?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
udp: get rid of sk_prot_clear_portaddr_nulls()
[linux-2.6/btrfs-unstable.git] / net / ipv4 / udp.c
blobf0ebb0bd1e11afc344445458b22c58cd8cb5b8f8
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * The User Datagram Protocol (UDP).
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
11 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
12 * Hirokazu Takahashi, <taka@valinux.co.jp>
13 *
14 * Fixes:
15 * Alan Cox : verify_area() calls
16 * Alan Cox : stopped close while in use off icmp
17 * messages. Not a fix but a botch that
18 * for udp at least is 'valid'.
19 * Alan Cox : Fixed icmp handling properly
20 * Alan Cox : Correct error for oversized datagrams
21 * Alan Cox : Tidied select() semantics.
22 * Alan Cox : udp_err() fixed properly, also now
23 * select and read wake correctly on errors
24 * Alan Cox : udp_send verify_area moved to avoid mem leak
25 * Alan Cox : UDP can count its memory
26 * Alan Cox : send to an unknown connection causes
27 * an ECONNREFUSED off the icmp, but
28 * does NOT close.
29 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
31 * bug no longer crashes it.
32 * Fred Van Kempen : Net2e support for sk->broadcast.
33 * Alan Cox : Uses skb_free_datagram
34 * Alan Cox : Added get/set sockopt support.
35 * Alan Cox : Broadcasting without option set returns EACCES.
36 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
37 * Alan Cox : Use ip_tos and ip_ttl
38 * Alan Cox : SNMP Mibs
39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
40 * Matt Dillon : UDP length checks.
41 * Alan Cox : Smarter af_inet used properly.
42 * Alan Cox : Use new kernel side addressing.
43 * Alan Cox : Incorrect return on truncated datagram receive.
44 * Arnt Gulbrandsen : New udp_send and stuff
45 * Alan Cox : Cache last socket
46 * Alan Cox : Route cache
47 * Jon Peatfield : Minor efficiency fix to sendto().
48 * Mike Shaver : RFC1122 checks.
49 * Alan Cox : Nonblocking error fix.
50 * Willy Konynenberg : Transparent proxying support.
51 * Mike McLagan : Routing by source
52 * David S. Miller : New socket lookup architecture.
53 * Last socket cache retained as it
54 * does have a high hit rate.
55 * Olaf Kirch : Don't linearise iovec on sendmsg.
56 * Andi Kleen : Some cleanups, cache destination entry
57 * for connect.
58 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
59 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
60 * return ENOTCONN for unconnected sockets (POSIX)
61 * Janos Farkas : don't deliver multi/broadcasts to a different
62 * bound-to-device socket
63 * Hirokazu Takahashi : HW checksumming for outgoing UDP
64 * datagrams.
65 * Hirokazu Takahashi : sendfile() on UDP works now.
66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
69 * a single port at the same time.
70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
71 * James Chapman : Add L2TP encapsulation type.
72 *
73 *
74 * This program is free software; you can redistribute it and/or
75 * modify it under the terms of the GNU General Public License
76 * as published by the Free Software Foundation; either version
77 * 2 of the License, or (at your option) any later version.
78 */
79
80 #define pr_fmt(fmt) "UDP: " fmt
81
82 #include <asm/uaccess.h>
83 #include <asm/ioctls.h>
84 #include <linux/bootmem.h>
85 #include <linux/highmem.h>
86 #include <linux/swap.h>
87 #include <linux/types.h>
88 #include <linux/fcntl.h>
89 #include <linux/module.h>
90 #include <linux/socket.h>
91 #include <linux/sockios.h>
92 #include <linux/igmp.h>
93 #include <linux/inetdevice.h>
94 #include <linux/in.h>
95 #include <linux/errno.h>
96 #include <linux/timer.h>
97 #include <linux/mm.h>
98 #include <linux/inet.h>
99 #include <linux/netdevice.h>
100 #include <linux/slab.h>
101 #include <net/tcp_states.h>
102 #include <linux/skbuff.h>
103 #include <linux/proc_fs.h>
104 #include <linux/seq_file.h>
105 #include <net/net_namespace.h>
106 #include <net/icmp.h>
107 #include <net/inet_hashtables.h>
108 #include <net/route.h>
109 #include <net/checksum.h>
110 #include <net/xfrm.h>
111 #include <trace/events/udp.h>
112 #include <linux/static_key.h>
113 #include <trace/events/skb.h>
114 #include <net/busy_poll.h>
115 #include "udp_impl.h"
116 #include <net/sock_reuseport.h>
117 #include <net/addrconf.h>
118
119 struct udp_table udp_table __read_mostly;
120 EXPORT_SYMBOL(udp_table);
121
122 long sysctl_udp_mem[3] __read_mostly;
123 EXPORT_SYMBOL(sysctl_udp_mem);
124
125 int sysctl_udp_rmem_min __read_mostly;
126 EXPORT_SYMBOL(sysctl_udp_rmem_min);
127
128 int sysctl_udp_wmem_min __read_mostly;
129 EXPORT_SYMBOL(sysctl_udp_wmem_min);
130
131 atomic_long_t udp_memory_allocated;
132 EXPORT_SYMBOL(udp_memory_allocated);
133
134 #define MAX_UDP_PORTS 65536
135 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
136
137 static int udp_lib_lport_inuse(struct net *net, __u16 num,
138 const struct udp_hslot *hslot,
139 unsigned long *bitmap,
140 struct sock *sk,
141 int (*saddr_comp)(const struct sock *sk1,
142 const struct sock *sk2,
143 bool match_wildcard),
144 unsigned int log)
145 {
146 struct sock *sk2;
147 kuid_t uid = sock_i_uid(sk);
148
149 sk_for_each(sk2, &hslot->head) {
150 if (net_eq(sock_net(sk2), net) &&
151 sk2 != sk &&
152 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
153 (!sk2->sk_reuse || !sk->sk_reuse) &&
154 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
155 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
156 (!sk2->sk_reuseport || !sk->sk_reuseport ||
157 rcu_access_pointer(sk->sk_reuseport_cb) ||
158 !uid_eq(uid, sock_i_uid(sk2))) &&
159 saddr_comp(sk, sk2, true)) {
160 if (!bitmap)
161 return 1;
162 __set_bit(udp_sk(sk2)->udp_port_hash >> log, bitmap);
163 }
164 }
165 return 0;
166 }
167
168 /*
169 * Note: we still hold spinlock of primary hash chain, so no other writer
170 * can insert/delete a socket with local_port == num
171 */
172 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
173 struct udp_hslot *hslot2,
174 struct sock *sk,
175 int (*saddr_comp)(const struct sock *sk1,
176 const struct sock *sk2,
177 bool match_wildcard))
178 {
179 struct sock *sk2;
180 kuid_t uid = sock_i_uid(sk);
181 int res = 0;
182
183 spin_lock(&hslot2->lock);
184 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
185 if (net_eq(sock_net(sk2), net) &&
186 sk2 != sk &&
187 (udp_sk(sk2)->udp_port_hash == num) &&
188 (!sk2->sk_reuse || !sk->sk_reuse) &&
189 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
190 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
191 (!sk2->sk_reuseport || !sk->sk_reuseport ||
192 rcu_access_pointer(sk->sk_reuseport_cb) ||
193 !uid_eq(uid, sock_i_uid(sk2))) &&
194 saddr_comp(sk, sk2, true)) {
195 res = 1;
196 break;
197 }
198 }
199 spin_unlock(&hslot2->lock);
200 return res;
201 }
202
203 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot,
204 int (*saddr_same)(const struct sock *sk1,
205 const struct sock *sk2,
206 bool match_wildcard))
207 {
208 struct net *net = sock_net(sk);
209 kuid_t uid = sock_i_uid(sk);
210 struct sock *sk2;
211
212 sk_for_each(sk2, &hslot->head) {
213 if (net_eq(sock_net(sk2), net) &&
214 sk2 != sk &&
215 sk2->sk_family == sk->sk_family &&
216 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
217 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
218 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
219 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
220 (*saddr_same)(sk, sk2, false)) {
221 return reuseport_add_sock(sk, sk2);
222 }
223 }
224
225 /* Initial allocation may have already happened via setsockopt */
226 if (!rcu_access_pointer(sk->sk_reuseport_cb))
227 return reuseport_alloc(sk);
228 return 0;
229 }
230
231 /**
232 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
233 *
234 * @sk: socket struct in question
235 * @snum: port number to look up
236 * @saddr_comp: AF-dependent comparison of bound local IP addresses
237 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
238 * with NULL address
239 */
240 int udp_lib_get_port(struct sock *sk, unsigned short snum,
241 int (*saddr_comp)(const struct sock *sk1,
242 const struct sock *sk2,
243 bool match_wildcard),
244 unsigned int hash2_nulladdr)
245 {
246 struct udp_hslot *hslot, *hslot2;
247 struct udp_table *udptable = sk->sk_prot->h.udp_table;
248 int error = 1;
249 struct net *net = sock_net(sk);
250
251 if (!snum) {
252 int low, high, remaining;
253 unsigned int rand;
254 unsigned short first, last;
255 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
256
257 inet_get_local_port_range(net, &low, &high);
258 remaining = (high - low) + 1;
259
260 rand = prandom_u32();
261 first = reciprocal_scale(rand, remaining) + low;
262 /*
263 * force rand to be an odd multiple of UDP_HTABLE_SIZE
264 */
265 rand = (rand | 1) * (udptable->mask + 1);
266 last = first + udptable->mask + 1;
267 do {
268 hslot = udp_hashslot(udptable, net, first);
269 bitmap_zero(bitmap, PORTS_PER_CHAIN);
270 spin_lock_bh(&hslot->lock);
271 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
272 saddr_comp, udptable->log);
273
274 snum = first;
275 /*
276 * Iterate on all possible values of snum for this hash.
