]>
Commit | Line | Data |
---|---|---|
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/xfrm.h> | |
49 | #include <net/inet_hashtables.h> | |
50 | #include <net/inet6_hashtables.h> | |
51 | #include <net/busy_poll.h> | |
52 | #include <net/sock_reuseport.h> | |
53 | ||
54 | #include <linux/proc_fs.h> | |
55 | #include <linux/seq_file.h> | |
56 | #include <trace/events/skb.h> | |
57 | #include "udp_impl.h" | |
58 | ||
59 | static bool udp6_lib_exact_dif_match(struct net *net, struct sk_buff *skb) | |
60 | { | |
61 | #if defined(CONFIG_NET_L3_MASTER_DEV) | |
62 | if (!net->ipv4.sysctl_udp_l3mdev_accept && | |
63 | skb && ipv6_l3mdev_skb(IP6CB(skb)->flags)) | |
64 | return true; | |
65 | #endif | |
66 | return false; | |
67 | } | |
68 | ||
69 | static u32 udp6_ehashfn(const struct net *net, | |
70 | const struct in6_addr *laddr, | |
71 | const u16 lport, | |
72 | const struct in6_addr *faddr, | |
73 | const __be16 fport) | |
74 | { | |
75 | static u32 udp6_ehash_secret __read_mostly; | |
76 | static u32 udp_ipv6_hash_secret __read_mostly; | |
77 | ||
78 | u32 lhash, fhash; | |
79 | ||
80 | net_get_random_once(&udp6_ehash_secret, | |
81 | sizeof(udp6_ehash_secret)); | |
82 | net_get_random_once(&udp_ipv6_hash_secret, | |
83 | sizeof(udp_ipv6_hash_secret)); | |
84 | ||
85 | lhash = (__force u32)laddr->s6_addr32[3]; | |
86 | fhash = __ipv6_addr_jhash(faddr, udp_ipv6_hash_secret); | |
87 | ||
88 | return __inet6_ehashfn(lhash, lport, fhash, fport, | |
89 | udp_ipv6_hash_secret + net_hash_mix(net)); | |
90 | } | |
91 | ||
92 | static u32 udp6_portaddr_hash(const struct net *net, | |
93 | const struct in6_addr *addr6, | |
94 | unsigned int port) | |
95 | { | |
96 | unsigned int hash, mix = net_hash_mix(net); | |
97 | ||
98 | if (ipv6_addr_any(addr6)) | |
99 | hash = jhash_1word(0, mix); | |
100 | else if (ipv6_addr_v4mapped(addr6)) | |
101 | hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix); | |
102 | else | |
103 | hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix); | |
104 | ||
105 | return hash ^ port; | |
106 | } | |
107 | ||
108 | int udp_v6_get_port(struct sock *sk, unsigned short snum) | |
109 | { | |
110 | unsigned int hash2_nulladdr = | |
111 | udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum); | |
112 | unsigned int hash2_partial = | |
113 | udp6_portaddr_hash(sock_net(sk), &sk->sk_v6_rcv_saddr, 0); | |
114 | ||
115 | /* precompute partial secondary hash */ | |
116 | udp_sk(sk)->udp_portaddr_hash = hash2_partial; | |
117 | return udp_lib_get_port(sk, snum, hash2_nulladdr); | |
118 | } | |
119 | ||
120 | static void udp_v6_rehash(struct sock *sk) | |
121 | { | |
122 | u16 new_hash = udp6_portaddr_hash(sock_net(sk), | |
123 | &sk->sk_v6_rcv_saddr, | |
124 | inet_sk(sk)->inet_num); | |
125 | ||
126 | udp_lib_rehash(sk, new_hash); | |
127 | } | |
128 | ||
129 | static int compute_score(struct sock *sk, struct net *net, | |
130 | const struct in6_addr *saddr, __be16 sport, | |
131 | const struct in6_addr *daddr, unsigned short hnum, | |
132 | int dif, bool exact_dif) | |
133 | { | |
134 | int score; | |
135 | struct inet_sock *inet; | |
136 | ||
137 | if (!net_eq(sock_net(sk), net) || | |
138 | udp_sk(sk)->udp_port_hash != hnum || | |
139 | sk->sk_family != PF_INET6) | |
140 | return -1; | |
141 | ||
142 | score = 0; | |
143 | inet = inet_sk(sk); | |
144 | ||
145 | if (inet->inet_dport) { | |
146 | if (inet->inet_dport != sport) | |
147 | return -1; | |
148 | score++; | |
149 | } | |
150 | ||
151 | if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) { | |
152 | if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr)) | |
153 | return -1; | |
154 | score++; | |
155 | } | |
156 | ||
157 | if (!ipv6_addr_any(&sk->sk_v6_daddr)) { | |
158 | if (!ipv6_addr_equal(&sk->sk_v6_daddr, saddr)) | |
159 | return -1; | |
160 | score++; | |
161 | } | |
162 | ||
163 | if (sk->sk_bound_dev_if || exact_dif) { | |
164 | if (sk->sk_bound_dev_if != dif) | |
165 | return -1; | |
166 | score++; | |
167 | } | |
168 | ||
169 | if (sk->sk_incoming_cpu == raw_smp_processor_id()) | |
170 | score++; | |
171 | ||
172 | return score; | |
173 | } | |
174 | ||
175 | /* called with rcu_read_lock() */ | |
176 | static struct sock *udp6_lib_lookup2(struct net *net, | |
177 | const struct in6_addr *saddr, __be16 sport, | |
178 | const struct in6_addr *daddr, unsigned int hnum, int dif, | |
179 | bool exact_dif, struct udp_hslot *hslot2, | |
180 | struct sk_buff *skb) | |
181 | { | |
182 | struct sock *sk, *result; | |
183 | int score, badness, matches = 0, reuseport = 0; | |
184 | u32 hash = 0; | |
185 | ||
186 | result = NULL; | |
187 | badness = -1; | |
188 | udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { | |
189 | score = compute_score(sk, net, saddr, sport, | |
190 | daddr, hnum, dif, exact_dif); | |
191 | if (score > badness) { | |
192 | reuseport = sk->sk_reuseport; | |
193 | if (reuseport) { | |
194 | hash = udp6_ehashfn(net, daddr, hnum, | |
195 | saddr, sport); | |
196 | ||
197 | result = reuseport_select_sock(sk, hash, skb, | |
198 | sizeof(struct udphdr)); | |
199 | if (result) | |
200 | return result; | |
201 | matches = 1; | |
202 | } | |
203 | result = sk; | |
204 | badness = score; | |
205 | } else if (score == badness && reuseport) { | |
206 | matches++; | |
207 | if (reciprocal_scale(hash, matches) == 0) | |
208 | result = sk; | |
209 | hash = next_pseudo_random32(hash); | |
210 | } | |
211 | } | |
212 | return result; | |
213 | } | |
214 | ||
215 | /* rcu_read_lock() must be held */ | |
216 | struct sock *__udp6_lib_lookup(struct net *net, | |
217 | const struct in6_addr *saddr, __be16 sport, | |
218 | const struct in6_addr *daddr, __be16 dport, | |
219 | int dif, struct udp_table *udptable, | |
220 | struct sk_buff *skb) | |
221 | { | |
222 | struct sock *sk, *result; | |
223 | unsigned short hnum = ntohs(dport); | |
224 | unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); | |
225 | struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; | |
226 | bool exact_dif = udp6_lib_exact_dif_match(net, skb); | |
227 | int score, badness, matches = 0, reuseport = 0; | |
228 | u32 hash = 0; | |
229 | ||
230 | if (hslot->count > 10) { | |
231 | hash2 = udp6_portaddr_hash(net, daddr, hnum); | |
232 | slot2 = hash2 & udptable->mask; | |
233 | hslot2 = &udptable->hash2[slot2]; | |
234 | if (hslot->count < hslot2->count) | |
235 | goto begin; | |
236 | ||
237 | result = udp6_lib_lookup2(net, saddr, sport, | |
238 | daddr, hnum, dif, exact_dif, | |
239 | hslot2, skb); | |
240 | if (!result) { | |
