?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[cris-mirror.git] / net / ipv4 / udp.c
blob5fdcb8d108d468ab33e8f29a21efdd643d27a49f
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
118 struct udp_table udp_table __read_mostly;
119 EXPORT_SYMBOL(udp_table);
120
121 long sysctl_udp_mem[3] __read_mostly;
122 EXPORT_SYMBOL(sysctl_udp_mem);
123
124 int sysctl_udp_rmem_min __read_mostly;
125 EXPORT_SYMBOL(sysctl_udp_rmem_min);
126
127 int sysctl_udp_wmem_min __read_mostly;
128 EXPORT_SYMBOL(sysctl_udp_wmem_min);
129
130 atomic_long_t udp_memory_allocated;
131 EXPORT_SYMBOL(udp_memory_allocated);
132
133 #define MAX_UDP_PORTS 65536
134 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
135
136 static int udp_lib_lport_inuse(struct net *net, __u16 num,
137 const struct udp_hslot *hslot,
138 unsigned long *bitmap,
139 struct sock *sk,
140 int (*saddr_comp)(const struct sock *sk1,
141 const struct sock *sk2,
142 bool match_wildcard),
143 unsigned int log)
144 {
145 struct sock *sk2;
146 kuid_t uid = sock_i_uid(sk);
147
148 sk_for_each(sk2, &hslot->head) {
149 if (net_eq(sock_net(sk2), net) &&
150 sk2 != sk &&
151 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
152 (!sk2->sk_reuse || !sk->sk_reuse) &&
153 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
154 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
155 (!sk2->sk_reuseport || !sk->sk_reuseport ||
156 rcu_access_pointer(sk->sk_reuseport_cb) ||
157 !uid_eq(uid, sock_i_uid(sk2))) &&
158 saddr_comp(sk, sk2, true)) {
159 if (!bitmap)
160 return 1;
161 __set_bit(udp_sk(sk2)->udp_port_hash >> log, bitmap);
162 }
163 }
164 return 0;
165 }
166
167 /*
168 * Note: we still hold spinlock of primary hash chain, so no other writer
169 * can insert/delete a socket with local_port == num
170 */
171 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
172 struct udp_hslot *hslot2,
173 struct sock *sk,
174 int (*saddr_comp)(const struct sock *sk1,
175 const struct sock *sk2,
176 bool match_wildcard))
177 {
178 struct sock *sk2;
179 kuid_t uid = sock_i_uid(sk);
180 int res = 0;
181
182 spin_lock(&hslot2->lock);
183 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
184 if (net_eq(sock_net(sk2), net) &&
185 sk2 != sk &&
186 (udp_sk(sk2)->udp_port_hash == num) &&
187 (!sk2->sk_reuse || !sk->sk_reuse) &&
188 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
189 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
190 (!sk2->sk_reuseport || !sk->sk_reuseport ||
191 rcu_access_pointer(sk->sk_reuseport_cb) ||
192 !uid_eq(uid, sock_i_uid(sk2))) &&
193 saddr_comp(sk, sk2, true)) {
194 res = 1;
195 break;
196 }
197 }
198 spin_unlock(&hslot2->lock);
199 return res;
200 }
201
202 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot,
203 int (*saddr_same)(const struct sock *sk1,
204 const struct sock *sk2,
205 bool match_wildcard))
206 {
207 struct net *net = sock_net(sk);
208 kuid_t uid = sock_i_uid(sk);
209 struct sock *sk2;
210
211 sk_for_each(sk2, &hslot->head) {
212 if (net_eq(sock_net(sk2), net) &&
213 sk2 != sk &&
214 sk2->sk_family == sk->sk_family &&
215 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
216 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
217 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
218 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
219 (*saddr_same)(sk, sk2, false)) {
220 return reuseport_add_sock(sk, sk2);
221 }
222 }
223
224 /* Initial allocation may have already happened via setsockopt */
225 if (!rcu_access_pointer(sk->sk_reuseport_cb))
226 return reuseport_alloc(sk);
227 return 0;
228 }
229
230 /**
231 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
232 *
233 * @sk: socket struct in question
234 * @snum: port number to look up
235 * @saddr_comp: AF-dependent comparison of bound local IP addresses
236 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
237 * with NULL address
238 */
239 int udp_lib_get_port(struct sock *sk, unsigned short snum,
240 int (*saddr_comp)(const struct sock *sk1,
241 const struct sock *sk2,
242 bool match_wildcard),
243 unsigned int hash2_nulladdr)
244 {
245 struct udp_hslot *hslot, *hslot2;
246 struct udp_table *udptable = sk->sk_prot->h.udp_table;
247 int error = 1;
248 struct net *net = sock_net(sk);
249
250 if (!snum) {
251 int low, high, remaining;
252 unsigned int rand;
253 unsigned short first, last;
254 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
255
256 inet_get_local_port_range(net, &low, &high);
257 remaining = (high - low) + 1;
258
259 rand = prandom_u32();
260 first = reciprocal_scale(rand, remaining) + low;
261 /*
262 * force rand to be an odd multiple of UDP_HTABLE_SIZE
263 */
264 rand = (rand | 1) * (udptable->mask + 1);
265 last = first + udptable->mask + 1;
266 do {
267 hslot = udp_hashslot(udptable, net, first);
268 bitmap_zero(bitmap, PORTS_PER_CHAIN);
269 spin_lock_bh(&hslot->lock);
270 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
271 saddr_comp, udptable->log);
272
273 snum = first;
274 /*
275 * Iterate on all possible values of snum for this hash.
276 * Using steps of an odd multiple of UDP_HTABLE_SIZE
277 * give us randomization and full range coverage.
278 */
279 do {
280 if (low <= snum && snum <= high &&
281 !test_bit(snum >> udptable->log, bitmap) &&
282 !inet_is_local_reserved_port(net, snum))
283 goto found;
284 snum += rand;
285 } while (snum != first);
286 spin_unlock_bh(&hslot->lock);
287 } while (++first != last);
288 goto fail;
289 } else {
290 hslot = udp_hashslot(udptable, net, snum);
291 spin_lock_bh(&hslot->lock);
292 if (hslot->count > 10) {
293 int exist;
294 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
295
296 slot2 &= udptable->mask;
297 hash2_nulladdr &= udptable->mask;
298
299 hslot2 = udp_hashslot2(udptable, slot2);
300 if (hslot->count < hslot2->count)
301 goto scan_primary_hash;
302
303 exist = udp_lib_lport_inuse2(net, snum, hslot2,
304 sk, saddr_comp);
305 if (!exist && (hash2_nulladdr != slot2)) {
306 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
307 exist = udp_lib_lport_inuse2(net, snum, hslot2,
308 sk, saddr_comp);
309 }
310 if (exist)
311 goto fail_unlock;
312 else
313 goto found;
314 }
315 scan_primary_hash:
316 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
317 saddr_comp, 0))
318 goto fail_unlock;
319 }
320 found:
321 inet_sk(sk)->inet_num = snum;
322 udp_sk(sk)->udp_port_hash = snum;
323 udp_sk(sk)->udp_portaddr_hash ^= snum;
324 if (sk_unhashed(sk)) {
325 if (sk->sk_reuseport &&
326 udp_reuseport_add_sock(sk, hslot, saddr_comp)) {
327 inet_sk(sk)->inet_num = 0;
328 udp_sk(sk)->udp_port_hash = 0;
329 udp_sk(sk)->udp_portaddr_hash ^= snum;
330 goto fail_unlock;
331 }
332
333 sk_add_node_rcu(sk, &hslot->head);
334 hslot->count++;
335 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
336
337 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
338 spin_lock(&hslot2->lock);
339 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
340 sk->sk_family == AF_INET6)
341 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
342 &hslot2->head);
343 else
344 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
345 &hslot2->head);
346 hslot2->count++;
347 spin_unlock(&hslot2->lock);
348 }
349 sock_set_flag(sk, SOCK_RCU_FREE);
350 error = 0;
351 fail_unlock:
352 spin_unlock_bh(&hslot->lock);
353 fail:
354 return error;
355 }
356 EXPORT_SYMBOL(udp_lib_get_port);
357
358 /* match_wildcard == true: 0.0.0.0 equals to any IPv4 addresses
359 * match_wildcard == false: addresses must be exactly the same, i.e.
