1235 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1236
1237 ipcm6_init(&ipc6);
1238 ipc6.gso_size = up->gso_size;
1239 ipc6.sockc.tsflags = sk->sk_tsflags;
1240 ipc6.sockc.mark = sk->sk_mark;
1241
1242 /* destination address check */
1243 if (sin6) {
1244 if (addr_len < offsetof(struct sockaddr, sa_data))
1245 return -EINVAL;
1246
1247 switch (sin6->sin6_family) {
1248 case AF_INET6:
1249 if (addr_len < SIN6_LEN_RFC2133)
1250 return -EINVAL;
1251 daddr = &sin6->sin6_addr;
1252 if (ipv6_addr_any(daddr) &&
1253 ipv6_addr_v4mapped(&np->saddr))
1254 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1255 daddr);
1256 break;
1257 case AF_INET:
1258 goto do_udp_sendmsg;
1259 case AF_UNSPEC:
1260 msg->msg_name = sin6 = NULL;
1261 msg->msg_namelen = addr_len = 0;
1262 daddr = NULL;
1263 break;
1264 default:
1265 return -EINVAL;
1266 }
1267 } else if (!up->pending) {
1268 if (sk->sk_state != TCP_ESTABLISHED)
1269 return -EDESTADDRREQ;
1270 daddr = &sk->sk_v6_daddr;
1271 } else
1272 daddr = NULL;
1273
1274 if (daddr) {
1275 if (ipv6_addr_v4mapped(daddr)) {
1276 struct sockaddr_in sin;
1277 sin.sin_family = AF_INET;
1278 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1279 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1280 msg->msg_name = &sin;
1281 msg->msg_namelen = sizeof(sin);
1282 do_udp_sendmsg:
1283 if (__ipv6_only_sock(sk))
1284 return -ENETUNREACH;
1285 return udp_sendmsg(sk, msg, len);
1286 }
1287 }
1288
1289 if (up->pending == AF_INET)
1290 return udp_sendmsg(sk, msg, len);
1291
1292 /* Rough check on arithmetic overflow,
1293 better check is made in ip6_append_data().
1294 */
1295 if (len > INT_MAX - sizeof(struct udphdr))
1296 return -EMSGSIZE;
1297
1298 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1299 if (up->pending) {
1300 /*
1301 * There are pending frames.
1302 * The socket lock must be held while it's corked.
1303 */
1304 lock_sock(sk);
1305 if (likely(up->pending)) {
1306 if (unlikely(up->pending != AF_INET6)) {
1307 release_sock(sk);
1308 return -EAFNOSUPPORT;
1309 }
1310 dst = NULL;
1311 goto do_append_data;
1312 }
1313 release_sock(sk);
1314 }
1315 ulen += sizeof(struct udphdr);
1316
1317 memset(&fl6, 0, sizeof(fl6));
1318
1319 if (sin6) {
1320 if (sin6->sin6_port == 0)
1321 return -EINVAL;
1322
1323 fl6.fl6_dport = sin6->sin6_port;
1324 daddr = &sin6->sin6_addr;
1325
1326 if (np->sndflow) {
1327 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1328 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1329 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1330 if (IS_ERR(flowlabel))
1331 return -EINVAL;
1332 }
1333 }
1334
1335 /*
1336 * Otherwise it will be difficult to maintain
1337 * sk->sk_dst_cache.