241 | unsigned int old_slot2 = slot2; | |
242 | hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum); | |
243 | slot2 = hash2 & udptable->mask; | |
244 | /* avoid searching the same slot again. */ | |
245 | if (unlikely(slot2 == old_slot2)) | |
246 | return result; | |
247 | ||
248 | hslot2 = &udptable->hash2[slot2]; | |
249 | if (hslot->count < hslot2->count) | |
250 | goto begin; | |
251 | ||
252 | result = udp6_lib_lookup2(net, saddr, sport, | |
253 | daddr, hnum, dif, | |
254 | exact_dif, hslot2, | |
255 | skb); | |
256 | } | |
257 | return result; | |
258 | } | |
259 | begin: | |
260 | result = NULL; | |
261 | badness = -1; | |
262 | sk_for_each_rcu(sk, &hslot->head) { | |
263 | score = compute_score(sk, net, saddr, sport, daddr, hnum, dif, | |
264 | exact_dif); | |
265 | if (score > badness) { | |
266 | reuseport = sk->sk_reuseport; | |
267 | if (reuseport) { | |
268 | hash = udp6_ehashfn(net, daddr, hnum, | |
269 | saddr, sport); | |
270 | result = reuseport_select_sock(sk, hash, skb, | |
271 | sizeof(struct udphdr)); | |
272 | if (result) | |
273 | return result; | |
274 | matches = 1; | |
275 | } | |
276 | result = sk; | |
277 | badness = score; | |
278 | } else if (score == badness && reuseport) { | |
279 | matches++; | |
280 | if (reciprocal_scale(hash, matches) == 0) | |
281 | result = sk; | |
282 | hash = next_pseudo_random32(hash); | |
283 | } | |
284 | } | |
285 | return result; | |
286 | } | |
287 | EXPORT_SYMBOL_GPL(__udp6_lib_lookup); | |
288 | ||
289 | static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, | |
290 | __be16 sport, __be16 dport, | |
291 | struct udp_table *udptable) | |
292 | { | |
293 | const struct ipv6hdr *iph = ipv6_hdr(skb); | |
294 | ||
295 | return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, | |
296 | &iph->daddr, dport, inet6_iif(skb), | |
297 | udptable, skb); | |
298 | } | |
299 | ||
300 | struct sock *udp6_lib_lookup_skb(struct sk_buff *skb, | |
301 | __be16 sport, __be16 dport) | |
302 | { | |
303 | const struct ipv6hdr *iph = ipv6_hdr(skb); | |
304 | ||
305 | return __udp6_lib_lookup(dev_net(skb->dev), &iph->saddr, sport, | |
306 | &iph->daddr, dport, inet6_iif(skb), | |
307 | &udp_table, skb); | |
308 | } | |
309 | EXPORT_SYMBOL_GPL(udp6_lib_lookup_skb); | |
310 | ||
311 | /* Must be called under rcu_read_lock(). | |
312 | * Does increment socket refcount. | |
313 | */ | |
314 | #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \ | |
315 | IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \ | |
316 | IS_ENABLED(CONFIG_NF_SOCKET_IPV6) | |
317 | struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, | |
318 | const struct in6_addr *daddr, __be16 dport, int dif) | |
319 | { | |
320 | struct sock *sk; | |
321 | ||
322 | sk = __udp6_lib_lookup(net, saddr, sport, daddr, dport, | |
323 | dif, &udp_table, NULL); | |
324 | if (sk && !refcount_inc_not_zero(&sk->sk_refcnt)) | |
325 | sk = NULL; | |
326 | return sk; | |
327 | } | |
328 | EXPORT_SYMBOL_GPL(udp6_lib_lookup); | |
329 | #endif | |
330 | ||
331 | /* do not use the scratch area len for jumbogram: their length execeeds the | |
332 | * scratch area space; note that the IP6CB flags is still in the first | |
333 | * cacheline, so checking for jumbograms is cheap | |
334 | */ | |
335 | static int udp6_skb_len(struct sk_buff *skb) | |
336 | { | |
337 | return unlikely(inet6_is_jumbogram(skb)) ? skb->len : udp_skb_len(skb); | |
338 | } | |
339 | ||
340 | /* | |
341 | * This should be easy, if there is something there we | |
342 | * return it, otherwise we block. | |
343 | */ | |
344 | ||
345 | int udpv6_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, | |
346 | int noblock, int flags, int *addr_len) | |
347 | { | |
348 | struct ipv6_pinfo *np = inet6_sk(sk); | |
349 | struct inet_sock *inet = inet_sk(sk); | |
350 | struct sk_buff *skb; | |
351 | unsigned int ulen, copied; | |
352 | int peeked, peeking, off; | |
353 | int err; | |
354 | int is_udplite = IS_UDPLITE(sk); | |
355 | bool checksum_valid = false; | |
356 | int is_udp4; | |
357 | ||
358 | if (flags & MSG_ERRQUEUE) | |
359 | return ipv6_recv_error(sk, msg, len, addr_len); | |
360 | ||
361 | if (np->rxpmtu && np->rxopt.bits.rxpmtu) | |
362 | return ipv6_recv_rxpmtu(sk, msg, len, addr_len); | |
363 | ||
364 | try_again: | |
365 | peeking = flags & MSG_PEEK; | |
366 | off = sk_peek_offset(sk, flags); | |
367 | skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err); | |
368 | if (!skb) | |
369 | return err; | |
370 | ||
371 | ulen = udp6_skb_len(skb); | |
372 | copied = len; | |
373 | if (copied > ulen - off) | |
374 | copied = ulen - off; | |
375 | else if (copied < ulen) | |
376 | msg->msg_flags |= MSG_TRUNC; | |
377 | ||
378 | is_udp4 = (skb->protocol == htons(ETH_P_IP)); | |
379 | ||
380 | /* | |
381 | * If checksum is needed at all, try to do it while copying the | |
382 | * data. If the data is truncated, or if we only want a partial | |
383 | * coverage checksum (UDP-Lite), do it before the copy. | |
384 | */ | |
385 | ||
386 | if (copied < ulen || peeking || | |
387 | (is_udplite && UDP_SKB_CB(skb)->partial_cov)) { | |
388 | checksum_valid = udp_skb_csum_unnecessary(skb) || | |
389 | !__udp_lib_checksum_complete(skb); | |
390 | if (!checksum_valid) | |
391 | goto csum_copy_err; | |
392 | } | |
393 | ||
394 | if (checksum_valid || udp_skb_csum_unnecessary(skb)) { | |
395 | if (udp_skb_is_linear(skb)) | |
396 | err = copy_linear_skb(skb, copied, off, &msg->msg_iter); | |
397 | else | |
398 | err = skb_copy_datagram_msg(skb, off, msg, copied); | |
399 | } else { | |
400 | err = skb_copy_and_csum_datagram_msg(skb, off, msg); | |
401 | if (err == -EINVAL) | |
402 | goto csum_copy_err; | |
403 | } | |
404 | if (unlikely(err)) { | |
405 | if (!peeked) { | |
406 | atomic_inc(&sk->sk_drops); | |
407 | if (is_udp4) | |
408 | UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, | |
409 | is_udplite); | |
410 | else | |
411 | UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, | |
412 | is_udplite); | |
413 | } | |
414 | kfree_skb(skb); | |
415 | return err; | |
416 | } | |
417 | if (!peeked) { | |
418 | if (is_udp4) | |
419 | UDP_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS, | |
420 | is_udplite); | |
421 | else | |
422 | UDP6_INC_STATS(sock_net(sk), UDP_MIB_INDATAGRAMS, | |
423 | is_udplite); | |
424 | } | |
425 | ||
426 | sock_recv_ts_and_drops(msg, sk, skb); | |
427 | ||
428 | /* Copy the address. */ | |
429 | if (msg->msg_name) { | |
430 | DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); | |
431 | sin6->sin6_family = AF_INET6; | |
