]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * UDP over IPv6 | |
3 | * Linux INET6 implementation | |
4 | * | |
5 | * Authors: | |
6 | * Pedro Roque <[email protected]> | |
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 <asm/uaccess.h> | |
39 | ||
40 | #include <net/ndisc.h> | |
41 | #include <net/protocol.h> | |
42 | #include <net/transp_v6.h> | |
43 | #include <net/ip6_route.h> | |
44 | #include <net/raw.h> | |
45 | #include <net/tcp_states.h> | |
46 | #include <net/ip6_checksum.h> | |
47 | #include <net/xfrm.h> | |
48 | ||
49 | #include <linux/proc_fs.h> | |
50 | #include <linux/seq_file.h> | |
51 | #include "udp_impl.h" | |
52 | ||
53 | int ipv6_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2) | |
54 | { | |
55 | const struct in6_addr *sk_rcv_saddr6 = &inet6_sk(sk)->rcv_saddr; | |
56 | const struct in6_addr *sk2_rcv_saddr6 = inet6_rcv_saddr(sk2); | |
57 | __be32 sk1_rcv_saddr = sk_rcv_saddr(sk); | |
58 | __be32 sk2_rcv_saddr = sk_rcv_saddr(sk2); | |
59 | int sk_ipv6only = ipv6_only_sock(sk); | |
60 | int sk2_ipv6only = inet_v6_ipv6only(sk2); | |
61 | int addr_type = ipv6_addr_type(sk_rcv_saddr6); | |
62 | int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED; | |
63 | ||
64 | /* if both are mapped, treat as IPv4 */ | |
65 | if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) | |
66 | return (!sk2_ipv6only && | |
67 | (!sk1_rcv_saddr || !sk2_rcv_saddr || | |
68 | sk1_rcv_saddr == sk2_rcv_saddr)); | |
69 | ||
70 | if (addr_type2 == IPV6_ADDR_ANY && | |
71 | !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED)) | |
72 | return 1; | |
73 | ||
74 | if (addr_type == IPV6_ADDR_ANY && | |
75 | !(sk_ipv6only && addr_type2 == IPV6_ADDR_MAPPED)) | |
76 | return 1; | |
77 | ||
78 | if (sk2_rcv_saddr6 && | |
79 | ipv6_addr_equal(sk_rcv_saddr6, sk2_rcv_saddr6)) | |
80 | return 1; | |
81 | ||
82 | return 0; | |
83 | } | |
84 | ||
85 | static unsigned int udp6_portaddr_hash(struct net *net, | |
86 | const struct in6_addr *addr6, | |
87 | unsigned int port) | |
88 | { | |
89 | unsigned int hash, mix = net_hash_mix(net); | |
90 | ||
91 | if (ipv6_addr_any(addr6)) | |
92 | hash = jhash_1word(0, mix); | |
93 | else if (ipv6_addr_v4mapped(addr6)) | |
94 | hash = jhash_1word((__force u32)addr6->s6_addr32[3], mix); | |
95 | else | |
96 | hash = jhash2((__force u32 *)addr6->s6_addr32, 4, mix); | |
97 | ||
98 | return hash ^ port; | |
99 | } | |
100 | ||
101 | ||
102 | int udp_v6_get_port(struct sock *sk, unsigned short snum) | |
103 | { | |
104 | unsigned int hash2_nulladdr = | |
105 | udp6_portaddr_hash(sock_net(sk), &in6addr_any, snum); | |
106 | unsigned int hash2_partial = | |
107 | udp6_portaddr_hash(sock_net(sk), &inet6_sk(sk)->rcv_saddr, 0); | |
108 | ||
109 | /* precompute partial secondary hash */ | |
110 | udp_sk(sk)->udp_portaddr_hash = hash2_partial; | |
111 | return udp_lib_get_port(sk, snum, ipv6_rcv_saddr_equal, hash2_nulladdr); | |
112 | } | |
113 | ||
114 | static void udp_v6_rehash(struct sock *sk) | |
115 | { | |
116 | u16 new_hash = udp6_portaddr_hash(sock_net(sk), | |
117 | &inet6_sk(sk)->rcv_saddr, | |
118 | inet_sk(sk)->inet_num); | |
119 | ||
120 | udp_lib_rehash(sk, new_hash); | |
121 | } | |
122 | ||
123 | static inline int compute_score(struct sock *sk, struct net *net, | |
124 | unsigned short hnum, | |
125 | const struct in6_addr *saddr, __be16 sport, | |
126 | const struct in6_addr *daddr, __be16 dport, | |
127 | int dif) | |
128 | { | |
129 | int score = -1; | |
130 | ||
131 | if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum && | |
132 | sk->sk_family == PF_INET6) { | |
133 | struct ipv6_pinfo *np = inet6_sk(sk); | |
134 | struct inet_sock *inet = inet_sk(sk); | |
135 | ||
136 | score = 0; | |
137 | if (inet->inet_dport) { | |
138 | if (inet->inet_dport != sport) | |
139 | return -1; | |
140 | score++; | |
141 | } | |
142 | if (!ipv6_addr_any(&np->rcv_saddr)) { | |
143 | if (!ipv6_addr_equal(&np->rcv_saddr, daddr)) | |
144 | return -1; | |
145 | score++; | |
146 | } | |
147 | if (!ipv6_addr_any(&np->daddr)) { | |
148 | if (!ipv6_addr_equal(&np->daddr, saddr)) | |
149 | return -1; | |
150 | score++; | |
151 | } | |
152 | if (sk->sk_bound_dev_if) { | |
153 | if (sk->sk_bound_dev_if != dif) | |
154 | return -1; | |
155 | score++; | |
156 | } | |
157 | } | |
158 | return score; | |
159 | } | |
160 | ||
161 | #define SCORE2_MAX (1 + 1 + 1) | |
162 | static inline int compute_score2(struct sock *sk, struct net *net, | |
163 | const struct in6_addr *saddr, __be16 sport, | |
164 | const struct in6_addr *daddr, unsigned short hnum, | |
165 | int dif) | |
166 | { | |
167 | int score = -1; | |
168 | ||
169 | if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum && | |
170 | sk->sk_family == PF_INET6) { | |
171 | struct ipv6_pinfo *np = inet6_sk(sk); | |
172 | struct inet_sock *inet = inet_sk(sk); | |
173 | ||
174 | if (!ipv6_addr_equal(&np->rcv_saddr, daddr)) | |
175 | return -1; | |
176 | score = 0; | |
177 | if (inet->inet_dport) { | |
178 | if (inet->inet_dport != sport) | |
179 | return -1; | |
180 | score++; | |
181 | } | |
182 | if (!ipv6_addr_any(&np->daddr)) { | |
183 | if (!ipv6_addr_equal(&np->daddr, saddr)) | |
184 | return -1; | |
185 | score++; | |
186 | } | |
187 | if (sk->sk_bound_dev_if) { | |
188 | if (sk->sk_bound_dev_if != dif) | |
189 | return -1; | |
190 | score++; | |
191 | } | |
192 | } | |
193 | return score; | |
194 | } | |
195 | ||
196 | ||
197 | /* called with read_rcu_lock() */ | |
198 | static struct sock *udp6_lib_lookup2(struct net *net, | |
199 | const struct in6_addr *saddr, __be16 sport, | |
200 | const struct in6_addr *daddr, unsigned int hnum, int dif, | |
201 | struct udp_hslot *hslot2, unsigned int slot2) | |
202 | { | |
203 | struct sock *sk, *result; | |
204 | struct hlist_nulls_node *node; | |
205 | int score, badness; | |
206 | ||
207 | begin: | |
208 | result = NULL; | |
209 | badness = -1; | |
210 | udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) { | |
211 | score = compute_score2(sk, net, saddr, sport, | |
212 | daddr, hnum, dif); | |
213 | if (score > badness) { | |
214 | result = sk; | |
215 | badness = score; | |
216 | if (score == SCORE2_MAX) | |
217 | goto exact_match; | |
218 | } | |
219 | } | |
220 | /* | |
221 | * if the nulls value we got at the end of this lookup is | |
222 | * not the expected one, we must restart lookup. | |
223 | * We probably met an item that was moved to another chain. | |
224 | */ | |