277 * Using steps of an odd multiple of UDP_HTABLE_SIZE
278 * give us randomization and full range coverage.
279 */
280 do {
281 if (low <= snum && snum <= high &&
282 !test_bit(snum >> udptable->log, bitmap) &&
283 !inet_is_local_reserved_port(net, snum))
284 goto found;
285 snum += rand;
286 } while (snum != first);
287 spin_unlock_bh(&hslot->lock);
288 } while (++first != last);
289 goto fail;
290 } else {
291 hslot = udp_hashslot(udptable, net, snum);
292 spin_lock_bh(&hslot->lock);
293 if (hslot->count > 10) {
294 int exist;
295 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
296
297 slot2 &= udptable->mask;
298 hash2_nulladdr &= udptable->mask;
299
300 hslot2 = udp_hashslot2(udptable, slot2);
301 if (hslot->count < hslot2->count)
302 goto scan_primary_hash;
303
304 exist = udp_lib_lport_inuse2(net, snum, hslot2,
305 sk, saddr_comp);
306 if (!exist && (hash2_nulladdr != slot2)) {
307 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
308 exist = udp_lib_lport_inuse2(net, snum, hslot2,
309 sk, saddr_comp);
310 }
311 if (exist)
312 goto fail_unlock;
313 else
314 goto found;
315 }
316 scan_primary_hash:
317 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
318 saddr_comp, 0))
319 goto fail_unlock;
320 }
321 found:
322 inet_sk(sk)->inet_num = snum;
323 udp_sk(sk)->udp_port_hash = snum;
324 udp_sk(sk)->udp_portaddr_hash ^= snum;
325 if (sk_unhashed(sk)) {
326 if (sk->sk_reuseport &&
327 udp_reuseport_add_sock(sk, hslot, saddr_comp)) {
328 inet_sk(sk)->inet_num = 0;
329 udp_sk(sk)->udp_port_hash = 0;
330 udp_sk(sk)->udp_portaddr_hash ^= snum;
331 goto fail_unlock;
332 }
333
334 sk_add_node_rcu(sk, &hslot->head);
335 hslot->count++;
336 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
337
338 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
339 spin_lock(&hslot2->lock);
340 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
341 sk->sk_family == AF_INET6)
342 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
343 &hslot2->head);
344 else
345 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
346 &hslot2->head);
347 hslot2->count++;
348 spin_unlock(&hslot2->lock);
349 }
350 sock_set_flag(sk, SOCK_RCU_FREE);
351 error = 0;
352 fail_unlock:
353 spin_unlock_bh(&hslot->lock);
354 fail:
355 return error;
356 }
357 EXPORT_SYMBOL(udp_lib_get_port);
358
359 /* match_wildcard == true: 0.0.0.0 equals to any IPv4 addresses
360 * match_wildcard == false: addresses must be exactly the same, i.e.
361 * 0.0.0.0 only equals to 0.0.0.0
362 */
363 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2,
364 bool match_wildcard)
365 {
366 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
367
368 if (!ipv6_only_sock(sk2)) {
369 if (inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)
370 return 1;
371 if (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr)
372 return match_wildcard;
373 }
374 return 0;
375 }
376
377 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr,
378 unsigned int port)
379 {
380 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
381 }
382
383 int udp_v4_get_port(struct sock *sk, unsigned short snum)
384 {
385 unsigned int hash2_nulladdr =
386 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
387 unsigned int hash2_partial =
388 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
389
390 /* precompute partial secondary hash */
391 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
392 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
393 }
394
395 static int compute_score(struct sock *sk, struct net *net,
396 __be32 saddr, __be16 sport,
397 __be32 daddr, unsigned short hnum, int dif)
398 {
399 int score;
400 struct inet_sock *inet;
401
402 if (!net_eq(sock_net(sk), net) ||
403 udp_sk(sk)->udp_port_hash != hnum ||
404 ipv6_only_sock(sk))
405 return -1;
406
407 score = (sk->sk_family == PF_INET) ? 2 : 1;
408 inet = inet_sk(sk);
409
410 if (inet->inet_rcv_saddr) {
411 if (inet->inet_rcv_saddr != daddr)
412 return -1;
413 score += 4;
414 }
415
416 if (inet->inet_daddr) {
417 if (inet->inet_daddr != saddr)
418 return -1;
419 score += 4;
420 }
421
422 if (inet->inet_dport) {
423 if (inet->inet_dport != sport)
424 return -1;
425 score += 4;
426 }
427
428 if (sk->sk_bound_dev_if) {
429 if (sk->sk_bound_dev_if != dif)
430 return -1;
431 score += 4;
432 }
433 if (sk->sk_incoming_cpu == raw_smp_processor_id())
434 score++;
435 return score;
436 }
437
438 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
439 const __u16 lport, const __be32 faddr,
440 const __be16 fport)
441 {
442 static u32 udp_ehash_secret __read_mostly;
443
444 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
445
446 return __inet_ehashfn(laddr, lport, faddr, fport,
447 udp_ehash_secret + net_hash_mix(net));
448 }
449
450 /* called with rcu_read_lock() */
451 static struct sock *udp4_lib_lookup2(struct net *net,
452 __be32 saddr, __be16 sport,
453 __be32 daddr, unsigned int hnum, int dif,
454 struct udp_hslot *hslot2,
455 struct sk_buff *skb)
456 {
457 struct sock *sk, *result;
458 int score, badness, matches = 0, reuseport = 0;
459 u32 hash = 0;
460
461 result = NULL;
462 badness = 0;
463 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
464 score = compute_score(sk, net, saddr, sport,
465 daddr, hnum, dif);
466 if (score > badness) {
467 reuseport = sk->sk_reuseport;
468 if (reuseport) {
469 hash = udp_ehashfn(net, daddr, hnum,
470 saddr, sport);
471 result = reuseport_select_sock(sk, hash, skb,
472 sizeof(struct udphdr));
473 if (result)
474 return result;
475 matches = 1;
476 }
477 badness = score;
478 result = sk;
479 } else if (score == badness && reuseport) {
480 matches++;
481 if (reciprocal_scale(hash, matches) == 0)
482 result = sk;
483 hash = next_pseudo_random32(hash);
484 }
485 }
486 return result;
487 }
488
489 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
490 * harder than this. -DaveM
491 */
492 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
493 __be16 sport, __be32 daddr, __be16 dport,
494 int dif, struct udp_table *udptable, struct sk_buff *skb)
495 {
496 struct sock *sk, *result;
497 unsigned short hnum = ntohs(dport);
498 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
499 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
500 int score, badness, matches = 0, reuseport = 0;
501 u32 hash = 0;
502
503 if (hslot->count > 10) {
504 hash2 = udp4_portaddr_hash(net, daddr, hnum);
505 slot2 = hash2 & udptable->mask;
506 hslot2 = &udptable->hash2[slot2];
507 if (hslot->count < hslot2->count)
508 goto begin;
509
510 result = udp4_lib_lookup2(net, saddr, sport,
511 daddr, hnum, dif,
512 hslot2, skb);
513 if (!result) {
514 unsigned int old_slot2 = slot2;
515 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
516 slot2 = hash2 & udptable->mask;
517 /* avoid searching the same slot again. */
518 if (unlikely(slot2 == old_slot2))
519 return result;
520
521 hslot2 = &udptable->hash2[slot2];
522 if (hslot->count < hslot2->count)
523 goto begin;
524
525 result = udp4_lib_lookup2(net, saddr, sport,
526 daddr, hnum, dif,
527 hslot2, skb);
528 }
529 return result;
530 }
531 begin:
532 result = NULL;
533 badness = 0;
534 sk_for_each_rcu(sk, &hslot->head) {
535 score = compute_score(sk, net, saddr, sport,
536 daddr, hnum, dif);
537 if (score > badness) {
538 reuseport = sk->sk_reuseport;
539 if (reuseport) {
540 hash = udp_ehashfn(net, daddr, hnum,
541 saddr, sport);
542 result = reuseport_select_sock(sk, hash, skb,
543 sizeof(struct udphdr));
544 if (result)
545 return result;
546 matches = 1;
547 }
548 result = sk;
549 badness = score;
550 } else if (score == badness && reuseport) {
551 matches++;
552 if (reciprocal_scale(hash, matches) == 0)
553 result = sk;
554 hash = next_pseudo_random32(hash);
555 }
556 }
557 return result;
558 }
559 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
560
561 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
562 __be16 sport, __be16 dport,
563 struct udp_table *udptable)
564 {
565 const struct iphdr *iph = ip_hdr(skb);
566
567 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
568 iph->daddr, dport, inet_iif(skb),
569 udptable, skb);
570 }
571
572 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
573 __be16 sport, __be16 dport)
574 {
575 return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table);
576 }
577 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
578
579 /* Must be called under rcu_read_lock().
580 * Does increment socket refcount.
581 */
582 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
583 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY)
584 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
585 __be32 daddr, __be16 dport, int dif)
586 {
587 struct sock *sk;
588
589 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
590 dif, &udp_table, NULL);
591 if (sk && !atomic_inc_not_zero(&sk->sk_refcnt))
592 sk = NULL;
593 return sk;
594 }
595 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
596 #endif
597
598 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
599 __be16 loc_port, __be32 loc_addr,
600 __be16 rmt_port, __be32 rmt_addr,
601 int dif, unsigned short hnum)
602 {
603 struct inet_sock *inet = inet_sk(sk);
604
605 if (!net_eq(sock_net(sk), net) ||
606 udp_sk(sk)->udp_port_hash != hnum ||
607 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
608 (inet->inet_dport != rmt_port && inet->inet_dport) ||
609 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
610 ipv6_only_sock(sk) ||
611 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
612 return false;
613 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif))
614 return false;
615 return true;
616 }
617
618 /*
619 * This routine is called by the ICMP module when it gets some
620 * sort of error condition. If err < 0 then the socket should
621 * be closed and the error returned to the user. If err > 0
622 * it's just the icmp type << 8 | icmp code.