1338 */
1339 if (sk->sk_state == TCP_ESTABLISHED &&
1340 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1341 daddr = &sk->sk_v6_daddr;
1342
1343 if (addr_len >= sizeof(struct sockaddr_in6) &&
1344 sin6->sin6_scope_id &&
1345 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1346 fl6.flowi6_oif = sin6->sin6_scope_id;
1347 } else {
1348 if (sk->sk_state != TCP_ESTABLISHED)
1349 return -EDESTADDRREQ;
1350
1351 fl6.fl6_dport = inet->inet_dport;
1352 daddr = &sk->sk_v6_daddr;
1353 fl6.flowlabel = np->flow_label;
1354 connected = true;
1355 }
1356
1357 if (!fl6.flowi6_oif)
1358 fl6.flowi6_oif = sk->sk_bound_dev_if;
1359
1360 if (!fl6.flowi6_oif)
1361 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1362
1363 fl6.flowi6_mark = ipc6.sockc.mark;
1364 fl6.flowi6_uid = sk->sk_uid;
1365
1366 if (msg->msg_controllen) {
1367 opt = &opt_space;
1368 memset(opt, 0, sizeof(struct ipv6_txoptions));
1369 opt->tot_len = sizeof(*opt);
1370 ipc6.opt = opt;
1371
1372 err = udp_cmsg_send(sk, msg, &ipc6.gso_size);
1373 if (err > 0)
1374 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6,
1375 &ipc6);
1376 if (err < 0) {
1377 fl6_sock_release(flowlabel);
1378 return err;
1379 }
1380 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1381 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1382 if (IS_ERR(flowlabel))
1383 return -EINVAL;
1384 }
1385 if (!(opt->opt_nflen|opt->opt_flen))
1386 opt = NULL;
1387 connected = false;
1388 }
1389 if (!opt) {
1390 opt = txopt_get(np);
1391 opt_to_free = opt;
1392 }
1393 if (flowlabel)
1394 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1395 opt = ipv6_fixup_options(&opt_space, opt);
1396 ipc6.opt = opt;
1397
1398 fl6.flowi6_proto = sk->sk_protocol;
1399 fl6.daddr = *daddr;
1400 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1401 fl6.saddr = np->saddr;
1402 fl6.fl6_sport = inet->inet_sport;
1403
1404 if (cgroup_bpf_enabled && !connected) {
1405 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk,
1406 (struct sockaddr *)sin6, &fl6.saddr);
1407 if (err)
1408 goto out_no_dst;
1409 if (sin6) {
1410 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
1411 /* BPF program rewrote IPv6-only by IPv4-mapped
1412 * IPv6. It's currently unsupported.
1413 */
1414 err = -ENOTSUPP;
1415 goto out_no_dst;
1416 }
1417 if (sin6->sin6_port == 0) {
1418 /* BPF program set invalid port. Reject it. */
1419 err = -EINVAL;
1420 goto out_no_dst;
1421 }
1422 fl6.fl6_dport = sin6->sin6_port;
1423 fl6.daddr = sin6->sin6_addr;
1424 }
1425 }
1426
1427 if (ipv6_addr_any(&fl6.daddr))
1428 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1429
1430 final_p = fl6_update_dst(&fl6, opt, &final);
1431 if (final_p)
1432 connected = false;
1433
1434 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1435 fl6.flowi6_oif = np->mcast_oif;
1436 connected = false;
1437 } else if (!fl6.flowi6_oif)
1438 fl6.flowi6_oif = np->ucast_oif;
1439
1440 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1441
1442 if (ipc6.tclass < 0)
1443 ipc6.tclass = np->tclass;
1444
1445 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel);
1446
1447 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, connected);
1448 if (IS_ERR(dst)) {
1449 err = PTR_ERR(dst);
1450 dst = NULL;
1451 goto out;
1452 }
1453
1454 if (ipc6.hlimit < 0)
1455 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1456
1457 if (msg->msg_flags&MSG_CONFIRM)
1458 goto do_confirm;
1459 back_from_confirm:
1460
1461 /* Lockless fast path for the non-corking case */
1462 if (!corkreq) {
1463 struct inet_cork_full cork;
1464 struct sk_buff *skb;
1465
1466 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1467 sizeof(struct udphdr), &ipc6,
1468 &fl6, (struct rt6_info *)dst,
1469 msg->msg_flags, &cork);
1470 err = PTR_ERR(skb);
1471 if (!IS_ERR_OR_NULL(skb))