225 | if (get_nulls_value(node) != slot2) | |
226 | goto begin; | |
227 | ||
228 | if (result) { | |
229 | exact_match: | |
230 | if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) | |
231 | result = NULL; | |
232 | else if (unlikely(compute_score2(result, net, saddr, sport, | |
233 | daddr, hnum, dif) < badness)) { | |
234 | sock_put(result); | |
235 | goto begin; | |
236 | } | |
237 | } | |
238 | return result; | |
239 | } | |
240 | ||
241 | struct sock *__udp6_lib_lookup(struct net *net, | |
242 | const struct in6_addr *saddr, __be16 sport, | |
243 | const struct in6_addr *daddr, __be16 dport, | |
244 | int dif, struct udp_table *udptable) | |
245 | { | |
246 | struct sock *sk, *result; | |
247 | struct hlist_nulls_node *node; | |
248 | unsigned short hnum = ntohs(dport); | |
249 | unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask); | |
250 | struct udp_hslot *hslot2, *hslot = &udptable->hash[slot]; | |
251 | int score, badness; | |
252 | ||
253 | rcu_read_lock(); | |
254 | if (hslot->count > 10) { | |
255 | hash2 = udp6_portaddr_hash(net, daddr, hnum); | |
256 | slot2 = hash2 & udptable->mask; | |
257 | hslot2 = &udptable->hash2[slot2]; | |
258 | if (hslot->count < hslot2->count) | |
259 | goto begin; | |
260 | ||
261 | result = udp6_lib_lookup2(net, saddr, sport, | |
262 | daddr, hnum, dif, | |
263 | hslot2, slot2); | |
264 | if (!result) { | |
265 | hash2 = udp6_portaddr_hash(net, &in6addr_any, hnum); | |
266 | slot2 = hash2 & udptable->mask; | |
267 | hslot2 = &udptable->hash2[slot2]; | |
268 | if (hslot->count < hslot2->count) | |
269 | goto begin; | |
270 | ||
271 | result = udp6_lib_lookup2(net, saddr, sport, | |
272 | &in6addr_any, hnum, dif, | |
273 | hslot2, slot2); | |
274 | } | |
275 | rcu_read_unlock(); | |
276 | return result; | |
277 | } | |
278 | begin: | |
279 | result = NULL; | |
280 | badness = -1; | |
281 | sk_nulls_for_each_rcu(sk, node, &hslot->head) { | |
282 | score = compute_score(sk, net, hnum, saddr, sport, daddr, dport, dif); | |
283 | if (score > badness) { | |
284 | result = sk; | |
285 | badness = score; | |
286 | } | |
287 | } | |
288 | /* | |
289 | * if the nulls value we got at the end of this lookup is | |
290 | * not the expected one, we must restart lookup. | |
291 | * We probably met an item that was moved to another chain. | |
292 | */ | |
293 | if (get_nulls_value(node) != slot) | |
294 | goto begin; | |
295 | ||
296 | if (result) { | |
297 | if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2))) | |
298 | result = NULL; | |
299 | else if (unlikely(compute_score(result, net, hnum, saddr, sport, | |
300 | daddr, dport, dif) < badness)) { | |
301 | sock_put(result); | |
302 | goto begin; | |
303 | } | |
304 | } | |
305 | rcu_read_unlock(); | |
306 | return result; | |
307 | } | |
308 | EXPORT_SYMBOL_GPL(__udp6_lib_lookup); | |
309 | ||
310 | static struct sock *__udp6_lib_lookup_skb(struct sk_buff *skb, | |
311 | __be16 sport, __be16 dport, | |
312 | struct udp_table *udptable) | |
313 | { | |
314 | struct sock *sk; | |
315 | const struct ipv6hdr *iph = ipv6_hdr(skb); | |
316 | ||
317 | if (unlikely(sk = skb_steal_sock(skb))) | |
318 | return sk; | |
319 | return __udp6_lib_lookup(dev_net(skb_dst(skb)->dev), &iph->saddr, sport, | |
320 | &iph->daddr, dport, inet6_iif(skb), | |
321 | udptable); | |
322 | } | |
323 | ||
324 | struct sock *udp6_lib_lookup(struct net *net, const struct in6_addr *saddr, __be16 sport, | |
325 | const struct in6_addr *daddr, __be16 dport, int dif) | |
326 | { | |
327 | return __udp6_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table); | |
328 | } | |
329 | EXPORT_SYMBOL_GPL(udp6_lib_lookup); | |
330 | ||
331 | ||
332 | /* | |
333 | * This should be easy, if there is something there we | |
334 | * return it, otherwise we block. | |
335 | */ | |
336 | ||
337 | int udpv6_recvmsg(struct kiocb *iocb, struct sock *sk, | |
338 | struct msghdr *msg, size_t len, | |
339 | int noblock, int flags, int *addr_len) | |
340 | { | |
341 | struct ipv6_pinfo *np = inet6_sk(sk); | |
342 | struct inet_sock *inet = inet_sk(sk); | |
343 | struct sk_buff *skb; | |
344 | unsigned int ulen, copied; | |
345 | int peeked, off = 0; | |
346 | int err; | |
347 | int is_udplite = IS_UDPLITE(sk); | |
348 | int is_udp4; | |
349 | bool slow; | |
350 | ||
351 | if (addr_len) | |
352 | *addr_len = sizeof(struct sockaddr_in6); | |
353 | ||
354 | if (flags & MSG_ERRQUEUE) | |
355 | return ipv6_recv_error(sk, msg, len); | |
356 | ||
357 | if (np->rxpmtu && np->rxopt.bits.rxpmtu) | |
358 | return ipv6_recv_rxpmtu(sk, msg, len); | |
359 | ||
360 | try_again: | |
361 | skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), | |
362 | &peeked, &off, &err); | |
363 | if (!skb) | |
364 | goto out; | |
365 | ||
366 | ulen = skb->len - sizeof(struct udphdr); | |
367 | copied = len; | |
368 | if (copied > ulen) | |
369 | copied = ulen; | |
370 | else if (copied < ulen) | |
371 | msg->msg_flags |= MSG_TRUNC; | |
372 | ||
373 | is_udp4 = (skb->protocol == htons(ETH_P_IP)); | |
374 | ||
375 | /* | |
376 | * If checksum is needed at all, try to do it while copying the | |
377 | * data. If the data is truncated, or if we only want a partial | |
378 | * coverage checksum (UDP-Lite), do it before the copy. | |
379 | */ | |
380 | ||
381 | if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { | |
382 | if (udp_lib_checksum_complete(skb)) | |
383 | goto csum_copy_err; | |
384 | } | |
385 | ||
386 | if (skb_csum_unnecessary(skb)) | |
387 | err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), | |
388 | msg->msg_iov, copied ); | |
389 | else { | |
390 | err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov); | |
391 | if (err == -EINVAL) | |
392 | goto csum_copy_err; | |
393 | } | |
394 | if (err) | |
395 | goto out_free; | |
396 | ||
397 | if (!peeked) { | |
398 | if (is_udp4) | |
399 | UDP_INC_STATS_USER(sock_net(sk), | |
400 | UDP_MIB_INDATAGRAMS, is_udplite); | |
401 | else | |
402 | UDP6_INC_STATS_USER(sock_net(sk), | |
403 | UDP_MIB_INDATAGRAMS, is_udplite); | |
404 | } | |
405 | ||
406 | sock_recv_ts_and_drops(msg, sk, skb); | |
407 | ||
408 | /* Copy the address. */ | |
409 | if (msg->msg_name) { | |
410 | struct sockaddr_in6 *sin6; | |
411 | ||
412 | sin6 = (struct sockaddr_in6 *) msg->msg_name; | |
413 | sin6->sin6_family = AF_INET6; | |
414 | sin6->sin6_port = udp_hdr(skb)->source; | |
415 | sin6->sin6_flowinfo = 0; | |
416 | sin6->sin6_scope_id = 0; | |