360 * 0.0.0.0 only equals to 0.0.0.0
361 */
362 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2,
363 bool match_wildcard)
364 {
365 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
366
367 if (!ipv6_only_sock(sk2)) {
368 if (inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)
369 return 1;
370 if (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr)
371 return match_wildcard;
372 }
373 return 0;
374 }
375
376 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr,
377 unsigned int port)
378 {
379 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
380 }
381
382 int udp_v4_get_port(struct sock *sk, unsigned short snum)
383 {
384 unsigned int hash2_nulladdr =
385 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
386 unsigned int hash2_partial =
387 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
388
389 /* precompute partial secondary hash */
390 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
391 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
392 }
393
394 static int compute_score(struct sock *sk, struct net *net,
395 __be32 saddr, __be16 sport,
396 __be32 daddr, unsigned short hnum, int dif)
397 {
398 int score;
399 struct inet_sock *inet;
400
401 if (!net_eq(sock_net(sk), net) ||
402 udp_sk(sk)->udp_port_hash != hnum ||
403 ipv6_only_sock(sk))
404 return -1;
405
406 score = (sk->sk_family == PF_INET) ? 2 : 1;
407 inet = inet_sk(sk);
408
409 if (inet->inet_rcv_saddr) {
410 if (inet->inet_rcv_saddr != daddr)
411 return -1;
412 score += 4;
413 }
414
415 if (inet->inet_daddr) {
416 if (inet->inet_daddr != saddr)
417 return -1;
418 score += 4;
419 }
420
421 if (inet->inet_dport) {
422 if (inet->inet_dport != sport)
423 return -1;
424 score += 4;
425 }
426
427 if (sk->sk_bound_dev_if) {
428 if (sk->sk_bound_dev_if != dif)
429 return -1;
430 score += 4;
431 }
432 if (sk->sk_incoming_cpu == raw_smp_processor_id())
433 score++;
434 return score;
435 }
436
437 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
438 const __u16 lport, const __be32 faddr,
439 const __be16 fport)
440 {
441 static u32 udp_ehash_secret __read_mostly;
442
443 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
444
445 return __inet_ehashfn(laddr, lport, faddr, fport,
446 udp_ehash_secret + net_hash_mix(net));
447 }
448
449 /* called with rcu_read_lock() */
450 static struct sock *udp4_lib_lookup2(struct net *net,
451 __be32 saddr, __be16 sport,
452 __be32 daddr, unsigned int hnum, int dif,
453 struct udp_hslot *hslot2,
454 struct sk_buff *skb)
455 {
456 struct sock *sk, *result;
457 int score, badness, matches = 0, reuseport = 0;
458 u32 hash = 0;
459
460 result = NULL;
461 badness = 0;
462 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
463 score = compute_score(sk, net, saddr, sport,
464 daddr, hnum, dif);
465 if (score > badness) {
466 reuseport = sk->sk_reuseport;
467 if (reuseport) {
468 hash = udp_ehashfn(net, daddr, hnum,
469 saddr, sport);
470 result = reuseport_select_sock(sk, hash, skb,
471 sizeof(struct udphdr));
472 if (result)
473 return result;
474 matches = 1;
475 }
476 badness = score;
477 result = sk;
478 } else if (score == badness && reuseport) {
479 matches++;
480 if (reciprocal_scale(hash, matches) == 0)
481 result = sk;
482 hash = next_pseudo_random32(hash);
483 }
484 }
485 return result;
486 }
487
488 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
489 * harder than this. -DaveM
490 */
491 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
492 __be16 sport, __be32 daddr, __be16 dport,
493 int dif, struct udp_table *udptable, struct sk_buff *skb)
494 {
495 struct sock *sk, *result;
496 unsigned short hnum = ntohs(dport);
497 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
498 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
499 int score, badness, matches = 0, reuseport = 0;
500 u32 hash = 0;
501
502 if (hslot->count > 10) {
503 hash2 = udp4_portaddr_hash(net, daddr, hnum);
504 slot2 = hash2 & udptable->mask;
505 hslot2 = &udptable->hash2[slot2];
506 if (hslot->count < hslot2->count)
507 goto begin;
508
509 result = udp4_lib_lookup2(net, saddr, sport,
510 daddr, hnum, dif,
511 hslot2, skb);
512 if (!result) {
513 unsigned int old_slot2 = slot2;
514 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
515 slot2 = hash2 & udptable->mask;
516 /* avoid searching the same slot again. */
517 if (unlikely(slot2 == old_slot2))
518 return result;
519
520 hslot2 = &udptable->hash2[slot2];
521 if (hslot->count < hslot2->count)
522 goto begin;
523
524 result = udp4_lib_lookup2(net, saddr, sport,
525 daddr, hnum, dif,
526 hslot2, skb);
527 }
528 return result;
529 }
530 begin:
531 result = NULL;
532 badness = 0;
533 sk_for_each_rcu(sk, &hslot->head) {
534 score = compute_score(sk, net, saddr, sport,
535 daddr, hnum, dif);
536 if (score > badness) {
537 reuseport = sk->sk_reuseport;
538 if (reuseport) {
539 hash = udp_ehashfn(net, daddr, hnum,
540 saddr, sport);
541 result = reuseport_select_sock(sk, hash, skb,
542 sizeof(struct udphdr));
543 if (result)
544 return result;
545 matches = 1;
546 }
547 result = sk;
548 badness = score;
549 } else if (score == badness && reuseport) {
550 matches++;
551 if (reciprocal_scale(hash, matches) == 0)
552 result = sk;
553 hash = next_pseudo_random32(hash);
554 }
555 }
556 return result;
557 }
558 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
559
560 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
561 __be16 sport, __be16 dport,
562 struct udp_table *udptable)
563 {
564 const struct iphdr *iph = ip_hdr(skb);
565
566 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
567 iph->daddr, dport, inet_iif(skb),
568 udptable, skb);
569 }
570
571 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
572 __be16 sport, __be16 dport)
573 {
574 return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table);
575 }
576 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
577
578 /* Must be called under rcu_read_lock().
579 * Does increment socket refcount.
580 */
581 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
582 IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY)
583 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
584 __be32 daddr, __be16 dport, int dif)
585 {
586 struct sock *sk;
587
588 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
589 dif, &udp_table, NULL);
590 if (sk && !atomic_inc_not_zero(&sk->sk_refcnt))
591 sk = NULL;
592 return sk;
593 }
594 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
595 #endif
596
597 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
598 __be16 loc_port, __be32 loc_addr,
599 __be16 rmt_port, __be32 rmt_addr,
600 int dif, unsigned short hnum)
601 {
602 struct inet_sock *inet = inet_sk(sk);
603
604 if (!net_eq(sock_net(sk), net) ||
605 udp_sk(sk)->udp_port_hash != hnum ||
606 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
607 (inet->inet_dport != rmt_port && inet->inet_dport) ||
608 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
609 ipv6_only_sock(sk) ||
610 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
611 return false;
612 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif))
613 return false;
614 return true;
615 }
616
617 /*
618 * This routine is called by the ICMP module when it gets some
619 * sort of error condition. If err < 0 then the socket should
620 * be closed and the error returned to the user. If err > 0
621 * it's just the icmp type << 8 | icmp code.
622 * Header points to the ip header of the error packet. We move
623 * on past this. Then (as it used to claim before adjustment)
624 * header points to the first 8 bytes of the udp header. We need
625 * to find the appropriate port.
626 */
627
628 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
629 {
630 struct inet_sock *inet;
631 const struct iphdr *iph = (const struct iphdr *)skb->data;
632 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
633 const int type = icmp_hdr(skb)->type;
634 const int code = icmp_hdr(skb)->code;
635 struct sock *sk;
636 int harderr;
637 int err;
638 struct net *net = dev_net(skb->dev);
639
640 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
641 iph->saddr, uh->source, skb->dev->ifindex, udptable,
642 NULL);
643 if (!sk) {
644 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
645 return; /* No socket for error */
646 }
647
648 err = 0;
649 harderr = 0;
650 inet = inet_sk(sk);
651
652 switch (type) {
653 default:
654 case ICMP_TIME_EXCEEDED:
655 err = EHOSTUNREACH;
656 break;
657 case ICMP_SOURCE_QUENCH:
658 goto out;
659 case ICMP_PARAMETERPROB:
660 err = EPROTO;
661 harderr = 1;
662 break;
663 case ICMP_DEST_UNREACH:
664 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
665 ipv4_sk_update_pmtu(skb, sk, info);
666 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
667 err = EMSGSIZE;
668 harderr = 1;
669 break;
670 }
671 goto out;
672 }
673 err = EHOSTUNREACH;
674 if (code <= NR_ICMP_UNREACH) {
675 harderr = icmp_err_convert[code].fatal;
676 err = icmp_err_convert[code].errno;
677 }
678 break;
679 case ICMP_REDIRECT:
680 ipv4_sk_redirect(skb, sk);
681 goto out;
682 }
683
684 /*
685 * RFC1122: OK. Passes ICMP errors back to application, as per
686 * 4.1.3.3.
687 */
688 if (!inet->recverr) {
689 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
690 goto out;
691 } else
692 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
693
694 sk->sk_err = err;
695 sk->sk_error_report(sk);
696 out:
697 return;
698 }
699
700 void udp_err(struct sk_buff *skb, u32 info)
701 {
702 __udp4_lib_err(skb, info, &udp_table);
703 }
704
705 /*
706 * Throw away all pending data and cancel the corking. Socket is locked.
707 */
708 void udp_flush_pending_frames(struct sock *sk)
709 {
710 struct udp_sock *up = udp_sk(sk);
711
712 if (up->pending) {
713 up->len = 0;
714 up->pending = 0;
715 ip_flush_pending_frames(sk);
716 }
717 }
718 EXPORT_SYMBOL(udp_flush_pending_frames);
719
720 /**
721 * udp4_hwcsum - handle outgoing HW checksumming
722 * @skb: sk_buff containing the filled-in UDP header
723 * (checksum field must be zeroed out)
724 * @src: source IP address
725 * @dst: destination IP address
726 */
727 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
728 {
729 struct udphdr *uh = udp_hdr(skb);
730 int offset = skb_transport_offset(skb);
731 int len = skb->len - offset;
732 int hlen = len;
733 __wsum csum = 0;
734
735 if (!skb_has_frag_list(skb)) {
736 /*
737 * Only one fragment on the socket.