623 * Header points to the ip header of the error packet. We move
624 * on past this. Then (as it used to claim before adjustment)
625 * header points to the first 8 bytes of the udp header. We need
626 * to find the appropriate port.
627 */
628
629 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
630 {
631 struct inet_sock *inet;
632 const struct iphdr *iph = (const struct iphdr *)skb->data;
633 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
634 const int type = icmp_hdr(skb)->type;
635 const int code = icmp_hdr(skb)->code;
636 struct sock *sk;
637 int harderr;
638 int err;
639 struct net *net = dev_net(skb->dev);
640
641 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
642 iph->saddr, uh->source, skb->dev->ifindex, udptable,
643 NULL);
644 if (!sk) {
645 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
646 return; /* No socket for error */
647 }
648
649 err = 0;
650 harderr = 0;
651 inet = inet_sk(sk);
652
653 switch (type) {
654 default:
655 case ICMP_TIME_EXCEEDED:
656 err = EHOSTUNREACH;
657 break;
658 case ICMP_SOURCE_QUENCH:
659 goto out;
660 case ICMP_PARAMETERPROB:
661 err = EPROTO;
662 harderr = 1;
663 break;
664 case ICMP_DEST_UNREACH:
665 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
666 ipv4_sk_update_pmtu(skb, sk, info);
667 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
668 err = EMSGSIZE;
669 harderr = 1;
670 break;
671 }
672 goto out;
673 }
674 err = EHOSTUNREACH;
675 if (code <= NR_ICMP_UNREACH) {
676 harderr = icmp_err_convert[code].fatal;
677 err = icmp_err_convert[code].errno;
678 }
679 break;
680 case ICMP_REDIRECT:
681 ipv4_sk_redirect(skb, sk);
682 goto out;
683 }
684
685 /*
686 * RFC1122: OK. Passes ICMP errors back to application, as per
687 * 4.1.3.3.
688 */
689 if (!inet->recverr) {
690 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
691 goto out;
692 } else
693 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
694
695 sk->sk_err = err;
696 sk->sk_error_report(sk);
697 out:
698 return;
699 }
700
701 void udp_err(struct sk_buff *skb, u32 info)
702 {
703 __udp4_lib_err(skb, info, &udp_table);
704 }
705
706 /*
707 * Throw away all pending data and cancel the corking. Socket is locked.
708 */
709 void udp_flush_pending_frames(struct sock *sk)
710 {
711 struct udp_sock *up = udp_sk(sk);
712
713 if (up->pending) {
714 up->len = 0;
715 up->pending = 0;
716 ip_flush_pending_frames(sk);
717 }
718 }
719 EXPORT_SYMBOL(udp_flush_pending_frames);
720
721 /**
722 * udp4_hwcsum - handle outgoing HW checksumming
723 * @skb: sk_buff containing the filled-in UDP header
724 * (checksum field must be zeroed out)
725 * @src: source IP address
726 * @dst: destination IP address
727 */
728 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
729 {
730 struct udphdr *uh = udp_hdr(skb);
731 int offset = skb_transport_offset(skb);
732 int len = skb->len - offset;
733 int hlen = len;
734 __wsum csum = 0;
735
736 if (!skb_has_frag_list(skb)) {
737 /*
738 * Only one fragment on the socket.
739 */
740 skb->csum_start = skb_transport_header(skb) - skb->head;
741 skb->csum_offset = offsetof(struct udphdr, check);
742 uh->check = ~csum_tcpudp_magic(src, dst, len,
743 IPPROTO_UDP, 0);
744 } else {
745 struct sk_buff *frags;
746
747 /*
748 * HW-checksum won't work as there are two or more
749 * fragments on the socket so that all csums of sk_buffs
750 * should be together
751 */
752 skb_walk_frags(skb, frags) {
753 csum = csum_add(csum, frags->csum);
754 hlen -= frags->len;
755 }
756
757 csum = skb_checksum(skb, offset, hlen, csum);
758 skb->ip_summed = CHECKSUM_NONE;
759
760 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
761 if (uh->check == 0)
762 uh->check = CSUM_MANGLED_0;
763 }
764 }
765 EXPORT_SYMBOL_GPL(udp4_hwcsum);
766
767 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
768 * for the simple case like when setting the checksum for a UDP tunnel.
769 */
770 void udp_set_csum(bool nocheck, struct sk_buff *skb,
771 __be32 saddr, __be32 daddr, int len)
772 {
773 struct udphdr *uh = udp_hdr(skb);
774
775 if (nocheck) {
776 uh->check = 0;
777 } else if (skb_is_gso(skb)) {
778 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
779 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
780 uh->check = 0;
781 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
782 if (uh->check == 0)
783 uh->check = CSUM_MANGLED_0;
784 } else {
785 skb->ip_summed = CHECKSUM_PARTIAL;
786 skb->csum_start = skb_transport_header(skb) - skb->head;
787 skb->csum_offset = offsetof(struct udphdr, check);
788 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
789 }
790 }
791 EXPORT_SYMBOL(udp_set_csum);
792
793 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
794 {
795 struct sock *sk = skb->sk;
796 struct inet_sock *inet = inet_sk(sk);
797 struct udphdr *uh;
798 int err = 0;
799 int is_udplite = IS_UDPLITE(sk);
800 int offset = skb_transport_offset(skb);
801 int len = skb->len - offset;
802 __wsum csum = 0;
803
804 /*
805 * Create a UDP header
806 */
807 uh = udp_hdr(skb);
808 uh->source = inet->inet_sport;
809 uh->dest = fl4->fl4_dport;
810 uh->len = htons(len);
811 uh->check = 0;
812
813 if (is_udplite) /* UDP-Lite */
814 csum = udplite_csum(skb);
815
816 else if (sk->sk_no_check_tx) { /* UDP csum disabled */
817
818 skb->ip_summed = CHECKSUM_NONE;
819 goto send;
820
821 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
822
823 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
824 goto send;
825
826 } else
827 csum = udp_csum(skb);
828
829 /* add protocol-dependent pseudo-header */
830 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
831 sk->sk_protocol, csum);
832 if (uh->check == 0)
833 uh->check = CSUM_MANGLED_0;
834
835 send:
836 err = ip_send_skb(sock_net(sk), skb);
837 if (err) {
838 if (err == -ENOBUFS && !inet->recverr) {
839 UDP_INC_STATS(sock_net(sk),
840 UDP_MIB_SNDBUFERRORS, is_udplite);
841 err = 0;
842 }
843 } else
844 UDP_INC_STATS(sock_net(sk),
845 UDP_MIB_OUTDATAGRAMS, is_udplite);
846 return err;
847 }
848
849 /*
850 * Push out all pending data as one UDP datagram. Socket is locked.
851 */
852 int udp_push_pending_frames(struct sock *sk)
853 {
854 struct udp_sock *up = udp_sk(sk);
855 struct inet_sock *inet = inet_sk(sk);
856 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
857 struct sk_buff *skb;
858 int err = 0;
859
860 skb = ip_finish_skb(sk, fl4);
861 if (!skb)
862 goto out;
863
864 err = udp_send_skb(skb, fl4);
865
866 out:
867 up->len = 0;
868 up->pending = 0;
869 return err;
870 }
871 EXPORT_SYMBOL(udp_push_pending_frames);
872
873 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
874 {
875 struct inet_sock *inet = inet_sk(sk);
876 struct udp_sock *up = udp_sk(sk);
877 struct flowi4 fl4_stack;
878 struct flowi4 *fl4;
879 int ulen = len;
880 struct ipcm_cookie ipc;
881 struct rtable *rt = NULL;
882 int free = 0;
883 int connected = 0;
884 __be32 daddr, faddr, saddr;
885 __be16 dport;
886 u8 tos;
887 int err, is_udplite = IS_UDPLITE(sk);
888 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
889 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
890 struct sk_buff *skb;
891 struct ip_options_data opt_copy;
892
893 if (len > 0xFFFF)
894 return -EMSGSIZE;
895
896 /*
897 * Check the flags.
898 */
899
900 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
901 return -EOPNOTSUPP;
902
903 ipc.opt = NULL;
904 ipc.tx_flags = 0;
905 ipc.ttl = 0;
906 ipc.tos = -1;
907
908 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
909
910 fl4 = &inet->cork.fl.u.ip4;
911 if (up->pending) {
912 /*
913 * There are pending frames.
914 * The socket lock must be held while it's corked.
915 */
916 lock_sock(sk);
917 if (likely(up->pending)) {
918 if (unlikely(up->pending != AF_INET)) {
919 release_sock(sk);
920 return -EINVAL;
921 }
922 goto do_append_data;
923 }
924 release_sock(sk);
925 }
926 ulen += sizeof(struct udphdr);
927
928 /*
929 * Get and verify the address.
930 */
931 if (msg->msg_name) {
932 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
933 if (msg->msg_namelen < sizeof(*usin))
934 return -EINVAL;
935 if (usin->sin_family != AF_INET) {
936 if (usin->sin_family != AF_UNSPEC)
937 return -EAFNOSUPPORT;
938 }
939
940 daddr = usin->sin_addr.s_addr;
941 dport = usin->sin_port;
942 if (dport == 0)
943 return -EINVAL;
944 } else {
945 if (sk->sk_state != TCP_ESTABLISHED)
946 return -EDESTADDRREQ;
947 daddr = inet->inet_daddr;
948 dport = inet->inet_dport;
949 /* Open fast path for connected socket.
