]> git.proxmox.com Git - mirror_ubuntu-focal-kernel.git/blob - net/ipv6/udp.c
projects / mirror_ubuntu-focal-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'sh-pfc-for-v5.1-tag2' of git://git.kernel.org/pub/scm/linux/kernel/git...
[mirror_ubuntu-focal-kernel.git] / net / ipv6 / udp.c
1 /*
2 * UDP over IPv6
3 * Linux INET6 implementation
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 *
8 * Based on linux/ipv4/udp.c
9 *
10 * Fixes:
11 * Hideaki YOSHIFUJI : sin6_scope_id support
12 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
13 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
14 * a single port at the same time.
15 * Kazunori MIYAZAWA @USAGI: change process style to use ip6_append_data
16 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/udp6 to seq_file.
17 *
18 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License
20 * as published by the Free Software Foundation; either version
21 * 2 of the License, or (at your option) any later version.
22 */
23
24 #include <linux/errno.h>
25 #include <linux/types.h>
26 #include <linux/socket.h>
27 #include <linux/sockios.h>
28 #include <linux/net.h>
29 #include <linux/in6.h>
30 #include <linux/netdevice.h>
31 #include <linux/if_arp.h>
32 #include <linux/ipv6.h>
33 #include <linux/icmpv6.h>
34 #include <linux/init.h>
35 #include <linux/module.h>
36 #include <linux/skbuff.h>
37 #include <linux/slab.h>
38 #include <linux/uaccess.h>
39
40 #include <net/addrconf.h>
41 #include <net/ndisc.h>
42 #include <net/protocol.h>
43 #include <net/transp_v6.h>
44 #include <net/ip6_route.h>
45 #include <net/raw.h>
46 #include <net/tcp_states.h>
47 #include <net/ip6_checksum.h>
48 #include <net/ip6_tunnel.h>
49 #include <net/xfrm.h>
50 #include <net/inet_hashtables.h>
51 #include <net/inet6_hashtables.h>
52 #include <net/busy_poll.h>
53 #include <net/sock_reuseport.h>
54
55 #include <linux/proc_fs.h>
56 #include <linux/seq_file.h>
57 #include <trace/events/skb.h>
58 #include "udp_impl.h"
59
60 static bool udp6_lib_exact_dif_match(struct net *net, struct sk_buff *skb)
61 {
62 #if defined(CONFIG_NET_L3_MASTER_DEV)
63 if (!net->ipv4.sysctl_udp_l3mdev_accept &&
64 skb && ipv6_l3mdev_skb(IP6CB(skb)->flags))
65 return true;
66 #endif
67 return false;
68 }
69
70 static u32 udp6_ehashfn(const struct net *net,
71 const struct in6_addr *laddr,
72 const u16 lport,
73 const struct in6_addr *faddr,
74 const __be16 fport)
75 {
76 static u32 udp6_ehash_secret __read_mostly;
77 static u32 udp_ipv6_hash_secret __read_mostly;
78
79 u32 lhash, fhash;
80
81 net_get_random_once(&udp6_ehash_secret,
82 sizeof(udp6_ehash_secret));
83 net_get_random_once(&udp_ipv6_hash_secret,
84 sizeof(udp_ipv6_hash_secret));
85
86 lhash = (__force u32)laddr->s6_addr32[3];
87 fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret);
88
89 return __inet6_ehashfn(lhash, lport, fhash, fport,
90 udp_ipv6_hash_secret + net_hash_mix(net));
91 }
92
93 int udp_v6_get_port(struct sock *sk, unsigned short snum)
94 {
95 unsigned int hash2_nulladdr =
96 ipv6_portaddr_hash(sock_net(sk), &in6addr_any, snum);
97 unsigned int hash2_partial =
98 ipv6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0);
99
100 /* precompute partial secondary hash */
101 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
102 return udp_lib_get_port(sk, snum, hash2_nulladdr);
103 }
104
105 void udp_v6_rehash(struct sock *sk)
106 {
107 u16 new_hash = ipv6_portaddr_hash(sock_net(sk),
108 &sk->sk_v6_rcv_saddr,
109 inet_sk(sk)->inet_num);
110
111 udp_lib_rehash(sk, new_hash);
112 }
113
114 static int compute_score(struct sock *sk, struct net *net,
115 const struct in6_addr *saddr, __be16 sport,
116 const struct in6_addr *daddr, unsigned short hnum,
117 int dif, int sdif, bool exact_dif)
118 {
119 int score;
120 struct inet_sock *inet;
121 bool dev_match;
122
123 if (!net_eq(sock_net(sk), net) ||
124 udp_sk(sk)->udp_port_hash != hnum ||
125 sk->sk_family != PF_INET6)
126 return -1;
127
128 if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
129 return -1;
130
131 score = 0;
132 inet = inet_sk(sk);
133
134 if (inet->inet_dport) {
135 if (inet->inet_dport != sport)
136 return -1;
137 score++;
138 }
139
140 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
141 if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr))
142 return -1;
143 score++;
144 }
145
146 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif);
147 if (!dev_match)
148 return -1;
149 score++;
150
151 if (sk->sk_incoming_cpu == raw_smp_processor_id())
152 score++;
153
154 return score;
155 }
156
157 /* called with rcu_read_lock() */
158 static struct sock *udp6_lib_lookup2(struct net *net,
159 const struct in6_addr *saddr, __be16 sport,
160 const struct in6_addr *daddr, unsigned int hnum,
161 int dif, int sdif, bool exact_dif,
162 struct udp_hslot *hslot2, struct sk_buff *skb)
163 {
164 struct sock *sk, *result;
165 int score, badness;
166 u32 hash = 0;
167
168 result = NULL;
169 badness = -1;
170 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
171 score = compute_score(sk, net, saddr, sport,
172 daddr, hnum, dif, sdif, exact_dif);
173 if (score > badness) {
174 if (sk->sk_reuseport) {
175 hash = udp6_ehashfn(net, daddr, hnum,
176 saddr, sport);
177
178 result = reuseport_select_sock(sk, hash, skb,
179 sizeof(struct udphdr));
180 if (result)
181 return result;
182 }
183 result = sk;
184 badness = score;
185 }
186 }
187 return result;
188 }
189
190 /* rcu_read_lock() must be held */
191 struct sock *__udp6_lib_lookup(struct net *net,
192 const struct in6_addr *saddr, __be16 sport,
193 const struct in6_addr *daddr, __be16 dport,
194 int dif, int sdif, struct udp_table *udptable,
195 struct sk_buff *skb)
196 {
197 unsigned short hnum = ntohs(dport);
198 unsigned int hash2, slot2;
199 struct udp_hslot *hslot2;
200 struct sock *result;
201 bool exact_dif = udp6_lib_exact_dif_match(net, skb);
202
203 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
204 slot2 = hash2 & udptable->mask;
205 hslot2 = &udptable->hash2[slot2];
206
207 result = udp6_lib_lookup2(net, saddr, sport,
208 daddr, hnum, dif, sdif, exact_dif,
209 hslot2, skb);
210 if (!result) {
211 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
212 slot2 = hash2 & udptable->mask;
213
214 hslot2 = &udptable->hash2[slot2];
215
216 result = udp6_lib_lookup2(net, saddr, sport,
217 &in6addr_any, hnum, dif, sdif,
218 exact_dif, hslot2,
219 skb);
220 }
221 if (unlikely(IS_ERR(result)))
222 return NULL;
223 return result;
224 }
225 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
226
227 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
228 __be16 sport, __be16 dport,
229 struct udp_table *udptable)
230 {
231 const struct ipv6hdr *iph = ipv6_hdr(skb);
232
233 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
234 &iph->daddr, dport, inet6_iif(skb),
235 inet6_sdif(skb), udptable, skb);
236 }
237
238 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb,
239 __be16 sport, __be16 dport)
240 {
241 const struct ipv6hdr *iph = ipv6_hdr(skb);
242
243 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
244 &iph->daddr, dport, inet6_iif(skb),
245 inet6_sdif(skb), &udp_table, skb);
246 }
247 EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb);
248
249 /* Must be called under rcu_read_lock().