1472 err = udp_v6_send_skb(skb, &fl6, &cork.base);
1473 goto out;
1474 }
1475
1476 lock_sock(sk);
1477 if (unlikely(up->pending)) {
1478 /* The socket is already corked while preparing it. */
1479 /* ... which is an evident application bug. --ANK */
1480 release_sock(sk);
1481
1482 net_dbg_ratelimited("udp cork app bug 2\n");
1483 err = -EINVAL;
1484 goto out;
1485 }
1486
1487 up->pending = AF_INET6;
1488
1489 do_append_data:
1490 if (ipc6.dontfrag < 0)
1491 ipc6.dontfrag = np->dontfrag;
1492 up->len += ulen;
1493 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1494 &ipc6, &fl6, (struct rt6_info *)dst,
1495 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1496 if (err)
1497 udp_v6_flush_pending_frames(sk);
1498 else if (!corkreq)
1499 err = udp_v6_push_pending_frames(sk);
1500 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1501 up->pending = 0;
1502
1503 if (err > 0)
1504 err = np->recverr ? net_xmit_errno(err) : 0;
1505 release_sock(sk);
1506
1507 out:
1508 dst_release(dst);
1509 out_no_dst:
1510 fl6_sock_release(flowlabel);
1511 txopt_put(opt_to_free);
1512 if (!err)
1513 return len;
1514 /*
1515 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1516 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1517 * we don't have a good statistic (IpOutDiscards but it can be too many
1518 * things). We could add another new stat but at least for now that
1519 * seems like overkill.
1520 */
1521 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1522 UDP6_INC_STATS(sock_net(sk),
1523 UDP_MIB_SNDBUFERRORS, is_udplite);
1524 }
1525 return err;
1526
1527 do_confirm:
1528 if (msg->msg_flags & MSG_PROBE)
1529 dst_confirm_neigh(dst, &fl6.daddr);
1530 if (!(msg->msg_flags&MSG_PROBE) || len)
1531 goto back_from_confirm;
1532 err = 0;
1533 goto out;
1534 }
1535
1536 void udpv6_destroy_sock(struct sock *sk)
1537 {
1538 struct udp_sock *up = udp_sk(sk);
1539 lock_sock(sk);
1540 udp_v6_flush_pending_frames(sk);
1541 release_sock(sk);
1542
1543 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
1544 if (up->encap_type) {
1545 void (*encap_destroy)(struct sock *sk);
1546 encap_destroy = READ_ONCE(up->encap_destroy);
1547 if (encap_destroy)
1548 encap_destroy(sk);
1549 }
1550 if (up->encap_enabled)
1551 static_branch_dec(&udpv6_encap_needed_key);
1552 }
1553
1554 inet6_destroy_sock(sk);
1555 }
1556
1557 /*
1558 * Socket option code for UDP
1559 */
1560 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1561 char __user *optval, unsigned int optlen)
1562 {
1563 if (level == SOL_UDP || level == SOL_UDPLITE)
1564 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1565 udp_v6_push_pending_frames);
1566 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1567 }
1568
1569 #ifdef CONFIG_COMPAT
1570 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1571 char __user *optval, unsigned int optlen)
1572 {
1573 if (level == SOL_UDP || level == SOL_UDPLITE)
1574 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1575 udp_v6_push_pending_frames);
1576 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1577 }
1578 #endif
1579
1580 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1581 char __user *optval, int __user *optlen)
1582 {
1583 if (level == SOL_UDP || level == SOL_UDPLITE)
1584 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1585 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1586 }
1587
1588 #ifdef CONFIG_COMPAT
1589 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1590 char __user *optval, int __user *optlen)
1591 {
1592 if (level == SOL_UDP || level == SOL_UDPLITE)
1593 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1594 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1595 }
1596 #endif
1597
1598 /* thinking of making this const? Don't.
1599 * early_demux can change based on sysctl.