432 | sin6->sin6_port = udp_hdr(skb)->source; | |
433 | sin6->sin6_flowinfo = 0; | |
434 | ||
435 | if (is_udp4) { | |
436 | ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, | |
437 | &sin6->sin6_addr); | |
438 | sin6->sin6_scope_id = 0; | |
439 | } else { | |
440 | sin6->sin6_addr = ipv6_hdr(skb)->saddr; | |
441 | sin6->sin6_scope_id = | |
442 | ipv6_iface_scope_id(&sin6->sin6_addr, | |
443 | inet6_iif(skb)); | |
444 | } | |
445 | *addr_len = sizeof(*sin6); | |
446 | } | |
447 | ||
448 | if (np->rxopt.all) | |
449 | ip6_datagram_recv_common_ctl(sk, msg, skb); | |
450 | ||
451 | if (is_udp4) { | |
452 | if (inet->cmsg_flags) | |
453 | ip_cmsg_recv_offset(msg, sk, skb, | |
454 | sizeof(struct udphdr), off); | |
455 | } else { | |
456 | if (np->rxopt.all) | |
457 | ip6_datagram_recv_specific_ctl(sk, msg, skb); | |
458 | } | |
459 | ||
460 | err = copied; | |
461 | if (flags & MSG_TRUNC) | |
462 | err = ulen; | |
463 | ||
464 | skb_consume_udp(sk, skb, peeking ? -err : err); | |
465 | return err; | |
466 | ||
467 | csum_copy_err: | |
468 | if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags, | |
469 | udp_skb_destructor)) { | |
470 | if (is_udp4) { | |
471 | UDP_INC_STATS(sock_net(sk), | |
472 | UDP_MIB_CSUMERRORS, is_udplite); | |
473 | UDP_INC_STATS(sock_net(sk), | |
474 | UDP_MIB_INERRORS, is_udplite); | |
475 | } else { | |
476 | UDP6_INC_STATS(sock_net(sk), | |
477 | UDP_MIB_CSUMERRORS, is_udplite); | |
478 | UDP6_INC_STATS(sock_net(sk), | |
479 | UDP_MIB_INERRORS, is_udplite); | |
480 | } | |
481 | } | |
482 | kfree_skb(skb); | |
483 | ||
484 | /* starting over for a new packet, but check if we need to yield */ | |
485 | cond_resched(); | |
486 | msg->msg_flags &= ~MSG_TRUNC; | |
487 | goto try_again; | |
488 | } | |
489 | ||
490 | void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, | |
491 | u8 type, u8 code, int offset, __be32 info, | |
492 | struct udp_table *udptable) | |
493 | { | |
494 | struct ipv6_pinfo *np; | |
495 | const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; | |
496 | const struct in6_addr *saddr = &hdr->saddr; | |
497 | const struct in6_addr *daddr = &hdr->daddr; | |
498 | struct udphdr *uh = (struct udphdr *)(skb->data+offset); | |
499 | struct sock *sk; | |
500 | int harderr; | |
501 | int err; | |
502 | struct net *net = dev_net(skb->dev); | |
503 | ||
504 | sk = __udp6_lib_lookup(net, daddr, uh->dest, saddr, uh->source, | |
505 | inet6_iif(skb), udptable, skb); | |
506 | if (!sk) { | |
507 | __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), | |
508 | ICMP6_MIB_INERRORS); | |
509 | return; | |
510 | } | |
511 | ||
512 | harderr = icmpv6_err_convert(type, code, &err); | |
513 | np = inet6_sk(sk); | |
514 | ||
515 | if (type == ICMPV6_PKT_TOOBIG) { | |
516 | if (!ip6_sk_accept_pmtu(sk)) | |
517 | goto out; | |
518 | ip6_sk_update_pmtu(skb, sk, info); | |
519 | if (np->pmtudisc != IPV6_PMTUDISC_DONT) | |
520 | harderr = 1; | |
521 | } | |
522 | if (type == NDISC_REDIRECT) { | |
523 | ip6_sk_redirect(skb, sk); | |
524 | goto out; | |
525 | } | |
526 | ||
527 | if (!np->recverr) { | |
528 | if (!harderr || sk->sk_state != TCP_ESTABLISHED) | |
529 | goto out; | |
530 | } else { | |
531 | ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); | |
532 | } | |
533 | ||
534 | sk->sk_err = err; | |
535 | sk->sk_error_report(sk); | |
536 | out: | |
537 | return; | |
538 | } | |
539 | ||
540 | static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) | |
541 | { | |
542 | int rc; | |
543 | ||
544 | if (!ipv6_addr_any(&sk->sk_v6_daddr)) { | |
545 | sock_rps_save_rxhash(sk, skb); | |
546 | sk_mark_napi_id(sk, skb); | |
547 | sk_incoming_cpu_update(sk); | |
548 | } else { | |
549 | sk_mark_napi_id_once(sk, skb); | |
550 | } | |
551 | ||
552 | rc = __udp_enqueue_schedule_skb(sk, skb); | |
553 | if (rc < 0) { | |
554 | int is_udplite = IS_UDPLITE(sk); | |
555 | ||
556 | /* Note that an ENOMEM error is charged twice */ | |
557 | if (rc == -ENOMEM) | |
558 | UDP6_INC_STATS(sock_net(sk), | |
559 | UDP_MIB_RCVBUFERRORS, is_udplite); | |
560 | UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); | |
561 | kfree_skb(skb); | |
562 | return -1; | |
563 | } | |
564 | ||
565 | return 0; | |
566 | } | |
567 | ||
568 | static __inline__ void udpv6_err(struct sk_buff *skb, | |
569 | struct inet6_skb_parm *opt, u8 type, | |
570 | u8 code, int offset, __be32 info) | |
571 | { | |
572 | __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); | |
573 | } | |
574 | ||
575 | static struct static_key udpv6_encap_needed __read_mostly; | |
576 | void udpv6_encap_enable(void) | |
577 | { | |
578 | if (!static_key_enabled(&udpv6_encap_needed)) | |
579 | static_key_slow_inc(&udpv6_encap_needed); | |
580 | } | |
581 | EXPORT_SYMBOL(udpv6_encap_enable); | |
582 | ||
583 | static int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) | |
584 | { | |
585 | struct udp_sock *up = udp_sk(sk); | |
586 | int is_udplite = IS_UDPLITE(sk); | |
587 | ||
588 | if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) | |
589 | goto drop; | |
590 | ||
591 | if (static_key_false(&udpv6_encap_needed) && up->encap_type) { | |
592 | int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); | |
593 | ||
594 | /* | |
595 | * This is an encapsulation socket so pass the skb to | |
596 | * the socket's udp_encap_rcv() hook. Otherwise, just | |
597 | * fall through and pass this up the UDP socket. | |
598 | * up->encap_rcv() returns the following value: | |
599 | * =0 if skb was successfully passed to the encap | |
600 | * handler or was discarded by it. | |
601 | * >0 if skb should be passed on to UDP. | |
602 | * <0 if skb should be resubmitted as proto -N | |
603 | */ | |
604 | ||
605 | /* if we're overly short, let UDP handle it */ | |
606 | encap_rcv = ACCESS_ONCE(up->encap_rcv); | |
607 | if (encap_rcv) { | |
608 | int ret; | |
609 | ||
610 | /* Verify checksum before giving to encap */ | |
611 | if (udp_lib_checksum_complete(skb)) | |
612 | goto csum_error; | |
613 | ||
614 | ret = encap_rcv(sk, skb); | |
615 | if (ret <= 0) { | |
616 | __UDP_INC_STATS(sock_net(sk), | |
617 | UDP_MIB_INDATAGRAMS, | |
618 | is_udplite); | |
619 | return -ret; | |
620 | } | |
621 | } | |
622 | ||
623 | /* FALLTHROUGH -- it's a UDP Packet */ | |
624 | } | |
625 | ||
626 | /* | |
627 | * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). | |
628 | */ | |
629 | if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { | |
630 | ||
631 | if (up->pcrlen == 0) { /* full coverage was set */ | |
