?  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
blob08eed5e16df0d26340bbfabd96ac6dad77895465
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 struct hlist_nulls_node *node;
147 kuid_t uid = sock_i_uid(sk);
148
149 sk_nulls_for_each(sk2, node, &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 struct hlist_nulls_node *node;
181 kuid_t uid = sock_i_uid(sk);
182 int res = 0;
183
184 spin_lock(&hslot2->lock);
185 udp_portaddr_for_each_entry(sk2, node, &hslot2->head) {
186 if (net_eq(sock_net(sk2), net) &&
187 sk2 != sk &&
188 (udp_sk(sk2)->udp_port_hash == num) &&
189 (!sk2->sk_reuse || !sk->sk_reuse) &&
190 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
191 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
192 (!sk2->sk_reuseport || !sk->sk_reuseport ||
193 rcu_access_pointer(sk->sk_reuseport_cb) ||
194 !uid_eq(uid, sock_i_uid(sk2))) &&
195 saddr_comp(sk, sk2, true)) {
196 res = 1;
197 break;
198 }
199 }
200 spin_unlock(&hslot2->lock);
201 return res;
202 }
203
204 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot,
205 int (*saddr_same)(const struct sock *sk1,
206 const struct sock *sk2,
207 bool match_wildcard))
208 {
209 struct net *net = sock_net(sk);
210 struct hlist_nulls_node *node;
211 kuid_t uid = sock_i_uid(sk);
212 struct sock *sk2;
213
214 sk_nulls_for_each(sk2, node, &hslot->head) {
215 if (net_eq(sock_net(sk2), net) &&
216 sk2 != sk &&
217 sk2->sk_family == sk->sk_family &&
218 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
219 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
220 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
221 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
222 (*saddr_same)(sk, sk2, false)) {
223 return reuseport_add_sock(sk, sk2);
224 }
225 }
226
227 /* Initial allocation may have already happened via setsockopt */
228 if (!rcu_access_pointer(sk->sk_reuseport_cb))
229 return reuseport_alloc(sk);
230 return 0;
231 }
232
233 /**
234 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
235 *
236 * @sk: socket struct in question
237 * @snum: port number to look up
238 * @saddr_comp: AF-dependent comparison of bound local IP addresses
239 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
240 * with NULL address
241 */
242 int udp_lib_get_port(struct sock *sk, unsigned short snum,
243 int (*saddr_comp)(const struct sock *sk1,
244 const struct sock *sk2,
245 bool match_wildcard),
246 unsigned int hash2_nulladdr)
247 {
248 struct udp_hslot *hslot, *hslot2;
249 struct udp_table *udptable = sk->sk_prot->h.udp_table;
250 int error = 1;
251 struct net *net = sock_net(sk);
252
253 if (!snum) {
254 int low, high, remaining;
255 unsigned int rand;
256 unsigned short first, last;
257 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
258
259 inet_get_local_port_range(net, &low, &high);
260 remaining = (high - low) + 1;
261
262 rand = prandom_u32();
263 first = reciprocal_scale(rand, remaining) + low;
264 /*
265 * force rand to be an odd multiple of UDP_HTABLE_SIZE
266 */
267 rand = (rand | 1) * (udptable->mask + 1);
268 last = first + udptable->mask + 1;
269 do {
270 hslot = udp_hashslot(udptable, net, first);
271 bitmap_zero(bitmap, PORTS_PER_CHAIN);
272 spin_lock_bh(&hslot->lock);
273 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
274 saddr_comp, udptable->log);
275
276 snum = first;
277 /*
278 * Iterate on all possible values of snum for this hash.
279 * Using steps of an odd multiple of UDP_HTABLE_SIZE
280 * give us randomization and full range coverage.
