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
udp6_ehashfn(const struct net * net,const struct in6_addr * laddr,const u16 lport,const struct in6_addr * faddr,const __be16 fport)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
udp_v6_get_port(struct sock * sk,unsigned short snum)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
udp_v6_rehash(struct sock * sk)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
compute_score(struct sock * sk,struct net * net,const struct in6_addr * saddr,__be16 sport,const struct in6_addr * daddr,unsigned short hnum,int dif,int sdif)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
lookup_reuseport(struct net * net,struct sock * sk,struct sk_buff * skb,const struct in6_addr * saddr,__be16 sport,const struct in6_addr * daddr,unsigned int hnum)144 static struct sock *lookup_reuseport(struct net *net, struct sock *sk,
145 struct sk_buff *skb,
146 const struct in6_addr *saddr,
147 __be16 sport,
148 const struct in6_addr *daddr,
149 unsigned int hnum)
150 {
151 struct sock *reuse_sk = NULL;
152 u32 hash;
153
154 if (sk->sk_reuseport && sk->sk_state != TCP_ESTABLISHED) {
155 hash = udp6_ehashfn(net, daddr, hnum, saddr, sport);
156 reuse_sk = reuseport_select_sock(sk, hash, skb,
157 sizeof(struct udphdr));
158 }
159 return reuse_sk;
160 }
161
162 /* called with rcu_read_lock() */
udp6_lib_lookup2(struct net * net,const struct in6_addr * saddr,__be16 sport,const struct in6_addr * daddr,unsigned int hnum,int dif,int sdif,struct udp_hslot * hslot2,struct sk_buff * skb)163 static struct sock *udp6_lib_lookup2(struct net *net,
164 const struct in6_addr *saddr, __be16 sport,
165 const struct in6_addr *daddr, unsigned int hnum,
166 int dif, int sdif, struct udp_hslot *hslot2,
167 struct sk_buff *skb)
168 {
169 struct sock *sk, *result;
170 int score, badness;
171
172 result = NULL;
173 badness = -1;
174 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
175 score = compute_score(sk, net, saddr, sport,
176 daddr, hnum, dif, sdif);
177 if (score > badness) {
178 result = lookup_reuseport(net, sk, skb,
179 saddr, sport, daddr, hnum);
180 /* Fall back to scoring if group has connections */
181 if (result && !reuseport_has_conns(sk, false))
182 return result;
183
184 result = result ? : sk;
185 badness = score;
186 }
187 }
188 return result;
189 }
190
udp6_lookup_run_bpf(struct net * net,struct udp_table * udptable,struct sk_buff * skb,const struct in6_addr * saddr,__be16 sport,const struct in6_addr * daddr,u16 hnum)191 static inline struct sock *udp6_lookup_run_bpf(struct net *net,
192 struct udp_table *udptable,
193 struct sk_buff *skb,
194 const struct in6_addr *saddr,
195 __be16 sport,
196 const struct in6_addr *daddr,
197 u16 hnum)
198 {
199 struct sock *sk, *reuse_sk;
200 bool no_reuseport;
201
202 if (udptable != &udp_table)
203 return NULL; /* only UDP is supported */
204
205 no_reuseport = bpf_sk_lookup_run_v6(net, IPPROTO_UDP,
206 saddr, sport, daddr, hnum, &sk);
207 if (no_reuseport || IS_ERR_OR_NULL(sk))
208 return sk;
209
210 reuse_sk = lookup_reuseport(net, sk, skb, saddr, sport, daddr, hnum);
211 if (reuse_sk)
212 sk = reuse_sk;
213 return sk;
214 }
215
216 /* rcu_read_lock() must be held */
__udp6_lib_lookup(struct net * net,const struct in6_addr * saddr,__be16 sport,const struct in6_addr * daddr,__be16 dport,int dif,int sdif,struct udp_table * udptable,struct sk_buff * skb)217 struct sock *__udp6_lib_lookup(struct net *net,
218 const struct in6_addr *saddr, __be16 sport,
219 const struct in6_addr *daddr, __be16 dport,
220 int dif, int sdif, struct udp_table *udptable,
221 struct sk_buff *skb)
222 {
223 unsigned short hnum = ntohs(dport);
224 unsigned int hash2, slot2;
225 struct udp_hslot *hslot2;
226 struct sock *result, *sk;
227
228 hash2 = ipv6_portaddr_hash(net, daddr, hnum);
229 slot2 = hash2 & udptable->mask;
230 hslot2 = &udptable->hash2[slot2];
231
232 /* Lookup connected or non-wildcard sockets */
233 result = udp6_lib_lookup2(net, saddr, sport,
234 daddr, hnum, dif, sdif,
235 hslot2, skb);
236 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
237 goto done;
238
239 /* Lookup redirect from BPF */
240 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
241 sk = udp6_lookup_run_bpf(net, udptable, skb,
242 saddr, sport, daddr, hnum);
243 if (sk) {
244 result = sk;
245 goto done;
246 }
247 }
248
249 /* Got non-wildcard socket or error on first lookup */
250 if (result)
251 goto done;
252
253 /* Lookup wildcard sockets */
254 hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
255 slot2 = hash2 & udptable->mask;
256 hslot2 = &udptable->hash2[slot2];
257
258 result = udp6_lib_lookup2(net, saddr, sport,
259 &in6addr_any, hnum, dif, sdif,
260 hslot2, skb);
261 done:
262 if (IS_ERR(result))
263 return NULL;
264 return result;
265 }
266 EXPORT_SYMBOL_GPL(__udp6_lib_lookup);
267
__udp6_lib_lookup_skb(struct sk_buff * skb,__be16 sport,__be16 dport,struct udp_table * udptable)268 static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb,
269 __be16 sport, __be16 dport,
270 struct udp_table *udptable)
271 {
272 const struct ipv6hdr *iph = ipv6_hdr(skb);
273
274 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
275 &iph->daddr, dport, inet6_iif(skb),
276 inet6_sdif(skb), udptable, skb);
277 }
278
udp6_lib_lookup_skb(const struct sk_buff * skb,__be16 sport,__be16 dport)279 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb,
280 __be16 sport, __be16 dport)
281 {
282 const struct ipv6hdr *iph = ipv6_hdr(skb);
283
284 return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport,
285 &iph->daddr, dport, inet6_iif(skb),
286 inet6_sdif(skb), &udp_table, NULL);
287 }
288
289 /* Must be called under rcu_read_lock().
290 * Does increment socket refcount.
291 */
292 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV6) || IS_ENABLED(CONFIG_NF_SOCKET_IPV6)
udp6_lib_lookup(struct net * net,const struct in6_addr * saddr,__be16 sport,const struct in6_addr * daddr,__be16 dport,int dif)293 struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport,
294 const struct in6_addr *daddr, __be16 dport, int dif)
295 {
296 struct sock *sk;
297
298 sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport,
299 dif, 0, &udp_table, NULL);
300 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
301 sk = NULL;
302 return sk;
303 }
304 EXPORT_SYMBOL_GPL(udp6_lib_lookup);
305 #endif
306
307 /* do not use the scratch area len for jumbogram: their length execeeds the
308 * scratch area space; note that the IP6CB flags is still in the first
309 * cacheline, so checking for jumbograms is cheap
310 */
udp6_skb_len(struct sk_buff * skb)311 static int udp6_skb_len(struct sk_buff *skb)
312 {
313 return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb);
314 }
315
316 /*
317 * This should be easy, if there is something there we
318 * return it, otherwise we block.