632 | net_dbg_ratelimited("UDPLITE6: partial coverage %d while full coverage %d requested\n", | |
633 | UDP_SKB_CB(skb)->cscov, skb->len); | |
634 | goto drop; | |
635 | } | |
636 | if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { | |
637 | net_dbg_ratelimited("UDPLITE6: coverage %d too small, need min %d\n", | |
638 | UDP_SKB_CB(skb)->cscov, up->pcrlen); | |
639 | goto drop; | |
640 | } | |
641 | } | |
642 | ||
643 | prefetch(&sk->sk_rmem_alloc); | |
644 | if (rcu_access_pointer(sk->sk_filter) && | |
645 | udp_lib_checksum_complete(skb)) | |
646 | goto csum_error; | |
647 | ||
648 | if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) | |
649 | goto drop; | |
650 | ||
651 | udp_csum_pull_header(skb); | |
652 | ||
653 | skb_dst_drop(skb); | |
654 | ||
655 | return __udpv6_queue_rcv_skb(sk, skb); | |
656 | ||
657 | csum_error: | |
658 | __UDP6_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite); | |
659 | drop: | |
660 | __UDP6_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite); | |
661 | atomic_inc(&sk->sk_drops); | |
662 | kfree_skb(skb); | |
663 | return -1; | |
664 | } | |
665 | ||
666 | static bool __udp_v6_is_mcast_sock(struct net *net, struct sock *sk, | |
667 | __be16 loc_port, const struct in6_addr *loc_addr, | |
668 | __be16 rmt_port, const struct in6_addr *rmt_addr, | |
669 | int dif, unsigned short hnum) | |
670 | { | |
671 | struct inet_sock *inet = inet_sk(sk); | |
672 | ||
673 | if (!net_eq(sock_net(sk), net)) | |
674 | return false; | |
675 | ||
676 | if (udp_sk(sk)->udp_port_hash != hnum || | |
677 | sk->sk_family != PF_INET6 || | |
678 | (inet->inet_dport && inet->inet_dport != rmt_port) || | |
679 | (!ipv6_addr_any(&sk->sk_v6_daddr) && | |
680 | !ipv6_addr_equal(&sk->sk_v6_daddr, rmt_addr)) || | |
681 | (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif) || | |
682 | (!ipv6_addr_any(&sk->sk_v6_rcv_saddr) && | |
683 | !ipv6_addr_equal(&sk->sk_v6_rcv_saddr, loc_addr))) | |
684 | return false; | |
685 | if (!inet6_mc_check(sk, loc_addr, rmt_addr)) | |
686 | return false; | |
687 | return true; | |
688 | } | |
689 | ||
690 | static void udp6_csum_zero_error(struct sk_buff *skb) | |
691 | { | |
692 | /* RFC 2460 section 8.1 says that we SHOULD log | |
693 | * this error. Well, it is reasonable. | |
694 | */ | |
695 | net_dbg_ratelimited("IPv6: udp checksum is 0 for [%pI6c]:%u->[%pI6c]:%u\n", | |
696 | &ipv6_hdr(skb)->saddr, ntohs(udp_hdr(skb)->source), | |
697 | &ipv6_hdr(skb)->daddr, ntohs(udp_hdr(skb)->dest)); | |
698 | } | |
699 | ||
700 | /* | |
701 | * Note: called only from the BH handler context, | |
702 | * so we don't need to lock the hashes. | |
703 | */ | |
704 | static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, | |
705 | const struct in6_addr *saddr, const struct in6_addr *daddr, | |
706 | struct udp_table *udptable, int proto) | |
707 | { | |
708 | struct sock *sk, *first = NULL; | |
709 | const struct udphdr *uh = udp_hdr(skb); | |
710 | unsigned short hnum = ntohs(uh->dest); | |
711 | struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum); | |
712 | unsigned int offset = offsetof(typeof(*sk), sk_node); | |
713 | unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10); | |
714 | int dif = inet6_iif(skb); | |
715 | struct hlist_node *node; | |
716 | struct sk_buff *nskb; | |
717 | ||
718 | if (use_hash2) { | |
719 | hash2_any = udp6_portaddr_hash(net, &in6addr_any, hnum) & | |
720 | udptable->mask; | |
721 | hash2 = udp6_portaddr_hash(net, daddr, hnum) & udptable->mask; | |
722 | start_lookup: | |
723 | hslot = &udptable->hash2[hash2]; | |
724 | offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node); | |
725 | } | |
726 | ||
727 | sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) { | |
728 | if (!__udp_v6_is_mcast_sock(net, sk, uh->dest, daddr, | |
729 | uh->source, saddr, dif, hnum)) | |
730 | continue; | |
731 | /* If zero checksum and no_check is not on for | |
732 | * the socket then skip it. | |
733 | */ | |
734 | if (!uh->check && !udp_sk(sk)->no_check6_rx) | |
735 | continue; | |
736 | if (!first) { | |
737 | first = sk; | |
738 | continue; | |
739 | } | |
740 | nskb = skb_clone(skb, GFP_ATOMIC); | |
741 | if (unlikely(!nskb)) { | |
742 | atomic_inc(&sk->sk_drops); | |
743 | __UDP6_INC_STATS(net, UDP_MIB_RCVBUFERRORS, | |
744 | IS_UDPLITE(sk)); | |
745 | __UDP6_INC_STATS(net, UDP_MIB_INERRORS, | |
746 | IS_UDPLITE(sk)); | |
747 | continue; | |
748 | } | |
749 | ||
750 | if (udpv6_queue_rcv_skb(sk, nskb) > 0) | |
751 | consume_skb(nskb); | |
752 | } | |
753 | ||
754 | /* Also lookup *:port if we are using hash2 and haven't done so yet. */ | |
755 | if (use_hash2 && hash2 != hash2_any) { | |
756 | hash2 = hash2_any; | |
757 | goto start_lookup; | |
758 | } | |
759 | ||
760 | if (first) { | |
761 | if (udpv6_queue_rcv_skb(first, skb) > 0) | |
762 | consume_skb(skb); | |
763 | } else { | |
764 | kfree_skb(skb); | |
765 | __UDP6_INC_STATS(net, UDP_MIB_IGNOREDMULTI, | |
766 | proto == IPPROTO_UDPLITE); | |
767 | } | |
768 | return 0; | |
769 | } | |
770 | ||
771 | static void udp6_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst) | |
772 | { | |
773 | if (udp_sk_rx_dst_set(sk, dst)) { | |
774 | const struct rt6_info *rt = (const struct rt6_info *)dst; | |
775 | ||
776 | inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt); | |
777 | } | |
778 | } | |
779 | ||
780 | int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, | |
781 | int proto) | |
782 | { | |
783 | const struct in6_addr *saddr, *daddr; | |
784 | struct net *net = dev_net(skb->dev); | |
785 | struct udphdr *uh; | |
786 | struct sock *sk; | |
787 | u32 ulen = 0; | |
788 | ||
789 | if (!pskb_may_pull(skb, sizeof(struct udphdr))) | |
790 | goto discard; | |
791 | ||
792 | saddr = &ipv6_hdr(skb)->saddr; | |
793 | daddr = &ipv6_hdr(skb)->daddr; | |
794 | uh = udp_hdr(skb); | |
795 | ||
796 | ulen = ntohs(uh->len); | |
797 | if (ulen > skb->len) | |
798 | goto short_packet; | |
799 | ||
800 | if (proto == IPPROTO_UDP) { | |
801 | /* UDP validates ulen. */ | |
802 | ||
803 | /* Check for jumbo payload */ | |
804 | if (ulen == 0) | |
805 | ulen = skb->len; | |
806 | ||
807 | if (ulen < sizeof(*uh)) | |
808 | goto short_packet; | |
809 | ||
810 | if (ulen < skb->len) { | |
811 | if (pskb_trim_rcsum(skb, ulen)) | |
812 | goto short_packet; | |
813 | saddr = &ipv6_hdr(skb)->saddr; | |
814 | daddr = &ipv6_hdr(skb)->daddr; | |
815 | uh = udp_hdr(skb); | |
816 | } | |
817 | } | |
818 | ||
819 | if (udp6_csum_init(skb, uh, proto)) | |
820 | goto csum_error; | |
821 | ||