281 */
282 do {
283 if (low <= snum && snum <= high &&
284 !test_bit(snum >> udptable->log, bitmap) &&
285 !inet_is_local_reserved_port(net, snum))
286 goto found;
287 snum += rand;
288 } while (snum != first);
289 spin_unlock_bh(&hslot->lock);
290 } while (++first != last);
291 goto fail;
292 } else {
293 hslot = udp_hashslot(udptable, net, snum);
294 spin_lock_bh(&hslot->lock);
295 if (hslot->count > 10) {
296 int exist;
297 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
298
299 slot2 &= udptable->mask;
300 hash2_nulladdr &= udptable->mask;
301
302 hslot2 = udp_hashslot2(udptable, slot2);
303 if (hslot->count < hslot2->count)
304 goto scan_primary_hash;
305
306 exist = udp_lib_lport_inuse2(net, snum, hslot2,
307 sk, saddr_comp);
308 if (!exist && (hash2_nulladdr != slot2)) {
309 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
310 exist = udp_lib_lport_inuse2(net, snum, hslot2,
311 sk, saddr_comp);
312 }
313 if (exist)
314 goto fail_unlock;
315 else
316 goto found;
317 }
318 scan_primary_hash:
319 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
320 saddr_comp, 0))
321 goto fail_unlock;
322 }
323 found:
324 inet_sk(sk)->inet_num = snum;
325 udp_sk(sk)->udp_port_hash = snum;
326 udp_sk(sk)->udp_portaddr_hash ^= snum;
327 if (sk_unhashed(sk)) {
328 if (sk->sk_reuseport &&
329 udp_reuseport_add_sock(sk, hslot, saddr_comp)) {
330 inet_sk(sk)->inet_num = 0;
331 udp_sk(sk)->udp_port_hash = 0;
332 udp_sk(sk)->udp_portaddr_hash ^= snum;
333 goto fail_unlock;
334 }
335
336 sk_nulls_add_node_rcu(sk, &hslot->head);
337 hslot->count++;
338 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
339
340 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
341 spin_lock(&hslot2->lock);
342 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
343 &hslot2->head);
344 hslot2->count++;
345 spin_unlock(&hslot2->lock);
346 }
347 error = 0;
348 fail_unlock:
349 spin_unlock_bh(&hslot->lock);
350 fail:
351 return error;
352 }
353 EXPORT_SYMBOL(udp_lib_get_port);
354
355 /* match_wildcard == true: 0.0.0.0 equals to any IPv4 addresses
356 * match_wildcard == false: addresses must be exactly the same, i.e.
357 * 0.0.0.0 only equals to 0.0.0.0
358 */
359 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2,
360 bool match_wildcard)
361 {
362 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
363
364 if (!ipv6_only_sock(sk2)) {
365 if (inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)
366 return 1;
367 if (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr)
368 return match_wildcard;
369 }
370 return 0;
371 }
372
373 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr,
374 unsigned int port)
375 {
376 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
377 }
378
379 int udp_v4_get_port(struct sock *sk, unsigned short snum)
380 {
381 unsigned int hash2_nulladdr =
382 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
383 unsigned int hash2_partial =
384 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
385
386 /* precompute partial secondary hash */
387 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
388 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
389 }
390
391 static inline int compute_score(struct sock *sk, struct net *net,
392 __be32 saddr, unsigned short hnum, __be16 sport,
393 __be32 daddr, __be16 dport, int dif)
394 {
395 int score;
396 struct inet_sock *inet;
397