738 */
739 skb->csum_start = skb_transport_header(skb) - skb->head;
740 skb->csum_offset = offsetof(struct udphdr, check);
741 uh->check = ~csum_tcpudp_magic(src, dst, len,
742 IPPROTO_UDP, 0);
743 } else {
744 struct sk_buff *frags;
745
746 /*
747 * HW-checksum won't work as there are two or more
748 * fragments on the socket so that all csums of sk_buffs
749 * should be together
750 */
751 skb_walk_frags(skb, frags) {
752 csum = csum_add(csum, frags->csum);
753 hlen -= frags->len;
754 }
755
756 csum = skb_checksum(skb, offset, hlen, csum);
757 skb->ip_summed = CHECKSUM_NONE;
758
759 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
760 if (uh->check == 0)
761 uh->check = CSUM_MANGLED_0;
762 }
763 }
764 EXPORT_SYMBOL_GPL(udp4_hwcsum);
765
766 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
767 * for the simple case like when setting the checksum for a UDP tunnel.
768 */
769 void udp_set_csum(bool nocheck, struct sk_buff *skb,
770 __be32 saddr, __be32 daddr, int len)
771 {
772 struct udphdr *uh = udp_hdr(skb);
773
774 if (nocheck) {
775 uh->check = 0;
776 } else if (skb_is_gso(skb)) {
777 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
778 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
779 uh->check = 0;
780 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
781 if (uh->check == 0)
782 uh->check = CSUM_MANGLED_0;
783 } else {
784 skb->ip_summed = CHECKSUM_PARTIAL;
785 skb->csum_start = skb_transport_header(skb) - skb->head;
786 skb->csum_offset = offsetof(struct udphdr, check);
787 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
788 }
789 }
790 EXPORT_SYMBOL(udp_set_csum);
791
792 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
793 {
794 struct sock *sk = skb->sk;
795 struct inet_sock *inet = inet_sk(sk);
796 struct udphdr *uh;
797 int err = 0;
798 int is_udplite = IS_UDPLITE(sk);
799 int offset = skb_transport_offset(skb);
800 int len = skb->len - offset;
801 __wsum csum = 0;
802
803 /*
804 * Create a UDP header
805 */
806 uh = udp_hdr(skb);
807 uh->source = inet->inet_sport;
808 uh->dest = fl4->fl4_dport;
809 uh->len = htons(len);
810 uh->check = 0;
811
812 if (is_udplite) /* UDP-Lite */
813 csum = udplite_csum(skb);
814
815 else if (sk->sk_no_check_tx) { /* UDP csum disabled */
816
817 skb->ip_summed = CHECKSUM_NONE;
818 goto send;
819
820 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
821
822 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
823 goto send;
824
825 } else
826 csum = udp_csum(skb);
827
828 /* add protocol-dependent pseudo-header */
829 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
830 sk->sk_protocol, csum);
831 if (uh->check == 0)
832 uh->check = CSUM_MANGLED_0;
833
834 send:
835 err = ip_send_skb(sock_net(sk), skb);
836 if (err) {
837 if (err == -ENOBUFS && !inet->recverr) {
838 UDP_INC_STATS(sock_net(sk),
839 UDP_MIB_SNDBUFERRORS, is_udplite);
840 err = 0;
841 }
842 } else
843 UDP_INC_STATS(sock_net(sk),
844 UDP_MIB_OUTDATAGRAMS, is_udplite);
845 return err;
846 }
847
848 /*
849 * Push out all pending data as one UDP datagram. Socket is locked.
850 */
851 int udp_push_pending_frames(struct sock *sk)
852 {
853 struct udp_sock *up = udp_sk(sk);
854 struct inet_sock *inet = inet_sk(sk);
855 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
856 struct sk_buff *skb;
857 int err = 0;
858
859 skb = ip_finish_skb(sk, fl4);
860 if (!skb)
861 goto out;
862
863 err = udp_send_skb(skb, fl4);
864
865 out:
866 up->len = 0;
867 up->pending = 0;
868 return err;
869 }
870 EXPORT_SYMBOL(udp_push_pending_frames);
871
872 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
873 {
874 struct inet_sock *inet = inet_sk(sk);
875 struct udp_sock *up = udp_sk(sk);
876 struct flowi4 fl4_stack;
877 struct flowi4 *fl4;
878 int ulen = len;
879 struct ipcm_cookie ipc;
880 struct rtable *rt = NULL;
881 int free = 0;
882 int connected = 0;
883 __be32 daddr, faddr, saddr;
884 __be16 dport;
885 u8 tos;
886 int err, is_udplite = IS_UDPLITE(sk);
887 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
888 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
889 struct sk_buff *skb;
890 struct ip_options_data opt_copy;
891
892 if (len > 0xFFFF)
893 return -EMSGSIZE;
894
895 /*
896 * Check the flags.
897 */
898
899 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
900 return -EOPNOTSUPP;
901
902 ipc.opt = NULL;
903 ipc.tx_flags = 0;
904 ipc.ttl = 0;
905 ipc.tos = -1;
906
907 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
908
909 fl4 = &inet->cork.fl.u.ip4;
910 if (up->pending) {
911 /*
912 * There are pending frames.
913 * The socket lock must be held while it's corked.
914 */
915 lock_sock(sk);
916 if (likely(up->pending)) {
917 if (unlikely(up->pending != AF_INET)) {
918 release_sock(sk);
919 return -EINVAL;
920 }
921 goto do_append_data;
922 }
923 release_sock(sk);
924 }
925 ulen += sizeof(struct udphdr);
926
927 /*
928 * Get and verify the address.
929 */
930 if (msg->msg_name) {
931 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
932 if (msg->msg_namelen < sizeof(*usin))
933 return -EINVAL;
934 if (usin->sin_family != AF_INET) {
935 if (usin->sin_family != AF_UNSPEC)
936 return -EAFNOSUPPORT;
937 }
938
939 daddr = usin->sin_addr.s_addr;
940 dport = usin->sin_port;
941 if (dport == 0)
942 return -EINVAL;
943 } else {
944 if (sk->sk_state != TCP_ESTABLISHED)
945 return -EDESTADDRREQ;
946 daddr = inet->inet_daddr;
947 dport = inet->inet_dport;
948 /* Open fast path for connected socket.
949 Route will not be used, if at least one option is set.
950 */
951 connected = 1;
952 }
953
954 ipc.sockc.tsflags = sk->sk_tsflags;
955 ipc.addr = inet->inet_saddr;
956 ipc.oif = sk->sk_bound_dev_if;
957
958 if (msg->msg_controllen) {
959 err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6);
960 if (unlikely(err)) {
961 kfree(ipc.opt);
962 return err;
963 }
964 if (ipc.opt)
965 free = 1;
966 connected = 0;
967 }
968 if (!ipc.opt) {
969 struct ip_options_rcu *inet_opt;
970
971 rcu_read_lock();
972 inet_opt = rcu_dereference(inet->inet_opt);
973 if (inet_opt) {
974 memcpy(&opt_copy, inet_opt,
975 sizeof(*inet_opt) + inet_opt->opt.optlen);
976 ipc.opt = &opt_copy.opt;
977 }
978 rcu_read_unlock();
979 }
980
981 saddr = ipc.addr;
982 ipc.addr = faddr = daddr;
983
984 sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags);
985
986 if (ipc.opt && ipc.opt->opt.srr) {
987 if (!daddr)
988 return -EINVAL;
989 faddr = ipc.opt->opt.faddr;
990 connected = 0;
991 }
992 tos = get_rttos(&ipc, inet);
993 if (sock_flag(sk, SOCK_LOCALROUTE) ||
994 (msg->msg_flags & MSG_DONTROUTE) ||
995 (ipc.opt && ipc.opt->opt.is_strictroute)) {
996 tos |= RTO_ONLINK;
997 connected = 0;
998 }
999
1000 if (ipv4_is_multicast(daddr)) {
1001 if (!ipc.oif)
1002 ipc.oif = inet->mc_index;
1003 if (!saddr)
1004 saddr = inet->mc_addr;
1005 connected = 0;
1006 } else if (!ipc.oif)
1007 ipc.oif = inet->uc_index;
1008
1009 if (connected)
1010 rt = (struct rtable *)sk_dst_check(sk, 0);
1011
1012 if (!rt) {
1013 struct net *net = sock_net(sk);
1014 __u8 flow_flags = inet_sk_flowi_flags(sk);
1015
1016 fl4 = &fl4_stack;
1017
1018 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1019 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1020 flow_flags,
1021 faddr, saddr, dport, inet->inet_sport);
1022
1023 if (!saddr && ipc.oif) {
1024 err = l3mdev_get_saddr(net, ipc.oif, fl4);
1025 if (err < 0)
1026 goto out;
1027 }
1028
1029 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1030 rt = ip_route_output_flow(net, fl4, sk);
1031 if (IS_ERR(rt)) {
1032 err = PTR_ERR(rt);
1033 rt = NULL;
1034 if (err == -ENETUNREACH)
1035 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1036 goto out;
1037 }
1038
1039 err = -EACCES;
1040 if ((rt->rt_flags & RTCF_BROADCAST) &&
1041 !sock_flag(sk, SOCK_BROADCAST))
1042 goto out;
1043 if (connected)
1044 sk_dst_set(sk, dst_clone(&rt->dst));
1045 }
1046
1047 if (msg->msg_flags&MSG_CONFIRM)
1048 goto do_confirm;
1049 back_from_confirm:
1050
1051 saddr = fl4->saddr;
1052 if (!ipc.addr)
1053 daddr = ipc.addr = fl4->daddr;
1054
1055 /* Lockless fast path for the non-corking case. */
1056 if (!corkreq) {
1057 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1058 sizeof(struct udphdr), &ipc, &rt,
1059 msg->msg_flags);
1060 err = PTR_ERR(skb);
1061 if (!IS_ERR_OR_NULL(skb))
1062 err = udp_send_skb(skb, fl4);
1063 goto out;
1064 }
1065
1066 lock_sock(sk);
1067 if (unlikely(up->pending)) {
1068 /* The socket is already corked while preparing it. */
1069 /* ... which is an evident application bug. --ANK */
1070 release_sock(sk);
1071
1072 net_dbg_ratelimited("cork app bug 2\n");
1073 err = -EINVAL;
1074 goto out;
1075 }
1076 /*
1077 * Now cork the socket to pend data.