950 Route will not be used, if at least one option is set.
951 */
952 connected = 1;
953 }
954
955 ipc.sockc.tsflags = sk->sk_tsflags;
956 ipc.addr = inet->inet_saddr;
957 ipc.oif = sk->sk_bound_dev_if;
958
959 if (msg->msg_controllen) {
960 err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6);
961 if (unlikely(err)) {
962 kfree(ipc.opt);
963 return err;
964 }
965 if (ipc.opt)
966 free = 1;
967 connected = 0;
968 }
969 if (!ipc.opt) {
970 struct ip_options_rcu *inet_opt;
971
972 rcu_read_lock();
973 inet_opt = rcu_dereference(inet->inet_opt);
974 if (inet_opt) {
975 memcpy(&opt_copy, inet_opt,
976 sizeof(*inet_opt) + inet_opt->opt.optlen);
977 ipc.opt = &opt_copy.opt;
978 }
979 rcu_read_unlock();
980 }
981
982 saddr = ipc.addr;
983 ipc.addr = faddr = daddr;
984
985 sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags);
986
987 if (ipc.opt && ipc.opt->opt.srr) {
988 if (!daddr)
989 return -EINVAL;
990 faddr = ipc.opt->opt.faddr;
991 connected = 0;
992 }
993 tos = get_rttos(&ipc, inet);
994 if (sock_flag(sk, SOCK_LOCALROUTE) ||
995 (msg->msg_flags & MSG_DONTROUTE) ||
996 (ipc.opt && ipc.opt->opt.is_strictroute)) {
997 tos |= RTO_ONLINK;
998 connected = 0;
999 }
1000
1001 if (ipv4_is_multicast(daddr)) {
1002 if (!ipc.oif)
1003 ipc.oif = inet->mc_index;
1004 if (!saddr)
1005 saddr = inet->mc_addr;
1006 connected = 0;
1007 } else if (!ipc.oif)
1008 ipc.oif = inet->uc_index;
1009
1010 if (connected)
1011 rt = (struct rtable *)sk_dst_check(sk, 0);
1012
1013 if (!rt) {
1014 struct net *net = sock_net(sk);
1015 __u8 flow_flags = inet_sk_flowi_flags(sk);
1016
1017 fl4 = &fl4_stack;
1018
1019 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1020 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1021 flow_flags,
1022 faddr, saddr, dport, inet->inet_sport);
1023
1024 if (!saddr && ipc.oif) {
1025 err = l3mdev_get_saddr(net, ipc.oif, fl4);
1026 if (err < 0)
1027 goto out;
1028 }
1029
1030 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1031 rt = ip_route_output_flow(net, fl4, sk);
1032 if (IS_ERR(rt)) {
1033 err = PTR_ERR(rt);
1034 rt = NULL;
1035 if (err == -ENETUNREACH)
1036 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1037 goto out;
1038 }
1039
1040 err = -EACCES;
1041 if ((rt->rt_flags & RTCF_BROADCAST) &&
1042 !sock_flag(sk, SOCK_BROADCAST))
1043 goto out;
1044 if (connected)
1045 sk_dst_set(sk, dst_clone(&rt->dst));
1046 }
1047
1048 if (msg->msg_flags&MSG_CONFIRM)
1049 goto do_confirm;
1050 back_from_confirm:
1051
1052 saddr = fl4->saddr;
1053 if (!ipc.addr)
1054 daddr = ipc.addr = fl4->daddr;
1055
1056 /* Lockless fast path for the non-corking case. */
1057 if (!corkreq) {
1058 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1059 sizeof(struct udphdr), &ipc, &rt,
1060 msg->msg_flags);
1061 err = PTR_ERR(skb);
1062 if (!IS_ERR_OR_NULL(skb))
1063 err = udp_send_skb(skb, fl4);
1064 goto out;
1065 }
1066
1067 lock_sock(sk);
1068 if (unlikely(up->pending)) {
1069 /* The socket is already corked while preparing it. */
1070 /* ... which is an evident application bug. --ANK */
1071 release_sock(sk);
1072
1073 net_dbg_ratelimited("cork app bug 2\n");
1074 err = -EINVAL;
1075 goto out;
1076 }
1077 /*
1078 * Now cork the socket to pend data.
1079 */
1080 fl4 = &inet->cork.fl.u.ip4;
1081 fl4->daddr = daddr;
1082 fl4->saddr = saddr;
1083 fl4->fl4_dport = dport;
1084 fl4->fl4_sport = inet->inet_sport;
1085 up->pending = AF_INET;
1086
1087 do_append_data:
1088 up->len += ulen;
1089 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1090 sizeof(struct udphdr), &ipc, &rt,
1091 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1092 if (err)
1093 udp_flush_pending_frames(sk);
1094 else if (!corkreq)
1095 err = udp_push_pending_frames(sk);
1096 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1097 up->pending = 0;
1098 release_sock(sk);
1099
1100 out:
1101 ip_rt_put(rt);
1102 if (free)
1103 kfree(ipc.opt);
1104 if (!err)
1105 return len;
1106 /*
1107 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1108 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1109 * we don't have a good statistic (IpOutDiscards but it can be too many
1110 * things). We could add another new stat but at least for now that
1111 * seems like overkill.
1112 */
1113 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1114 UDP_INC_STATS(sock_net(sk),
1115 UDP_MIB_SNDBUFERRORS, is_udplite);
1116 }
1117 return err;
1118
1119 do_confirm:
1120 dst_confirm(&rt->dst);
1121 if (!(msg->msg_flags&MSG_PROBE) || len)
1122 goto back_from_confirm;
1123 err = 0;
1124 goto out;
1125 }
1126 EXPORT_SYMBOL(udp_sendmsg);
1127
1128 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1129 size_t size, int flags)
1130 {
1131 struct inet_sock *inet = inet_sk(sk);
1132 struct udp_sock *up = udp_sk(sk);
1133 int ret;
1134
1135 if (flags & MSG_SENDPAGE_NOTLAST)
1136 flags |= MSG_MORE;
1137
1138 if (!up->pending) {
1139 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1140
1141 /* Call udp_sendmsg to specify destination address which
1142 * sendpage interface can't pass.
1143 * This will succeed only when the socket is connected.
1144 */
1145 ret = udp_sendmsg(sk, &msg, 0);
1146 if (ret < 0)
1147 return ret;
1148 }
1149
1150 lock_sock(sk);
1151
1152 if (unlikely(!up->pending)) {
1153 release_sock(sk);
1154
1155 net_dbg_ratelimited("udp cork app bug 3\n");
1156 return -EINVAL;
1157 }
1158
1159 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1160 page, offset, size, flags);
1161 if (ret == -EOPNOTSUPP) {
1162 release_sock(sk);
1163 return sock_no_sendpage(sk->sk_socket, page, offset,
1164 size, flags);
1165 }
1166 if (ret < 0) {
1167 udp_flush_pending_frames(sk);
1168 goto out;
1169 }
1170
1171 up->len += size;
1172 if (!(up->corkflag || (flags&MSG_MORE)))
1173 ret = udp_push_pending_frames(sk);
1174 if (!ret)
1175 ret = size;
1176 out:
1177 release_sock(sk);
1178 return ret;
1179 }
1180
1181 /**
1182 * first_packet_length - return length of first packet in receive queue
1183 * @sk: socket
1184 *
1185 * Drops all bad checksum frames, until a valid one is found.
1186 * Returns the length of found skb, or 0 if none is found.
1187 */
1188 static unsigned int first_packet_length(struct sock *sk)
1189 {
1190 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1191 struct sk_buff *skb;
1192 unsigned int res;
1193
1194 __skb_queue_head_init(&list_kill);
1195
1196 spin_lock_bh(&rcvq->lock);
1197 while ((skb = skb_peek(rcvq)) != NULL &&
1198 udp_lib_checksum_complete(skb)) {
1199 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1200 IS_UDPLITE(sk));
1201 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1202 IS_UDPLITE(sk));
1203 atomic_inc(&sk->sk_drops);
1204 __skb_unlink(skb, rcvq);
1205 __skb_queue_tail(&list_kill, skb);
1206 }
1207 res = skb ? skb->len : 0;
1208 spin_unlock_bh(&rcvq->lock);
1209
1210 if (!skb_queue_empty(&list_kill)) {
1211 bool slow = lock_sock_fast(sk);
1212
1213 __skb_queue_purge(&list_kill);
1214 sk_mem_reclaim_partial(sk);
1215 unlock_sock_fast(sk, slow);
1216 }
1217 return res;
1218 }
1219
1220 /*
1221 * IOCTL requests applicable to the UDP protocol
1222 */
1223
1224 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1225 {
1226 switch (cmd) {
1227 case SIOCOUTQ:
1228 {
1229 int amount = sk_wmem_alloc_get(sk);
1230
1231 return put_user(amount, (int __user *)arg);
1232 }
1233
1234 case SIOCINQ:
1235 {
1236 unsigned int amount = first_packet_length(sk);
1237
1238 return put_user(amount, (int __user *)arg);
1239 }
1240
1241 default:
1242 return -ENOIOCTLCMD;
1243 }
1244
1245 return 0;
1246 }
1247 EXPORT_SYMBOL(udp_ioctl);
1248
1249 /*
1250 * This should be easy, if there is something there we
1251 * return it, otherwise we block.