417 | ||
418 | if (is_udp4) | |
419 | ipv6_addr_set_v4mapped(ip_hdr(skb)->saddr, | |
420 | &sin6->sin6_addr); | |
421 | else { | |
422 | sin6->sin6_addr = ipv6_hdr(skb)->saddr; | |
423 | if (ipv6_addr_type(&sin6->sin6_addr) & IPV6_ADDR_LINKLOCAL) | |
424 | sin6->sin6_scope_id = IP6CB(skb)->iif; | |
425 | } | |
426 | ||
427 | } | |
428 | if (is_udp4) { | |
429 | if (inet->cmsg_flags) | |
430 | ip_cmsg_recv(msg, skb); | |
431 | } else { | |
432 | if (np->rxopt.all) | |
433 | datagram_recv_ctl(sk, msg, skb); | |
434 | } | |
435 | ||
436 | err = copied; | |
437 | if (flags & MSG_TRUNC) | |
438 | err = ulen; | |
439 | ||
440 | out_free: | |
441 | skb_free_datagram_locked(sk, skb); | |
442 | out: | |
443 | return err; | |
444 | ||
445 | csum_copy_err: | |
446 | slow = lock_sock_fast(sk); | |
447 | if (!skb_kill_datagram(sk, skb, flags)) { | |
448 | if (is_udp4) | |
449 | UDP_INC_STATS_USER(sock_net(sk), | |
450 | UDP_MIB_INERRORS, is_udplite); | |
451 | else | |
452 | UDP6_INC_STATS_USER(sock_net(sk), | |
453 | UDP_MIB_INERRORS, is_udplite); | |
454 | } | |
455 | unlock_sock_fast(sk, slow); | |
456 | ||
457 | if (noblock) | |
458 | return -EAGAIN; | |
459 | ||
460 | /* starting over for a new packet */ | |
461 | msg->msg_flags &= ~MSG_TRUNC; | |
462 | goto try_again; | |
463 | } | |
464 | ||
465 | void __udp6_lib_err(struct sk_buff *skb, struct inet6_skb_parm *opt, | |
466 | u8 type, u8 code, int offset, __be32 info, | |
467 | struct udp_table *udptable) | |
468 | { | |
469 | struct ipv6_pinfo *np; | |
470 | const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; | |
471 | const struct in6_addr *saddr = &hdr->saddr; | |
472 | const struct in6_addr *daddr = &hdr->daddr; | |
473 | struct udphdr *uh = (struct udphdr*)(skb->data+offset); | |
474 | struct sock *sk; | |
475 | int err; | |
476 | ||
477 | sk = __udp6_lib_lookup(dev_net(skb->dev), daddr, uh->dest, | |
478 | saddr, uh->source, inet6_iif(skb), udptable); | |
479 | if (sk == NULL) | |
480 | return; | |
481 | ||
482 | np = inet6_sk(sk); | |
483 | ||
484 | if (!icmpv6_err_convert(type, code, &err) && !np->recverr) | |
485 | goto out; | |
486 | ||
487 | if (sk->sk_state != TCP_ESTABLISHED && !np->recverr) | |
488 | goto out; | |
489 | ||
490 | if (np->recverr) | |
491 | ipv6_icmp_error(sk, skb, err, uh->dest, ntohl(info), (u8 *)(uh+1)); | |
492 | ||
493 | sk->sk_err = err; | |
494 | sk->sk_error_report(sk); | |
495 | out: | |
496 | sock_put(sk); | |
497 | } | |
498 | ||
499 | static int __udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) | |
500 | { | |
501 | int rc; | |
502 | ||
503 | if (!ipv6_addr_any(&inet6_sk(sk)->daddr)) | |
504 | sock_rps_save_rxhash(sk, skb); | |
505 | ||
506 | rc = sock_queue_rcv_skb(sk, skb); | |
507 | if (rc < 0) { | |
508 | int is_udplite = IS_UDPLITE(sk); | |
509 | ||
510 | /* Note that an ENOMEM error is charged twice */ | |
511 | if (rc == -ENOMEM) | |
512 | UDP6_INC_STATS_BH(sock_net(sk), | |
513 | UDP_MIB_RCVBUFERRORS, is_udplite); | |
514 | UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); | |
515 | kfree_skb(skb); | |
516 | return -1; | |
517 | } | |
518 | return 0; | |
519 | } | |
520 | ||
521 | static __inline__ void udpv6_err(struct sk_buff *skb, | |
522 | struct inet6_skb_parm *opt, u8 type, | |
523 | u8 code, int offset, __be32 info ) | |
524 | { | |
525 | __udp6_lib_err(skb, opt, type, code, offset, info, &udp_table); | |
526 | } | |
527 | ||
528 | static struct static_key udpv6_encap_needed __read_mostly; | |
529 | void udpv6_encap_enable(void) | |
530 | { | |
531 | if (!static_key_enabled(&udpv6_encap_needed)) | |
532 | static_key_slow_inc(&udpv6_encap_needed); | |
533 | } | |
534 | EXPORT_SYMBOL(udpv6_encap_enable); | |
535 | ||
536 | int udpv6_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) | |
537 | { | |
538 | struct udp_sock *up = udp_sk(sk); | |
539 | int rc; | |
540 | int is_udplite = IS_UDPLITE(sk); | |
541 | ||
542 | if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) | |
543 | goto drop; | |
544 | ||
545 | if (static_key_false(&udpv6_encap_needed) && up->encap_type) { | |
546 | int (*encap_rcv)(struct sock *sk, struct sk_buff *skb); | |
547 | ||
548 | /* | |
549 | * This is an encapsulation socket so pass the skb to | |
550 | * the socket's udp_encap_rcv() hook. Otherwise, just | |
551 | * fall through and pass this up the UDP socket. | |
552 | * up->encap_rcv() returns the following value: | |
553 | * =0 if skb was successfully passed to the encap | |
554 | * handler or was discarded by it. | |
555 | * >0 if skb should be passed on to UDP. | |
556 | * <0 if skb should be resubmitted as proto -N | |
557 | */ | |
558 | ||
559 | /* if we're overly short, let UDP handle it */ | |
560 | encap_rcv = ACCESS_ONCE(up->encap_rcv); | |
561 | if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) { | |
562 | int ret; | |
563 | ||
564 | ret = encap_rcv(sk, skb); | |
565 | if (ret <= 0) { | |
566 | UDP_INC_STATS_BH(sock_net(sk), | |
567 | UDP_MIB_INDATAGRAMS, | |
568 | is_udplite); | |
569 | return -ret; | |
570 | } | |
571 | } | |
572 | ||
573 | /* FALLTHROUGH -- it's a UDP Packet */ | |
574 | } | |
575 | ||
576 | /* | |
577 | * UDP-Lite specific tests, ignored on UDP sockets (see net/ipv4/udp.c). | |
578 | */ | |
579 | if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { | |
580 | ||
581 | if (up->pcrlen == 0) { /* full coverage was set */ | |
582 | LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: partial coverage" | |
583 | " %d while full coverage %d requested\n", | |
584 | UDP_SKB_CB(skb)->cscov, skb->len); | |
585 | goto drop; | |
586 | } | |
587 | if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { | |
588 | LIMIT_NETDEBUG(KERN_WARNING "UDPLITE6: coverage %d " | |
589 | "too small, need min %d\n", | |
590 | UDP_SKB_CB(skb)->cscov, up->pcrlen); | |
591 | goto drop; | |
592 | } | |
593 | } | |
594 | ||
595 | if (rcu_access_pointer(sk->sk_filter)) { | |
596 | if (udp_lib_checksum_complete(skb)) | |
597 | goto drop; | |
598 | } | |
599 | ||
600 | if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) | |
601 | goto drop; | |
602 | ||
603 | skb_dst_drop(skb); | |
604 | ||
605 | bh_lock_sock(sk); | |
606 | rc = 0; | |
607 | if (!sock_owned_by_user(sk)) | |
608 | rc = __udpv6_queue_rcv_skb(sk, skb); | |
609 | else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { | |
610 | bh_unlock_sock(sk); | |
611 | goto drop; | |
612 | } | |
613 | bh_unlock_sock(sk); | |
614 | ||
615 | return rc; | |