1600 */
1601 static struct inet6_protocol udpv6_protocol = {
1602 .early_demux = udp_v6_early_demux,
1603 .early_demux_handler = udp_v6_early_demux,
1604 .handler = udpv6_rcv,
1605 .err_handler = udpv6_err,
1606 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1607 };
1608
1609 /* ------------------------------------------------------------------------ */
1610 #ifdef CONFIG_PROC_FS
1611 int udp6_seq_show(struct seq_file *seq, void *v)
1612 {
1613 if (v == SEQ_START_TOKEN) {
1614 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1615 } else {
1616 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1617 struct inet_sock *inet = inet_sk(v);
1618 __u16 srcp = ntohs(inet->inet_sport);
1619 __u16 destp = ntohs(inet->inet_dport);
1620 __ip6_dgram_sock_seq_show(seq, v, srcp, destp,
1621 udp_rqueue_get(v), bucket);
1622 }
1623 return 0;
1624 }
1625
1626 const struct seq_operations udp6_seq_ops = {
1627 .start = udp_seq_start,
1628 .next = udp_seq_next,
1629 .stop = udp_seq_stop,
1630 .show = udp6_seq_show,
1631 };
1632 EXPORT_SYMBOL(udp6_seq_ops);
1633
1634 static struct udp_seq_afinfo udp6_seq_afinfo = {
1635 .family = AF_INET6,
1636 .udp_table = &udp_table,
1637 };
1638
1639 int __net_init udp6_proc_init(struct net *net)
1640 {
1641 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops,
1642 sizeof(struct udp_iter_state), &udp6_seq_afinfo))
1643 return -ENOMEM;
1644 return 0;
1645 }
1646
1647 void udp6_proc_exit(struct net *net)
1648 {
1649 remove_proc_entry("udp6", net->proc_net);
1650 }
1651 #endif /* CONFIG_PROC_FS */
1652
1653 /* ------------------------------------------------------------------------ */
1654
1655 struct proto udpv6_prot = {
1656 .name = "UDPv6",
1657 .owner = THIS_MODULE,
1658 .close = udp_lib_close,
1659 .pre_connect = udpv6_pre_connect,
1660 .connect = ip6_datagram_connect,
1661 .disconnect = udp_disconnect,
1662 .ioctl = udp_ioctl,
1663 .init = udp_init_sock,
1664 .destroy = udpv6_destroy_sock,
1665 .setsockopt = udpv6_setsockopt,
1666 .getsockopt = udpv6_getsockopt,
1667 .sendmsg = udpv6_sendmsg,
1668 .recvmsg = udpv6_recvmsg,
1669 .release_cb = ip6_datagram_release_cb,
1670 .hash = udp_lib_hash,
1671 .unhash = udp_lib_unhash,
1672 .rehash = udp_v6_rehash,
1673 .get_port = udp_v6_get_port,
1674 .memory_allocated = &udp_memory_allocated,
1675 .sysctl_mem = sysctl_udp_mem,
1676 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
1677 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
1678 .obj_size = sizeof(struct udp6_sock),
1679 .h.udp_table = &udp_table,
1680 #ifdef CONFIG_COMPAT
1681 .compat_setsockopt = compat_udpv6_setsockopt,
1682 .compat_getsockopt = compat_udpv6_getsockopt,
1683 #endif
1684 .diag_destroy = udp_abort,
1685 };
1686
1687 static struct inet_protosw udpv6_protosw = {
1688 .type = SOCK_DGRAM,
1689 .protocol = IPPROTO_UDP,
1690 .prot = &udpv6_prot,
1691 .ops = &inet6_dgram_ops,
1692 .flags = INET_PROTOSW_PERMANENT,
1693 };
1694
1695 int __init udpv6_init(void)
1696 {
1697 int ret;
1698
1699 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1700 if (ret)
1701 goto out;
1702
1703 ret = inet6_register_protosw(&udpv6_protosw);
1704 if (ret)
1705 goto out_udpv6_protocol;
1706 out:
1707 return ret;
1708
1709 out_udpv6_protocol:
1710 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1711 goto out;
1712 }
1713
1714 void udpv6_exit(void)
1715 {
1716 inet6_unregister_protosw(&udpv6_protosw);
1717 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1718 }
1719
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.