1252 */
1253
1254 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1255 int flags, int *addr_len)
1256 {
1257 struct inet_sock *inet = inet_sk(sk);
1258 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1259 struct sk_buff *skb;
1260 unsigned int ulen, copied;
1261 int peeked, peeking, off;
1262 int err;
1263 int is_udplite = IS_UDPLITE(sk);
1264 bool checksum_valid = false;
1265 bool slow;
1266
1267 if (flags & MSG_ERRQUEUE)
1268 return ip_recv_error(sk, msg, len, addr_len);
1269
1270 try_again:
1271 peeking = off = sk_peek_offset(sk, flags);
1272 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1273 &peeked, &off, &err);
1274 if (!skb)
1275 return err;
1276
1277 ulen = skb->len;
1278 copied = len;
1279 if (copied > ulen - off)
1280 copied = ulen - off;
1281 else if (copied < ulen)
1282 msg->msg_flags |= MSG_TRUNC;
1283
1284 /*
1285 * If checksum is needed at all, try to do it while copying the
1286 * data. If the data is truncated, or if we only want a partial
1287 * coverage checksum (UDP-Lite), do it before the copy.
1288 */
1289
1290 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov || peeking) {
1291 checksum_valid = !udp_lib_checksum_complete(skb);
1292 if (!checksum_valid)
1293 goto csum_copy_err;
1294 }
1295
1296 if (checksum_valid || skb_csum_unnecessary(skb))
1297 err = skb_copy_datagram_msg(skb, off, msg, copied);
1298 else {
1299 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1300
1301 if (err == -EINVAL)
1302 goto csum_copy_err;
1303 }
1304
1305 if (unlikely(err)) {
1306 trace_kfree_skb(skb, udp_recvmsg);
1307 if (!peeked) {
1308 atomic_inc(&sk->sk_drops);
1309 UDP_INC_STATS(sock_net(sk),
1310 UDP_MIB_INERRORS, is_udplite);
1311 }
1312 skb_free_datagram_locked(sk, skb);
1313 return err;
1314 }
1315
1316 if (!peeked)
1317 UDP_INC_STATS(sock_net(sk),
1318 UDP_MIB_INDATAGRAMS, is_udplite);
1319
1320 sock_recv_ts_and_drops(msg, sk, skb);
1321
1322 /* Copy the address. */
1323 if (sin) {
1324 sin->sin_family = AF_INET;
1325 sin->sin_port = udp_hdr(skb)->source;
1326 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1327 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1328 *addr_len = sizeof(*sin);
1329 }
1330 if (inet->cmsg_flags)
1331 ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr) + off);
1332
1333 err = copied;
1334 if (flags & MSG_TRUNC)
1335 err = ulen;
1336
1337 __skb_free_datagram_locked(sk, skb, peeking ? -err : err);
1338 return err;
1339
1340 csum_copy_err:
1341 slow = lock_sock_fast(sk);
1342 if (!skb_kill_datagram(sk, skb, flags)) {
1343 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1344 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1345 }
1346 unlock_sock_fast(sk, slow);
1347
1348 /* starting over for a new packet, but check if we need to yield */
1349 cond_resched();
1350 msg->msg_flags &= ~MSG_TRUNC;
1351 goto try_again;
1352 }
1353
1354 int udp_disconnect(struct sock *sk, int flags)
1355 {
1356 struct inet_sock *inet = inet_sk(sk);
1357 /*
1358 * 1003.1g - break association.
1359 */
1360
1361 sk->sk_state = TCP_CLOSE;
1362 inet->inet_daddr = 0;
1363 inet->inet_dport = 0;
1364 sock_rps_reset_rxhash(sk);
1365 sk->sk_bound_dev_if = 0;
1366 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1367 inet_reset_saddr(sk);
1368
1369 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1370 sk->sk_prot->unhash(sk);
1371 inet->inet_sport = 0;
1372 }
1373 sk_dst_reset(sk);
1374 return 0;
1375 }
1376 EXPORT_SYMBOL(udp_disconnect);
1377
1378 void udp_lib_unhash(struct sock *sk)
1379 {
1380 if (sk_hashed(sk)) {
1381 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1382 struct udp_hslot *hslot, *hslot2;
1383
1384 hslot = udp_hashslot(udptable, sock_net(sk),
1385 udp_sk(sk)->udp_port_hash);
1386 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1387
1388 spin_lock_bh(&hslot->lock);
1389 if (rcu_access_pointer(sk->sk_reuseport_cb))
1390 reuseport_detach_sock(sk);
1391 if (sk_del_node_init_rcu(sk)) {
1392 hslot->count--;
1393 inet_sk(sk)->inet_num = 0;
1394 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1395
1396 spin_lock(&hslot2->lock);
1397 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1398 hslot2->count--;
1399 spin_unlock(&hslot2->lock);
1400 }
1401 spin_unlock_bh(&hslot->lock);
1402 }
1403 }
1404 EXPORT_SYMBOL(udp_lib_unhash);
1405
1406 /*
1407 * inet_rcv_saddr was changed, we must rehash secondary hash
1408 */
1409 void udp_lib_rehash(struct sock *sk, u16 newhash)
1410 {
1411 if (sk_hashed(sk)) {
1412 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1413 struct udp_hslot *hslot, *hslot2, *nhslot2;
1414
1415 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1416 nhslot2 = udp_hashslot2(udptable, newhash);
1417 udp_sk(sk)->udp_portaddr_hash = newhash;
1418
1419 if (hslot2 != nhslot2 ||
1420 rcu_access_pointer(sk->sk_reuseport_cb)) {
1421 hslot = udp_hashslot(udptable, sock_net(sk),
1422 udp_sk(sk)->udp_port_hash);
1423 /* we must lock primary chain too */
1424 spin_lock_bh(&hslot->lock);
1425 if (rcu_access_pointer(sk->sk_reuseport_cb))
1426 reuseport_detach_sock(sk);
1427
1428 if (hslot2 != nhslot2) {
1429 spin_lock(&hslot2->lock);
1430 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1431 hslot2->count--;
1432 spin_unlock(&hslot2->lock);
1433
1434 spin_lock(&nhslot2->lock);
1435 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1436 &nhslot2->head);
1437 nhslot2->count++;
1438 spin_unlock(&nhslot2->lock);
1439 }
1440
1441 spin_unlock_bh(&hslot->lock);
1442 }
1443 }
1444 }
1445 EXPORT_SYMBOL(udp_lib_rehash);
1446
1447 static void udp_v4_rehash(struct sock *sk)
1448 {
1449 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1450 inet_sk(sk)->inet_rcv_saddr,
1451 inet_sk(sk)->inet_num);
1452 udp_lib_rehash(sk, new_hash);
1453 }
1454
1455 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1456 {
1457 int rc;
1458
1459 if (inet_sk(sk)->inet_daddr) {
1460 sock_rps_save_rxhash(sk, skb);
1461 sk_mark_napi_id(sk, skb);
1462 sk_incoming_cpu_update(sk);
1463 }
1464
1465 rc = __sock_queue_rcv_skb(sk, skb);
1466 if (rc < 0) {
1467 int is_udplite = IS_UDPLITE(sk);
1468
1469 /* Note that an ENOMEM error is charged twice */
1470 if (rc == -ENOMEM)
1471 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1472 is_udplite);
1473 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1474 kfree_skb(skb);
1475 trace_udp_fail_queue_rcv_skb(rc, sk);
1476 return -1;
1477 }
1478
1479 return 0;
1480
1481 }
1482
1483 static struct static_key udp_encap_needed __read_mostly;
1484 void udp_encap_enable(void)
1485 {
1486 if (!static_key_enabled(&udp_encap_needed))
1487 static_key_slow_inc(&udp_encap_needed);
1488 }
1489 EXPORT_SYMBOL(udp_encap_enable);
1490
1491 /* returns:
1492 * -1: error
1493 * 0: success
1494 * >0: "udp encap" protocol resubmission
1495 *
1496 * Note that in the success and error cases, the skb is assumed to
1497 * have either been requeued or freed.
1498 */
1499 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1500 {
1501 struct udp_sock *up = udp_sk(sk);
1502 int rc;
1503 int is_udplite = IS_UDPLITE(sk);
1504
1505 /*
1506 * Charge it to the socket, dropping if the queue is full.
1507 */
1508 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1509 goto drop;
1510 nf_reset(skb);
1511
1512 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1513 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1514
1515 /*
1516 * This is an encapsulation socket so pass the skb to
1517 * the socket's udp_encap_rcv() hook. Otherwise, just
1518 * fall through and pass this up the UDP socket.
1519 * up->encap_rcv() returns the following value:
1520 * =0 if skb was successfully passed to the encap
1521 * handler or was discarded by it.
1522 * >0 if skb should be passed on to UDP.
1523 * <0 if skb should be resubmitted as proto -N
1524 */
1525
1526 /* if we're overly short, let UDP handle it */
1527 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1528 if (encap_rcv) {
1529 int ret;
1530
1531 /* Verify checksum before giving to encap */
1532 if (udp_lib_checksum_complete(skb))
1533 goto csum_error;
1534
1535 ret = encap_rcv(sk, skb);
1536 if (ret <= 0) {
1537 __UDP_INC_STATS(sock_net(sk),
1538 UDP_MIB_INDATAGRAMS,
1539 is_udplite);
1540 return -ret;
1541 }
1542 }
1543
1544 /* FALLTHROUGH -- it's a UDP Packet */
1545 }
1546
1547 /*
1548 * UDP-Lite specific tests, ignored on UDP sockets
1549 */
1550 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1551
1552 /*
1553 * MIB statistics other than incrementing the error count are
1554 * disabled for the following two types of errors: these depend
1555 * on the application settings, not on the functioning of the
1556 * protocol stack as such.
1557 *
1558 * RFC 3828 here recommends (sec 3.3): "There should also be a
1559 * way ... to ... at least let the receiving application block
1560 * delivery of packets with coverage values less than a value
1561 * provided by the application."