616 | drop: | |
617 | UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); | |
618 | atomic_inc(&sk->sk_drops); | |
619 | kfree_skb(skb); | |
620 | return -1; | |
621 | } | |
622 | ||
623 | static struct sock *udp_v6_mcast_next(struct net *net, struct sock *sk, | |
624 | __be16 loc_port, const struct in6_addr *loc_addr, | |
625 | __be16 rmt_port, const struct in6_addr *rmt_addr, | |
626 | int dif) | |
627 | { | |
628 | struct hlist_nulls_node *node; | |
629 | struct sock *s = sk; | |
630 | unsigned short num = ntohs(loc_port); | |
631 | ||
632 | sk_nulls_for_each_from(s, node) { | |
633 | struct inet_sock *inet = inet_sk(s); | |
634 | ||
635 | if (!net_eq(sock_net(s), net)) | |
636 | continue; | |
637 | ||
638 | if (udp_sk(s)->udp_port_hash == num && | |
639 | s->sk_family == PF_INET6) { | |
640 | struct ipv6_pinfo *np = inet6_sk(s); | |
641 | if (inet->inet_dport) { | |
642 | if (inet->inet_dport != rmt_port) | |
643 | continue; | |
644 | } | |
645 | if (!ipv6_addr_any(&np->daddr) && | |
646 | !ipv6_addr_equal(&np->daddr, rmt_addr)) | |
647 | continue; | |
648 | ||
649 | if (s->sk_bound_dev_if && s->sk_bound_dev_if != dif) | |
650 | continue; | |
651 | ||
652 | if (!ipv6_addr_any(&np->rcv_saddr)) { | |
653 | if (!ipv6_addr_equal(&np->rcv_saddr, loc_addr)) | |
654 | continue; | |
655 | } | |
656 | if (!inet6_mc_check(s, loc_addr, rmt_addr)) | |
657 | continue; | |
658 | return s; | |
659 | } | |
660 | } | |
661 | return NULL; | |
662 | } | |
663 | ||
664 | static void flush_stack(struct sock **stack, unsigned int count, | |
665 | struct sk_buff *skb, unsigned int final) | |
666 | { | |
667 | struct sk_buff *skb1 = NULL; | |
668 | struct sock *sk; | |
669 | unsigned int i; | |
670 | ||
671 | for (i = 0; i < count; i++) { | |
672 | sk = stack[i]; | |
673 | if (likely(skb1 == NULL)) | |
674 | skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC); | |
675 | if (!skb1) { | |
676 | atomic_inc(&sk->sk_drops); | |
677 | UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, | |
678 | IS_UDPLITE(sk)); | |
679 | UDP6_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, | |
680 | IS_UDPLITE(sk)); | |
681 | } | |
682 | ||
683 | if (skb1 && udpv6_queue_rcv_skb(sk, skb1) <= 0) | |
684 | skb1 = NULL; | |
685 | } | |
686 | if (unlikely(skb1)) | |
687 | kfree_skb(skb1); | |
688 | } | |
689 | /* | |
690 | * Note: called only from the BH handler context, | |
691 | * so we don't need to lock the hashes. | |
692 | */ | |
693 | static int __udp6_lib_mcast_deliver(struct net *net, struct sk_buff *skb, | |
694 | const struct in6_addr *saddr, const struct in6_addr *daddr, | |
695 | struct udp_table *udptable) | |
696 | { | |
697 | struct sock *sk, *stack[256 / sizeof(struct sock *)]; | |
698 | const struct udphdr *uh = udp_hdr(skb); | |
699 | struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest)); | |
700 | int dif; | |
701 | unsigned int i, count = 0; | |
702 | ||
703 | spin_lock(&hslot->lock); | |
704 | sk = sk_nulls_head(&hslot->head); | |
705 | dif = inet6_iif(skb); | |
706 | sk = udp_v6_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif); | |
707 | while (sk) { | |
708 | stack[count++] = sk; | |
709 | sk = udp_v6_mcast_next(net, sk_nulls_next(sk), uh->dest, daddr, | |
710 | uh->source, saddr, dif); | |
711 | if (unlikely(count == ARRAY_SIZE(stack))) { | |
712 | if (!sk) | |
713 | break; | |
714 | flush_stack(stack, count, skb, ~0); | |
715 | count = 0; | |
716 | } | |
717 | } | |
718 | /* | |
719 | * before releasing the lock, we must take reference on sockets | |
720 | */ | |
721 | for (i = 0; i < count; i++) | |
722 | sock_hold(stack[i]); | |
723 | ||
724 | spin_unlock(&hslot->lock); | |
725 | ||
726 | if (count) { | |
727 | flush_stack(stack, count, skb, count - 1); | |
728 | ||
729 | for (i = 0; i < count; i++) | |
730 | sock_put(stack[i]); | |
731 | } else { | |
732 | kfree_skb(skb); | |
733 | } | |
734 | return 0; | |
735 | } | |
736 | ||
737 | static inline int udp6_csum_init(struct sk_buff *skb, struct udphdr *uh, | |
738 | int proto) | |
739 | { | |
740 | int err; | |
741 | ||
742 | UDP_SKB_CB(skb)->partial_cov = 0; | |
743 | UDP_SKB_CB(skb)->cscov = skb->len; | |
744 | ||
745 | if (proto == IPPROTO_UDPLITE) { | |
746 | err = udplite_checksum_init(skb, uh); | |
747 | if (err) | |
748 | return err; | |
749 | } | |
750 | ||
751 | if (uh->check == 0) { | |
752 | /* RFC 2460 section 8.1 says that we SHOULD log | |
753 | this error. Well, it is reasonable. | |
754 | */ | |
755 | LIMIT_NETDEBUG(KERN_INFO "IPv6: udp checksum is 0\n"); | |
756 | return 1; | |
757 | } | |
758 | if (skb->ip_summed == CHECKSUM_COMPLETE && | |
759 | !csum_ipv6_magic(&ipv6_hdr(skb)->saddr, &ipv6_hdr(skb)->daddr, | |
760 | skb->len, proto, skb->csum)) | |
761 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
762 | ||
763 | if (!skb_csum_unnecessary(skb)) | |
764 | skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr, | |
765 | &ipv6_hdr(skb)->daddr, | |
766 | skb->len, proto, 0)); | |
767 | ||
768 | return 0; | |
769 | } | |
770 | ||
771 | int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable, | |
772 | int proto) | |
773 | { | |
774 | struct net *net = dev_net(skb->dev); | |
775 | struct sock *sk; | |
776 | struct udphdr *uh; | |
777 | const struct in6_addr *saddr, *daddr; | |
778 | u32 ulen = 0; | |
779 | ||
780 | if (!pskb_may_pull(skb, sizeof(struct udphdr))) | |
781 | goto discard; | |
782 | ||
783 | saddr = &ipv6_hdr(skb)->saddr; | |
784 | daddr = &ipv6_hdr(skb)->daddr; | |
785 | uh = udp_hdr(skb); | |
786 | ||
787 | ulen = ntohs(uh->len); | |
788 | if (ulen > skb->len) | |
789 | goto short_packet; | |
790 | ||
791 | if (proto == IPPROTO_UDP) { | |
792 | /* UDP validates ulen. */ | |
793 | ||
794 | /* Check for jumbo payload */ | |
795 | if (ulen == 0) | |
796 | ulen = skb->len; | |
797 | ||
798 | if (ulen < sizeof(*uh)) | |
799 | goto short_packet; | |
800 | ||
801 | if (ulen < skb->len) { | |
802 | if (pskb_trim_rcsum(skb, ulen)) | |
803 | goto short_packet; | |
804 | saddr = &ipv6_hdr(skb)->saddr; | |
805 | daddr = &ipv6_hdr(skb)->daddr; | |
806 | uh = udp_hdr(skb); | |
807 | } | |
808 | } | |
809 | ||
810 | if (udp6_csum_init(skb, uh, proto)) | |
811 | goto discard; | |
812 | ||
813 | /* | |
814 | * Multicast receive code | |
815 | */ | |
816 | if (ipv6_addr_is_multicast(daddr)) | |
817 | return __udp6_lib_mcast_deliver(net, skb, | |
818 | saddr, daddr, udptable); | |
819 | ||
820 | /* Unicast */ | |
821 | ||