319 */
320
udpv6_recvmsg(struct sock * sk,struct msghdr * msg,size_t len,int noblock,int flags,int * addr_len)321 int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
322 int noblock, int flags, int *addr_len)
323 {
324 struct ipv6_pinfo *np = inet6_sk(sk);
325 struct inet_sock *inet = inet_sk(sk);
326 struct sk_buff *skb;
327 unsigned int ulen, copied;
328 int off, err, peeking = flags & MSG_PEEK;
329 int is_udplite = IS_UDPLITE(sk);
330 struct udp_mib __percpu *mib;
331 bool checksum_valid = false;
332 int is_udp4;
333
334 if (flags & MSG_ERRQUEUE)
335 return ipv6_recv_error(sk, msg, len, addr_len);
336
337 if (np->rxpmtu && np->rxopt.bits.rxpmtu)
338 return ipv6_recv_rxpmtu(sk, msg, len, addr_len);
339
340 try_again:
341 off = sk_peek_offset(sk, flags);
342 skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
343 if (!skb)
344 return err;
345
346 ulen = udp6_skb_len(skb);
347 copied = len;
348 if (copied > ulen - off)
349 copied = ulen - off;
350 else if (copied < ulen)
351 msg->msg_flags |= MSG_TRUNC;
352
353 is_udp4 = (skb->protocol == htons(ETH_P_IP));
354 mib = __UDPX_MIB(sk, is_udp4);
355
356 /*
357 * If checksum is needed at all, try to do it while copying the
358 * data. If the data is truncated, or if we only want a partial
359 * coverage checksum (UDP-Lite), do it before the copy.
360 */
361
362 if (copied < ulen || peeking ||
363 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
364 checksum_valid = udp_skb_csum_unnecessary(skb) ||
365 !__udp_lib_checksum_complete(skb);
366 if (!checksum_valid)
367 goto csum_copy_err;
368 }
369
370 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
371 if (udp_skb_is_linear(skb))
372 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
373 else
374 err = skb_copy_datagram_msg(skb, off, msg, copied);
375 } else {
376 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
377 if (err == -EINVAL)
378 goto csum_copy_err;
379 }
380 if (unlikely(err)) {
381 if (!peeking) {
382 atomic_inc(&sk->sk_drops);
383 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
384 }
385 kfree_skb(skb);
386 return err;
387 }
388 if (!peeking)
389 SNMP_INC_STATS(mib, UDP_MIB_INDATAGRAMS);
390
391 sock_recv_ts_and_drops(msg, sk, skb);
392
393 /* Copy the address. */
394 if (msg->msg_name) {
395 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
396 sin6->sin6_family = AF_INET6;
397 sin6->sin6_port = udp_hdr(skb)->source;
398 sin6->sin6_flowinfo = 0;
399
400 if (is_udp4) {
401 ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr,
402 &sin6->sin6_addr);
403 sin6->sin6_scope_id = 0;
404 } else {
405 sin6->sin6_addr = ipv6_hdr(skb)->saddr;
406 sin6->sin6_scope_id =
407 ipv6_iface_scope_id(&sin6->sin6_addr,
408 inet6_iif(skb));
409 }
410 *addr_len = sizeof(*sin6);
411
412 BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk,
413 (struct sockaddr *)sin6);
414 }
415
416 if (udp_sk(sk)->gro_enabled)
417 udp_cmsg_recv(msg, sk, skb);
418
419 if (np->rxopt.all)
420 ip6_datagram_recv_common_ctl(sk, msg, skb);
421
422 if (is_udp4) {
423 if (inet->cmsg_flags)
424 ip_cmsg_recv_offset(msg, sk, skb,
425 sizeof(struct udphdr), off);
426 } else {
427 if (np->rxopt.all)
428 ip6_datagram_recv_specific_ctl(sk, msg, skb);
429 }
430
431 err = copied;
432 if (flags & MSG_TRUNC)
433 err = ulen;
434
435 skb_consume_udp(sk, skb, peeking ? -err : err);
436 return err;
437
438 csum_copy_err:
439 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
440 udp_skb_destructor)) {
441 SNMP_INC_STATS(mib, UDP_MIB_CSUMERRORS);
442 SNMP_INC_STATS(mib, UDP_MIB_INERRORS);
443 }
444 kfree_skb(skb);
445
446 /* starting over for a new packet, but check if we need to yield */
447 cond_resched();
448 msg->msg_flags &= ~MSG_TRUNC;
449 goto try_again;
450 }
451
452 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
udpv6_encap_enable(void)453 void udpv6_encap_enable(void)
454 {
455 static_branch_inc(&udpv6_encap_needed_key);
456 }
457 EXPORT_SYMBOL(udpv6_encap_enable);
458
459 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
460 * through error handlers in encapsulations looking for a match.
461 */
__udp6_lib_err_encap_no_sk(struct sk_buff * skb,struct inet6_skb_parm * opt,u8 type,u8 code,int offset,__be32 info)462 static int __udp6_lib_err_encap_no_sk(struct sk_buff *skb,
463 struct inet6_skb_parm *opt,
464 u8 type, u8 code, int offset, __be32 info)
465 {
466 int i;
467
468 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
469 int (*handler)(struct sk_buff *skb, struct inet6_skb_parm *opt,
470 u8 type, u8 code, int offset, __be32 info);
471 const struct ip6_tnl_encap_ops *encap;
472
473 encap = rcu_dereference(ip6tun_encaps[i]);
474 if (!encap)
475 continue;
476 handler = encap->err_handler;
477 if (handler && !handler(skb, opt, type, code, offset, info))
478 return 0;
479 }
480
481 return -ENOENT;
482 }
483
484 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
485 * reversing source and destination port: this will match tunnels that force the
486 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
487 * lwtunnels might actually break this assumption by being configured with
488 * different destination ports on endpoints, in this case we won't be able to
489 * trace ICMP messages back to them.
490 *
491 * If this doesn't match any socket, probe tunnels with arbitrary destination
492 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
493 * we've sent packets to won't necessarily match the local destination port.
494 *
495 * Then ask the tunnel implementation to match the error against a valid
496 * association.
497 *
498 * Return an error if we can't find a match, the socket if we need further
499 * processing, zero otherwise.