1562 */
1563 if (up->pcrlen == 0) { /* full coverage was set */
1564 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1565 UDP_SKB_CB(skb)->cscov, skb->len);
1566 goto drop;
1567 }
1568 /* The next case involves violating the min. coverage requested
1569 * by the receiver. This is subtle: if receiver wants x and x is
1570 * greater than the buffersize/MTU then receiver will complain
1571 * that it wants x while sender emits packets of smaller size y.
1572 * Therefore the above ...()->partial_cov statement is essential.
1573 */
1574 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1575 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1576 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1577 goto drop;
1578 }
1579 }
1580
1581 if (rcu_access_pointer(sk->sk_filter) &&
1582 udp_lib_checksum_complete(skb))
1583 goto csum_error;
1584
1585 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
1586 goto drop;
1587
1588 udp_csum_pull_header(skb);
1589 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
1590 __UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1591 is_udplite);
1592 goto drop;
1593 }
1594
1595 rc = 0;
1596
1597 ipv4_pktinfo_prepare(sk, skb);
1598 bh_lock_sock(sk);
1599 if (!sock_owned_by_user(sk))
1600 rc = __udp_queue_rcv_skb(sk, skb);
1601 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1602 bh_unlock_sock(sk);
1603 goto drop;
1604 }
1605 bh_unlock_sock(sk);
1606
1607 return rc;
1608
1609 csum_error:
1610 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1611 drop:
1612 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1613 atomic_inc(&sk->sk_drops);
1614 kfree_skb(skb);
1615 return -1;
1616 }
1617
1618 /* For TCP sockets, sk_rx_dst is protected by socket lock
1619 * For UDP, we use xchg() to guard against concurrent changes.
1620 */
1621 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1622 {
1623 struct dst_entry *old;
1624
1625 dst_hold(dst);
1626 old = xchg(&sk->sk_rx_dst, dst);
1627 dst_release(old);
1628 }
1629
1630 /*
1631 * Multicasts and broadcasts go to each listener.
1632 *
1633 * Note: called only from the BH handler context.
1634 */
1635 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1636 struct udphdr *uh,
1637 __be32 saddr, __be32 daddr,
1638 struct udp_table *udptable,
1639 int proto)
1640 {
1641 struct sock *sk, *first = NULL;
1642 unsigned short hnum = ntohs(uh->dest);
1643 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1644 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1645 unsigned int offset = offsetof(typeof(*sk), sk_node);
1646 int dif = skb->dev->ifindex;
1647 struct hlist_node *node;
1648 struct sk_buff *nskb;
1649
1650 if (use_hash2) {
1651 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1652 udp_table.mask;
1653 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask;
1654 start_lookup:
1655 hslot = &udp_table.hash2[hash2];
1656 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1657 }
1658
1659 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
1660 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
1661 uh->source, saddr, dif, hnum))
1662 continue;
1663
1664 if (!first) {
1665 first = sk;
1666 continue;
1667 }
1668 nskb = skb_clone(skb, GFP_ATOMIC);
1669
1670 if (unlikely(!nskb)) {
1671 atomic_inc(&sk->sk_drops);
1672 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
1673 IS_UDPLITE(sk));
1674 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
1675 IS_UDPLITE(sk));
1676 continue;
1677 }
1678 if (udp_queue_rcv_skb(sk, nskb) > 0)
1679 consume_skb(nskb);
1680 }
1681
1682 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
1683 if (use_hash2 && hash2 != hash2_any) {
1684 hash2 = hash2_any;
1685 goto start_lookup;
1686 }
1687
1688 if (first) {
1689 if (udp_queue_rcv_skb(first, skb) > 0)
1690 consume_skb(skb);
1691 } else {
1692 kfree_skb(skb);
1693 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
1694 proto == IPPROTO_UDPLITE);
1695 }
1696 return 0;
1697 }
1698
1699 /* Initialize UDP checksum. If exited with zero value (success),
1700 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1701 * Otherwise, csum completion requires chacksumming packet body,
1702 * including udp header and folding it to skb->csum.
1703 */
1704 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1705 int proto)
1706 {
1707 int err;
1708
1709 UDP_SKB_CB(skb)->partial_cov = 0;
1710 UDP_SKB_CB(skb)->cscov = skb->len;
1711
1712 if (proto == IPPROTO_UDPLITE) {
1713 err = udplite_checksum_init(skb, uh);
1714 if (err)
1715 return err;
1716 }
1717
1718 /* Note, we are only interested in != 0 or == 0, thus the
1719 * force to int.
1720 */
1721 return (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
1722 inet_compute_pseudo);
1723 }
1724
1725 /*
1726 * All we need to do is get the socket, and then do a checksum.
1727 */
1728
1729 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1730 int proto)
1731 {
1732 struct sock *sk;
1733 struct udphdr *uh;
1734 unsigned short ulen;
1735 struct rtable *rt = skb_rtable(skb);
1736 __be32 saddr, daddr;
1737 struct net *net = dev_net(skb->dev);
1738
1739 /*
1740 * Validate the packet.
1741 */
1742 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1743 goto drop; /* No space for header. */
1744
1745 uh = udp_hdr(skb);
1746 ulen = ntohs(uh->len);
1747 saddr = ip_hdr(skb)->saddr;
1748 daddr = ip_hdr(skb)->daddr;
1749
1750 if (ulen > skb->len)
1751 goto short_packet;
1752
1753 if (proto == IPPROTO_UDP) {
1754 /* UDP validates ulen. */
1755 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1756 goto short_packet;
1757 uh = udp_hdr(skb);
1758 }
1759
1760 if (udp4_csum_init(skb, uh, proto))
1761 goto csum_error;
1762
1763 sk = skb_steal_sock(skb);
1764 if (sk) {
1765 struct dst_entry *dst = skb_dst(skb);
1766 int ret;
1767
1768 if (unlikely(sk->sk_rx_dst != dst))
1769 udp_sk_rx_dst_set(sk, dst);
1770
1771 ret = udp_queue_rcv_skb(sk, skb);
1772 sock_put(sk);
1773 /* a return value > 0 means to resubmit the input, but
1774 * it wants the return to be -protocol, or 0
1775 */
1776 if (ret > 0)
1777 return -ret;
1778 return 0;
1779 }
1780
1781 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1782 return __udp4_lib_mcast_deliver(net, skb, uh,
1783 saddr, daddr, udptable, proto);
1784
1785 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1786 if (sk) {
1787 int ret;
1788
1789 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
1790 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
1791 inet_compute_pseudo);
1792
1793 ret = udp_queue_rcv_skb(sk, skb);
1794
1795 /* a return value > 0 means to resubmit the input, but
1796 * it wants the return to be -protocol, or 0
1797 */
1798 if (ret > 0)
1799 return -ret;
1800 return 0;
1801 }
1802
1803 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1804 goto drop;
1805 nf_reset(skb);
1806
1807 /* No socket. Drop packet silently, if checksum is wrong */
1808 if (udp_lib_checksum_complete(skb))
1809 goto csum_error;
1810
1811 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1812 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1813
1814 /*
1815 * Hmm. We got an UDP packet to a port to which we
1816 * don't wanna listen. Ignore it.
1817 */
1818 kfree_skb(skb);
1819 return 0;
1820
1821 short_packet:
1822 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1823 proto == IPPROTO_UDPLITE ? "Lite" : "",
1824 &saddr, ntohs(uh->source),
1825 ulen, skb->len,
1826 &daddr, ntohs(uh->dest));
1827 goto drop;
1828
1829 csum_error:
1830 /*
1831 * RFC1122: OK. Discards the bad packet silently (as far as
1832 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1833 */
1834 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1835 proto == IPPROTO_UDPLITE ? "Lite" : "",
1836 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1837 ulen);
1838 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1839 drop:
1840 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1841 kfree_skb(skb);
1842 return 0;
1843 }
1844
1845 /* We can only early demux multicast if there is a single matching socket.
1846 * If more than one socket found returns NULL
1847 */
1848 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
1849 __be16 loc_port, __be32 loc_addr,
1850 __be16 rmt_port, __be32 rmt_addr,
1851 int dif)
1852 {
1853 struct sock *sk, *result;
1854 unsigned short hnum = ntohs(loc_port);
1855 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
1856 struct udp_hslot *hslot = &udp_table.hash[slot];
1857
1858 /* Do not bother scanning a too big list */
1859 if (hslot->count > 10)
1860 return NULL;
1861
1862 result = NULL;
1863 sk_for_each_rcu(sk, &hslot->head) {
1864 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
1865 rmt_port, rmt_addr, dif, hnum)) {
1866 if (result)
1867 return NULL;
1868 result = sk;
1869 }
1870 }
1871
1872 return result;
1873 }
1874
1875 /* For unicast we should only early demux connected sockets or we can
1876 * break forwarding setups. The chains here can be long so only check
1877 * if the first socket is an exact match and if not move on.