822 | /* | |
823 | * check socket cache ... must talk to Alan about his plans | |
824 | * for sock caches... i'll skip this for now. | |
825 | */ | |
826 | sk = __udp6_lib_lookup_skb(skb, uh->source, uh->dest, udptable); | |
827 | if (sk != NULL) { | |
828 | int ret = udpv6_queue_rcv_skb(sk, skb); | |
829 | sock_put(sk); | |
830 | ||
831 | /* a return value > 0 means to resubmit the input, but | |
832 | * it wants the return to be -protocol, or 0 | |
833 | */ | |
834 | if (ret > 0) | |
835 | return -ret; | |
836 | ||
837 | return 0; | |
838 | } | |
839 | ||
840 | if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) | |
841 | goto discard; | |
842 | ||
843 | if (udp_lib_checksum_complete(skb)) | |
844 | goto discard; | |
845 | ||
846 | UDP6_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); | |
847 | icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0); | |
848 | ||
849 | kfree_skb(skb); | |
850 | return 0; | |
851 | ||
852 | short_packet: | |
853 | LIMIT_NETDEBUG(KERN_DEBUG "UDP%sv6: short packet: From [%pI6c]:%u %d/%d to [%pI6c]:%u\n", | |
854 | proto == IPPROTO_UDPLITE ? "-Lite" : "", | |
855 | saddr, | |
856 | ntohs(uh->source), | |
857 | ulen, | |
858 | skb->len, | |
859 | daddr, | |
860 | ntohs(uh->dest)); | |
861 | ||
862 | discard: | |
863 | UDP6_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); | |
864 | kfree_skb(skb); | |
865 | return 0; | |
866 | } | |
867 | ||
868 | static __inline__ int udpv6_rcv(struct sk_buff *skb) | |
869 | { | |
870 | return __udp6_lib_rcv(skb, &udp_table, IPPROTO_UDP); | |
871 | } | |
872 | ||
873 | /* | |
874 | * Throw away all pending data and cancel the corking. Socket is locked. | |
875 | */ | |
876 | static void udp_v6_flush_pending_frames(struct sock *sk) | |
877 | { | |
878 | struct udp_sock *up = udp_sk(sk); | |
879 | ||
880 | if (up->pending == AF_INET) | |
881 | udp_flush_pending_frames(sk); | |
882 | else if (up->pending) { | |
883 | up->len = 0; | |
884 | up->pending = 0; | |
885 | ip6_flush_pending_frames(sk); | |
886 | } | |
887 | } | |
888 | ||
889 | /** | |
890 | * udp6_hwcsum_outgoing - handle outgoing HW checksumming | |
891 | * @sk: socket we are sending on | |
892 | * @skb: sk_buff containing the filled-in UDP header | |
893 | * (checksum field must be zeroed out) | |
894 | */ | |
895 | static void udp6_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, | |
896 | const struct in6_addr *saddr, | |
897 | const struct in6_addr *daddr, int len) | |
898 | { | |
899 | unsigned int offset; | |
900 | struct udphdr *uh = udp_hdr(skb); | |
901 | __wsum csum = 0; | |
902 | ||
903 | if (skb_queue_len(&sk->sk_write_queue) == 1) { | |
904 | /* Only one fragment on the socket. */ | |
905 | skb->csum_start = skb_transport_header(skb) - skb->head; | |
906 | skb->csum_offset = offsetof(struct udphdr, check); | |
907 | uh->check = ~csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, 0); | |
908 | } else { | |
909 | /* | |
910 | * HW-checksum won't work as there are two or more | |
911 | * fragments on the socket so that all csums of sk_buffs | |
912 | * should be together | |
913 | */ | |
914 | offset = skb_transport_offset(skb); | |
915 | skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
916 | ||
917 | skb->ip_summed = CHECKSUM_NONE; | |
918 | ||
919 | skb_queue_walk(&sk->sk_write_queue, skb) { | |
920 | csum = csum_add(csum, skb->csum); | |
921 | } | |
922 | ||
923 | uh->check = csum_ipv6_magic(saddr, daddr, len, IPPROTO_UDP, | |
924 | csum); | |
925 | if (uh->check == 0) | |
926 | uh->check = CSUM_MANGLED_0; | |
927 | } | |
928 | } | |
929 | ||
930 | /* | |
931 | * Sending | |
932 | */ | |
933 | ||
934 | static int udp_v6_push_pending_frames(struct sock *sk) | |
935 | { | |
936 | struct sk_buff *skb; | |
937 | struct udphdr *uh; | |
938 | struct udp_sock *up = udp_sk(sk); | |
939 | struct inet_sock *inet = inet_sk(sk); | |
940 | struct flowi6 *fl6 = &inet->cork.fl.u.ip6; | |
941 | int err = 0; | |
942 | int is_udplite = IS_UDPLITE(sk); | |
943 | __wsum csum = 0; | |
944 | ||
945 | /* Grab the skbuff where UDP header space exists. */ | |
946 | if ((skb = skb_peek(&sk->sk_write_queue)) == NULL) | |
947 | goto out; | |
948 | ||
949 | /* | |
950 | * Create a UDP header | |
951 | */ | |
952 | uh = udp_hdr(skb); | |
953 | uh->source = fl6->fl6_sport; | |
954 | uh->dest = fl6->fl6_dport; | |
955 | uh->len = htons(up->len); | |
956 | uh->check = 0; | |
957 | ||
958 | if (is_udplite) | |
959 | csum = udplite_csum_outgoing(sk, skb); | |
960 | else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ | |
961 | udp6_hwcsum_outgoing(sk, skb, &fl6->saddr, &fl6->daddr, | |
962 | up->len); | |
963 | goto send; | |
964 | } else | |
965 | csum = udp_csum_outgoing(sk, skb); | |
966 | ||
967 | /* add protocol-dependent pseudo-header */ | |
968 | uh->check = csum_ipv6_magic(&fl6->saddr, &fl6->daddr, | |
969 | up->len, fl6->flowi6_proto, csum); | |
970 | if (uh->check == 0) | |
971 | uh->check = CSUM_MANGLED_0; | |
972 | ||
973 | send: | |
974 | err = ip6_push_pending_frames(sk); | |
975 | if (err) { | |
976 | if (err == -ENOBUFS && !inet6_sk(sk)->recverr) { | |
977 | UDP6_INC_STATS_USER(sock_net(sk), | |
978 | UDP_MIB_SNDBUFERRORS, is_udplite); | |
979 | err = 0; | |
980 | } | |
981 | } else | |
982 | UDP6_INC_STATS_USER(sock_net(sk), | |
983 | UDP_MIB_OUTDATAGRAMS, is_udplite); | |
984 | out: | |
985 | up->len = 0; | |
986 | up->pending = 0; | |
987 | return err; | |
988 | } | |
989 | ||
990 | int udpv6_sendmsg(struct kiocb *iocb, struct sock *sk, | |
991 | struct msghdr *msg, size_t len) | |
992 | { | |
993 | struct ipv6_txoptions opt_space; | |
994 | struct udp_sock *up = udp_sk(sk); | |
995 | struct inet_sock *inet = inet_sk(sk); | |
996 | struct ipv6_pinfo *np = inet6_sk(sk); | |
997 | struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) msg->msg_name; | |
998 | struct in6_addr *daddr, *final_p, final; | |
999 | struct ipv6_txoptions *opt = NULL; | |
1000 | struct ip6_flowlabel *flowlabel = NULL; | |
1001 | struct flowi6 fl6; | |
1002 | struct dst_entry *dst; | |
1003 | int addr_len = msg->msg_namelen; | |
1004 | int ulen = len; | |
1005 | int hlimit = -1; | |
1006 | int tclass = -1; | |
1007 | int dontfrag = -1; | |
1008 | int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; | |
1009 | int err; | |
1010 | int connected = 0; | |
1011 | int is_udplite = IS_UDPLITE(sk); | |
1012 | int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); | |
1013 | ||