250 * Does increment socket refcount.
251 */
252 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
253 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
254 const struct in6_addr *daddr, __be16 dport, int dif)
255 {
256 struct sock *sk;
257
258 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport,
259 dif, 0, &udp_table, NULL);
260 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
261 sk = NULL;
262 return sk;
263 }
264 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
265 #endif
266
267 /* do not use the scratch area len for jumbogram: their length execeeds the
268 * scratch area space; note that the IP6CB flags is still in the first
269 * cacheline, so checking for jumbograms is cheap
270 */
271 static int udp6_skb_len(struct sk_buff *skb)
272 {
273 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
274 }
275
276 /*
277 * This should be easy, if there is something there we
278 * return it, otherwise we block.
279 */
280
281 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
282 int noblock, int flags, int *addr_len)
283 {
284 struct ipv6_pinfo *np = inet6_sk(sk);
285 struct inet_sock *inet = inet_sk(sk);
286 struct sk_buff *skb;
287 unsigned int ulen, copied;
288 int peeked, peeking, off;
289 int err;
290 int is_udplite = IS_UDPLITE(sk);
291 bool checksum_valid = false;
292 struct udp_mib *mib;
293 int is_udp4;
294
295 if (flags & MSG_ERRQUEUE)
296 return ipv6_recv_error(sk, msg, len, addr_len);
297
298 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
299 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
300
301 try_again:
302 peeking = flags & MSG_PEEK;
303 off = sk_peek_offset(sk, flags);
304 skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
305 if (!skb)
306 return err;
307
308 ulen = udp6_skb_len(skb);
309 copied = len;
310 if (copied > ulen - off)
311 copied = ulen - off;
312 else if (copied < ulen)
313 msg->msg_flags |= MSG_TRUNC;
314
315 is_udp4 = (skb->protocol == htons(ETH_P_IP));
316 mib = __UDPX_MIB(sk, is_udp4);
317
318 /*
319 * If checksum is needed at all, try to do it while copying the
320 * data. If the data is truncated, or if we only want a partial
321 * coverage checksum (UDP-Lite), do it before the copy.
322 */
323
324 if (copied < ulen || peeking ||
325 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
326 checksum_valid = udp_skb_csum_unnecessary(skb) ||
327 !__udp_lib_checksum_complete(skb);
328 if (!checksum_valid)
329 goto csum_copy_err;
330 }
331
332 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
333 if (udp_skb_is_linear(skb))
334 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
335 else
336 err = skb_copy_datagram_msg(skb, off, msg, copied);
337 } else {
338 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
339 if (err == -EINVAL)
340 goto csum_copy_err;
341 }
342 if (unlikely(err)) {
343 if (!peeked) {
344 atomic_inc(&sk->sk_drops);
345 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
346 }
347 kfree_skb(skb);
348 return err;
349 }
350 if (!peeked)
351 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS);
352
353 sock_recv_ts_and_drops(msg, sk, skb);
354
355 /* Copy the address. */
356 if (msg->msg_name) {
357 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
358 sin6->sin6_family = AF_INET6;
359 sin6->sin6_port = udp_hdr(skb)->source;
360 sin6->sin6_flowinfo = 0;
361
362 if (is_udp4) {
363 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
364 &sin6->sin6_addr);
365 sin6->sin6_scope_id = 0;
366 } else {
367 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
368 sin6->sin6_scope_id =
369 ipv6_iface_scope_id(&sin6->sin6_addr,
370 inet6_iif(skb));
371 }
372 *addr_len = sizeof(*sin6);
373 }
374
375 if (udp_sk(sk)->gro_enabled)
376 udp_cmsg_recv(msg, sk, skb);
377
378 if (np->rxopt.all)
379 ip6_datagram_recv_common_ctl(sk, msg, skb);
380
381 if (is_udp4) {
382 if (inet->cmsg_flags)
383 ip_cmsg_recv_offset(msg, sk, skb,
384 sizeof(struct udphdr), off);
385 } else {
386 if (np->rxopt.all)
387 ip6_datagram_recv_specific_ctl(sk, msg, skb);
388 }
389
390 err = copied;
391 if (flags & MSG_TRUNC)
392 err = ulen;
393
394 skb_consume_udp(sk, skb, peeking ? -err : err);
395 return err;
396
397 csum_copy_err:
398 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
399 udp_skb_destructor)) {
400 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS);
401 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
402 }
403 kfree_skb(skb);
404
405 /* starting over for a new packet, but check if we need to yield */
406 cond_resched();
407 msg->msg_flags &= ~MSG_TRUNC;
408 goto try_again;
409 }
410
411 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
412 void udpv6_encap_enable(void)
413 {
414 static_branch_inc(&udpv6_encap_needed_key);
415 }
416 EXPORT_SYMBOL(udpv6_encap_enable);
417
418 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
419 * through error handlers in encapsulations looking for a match.