1078 */
1079 fl4 = &inet->cork.fl.u.ip4;
1080 fl4->daddr = daddr;
1081 fl4->saddr = saddr;
1082 fl4->fl4_dport = dport;
1083 fl4->fl4_sport = inet->inet_sport;
1084 up->pending = AF_INET;
1085
1086 do_append_data:
1087 up->len += ulen;
1088 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1089 sizeof(struct udphdr), &ipc, &rt,
1090 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1091 if (err)
1092 udp_flush_pending_frames(sk);
1093 else if (!corkreq)
1094 err = udp_push_pending_frames(sk);
1095 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1096 up->pending = 0;
1097 release_sock(sk);
1098
1099 out:
1100 ip_rt_put(rt);
1101 if (free)
1102 kfree(ipc.opt);
1103 if (!err)
1104 return len;
1105 /*
1106 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1107 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1108 * we don't have a good statistic (IpOutDiscards but it can be too many
1109 * things). We could add another new stat but at least for now that
1110 * seems like overkill.
1111 */
1112 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1113 UDP_INC_STATS(sock_net(sk),
1114 UDP_MIB_SNDBUFERRORS, is_udplite);
1115 }
1116 return err;
1117
1118 do_confirm:
1119 dst_confirm(&rt->dst);
1120 if (!(msg->msg_flags&MSG_PROBE) || len)
1121 goto back_from_confirm;
1122 err = 0;
1123 goto out;
1124 }
1125 EXPORT_SYMBOL(udp_sendmsg);
1126
1127 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1128 size_t size, int flags)
1129 {
1130 struct inet_sock *inet = inet_sk(sk);
1131 struct udp_sock *up = udp_sk(sk);
1132 int ret;
1133
1134 if (flags & MSG_SENDPAGE_NOTLAST)
1135 flags |= MSG_MORE;
1136
1137 if (!up->pending) {
1138 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1139
1140 /* Call udp_sendmsg to specify destination address which
1141 * sendpage interface can't pass.
1142 * This will succeed only when the socket is connected.
1143 */
1144 ret = udp_sendmsg(sk, &msg, 0);
1145 if (ret < 0)
1146 return ret;
1147 }
1148
1149 lock_sock(sk);
1150
1151 if (unlikely(!up->pending)) {
1152 release_sock(sk);
1153
1154 net_dbg_ratelimited("udp cork app bug 3\n");
1155 return -EINVAL;
1156 }
1157
1158 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1159 page, offset, size, flags);
1160 if (ret == -EOPNOTSUPP) {
1161 release_sock(sk);
1162 return sock_no_sendpage(sk->sk_socket, page, offset,
1163 size, flags);
1164 }
1165 if (ret < 0) {
1166 udp_flush_pending_frames(sk);
1167 goto out;
1168 }
1169
1170 up->len += size;
1171 if (!(up->corkflag || (flags&MSG_MORE)))
1172 ret = udp_push_pending_frames(sk);
1173 if (!ret)
1174 ret = size;
1175 out:
1176 release_sock(sk);
1177 return ret;
1178 }
1179
1180 /**
1181 * first_packet_length - return length of first packet in receive queue
1182 * @sk: socket
1183 *
1184 * Drops all bad checksum frames, until a valid one is found.
1185 * Returns the length of found skb, or -1 if none is found.
1186 */
1187 static int first_packet_length(struct sock *sk)
1188 {
1189 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1190 struct sk_buff *skb;
1191 int res;
1192
1193 __skb_queue_head_init(&list_kill);
1194
1195 spin_lock_bh(&rcvq->lock);
1196 while ((skb = skb_peek(rcvq)) != NULL &&
1197 udp_lib_checksum_complete(skb)) {
1198 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1199 IS_UDPLITE(sk));
1200 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1201 IS_UDPLITE(sk));
1202 atomic_inc(&sk->sk_drops);
1203 __skb_unlink(skb, rcvq);
1204 __skb_queue_tail(&list_kill, skb);
1205 }
1206 res = skb ? skb->len : -1;
1207 spin_unlock_bh(&rcvq->lock);
1208
1209 if (!skb_queue_empty(&list_kill)) {
1210 bool slow = lock_sock_fast(sk);
1211
1212 __skb_queue_purge(&list_kill);
1213 sk_mem_reclaim_partial(sk);
1214 unlock_sock_fast(sk, slow);
1215 }
1216 return res;
1217 }
1218
1219 /*
1220 * IOCTL requests applicable to the UDP protocol
1221 */
1222
1223 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1224 {
1225 switch (cmd) {
1226 case SIOCOUTQ:
1227 {
1228 int amount = sk_wmem_alloc_get(sk);
1229
1230 return put_user(amount, (int __user *)arg);
1231 }
1232
1233 case SIOCINQ:
1234 {
1235 int amount = max_t(int, 0, first_packet_length(sk));
1236
1237 return put_user(amount, (int __user *)arg);
1238 }
1239
1240 default:
1241 return -ENOIOCTLCMD;
1242 }
1243
1244 return 0;
1245 }
1246 EXPORT_SYMBOL(udp_ioctl);
1247
1248 /*
1249 * This should be easy, if there is something there we
1250 * return it, otherwise we block.
1251 */
1252
1253 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1254 int flags, int *addr_len)
1255 {
1256 struct inet_sock *inet = inet_sk(sk);
1257 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1258 struct sk_buff *skb;
1259 unsigned int ulen, copied;
1260 int peeked, peeking, off;
1261 int err;
1262 int is_udplite = IS_UDPLITE(sk);
1263 bool checksum_valid = false;
1264 bool slow;
1265
1266 if (flags & MSG_ERRQUEUE)
1267 return ip_recv_error(sk, msg, len, addr_len);
1268
1269 try_again:
1270 peeking = off = sk_peek_offset(sk, flags);
1271 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1272 &peeked, &off, &err);
1273 if (!skb)
1274 return err;
1275
1276 ulen = skb->len;
1277 copied = len;
1278 if (copied > ulen - off)
1279 copied = ulen - off;
1280 else if (copied < ulen)
1281 msg->msg_flags |= MSG_TRUNC;
1282
1283 /*
1284 * If checksum is needed at all, try to do it while copying the
1285 * data. If the data is truncated, or if we only want a partial
1286 * coverage checksum (UDP-Lite), do it before the copy.
1287 */
1288
1289 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov || peeking) {
1290 checksum_valid = !udp_lib_checksum_complete(skb);
1291 if (!checksum_valid)
1292 goto csum_copy_err;
1293 }
1294
1295 if (checksum_valid || skb_csum_unnecessary(skb))
1296 err = skb_copy_datagram_msg(skb, off, msg, copied);
1297 else {
1298 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1299
1300 if (err == -EINVAL)
1301 goto csum_copy_err;
1302 }
1303
1304 if (unlikely(err)) {
1305 trace_kfree_skb(skb, udp_recvmsg);
1306 if (!peeked) {
1307 atomic_inc(&sk->sk_drops);
1308 UDP_INC_STATS(sock_net(sk),
1309 UDP_MIB_INERRORS, is_udplite);
1310 }
1311 skb_free_datagram_locked(sk, skb);
1312 return err;
1313 }
1314
1315 if (!peeked)
1316 UDP_INC_STATS(sock_net(sk),
1317 UDP_MIB_INDATAGRAMS, is_udplite);
1318
1319 sock_recv_ts_and_drops(msg, sk, skb);
1320
1321 /* Copy the address. */
1322 if (sin) {
1323 sin->sin_family = AF_INET;
1324 sin->sin_port = udp_hdr(skb)->source;
1325 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1326 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1327 *addr_len = sizeof(*sin);
1328 }
1329 if (inet->cmsg_flags)
1330 ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr) + off);
1331
1332 err = copied;
1333 if (flags & MSG_TRUNC)
1334 err = ulen;
1335
1336 __skb_free_datagram_locked(sk, skb, peeking ? -err : err);
1337 return err;
1338
1339 csum_copy_err:
1340 slow = lock_sock_fast(sk);
1341 if (!skb_kill_datagram(sk, skb, flags)) {
1342 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1343 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1344 }
1345 unlock_sock_fast(sk, slow);
1346
1347 /* starting over for a new packet, but check if we need to yield */
1348 cond_resched();
1349 msg->msg_flags &= ~MSG_TRUNC;
1350 goto try_again;
1351 }
1352
1353 int udp_disconnect(struct sock *sk, int flags)
1354 {
1355 struct inet_sock *inet = inet_sk(sk);
1356 /*
1357 * 1003.1g - break association.