500 */
__udp6_lib_err_encap(struct net * net,const struct ipv6hdr * hdr,int offset,struct udphdr * uh,struct udp_table * udptable,struct sk_buff * skb,struct inet6_skb_parm * opt,u8 type,u8 code,__be32 info)501 static struct sock *__udp6_lib_err_encap(struct net *net,
502 const struct ipv6hdr *hdr, int offset,
503 struct udphdr *uh,
504 struct udp_table *udptable,
505 struct sk_buff *skb,
506 struct inet6_skb_parm *opt,
507 u8 type, u8 code, __be32 info)
508 {
509 int network_offset, transport_offset;
510 struct sock *sk;
511
512 network_offset = skb_network_offset(skb);
513 transport_offset = skb_transport_offset(skb);
514
515 /* Network header needs to point to the outer IPv6 header inside ICMP */
516 skb_reset_network_header(skb);
517
518 /* Transport header needs to point to the UDP header */
519 skb_set_transport_header(skb, offset);
520
521 sk = __udp6_lib_lookup(net, &hdr->daddr, uh->source,
522 &hdr->saddr, uh->dest,
523 inet6_iif(skb), 0, udptable, skb);
524 if (sk) {
525 int (*lookup)(struct sock *sk, struct sk_buff *skb);
526 struct udp_sock *up = udp_sk(sk);
527
528 lookup = READ_ONCE(up->encap_err_lookup);
529 if (!lookup || lookup(sk, skb))
530 sk = NULL;
531 }
532
533 if (!sk) {
534 sk = ERR_PTR(__udp6_lib_err_encap_no_sk(skb, opt, type, code,
535 offset, info));
536 }
537
538 skb_set_transport_header(skb, transport_offset);
539 skb_set_network_header(skb, network_offset);
540
541 return sk;
542 }
543
__udp6_lib_err(struct sk_buff * skb,struct inet6_skb_parm * opt,u8 type,u8 code,int offset,__be32 info,struct udp_table * udptable)544 int __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
545 u8 type, u8 code, int offset, __be32 info,
546 struct udp_table *udptable)
547 {
548 struct ipv6_pinfo *np;
549 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
550 const struct in6_addr *saddr = &hdr->saddr;
551 const struct in6_addr *daddr = &hdr->daddr;
552 struct udphdr *uh = (struct udphdr *)(skb->data+offset);
553 bool tunnel = false;
554 struct sock *sk;
555 int harderr;
556 int err;
557 struct net *net = dev_net(skb->dev);
558
559 sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source,
560 inet6_iif(skb), inet6_sdif(skb), udptable, NULL);
561 if (!sk || udp_sk(sk)->encap_type) {
562 /* No socket for error: try tunnels before discarding */
563 sk = ERR_PTR(-ENOENT);
564 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
565 sk = __udp6_lib_err_encap(net, hdr, offset, uh,
566 udptable, skb,
567 opt, type, code, info);
568 if (!sk)
569 return 0;
570 }
571
572 if (IS_ERR(sk)) {
573 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
574 ICMP6_MIB_INERRORS);
575 return PTR_ERR(sk);
576 }
577
578 tunnel = true;
579 }
580
581 harderr = icmpv6_err_convert(type, code, &err);
582 np = inet6_sk(sk);
583
584 if (type == ICMPV6_PKT_TOOBIG) {
585 if (!ip6_sk_accept_pmtu(sk))
586 goto out;
587 ip6_sk_update_pmtu(skb, sk, info);
588 if (np->pmtudisc != IPV6_PMTUDISC_DONT)
589 harderr = 1;
590 }
591 if (type == NDISC_REDIRECT) {
592 if (tunnel) {
593 ip6_redirect(skb, sock_net(sk), inet6_iif(skb),
594 sk->sk_mark, sk->sk_uid);
595 } else {
596 ip6_sk_redirect(skb, sk);
597 }
598 goto out;
599 }
600
601 /* Tunnels don't have an application socket: don't pass errors back */
602 if (tunnel)
603 goto out;
604
605 if (!np->recverr) {
606 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
607 goto out;
608 } else {
609 ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1));
610 }
611
612 sk->sk_err = err;
613 sk->sk_error_report(sk);
614 out:
615 return 0;
616 }
617
__udpv6_queue_rcv_skb(struct sock * sk,struct sk_buff * skb)618 static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
619 {
620 int rc;
621
622 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
623 sock_rps_save_rxhash(sk, skb);
624 sk_mark_napi_id(sk, skb);
625 sk_incoming_cpu_update(sk);
626 } else {
627 sk_mark_napi_id_once(sk, skb);
628 }
629
630 rc = __udp_enqueue_schedule_skb(sk, skb);
631 if (rc < 0) {
632 int is_udplite = IS_UDPLITE(sk);
633
634 /* Note that an ENOMEM error is charged twice */
635 if (rc == -ENOMEM)
636 UDP6_INC_STATS(sock_net(sk),
637 UDP_MIB_RCVBUFERRORS, is_udplite);
638 else
639 UDP6_INC_STATS(sock_net(sk),
640 UDP_MIB_MEMERRORS, is_udplite);
641 UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
642 kfree_skb(skb);
643 return -1;
644 }
645
646 return 0;
647 }
648
udpv6_err(struct sk_buff * skb,struct inet6_skb_parm * opt,u8 type,u8 code,int offset,__be32 info)649 static __inline__ int udpv6_err(struct sk_buff *skb,
650 struct inet6_skb_parm *opt, u8 type,
651 u8 code, int offset, __be32 info)
652 {
653 return __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table);
654 }
655
udpv6_queue_rcv_one_skb(struct sock * sk,struct sk_buff * skb)656 static int udpv6_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
657 {
658 struct udp_sock *up = udp_sk(sk);
659 int is_udplite = IS_UDPLITE(sk);
660
661 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
662 goto drop;
663
664 if (static_branch_unlikely(&udpv6_encap_needed_key) && up->encap_type) {
665 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
666
667 /*
668 * This is an encapsulation socket so pass the skb to
669 * the socket's udp_encap_rcv() hook. Otherwise, just
670 * fall through and pass this up the UDP socket.
671 * up->encap_rcv() returns the following value:
672 * =0 if skb was successfully passed to the encap
673 * handler or was discarded by it.
674 * >0 if skb should be passed on to UDP.
675 * <0 if skb should be resubmitted as proto -N
676 */
677
678 /* if we're overly short, let UDP handle it */
679 encap_rcv = READ_ONCE(up->encap_rcv);
680 if (encap_rcv) {
681 int ret;
682
683 /* Verify checksum before giving to encap */
684 if (udp_lib_checksum_complete(skb))
685 goto csum_error;
686
687 ret = encap_rcv(sk, skb);
688 if (ret <= 0) {
689 __UDP_INC_STATS(sock_net(sk),
690 UDP_MIB_INDATAGRAMS,
691 is_udplite);
692 return -ret;
693 }
694 }
695
696 /* FALLTHROUGH -- it's a UDP Packet */
697 }
698
699 /*
700 * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c).