420 */
421 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb,
422 struct inet6_skb_parm *opt,
423 u8 type, u8 code, int offset, u32 info)
424 {
425 int i;
426
427 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
428 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt,
429 u8 type, u8 code, int offset, u32 info);
430
431 if (!ip6tun_encaps[i])
432 continue;
433 handler = rcu_dereference(ip6tun_encaps[i]->err_handler);
434 if (handler && !handler(skb, opt, type, code, offset, info))
435 return 0;
436 }
437
438 return -ENOENT;
439 }
440
441 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
442 * reversing source and destination port: this will match tunnels that force the
443 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
444 * lwtunnels might actually break this assumption by being configured with
445 * different destination ports on endpoints, in this case we won't be able to
446 * trace ICMP messages back to them.
447 *
448 * If this doesn't match any socket, probe tunnels with arbitrary destination
449 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
450 * we've sent packets to won't necessarily match the local destination port.
451 *
452 * Then ask the tunnel implementation to match the error against a valid
453 * association.
454 *
455 * Return an error if we can't find a match, the socket if we need further
456 * processing, zero otherwise.
457 */
458 static struct sock *__udp6_lib_err_encap(struct net *net,
459 const struct ipv6hdr *hdr, int offset,
460 struct udphdr *uh,
461 struct udp_table *udptable,
462 struct sk_buff *skb,
463 struct inet6_skb_parm *opt,
464 u8 type, u8 code, __be32 info)
465 {
466 int network_offset, transport_offset;
467 struct sock *sk;
468
469 network_offset = skb_network_offset(skb);
470 transport_offset = skb_transport_offset(skb);
471
472 /* Network header needs to point to the outer IPv6 header inside ICMP */
473 skb_reset_network_header(skb);
474
475 /* Transport header needs to point to the UDP header */
476 skb_set_transport_header(skb, offset);
477
478 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source,
479 &hdr->saddr, uh->dest,
480 inet6_iif(skb), 0, udptable, skb);
481 if (sk) {
482 int (*lookup)(struct sock *sk, struct sk_buff *skb);
483 struct udp_sock *up = udp_sk(sk);
484
485 lookup = READ_ONCE(up->encap_err_lookup);
486 if (!lookup || lookup(sk, skb))
487 sk = NULL;
488 }
489
490 if (!sk) {
491 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code,
492 offset, info));
493 }
494
495 skb_set_transport_header(skb, transport_offset);
496 skb_set_network_header(skb, network_offset);
497
498 return sk;
499 }
500
501 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
502 u8 type, u8 code, int offset, __be32 info,
503 struct udp_table *udptable)
504 {
505 struct ipv6_pinfo *np;
506 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
507 const struct in6_addr *saddr = &hdr->saddr;
508 const struct in6_addr *daddr = &hdr->daddr;
509 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
510 bool tunnel = false;
511 struct sock *sk;
512 int harderr;
513 int err;
514 struct net *net = dev_net(skb->dev);
515
516 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
517 inet6_iif(skb), inet6_sdif(skb), udptable, skb);
518 if (!sk) {
519 /* No socket for error: try tunnels before discarding */
520 sk = ERR_PTR(-ENOENT);
521 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
522 sk = __udp6_lib_err_encap(net, hdr, offset, uh,
523 udptable, skb,
524 opt, type, code, info);
525 if (!sk)
526 return 0;
527 }
528
529 if (IS_ERR(sk)) {
530 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
531 ICMP6_MIB_INERRORS);
532 return PTR_ERR(sk);
533 }
534
535 tunnel = true;
536 }
537
538 harderr = icmpv6_err_convert(type, code, &err);
539 np = inet6_sk(sk);
540
541 if (type == ICMPV6_PKT_TOOBIG) {
542 if (!ip6_sk_accept_pmtu(sk))
543 goto out;
544 ip6_sk_update_pmtu(skb, sk, info);
545 if (np->pmtudisc != IPV6_PMTUDISC_DONT)
546 harderr = 1;
547 }
548 if (type == NDISC_REDIRECT) {
549 if (tunnel) {
550 ip6_redirect(skb, sock_net(sk), inet6_iif(skb),
551 sk->sk_mark, sk->sk_uid);
552 } else {
553 ip6_sk_redirect(skb, sk);
554 }
555 goto out;
556 }
557
558 /* Tunnels don't have an application socket: don't pass errors back */
559 if (tunnel)
560 goto out;
561
562 if (!np->recverr) {
563 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
564 goto out;
565 } else {
566 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
567 }
568
569 sk->sk_err = err;
570 sk->sk_error_report(sk);
571 out:
572 return 0;
573 }
574
575 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
576 {
577 int rc;
578
579 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
580 sock_rps_save_rxhash(sk, skb);
581 sk_mark_napi_id(sk, skb);
582 sk_incoming_cpu_update(sk);
583 } else {
584 sk_mark_napi_id_once(sk, skb);
585 }
586
587 rc = __udp_enqueue_schedule_skb(sk, skb);
588 if (rc < 0) {
589 int is_udplite = IS_UDPLITE(sk);
590
591 /* Note that an ENOMEM error is charged twice */
592 if (rc == -ENOMEM)
593 UDP6_INC_STATS(sock_net(sk),
594 UDP_MIB_RCVBUFERRORS, is_udplite);
595 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
596 kfree_skb(skb);
597 return -1;
598 }
599
600 return 0;
601 }
602
603 static __inline__ int udpv6_err(struct sk_buff *skb,
604 struct inet6_skb_parm *opt, u8 type,
605 u8 code, int offset, __be32 info)
606 {
607 return __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
608 }
609
610 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
611 {
612 struct udp_sock *up = udp_sk(sk);
613 int is_udplite = IS_UDPLITE(sk);
614
615 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
616 goto drop;
617
618 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) {
619 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
620
621 /*
622 * This is an encapsulation socket so pass the skb to
623 * the socket's udp_encap_rcv() hook. Otherwise, just
624 * fall through and pass this up the UDP socket.
625 * up->encap_rcv() returns the following value:
626 * =0 if skb was successfully passed to the encap
627 * handler or was discarded by it.
628 * >0 if skb should be passed on to UDP.