1014 | /* destination address check */ | |
1015 | if (sin6) { | |
1016 | if (addr_len < offsetof(struct sockaddr, sa_data)) | |
1017 | return -EINVAL; | |
1018 | ||
1019 | switch (sin6->sin6_family) { | |
1020 | case AF_INET6: | |
1021 | if (addr_len < SIN6_LEN_RFC2133) | |
1022 | return -EINVAL; | |
1023 | daddr = &sin6->sin6_addr; | |
1024 | break; | |
1025 | case AF_INET: | |
1026 | goto do_udp_sendmsg; | |
1027 | case AF_UNSPEC: | |
1028 | msg->msg_name = sin6 = NULL; | |
1029 | msg->msg_namelen = addr_len = 0; | |
1030 | daddr = NULL; | |
1031 | break; | |
1032 | default: | |
1033 | return -EINVAL; | |
1034 | } | |
1035 | } else if (!up->pending) { | |
1036 | if (sk->sk_state != TCP_ESTABLISHED) | |
1037 | return -EDESTADDRREQ; | |
1038 | daddr = &np->daddr; | |
1039 | } else | |
1040 | daddr = NULL; | |
1041 | ||
1042 | if (daddr) { | |
1043 | if (ipv6_addr_v4mapped(daddr)) { | |
1044 | struct sockaddr_in sin; | |
1045 | sin.sin_family = AF_INET; | |
1046 | sin.sin_port = sin6 ? sin6->sin6_port : inet->inet_dport; | |
1047 | sin.sin_addr.s_addr = daddr->s6_addr32[3]; | |
1048 | msg->msg_name = &sin; | |
1049 | msg->msg_namelen = sizeof(sin); | |
1050 | do_udp_sendmsg: | |
1051 | if (__ipv6_only_sock(sk)) | |
1052 | return -ENETUNREACH; | |
1053 | return udp_sendmsg(iocb, sk, msg, len); | |
1054 | } | |
1055 | } | |
1056 | ||
1057 | if (up->pending == AF_INET) | |
1058 | return udp_sendmsg(iocb, sk, msg, len); | |
1059 | ||
1060 | /* Rough check on arithmetic overflow, | |
1061 | better check is made in ip6_append_data(). | |
1062 | */ | |
1063 | if (len > INT_MAX - sizeof(struct udphdr)) | |
1064 | return -EMSGSIZE; | |
1065 | ||
1066 | if (up->pending) { | |
1067 | /* | |
1068 | * There are pending frames. | |
1069 | * The socket lock must be held while it's corked. | |
1070 | */ | |
1071 | lock_sock(sk); | |
1072 | if (likely(up->pending)) { | |
1073 | if (unlikely(up->pending != AF_INET6)) { | |
1074 | release_sock(sk); | |
1075 | return -EAFNOSUPPORT; | |
1076 | } | |
1077 | dst = NULL; | |
1078 | goto do_append_data; | |
1079 | } | |
1080 | release_sock(sk); | |
1081 | } | |
1082 | ulen += sizeof(struct udphdr); | |
1083 | ||
1084 | memset(&fl6, 0, sizeof(fl6)); | |
1085 | ||
1086 | if (sin6) { | |
1087 | if (sin6->sin6_port == 0) | |
1088 | return -EINVAL; | |
1089 | ||
1090 | fl6.fl6_dport = sin6->sin6_port; | |
1091 | daddr = &sin6->sin6_addr; | |
1092 | ||
1093 | if (np->sndflow) { | |
1094 | fl6.flowlabel = sin6->sin6_flowinfo&IPV6_FLOWINFO_MASK; | |
1095 | if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { | |
1096 | flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); | |
1097 | if (flowlabel == NULL) | |
1098 | return -EINVAL; | |
1099 | daddr = &flowlabel->dst; | |
1100 | } | |
1101 | } | |
1102 | ||
1103 | /* | |
1104 | * Otherwise it will be difficult to maintain | |
1105 | * sk->sk_dst_cache. | |
1106 | */ | |
1107 | if (sk->sk_state == TCP_ESTABLISHED && | |
1108 | ipv6_addr_equal(daddr, &np->daddr)) | |
1109 | daddr = &np->daddr; | |
1110 | ||
1111 | if (addr_len >= sizeof(struct sockaddr_in6) && | |
1112 | sin6->sin6_scope_id && | |
1113 | ipv6_addr_type(daddr)&IPV6_ADDR_LINKLOCAL) | |
1114 | fl6.flowi6_oif = sin6->sin6_scope_id; | |
1115 | } else { | |
1116 | if (sk->sk_state != TCP_ESTABLISHED) | |
1117 | return -EDESTADDRREQ; | |
1118 | ||
1119 | fl6.fl6_dport = inet->inet_dport; | |
1120 | daddr = &np->daddr; | |
1121 | fl6.flowlabel = np->flow_label; | |
1122 | connected = 1; | |
1123 | } | |
1124 | ||
1125 | if (!fl6.flowi6_oif) | |
1126 | fl6.flowi6_oif = sk->sk_bound_dev_if; | |
1127 | ||
1128 | if (!fl6.flowi6_oif) | |
1129 | fl6.flowi6_oif = np->sticky_pktinfo.ipi6_ifindex; | |
1130 | ||
1131 | fl6.flowi6_mark = sk->sk_mark; | |
1132 | ||
1133 | if (msg->msg_controllen) { | |
1134 | opt = &opt_space; | |
1135 | memset(opt, 0, sizeof(struct ipv6_txoptions)); | |
1136 | opt->tot_len = sizeof(*opt); | |
1137 | ||
1138 | err = datagram_send_ctl(sock_net(sk), sk, msg, &fl6, opt, | |
1139 | &hlimit, &tclass, &dontfrag); | |
1140 | if (err < 0) { | |
1141 | fl6_sock_release(flowlabel); | |
1142 | return err; | |
1143 | } | |
1144 | if ((fl6.flowlabel&IPV6_FLOWLABEL_MASK) && !flowlabel) { | |
1145 | flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); | |
1146 | if (flowlabel == NULL) | |
1147 | return -EINVAL; | |
1148 | } | |
1149 | if (!(opt->opt_nflen|opt->opt_flen)) | |
1150 | opt = NULL; | |
1151 | connected = 0; | |
1152 | } | |
1153 | if (opt == NULL) | |
1154 | opt = np->opt; | |
1155 | if (flowlabel) | |
1156 | opt = fl6_merge_options(&opt_space, flowlabel, opt); | |
1157 | opt = ipv6_fixup_options(&opt_space, opt); | |
1158 | ||
1159 | fl6.flowi6_proto = sk->sk_protocol; | |
1160 | if (!ipv6_addr_any(daddr)) | |
1161 | fl6.daddr = *daddr; | |
1162 | else | |
1163 | fl6.daddr.s6_addr[15] = 0x1; /* :: means loopback (BSD'ism) */ | |
1164 | if (ipv6_addr_any(&fl6.saddr) && !ipv6_addr_any(&np->saddr)) | |
1165 | fl6.saddr = np->saddr; | |
1166 | fl6.fl6_sport = inet->inet_sport; | |
1167 | ||
1168 | final_p = fl6_update_dst(&fl6, opt, &final); | |
1169 | if (final_p) | |
1170 | connected = 0; | |
1171 | ||
1172 | if (!fl6.flowi6_oif && ipv6_addr_is_multicast(&fl6.daddr)) { | |
1173 | fl6.flowi6_oif = np->mcast_oif; | |
1174 | connected = 0; | |
1175 | } else if (!fl6.flowi6_oif) | |
1176 | fl6.flowi6_oif = np->ucast_oif; | |
1177 | ||
1178 | security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); | |
1179 | ||
1180 | dst = ip6_sk_dst_lookup_flow(sk, &fl6, final_p, true); | |
1181 | if (IS_ERR(dst)) { | |
1182 | err = PTR_ERR(dst); | |
1183 | dst = NULL; | |
1184 | goto out; | |
1185 | } | |
1186 | ||
1187 | if (hlimit < 0) { | |
1188 | if (ipv6_addr_is_multicast(&fl6.daddr)) | |
1189 | hlimit = np->mcast_hops; | |
1190 | else | |
1191 | hlimit = np->hop_limit; | |
1192 | if (hlimit < 0) | |
1193 | hlimit = ip6_dst_hoplimit(dst); | |
1194 | } | |
1195 | ||
1196 | if (tclass < 0) | |
1197 | tclass = np->tclass; | |
1198 | ||
1199 | if (dontfrag < 0) | |
1200 | dontfrag = np->dontfrag; | |
1201 | ||
1202 | if (msg->msg_flags&MSG_CONFIRM) | |
1203 | goto do_confirm; | |
1204 | back_from_confirm: | |
1205 | ||
1206 | lock_sock(sk); | |
1207 | if (unlikely(up->pending)) { | |
1208 | /* The socket is already corked while preparing it. */ | |
1209 | /* ... which is an evident application bug. --ANK */ | |
1210 | release_sock(sk); | |
1211 | ||