1358 */
1359
1360 sk->sk_state = TCP_CLOSE;
1361 inet->inet_daddr = 0;
1362 inet->inet_dport = 0;
1363 sock_rps_reset_rxhash(sk);
1364 sk->sk_bound_dev_if = 0;
1365 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1366 inet_reset_saddr(sk);
1367
1368 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1369 sk->sk_prot->unhash(sk);
1370 inet->inet_sport = 0;
1371 }
1372 sk_dst_reset(sk);
1373 return 0;
1374 }
1375 EXPORT_SYMBOL(udp_disconnect);
1376
1377 void udp_lib_unhash(struct sock *sk)
1378 {
1379 if (sk_hashed(sk)) {
1380 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1381 struct udp_hslot *hslot, *hslot2;
1382
1383 hslot = udp_hashslot(udptable, sock_net(sk),
1384 udp_sk(sk)->udp_port_hash);
1385 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1386
1387 spin_lock_bh(&hslot->lock);
1388 if (rcu_access_pointer(sk->sk_reuseport_cb))
1389 reuseport_detach_sock(sk);
1390 if (sk_del_node_init_rcu(sk)) {
1391 hslot->count--;
1392 inet_sk(sk)->inet_num = 0;
1393 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1394
1395 spin_lock(&hslot2->lock);
1396 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1397 hslot2->count--;
1398 spin_unlock(&hslot2->lock);
1399 }
1400 spin_unlock_bh(&hslot->lock);
1401 }
1402 }
1403 EXPORT_SYMBOL(udp_lib_unhash);
1404
1405 /*
1406 * inet_rcv_saddr was changed, we must rehash secondary hash
1407 */
1408 void udp_lib_rehash(struct sock *sk, u16 newhash)
1409 {
1410 if (sk_hashed(sk)) {
1411 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1412 struct udp_hslot *hslot, *hslot2, *nhslot2;
1413
1414 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1415 nhslot2 = udp_hashslot2(udptable, newhash);
1416 udp_sk(sk)->udp_portaddr_hash = newhash;
1417
1418 if (hslot2 != nhslot2 ||
1419 rcu_access_pointer(sk->sk_reuseport_cb)) {
1420 hslot = udp_hashslot(udptable, sock_net(sk),
1421 udp_sk(sk)->udp_port_hash);
1422 /* we must lock primary chain too */
1423 spin_lock_bh(&hslot->lock);
1424 if (rcu_access_pointer(sk->sk_reuseport_cb))
1425 reuseport_detach_sock(sk);
1426
1427 if (hslot2 != nhslot2) {
1428 spin_lock(&hslot2->lock);
1429 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1430 hslot2->count--;
1431 spin_unlock(&hslot2->lock);
1432
1433 spin_lock(&nhslot2->lock);
1434 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1435 &nhslot2->head);
1436 nhslot2->count++;
1437 spin_unlock(&nhslot2->lock);
1438 }
1439
1440 spin_unlock_bh(&hslot->lock);
1441 }
1442 }
1443 }
1444 EXPORT_SYMBOL(udp_lib_rehash);
1445
1446 static void udp_v4_rehash(struct sock *sk)
1447 {
1448 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1449 inet_sk(sk)->inet_rcv_saddr,
1450 inet_sk(sk)->inet_num);
1451 udp_lib_rehash(sk, new_hash);
1452 }
1453
1454 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1455 {
1456 int rc;
1457
1458 if (inet_sk(sk)->inet_daddr) {
1459 sock_rps_save_rxhash(sk, skb);
1460 sk_mark_napi_id(sk, skb);
1461 sk_incoming_cpu_update(sk);
1462 }
1463
1464 rc = __sock_queue_rcv_skb(sk, skb);
1465 if (rc < 0) {
1466 int is_udplite = IS_UDPLITE(sk);
1467
1468 /* Note that an ENOMEM error is charged twice */
1469 if (rc == -ENOMEM)
1470 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1471 is_udplite);
1472 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1473 kfree_skb(skb);
1474 trace_udp_fail_queue_rcv_skb(rc, sk);
1475 return -1;
1476 }
1477
1478 return 0;
1479
1480 }
1481
1482 static struct static_key udp_encap_needed __read_mostly;
1483 void udp_encap_enable(void)
1484 {
1485 if (!static_key_enabled(&udp_encap_needed))
1486 static_key_slow_inc(&udp_encap_needed);
1487 }
1488 EXPORT_SYMBOL(udp_encap_enable);
1489
1490 /* returns:
1491 * -1: error
1492 * 0: success
1493 * >0: "udp encap" protocol resubmission
1494 *
1495 * Note that in the success and error cases, the skb is assumed to
1496 * have either been requeued or freed.
1497 */
1498 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1499 {
1500 struct udp_sock *up = udp_sk(sk);
1501 int rc;
1502 int is_udplite = IS_UDPLITE(sk);
1503
1504 /*
1505 * Charge it to the socket, dropping if the queue is full.
1506 */
1507 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1508 goto drop;
1509 nf_reset(skb);
1510
1511 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1512 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1513
1514 /*
1515 * This is an encapsulation socket so pass the skb to
1516 * the socket's udp_encap_rcv() hook. Otherwise, just
1517 * fall through and pass this up the UDP socket.
1518 * up->encap_rcv() returns the following value:
1519 * =0 if skb was successfully passed to the encap
1520 * handler or was discarded by it.
1521 * >0 if skb should be passed on to UDP.
1522 * <0 if skb should be resubmitted as proto -N
1523 */
1524
1525 /* if we're overly short, let UDP handle it */
1526 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1527 if (encap_rcv) {
1528 int ret;
1529
1530 /* Verify checksum before giving to encap */
1531 if (udp_lib_checksum_complete(skb))
1532 goto csum_error;
1533
1534 ret = encap_rcv(sk, skb);
1535 if (ret <= 0) {
1536 __UDP_INC_STATS(sock_net(sk),
1537 UDP_MIB_INDATAGRAMS,
1538 is_udplite);
1539 return -ret;
1540 }
1541 }
1542
1543 /* FALLTHROUGH -- it's a UDP Packet */
1544 }
1545
1546 /*
1547 * UDP-Lite specific tests, ignored on UDP sockets
1548 */
1549 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1550
1551 /*
1552 * MIB statistics other than incrementing the error count are
1553 * disabled for the following two types of errors: these depend
1554 * on the application settings, not on the functioning of the
1555 * protocol stack as such.
1556 *
1557 * RFC 3828 here recommends (sec 3.3): "There should also be a
1558 * way ... to ... at least let the receiving application block
1559 * delivery of packets with coverage values less than a value
1560 * provided by the application."
1561 */
1562 if (up->pcrlen == 0) { /* full coverage was set */
1563 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1564 UDP_SKB_CB(skb)->cscov, skb->len);
1565 goto drop;
1566 }
1567 /* The next case involves violating the min. coverage requested
1568 * by the receiver. This is subtle: if receiver wants x and x is
1569 * greater than the buffersize/MTU then receiver will complain
1570 * that it wants x while sender emits packets of smaller size y.
1571 * Therefore the above ...()->partial_cov statement is essential.
1572 */
1573 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1574 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1575 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1576 goto drop;
1577 }
1578 }
1579
1580 if (rcu_access_pointer(sk->sk_filter) &&
1581 udp_lib_checksum_complete(skb))
1582 goto csum_error;
1583
1584 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
1585 goto drop;
1586
1587 udp_csum_pull_header(skb);
1588 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
1589 __UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1590 is_udplite);
1591 goto drop;
1592 }
1593
1594 rc = 0;
1595
1596 ipv4_pktinfo_prepare(sk, skb);
1597 bh_lock_sock(sk);
1598 if (!sock_owned_by_user(sk))
1599 rc = __udp_queue_rcv_skb(sk, skb);
1600 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1601 bh_unlock_sock(sk);
1602 goto drop;
1603 }
1604 bh_unlock_sock(sk);
1605
1606 return rc;
1607
1608 csum_error:
1609 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1610 drop:
1611 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1612 atomic_inc(&sk->sk_drops);
1613 kfree_skb(skb);
1614 return -1;
1615 }
1616
1617 /* For TCP sockets, sk_rx_dst is protected by socket lock
1618 * For UDP, we use xchg() to guard against concurrent changes.