629 * <0 if skb should be resubmitted as proto -N
630 */
631
632 /* if we're overly short, let UDP handle it */
633 encap_rcv = READ_ONCE(up->encap_rcv);
634 if (encap_rcv) {
635 int ret;
636
637 /* Verify checksum before giving to encap */
638 if (udp_lib_checksum_complete(skb))
639 goto csum_error;
640
641 ret = encap_rcv(sk, skb);
642 if (ret <= 0) {
643 __UDP_INC_STATS(sock_net(sk),
644 UDP_MIB_INDATAGRAMS,
645 is_udplite);
646 return -ret;
647 }
648 }
649
650 /* FALLTHROUGH -- it's a UDP Packet */
651 }
652
653 /*
654 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
655 */
656 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
657
658 if (up->pcrlen == 0) { /* full coverage was set */
659 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
660 UDP_SKB_CB(skb)->cscov, skb->len);
661 goto drop;
662 }
663 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
664 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
665 UDP_SKB_CB(skb)->cscov, up->pcrlen);
666 goto drop;
667 }
668 }
669
670 prefetch(&sk->sk_rmem_alloc);
671 if (rcu_access_pointer(sk->sk_filter) &&
672 udp_lib_checksum_complete(skb))
673 goto csum_error;
674
675 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
676 goto drop;
677
678 udp_csum_pull_header(skb);
679
680 skb_dst_drop(skb);
681
682 return __udpv6_queue_rcv_skb(sk, skb);
683
684 csum_error:
685 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
686 drop:
687 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
688 atomic_inc(&sk->sk_drops);
689 kfree_skb(skb);
690 return -1;
691 }
692
693 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
694 {
695 struct sk_buff *next, *segs;
696 int ret;
697
698 if (likely(!udp_unexpected_gso(sk, skb)))
699 return udpv6_queue_rcv_one_skb(sk, skb);
700
701 __skb_push(skb, -skb_mac_offset(skb));
702 segs = udp_rcv_segment(sk, skb, false);
703 for (skb = segs; skb; skb = next) {
704 next = skb->next;
705 __skb_pull(skb, skb_transport_offset(skb));
706
707 ret = udpv6_queue_rcv_one_skb(sk, skb);
708 if (ret > 0)
709 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret,
710 true);
711 }
712 return 0;
713 }
714
715 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
716 __be16 loc_port, const struct in6_addr *loc_addr,
717 __be16 rmt_port, const struct in6_addr *rmt_addr,
718 int dif, int sdif, unsigned short hnum)
719 {
720 struct inet_sock *inet = inet_sk(sk);
721
722 if (!net_eq(sock_net(sk), net))
723 return false;
724
725 if (udp_sk(sk)->udp_port_hash != hnum ||
726 sk->sk_family != PF_INET6 ||
727 (inet->inet_dport && inet->inet_dport != rmt_port) ||
728 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
729 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
730 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif) ||
731 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
732 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
733 return false;
734 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
735 return false;
736 return true;
737 }
738
739 static void udp6_csum_zero_error(struct sk_buff *skb)
740 {
741 /* RFC 2460 section 8.1 says that we SHOULD log
742 * this error. Well, it is reasonable.
743 */
744 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
745 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
746 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
747 }
748
749 /*
750 * Note: called only from the BH handler context,
751 * so we don't need to lock the hashes.
752 */
753 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
754 const struct in6_addr *saddr, const struct in6_addr *daddr,
755 struct udp_table *udptable, int proto)
756 {
757 struct sock *sk, *first = NULL;
758 const struct udphdr *uh = udp_hdr(skb);
759 unsigned short hnum = ntohs(uh->dest);
760 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
761 unsigned int offset = offsetof(typeof(*sk), sk_node);
762 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
763 int dif = inet6_iif(skb);
764 int sdif = inet6_sdif(skb);
765 struct hlist_node *node;
766 struct sk_buff *nskb;
767
768 if (use_hash2) {
769 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) &
770 udptable->mask;
771 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask;
772 start_lookup:
773 hslot = &udptable->hash2[hash2];
774 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
775 }
776
777 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
778 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
779 uh->source, saddr, dif, sdif,
780 hnum))
781 continue;
782 /* If zero checksum and no_check is not on for
783 * the socket then skip it.
784 */
785 if (!uh->check && !udp_sk(sk)->no_check6_rx)
786 continue;
787 if (!first) {
788 first = sk;
789 continue;
790 }
791 nskb = skb_clone(skb, GFP_ATOMIC);
792 if (unlikely(!nskb)) {
793 atomic_inc(&sk->sk_drops);
794 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
795 IS_UDPLITE(sk));
796 __UDP6_INC_STATS(net, UDP_MIB_INERRORS,
797 IS_UDPLITE(sk));
798 continue;
799 }
800
801 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
802 consume_skb(nskb);
803 }
804
805 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
806 if (use_hash2 && hash2 != hash2_any) {
807 hash2 = hash2_any;
808 goto start_lookup;
809 }
810
811 if (first) {
812 if (udpv6_queue_rcv_skb(first, skb) > 0)
813 consume_skb(skb);
814 } else {
815 kfree_skb(skb);
816 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
817 proto == IPPROTO_UDPLITE);
818 }
819 return 0;
820 }
821
822 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
823 {
824 if (udp_sk_rx_dst_set(sk, dst)) {
825 const struct rt6_info *rt = (const struct rt6_info *)dst;
826
827 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
828 }
829 }
830
831 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
832 * return code conversion for ip layer consumption
833 */
834 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
835 struct udphdr *uh)
836 {
837 int ret;
838
839 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
840 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
841 ip6_compute_pseudo);
842
843 ret = udpv6_queue_rcv_skb(sk, skb);
844
845 /* a return value > 0 means to resubmit the input */
846 if (ret > 0)
847 return ret;
848 return 0;
849 }
850
851 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
852 int proto)
853 {
854 const struct in6_addr *saddr, *daddr;
855 struct net *net = dev_net(skb->dev);