822 | /* Check if the socket is already available, e.g. due to early demux */ | |
823 | sk = skb_steal_sock(skb); | |
824 | if (sk) { | |
825 | struct dst_entry *dst = skb_dst(skb); | |
826 | int ret; | |
827 | ||
828 | if (unlikely(sk->sk_rx_dst != dst)) | |
829 | udp6_sk_rx_dst_set(sk, dst); | |
830 | ||
831 | ret = udpv6_queue_rcv_skb(sk, skb); | |
832 | sock_put(sk); | |
833 | ||
834 | /* a return value > 0 means to resubmit the input */ | |
835 | if (ret > 0) | |
836 | return ret; | |
837 | return 0; | |
838 | } | |
839 | ||
840 | /* | |
841 | * Multicast receive code | |
842 | */ | |
843 | if (ipv6_addr_is_multicast(daddr)) | |
844 | return __udp6_lib_mcast_deliver(net, skb, | |
845 | saddr, daddr, udptable, proto); | |
846 | ||
847 | /* Unicast */ | |
848 | sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); | |
849 | if (sk) { | |
850 | int ret; | |
851 | ||
852 | if (!uh->check && !udp_sk(sk)->no_check6_rx) { | |
853 | udp6_csum_zero_error(skb); | |
854 | goto csum_error; | |
855 | } | |
856 | ||
857 | if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk)) | |
858 | skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check, | |
859 | ip6_compute_pseudo); | |
860 | ||
861 | ret = udpv6_queue_rcv_skb(sk, skb); | |
862 | ||
863 | /* a return value > 0 means to resubmit the input */ | |
864 | if (ret > 0) | |
865 | return ret; | |
866 | ||
867 | return 0; | |
868 | } | |
869 | ||
870 | if (!uh->check) { | |
871 | udp6_csum_zero_error(skb); | |
872 | goto csum_error; | |
873 | } | |
874 | ||
875 | if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) | |
876 | goto discard; | |
877 | ||
878 | if (udp_lib_checksum_complete(skb)) | |
879 | goto csum_error; | |
880 | ||
881 | __UDP6_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); | |
882 | icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); | |
883 | ||
884 | kfree_skb(skb); | |
885 | return 0; | |
886 | ||
887 | short_packet: | |
888 | net_dbg_ratelimited("UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", | |
889 | proto == IPPROTO_UDPLITE ? "-Lite" : "", | |
890 | saddr, ntohs(uh->source), | |
891 | ulen, skb->len, | |
892 | daddr, ntohs(uh->dest)); | |
893 | goto discard; | |
894 | csum_error: | |
895 | __UDP6_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE); | |
896 | discard: | |
897 | __UDP6_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); | |
898 | kfree_skb(skb); | |
899 | return 0; | |
900 | } | |
901 | ||
902 | ||
903 | static struct sock *__udp6_lib_demux_lookup(struct net *net, | |
904 | __be16 loc_port, const struct in6_addr *loc_addr, | |
905 | __be16 rmt_port, const struct in6_addr *rmt_addr, | |
906 | int dif) | |
907 | { | |
908 | unsigned short hnum = ntohs(loc_port); | |
909 | unsigned int hash2 = udp6_portaddr_hash(net, loc_addr, hnum); | |
910 | unsigned int slot2 = hash2 & udp_table.mask; | |
911 | struct udp_hslot *hslot2 = &udp_table.hash2[slot2]; | |
912 | const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum); | |
913 | struct sock *sk; | |
914 | ||
915 | udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) { | |
916 | if (sk->sk_state == TCP_ESTABLISHED && | |
917 | INET6_MATCH(sk, net, rmt_addr, loc_addr, ports, dif)) | |
918 | return sk; | |
919 | /* Only check first socket in chain */ | |
920 | break; | |
921 | } | |
922 | return NULL; | |
923 | } | |
924 | ||
925 | static void udp_v6_early_demux(struct sk_buff *skb) | |
926 | { | |
927 | struct net *net = dev_net(skb->dev); | |
928 | const struct udphdr *uh; | |
929 | struct sock *sk; | |
930 | struct dst_entry *dst; | |
931 | int dif = skb->dev->ifindex; | |
932 | ||
933 | if (!pskb_may_pull(skb, skb_transport_offset(skb) + | |
934 | sizeof(struct udphdr))) | |
935 | return; | |
936 | ||
937 | uh = udp_hdr(skb); | |
938 | ||
939 | if (skb->pkt_type == PACKET_HOST) | |
940 | sk = __udp6_lib_demux_lookup(net, uh->dest, | |
941 | &ipv6_hdr(skb)->daddr, | |
942 | uh->source, &ipv6_hdr(skb)->saddr, | |
943 | dif); | |
944 | else | |
945 | return; | |
946 | ||
947 | if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt)) | |
948 | return; | |
949 | ||
950 | skb->sk = sk; | |
951 | skb->destructor = sock_efree; | |
952 | dst = READ_ONCE(sk->sk_rx_dst); | |
953 | ||
954 | if (dst) | |
955 | dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie); | |
956 | if (dst) { | |
957 | /* set noref for now. | |
958 | * any place which wants to hold dst has to call | |
959 | * dst_hold_safe() | |
960 | */ | |
961 | skb_dst_set_noref(skb, dst); | |
962 | } | |
963 | } | |
964 | ||
965 | static __inline__ int udpv6_rcv(struct sk_buff *skb) | |
966 | { | |
967 | return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); | |
968 | } | |
969 | ||
970 | /* | |
971 | * Throw away all pending data and cancel the corking. Socket is locked. | |
972 | */ | |
973 | static void udp_v6_flush_pending_frames(struct sock *sk) | |
974 | { | |
975 | struct udp_sock *up = udp_sk(sk); | |
976 | ||
977 | if (up->pending == AF_INET) | |
978 | udp_flush_pending_frames(sk); | |
979 | else if (up->pending) { | |
980 | up->len = 0; | |
981 | up->pending = 0; | |
982 | ip6_flush_pending_frames(sk); | |
983 | } | |
984 | } | |
985 | ||
986 | /** | |
987 | * udp6_hwcsum_outgoing - handle outgoing HW checksumming | |
988 | * @sk: socket we are sending on | |
989 | * @skb: sk_buff containing the filled-in UDP header | |
990 | * (checksum field must be zeroed out) | |
991 | */ | |
992 | static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, | |
993 | const struct in6_addr *saddr, | |
994 | const struct in6_addr *daddr, int len) | |
995 | { | |
996 | unsigned int offset; | |
997 | struct udphdr *uh = udp_hdr(skb); | |
998 | struct sk_buff *frags = skb_shinfo(skb)->frag_list; | |
999 | __wsum csum = 0; | |
1000 | ||
1001 | if (!frags) { | |
1002 | /* Only one fragment on the socket. */ | |
1003 | skb->csum_start = skb_transport_header(skb) - skb->head; | |
1004 | skb->csum_offset = offsetof(struct udphdr, check); | |
1005 | uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); | |
1006 | } else { | |
1007 | /* | |
1008 | * HW-checksum won't work as there are two or more | |
1009 | * fragments on the socket so that all csums of sk_buffs | |
1010 | * should be together | |
1011 | */ | |
1012 | offset = skb_transport_offset(skb); | |
1013 | skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
1014 | csum = skb->csum; | |
1015 | ||
1016 | skb->ip_summed = CHECKSUM_NONE; | |
1017 | ||
1018 | do { | |
1019 | csum = csum_add(csum, frags->csum); | |
1020 | } while ((frags = frags->next)); | |