398 if (!net_eq(sock_net(sk), net) ||
399 udp_sk(sk)->udp_port_hash != hnum ||
400 ipv6_only_sock(sk))
401 return -1;
402
403 score = (sk->sk_family == PF_INET) ? 2 : 1;
404 inet = inet_sk(sk);
405
406 if (inet->inet_rcv_saddr) {
407 if (inet->inet_rcv_saddr != daddr)
408 return -1;
409 score += 4;
410 }
411
412 if (inet->inet_daddr) {
413 if (inet->inet_daddr != saddr)
414 return -1;
415 score += 4;
416 }
417
418 if (inet->inet_dport) {
419 if (inet->inet_dport != sport)
420 return -1;
421 score += 4;
422 }
423
424 if (sk->sk_bound_dev_if) {
425 if (sk->sk_bound_dev_if != dif)
426 return -1;
427 score += 4;
428 }
429 if (sk->sk_incoming_cpu == raw_smp_processor_id())
430 score++;
431 return score;
432 }
433
434 /*
435 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
436 */
437 static inline int compute_score2(struct sock *sk, struct net *net,
438 __be32 saddr, __be16 sport,
439 __be32 daddr, unsigned int hnum, int dif)
440 {
441 int score;
442 struct inet_sock *inet;
443
444 if (!net_eq(sock_net(sk), net) ||
445 ipv6_only_sock(sk))
446 return -1;
447
448 inet = inet_sk(sk);
449
450 if (inet->inet_rcv_saddr != daddr ||
451 inet->inet_num != hnum)
452 return -1;
453
454 score = (sk->sk_family == PF_INET) ? 2 : 1;
455
456 if (inet->inet_daddr) {
457 if (inet->inet_daddr != saddr)
458 return -1;
459 score += 4;
460 }
461
462 if (inet->inet_dport) {
463 if (inet->inet_dport != sport)
464 return -1;
465 score += 4;
466 }
467
468 if (sk->sk_bound_dev_if) {
469 if (sk->sk_bound_dev_if != dif)
470 return -1;
471 score += 4;
472 }
473
474 if (sk->sk_incoming_cpu == raw_smp_processor_id())
475 score++;
476
477 return score;
478 }
479
480 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
481 const __u16 lport, const __be32 faddr,
482 const __be16 fport)
483 {
484 static u32 udp_ehash_secret __read_mostly;
485
486 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
487
488 return __inet_ehashfn(laddr, lport, faddr, fport,
489 udp_ehash_secret + net_hash_mix(net));
490 }
491
492 /* called with read_rcu_lock() */
493 static struct sock *udp4_lib_lookup2(struct net *net,
494 __be32 saddr, __be16 sport,
495 __be32 daddr, unsigned int hnum, int dif,
496 struct udp_hslot *hslot2, unsigned int slot2,
497 struct sk_buff *skb)
498 {
499 struct sock *sk, *result;
500 struct hlist_nulls_node *node;
501 int score, badness, matches = 0, reuseport = 0;
502 bool select_ok = true;
503 u32 hash = 0;
504
505 begin:
506 result = NULL;
507 badness = 0;
508 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
509 score = compute_score2(sk, net, saddr, sport,
510 daddr, hnum, dif);
511 if (score > badness) {
512 result = sk;
513 badness = score;
514 reuseport = sk->sk_reuseport;
515 if (reuseport) {
516 hash = udp_ehashfn(net, daddr, hnum,
517 saddr, sport);
518 if (select_ok) {
519 struct sock *sk2;
520
521 sk2 = reuseport_select_sock(sk, hash, skb,
522 sizeof(struct udphdr));
523 if (sk2) {
524 result = sk2;
525 select_ok = false;
526 goto found;
527 }
528 }
529 matches = 1;
530 }
531 } else if (score == badness && reuseport) {
532 matches++;
533 if (reciprocal_scale(hash, matches) == 0)
534 result = sk;
535 hash = next_pseudo_random32(hash);
536 }
537 }
538 /*
539 * if the nulls value we got at the end of this lookup is
540 * not the expected one, we must restart lookup.
541 * We probably met an item that was moved to another chain.