856 struct udphdr *uh;
857 struct sock *sk;
858 u32 ulen = 0;
859
860 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
861 goto discard;
862
863 saddr = &ipv6_hdr(skb)->saddr;
864 daddr = &ipv6_hdr(skb)->daddr;
865 uh = udp_hdr(skb);
866
867 ulen = ntohs(uh->len);
868 if (ulen > skb->len)
869 goto short_packet;
870
871 if (proto == IPPROTO_UDP) {
872 /* UDP validates ulen. */
873
874 /* Check for jumbo payload */
875 if (ulen == 0)
876 ulen = skb->len;
877
878 if (ulen < sizeof(*uh))
879 goto short_packet;
880
881 if (ulen < skb->len) {
882 if (pskb_trim_rcsum(skb, ulen))
883 goto short_packet;
884 saddr = &ipv6_hdr(skb)->saddr;
885 daddr = &ipv6_hdr(skb)->daddr;
886 uh = udp_hdr(skb);
887 }
888 }
889
890 if (udp6_csum_init(skb, uh, proto))
891 goto csum_error;
892
893 /* Check if the socket is already available, e.g. due to early demux */
894 sk = skb_steal_sock(skb);
895 if (sk) {
896 struct dst_entry *dst = skb_dst(skb);
897 int ret;
898
899 if (unlikely(sk->sk_rx_dst != dst))
900 udp6_sk_rx_dst_set(sk, dst);
901
902 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
903 sock_put(sk);
904 goto report_csum_error;
905 }
906
907 ret = udp6_unicast_rcv_skb(sk, skb, uh);
908 sock_put(sk);
909 return ret;
910 }
911
912 /*
913 * Multicast receive code
914 */
915 if (ipv6_addr_is_multicast(daddr))
916 return __udp6_lib_mcast_deliver(net, skb,
917 saddr, daddr, udptable, proto);
918
919 /* Unicast */
920 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
921 if (sk) {
922 if (!uh->check && !udp_sk(sk)->no_check6_rx)
923 goto report_csum_error;
924 return udp6_unicast_rcv_skb(sk, skb, uh);
925 }
926
927 if (!uh->check)
928 goto report_csum_error;
929
930 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
931 goto discard;
932
933 if (udp_lib_checksum_complete(skb))
934 goto csum_error;
935
936 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
937 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
938
939 kfree_skb(skb);
940 return 0;
941
942 short_packet:
943 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
944 proto == IPPROTO_UDPLITE ? "-Lite" : "",
945 saddr, ntohs(uh->source),
946 ulen, skb->len,
947 daddr, ntohs(uh->dest));
948 goto discard;
949
950 report_csum_error:
951 udp6_csum_zero_error(skb);
952 csum_error:
953 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
954 discard:
955 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
956 kfree_skb(skb);
957 return 0;
958 }
959
960
961 static struct sock *__udp6_lib_demux_lookup(struct net *net,
962 __be16 loc_port, const struct in6_addr *loc_addr,
963 __be16 rmt_port, const struct in6_addr *rmt_addr,
964 int dif, int sdif)
965 {
966 unsigned short hnum = ntohs(loc_port);
967 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum);
968 unsigned int slot2 = hash2 & udp_table.mask;
969 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
970 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
971 struct sock *sk;
972
973 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
974 if (sk->sk_state == TCP_ESTABLISHED &&
975 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif))
976 return sk;
977 /* Only check first socket in chain */
978 break;
979 }
980 return NULL;
981 }
982
983 static void udp_v6_early_demux(struct sk_buff *skb)
984 {
985 struct net *net = dev_net(skb->dev);
986 const struct udphdr *uh;
987 struct sock *sk;
988 struct dst_entry *dst;
989 int dif = skb->dev->ifindex;
990 int sdif = inet6_sdif(skb);
991
992 if (!pskb_may_pull(skb, skb_transport_offset(skb) +
993 sizeof(struct udphdr)))
994 return;
995
996 uh = udp_hdr(skb);
997
998 if (skb->pkt_type == PACKET_HOST)
999 sk = __udp6_lib_demux_lookup(net, uh->dest,
1000 &ipv6_hdr(skb)->daddr,
1001 uh->source, &ipv6_hdr(skb)->saddr,
1002 dif, sdif);
1003 else
1004 return;
1005
1006 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
1007 return;
1008
1009 skb->sk = sk;
1010 skb->destructor = sock_efree;
1011 dst = READ_ONCE(sk->sk_rx_dst);
1012
1013 if (dst)
1014 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1015 if (dst) {
1016 /* set noref for now.
1017 * any place which wants to hold dst has to call
1018 * dst_hold_safe()
1019 */
1020 skb_dst_set_noref(skb, dst);
1021 }
1022 }
1023
1024 static __inline__ int udpv6_rcv(struct sk_buff *skb)
1025 {
1026 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1027 }
1028
1029 /*
1030 * Throw away all pending data and cancel the corking. Socket is locked.
1031 */
1032 static void udp_v6_flush_pending_frames(struct sock *sk)
1033 {
1034 struct udp_sock *up = udp_sk(sk);
1035
1036 if (up->pending == AF_INET)
1037 udp_flush_pending_frames(sk);
1038 else if (up->pending) {
1039 up->len = 0;
1040 up->pending = 0;
1041 ip6_flush_pending_frames(sk);
1042 }
1043 }
1044
1045 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
1046 int addr_len)
1047 {
1048 /* The following checks are replicated from __ip6_datagram_connect()
1049 * and intended to prevent BPF program called below from accessing
1050 * bytes that are out of the bound specified by user in addr_len.
1051 */
1052 if (uaddr->sa_family == AF_INET) {
1053 if (__ipv6_only_sock(sk))
1054 return -EAFNOSUPPORT;
1055 return udp_pre_connect(sk, uaddr, addr_len);
1056 }
1057
1058 if (addr_len < SIN6_LEN_RFC2133)
1059 return -EINVAL;
1060
1061 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr);
1062 }
1063
1064 /**
1065 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
1066 * @sk: socket we are sending on
1067 * @skb: sk_buff containing the filled-in UDP header
1068 * (checksum field must be zeroed out)
1069 */
1070 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
1071 const struct in6_addr *saddr,
1072 const struct in6_addr *daddr, int len)
1073 {
1074 unsigned int offset;
1075 struct udphdr *uh = udp_hdr(skb);
1076 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
1077 __wsum csum = 0;
1078
1079 if (!frags) {
1080 /* Only one fragment on the socket. */
1081 skb->csum_start = skb_transport_header(skb) - skb->head;
1082 skb->csum_offset = offsetof(struct udphdr, check);
1083 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1084 } else {
1085 /*
1086 * HW-checksum won't work as there are two or more
1087 * fragments on the socket so that all csums of sk_buffs
1088 * should be together
1089 */
1090 offset = skb_transport_offset(skb);
1091 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1092 csum = skb->csum;
1093