701 */
702 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
703
704 if (up->pcrlen == 0) { /* full coverage was set */
705 net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n",
706 UDP_SKB_CB(skb)->cscov, skb->len);
707 goto drop;
708 }
709 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
710 net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n",
711 UDP_SKB_CB(skb)->cscov, up->pcrlen);
712 goto drop;
713 }
714 }
715
716 prefetch(&sk->sk_rmem_alloc);
717 if (rcu_access_pointer(sk->sk_filter) &&
718 udp_lib_checksum_complete(skb))
719 goto csum_error;
720
721 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
722 goto drop;
723
724 udp_csum_pull_header(skb);
725
726 skb_dst_drop(skb);
727
728 return __udpv6_queue_rcv_skb(sk, skb);
729
730 csum_error:
731 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
732 drop:
733 __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
734 atomic_inc(&sk->sk_drops);
735 kfree_skb(skb);
736 return -1;
737 }
738
udpv6_queue_rcv_skb(struct sock * sk,struct sk_buff * skb)739 static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
740 {
741 struct sk_buff *next, *segs;
742 int ret;
743
744 if (likely(!udp_unexpected_gso(sk, skb)))
745 return udpv6_queue_rcv_one_skb(sk, skb);
746
747 __skb_push(skb, -skb_mac_offset(skb));
748 segs = udp_rcv_segment(sk, skb, false);
749 skb_list_walk_safe(segs, skb, next) {
750 __skb_pull(skb, skb_transport_offset(skb));
751
752 udp_post_segment_fix_csum(skb);
753 ret = udpv6_queue_rcv_one_skb(sk, skb);
754 if (ret > 0)
755 ip6_protocol_deliver_rcu(dev_net(skb->dev), skb, ret,
756 true);
757 }
758 return 0;
759 }
760
__udp_v6_is_mcast_sock(struct net * net,struct sock * sk,__be16 loc_port,const struct in6_addr * loc_addr,__be16 rmt_port,const struct in6_addr * rmt_addr,int dif,int sdif,unsigned short hnum)761 static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk,
762 __be16 loc_port, const struct in6_addr *loc_addr,
763 __be16 rmt_port, const struct in6_addr *rmt_addr,
764 int dif, int sdif, unsigned short hnum)
765 {
766 struct inet_sock *inet = inet_sk(sk);
767
768 if (!net_eq(sock_net(sk), net))
769 return false;
770
771 if (udp_sk(sk)->udp_port_hash != hnum ||
772 sk->sk_family != PF_INET6 ||
773 (inet->inet_dport && inet->inet_dport != rmt_port) ||
774 (!ipv6_addr_any(&sk->sk_v6_daddr) &&
775 !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) ||
776 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif) ||
777 (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) &&
778 !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr)))
779 return false;
780 if (!inet6_mc_check(sk, loc_addr, rmt_addr))
781 return false;
782 return true;
783 }
784
udp6_csum_zero_error(struct sk_buff * skb)785 static void udp6_csum_zero_error(struct sk_buff *skb)
786 {
787 /* RFC 2460 section 8.1 says that we SHOULD log
788 * this error. Well, it is reasonable.
789 */
790 net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n",
791 &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source),
792 &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest));
793 }
794
795 /*
796 * Note: called only from the BH handler context,
797 * so we don't need to lock the hashes.
798 */
__udp6_lib_mcast_deliver(struct net * net,struct sk_buff * skb,const struct in6_addr * saddr,const struct in6_addr * daddr,struct udp_table * udptable,int proto)799 static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
800 const struct in6_addr *saddr, const struct in6_addr *daddr,
801 struct udp_table *udptable, int proto)
802 {
803 struct sock *sk, *first = NULL;
804 const struct udphdr *uh = udp_hdr(skb);
805 unsigned short hnum = ntohs(uh->dest);
806 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
807 unsigned int offset = offsetof(typeof(*sk), sk_node);
808 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
809 int dif = inet6_iif(skb);
810 int sdif = inet6_sdif(skb);
811 struct hlist_node *node;
812 struct sk_buff *nskb;
813
814 if (use_hash2) {
815 hash2_any = ipv6_portaddr_hash(net, &in6addr_any, hnum) &
816 udptable->mask;
817 hash2 = ipv6_portaddr_hash(net, daddr, hnum) & udptable->mask;
818 start_lookup:
819 hslot = &udptable->hash2[hash2];
820 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
821 }
822
823 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
824 if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr,
825 uh->source, saddr, dif, sdif,
826 hnum))
827 continue;
828 /* If zero checksum and no_check is not on for
829 * the socket then skip it.
830 */
831 if (!uh->check && !udp_sk(sk)->no_check6_rx)
832 continue;
833 if (!first) {
834 first = sk;
835 continue;
836 }
837 nskb = skb_clone(skb, GFP_ATOMIC);
838 if (unlikely(!nskb)) {
839 atomic_inc(&sk->sk_drops);
840 __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
841 IS_UDPLITE(sk));
842 __UDP6_INC_STATS(net, UDP_MIB_INERRORS,
843 IS_UDPLITE(sk));
844 continue;
845 }
846
847 if (udpv6_queue_rcv_skb(sk, nskb) > 0)
848 consume_skb(nskb);
849 }
850
851 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
852 if (use_hash2 && hash2 != hash2_any) {
853 hash2 = hash2_any;
854 goto start_lookup;
855 }
856
857 if (first) {
858 if (udpv6_queue_rcv_skb(first, skb) > 0)
859 consume_skb(skb);
860 } else {
861 kfree_skb(skb);
862 __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
863 proto == IPPROTO_UDPLITE);
864 }
865 return 0;
866 }
867
udp6_sk_rx_dst_set(struct sock * sk,struct dst_entry * dst)868 static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
869 {
870 if (udp_sk_rx_dst_set(sk, dst)) {
871 const struct rt6_info *rt = (const struct rt6_info *)dst;
872
873 inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
874 }
875 }
876
877 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
878 * return code conversion for ip layer consumption
879 */
udp6_unicast_rcv_skb(struct sock * sk,struct sk_buff * skb,struct udphdr * uh)880 static int udp6_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
881 struct udphdr *uh)
882 {
883 int ret;
884
885 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