1021 | ||
1022 | uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, | |
1023 | csum); | |
1024 | if (uh->check == 0) | |
1025 | uh->check = CSUM_MANGLED_0; | |
1026 | } | |
1027 | } | |
1028 | ||
1029 | /* | |
1030 | * Sending | |
1031 | */ | |
1032 | ||
1033 | static int udp_v6_send_skb(struct sk_buff *skb, struct flowi6 *fl6) | |
1034 | { | |
1035 | struct sock *sk = skb->sk; | |
1036 | struct udphdr *uh; | |
1037 | int err = 0; | |
1038 | int is_udplite = IS_UDPLITE(sk); | |
1039 | __wsum csum = 0; | |
1040 | int offset = skb_transport_offset(skb); | |
1041 | int len = skb->len - offset; | |
1042 | ||
1043 | /* | |
1044 | * Create a UDP header | |
1045 | */ | |
1046 | uh = udp_hdr(skb); | |
1047 | uh->source = fl6->fl6_sport; | |
1048 | uh->dest = fl6->fl6_dport; | |
1049 | uh->len = htons(len); | |
1050 | uh->check = 0; | |
1051 | ||
1052 | if (is_udplite) | |
1053 | csum = udplite_csum(skb); | |
1054 | else if (udp_sk(sk)->no_check6_tx) { /* UDP csum disabled */ | |
1055 | skb->ip_summed = CHECKSUM_NONE; | |
1056 | goto send; | |
1057 | } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ | |
1058 | udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, len); | |
1059 | goto send; | |
1060 | } else | |
1061 | csum = udp_csum(skb); | |
1062 | ||
1063 | /* add protocol-dependent pseudo-header */ | |
1064 | uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, | |
1065 | len, fl6->flowi6_proto, csum); | |
1066 | if (uh->check == 0) | |
1067 | uh->check = CSUM_MANGLED_0; | |
1068 | ||
1069 | send: | |
1070 | err = ip6_send_skb(skb); | |
1071 | if (err) { | |
1072 | if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { | |
1073 | UDP6_INC_STATS(sock_net(sk), | |
1074 | UDP_MIB_SNDBUFERRORS, is_udplite); | |
1075 | err = 0; | |
1076 | } | |
1077 | } else { | |
1078 | UDP6_INC_STATS(sock_net(sk), | |
1079 | UDP_MIB_OUTDATAGRAMS, is_udplite); | |
1080 | } | |
1081 | return err; | |
1082 | } | |
1083 | ||
1084 | static int udp_v6_push_pending_frames(struct sock *sk) | |
1085 | { | |
1086 | struct sk_buff *skb; | |
1087 | struct udp_sock *up = udp_sk(sk); | |
1088 | struct flowi6 fl6; | |
1089 | int err = 0; | |
1090 | ||
1091 | if (up->pending == AF_INET) | |
1092 | return udp_push_pending_frames(sk); | |
1093 | ||
1094 | /* ip6_finish_skb will release the cork, so make a copy of | |
1095 | * fl6 here. | |
1096 | */ | |
1097 | fl6 = inet_sk(sk)->cork.fl.u.ip6; | |
1098 | ||
1099 | skb = ip6_finish_skb(sk); | |
1100 | if (!skb) | |
1101 | goto out; | |
1102 | ||
1103 | err = udp_v6_send_skb(skb, &fl6); | |
1104 | ||
1105 | out: | |
1106 | up->len = 0; | |
1107 | up->pending = 0; | |
1108 | return err; | |
1109 | } | |
1110 | ||
1111 | int udpv6_sendmsg(struct sock *sk, struct msghdr *msg, size_t len) | |
1112 | { | |
1113 | struct ipv6_txoptions opt_space; | |
1114 | struct udp_sock *up = udp_sk(sk); | |
1115 | struct inet_sock *inet = inet_sk(sk); | |
1116 | struct ipv6_pinfo *np = inet6_sk(sk); | |
1117 | DECLARE_SOCKADDR(struct sockaddr_in6 *, sin6, msg->msg_name); | |
1118 | struct in6_addr *daddr, *final_p, final; | |
1119 | struct ipv6_txoptions *opt = NULL; | |
1120 | struct ipv6_txoptions *opt_to_free = NULL; | |
1121 | struct ip6_flowlabel *flowlabel = NULL; | |
1122 | struct flowi6 fl6; | |
1123 | struct dst_entry *dst; | |
1124 | struct ipcm6_cookie ipc6; | |
1125 | int addr_len = msg->msg_namelen; | |
1126 | int ulen = len; | |
1127 | int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; | |
1128 | int err; | |
1129 | int connected = 0; | |
1130 | int is_udplite = IS_UDPLITE(sk); | |
1131 | int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); | |
1132 | struct sockcm_cookie sockc; | |
1133 | ||
1134 | ipc6.hlimit = -1; | |
1135 | ipc6.tclass = -1; | |
1136 | ipc6.dontfrag = -1; | |
1137 | sockc.tsflags = sk->sk_tsflags; | |
1138 | ||
1139 | /* destination address check */ | |
1140 | if (sin6) { | |
1141 | if (addr_len < offsetof(struct sockaddr, sa_data)) | |
1142 | return -EINVAL; | |
1143 | ||
1144 | switch (sin6->sin6_family) { | |
1145 | case AF_INET6: | |
1146 | if (addr_len < SIN6_LEN_RFC2133) | |
1147 | return -EINVAL; | |
1148 | daddr = &sin6->sin6_addr; | |
1149 | if (ipv6_addr_any(daddr) && | |
1150 | ipv6_addr_v4mapped(&np->saddr)) | |
1151 | ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), | |
1152 | daddr); | |
1153 | break; | |
1154 | case AF_INET: | |
1155 | goto do_udp_sendmsg; | |
1156 | case AF_UNSPEC: | |
1157 | msg->msg_name = sin6 = NULL; | |
1158 | msg->msg_namelen = addr_len = 0; | |
1159 | daddr = NULL; | |
1160 | break; | |
1161 | default: | |
1162 | return -EINVAL; | |
1163 | } | |
1164 | } else if (!up->pending) { | |
1165 | if (sk->sk_state != TCP_ESTABLISHED) | |
1166 | return -EDESTADDRREQ; | |
1167 | daddr = &sk->sk_v6_daddr; | |
1168 | } else | |
1169 | daddr = NULL; | |
1170 | ||
1171 | if (daddr) { | |
1172 | if (ipv6_addr_v4mapped(daddr)) { | |
1173 | struct sockaddr_in sin; | |
1174 | sin.sin_family = AF_INET; | |
1175 | sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; | |
1176 | sin.sin_addr.s_addr = daddr->s6_addr32[3]; | |
1177 | msg->msg_name = &sin; | |
1178 | msg->msg_namelen = sizeof(sin); | |
1179 | do_udp_sendmsg: | |
1180 | if (__ipv6_only_sock(sk)) | |
1181 | return -ENETUNREACH; | |
1182 | return udp_sendmsg(sk, msg, len); | |
1183 | } | |
1184 | } | |
1185 | ||
1186 | if (up->pending == AF_INET) | |
1187 | return udp_sendmsg(sk, msg, len); | |
1188 | ||
1189 | /* Rough check on arithmetic overflow, | |
1190 | better check is made in ip6_append_data(). | |
1191 | */ | |
1192 | if (len > INT_MAX - sizeof(struct udphdr)) | |
1193 | return -EMSGSIZE; | |
1194 | ||
1195 | getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; | |
1196 | if (up->pending) { | |
1197 | /* | |
1198 | * There are pending frames. | |
1199 | * The socket lock must be held while it's corked. | |
1200 | */ | |
1201 | lock_sock(sk); | |
1202 | if (likely(up->pending)) { | |
1203 | if (unlikely(up->pending != AF_INET6)) { | |
1204 | release_sock(sk); | |
1205 | return -EAFNOSUPPORT; | |
1206 | } | |
1207 | dst = NULL; | |
1208 | goto do_append_data; | |
1209 | } | |
1210 | release_sock(sk); | |
1211 | } | |
1212 | ulen += sizeof(struct udphdr); | |
1213 | ||
1214 | memset(&fl6, 0, sizeof(fl6)); | |
1215 | ||
1216 | if (sin6) { | |
1217 | if (sin6->sin6_port == 0) | |
1218 | return -EINVAL; | |
1219 | ||
1220 | fl6.fl6_dport = sin6->sin6_port; | |
1221 | daddr = &sin6->sin6_addr; | |
1222 | ||
1223 | if (np->sndflow) { | |