1094 skb->ip_summed = CHECKSUM_NONE;
1095
1096 do {
1097 csum = csum_add(csum, frags->csum);
1098 } while ((frags = frags->next));
1099
1100 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1101 csum);
1102 if (uh->check == 0)
1103 uh->check = CSUM_MANGLED_0;
1104 }
1105 }
1106
1107 /*
1108 * Sending
1109 */
1110
1111 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6,
1112 struct inet_cork *cork)
1113 {
1114 struct sock *sk = skb->sk;
1115 struct udphdr *uh;
1116 int err = 0;
1117 int is_udplite = IS_UDPLITE(sk);
1118 __wsum csum = 0;
1119 int offset = skb_transport_offset(skb);
1120 int len = skb->len - offset;
1121
1122 /*
1123 * Create a UDP header
1124 */
1125 uh = udp_hdr(skb);
1126 uh->source = fl6->fl6_sport;
1127 uh->dest = fl6->fl6_dport;
1128 uh->len = htons(len);
1129 uh->check = 0;
1130
1131 if (cork->gso_size) {
1132 const int hlen = skb_network_header_len(skb) +
1133 sizeof(struct udphdr);
1134
1135 if (hlen + cork->gso_size > cork->fragsize) {
1136 kfree_skb(skb);
1137 return -EINVAL;
1138 }
1139 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
1140 kfree_skb(skb);
1141 return -EINVAL;
1142 }
1143 if (udp_sk(sk)->no_check6_tx) {
1144 kfree_skb(skb);
1145 return -EINVAL;
1146 }
1147 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
1148 dst_xfrm(skb_dst(skb))) {
1149 kfree_skb(skb);
1150 return -EIO;
1151 }
1152
1153 skb_shinfo(skb)->gso_size = cork->gso_size;
1154 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1155 goto csum_partial;
1156 }
1157
1158 if (is_udplite)
1159 csum = udplite_csum(skb);
1160 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
1161 skb->ip_summed = CHECKSUM_NONE;
1162 goto send;
1163 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1164 csum_partial:
1165 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1166 goto send;
1167 } else
1168 csum = udp_csum(skb);
1169
1170 /* add protocol-dependent pseudo-header */
1171 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1172 len, fl6->flowi6_proto, csum);
1173 if (uh->check == 0)
1174 uh->check = CSUM_MANGLED_0;
1175
1176 send:
1177 err = ip6_send_skb(skb);
1178 if (err) {
1179 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1180 UDP6_INC_STATS(sock_net(sk),
1181 UDP_MIB_SNDBUFERRORS, is_udplite);
1182 err = 0;
1183 }
1184 } else {
1185 UDP6_INC_STATS(sock_net(sk),
1186 UDP_MIB_OUTDATAGRAMS, is_udplite);
1187 }
1188 return err;
1189 }
1190
1191 static int udp_v6_push_pending_frames(struct sock *sk)
1192 {
1193 struct sk_buff *skb;
1194 struct udp_sock *up = udp_sk(sk);
1195 struct flowi6 fl6;
1196 int err = 0;
1197
1198 if (up->pending == AF_INET)
1199 return udp_push_pending_frames(sk);
1200
1201 /* ip6_finish_skb will release the cork, so make a copy of
1202 * fl6 here.
1203 */
1204 fl6 = inet_sk(sk)->cork.fl.u.ip6;
1205
1206 skb = ip6_finish_skb(sk);
1207 if (!skb)
1208 goto out;
1209
1210 err = udp_v6_send_skb(skb, &fl6, &inet_sk(sk)->cork.base);
1211
1212 out:
1213 up->len = 0;
1214 up->pending = 0;
1215 return err;
1216 }
1217
1218 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1219 {
1220 struct ipv6_txoptions opt_space;
1221 struct udp_sock *up = udp_sk(sk);
1222 struct inet_sock *inet = inet_sk(sk);
1223 struct ipv6_pinfo *np = inet6_sk(sk);
1224 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1225 struct in6_addr *daddr, *final_p, final;
1226 struct ipv6_txoptions *opt = NULL;
1227 struct ipv6_txoptions *opt_to_free = NULL;
1228 struct ip6_flowlabel *flowlabel = NULL;
1229 struct flowi6 fl6;
1230 struct dst_entry *dst;
1231 struct ipcm6_cookie ipc6;
1232 int addr_len = msg->msg_namelen;
1233 bool connected = false;
1234 int ulen = len;
1235 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1236 int err;
1237 int is_udplite = IS_UDPLITE(sk);
1238 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1239
1240 ipcm6_init(&ipc6);
1241 ipc6.gso_size = up->gso_size;
1242 ipc6.sockc.tsflags = sk->sk_tsflags;
1243
1244 /* destination address check */
1245 if (sin6) {
1246 if (addr_len < offsetof(struct sockaddr, sa_data))
1247 return -EINVAL;
1248
1249 switch (sin6->sin6_family) {
1250 case AF_INET6:
1251 if (addr_len < SIN6_LEN_RFC2133)
1252 return -EINVAL;
1253 daddr = &sin6->sin6_addr;
1254 if (ipv6_addr_any(daddr) &&
1255 ipv6_addr_v4mapped(&np->saddr))
1256 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1257 daddr);
1258 break;
1259 case AF_INET:
1260 goto do_udp_sendmsg;
1261 case AF_UNSPEC:
1262 msg->msg_name = sin6 = NULL;
1263 msg->msg_namelen = addr_len = 0;
1264 daddr = NULL;
1265 break;
1266 default:
1267 return -EINVAL;
1268 }
1269 } else if (!up->pending) {
1270 if (sk->sk_state != TCP_ESTABLISHED)
1271 return -EDESTADDRREQ;
1272 daddr = &sk->sk_v6_daddr;
1273 } else
1274 daddr = NULL;
1275
1276 if (daddr) {
1277 if (ipv6_addr_v4mapped(daddr)) {
1278 struct sockaddr_in sin;
1279 sin.sin_family = AF_INET;
1280 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1281 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1282 msg->msg_name = &sin;
1283 msg->msg_namelen = sizeof(sin);
1284 do_udp_sendmsg:
1285 if (__ipv6_only_sock(sk))
1286 return -ENETUNREACH;
1287 return udp_sendmsg(sk, msg, len);
1288 }
1289 }
1290
1291 if (up->pending == AF_INET)
1292 return udp_sendmsg(sk, msg, len);
1293
1294 /* Rough check on arithmetic overflow,
1295 better check is made in ip6_append_data().
1296 */
1297 if (len > INT_MAX - sizeof(struct udphdr))
1298 return -EMSGSIZE;
1299
1300 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1301 if (up->pending) {
1302 /*
1303 * There are pending frames.
1304 * The socket lock must be held while it's corked.
1305 */
1306 lock_sock(sk);
1307 if (likely(up->pending)) {
1308 if (unlikely(up->pending != AF_INET6)) {
1309 release_sock(sk);
1310 return -EAFNOSUPPORT;
1311 }
1312 dst = NULL;
1313 goto do_append_data;
1314 }
1315 release_sock(sk);
1316 }
1317 ulen += sizeof(struct udphdr);
1318
1319 memset(&fl6, 0, sizeof(fl6));
1320
1321 if (sin6) {
1322 if (sin6->sin6_port == 0)
1323 return -EINVAL;
1324
1325 fl6.fl6_dport = sin6->sin6_port;
1326 daddr = &sin6->sin6_addr;
1327
1328 if (np->sndflow) {
1329 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1330 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1331 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1332 if (!flowlabel)
1333 return -EINVAL;
1334 }
1335 }
1336
1337 /*
1338 * Otherwise it will be difficult to maintain
1339 * sk->sk_dst_cache.