1619 */
1620 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1621 {
1622 struct dst_entry *old;
1623
1624 dst_hold(dst);
1625 old = xchg(&sk->sk_rx_dst, dst);
1626 dst_release(old);
1627 }
1628
1629 /*
1630 * Multicasts and broadcasts go to each listener.
1631 *
1632 * Note: called only from the BH handler context.
1633 */
1634 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1635 struct udphdr *uh,
1636 __be32 saddr, __be32 daddr,
1637 struct udp_table *udptable,
1638 int proto)
1639 {
1640 struct sock *sk, *first = NULL;
1641 unsigned short hnum = ntohs(uh->dest);
1642 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1643 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1644 unsigned int offset = offsetof(typeof(*sk), sk_node);
1645 int dif = skb->dev->ifindex;
1646 struct hlist_node *node;
1647 struct sk_buff *nskb;
1648
1649 if (use_hash2) {
1650 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1651 udp_table.mask;
1652 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask;
1653 start_lookup:
1654 hslot = &udp_table.hash2[hash2];
1655 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1656 }
1657
1658 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
1659 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
1660 uh->source, saddr, dif, hnum))
1661 continue;
1662
1663 if (!first) {
1664 first = sk;
1665 continue;
1666 }
1667 nskb = skb_clone(skb, GFP_ATOMIC);
1668
1669 if (unlikely(!nskb)) {
1670 atomic_inc(&sk->sk_drops);
1671 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
1672 IS_UDPLITE(sk));
1673 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
1674 IS_UDPLITE(sk));
1675 continue;
1676 }
1677 if (udp_queue_rcv_skb(sk, nskb) > 0)
1678 consume_skb(nskb);
1679 }
1680
1681 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
1682 if (use_hash2 && hash2 != hash2_any) {
1683 hash2 = hash2_any;
1684 goto start_lookup;
1685 }
1686
1687 if (first) {
1688 if (udp_queue_rcv_skb(first, skb) > 0)
1689 consume_skb(skb);
1690 } else {
1691 kfree_skb(skb);
1692 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
1693 proto == IPPROTO_UDPLITE);
1694 }
1695 return 0;
1696 }
1697
1698 /* Initialize UDP checksum. If exited with zero value (success),
1699 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1700 * Otherwise, csum completion requires chacksumming packet body,
1701 * including udp header and folding it to skb->csum.
1702 */
1703 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1704 int proto)
1705 {
1706 int err;
1707
1708 UDP_SKB_CB(skb)->partial_cov = 0;
1709 UDP_SKB_CB(skb)->cscov = skb->len;
1710
1711 if (proto == IPPROTO_UDPLITE) {
1712 err = udplite_checksum_init(skb, uh);
1713 if (err)
1714 return err;
1715 }
1716
1717 /* Note, we are only interested in != 0 or == 0, thus the
1718 * force to int.
1719 */
1720 return (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
1721 inet_compute_pseudo);
1722 }
1723
1724 /*
1725 * All we need to do is get the socket, and then do a checksum.
1726 */
1727
1728 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1729 int proto)
1730 {
1731 struct sock *sk;
1732 struct udphdr *uh;
1733 unsigned short ulen;
1734 struct rtable *rt = skb_rtable(skb);
1735 __be32 saddr, daddr;
1736 struct net *net = dev_net(skb->dev);
1737
1738 /*
1739 * Validate the packet.
1740 */
1741 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1742 goto drop; /* No space for header. */
1743
1744 uh = udp_hdr(skb);
1745 ulen = ntohs(uh->len);
1746 saddr = ip_hdr(skb)->saddr;
1747 daddr = ip_hdr(skb)->daddr;
1748
1749 if (ulen > skb->len)
1750 goto short_packet;
1751
1752 if (proto == IPPROTO_UDP) {
1753 /* UDP validates ulen. */
1754 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1755 goto short_packet;
1756 uh = udp_hdr(skb);
1757 }
1758
1759 if (udp4_csum_init(skb, uh, proto))
1760 goto csum_error;
1761
1762 sk = skb_steal_sock(skb);
1763 if (sk) {
1764 struct dst_entry *dst = skb_dst(skb);
1765 int ret;
1766
1767 if (unlikely(sk->sk_rx_dst != dst))
1768 udp_sk_rx_dst_set(sk, dst);
1769
1770 ret = udp_queue_rcv_skb(sk, skb);
1771 sock_put(sk);
1772 /* a return value > 0 means to resubmit the input, but
1773 * it wants the return to be -protocol, or 0
1774 */
1775 if (ret > 0)
1776 return -ret;
1777 return 0;
1778 }
1779
1780 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1781 return __udp4_lib_mcast_deliver(net, skb, uh,
1782 saddr, daddr, udptable, proto);
1783
1784 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1785 if (sk) {
1786 int ret;
1787
1788 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
1789 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
1790 inet_compute_pseudo);
1791
1792 ret = udp_queue_rcv_skb(sk, skb);
1793
1794 /* a return value > 0 means to resubmit the input, but
1795 * it wants the return to be -protocol, or 0
1796 */
1797 if (ret > 0)
1798 return -ret;
1799 return 0;
1800 }
1801
1802 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1803 goto drop;
1804 nf_reset(skb);
1805
1806 /* No socket. Drop packet silently, if checksum is wrong */
1807 if (udp_lib_checksum_complete(skb))
1808 goto csum_error;
1809
1810 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1811 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1812
1813 /*
1814 * Hmm. We got an UDP packet to a port to which we
1815 * don't wanna listen. Ignore it.
1816 */
1817 kfree_skb(skb);
1818 return 0;
1819
1820 short_packet:
1821 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1822 proto == IPPROTO_UDPLITE ? "Lite" : "",
1823 &saddr, ntohs(uh->source),
1824 ulen, skb->len,
1825 &daddr, ntohs(uh->dest));
1826 goto drop;
1827
1828 csum_error:
1829 /*
1830 * RFC1122: OK. Discards the bad packet silently (as far as
1831 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1832 */
1833 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1834 proto == IPPROTO_UDPLITE ? "Lite" : "",
1835 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1836 ulen);
1837 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1838 drop:
1839 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1840 kfree_skb(skb);
1841 return 0;
1842 }
1843
1844 /* We can only early demux multicast if there is a single matching socket.
1845 * If more than one socket found returns NULL
1846 */
1847 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
1848 __be16 loc_port, __be32 loc_addr,
1849 __be16 rmt_port, __be32 rmt_addr,
1850 int dif)
1851 {
1852 struct sock *sk, *result;
1853 unsigned short hnum = ntohs(loc_port);
1854 unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
1855 struct udp_hslot *hslot = &udp_table.hash[slot];
1856
1857 /* Do not bother scanning a too big list */
1858 if (hslot->count > 10)
1859 return NULL;
1860
1861 result = NULL;
1862 sk_for_each_rcu(sk, &hslot->head) {
1863 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
1864 rmt_port, rmt_addr, dif, hnum)) {
1865 if (result)
1866 return NULL;
1867 result = sk;
1868 }
1869 }
1870
1871 return result;
1872 }
1873
1874 /* For unicast we should only early demux connected sockets or we can
1875 * break forwarding setups. The chains here can be long so only check
1876 * if the first socket is an exact match and if not move on.