1224 | fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; | |
1225 | if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { | |
1226 | flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); | |
1227 | if (!flowlabel) | |
1228 | return -EINVAL; | |
1229 | } | |
1230 | } | |
1231 | ||
1232 | /* | |
1233 | * Otherwise it will be difficult to maintain | |
1234 | * sk->sk_dst_cache. | |
1235 | */ | |
1236 | if (sk->sk_state == TCP_ESTABLISHED && | |
1237 | ipv6_addr_equal(daddr, &sk->sk_v6_daddr)) | |
1238 | daddr = &sk->sk_v6_daddr; | |
1239 | ||
1240 | if (addr_len >= sizeof(struct sockaddr_in6) && | |
1241 | sin6->sin6_scope_id && | |
1242 | __ipv6_addr_needs_scope_id(__ipv6_addr_type(daddr))) | |
1243 | fl6.flowi6_oif = sin6->sin6_scope_id; | |
1244 | } else { | |
1245 | if (sk->sk_state != TCP_ESTABLISHED) | |
1246 | return -EDESTADDRREQ; | |
1247 | ||
1248 | fl6.fl6_dport = inet->inet_dport; | |
1249 | daddr = &sk->sk_v6_daddr; | |
1250 | fl6.flowlabel = np->flow_label; | |
1251 | connected = 1; | |
1252 | } | |
1253 | ||
1254 | if (!fl6.flowi6_oif) | |
1255 | fl6.flowi6_oif = sk->sk_bound_dev_if; | |
1256 | ||
1257 | if (!fl6.flowi6_oif) | |
1258 | fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; | |
1259 | ||
1260 | fl6.flowi6_mark = sk->sk_mark; | |
1261 | fl6.flowi6_uid = sk->sk_uid; | |
1262 | ||
1263 | if (msg->msg_controllen) { | |
1264 | opt = &opt_space; | |
1265 | memset(opt, 0, sizeof(struct ipv6_txoptions)); | |
1266 | opt->tot_len = sizeof(*opt); | |
1267 | ipc6.opt = opt; | |
1268 | ||
1269 | err = ip6_datagram_send_ctl(sock_net(sk), sk, msg, &fl6, &ipc6, &sockc); | |
1270 | if (err < 0) { | |
1271 | fl6_sock_release(flowlabel); | |
1272 | return err; | |
1273 | } | |
1274 | if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { | |
1275 | flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); | |
1276 | if (!flowlabel) | |
1277 | return -EINVAL; | |
1278 | } | |
1279 | if (!(opt->opt_nflen|opt->opt_flen)) | |
1280 | opt = NULL; | |
1281 | connected = 0; | |
1282 | } | |
1283 | if (!opt) { | |
1284 | opt = txopt_get(np); | |
1285 | opt_to_free = opt; | |
1286 | } | |
1287 | if (flowlabel) | |
1288 | opt = fl6_merge_options(&opt_space, flowlabel, opt); | |
1289 | opt = ipv6_fixup_options(&opt_space, opt); | |
1290 | ipc6.opt = opt; | |
1291 | ||
1292 | fl6.flowi6_proto = sk->sk_protocol; | |
1293 | if (!ipv6_addr_any(daddr)) | |
1294 | fl6.daddr = *daddr; | |
1295 | else | |
1296 | fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ | |
1297 | if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) | |
1298 | fl6.saddr = np->saddr; | |
1299 | fl6.fl6_sport = inet->inet_sport; | |
1300 | ||
1301 | final_p = fl6_update_dst(&fl6, opt, &final); | |
1302 | if (final_p) | |
1303 | connected = 0; | |
1304 | ||
1305 | if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) { | |
1306 | fl6.flowi6_oif = np->mcast_oif; | |
1307 | connected = 0; | |
1308 | } else if (!fl6.flowi6_oif) | |
1309 | fl6.flowi6_oif = np->ucast_oif; | |
1310 | ||
1311 | security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); | |
1312 | ||
1313 | if (ipc6.tclass < 0) | |
1314 | ipc6.tclass = np->tclass; | |
1315 | ||
1316 | fl6.flowlabel = ip6_make_flowinfo(ipc6.tclass, fl6.flowlabel); | |
1317 | ||
1318 | dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p); | |
1319 | if (IS_ERR(dst)) { | |
1320 | err = PTR_ERR(dst); | |
1321 | dst = NULL; | |
1322 | goto out; | |
1323 | } | |
1324 | ||
1325 | if (ipc6.hlimit < 0) | |
1326 | ipc6.hlimit = ip6_sk_dst_hoplimit(np, &fl6, dst); | |
1327 | ||
1328 | if (msg->msg_flags&MSG_CONFIRM) | |
1329 | goto do_confirm; | |
1330 | back_from_confirm: | |
1331 | ||
1332 | /* Lockless fast path for the non-corking case */ | |
1333 | if (!corkreq) { | |
1334 | struct sk_buff *skb; | |
1335 | ||
1336 | skb = ip6_make_skb(sk, getfrag, msg, ulen, | |
1337 | sizeof(struct udphdr), &ipc6, | |
1338 | &fl6, (struct rt6_info *)dst, | |
1339 | msg->msg_flags, &sockc); | |
1340 | err = PTR_ERR(skb); | |
1341 | if (!IS_ERR_OR_NULL(skb)) | |
1342 | err = udp_v6_send_skb(skb, &fl6); | |
1343 | goto release_dst; | |
1344 | } | |
1345 | ||
1346 | lock_sock(sk); | |
1347 | if (unlikely(up->pending)) { | |
1348 | /* The socket is already corked while preparing it. */ | |
1349 | /* ... which is an evident application bug. --ANK */ | |
1350 | release_sock(sk); | |
1351 | ||
1352 | net_dbg_ratelimited("udp cork app bug 2\n"); | |
1353 | err = -EINVAL; | |
1354 | goto out; | |
1355 | } | |
1356 | ||
1357 | up->pending = AF_INET6; | |
1358 | ||
1359 | do_append_data: | |
1360 | if (ipc6.dontfrag < 0) | |
1361 | ipc6.dontfrag = np->dontfrag; | |
1362 | up->len += ulen; | |
1363 | err = ip6_append_data(sk, getfrag, msg, ulen, sizeof(struct udphdr), | |
1364 | &ipc6, &fl6, (struct rt6_info *)dst, | |
1365 | corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, &sockc); | |
1366 | if (err) | |
1367 | udp_v6_flush_pending_frames(sk); | |
1368 | else if (!corkreq) | |
1369 | err = udp_v6_push_pending_frames(sk); | |
1370 | else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) | |
1371 | up->pending = 0; | |
1372 | ||
1373 | if (err > 0) | |
1374 | err = np->recverr ? net_xmit_errno(err) : 0; | |
1375 | release_sock(sk); | |
1376 | ||
1377 | release_dst: | |
1378 | if (dst) { | |
1379 | if (connected) { | |
1380 | ip6_dst_store(sk, dst, | |
1381 | ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr) ? | |
1382 | &sk->sk_v6_daddr : NULL, | |
1383 | #ifdef CONFIG_IPV6_SUBTREES | |
1384 | ipv6_addr_equal(&fl6.saddr, &np->saddr) ? | |
1385 | &np->saddr : | |
1386 | #endif | |
1387 | NULL); | |
1388 | } else { | |
1389 | dst_release(dst); | |
1390 | } | |
1391 | dst = NULL; | |
1392 | } | |
1393 | ||
1394 | out: | |
1395 | dst_release(dst); | |
1396 | fl6_sock_release(flowlabel); | |
1397 | txopt_put(opt_to_free); | |
1398 | if (!err) | |
1399 | return len; | |
1400 | /* | |
1401 | * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting | |
1402 | * ENOBUFS might not be good (it's not tunable per se), but otherwise | |
1403 | * we don't have a good statistic (IpOutDiscards but it can be too many | |
1404 | * things). We could add another new stat but at least for now that | |
1405 | * seems like overkill. | |
1406 | */ | |
1407 | if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { | |
1408 | UDP6_INC_STATS(sock_net(sk), | |
1409 | UDP_MIB_SNDBUFERRORS, is_udplite); | |
1410 | } | |
1411 | return err; | |
1412 | ||
1413 | do_confirm: | |
1414 | if (msg->msg_flags & MSG_PROBE) | |