542 */
543 if (get_nulls_value(node) != slot2)
544 goto begin;
545 if (result) {
546 found:
547 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
548 result = NULL;
549 else if (unlikely(compute_score2(result, net, saddr, sport,
550 daddr, hnum, dif) < badness)) {
551 sock_put(result);
552 goto begin;
553 }
554 }
555 return result;
556 }
557
558 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
559 * harder than this. -DaveM
560 */
561 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
562 __be16 sport, __be32 daddr, __be16 dport,
563 int dif, struct udp_table *udptable, struct sk_buff *skb)
564 {
565 struct sock *sk, *result;
566 struct hlist_nulls_node *node;
567 unsigned short hnum = ntohs(dport);
568 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
569 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
570 int score, badness, matches = 0, reuseport = 0;
571 bool select_ok = true;
572 u32 hash = 0;
573
574 rcu_read_lock();
575 if (hslot->count > 10) {
576 hash2 = udp4_portaddr_hash(net, daddr, hnum);
577 slot2 = hash2 & udptable->mask;
578 hslot2 = &udptable->hash2[slot2];
579 if (hslot->count < hslot2->count)
580 goto begin;
581
582 result = udp4_lib_lookup2(net, saddr, sport,
583 daddr, hnum, dif,
584 hslot2, slot2, skb);
585 if (!result) {
586 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
587 slot2 = hash2 & udptable->mask;
588 hslot2 = &udptable->hash2[slot2];
589 if (hslot->count < hslot2->count)
590 goto begin;
591
592 result = udp4_lib_lookup2(net, saddr, sport,
593 htonl(INADDR_ANY), hnum, dif,
594 hslot2, slot2, skb);
595 }
596 rcu_read_unlock();
597 return result;
598 }
599 begin:
600 result = NULL;
601 badness = 0;
602 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
603 score = compute_score(sk, net, saddr, hnum, sport,
604 daddr, dport, dif);
605 if (score > badness) {
606 result = sk;
607 badness = score;
608 reuseport = sk->sk_reuseport;
609 if (reuseport) {
610 hash = udp_ehashfn(net, daddr, hnum,
611 saddr, sport);
612 if (select_ok) {
613 struct sock *sk2;
614
615 sk2 = reuseport_select_sock(sk, hash, skb,
616 sizeof(struct udphdr));
617 if (sk2) {
618 result = sk2;
619 select_ok = false;
620 goto found;
621 }
622 }
623 matches = 1;
624 }
625 } else if (score == badness && reuseport) {
626 matches++;
627 if (reciprocal_scale(hash, matches) == 0)
628 result = sk;
629 hash = next_pseudo_random32(hash);
630 }
631 }
632 /*
633 * if the nulls value we got at the end of this lookup is
634 * not the expected one, we must restart lookup.
635 * We probably met an item that was moved to another chain.
636 */
637 if (get_nulls_value(node) != slot)
638 goto begin;
639
640 if (result) {
641 found:
642 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
643 result = NULL;
644 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
645 daddr, dport, dif) < badness)) {
646 sock_put(result);
647 goto begin;
648 }
649 }
650 rcu_read_unlock();
651 return result;
652 }
653 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
654
655 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
656 __be16 sport, __be16 dport,
657 struct udp_table *udptable)
658 {
659 const struct iphdr *iph = ip_hdr(skb);
660
661 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
662 iph->daddr, dport, inet_iif(skb),
663 udptable, skb);
664 }
665
666 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
667 __be32 daddr, __be16 dport, int dif)
668 {
669 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif,
670 &udp_table, NULL);
671 }
672 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
673
674 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
675 __be16 loc_port, __be32 loc_addr,
676 __be16 rmt_port, __be32 rmt_addr,
677 int dif, unsigned short hnum)
678 {
679 struct inet_sock *inet = inet_sk(sk);
680
681 if (!net_eq(sock_net(sk), net) ||
682 udp_sk(sk)->udp_port_hash != hnum ||
683 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
684 (inet->inet_dport != rmt_port && inet->inet_dport) ||
685 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
686 ipv6_only_sock(sk) ||
687 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
688 return false;
689 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif))
690 return false;
691 return true;
692 }
693
694 /*
695 * This routine is called by the ICMP module when it gets some
696 * sort of error condition. If err < 0 then the socket should
697 * be closed and the error returned to the user. If err > 0
698 * it's just the icmp type << 8 | icmp code.