886 skb_checksum_try_convert(skb, IPPROTO_UDP, ip6_compute_pseudo);
887
888 ret = udpv6_queue_rcv_skb(sk, skb);
889
890 /* a return value > 0 means to resubmit the input */
891 if (ret > 0)
892 return ret;
893 return 0;
894 }
895
__udp6_lib_rcv(struct sk_buff * skb,struct udp_table * udptable,int proto)896 int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
897 int proto)
898 {
899 const struct in6_addr *saddr, *daddr;
900 struct net *net = dev_net(skb->dev);
901 struct udphdr *uh;
902 struct sock *sk;
903 bool refcounted;
904 u32 ulen = 0;
905
906 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
907 goto discard;
908
909 saddr = &ipv6_hdr(skb)->saddr;
910 daddr = &ipv6_hdr(skb)->daddr;
911 uh = udp_hdr(skb);
912
913 ulen = ntohs(uh->len);
914 if (ulen > skb->len)
915 goto short_packet;
916
917 if (proto == IPPROTO_UDP) {
918 /* UDP validates ulen. */
919
920 /* Check for jumbo payload */
921 if (ulen == 0)
922 ulen = skb->len;
923
924 if (ulen < sizeof(*uh))
925 goto short_packet;
926
927 if (ulen < skb->len) {
928 if (pskb_trim_rcsum(skb, ulen))
929 goto short_packet;
930 saddr = &ipv6_hdr(skb)->saddr;
931 daddr = &ipv6_hdr(skb)->daddr;
932 uh = udp_hdr(skb);
933 }
934 }
935
936 if (udp6_csum_init(skb, uh, proto))
937 goto csum_error;
938
939 /* Check if the socket is already available, e.g. due to early demux */
940 sk = skb_steal_sock(skb, &refcounted);
941 if (sk) {
942 struct dst_entry *dst = skb_dst(skb);
943 int ret;
944
945 if (unlikely(sk->sk_rx_dst != dst))
946 udp6_sk_rx_dst_set(sk, dst);
947
948 if (!uh->check && !udp_sk(sk)->no_check6_rx) {
949 if (refcounted)
950 sock_put(sk);
951 goto report_csum_error;
952 }
953
954 ret = udp6_unicast_rcv_skb(sk, skb, uh);
955 if (refcounted)
956 sock_put(sk);
957 return ret;
958 }
959
960 /*
961 * Multicast receive code
962 */
963 if (ipv6_addr_is_multicast(daddr))
964 return __udp6_lib_mcast_deliver(net, skb,
965 saddr, daddr, udptable, proto);
966
967 /* Unicast */
968 sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
969 if (sk) {
970 if (!uh->check && !udp_sk(sk)->no_check6_rx)
971 goto report_csum_error;
972 return udp6_unicast_rcv_skb(sk, skb, uh);
973 }
974
975 if (!uh->check)
976 goto report_csum_error;
977
978 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
979 goto discard;
980
981 if (udp_lib_checksum_complete(skb))
982 goto csum_error;
983
984 __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
985 icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
986
987 kfree_skb(skb);
988 return 0;
989
990 short_packet:
991 net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n",
992 proto == IPPROTO_UDPLITE ? "-Lite" : "",
993 saddr, ntohs(uh->source),
994 ulen, skb->len,
995 daddr, ntohs(uh->dest));
996 goto discard;
997
998 report_csum_error:
999 udp6_csum_zero_error(skb);
1000 csum_error:
1001 __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1002 discard:
1003 __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1004 kfree_skb(skb);
1005 return 0;
1006 }
1007
1008
__udp6_lib_demux_lookup(struct net * net,__be16 loc_port,const struct in6_addr * loc_addr,__be16 rmt_port,const struct in6_addr * rmt_addr,int dif,int sdif)1009 static struct sock *__udp6_lib_demux_lookup(struct net *net,
1010 __be16 loc_port, const struct in6_addr *loc_addr,
1011 __be16 rmt_port, const struct in6_addr *rmt_addr,
1012 int dif, int sdif)
1013 {
1014 unsigned short hnum = ntohs(loc_port);
1015 unsigned int hash2 = ipv6_portaddr_hash(net, loc_addr, hnum);
1016 unsigned int slot2 = hash2 & udp_table.mask;
1017 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
1018 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
1019 struct sock *sk;
1020
1021 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
1022 if (sk->sk_state == TCP_ESTABLISHED &&
1023 INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif, sdif))
1024 return sk;
1025 /* Only check first socket in chain */
1026 break;
1027 }
1028 return NULL;
1029 }
1030
udp_v6_early_demux(struct sk_buff * skb)1031 INDIRECT_CALLABLE_SCOPE void udp_v6_early_demux(struct sk_buff *skb)
1032 {
1033 struct net *net = dev_net(skb->dev);
1034 const struct udphdr *uh;
1035 struct sock *sk;
1036 struct dst_entry *dst;
1037 int dif = skb->dev->ifindex;
1038 int sdif = inet6_sdif(skb);
1039
1040 if (!pskb_may_pull(skb, skb_transport_offset(skb) +
1041 sizeof(struct udphdr)))
1042 return;
1043
1044 uh = udp_hdr(skb);
1045
1046 if (skb->pkt_type == PACKET_HOST)
1047 sk = __udp6_lib_demux_lookup(net, uh->dest,
1048 &ipv6_hdr(skb)->daddr,
1049 uh->source, &ipv6_hdr(skb)->saddr,
1050 dif, sdif);
1051 else
1052 return;
1053
1054 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
1055 return;
1056
1057 skb->sk = sk;
1058 skb->destructor = sock_efree;
1059 dst = READ_ONCE(sk->sk_rx_dst);
1060
1061 if (dst)
1062 dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie);
1063 if (dst) {
1064 /* set noref for now.
1065 * any place which wants to hold dst has to call
1066 * dst_hold_safe()
1067 */
1068 skb_dst_set_noref(skb, dst);
1069 }
1070 }
1071
udpv6_rcv(struct sk_buff * skb)1072 INDIRECT_CALLABLE_SCOPE int udpv6_rcv(struct sk_buff *skb)
1073 {
1074 return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1075 }
1076
1077 /*
1078 * Throw away all pending data and cancel the corking. Socket is locked.
1079 */
udp_v6_flush_pending_frames(struct sock * sk)1080 static void udp_v6_flush_pending_frames(struct sock *sk)
1081 {
1082 struct udp_sock *up = udp_sk(sk);
1083
1084 if (up->pending == AF_INET)
1085 udp_flush_pending_frames(sk);
1086 else if (up->pending) {
1087 up->len = 0;
1088 up->pending = 0;
1089 ip6_flush_pending_frames(sk);
1090 }
1091 }
1092
udpv6_pre_connect(struct sock * sk,struct sockaddr * uaddr,int addr_len)1093 static int udpv6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
1094 int addr_len)
1095 {
1096 if (addr_len < offsetofend(struct sockaddr, sa_family))
1097 return -EINVAL;
1098 /* The following checks are replicated from __ip6_datagram_connect()
1099 * and intended to prevent BPF program called below from accessing
1100 * bytes that are out of the bound specified by user in addr_len.