1212 | LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); | |
1213 | err = -EINVAL; | |
1214 | goto out; | |
1215 | } | |
1216 | ||
1217 | up->pending = AF_INET6; | |
1218 | ||
1219 | do_append_data: | |
1220 | up->len += ulen; | |
1221 | getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; | |
1222 | err = ip6_append_data(sk, getfrag, msg->msg_iov, ulen, | |
1223 | sizeof(struct udphdr), hlimit, tclass, opt, &fl6, | |
1224 | (struct rt6_info*)dst, | |
1225 | corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags, dontfrag); | |
1226 | if (err) | |
1227 | udp_v6_flush_pending_frames(sk); | |
1228 | else if (!corkreq) | |
1229 | err = udp_v6_push_pending_frames(sk); | |
1230 | else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) | |
1231 | up->pending = 0; | |
1232 | ||
1233 | if (dst) { | |
1234 | if (connected) { | |
1235 | ip6_dst_store(sk, dst, | |
1236 | ipv6_addr_equal(&fl6.daddr, &np->daddr) ? | |
1237 | &np->daddr : NULL, | |
1238 | #ifdef CONFIG_IPV6_SUBTREES | |
1239 | ipv6_addr_equal(&fl6.saddr, &np->saddr) ? | |
1240 | &np->saddr : | |
1241 | #endif | |
1242 | NULL); | |
1243 | } else { | |
1244 | dst_release(dst); | |
1245 | } | |
1246 | dst = NULL; | |
1247 | } | |
1248 | ||
1249 | if (err > 0) | |
1250 | err = np->recverr ? net_xmit_errno(err) : 0; | |
1251 | release_sock(sk); | |
1252 | out: | |
1253 | dst_release(dst); | |
1254 | fl6_sock_release(flowlabel); | |
1255 | if (!err) | |
1256 | return len; | |
1257 | /* | |
1258 | * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting | |
1259 | * ENOBUFS might not be good (it's not tunable per se), but otherwise | |
1260 | * we don't have a good statistic (IpOutDiscards but it can be too many | |
1261 | * things). We could add another new stat but at least for now that | |
1262 | * seems like overkill. | |
1263 | */ | |
1264 | if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { | |
1265 | UDP6_INC_STATS_USER(sock_net(sk), | |
1266 | UDP_MIB_SNDBUFERRORS, is_udplite); | |
1267 | } | |
1268 | return err; | |
1269 | ||
1270 | do_confirm: | |
1271 | dst_confirm(dst); | |
1272 | if (!(msg->msg_flags&MSG_PROBE) || len) | |
1273 | goto back_from_confirm; | |
1274 | err = 0; | |
1275 | goto out; | |
1276 | } | |
1277 | ||
1278 | void udpv6_destroy_sock(struct sock *sk) | |
1279 | { | |
1280 | lock_sock(sk); | |
1281 | udp_v6_flush_pending_frames(sk); | |
1282 | release_sock(sk); | |
1283 | ||
1284 | inet6_destroy_sock(sk); | |
1285 | } | |
1286 | ||
1287 | /* | |
1288 | * Socket option code for UDP | |
1289 | */ | |
1290 | int udpv6_setsockopt(struct sock *sk, int level, int optname, | |
1291 | char __user *optval, unsigned int optlen) | |
1292 | { | |
1293 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
1294 | return udp_lib_setsockopt(sk, level, optname, optval, optlen, | |
1295 | udp_v6_push_pending_frames); | |
1296 | return ipv6_setsockopt(sk, level, optname, optval, optlen); | |
1297 | } | |
1298 | ||
1299 | #ifdef CONFIG_COMPAT | |
1300 | int compat_udpv6_setsockopt(struct sock *sk, int level, int optname, | |
1301 | char __user *optval, unsigned int optlen) | |
1302 | { | |
1303 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
1304 | return udp_lib_setsockopt(sk, level, optname, optval, optlen, | |
1305 | udp_v6_push_pending_frames); | |
1306 | return compat_ipv6_setsockopt(sk, level, optname, optval, optlen); | |
1307 | } | |
1308 | #endif | |
1309 | ||
1310 | int udpv6_getsockopt(struct sock *sk, int level, int optname, | |
1311 | char __user *optval, int __user *optlen) | |
1312 | { | |
1313 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
1314 | return udp_lib_getsockopt(sk, level, optname, optval, optlen); | |
1315 | return ipv6_getsockopt(sk, level, optname, optval, optlen); | |
1316 | } | |
1317 | ||
1318 | #ifdef CONFIG_COMPAT | |
1319 | int compat_udpv6_getsockopt(struct sock *sk, int level, int optname, | |
1320 | char __user *optval, int __user *optlen) | |
1321 | { | |
1322 | if (level == SOL_UDP || level == SOL_UDPLITE) | |
1323 | return udp_lib_getsockopt(sk, level, optname, optval, optlen); | |
1324 | return compat_ipv6_getsockopt(sk, level, optname, optval, optlen); | |
1325 | } | |
1326 | #endif | |
1327 | ||
1328 | static int udp6_ufo_send_check(struct sk_buff *skb) | |
1329 | { | |
1330 | const struct ipv6hdr *ipv6h; | |
1331 | struct udphdr *uh; | |
1332 | ||
1333 | if (!pskb_may_pull(skb, sizeof(*uh))) | |
1334 | return -EINVAL; | |
1335 | ||
1336 | ipv6h = ipv6_hdr(skb); | |
1337 | uh = udp_hdr(skb); | |
1338 | ||
1339 | uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr, skb->len, | |
1340 | IPPROTO_UDP, 0); | |
1341 | skb->csum_start = skb_transport_header(skb) - skb->head; | |
1342 | skb->csum_offset = offsetof(struct udphdr, check); | |
1343 | skb->ip_summed = CHECKSUM_PARTIAL; | |
1344 | return 0; | |
1345 | } | |
1346 | ||
1347 | static struct sk_buff *udp6_ufo_fragment(struct sk_buff *skb, | |
1348 | netdev_features_t features) | |
1349 | { | |
1350 | struct sk_buff *segs = ERR_PTR(-EINVAL); | |
1351 | unsigned int mss; | |
1352 | unsigned int unfrag_ip6hlen, unfrag_len; | |
1353 | struct frag_hdr *fptr; | |
1354 | u8 *mac_start, *prevhdr; | |
1355 | u8 nexthdr; | |
1356 | u8 frag_hdr_sz = sizeof(struct frag_hdr); | |
1357 | int offset; | |
1358 | __wsum csum; | |
1359 | ||
1360 | mss = skb_shinfo(skb)->gso_size; | |
1361 | if (unlikely(skb->len <= mss)) | |
1362 | goto out; | |
1363 | ||
1364 | if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) { | |
1365 | /* Packet is from an untrusted source, reset gso_segs. */ | |
1366 | int type = skb_shinfo(skb)->gso_type; | |
1367 | ||
1368 | if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) || | |
1369 | !(type & (SKB_GSO_UDP)))) | |
1370 | goto out; | |
1371 | ||
1372 | skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss); | |
1373 | ||
1374 | segs = NULL; | |
1375 | goto out; | |
1376 | } | |
1377 | ||
1378 | /* Do software UFO. Complete and fill in the UDP checksum as HW cannot | |
1379 | * do checksum of UDP packets sent as multiple IP fragments. | |
1380 | */ | |
1381 | offset = skb_checksum_start_offset(skb); | |
1382 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
1383 | offset += skb->csum_offset; | |
1384 | *(__sum16 *)(skb->data + offset) = csum_fold(csum); | |
1385 | skb->ip_summed = CHECKSUM_NONE; | |
1386 | ||
1387 | /* Check if there is enough headroom to insert fragment header. */ | |
1388 | if ((skb_mac_header(skb) < skb->head + frag_hdr_sz) && | |