699 * Header points to the ip header of the error packet. We move
700 * on past this. Then (as it used to claim before adjustment)
701 * header points to the first 8 bytes of the udp header. We need
702 * to find the appropriate port.
703 */
704
705 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
706 {
707 struct inet_sock *inet;
708 const struct iphdr *iph = (const struct iphdr *)skb->data;
709 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
710 const int type = icmp_hdr(skb)->type;
711 const int code = icmp_hdr(skb)->code;
712 struct sock *sk;
713 int harderr;
714 int err;
715 struct net *net = dev_net(skb->dev);
716
717 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
718 iph->saddr, uh->source, skb->dev->ifindex, udptable,
719 NULL);
720 if (!sk) {
721 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
722 return; /* No socket for error */
723 }
724
725 err = 0;
726 harderr = 0;
727 inet = inet_sk(sk);
728
729 switch (type) {
730 default:
731 case ICMP_TIME_EXCEEDED:
732 err = EHOSTUNREACH;
733 break;
734 case ICMP_SOURCE_QUENCH:
735 goto out;
736 case ICMP_PARAMETERPROB:
737 err = EPROTO;
738 harderr = 1;
739 break;
740 case ICMP_DEST_UNREACH:
741 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
742 ipv4_sk_update_pmtu(skb, sk, info);
743 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
744 err = EMSGSIZE;
745 harderr = 1;
746 break;
747 }
748 goto out;
749 }
750 err = EHOSTUNREACH;
751 if (code <= NR_ICMP_UNREACH) {
752 harderr = icmp_err_convert[code].fatal;
753 err = icmp_err_convert[code].errno;
754 }
755 break;
756 case ICMP_REDIRECT:
757 ipv4_sk_redirect(skb, sk);
758 goto out;
759 }
760
761 /*
762 * RFC1122: OK. Passes ICMP errors back to application, as per
763 * 4.1.3.3.
764 */
765 if (!inet->recverr) {
766 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
767 goto out;
768 } else
769 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
770
771 sk->sk_err = err;
772 sk->sk_error_report(sk);
773 out:
774 sock_put(sk);
775 }
776
777 void udp_err(struct sk_buff *skb, u32 info)
778 {
779 __udp4_lib_err(skb, info, &udp_table);
780 }
781
782 /*
783 * Throw away all pending data and cancel the corking. Socket is locked.
784 */
785 void udp_flush_pending_frames(struct sock *sk)
786 {
787 struct udp_sock *up = udp_sk(sk);
788
789 if (up->pending) {
790 up->len = 0;
791 up->pending = 0;
792 ip_flush_pending_frames(sk);
793 }
794 }
795 EXPORT_SYMBOL(udp_flush_pending_frames);
796
797 /**
798 * udp4_hwcsum - handle outgoing HW checksumming
799 * @skb: sk_buff containing the filled-in UDP header
800 * (checksum field must be zeroed out)
801 * @src: source IP address
802 * @dst: destination IP address
803 */
804 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
805 {
806 struct udphdr *uh = udp_hdr(skb);
807 int offset = skb_transport_offset(skb);
808 int len = skb->len - offset;
809 int hlen = len;
810 __wsum csum = 0;
811
812 if (!skb_has_frag_list(skb)) {
813 /*
814 * Only one fragment on the socket.
815 */
816 skb->csum_start = skb_transport_header(skb) - skb->head;
817 skb->csum_offset = offsetof(struct udphdr, check);
818 uh->check = ~csum_tcpudp_magic(src, dst, len,
819 IPPROTO_UDP, 0);
820 } else {
821 struct sk_buff *frags;
822
823 /*
824 * HW-checksum won't work as there are two or more
825 * fragments on the socket so that all csums of sk_buffs
826 * should be together
827 */
828 skb_walk_frags(skb, frags) {
829 csum = csum_add(csum, frags->csum);
830 hlen -= frags->len;
831 }
832
833 csum = skb_checksum(skb, offset, hlen, csum);
834 skb->ip_summed = CHECKSUM_NONE;
835
836 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
837 if (uh->check == 0)
838 uh->check = CSUM_MANGLED_0;
839 }
840 }
841 EXPORT_SYMBOL_GPL(udp4_hwcsum);
842
843 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
844 * for the simple case like when setting the checksum for a UDP tunnel.