1340 */
1341 if (sk->sk_state == TCP_ESTABLISHED &&
1342 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1343 daddr = &sk->sk_v6_daddr;
1344
1345 if (addr_len >= sizeof(struct sockaddr_in6) &&
1346 sin6->sin6_scope_id &&
1347 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1348 fl6.flowi6_oif = sin6->sin6_scope_id;
1349 } else {
1350 if (sk->sk_state != TCP_ESTABLISHED)
1351 return -EDESTADDRREQ;
1352
1353 fl6.fl6_dport = inet->inet_dport;
1354 daddr = &sk->sk_v6_daddr;
1355 fl6.flowlabel = np->flow_label;
1356 connected = true;
1357 }
1358
1359 if (!fl6.flowi6_oif)
1360 fl6.flowi6_oif = sk->sk_bound_dev_if;
1361
1362 if (!fl6.flowi6_oif)
1363 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1364
1365 fl6.flowi6_mark = sk->sk_mark;
1366 fl6.flowi6_uid = sk->sk_uid;
1367
1368 if (msg->msg_controllen) {
1369 opt = &opt_space;
1370 memset(opt, 0, sizeof(struct ipv6_txoptions));
1371 opt->tot_len = sizeof(*opt);
1372 ipc6.opt = opt;
1373
1374 err = udp_cmsg_send(sk, msg, &ipc6.gso_size);
1375 if (err > 0)
1376 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6,
1377 &ipc6);
1378 if (err < 0) {
1379 fl6_sock_release(flowlabel);
1380 return err;
1381 }
1382 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1383 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1384 if (!flowlabel)
1385 return -EINVAL;
1386 }
1387 if (!(opt->opt_nflen|opt->opt_flen))
1388 opt = NULL;
1389 connected = false;
1390 }
1391 if (!opt) {
1392 opt = txopt_get(np);
1393 opt_to_free = opt;
1394 }
1395 if (flowlabel)
1396 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1397 opt = ipv6_fixup_options(&opt_space, opt);
1398 ipc6.opt = opt;
1399
1400 fl6.flowi6_proto = sk->sk_protocol;
1401 fl6.daddr = *daddr;
1402 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1403 fl6.saddr = np->saddr;
1404 fl6.fl6_sport = inet->inet_sport;
1405
1406 if (cgroup_bpf_enabled && !connected) {
1407 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk,
1408 (struct sockaddr *)sin6, &fl6.saddr);
1409 if (err)
1410 goto out_no_dst;
1411 if (sin6) {
1412 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
1413 /* BPF program rewrote IPv6-only by IPv4-mapped
1414 * IPv6. It's currently unsupported.
1415 */
1416 err = -ENOTSUPP;
1417 goto out_no_dst;
1418 }
1419 if (sin6->sin6_port == 0) {
1420 /* BPF program set invalid port. Reject it. */
1421 err = -EINVAL;
1422 goto out_no_dst;
1423 }
1424 fl6.fl6_dport = sin6->sin6_port;
1425 fl6.daddr = sin6->sin6_addr;
1426 }
1427 }
1428
1429 if (ipv6_addr_any(&fl6.daddr))
1430 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1431
1432 final_p = fl6_update_dst(&fl6, opt, &final);
1433 if (final_p)
1434 connected = false;
1435
1436 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1437 fl6.flowi6_oif = np->mcast_oif;
1438 connected = false;
1439 } else if (!fl6.flowi6_oif)
1440 fl6.flowi6_oif = np->ucast_oif;
1441
1442 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
1443
1444 if (ipc6.tclass < 0)
1445 ipc6.tclass = np->tclass;
1446
1447 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel);
1448
1449 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, connected);
1450 if (IS_ERR(dst)) {
1451 err = PTR_ERR(dst);
1452 dst = NULL;
1453 goto out;
1454 }
1455
1456 if (ipc6.hlimit < 0)
1457 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1458
1459 if (msg->msg_flags&MSG_CONFIRM)
1460 goto do_confirm;
1461 back_from_confirm:
1462
1463 /* Lockless fast path for the non-corking case */
1464 if (!corkreq) {
1465 struct inet_cork_full cork;
1466 struct sk_buff *skb;
1467
1468 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1469 sizeof(struct udphdr), &ipc6,
1470 &fl6, (struct rt6_info *)dst,
1471 msg->msg_flags, &cork);
1472 err = PTR_ERR(skb);
1473 if (!IS_ERR_OR_NULL(skb))
1474 err = udp_v6_send_skb(skb, &fl6, &cork.base);
1475 goto out;
1476 }
1477
1478 lock_sock(sk);
1479 if (unlikely(up->pending)) {
1480 /* The socket is already corked while preparing it. */
1481 /* ... which is an evident application bug. --ANK */
1482 release_sock(sk);
1483
1484 net_dbg_ratelimited("udp cork app bug 2\n");
1485 err = -EINVAL;
1486 goto out;
1487 }
1488
1489 up->pending = AF_INET6;
1490
1491 do_append_data:
1492 if (ipc6.dontfrag < 0)
1493 ipc6.dontfrag = np->dontfrag;
1494 up->len += ulen;
1495 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1496 &ipc6, &fl6, (struct rt6_info *)dst,
1497 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1498 if (err)
1499 udp_v6_flush_pending_frames(sk);
1500 else if (!corkreq)
1501 err = udp_v6_push_pending_frames(sk);
1502 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1503 up->pending = 0;
1504
1505 if (err > 0)
1506 err = np->recverr ? net_xmit_errno(err) : 0;
1507 release_sock(sk);
1508
1509 out:
1510 dst_release(dst);
1511 out_no_dst:
1512 fl6_sock_release(flowlabel);
1513 txopt_put(opt_to_free);
1514 if (!err)
1515 return len;
1516 /*
1517 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1518 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1519 * we don't have a good statistic (IpOutDiscards but it can be too many
1520 * things). We could add another new stat but at least for now that
1521 * seems like overkill.