1101 */
1102 if (uaddr->sa_family == AF_INET) {
1103 if (__ipv6_only_sock(sk))
1104 return -EAFNOSUPPORT;
1105 return udp_pre_connect(sk, uaddr, addr_len);
1106 }
1107
1108 if (addr_len < SIN6_LEN_RFC2133)
1109 return -EINVAL;
1110
1111 return BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr);
1112 }
1113
1114 /**
1115 * udp6_hwcsum_outgoing - handle outgoing HW checksumming
1116 * @sk: socket we are sending on
1117 * @skb: sk_buff containing the filled-in UDP header
1118 * (checksum field must be zeroed out)
1119 * @saddr: source address
1120 * @daddr: destination address
1121 * @len: length of packet
1122 */
udp6_hwcsum_outgoing(struct sock * sk,struct sk_buff * skb,const struct in6_addr * saddr,const struct in6_addr * daddr,int len)1123 static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
1124 const struct in6_addr *saddr,
1125 const struct in6_addr *daddr, int len)
1126 {
1127 unsigned int offset;
1128 struct udphdr *uh = udp_hdr(skb);
1129 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
1130 __wsum csum = 0;
1131
1132 if (!frags) {
1133 /* Only one fragment on the socket. */
1134 skb->csum_start = skb_transport_header(skb) - skb->head;
1135 skb->csum_offset = offsetof(struct udphdr, check);
1136 uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0);
1137 } else {
1138 /*
1139 * HW-checksum won't work as there are two or more
1140 * fragments on the socket so that all csums of sk_buffs
1141 * should be together
1142 */
1143 offset = skb_transport_offset(skb);
1144 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
1145 csum = skb->csum;
1146
1147 skb->ip_summed = CHECKSUM_NONE;
1148
1149 do {
1150 csum = csum_add(csum, frags->csum);
1151 } while ((frags = frags->next));
1152
1153 uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP,
1154 csum);
1155 if (uh->check == 0)
1156 uh->check = CSUM_MANGLED_0;
1157 }
1158 }
1159
1160 /*
1161 * Sending
1162 */
1163
udp_v6_send_skb(struct sk_buff * skb,struct flowi6 * fl6,struct inet_cork * cork)1164 static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6,
1165 struct inet_cork *cork)
1166 {
1167 struct sock *sk = skb->sk;
1168 struct udphdr *uh;
1169 int err = 0;
1170 int is_udplite = IS_UDPLITE(sk);
1171 __wsum csum = 0;
1172 int offset = skb_transport_offset(skb);
1173 int len = skb->len - offset;
1174 int datalen = len - sizeof(*uh);
1175
1176 /*
1177 * Create a UDP header
1178 */
1179 uh = udp_hdr(skb);
1180 uh->source = fl6->fl6_sport;
1181 uh->dest = fl6->fl6_dport;
1182 uh->len = htons(len);
1183 uh->check = 0;
1184
1185 if (cork->gso_size) {
1186 const int hlen = skb_network_header_len(skb) +
1187 sizeof(struct udphdr);
1188
1189 if (hlen + cork->gso_size > cork->fragsize) {
1190 kfree_skb(skb);
1191 return -EINVAL;
1192 }
1193 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
1194 kfree_skb(skb);
1195 return -EINVAL;
1196 }
1197 if (udp_sk(sk)->no_check6_tx) {
1198 kfree_skb(skb);
1199 return -EINVAL;
1200 }
1201 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
1202 dst_xfrm(skb_dst(skb))) {
1203 kfree_skb(skb);
1204 return -EIO;
1205 }
1206
1207 if (datalen > cork->gso_size) {
1208 skb_shinfo(skb)->gso_size = cork->gso_size;
1209 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
1210 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
1211 cork->gso_size);
1212 }
1213 goto csum_partial;
1214 }
1215
1216 if (is_udplite)
1217 csum = udplite_csum(skb);
1218 else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */
1219 skb->ip_summed = CHECKSUM_NONE;
1220 goto send;
1221 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
1222 csum_partial:
1223 udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len);
1224 goto send;
1225 } else
1226 csum = udp_csum(skb);
1227
1228 /* add protocol-dependent pseudo-header */
1229 uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr,
1230 len, fl6->flowi6_proto, csum);
1231 if (uh->check == 0)
1232 uh->check = CSUM_MANGLED_0;
1233
1234 send:
1235 err = ip6_send_skb(skb);
1236 if (err) {
1237 if (err == -ENOBUFS && !inet6_sk(sk)->recverr) {
1238 UDP6_INC_STATS(sock_net(sk),
1239 UDP_MIB_SNDBUFERRORS, is_udplite);
1240 err = 0;
1241 }
1242 } else {
1243 UDP6_INC_STATS(sock_net(sk),
1244 UDP_MIB_OUTDATAGRAMS, is_udplite);
1245 }
1246 return err;
1247 }
1248
udp_v6_push_pending_frames(struct sock * sk)1249 static int udp_v6_push_pending_frames(struct sock *sk)
1250 {
1251 struct sk_buff *skb;
1252 struct udp_sock *up = udp_sk(sk);
1253 struct flowi6 fl6;
1254 int err = 0;
1255
1256 if (up->pending == AF_INET)
1257 return udp_push_pending_frames(sk);
1258
1259 /* ip6_finish_skb will release the cork, so make a copy of
1260 * fl6 here.
1261 */
1262 fl6 = inet_sk(sk)->cork.fl.u.ip6;
1263
1264 skb = ip6_finish_skb(sk);
1265 if (!skb)
1266 goto out;
1267
1268 err = udp_v6_send_skb(skb, &fl6, &inet_sk(sk)->cork.base);
1269
1270 out:
1271 up->len = 0;
1272 up->pending = 0;
1273 return err;
1274 }
1275
udpv6_sendmsg(struct sock * sk,struct msghdr * msg,size_t len)1276 int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1277 {
1278 struct ipv6_txoptions opt_space;
1279 struct udp_sock *up = udp_sk(sk);
1280 struct inet_sock *inet = inet_sk(sk);
1281 struct ipv6_pinfo *np = inet6_sk(sk);
1282 DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name);
1283 struct in6_addr *daddr, *final_p, final;
1284 struct ipv6_txoptions *opt = NULL;
1285 struct ipv6_txoptions *opt_to_free = NULL;
1286 struct ip6_flowlabel *flowlabel = NULL;
1287 struct flowi6 fl6;
1288 struct dst_entry *dst;
1289 struct ipcm6_cookie ipc6;
1290 int addr_len = msg->msg_namelen;
1291 bool connected = false;
1292 int ulen = len;
1293 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
1294 int err;
1295 int is_udplite = IS_UDPLITE(sk);
1296 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1297
1298 ipcm6_init(&ipc6);
1299 ipc6.gso_size = up->gso_size;
1300 ipc6.sockc.tsflags = sk->sk_tsflags;
1301 ipc6.sockc.mark = sk->sk_mark;
1302
1303 /* destination address check */
1304 if (sin6) {
1305 if (addr_len < offsetof(struct sockaddr, sa_data))
1306 return -EINVAL;
1307
1308 switch (sin6->sin6_family) {
1309 case AF_INET6:
1310 if (addr_len < SIN6_LEN_RFC2133)
1311 return -EINVAL;
1312 daddr = &sin6->sin6_addr;
1313 if (ipv6_addr_any(daddr) &&
1314 ipv6_addr_v4mapped(&np->saddr))
1315 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
1316 daddr);
1317 break;
1318 case AF_INET:
1319 goto do_udp_sendmsg;
1320 case AF_UNSPEC:
1321 msg->msg_name = sin6 = NULL;
1322 msg->msg_namelen = addr_len = 0;
1323 daddr = NULL;
1324 break;
1325 default:
1326 return -EINVAL;
1327 }
1328 } else if (!up->pending) {
1329 if (sk->sk_state != TCP_ESTABLISHED)
1330 return -EDESTADDRREQ;
1331 daddr = &sk->sk_v6_daddr;
1332 } else
1333 daddr = NULL;
1334
1335 if (daddr) {
1336 if (ipv6_addr_v4mapped(daddr)) {
1337 struct sockaddr_in sin;
1338 sin.sin_family = AF_INET;
1339 sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport;
1340 sin.sin_addr.s_addr = daddr->s6_addr32[3];
1341 msg->msg_name = &sin;
1342 msg->msg_namelen = sizeof(sin);
1343 do_udp_sendmsg:
1344 if (__ipv6_only_sock(sk))
1345 return -ENETUNREACH;
1346 return udp_sendmsg(sk, msg, len);
1347 }
1348 }
1349
1350 if (up->pending == AF_INET)
1351 return udp_sendmsg(sk, msg, len);
1352
1353 /* Rough check on arithmetic overflow,
1354 better check is made in ip6_append_data().
1355 */
1356 if (len > INT_MAX - sizeof(struct udphdr))
1357 return -EMSGSIZE;
1358
1359 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1360 if (up->pending) {
1361 /*
1362 * There are pending frames.