845 */
846 void udp_set_csum(bool nocheck, struct sk_buff *skb,
847 __be32 saddr, __be32 daddr, int len)
848 {
849 struct udphdr *uh = udp_hdr(skb);
850
851 if (nocheck) {
852 uh->check = 0;
853 } else if (skb_is_gso(skb)) {
854 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
855 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
856 uh->check = 0;
857 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
858 if (uh->check == 0)
859 uh->check = CSUM_MANGLED_0;
860 } else {
861 skb->ip_summed = CHECKSUM_PARTIAL;
862 skb->csum_start = skb_transport_header(skb) - skb->head;
863 skb->csum_offset = offsetof(struct udphdr, check);
864 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
865 }
866 }
867 EXPORT_SYMBOL(udp_set_csum);
868
869 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
870 {
871 struct sock *sk = skb->sk;
872 struct inet_sock *inet = inet_sk(sk);
873 struct udphdr *uh;
874 int err = 0;
875 int is_udplite = IS_UDPLITE(sk);
876 int offset = skb_transport_offset(skb);
877 int len = skb->len - offset;
878 __wsum csum = 0;
879
880 /*
881 * Create a UDP header
882 */
883 uh = udp_hdr(skb);
884 uh->source = inet->inet_sport;
885 uh->dest = fl4->fl4_dport;
886 uh->len = htons(len);
887 uh->check = 0;
888
889 if (is_udplite) /* UDP-Lite */
890 csum = udplite_csum(skb);
891
892 else if (sk->sk_no_check_tx) { /* UDP csum disabled */
893
894 skb->ip_summed = CHECKSUM_NONE;
895 goto send;
896
897 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
898
899 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
900 goto send;
901
902 } else
903 csum = udp_csum(skb);
904
905 /* add protocol-dependent pseudo-header */
906 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
907 sk->sk_protocol, csum);
908 if (uh->check == 0)
909 uh->check = CSUM_MANGLED_0;
910
911 send:
912 err = ip_send_skb(sock_net(sk), skb);
913 if (err) {
914 if (err == -ENOBUFS && !inet->recverr) {
915 UDP_INC_STATS_USER(sock_net(sk),
916 UDP_MIB_SNDBUFERRORS, is_udplite);
917 err = 0;
918 }
919 } else
920 UDP_INC_STATS_USER(sock_net(sk),
921 UDP_MIB_OUTDATAGRAMS, is_udplite);
922 return err;
923 }
924
925 /*
926 * Push out all pending data as one UDP datagram. Socket is locked.
927 */
928 int udp_push_pending_frames(struct sock *sk)
929 {
930 struct udp_sock *up = udp_sk(sk);
931 struct inet_sock *inet = inet_sk(sk);
932 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
933 struct sk_buff *skb;
934 int err = 0;
935
936 skb = ip_finish_skb(sk, fl4);
937 if (!skb)
938 goto out;
939
940 err = udp_send_skb(skb, fl4);
941
942 out:
943 up->len = 0;
944 up->pending = 0;
945 return err;
946 }
947 EXPORT_SYMBOL(udp_push_pending_frames);
948
949 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
950 {
951 struct inet_sock *inet = inet_sk(sk);
952 struct udp_sock *up = udp_sk(sk);
953 struct flowi4 fl4_stack;
954 struct flowi4 *fl4;
955 int ulen = len;
956 struct ipcm_cookie ipc;
957 struct rtable *rt = NULL;
958 int free = 0;
959 int connected = 0;
960 __be32 daddr, faddr, saddr;
961 __be16 dport;
962 u8 tos;
963 int err, is_udplite = IS_UDPLITE(sk);
964 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
965 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
966 struct sk_buff *skb;
967 struct ip_options_data opt_copy;
968
969 if (len > 0xFFFF)
970 return -EMSGSIZE;
971
972 /*
973 * Check the flags.