1415 | dst_confirm_neigh(dst, &fl6.daddr); | |
1416 | if (!(msg->msg_flags&MSG_PROBE) || len) | |
1417 | goto back_from_confirm; | |
1418 | err = 0; | |
1419 | goto out; | |
1420 | } | |
1421 | ||
1422 | void udpv6_destroy_sock(struct sock *sk) | |
1423 | { | |
1424 | struct udp_sock *up = udp_sk(sk); | |
1425 | lock_sock(sk); | |
1426 | udp_v6_flush_pending_frames(sk); | |
1427 | release_sock(sk); | |
1428 | ||
1429 | if (static_key_false(&udpv6_encap_needed) && up->encap_type) { | |
1430 | void (*encap_destroy)(struct sock *sk); | |
1431 | encap_destroy = ACCESS_ONCE(up->encap_destroy); | |
1432 | if (encap_destroy) | |
1433 | encap_destroy(sk); | |
1434 | } | |
1435 | ||
1436 | inet6_destroy_sock(sk); | |
1437 | } | |
1438 | ||
1439 | /* | |
1440 | * Socket option code for UDP | |
1441 | */ | |
1442 | int udpv6_setsockopt(struct sock *sk, int level, int optname, | |
1443 | char __user *optval, unsigned int optlen) | |
1444 | { | |
1445 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
1446 | return udp_lib_setsockopt(sk, level, optname, optval, optlen, | |
1447 | udp_v6_push_pending_frames); | |
1448 | return ipv6_setsockopt(sk, level, optname, optval, optlen); | |
1449 | } | |
1450 | ||
1451 | #ifdef CONFIG_COMPAT | |
1452 | int compat_udpv6_setsockopt(struct sock *sk, int level, int optname, | |
1453 | char __user *optval, unsigned int optlen) | |
1454 | { | |
1455 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
1456 | return udp_lib_setsockopt(sk, level, optname, optval, optlen, | |
1457 | udp_v6_push_pending_frames); | |
1458 | return compat_ipv6_setsockopt(sk, level, optname, optval, optlen); | |
1459 | } | |
1460 | #endif | |
1461 | ||
1462 | int udpv6_getsockopt(struct sock *sk, int level, int optname, | |
1463 | char __user *optval, int __user *optlen) | |
1464 | { | |
1465 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
1466 | return udp_lib_getsockopt(sk, level, optname, optval, optlen); | |
1467 | return ipv6_getsockopt(sk, level, optname, optval, optlen); | |
1468 | } | |
1469 | ||
1470 | #ifdef CONFIG_COMPAT | |
1471 | int compat_udpv6_getsockopt(struct sock *sk, int level, int optname, | |
1472 | char __user *optval, int __user *optlen) | |
1473 | { | |
1474 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
1475 | return udp_lib_getsockopt(sk, level, optname, optval, optlen); | |
1476 | return compat_ipv6_getsockopt(sk, level, optname, optval, optlen); | |
1477 | } | |
1478 | #endif | |
1479 | ||
1480 | static struct inet6_protocol udpv6_protocol = { | |
1481 | .early_demux = udp_v6_early_demux, | |
1482 | .early_demux_handler = udp_v6_early_demux, | |
1483 | .handler = udpv6_rcv, | |
1484 | .err_handler = udpv6_err, | |
1485 | .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, | |
1486 | }; | |
1487 | ||
1488 | /* ------------------------------------------------------------------------ */ | |
1489 | #ifdef CONFIG_PROC_FS | |
1490 | int udp6_seq_show(struct seq_file *seq, void *v) | |
1491 | { | |
1492 | if (v == SEQ_START_TOKEN) { | |
1493 | seq_puts(seq, IPV6_SEQ_DGRAM_HEADER); | |
1494 | } else { | |
1495 | int bucket = ((struct udp_iter_state *)seq->private)->bucket; | |
1496 | struct inet_sock *inet = inet_sk(v); | |
1497 | __u16 srcp = ntohs(inet->inet_sport); | |
1498 | __u16 destp = ntohs(inet->inet_dport); | |
1499 | ip6_dgram_sock_seq_show(seq, v, srcp, destp, bucket); | |
1500 | } | |
1501 | return 0; | |
1502 | } | |
1503 | ||
1504 | static const struct file_operations udp6_afinfo_seq_fops = { | |
1505 | .owner = THIS_MODULE, | |
1506 | .open = udp_seq_open, | |
1507 | .read = seq_read, | |
1508 | .llseek = seq_lseek, | |
1509 | .release = seq_release_net | |
1510 | }; | |
1511 | ||
1512 | static struct udp_seq_afinfo udp6_seq_afinfo = { | |
1513 | .name = "udp6", | |
1514 | .family = AF_INET6, | |
1515 | .udp_table = &udp_table, | |
1516 | .seq_fops = &udp6_afinfo_seq_fops, | |
1517 | .seq_ops = { | |
1518 | .show = udp6_seq_show, | |
1519 | }, | |
1520 | }; | |
1521 | ||
1522 | int __net_init udp6_proc_init(struct net *net) | |
1523 | { | |
1524 | return udp_proc_register(net, &udp6_seq_afinfo); | |
1525 | } | |
1526 | ||
1527 | void udp6_proc_exit(struct net *net) | |
1528 | { | |
1529 | udp_proc_unregister(net, &udp6_seq_afinfo); | |
1530 | } | |
1531 | #endif /* CONFIG_PROC_FS */ | |
1532 | ||
1533 | /* ------------------------------------------------------------------------ */ | |
1534 | ||
1535 | struct proto udpv6_prot = { | |
1536 | .name = "UDPv6", | |
1537 | .owner = THIS_MODULE, | |
1538 | .close = udp_lib_close, | |
1539 | .connect = ip6_datagram_connect, | |
1540 | .disconnect = udp_disconnect, | |
1541 | .ioctl = udp_ioctl, | |
1542 | .init = udp_init_sock, | |
1543 | .destroy = udpv6_destroy_sock, | |
1544 | .setsockopt = udpv6_setsockopt, | |
1545 | .getsockopt = udpv6_getsockopt, | |
1546 | .sendmsg = udpv6_sendmsg, | |
1547 | .recvmsg = udpv6_recvmsg, | |
1548 | .release_cb = ip6_datagram_release_cb, | |
1549 | .hash = udp_lib_hash, | |
1550 | .unhash = udp_lib_unhash, | |
1551 | .rehash = udp_v6_rehash, | |
1552 | .get_port = udp_v6_get_port, | |
1553 | .memory_allocated = &udp_memory_allocated, | |
1554 | .sysctl_mem = sysctl_udp_mem, | |
1555 | .sysctl_wmem = &sysctl_udp_wmem_min, | |
1556 | .sysctl_rmem = &sysctl_udp_rmem_min, | |
1557 | .obj_size = sizeof(struct udp6_sock), | |
1558 | .h.udp_table = &udp_table, | |
1559 | #ifdef CONFIG_COMPAT | |
1560 | .compat_setsockopt = compat_udpv6_setsockopt, | |
1561 | .compat_getsockopt = compat_udpv6_getsockopt, | |
1562 | #endif | |
1563 | .diag_destroy = udp_abort, | |
1564 | }; | |
1565 | ||
1566 | static struct inet_protosw udpv6_protosw = { | |
1567 | .type = SOCK_DGRAM, | |
1568 | .protocol = IPPROTO_UDP, | |
1569 | .prot = &udpv6_prot, | |
1570 | .ops = &inet6_dgram_ops, | |
1571 | .flags = INET_PROTOSW_PERMANENT, | |
1572 | }; | |
1573 | ||
1574 | int __init udpv6_init(void) | |
1575 | { | |
1576 | int ret; | |
1577 | ||
1578 | ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); | |
1579 | if (ret) | |
1580 | goto out; | |
1581 | ||
1582 | ret = inet6_register_protosw(&udpv6_protosw); | |
1583 | if (ret) | |
1584 | goto out_udpv6_protocol; | |
1585 | out: | |
1586 | return ret; | |
1587 | ||
1588 | out_udpv6_protocol: | |
1589 | inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); | |
1590 | goto out; | |
1591 | } | |
1592 | ||
1593 | void udpv6_exit(void) | |
1594 | { | |
1595 | inet6_unregister_protosw(&udpv6_protosw); | |
1596 | inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); | |
1597 | } |