1363 * The socket lock must be held while it's corked.
1364 */
1365 lock_sock(sk);
1366 if (likely(up->pending)) {
1367 if (unlikely(up->pending != AF_INET6)) {
1368 release_sock(sk);
1369 return -EAFNOSUPPORT;
1370 }
1371 dst = NULL;
1372 goto do_append_data;
1373 }
1374 release_sock(sk);
1375 }
1376 ulen += sizeof(struct udphdr);
1377
1378 memset(&fl6, 0, sizeof(fl6));
1379
1380 if (sin6) {
1381 if (sin6->sin6_port == 0)
1382 return -EINVAL;
1383
1384 fl6.fl6_dport = sin6->sin6_port;
1385 daddr = &sin6->sin6_addr;
1386
1387 if (np->sndflow) {
1388 fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK;
1389 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
1390 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1391 if (IS_ERR(flowlabel))
1392 return -EINVAL;
1393 }
1394 }
1395
1396 /*
1397 * Otherwise it will be difficult to maintain
1398 * sk->sk_dst_cache.
1399 */
1400 if (sk->sk_state == TCP_ESTABLISHED &&
1401 ipv6_addr_equal(daddr, &sk->sk_v6_daddr))
1402 daddr = &sk->sk_v6_daddr;
1403
1404 if (addr_len >= sizeof(struct sockaddr_in6) &&
1405 sin6->sin6_scope_id &&
1406 __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr)))
1407 fl6.flowi6_oif = sin6->sin6_scope_id;
1408 } else {
1409 if (sk->sk_state != TCP_ESTABLISHED)
1410 return -EDESTADDRREQ;
1411
1412 fl6.fl6_dport = inet->inet_dport;
1413 daddr = &sk->sk_v6_daddr;
1414 fl6.flowlabel = np->flow_label;
1415 connected = true;
1416 }
1417
1418 if (!fl6.flowi6_oif)
1419 fl6.flowi6_oif = sk->sk_bound_dev_if;
1420
1421 if (!fl6.flowi6_oif)
1422 fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex;
1423
1424 fl6.flowi6_mark = ipc6.sockc.mark;
1425 fl6.flowi6_uid = sk->sk_uid;
1426
1427 if (msg->msg_controllen) {
1428 opt = &opt_space;
1429 memset(opt, 0, sizeof(struct ipv6_txoptions));
1430 opt->tot_len = sizeof(*opt);
1431 ipc6.opt = opt;
1432
1433 err = udp_cmsg_send(sk, msg, &ipc6.gso_size);
1434 if (err > 0)
1435 err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6,
1436 &ipc6);
1437 if (err < 0) {
1438 fl6_sock_release(flowlabel);
1439 return err;
1440 }
1441 if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) {
1442 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
1443 if (IS_ERR(flowlabel))
1444 return -EINVAL;
1445 }
1446 if (!(opt->opt_nflen|opt->opt_flen))
1447 opt = NULL;
1448 connected = false;
1449 }
1450 if (!opt) {
1451 opt = txopt_get(np);
1452 opt_to_free = opt;
1453 }
1454 if (flowlabel)
1455 opt = fl6_merge_options(&opt_space, flowlabel, opt);
1456 opt = ipv6_fixup_options(&opt_space, opt);
1457 ipc6.opt = opt;
1458
1459 fl6.flowi6_proto = sk->sk_protocol;
1460 fl6.daddr = *daddr;
1461 if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr))
1462 fl6.saddr = np->saddr;
1463 fl6.fl6_sport = inet->inet_sport;
1464
1465 if (cgroup_bpf_enabled(BPF_CGROUP_UDP6_SENDMSG) && !connected) {
1466 err = BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk,
1467 (struct sockaddr *)sin6, &fl6.saddr);
1468 if (err)
1469 goto out_no_dst;
1470 if (sin6) {
1471 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
1472 /* BPF program rewrote IPv6-only by IPv4-mapped
1473 * IPv6. It's currently unsupported.
1474 */
1475 err = -ENOTSUPP;
1476 goto out_no_dst;
1477 }
1478 if (sin6->sin6_port == 0) {
1479 /* BPF program set invalid port. Reject it. */
1480 err = -EINVAL;
1481 goto out_no_dst;
1482 }
1483 fl6.fl6_dport = sin6->sin6_port;
1484 fl6.daddr = sin6->sin6_addr;
1485 }
1486 }
1487
1488 if (ipv6_addr_any(&fl6.daddr))
1489 fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */
1490
1491 final_p = fl6_update_dst(&fl6, opt, &final);
1492 if (final_p)
1493 connected = false;
1494
1495 if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) {
1496 fl6.flowi6_oif = np->mcast_oif;
1497 connected = false;
1498 } else if (!fl6.flowi6_oif)
1499 fl6.flowi6_oif = np->ucast_oif;
1500
1501 security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6));
1502
1503 if (ipc6.tclass < 0)
1504 ipc6.tclass = np->tclass;
1505
1506 fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel);
1507
1508 dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, connected);
1509 if (IS_ERR(dst)) {
1510 err = PTR_ERR(dst);
1511 dst = NULL;
1512 goto out;
1513 }
1514
1515 if (ipc6.hlimit < 0)
1516 ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst);
1517
1518 if (msg->msg_flags&MSG_CONFIRM)
1519 goto do_confirm;
1520 back_from_confirm:
1521
1522 /* Lockless fast path for the non-corking case */
1523 if (!corkreq) {
1524 struct inet_cork_full cork;
1525 struct sk_buff *skb;
1526
1527 skb = ip6_make_skb(sk, getfrag, msg, ulen,
1528 sizeof(struct udphdr), &ipc6,
1529 &fl6, (struct rt6_info *)dst,
1530 msg->msg_flags, &cork);
1531 err = PTR_ERR(skb);
1532 if (!IS_ERR_OR_NULL(skb))
1533 err = udp_v6_send_skb(skb, &fl6, &cork.base);
1534 goto out;
1535 }
1536
1537 lock_sock(sk);
1538 if (unlikely(up->pending)) {
1539 /* The socket is already corked while preparing it. */
1540 /* ... which is an evident application bug. --ANK */
1541 release_sock(sk);
1542
1543 net_dbg_ratelimited("udp cork app bug 2\n");
1544 err = -EINVAL;
1545 goto out;
1546 }
1547
1548 up->pending = AF_INET6;
1549
1550 do_append_data:
1551 if (ipc6.dontfrag < 0)
1552 ipc6.dontfrag = np->dontfrag;
1553 up->len += ulen;
1554 err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr),
1555 &ipc6, &fl6, (struct rt6_info *)dst,
1556 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1557 if (err)
1558 udp_v6_flush_pending_frames(sk);
1559 else if (!corkreq)
1560 err = udp_v6_push_pending_frames(sk);
1561 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1562 up->pending = 0;
1563
1564 if (err > 0)
1565 err = np->recverr ? net_xmit_errno(err) : 0;
1566 release_sock(sk);
1567
1568 out:
1569 dst_release(dst);
1570 out_no_dst:
1571 fl6_sock_release(flowlabel);
1572 txopt_put(opt_to_free);
1573 if (!err)
1574 return len;
1575 /*
1576 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1577 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1578 * we don't have a good statistic (IpOutDiscards but it can be too many
1579 * things). We could add another new stat but at least for now that
1580 * seems like overkill.