1877 */
1878 static struct sock *__udp4_lib_demux_lookup(struct net *net,
1879 __be16 loc_port, __be32 loc_addr,
1880 __be16 rmt_port, __be32 rmt_addr,
1881 int dif)
1882 {
1883 unsigned short hnum = ntohs(loc_port);
1884 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
1885 unsigned int slot2 = hash2 & udp_table.mask;
1886 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
1887 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
1888 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
1889 struct sock *sk;
1890
1891 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
1892 if (INET_MATCH(sk, net, acookie, rmt_addr,
1893 loc_addr, ports, dif))
1894 return sk;
1895 /* Only check first socket in chain */
1896 break;
1897 }
1898 return NULL;
1899 }
1900
1901 void udp_v4_early_demux(struct sk_buff *skb)
1902 {
1903 struct net *net = dev_net(skb->dev);
1904 const struct iphdr *iph;
1905 const struct udphdr *uh;
1906 struct sock *sk = NULL;
1907 struct dst_entry *dst;
1908 int dif = skb->dev->ifindex;
1909 int ours;
1910
1911 /* validate the packet */
1912 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
1913 return;
1914
1915 iph = ip_hdr(skb);
1916 uh = udp_hdr(skb);
1917
1918 if (skb->pkt_type == PACKET_BROADCAST ||
1919 skb->pkt_type == PACKET_MULTICAST) {
1920 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
1921
1922 if (!in_dev)
1923 return;
1924
1925 /* we are supposed to accept bcast packets */
1926 if (skb->pkt_type == PACKET_MULTICAST) {
1927 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
1928 iph->protocol);
1929 if (!ours)
1930 return;
1931 }
1932
1933 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
1934 uh->source, iph->saddr, dif);
1935 } else if (skb->pkt_type == PACKET_HOST) {
1936 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
1937 uh->source, iph->saddr, dif);
1938 }
1939
1940 if (!sk || !atomic_inc_not_zero_hint(&sk->sk_refcnt, 2))
1941 return;
1942
1943 skb->sk = sk;
1944 skb->destructor = sock_efree;
1945 dst = READ_ONCE(sk->sk_rx_dst);
1946
1947 if (dst)
1948 dst = dst_check(dst, 0);
1949 if (dst) {
1950 /* DST_NOCACHE can not be used without taking a reference */
1951 if (dst->flags & DST_NOCACHE) {
1952 if (likely(atomic_inc_not_zero(&dst->__refcnt)))
1953 skb_dst_set(skb, dst);
1954 } else {
1955 skb_dst_set_noref(skb, dst);
1956 }
1957 }
1958 }
1959
1960 int udp_rcv(struct sk_buff *skb)
1961 {
1962 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1963 }
1964
1965 void udp_destroy_sock(struct sock *sk)
1966 {
1967 struct udp_sock *up = udp_sk(sk);
1968 bool slow = lock_sock_fast(sk);
1969 udp_flush_pending_frames(sk);
1970 unlock_sock_fast(sk, slow);
1971 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1972 void (*encap_destroy)(struct sock *sk);
1973 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1974 if (encap_destroy)
1975 encap_destroy(sk);
1976 }
1977 }
1978
1979 /*
1980 * Socket option code for UDP
1981 */
1982 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1983 char __user *optval, unsigned int optlen,
1984 int (*push_pending_frames)(struct sock *))
1985 {
1986 struct udp_sock *up = udp_sk(sk);
1987 int val, valbool;
1988 int err = 0;
1989 int is_udplite = IS_UDPLITE(sk);
1990
1991 if (optlen < sizeof(int))
1992 return -EINVAL;
1993
1994 if (get_user(val, (int __user *)optval))
1995 return -EFAULT;
1996
1997 valbool = val ? 1 : 0;
1998
1999 switch (optname) {
2000 case UDP_CORK:
2001 if (val != 0) {
2002 up->corkflag = 1;
2003 } else {
2004 up->corkflag = 0;
2005 lock_sock(sk);
2006 push_pending_frames(sk);
2007 release_sock(sk);
2008 }
2009 break;
2010
2011 case UDP_ENCAP:
2012 switch (val) {
2013 case 0:
2014 case UDP_ENCAP_ESPINUDP:
2015 case UDP_ENCAP_ESPINUDP_NON_IKE:
2016 up->encap_rcv = xfrm4_udp_encap_rcv;
2017 /* FALLTHROUGH */
2018 case UDP_ENCAP_L2TPINUDP:
2019 up->encap_type = val;
2020 udp_encap_enable();
2021 break;
2022 default:
2023 err = -ENOPROTOOPT;
2024 break;
2025 }
2026 break;
2027
2028 case UDP_NO_CHECK6_TX:
2029 up->no_check6_tx = valbool;
2030 break;
2031
2032 case UDP_NO_CHECK6_RX:
2033 up->no_check6_rx = valbool;
2034 break;
2035
2036 /*
2037 * UDP-Lite's partial checksum coverage (RFC 3828).
2038 */
2039 /* The sender sets actual checksum coverage length via this option.
2040 * The case coverage > packet length is handled by send module. */
2041 case UDPLITE_SEND_CSCOV:
2042 if (!is_udplite) /* Disable the option on UDP sockets */
2043 return -ENOPROTOOPT;
2044 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2045 val = 8;
2046 else if (val > USHRT_MAX)
2047 val = USHRT_MAX;
2048 up->pcslen = val;
2049 up->pcflag |= UDPLITE_SEND_CC;
2050 break;
2051
2052 /* The receiver specifies a minimum checksum coverage value. To make
2053 * sense, this should be set to at least 8 (as done below). If zero is
2054 * used, this again means full checksum coverage. */
2055 case UDPLITE_RECV_CSCOV:
2056 if (!is_udplite) /* Disable the option on UDP sockets */
2057 return -ENOPROTOOPT;
2058 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2059 val = 8;
2060 else if (val > USHRT_MAX)
2061 val = USHRT_MAX;
2062 up->pcrlen = val;
2063 up->pcflag |= UDPLITE_RECV_CC;
2064 break;
2065
2066 default:
2067 err = -ENOPROTOOPT;
2068 break;
2069 }
2070
2071 return err;
2072 }
2073 EXPORT_SYMBOL(udp_lib_setsockopt);
2074
2075 int udp_setsockopt(struct sock *sk, int level, int optname,
2076 char __user *optval, unsigned int optlen)
2077 {
2078 if (level == SOL_UDP || level == SOL_UDPLITE)
2079 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2080 udp_push_pending_frames);
2081 return ip_setsockopt(sk, level, optname, optval, optlen);
2082 }
2083
2084 #ifdef CONFIG_COMPAT
2085 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2086 char __user *optval, unsigned int optlen)
2087 {
2088 if (level == SOL_UDP || level == SOL_UDPLITE)
2089 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2090 udp_push_pending_frames);
2091 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2092 }
2093 #endif
2094
2095 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2096 char __user *optval, int __user *optlen)
2097 {
2098 struct udp_sock *up = udp_sk(sk);
2099 int val, len;
2100
2101 if (get_user(len, optlen))
2102 return -EFAULT;
2103
2104 len = min_t(unsigned int, len, sizeof(int));
2105
2106 if (len < 0)
2107 return -EINVAL;
2108
2109 switch (optname) {
2110 case UDP_CORK:
2111 val = up->corkflag;
2112 break;
2113
2114 case UDP_ENCAP:
2115 val = up->encap_type;
2116 break;
2117
2118 case UDP_NO_CHECK6_TX:
2119 val = up->no_check6_tx;
2120 break;
2121
2122 case UDP_NO_CHECK6_RX:
2123 val = up->no_check6_rx;
2124 break;
2125
2126 /* The following two cannot be changed on UDP sockets, the return is
2127 * always 0 (which corresponds to the full checksum coverage of UDP). */
2128 case UDPLITE_SEND_CSCOV:
2129 val = up->pcslen;
2130 break;
2131
2132 case UDPLITE_RECV_CSCOV:
2133 val = up->pcrlen;
2134 break;
2135
2136 default:
2137 return -ENOPROTOOPT;
2138 }
2139
2140 if (put_user(len, optlen))
2141 return -EFAULT;
2142 if (copy_to_user(optval, &val, len))
2143 return -EFAULT;
2144 return 0;
2145 }
2146 EXPORT_SYMBOL(udp_lib_getsockopt);
2147
2148 int udp_getsockopt(struct sock *sk, int level, int optname,
2149 char __user *optval, int __user *optlen)
2150 {
2151 if (level == SOL_UDP || level == SOL_UDPLITE)
2152 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2153 return ip_getsockopt(sk, level, optname, optval, optlen);
2154 }
2155
2156 #ifdef CONFIG_COMPAT
2157 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2158 char __user *optval, int __user *optlen)
2159 {
2160 if (level == SOL_UDP || level == SOL_UDPLITE)
2161 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2162 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2163 }
2164 #endif
2165 /**
2166 * udp_poll - wait for a UDP event.
2167 * @file - file struct
2168 * @sock - socket
2169 * @wait - poll table
2170 *
2171 * This is same as datagram poll, except for the special case of
2172 * blocking sockets. If application is using a blocking fd
2173 * and a packet with checksum error is in the queue;
2174 * then it could get return from select indicating data available
2175 * but then block when reading it. Add special case code
2176 * to work around these arguably broken applications.