1878 */
1879 static struct sock *__udp4_lib_demux_lookup(struct net *net,
1880 __be16 loc_port, __be32 loc_addr,
1881 __be16 rmt_port, __be32 rmt_addr,
1882 int dif)
1883 {
1884 unsigned short hnum = ntohs(loc_port);
1885 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
1886 unsigned int slot2 = hash2 & udp_table.mask;
1887 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
1888 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
1889 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
1890 struct sock *sk;
1891
1892 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
1893 if (INET_MATCH(sk, net, acookie, rmt_addr,
1894 loc_addr, ports, dif))
1895 return sk;
1896 /* Only check first socket in chain */
1897 break;
1898 }
1899 return NULL;
1900 }
1901
1902 void udp_v4_early_demux(struct sk_buff *skb)
1903 {
1904 struct net *net = dev_net(skb->dev);
1905 const struct iphdr *iph;
1906 const struct udphdr *uh;
1907 struct sock *sk = NULL;
1908 struct dst_entry *dst;
1909 int dif = skb->dev->ifindex;
1910 int ours;
1911
1912 /* validate the packet */
1913 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
1914 return;
1915
1916 iph = ip_hdr(skb);
1917 uh = udp_hdr(skb);
1918
1919 if (skb->pkt_type == PACKET_BROADCAST ||
1920 skb->pkt_type == PACKET_MULTICAST) {
1921 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
1922
1923 if (!in_dev)
1924 return;
1925
1926 /* we are supposed to accept bcast packets */
1927 if (skb->pkt_type == PACKET_MULTICAST) {
1928 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
1929 iph->protocol);
1930 if (!ours)
1931 return;
1932 }
1933
1934 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
1935 uh->source, iph->saddr, dif);
1936 } else if (skb->pkt_type == PACKET_HOST) {
1937 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
1938 uh->source, iph->saddr, dif);
1939 }
1940
1941 if (!sk || !atomic_inc_not_zero_hint(&sk->sk_refcnt, 2))
1942 return;
1943
1944 skb->sk = sk;
1945 skb->destructor = sock_efree;
1946 dst = READ_ONCE(sk->sk_rx_dst);
1947
1948 if (dst)
1949 dst = dst_check(dst, 0);
1950 if (dst) {
1951 /* DST_NOCACHE can not be used without taking a reference */
1952 if (dst->flags & DST_NOCACHE) {
1953 if (likely(atomic_inc_not_zero(&dst->__refcnt)))
1954 skb_dst_set(skb, dst);
1955 } else {
1956 skb_dst_set_noref(skb, dst);
1957 }
1958 }
1959 }
1960
1961 int udp_rcv(struct sk_buff *skb)
1962 {
1963 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1964 }
1965
1966 void udp_destroy_sock(struct sock *sk)
1967 {
1968 struct udp_sock *up = udp_sk(sk);
1969 bool slow = lock_sock_fast(sk);
1970 udp_flush_pending_frames(sk);
1971 unlock_sock_fast(sk, slow);
1972 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1973 void (*encap_destroy)(struct sock *sk);
1974 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1975 if (encap_destroy)
1976 encap_destroy(sk);
1977 }
1978 }
1979
1980 /*
1981 * Socket option code for UDP
1982 */
1983 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1984 char __user *optval, unsigned int optlen,
1985 int (*push_pending_frames)(struct sock *))
1986 {
1987 struct udp_sock *up = udp_sk(sk);
1988 int val, valbool;
1989 int err = 0;
1990 int is_udplite = IS_UDPLITE(sk);
1991
1992 if (optlen < sizeof(int))
1993 return -EINVAL;
1994
1995 if (get_user(val, (int __user *)optval))
1996 return -EFAULT;
1997
1998 valbool = val ? 1 : 0;
1999
2000 switch (optname) {
2001 case UDP_CORK:
2002 if (val != 0) {
2003 up->corkflag = 1;
2004 } else {
2005 up->corkflag = 0;
2006 lock_sock(sk);
2007 push_pending_frames(sk);
2008 release_sock(sk);
2009 }
2010 break;
2011
2012 case UDP_ENCAP:
2013 switch (val) {
2014 case 0:
2015 case UDP_ENCAP_ESPINUDP:
2016 case UDP_ENCAP_ESPINUDP_NON_IKE:
2017 up->encap_rcv = xfrm4_udp_encap_rcv;
2018 /* FALLTHROUGH */
2019 case UDP_ENCAP_L2TPINUDP:
2020 up->encap_type = val;
2021 udp_encap_enable();
2022 break;
2023 default:
2024 err = -ENOPROTOOPT;
2025 break;
2026 }
2027 break;
2028
2029 case UDP_NO_CHECK6_TX:
2030 up->no_check6_tx = valbool;
2031 break;
2032
2033 case UDP_NO_CHECK6_RX:
2034 up->no_check6_rx = valbool;
2035 break;
2036
2037 /*
2038 * UDP-Lite's partial checksum coverage (RFC 3828).
2039 */
2040 /* The sender sets actual checksum coverage length via this option.
2041 * The case coverage > packet length is handled by send module. */
2042 case UDPLITE_SEND_CSCOV:
2043 if (!is_udplite) /* Disable the option on UDP sockets */
2044 return -ENOPROTOOPT;
2045 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2046 val = 8;
2047 else if (val > USHRT_MAX)
2048 val = USHRT_MAX;
2049 up->pcslen = val;
2050 up->pcflag |= UDPLITE_SEND_CC;
2051 break;
2052
2053 /* The receiver specifies a minimum checksum coverage value. To make
2054 * sense, this should be set to at least 8 (as done below). If zero is
2055 * used, this again means full checksum coverage. */
2056 case UDPLITE_RECV_CSCOV:
2057 if (!is_udplite) /* Disable the option on UDP sockets */
2058 return -ENOPROTOOPT;
2059 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2060 val = 8;
2061 else if (val > USHRT_MAX)
2062 val = USHRT_MAX;
2063 up->pcrlen = val;
2064 up->pcflag |= UDPLITE_RECV_CC;
2065 break;
2066
2067 default:
2068 err = -ENOPROTOOPT;
2069 break;
2070 }
2071
2072 return err;
2073 }
2074 EXPORT_SYMBOL(udp_lib_setsockopt);
2075
2076 int udp_setsockopt(struct sock *sk, int level, int optname,
2077 char __user *optval, unsigned int optlen)
2078 {
2079 if (level == SOL_UDP || level == SOL_UDPLITE)
2080 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2081 udp_push_pending_frames);
2082 return ip_setsockopt(sk, level, optname, optval, optlen);
2083 }
2084
2085 #ifdef CONFIG_COMPAT
2086 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2087 char __user *optval, unsigned int optlen)
2088 {
2089 if (level == SOL_UDP || level == SOL_UDPLITE)
2090 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2091 udp_push_pending_frames);
2092 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2093 }
2094 #endif
2095
2096 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2097 char __user *optval, int __user *optlen)
2098 {
2099 struct udp_sock *up = udp_sk(sk);
2100 int val, len;
2101
2102 if (get_user(len, optlen))
2103 return -EFAULT;
2104
2105 len = min_t(unsigned int, len, sizeof(int));
2106
2107 if (len < 0)
2108 return -EINVAL;
2109
2110 switch (optname) {
2111 case UDP_CORK:
2112 val = up->corkflag;
2113 break;
2114
2115 case UDP_ENCAP:
2116 val = up->encap_type;
2117 break;
2118
2119 case UDP_NO_CHECK6_TX:
2120 val = up->no_check6_tx;
2121 break;
2122
2123 case UDP_NO_CHECK6_RX:
2124 val = up->no_check6_rx;
2125 break;
2126
2127 /* The following two cannot be changed on UDP sockets, the return is
2128 * always 0 (which corresponds to the full checksum coverage of UDP). */
2129 case UDPLITE_SEND_CSCOV:
2130 val = up->pcslen;
2131 break;
2132
2133 case UDPLITE_RECV_CSCOV:
2134 val = up->pcrlen;
2135 break;
2136
2137 default:
2138 return -ENOPROTOOPT;
2139 }
2140
2141 if (put_user(len, optlen))
2142 return -EFAULT;
2143 if (copy_to_user(optval, &val, len))
2144 return -EFAULT;
2145 return 0;
2146 }
2147 EXPORT_SYMBOL(udp_lib_getsockopt);
2148
2149 int udp_getsockopt(struct sock *sk, int level, int optname,
2150 char __user *optval, int __user *optlen)
2151 {
2152 if (level == SOL_UDP || level == SOL_UDPLITE)
2153 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2154 return ip_getsockopt(sk, level, optname, optval, optlen);
2155 }
2156
2157 #ifdef CONFIG_COMPAT
2158 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2159 char __user *optval, int __user *optlen)
2160 {
2161 if (level == SOL_UDP || level == SOL_UDPLITE)
2162 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2163 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2164 }
2165 #endif
2166 /**
2167 * udp_poll - wait for a UDP event.
2168 * @file - file struct
2169 * @sock - socket
2170 * @wait - poll table
2171 *
2172 * This is same as datagram poll, except for the special case of
2173 * blocking sockets. If application is using a blocking fd
2174 * and a packet with checksum error is in the queue;
2175 * then it could get return from select indicating data available
2176 * but then block when reading it. Add special case code
2177 * to work around these arguably broken applications.