1522 */
1523 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1524 UDP6_INC_STATS(sock_net(sk),
1525 UDP_MIB_SNDBUFERRORS, is_udplite);
1526 }
1527 return err;
1528
1529 do_confirm:
1530 if (msg->msg_flags & MSG_PROBE)
1531 dst_confirm_neigh(dst, &fl6.daddr);
1532 if (!(msg->msg_flags&MSG_PROBE) || len)
1533 goto back_from_confirm;
1534 err = 0;
1535 goto out;
1536 }
1537
1538 void udpv6_destroy_sock(struct sock *sk)
1539 {
1540 struct udp_sock *up = udp_sk(sk);
1541 lock_sock(sk);
1542 udp_v6_flush_pending_frames(sk);
1543 release_sock(sk);
1544
1545 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
1546 if (up->encap_type) {
1547 void (*encap_destroy)(struct sock *sk);
1548 encap_destroy = READ_ONCE(up->encap_destroy);
1549 if (encap_destroy)
1550 encap_destroy(sk);
1551 }
1552 if (up->encap_enabled)
1553 static_branch_dec(&udpv6_encap_needed_key);
1554 }
1555
1556 inet6_destroy_sock(sk);
1557 }
1558
1559 /*
1560 * Socket option code for UDP
1561 */
1562 int udpv6_setsockopt(struct sock *sk, int level, int optname,
1563 char __user *optval, unsigned int optlen)
1564 {
1565 if (level == SOL_UDP || level == SOL_UDPLITE)
1566 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1567 udp_v6_push_pending_frames);
1568 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1569 }
1570
1571 #ifdef CONFIG_COMPAT
1572 int compat_udpv6_setsockopt(struct sock *sk, int level, int optname,
1573 char __user *optval, unsigned int optlen)
1574 {
1575 if (level == SOL_UDP || level == SOL_UDPLITE)
1576 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1577 udp_v6_push_pending_frames);
1578 return compat_ipv6_setsockopt(sk, level, optname, optval, optlen);
1579 }
1580 #endif
1581
1582 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1583 char __user *optval, int __user *optlen)
1584 {
1585 if (level == SOL_UDP || level == SOL_UDPLITE)
1586 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1587 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1588 }
1589
1590 #ifdef CONFIG_COMPAT
1591 int compat_udpv6_getsockopt(struct sock *sk, int level, int optname,
1592 char __user *optval, int __user *optlen)
1593 {
1594 if (level == SOL_UDP || level == SOL_UDPLITE)
1595 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1596 return compat_ipv6_getsockopt(sk, level, optname, optval, optlen);
1597 }
1598 #endif
1599
1600 /* thinking of making this const? Don't.
1601 * early_demux can change based on sysctl.
1602 */
1603 static struct inet6_protocol udpv6_protocol = {
1604 .early_demux = udp_v6_early_demux,
1605 .early_demux_handler = udp_v6_early_demux,
1606 .handler = udpv6_rcv,
1607 .err_handler = udpv6_err,
1608 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1609 };
1610
1611 /* ------------------------------------------------------------------------ */
1612 #ifdef CONFIG_PROC_FS
1613 int udp6_seq_show(struct seq_file *seq, void *v)
1614 {
1615 if (v == SEQ_START_TOKEN) {
1616 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1617 } else {
1618 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1619 struct inet_sock *inet = inet_sk(v);
1620 __u16 srcp = ntohs(inet->inet_sport);
1621 __u16 destp = ntohs(inet->inet_dport);
1622 __ip6_dgram_sock_seq_show(seq, v, srcp, destp,
1623 udp_rqueue_get(v), bucket);
1624 }
1625 return 0;
1626 }
1627
1628 const struct seq_operations udp6_seq_ops = {
1629 .start = udp_seq_start,
1630 .next = udp_seq_next,
1631 .stop = udp_seq_stop,
1632 .show = udp6_seq_show,
1633 };
1634 EXPORT_SYMBOL(udp6_seq_ops);
1635
1636 static struct udp_seq_afinfo udp6_seq_afinfo = {
1637 .family = AF_INET6,
1638 .udp_table = &udp_table,
1639 };
1640
1641 int __net_init udp6_proc_init(struct net *net)
1642 {
1643 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops,
1644 sizeof(struct udp_iter_state), &udp6_seq_afinfo))
1645 return -ENOMEM;
1646 return 0;
1647 }
1648
1649 void udp6_proc_exit(struct net *net)
1650 {
1651 remove_proc_entry("udp6", net->proc_net);
1652 }
1653 #endif /* CONFIG_PROC_FS */
1654
1655 /* ------------------------------------------------------------------------ */
1656
1657 struct proto udpv6_prot = {
1658 .name = "UDPv6",
1659 .owner = THIS_MODULE,
1660 .close = udp_lib_close,
1661 .pre_connect = udpv6_pre_connect,
1662 .connect = ip6_datagram_connect,
1663 .disconnect = udp_disconnect,
1664 .ioctl = udp_ioctl,
1665 .init = udp_init_sock,
1666 .destroy = udpv6_destroy_sock,
1667 .setsockopt = udpv6_setsockopt,
1668 .getsockopt = udpv6_getsockopt,
1669 .sendmsg = udpv6_sendmsg,
1670 .recvmsg = udpv6_recvmsg,
1671 .release_cb = ip6_datagram_release_cb,
1672 .hash = udp_lib_hash,
1673 .unhash = udp_lib_unhash,
1674 .rehash = udp_v6_rehash,
1675 .get_port = udp_v6_get_port,
1676 .memory_allocated = &udp_memory_allocated,
1677 .sysctl_mem = sysctl_udp_mem,
1678 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
1679 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
1680 .obj_size = sizeof(struct udp6_sock),
1681 .h.udp_table = &udp_table,
1682 #ifdef CONFIG_COMPAT
1683 .compat_setsockopt = compat_udpv6_setsockopt,
1684 .compat_getsockopt = compat_udpv6_getsockopt,
1685 #endif
1686 .diag_destroy = udp_abort,
1687 };
1688
1689 static struct inet_protosw udpv6_protosw = {
1690 .type = SOCK_DGRAM,
1691 .protocol = IPPROTO_UDP,
1692 .prot = &udpv6_prot,
1693 .ops = &inet6_dgram_ops,
1694 .flags = INET_PROTOSW_PERMANENT,
1695 };
1696
1697 int __init udpv6_init(void)
1698 {
1699 int ret;
1700
1701 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1702 if (ret)
1703 goto out;
1704
1705 ret = inet6_register_protosw(&udpv6_protosw);
1706 if (ret)
1707 goto out_udpv6_protocol;
1708 out:
1709 return ret;
1710
1711 out_udpv6_protocol:
1712 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1713 goto out;
1714 }
1715
1716 void udpv6_exit(void)
1717 {
1718 inet6_unregister_protosw(&udpv6_protosw);
1719 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1720 }
Ubuntu Focal Linux kernel mirror