2177 */
2178 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2179 {
2180 unsigned int mask = datagram_poll(file, sock, wait);
2181 struct sock *sk = sock->sk;
2182
2183 sock_rps_record_flow(sk);
2184
2185 /* Check for false positives due to checksum errors */
2186 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2187 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2188 mask &= ~(POLLIN | POLLRDNORM);
2189
2190 return mask;
2191
2192 }
2193 EXPORT_SYMBOL(udp_poll);
2194
2195 struct proto udp_prot = {
2196 .name = "UDP",
2197 .owner = THIS_MODULE,
2198 .close = udp_lib_close,
2199 .connect = ip4_datagram_connect,
2200 .disconnect = udp_disconnect,
2201 .ioctl = udp_ioctl,
2202 .destroy = udp_destroy_sock,
2203 .setsockopt = udp_setsockopt,
2204 .getsockopt = udp_getsockopt,
2205 .sendmsg = udp_sendmsg,
2206 .recvmsg = udp_recvmsg,
2207 .sendpage = udp_sendpage,
2208 .backlog_rcv = __udp_queue_rcv_skb,
2209 .release_cb = ip4_datagram_release_cb,
2210 .hash = udp_lib_hash,
2211 .unhash = udp_lib_unhash,
2212 .rehash = udp_v4_rehash,
2213 .get_port = udp_v4_get_port,
2214 .memory_allocated = &udp_memory_allocated,
2215 .sysctl_mem = sysctl_udp_mem,
2216 .sysctl_wmem = &sysctl_udp_wmem_min,
2217 .sysctl_rmem = &sysctl_udp_rmem_min,
2218 .obj_size = sizeof(struct udp_sock),
2219 .h.udp_table = &udp_table,
2220 #ifdef CONFIG_COMPAT
2221 .compat_setsockopt = compat_udp_setsockopt,
2222 .compat_getsockopt = compat_udp_getsockopt,
2223 #endif
2224 .clear_sk = sk_prot_clear_portaddr_nulls,
2225 };
2226 EXPORT_SYMBOL(udp_prot);
2227
2228 /* ------------------------------------------------------------------------ */
2229 #ifdef CONFIG_PROC_FS
2230
2231 static struct sock *udp_get_first(struct seq_file *seq, int start)
2232 {
2233 struct sock *sk;
2234 struct udp_iter_state *state = seq->private;
2235 struct net *net = seq_file_net(seq);
2236
2237 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2238 ++state->bucket) {
2239 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2240
2241 if (hlist_empty(&hslot->head))
2242 continue;
2243
2244 spin_lock_bh(&hslot->lock);
2245 sk_for_each(sk, &hslot->head) {
2246 if (!net_eq(sock_net(sk), net))
2247 continue;
2248 if (sk->sk_family == state->family)
2249 goto found;
2250 }
2251 spin_unlock_bh(&hslot->lock);
2252 }
2253 sk = NULL;
2254 found:
2255 return sk;
2256 }
2257
2258 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2259 {
2260 struct udp_iter_state *state = seq->private;
2261 struct net *net = seq_file_net(seq);
2262
2263 do {
2264 sk = sk_next(sk);
2265 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2266
2267 if (!sk) {
2268 if (state->bucket <= state->udp_table->mask)
2269 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2270 return udp_get_first(seq, state->bucket + 1);
2271 }
2272 return sk;
2273 }
2274
2275 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2276 {
2277 struct sock *sk = udp_get_first(seq, 0);
2278
2279 if (sk)
2280 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2281 --pos;
2282 return pos ? NULL : sk;
2283 }
2284
2285 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2286 {
2287 struct udp_iter_state *state = seq->private;
2288 state->bucket = MAX_UDP_PORTS;
2289
2290 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2291 }
2292
2293 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2294 {
2295 struct sock *sk;
2296
2297 if (v == SEQ_START_TOKEN)
2298 sk = udp_get_idx(seq, 0);
2299 else
2300 sk = udp_get_next(seq, v);
2301
2302 ++*pos;
2303 return sk;
2304 }
2305
2306 static void udp_seq_stop(struct seq_file *seq, void *v)
2307 {
2308 struct udp_iter_state *state = seq->private;
2309
2310 if (state->bucket <= state->udp_table->mask)
2311 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2312 }
2313
2314 int udp_seq_open(struct inode *inode, struct file *file)
2315 {
2316 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2317 struct udp_iter_state *s;
2318 int err;
2319
2320 err = seq_open_net(inode, file, &afinfo->seq_ops,
2321 sizeof(struct udp_iter_state));
2322 if (err < 0)
2323 return err;
2324
2325 s = ((struct seq_file *)file->private_data)->private;
2326 s->family = afinfo->family;
2327 s->udp_table = afinfo->udp_table;
2328 return err;
2329 }
2330 EXPORT_SYMBOL(udp_seq_open);
2331
2332 /* ------------------------------------------------------------------------ */
2333 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2334 {
2335 struct proc_dir_entry *p;
2336 int rc = 0;
2337
2338 afinfo->seq_ops.start = udp_seq_start;
2339 afinfo->seq_ops.next = udp_seq_next;
2340 afinfo->seq_ops.stop = udp_seq_stop;
2341
2342 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2343 afinfo->seq_fops, afinfo);
2344 if (!p)
2345 rc = -ENOMEM;
2346 return rc;
2347 }
2348 EXPORT_SYMBOL(udp_proc_register);
2349
2350 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2351 {
2352 remove_proc_entry(afinfo->name, net->proc_net);
2353 }
2354 EXPORT_SYMBOL(udp_proc_unregister);
2355
2356 /* ------------------------------------------------------------------------ */
2357 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2358 int bucket)
2359 {
2360 struct inet_sock *inet = inet_sk(sp);
2361 __be32 dest = inet->inet_daddr;
2362 __be32 src = inet->inet_rcv_saddr;
2363 __u16 destp = ntohs(inet->inet_dport);
2364 __u16 srcp = ntohs(inet->inet_sport);
2365
2366 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2367 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2368 bucket, src, srcp, dest, destp, sp->sk_state,
2369 sk_wmem_alloc_get(sp),
2370 sk_rmem_alloc_get(sp),
2371 0, 0L, 0,
2372 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2373 0, sock_i_ino(sp),
2374 atomic_read(&sp->sk_refcnt), sp,
2375 atomic_read(&sp->sk_drops));
2376 }
2377
2378 int udp4_seq_show(struct seq_file *seq, void *v)
2379 {
2380 seq_setwidth(seq, 127);
2381 if (v == SEQ_START_TOKEN)
2382 seq_puts(seq, " sl local_address rem_address st tx_queue "
2383 "rx_queue tr tm->when retrnsmt uid timeout "
2384 "inode ref pointer drops");
2385 else {
2386 struct udp_iter_state *state = seq->private;
2387
2388 udp4_format_sock(v, seq, state->bucket);
2389 }
2390 seq_pad(seq, '\n');
2391 return 0;
2392 }
2393
2394 static const struct file_operations udp_afinfo_seq_fops = {
2395 .owner = THIS_MODULE,
2396 .open = udp_seq_open,
2397 .read = seq_read,
2398 .llseek = seq_lseek,
2399 .release = seq_release_net
2400 };
2401
2402 /* ------------------------------------------------------------------------ */
2403 static struct udp_seq_afinfo udp4_seq_afinfo = {
2404 .name = "udp",
2405 .family = AF_INET,
2406 .udp_table = &udp_table,
2407 .seq_fops = &udp_afinfo_seq_fops,
2408 .seq_ops = {
2409 .show = udp4_seq_show,
2410 },
2411 };
2412
2413 static int __net_init udp4_proc_init_net(struct net *net)
2414 {
2415 return udp_proc_register(net, &udp4_seq_afinfo);
2416 }
2417
2418 static void __net_exit udp4_proc_exit_net(struct net *net)
2419 {
2420 udp_proc_unregister(net, &udp4_seq_afinfo);
2421 }
2422
2423 static struct pernet_operations udp4_net_ops = {
2424 .init = udp4_proc_init_net,
2425 .exit = udp4_proc_exit_net,
2426 };
2427
2428 int __init udp4_proc_init(void)
2429 {
2430 return register_pernet_subsys(&udp4_net_ops);
2431 }
2432
2433 void udp4_proc_exit(void)
2434 {
2435 unregister_pernet_subsys(&udp4_net_ops);
2436 }
2437 #endif /* CONFIG_PROC_FS */
2438
2439 static __initdata unsigned long uhash_entries;
2440 static int __init set_uhash_entries(char *str)
2441 {
2442 ssize_t ret;
2443
2444 if (!str)
2445 return 0;
2446
2447 ret = kstrtoul(str, 0, &uhash_entries);
2448 if (ret)
2449 return 0;
2450
2451 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2452 uhash_entries = UDP_HTABLE_SIZE_MIN;
2453 return 1;
2454 }
2455 __setup("uhash_entries=", set_uhash_entries);
2456
2457 void __init udp_table_init(struct udp_table *table, const char *name)
2458 {
2459 unsigned int i;
2460
2461 table->hash = alloc_large_system_hash(name,
2462 2 * sizeof(struct udp_hslot),
2463 uhash_entries,
2464 21, /* one slot per 2 MB */
2465 0,
2466 &table->log,
2467 &table->mask,
2468 UDP_HTABLE_SIZE_MIN,
2469 64 * 1024);
2470
2471 table->hash2 = table->hash + (table->mask + 1);
2472 for (i = 0; i <= table->mask; i++) {
2473 INIT_HLIST_HEAD(&table->hash[i].head);
2474 table->hash[i].count = 0;
2475 spin_lock_init(&table->hash[i].lock);
2476 }
2477 for (i = 0; i <= table->mask; i++) {
2478 INIT_HLIST_HEAD(&table->hash2[i].head);
2479 table->hash2[i].count = 0;
2480 spin_lock_init(&table->hash2[i].lock);
2481 }
2482 }
2483
2484 u32 udp_flow_hashrnd(void)
2485 {
2486 static u32 hashrnd __read_mostly;
2487
2488 net_get_random_once(&hashrnd, sizeof(hashrnd));
2489
2490 return hashrnd;
2491 }
2492 EXPORT_SYMBOL(udp_flow_hashrnd);
2493
2494 void __init udp_init(void)
2495 {
2496 unsigned long limit;
2497
2498 udp_table_init(&udp_table, "UDP");
2499 limit = nr_free_buffer_pages() / 8;
2500 limit = max(limit, 128UL);
2501 sysctl_udp_mem[0] = limit / 4 * 3;
2502 sysctl_udp_mem[1] = limit;
2503 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2504
2505 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2506 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2507 }
CRIS linux kernel tree