1581 */
1582 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1583 UDP6_INC_STATS(sock_net(sk),
1584 UDP_MIB_SNDBUFERRORS, is_udplite);
1585 }
1586 return err;
1587
1588 do_confirm:
1589 if (msg->msg_flags & MSG_PROBE)
1590 dst_confirm_neigh(dst, &fl6.daddr);
1591 if (!(msg->msg_flags&MSG_PROBE) || len)
1592 goto back_from_confirm;
1593 err = 0;
1594 goto out;
1595 }
1596
udpv6_destroy_sock(struct sock * sk)1597 void udpv6_destroy_sock(struct sock *sk)
1598 {
1599 struct udp_sock *up = udp_sk(sk);
1600 lock_sock(sk);
1601 udp_v6_flush_pending_frames(sk);
1602 release_sock(sk);
1603
1604 if (static_branch_unlikely(&udpv6_encap_needed_key)) {
1605 if (up->encap_type) {
1606 void (*encap_destroy)(struct sock *sk);
1607 encap_destroy = READ_ONCE(up->encap_destroy);
1608 if (encap_destroy)
1609 encap_destroy(sk);
1610 }
1611 if (up->encap_enabled) {
1612 static_branch_dec(&udpv6_encap_needed_key);
1613 udp_encap_disable();
1614 }
1615 }
1616
1617 inet6_destroy_sock(sk);
1618 }
1619
1620 /*
1621 * Socket option code for UDP
1622 */
udpv6_setsockopt(struct sock * sk,int level,int optname,sockptr_t optval,unsigned int optlen)1623 int udpv6_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
1624 unsigned int optlen)
1625 {
1626 if (level == SOL_UDP || level == SOL_UDPLITE)
1627 return udp_lib_setsockopt(sk, level, optname,
1628 optval, optlen,
1629 udp_v6_push_pending_frames);
1630 return ipv6_setsockopt(sk, level, optname, optval, optlen);
1631 }
1632
udpv6_getsockopt(struct sock * sk,int level,int optname,char __user * optval,int __user * optlen)1633 int udpv6_getsockopt(struct sock *sk, int level, int optname,
1634 char __user *optval, int __user *optlen)
1635 {
1636 if (level == SOL_UDP || level == SOL_UDPLITE)
1637 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1638 return ipv6_getsockopt(sk, level, optname, optval, optlen);
1639 }
1640
1641 /* thinking of making this const? Don't.
1642 * early_demux can change based on sysctl.
1643 */
1644 static struct inet6_protocol udpv6_protocol = {
1645 .early_demux = udp_v6_early_demux,
1646 .early_demux_handler = udp_v6_early_demux,
1647 .handler = udpv6_rcv,
1648 .err_handler = udpv6_err,
1649 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1650 };
1651
1652 /* ------------------------------------------------------------------------ */
1653 #ifdef CONFIG_PROC_FS
udp6_seq_show(struct seq_file * seq,void * v)1654 int udp6_seq_show(struct seq_file *seq, void *v)
1655 {
1656 if (v == SEQ_START_TOKEN) {
1657 seq_puts(seq, IPV6_SEQ_DGRAM_HEADER);
1658 } else {
1659 int bucket = ((struct udp_iter_state *)seq->private)->bucket;
1660 struct inet_sock *inet = inet_sk(v);
1661 __u16 srcp = ntohs(inet->inet_sport);
1662 __u16 destp = ntohs(inet->inet_dport);
1663 __ip6_dgram_sock_seq_show(seq, v, srcp, destp,
1664 udp_rqueue_get(v), bucket);
1665 }
1666 return 0;
1667 }
1668
1669 const struct seq_operations udp6_seq_ops = {
1670 .start = udp_seq_start,
1671 .next = udp_seq_next,
1672 .stop = udp_seq_stop,
1673 .show = udp6_seq_show,
1674 };
1675 EXPORT_SYMBOL(udp6_seq_ops);
1676
1677 static struct udp_seq_afinfo udp6_seq_afinfo = {
1678 .family = AF_INET6,
1679 .udp_table = &udp_table,
1680 };
1681
udp6_proc_init(struct net * net)1682 int __net_init udp6_proc_init(struct net *net)
1683 {
1684 if (!proc_create_net_data("udp6", 0444, net->proc_net, &udp6_seq_ops,
1685 sizeof(struct udp_iter_state), &udp6_seq_afinfo))
1686 return -ENOMEM;
1687 return 0;
1688 }
1689
udp6_proc_exit(struct net * net)1690 void udp6_proc_exit(struct net *net)
1691 {
1692 remove_proc_entry("udp6", net->proc_net);
1693 }
1694 #endif /* CONFIG_PROC_FS */
1695
1696 /* ------------------------------------------------------------------------ */
1697
1698 struct proto udpv6_prot = {
1699 .name = "UDPv6",
1700 .owner = THIS_MODULE,
1701 .close = udp_lib_close,
1702 .pre_connect = udpv6_pre_connect,
1703 .connect = ip6_datagram_connect,
1704 .disconnect = udp_disconnect,
1705 .ioctl = udp_ioctl,
1706 .init = udp_init_sock,
1707 .destroy = udpv6_destroy_sock,
1708 .setsockopt = udpv6_setsockopt,
1709 .getsockopt = udpv6_getsockopt,
1710 .sendmsg = udpv6_sendmsg,
1711 .recvmsg = udpv6_recvmsg,
1712 .release_cb = ip6_datagram_release_cb,
1713 .hash = udp_lib_hash,
1714 .unhash = udp_lib_unhash,
1715 .rehash = udp_v6_rehash,
1716 .get_port = udp_v6_get_port,
1717 #ifdef CONFIG_BPF_SYSCALL
1718 .psock_update_sk_prot = udp_bpf_update_proto,
1719 #endif
1720 .memory_allocated = &udp_memory_allocated,
1721 .sysctl_mem = sysctl_udp_mem,
1722 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
1723 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
1724 .obj_size = sizeof(struct udp6_sock),
1725 .h.udp_table = &udp_table,
1726 .diag_destroy = udp_abort,
1727 };
1728
1729 static struct inet_protosw udpv6_protosw = {
1730 .type = SOCK_DGRAM,
1731 .protocol = IPPROTO_UDP,
1732 .prot = &udpv6_prot,
1733 .ops = &inet6_dgram_ops,
1734 .flags = INET_PROTOSW_PERMANENT,
1735 };
1736
udpv6_init(void)1737 int __init udpv6_init(void)
1738 {
1739 int ret;
1740
1741 ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP);
1742 if (ret)
1743 goto out;
1744
1745 ret = inet6_register_protosw(&udpv6_protosw);
1746 if (ret)
1747 goto out_udpv6_protocol;
1748 out:
1749 return ret;
1750
1751 out_udpv6_protocol:
1752 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1753 goto out;
1754 }
1755
udpv6_exit(void)1756 void udpv6_exit(void)
1757 {
1758 inet6_unregister_protosw(&udpv6_protosw);
1759 inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP);
1760 }
1761