1389 | pskb_expand_head(skb, frag_hdr_sz, 0, GFP_ATOMIC)) | |
1390 | goto out; | |
1391 | ||
1392 | /* Find the unfragmentable header and shift it left by frag_hdr_sz | |
1393 | * bytes to insert fragment header. | |
1394 | */ | |
1395 | unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr); | |
1396 | nexthdr = *prevhdr; | |
1397 | *prevhdr = NEXTHDR_FRAGMENT; | |
1398 | unfrag_len = skb_network_header(skb) - skb_mac_header(skb) + | |
1399 | unfrag_ip6hlen; | |
1400 | mac_start = skb_mac_header(skb); | |
1401 | memmove(mac_start-frag_hdr_sz, mac_start, unfrag_len); | |
1402 | ||
1403 | skb->mac_header -= frag_hdr_sz; | |
1404 | skb->network_header -= frag_hdr_sz; | |
1405 | ||
1406 | fptr = (struct frag_hdr *)(skb_network_header(skb) + unfrag_ip6hlen); | |
1407 | fptr->nexthdr = nexthdr; | |
1408 | fptr->reserved = 0; | |
1409 | ipv6_select_ident(fptr, (struct rt6_info *)skb_dst(skb)); | |
1410 | ||
1411 | /* Fragment the skb. ipv6 header and the remaining fields of the | |
1412 | * fragment header are updated in ipv6_gso_segment() | |
1413 | */ | |
1414 | segs = skb_segment(skb, features); | |
1415 | ||
1416 | out: | |
1417 | return segs; | |
1418 | } | |
1419 | ||
1420 | static const struct inet6_protocol udpv6_protocol = { | |
1421 | .handler = udpv6_rcv, | |
1422 | .err_handler = udpv6_err, | |
1423 | .gso_send_check = udp6_ufo_send_check, | |
1424 | .gso_segment = udp6_ufo_fragment, | |
1425 | .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, | |
1426 | }; | |
1427 | ||
1428 | /* ------------------------------------------------------------------------ */ | |
1429 | #ifdef CONFIG_PROC_FS | |
1430 | ||
1431 | static void udp6_sock_seq_show(struct seq_file *seq, struct sock *sp, int bucket) | |
1432 | { | |
1433 | struct inet_sock *inet = inet_sk(sp); | |
1434 | struct ipv6_pinfo *np = inet6_sk(sp); | |
1435 | const struct in6_addr *dest, *src; | |
1436 | __u16 destp, srcp; | |
1437 | ||
1438 | dest = &np->daddr; | |
1439 | src = &np->rcv_saddr; | |
1440 | destp = ntohs(inet->inet_dport); | |
1441 | srcp = ntohs(inet->inet_sport); | |
1442 | seq_printf(seq, | |
1443 | "%5d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " | |
1444 | "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d\n", | |
1445 | bucket, | |
1446 | src->s6_addr32[0], src->s6_addr32[1], | |
1447 | src->s6_addr32[2], src->s6_addr32[3], srcp, | |
1448 | dest->s6_addr32[0], dest->s6_addr32[1], | |
1449 | dest->s6_addr32[2], dest->s6_addr32[3], destp, | |
1450 | sp->sk_state, | |
1451 | sk_wmem_alloc_get(sp), | |
1452 | sk_rmem_alloc_get(sp), | |
1453 | 0, 0L, 0, | |
1454 | sock_i_uid(sp), 0, | |
1455 | sock_i_ino(sp), | |
1456 | atomic_read(&sp->sk_refcnt), sp, | |
1457 | atomic_read(&sp->sk_drops)); | |
1458 | } | |
1459 | ||
1460 | int udp6_seq_show(struct seq_file *seq, void *v) | |
1461 | { | |
1462 | if (v == SEQ_START_TOKEN) | |
1463 | seq_printf(seq, | |
1464 | " sl " | |
1465 | "local_address " | |
1466 | "remote_address " | |
1467 | "st tx_queue rx_queue tr tm->when retrnsmt" | |
1468 | " uid timeout inode ref pointer drops\n"); | |
1469 | else | |
1470 | udp6_sock_seq_show(seq, v, ((struct udp_iter_state *)seq->private)->bucket); | |
1471 | return 0; | |
1472 | } | |
1473 | ||
1474 | static const struct file_operations udp6_afinfo_seq_fops = { | |
1475 | .owner = THIS_MODULE, | |
1476 | .open = udp_seq_open, | |
1477 | .read = seq_read, | |
1478 | .llseek = seq_lseek, | |
1479 | .release = seq_release_net | |
1480 | }; | |
1481 | ||
1482 | static struct udp_seq_afinfo udp6_seq_afinfo = { | |
1483 | .name = "udp6", | |
1484 | .family = AF_INET6, | |
1485 | .udp_table = &udp_table, | |
1486 | .seq_fops = &udp6_afinfo_seq_fops, | |
1487 | .seq_ops = { | |
1488 | .show = udp6_seq_show, | |
1489 | }, | |
1490 | }; | |
1491 | ||
1492 | int __net_init udp6_proc_init(struct net *net) | |
1493 | { | |
1494 | return udp_proc_register(net, &udp6_seq_afinfo); | |
1495 | } | |
1496 | ||
1497 | void udp6_proc_exit(struct net *net) { | |
1498 | udp_proc_unregister(net, &udp6_seq_afinfo); | |
1499 | } | |
1500 | #endif /* CONFIG_PROC_FS */ | |
1501 | ||
1502 | /* ------------------------------------------------------------------------ */ | |
1503 | ||
1504 | struct proto udpv6_prot = { | |
1505 | .name = "UDPv6", | |
1506 | .owner = THIS_MODULE, | |
1507 | .close = udp_lib_close, | |
1508 | .connect = ip6_datagram_connect, | |
1509 | .disconnect = udp_disconnect, | |
1510 | .ioctl = udp_ioctl, | |
1511 | .destroy = udpv6_destroy_sock, | |
1512 | .setsockopt = udpv6_setsockopt, | |
1513 | .getsockopt = udpv6_getsockopt, | |
1514 | .sendmsg = udpv6_sendmsg, | |
1515 | .recvmsg = udpv6_recvmsg, | |
1516 | .backlog_rcv = __udpv6_queue_rcv_skb, | |
1517 | .hash = udp_lib_hash, | |
1518 | .unhash = udp_lib_unhash, | |
1519 | .rehash = udp_v6_rehash, | |
1520 | .get_port = udp_v6_get_port, | |
1521 | .memory_allocated = &udp_memory_allocated, | |
1522 | .sysctl_mem = sysctl_udp_mem, | |
1523 | .sysctl_wmem = &sysctl_udp_wmem_min, | |
1524 | .sysctl_rmem = &sysctl_udp_rmem_min, | |
1525 | .obj_size = sizeof(struct udp6_sock), | |
1526 | .slab_flags = SLAB_DESTROY_BY_RCU, | |
1527 | .h.udp_table = &udp_table, | |
1528 | #ifdef CONFIG_COMPAT | |
1529 | .compat_setsockopt = compat_udpv6_setsockopt, | |
1530 | .compat_getsockopt = compat_udpv6_getsockopt, | |
1531 | #endif | |
1532 | .clear_sk = sk_prot_clear_portaddr_nulls, | |
1533 | }; | |
1534 | ||
1535 | static struct inet_protosw udpv6_protosw = { | |
1536 | .type = SOCK_DGRAM, | |
1537 | .protocol = IPPROTO_UDP, | |
1538 | .prot = &udpv6_prot, | |
1539 | .ops = &inet6_dgram_ops, | |
1540 | .no_check = UDP_CSUM_DEFAULT, | |
1541 | .flags = INET_PROTOSW_PERMANENT, | |
1542 | }; | |
1543 | ||
1544 | ||
1545 | int __init udpv6_init(void) | |
1546 | { | |
1547 | int ret; | |
1548 | ||
1549 | ret = inet6_add_protocol(&udpv6_protocol, IPPROTO_UDP); | |
1550 | if (ret) | |
1551 | goto out; | |
1552 | ||
1553 | ret = inet6_register_protosw(&udpv6_protosw); | |
1554 | if (ret) | |
1555 | goto out_udpv6_protocol; | |
1556 | out: | |
1557 | return ret; | |
1558 | ||
1559 | out_udpv6_protocol: | |
1560 | inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); | |
1561 | goto out; | |
1562 | } | |
1563 | ||
1564 | void udpv6_exit(void) | |
1565 | { | |
1566 | inet6_unregister_protosw(&udpv6_protosw); | |
1567 | inet6_del_protocol(&udpv6_protocol, IPPROTO_UDP); | |
1568 | } |