2178 */
2179 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2180 {
2181 unsigned int mask = datagram_poll(file, sock, wait);
2182 struct sock *sk = sock->sk;
2183
2184 sock_rps_record_flow(sk);
2185
2186 /* Check for false positives due to checksum errors */
2187 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2188 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
2189 mask &= ~(POLLIN | POLLRDNORM);
2190
2191 return mask;
2192
2193 }
2194 EXPORT_SYMBOL(udp_poll);
2195
2196 int udp_abort(struct sock *sk, int err)
2197 {
2198 lock_sock(sk);
2199
2200 sk->sk_err = err;
2201 sk->sk_error_report(sk);
2202 udp_disconnect(sk, 0);
2203
2204 release_sock(sk);
2205
2206 return 0;
2207 }
2208 EXPORT_SYMBOL_GPL(udp_abort);
2209
2210 struct proto udp_prot = {
2211 .name = "UDP",
2212 .owner = THIS_MODULE,
2213 .close = udp_lib_close,
2214 .connect = ip4_datagram_connect,
2215 .disconnect = udp_disconnect,
2216 .ioctl = udp_ioctl,
2217 .destroy = udp_destroy_sock,
2218 .setsockopt = udp_setsockopt,
2219 .getsockopt = udp_getsockopt,
2220 .sendmsg = udp_sendmsg,
2221 .recvmsg = udp_recvmsg,
2222 .sendpage = udp_sendpage,
2223 .backlog_rcv = __udp_queue_rcv_skb,
2224 .release_cb = ip4_datagram_release_cb,
2225 .hash = udp_lib_hash,
2226 .unhash = udp_lib_unhash,
2227 .rehash = udp_v4_rehash,
2228 .get_port = udp_v4_get_port,
2229 .memory_allocated = &udp_memory_allocated,
2230 .sysctl_mem = sysctl_udp_mem,
2231 .sysctl_wmem = &sysctl_udp_wmem_min,
2232 .sysctl_rmem = &sysctl_udp_rmem_min,
2233 .obj_size = sizeof(struct udp_sock),
2234 .slab_flags = SLAB_DESTROY_BY_RCU,
2235 .h.udp_table = &udp_table,
2236 #ifdef CONFIG_COMPAT
2237 .compat_setsockopt = compat_udp_setsockopt,
2238 .compat_getsockopt = compat_udp_getsockopt,
2239 #endif
2240 .diag_destroy = udp_abort,
2241 };
2242 EXPORT_SYMBOL(udp_prot);
2243
2244 /* ------------------------------------------------------------------------ */
2245 #ifdef CONFIG_PROC_FS
2246
2247 static struct sock *udp_get_first(struct seq_file *seq, int start)
2248 {
2249 struct sock *sk;
2250 struct udp_iter_state *state = seq->private;
2251 struct net *net = seq_file_net(seq);
2252
2253 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2254 ++state->bucket) {
2255 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2256
2257 if (hlist_empty(&hslot->head))
2258 continue;
2259
2260 spin_lock_bh(&hslot->lock);
2261 sk_for_each(sk, &hslot->head) {
2262 if (!net_eq(sock_net(sk), net))
2263 continue;
2264 if (sk->sk_family == state->family)
2265 goto found;
2266 }
2267 spin_unlock_bh(&hslot->lock);
2268 }
2269 sk = NULL;
2270 found:
2271 return sk;
2272 }
2273
2274 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2275 {
2276 struct udp_iter_state *state = seq->private;
2277 struct net *net = seq_file_net(seq);
2278
2279 do {
2280 sk = sk_next(sk);
2281 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2282
2283 if (!sk) {
2284 if (state->bucket <= state->udp_table->mask)
2285 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2286 return udp_get_first(seq, state->bucket + 1);
2287 }
2288 return sk;
2289 }
2290
2291 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2292 {
2293 struct sock *sk = udp_get_first(seq, 0);
2294
2295 if (sk)
2296 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2297 --pos;
2298 return pos ? NULL : sk;
2299 }
2300
2301 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2302 {
2303 struct udp_iter_state *state = seq->private;
2304 state->bucket = MAX_UDP_PORTS;
2305
2306 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2307 }
2308
2309 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2310 {
2311 struct sock *sk;
2312
2313 if (v == SEQ_START_TOKEN)
2314 sk = udp_get_idx(seq, 0);
2315 else
2316 sk = udp_get_next(seq, v);
2317
2318 ++*pos;
2319 return sk;
2320 }
2321
2322 static void udp_seq_stop(struct seq_file *seq, void *v)
2323 {
2324 struct udp_iter_state *state = seq->private;
2325
2326 if (state->bucket <= state->udp_table->mask)
2327 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2328 }
2329
2330 int udp_seq_open(struct inode *inode, struct file *file)
2331 {
2332 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2333 struct udp_iter_state *s;
2334 int err;
2335
2336 err = seq_open_net(inode, file, &afinfo->seq_ops,
2337 sizeof(struct udp_iter_state));
2338 if (err < 0)
2339 return err;
2340
2341 s = ((struct seq_file *)file->private_data)->private;
2342 s->family = afinfo->family;
2343 s->udp_table = afinfo->udp_table;
2344 return err;
2345 }
2346 EXPORT_SYMBOL(udp_seq_open);
2347
2348 /* ------------------------------------------------------------------------ */
2349 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2350 {
2351 struct proc_dir_entry *p;
2352 int rc = 0;
2353
2354 afinfo->seq_ops.start = udp_seq_start;
2355 afinfo->seq_ops.next = udp_seq_next;
2356 afinfo->seq_ops.stop = udp_seq_stop;
2357
2358 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2359 afinfo->seq_fops, afinfo);
2360 if (!p)
2361 rc = -ENOMEM;
2362 return rc;
2363 }
2364 EXPORT_SYMBOL(udp_proc_register);
2365
2366 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2367 {
2368 remove_proc_entry(afinfo->name, net->proc_net);
2369 }
2370 EXPORT_SYMBOL(udp_proc_unregister);
2371
2372 /* ------------------------------------------------------------------------ */
2373 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2374 int bucket)
2375 {
2376 struct inet_sock *inet = inet_sk(sp);
2377 __be32 dest = inet->inet_daddr;
2378 __be32 src = inet->inet_rcv_saddr;
2379 __u16 destp = ntohs(inet->inet_dport);
2380 __u16 srcp = ntohs(inet->inet_sport);
2381
2382 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2383 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2384 bucket, src, srcp, dest, destp, sp->sk_state,
2385 sk_wmem_alloc_get(sp),
2386 sk_rmem_alloc_get(sp),
2387 0, 0L, 0,
2388 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2389 0, sock_i_ino(sp),
2390 atomic_read(&sp->sk_refcnt), sp,
2391 atomic_read(&sp->sk_drops));
2392 }
2393
2394 int udp4_seq_show(struct seq_file *seq, void *v)
2395 {
2396 seq_setwidth(seq, 127);
2397 if (v == SEQ_START_TOKEN)
2398 seq_puts(seq, " sl local_address rem_address st tx_queue "
2399 "rx_queue tr tm->when retrnsmt uid timeout "
2400 "inode ref pointer drops");
2401 else {
2402 struct udp_iter_state *state = seq->private;
2403
2404 udp4_format_sock(v, seq, state->bucket);
2405 }
2406 seq_pad(seq, '\n');
2407 return 0;
2408 }
2409
2410 static const struct file_operations udp_afinfo_seq_fops = {
2411 .owner = THIS_MODULE,
2412 .open = udp_seq_open,
2413 .read = seq_read,
2414 .llseek = seq_lseek,
2415 .release = seq_release_net
2416 };
2417
2418 /* ------------------------------------------------------------------------ */
2419 static struct udp_seq_afinfo udp4_seq_afinfo = {
2420 .name = "udp",
2421 .family = AF_INET,
2422 .udp_table = &udp_table,
2423 .seq_fops = &udp_afinfo_seq_fops,
2424 .seq_ops = {
2425 .show = udp4_seq_show,
2426 },
2427 };
2428
2429 static int __net_init udp4_proc_init_net(struct net *net)
2430 {
2431 return udp_proc_register(net, &udp4_seq_afinfo);
2432 }
2433
2434 static void __net_exit udp4_proc_exit_net(struct net *net)
2435 {
2436 udp_proc_unregister(net, &udp4_seq_afinfo);
2437 }
2438
2439 static struct pernet_operations udp4_net_ops = {
2440 .init = udp4_proc_init_net,
2441 .exit = udp4_proc_exit_net,
2442 };
2443
2444 int __init udp4_proc_init(void)
2445 {
2446 return register_pernet_subsys(&udp4_net_ops);
2447 }
2448
2449 void udp4_proc_exit(void)
2450 {
2451 unregister_pernet_subsys(&udp4_net_ops);
2452 }
2453 #endif /* CONFIG_PROC_FS */
2454
2455 static __initdata unsigned long uhash_entries;
2456 static int __init set_uhash_entries(char *str)
2457 {
2458 ssize_t ret;
2459
2460 if (!str)
2461 return 0;
2462
2463 ret = kstrtoul(str, 0, &uhash_entries);
2464 if (ret)
2465 return 0;
2466
2467 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2468 uhash_entries = UDP_HTABLE_SIZE_MIN;
2469 return 1;
2470 }
2471 __setup("uhash_entries=", set_uhash_entries);
2472
2473 void __init udp_table_init(struct udp_table *table, const char *name)
2474 {
2475 unsigned int i;
2476
2477 table->hash = alloc_large_system_hash(name,
2478 2 * sizeof(struct udp_hslot),
2479 uhash_entries,
2480 21, /* one slot per 2 MB */
2481 0,
2482 &table->log,
2483 &table->mask,
2484 UDP_HTABLE_SIZE_MIN,
2485 64 * 1024);
2486
2487 table->hash2 = table->hash + (table->mask + 1);
2488 for (i = 0; i <= table->mask; i++) {
2489 INIT_HLIST_HEAD(&table->hash[i].head);
2490 table->hash[i].count = 0;
2491 spin_lock_init(&table->hash[i].lock);
2492 }
2493 for (i = 0; i <= table->mask; i++) {
2494 INIT_HLIST_HEAD(&table->hash2[i].head);
2495 table->hash2[i].count = 0;
2496 spin_lock_init(&table->hash2[i].lock);
2497 }
2498 }
2499
2500 u32 udp_flow_hashrnd(void)
2501 {
2502 static u32 hashrnd __read_mostly;
2503
2504 net_get_random_once(&hashrnd, sizeof(hashrnd));
2505
2506 return hashrnd;
2507 }
2508 EXPORT_SYMBOL(udp_flow_hashrnd);
2509
2510 void __init udp_init(void)
2511 {
2512 unsigned long limit;
2513
2514 udp_table_init(&udp_table, "UDP");
2515 limit = nr_free_buffer_pages() / 8;
2516 limit = max(limit, 128UL);
2517 sysctl_udp_mem[0] = limit / 4 * 3;
2518 sysctl_udp_mem[1] = limit;
2519 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2520
2521 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2522 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2523 }
(deprecated) Btrfs devel tree