974 */
975
976 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
977 return -EOPNOTSUPP;
978
979 ipc.opt = NULL;
980 ipc.tx_flags = 0;
981 ipc.ttl = 0;
982 ipc.tos = -1;
983
984 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
985
986 fl4 = &inet->cork.fl.u.ip4;
987 if (up->pending) {
988 /*
989 * There are pending frames.
990 * The socket lock must be held while it's corked.
991 */
992 lock_sock(sk);
993 if (likely(up->pending)) {
994 if (unlikely(up->pending != AF_INET)) {
995 release_sock(sk);
996 return -EINVAL;
997 }
998 goto do_append_data;
999 }
1000 release_sock(sk);
1001 }
1002 ulen += sizeof(struct udphdr);
1003
1004 /*
1005 * Get and verify the address.
1006 */
1007 if (msg->msg_name) {
1008 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1009 if (msg->msg_namelen < sizeof(*usin))
1010 return -EINVAL;
1011 if (usin->sin_family != AF_INET) {
1012 if (usin->sin_family != AF_UNSPEC)
1013 return -EAFNOSUPPORT;
1014 }
1015
1016 daddr = usin->sin_addr.s_addr;
1017 dport = usin->sin_port;
1018 if (dport == 0)
1019 return -EINVAL;
1020 } else {
1021 if (sk->sk_state != TCP_ESTABLISHED)
1022 return -EDESTADDRREQ;
1023 daddr = inet->inet_daddr;
1024 dport = inet->inet_dport;
1025 /* Open fast path for connected socket.
1026 Route will not be used, if at least one option is set.
1027 */
1028 connected = 1;
1029 }
1030 ipc.addr = inet->inet_saddr;
1031
1032 ipc.oif = sk->sk_bound_dev_if;
1033
1034 sock_tx_timestamp(sk, &ipc.tx_flags);
1035
1036 if (msg->msg_controllen) {
1037 err = ip_cmsg_send(sock_net(sk), msg, &ipc,
1038 sk->sk_family == AF_INET6);
1039 if (unlikely(err)) {
1040 kfree(ipc.opt);
1041 return err;
1042 }
1043 if (ipc.opt)
1044 free = 1;
1045 connected = 0;
1046 }
1047 if (!ipc.opt) {
1048 struct ip_options_rcu *inet_opt;
1049
1050 rcu_read_lock();
1051 inet_opt = rcu_dereference(inet->inet_opt);
1052 if (inet_opt) {
1053 memcpy(&opt_copy, inet_opt,
1054 sizeof(*inet_opt) + inet_opt->opt.optlen);
1055 ipc.opt = &opt_copy.opt;
1056 }
1057 rcu_read_unlock();
1058 }
1059
1060 saddr = ipc.addr;
1061 ipc.addr = faddr = daddr;
1062
1063 if (ipc.opt && ipc.opt->opt.srr) {
1064 if (!daddr)
1065 return -EINVAL;
1066 faddr = ipc.opt->opt.faddr;
1067 connected = 0;
1068 }
1069 tos = get_rttos(&ipc, inet);
1070 if (sock_flag(sk, SOCK_LOCALROUTE) ||
1071 (msg->msg_flags & MSG_DONTROUTE) ||
1072 (ipc.opt && ipc.opt->opt.is_strictroute)) {
1073 tos |= RTO_ONLINK;
1074 connected = 0;
1075 }
1076
1077 if (ipv4_is_multicast(daddr)) {
1078 if (!ipc.oif)
1079 ipc.oif = inet->mc_index;
1080 if (!saddr)
1081 saddr = inet->mc_addr;
1082 connected = 0;
1083 } else if (!ipc.oif)
1084 ipc.oif = inet->uc_index;
1085
1086 if (connected)
1087 rt = (struct rtable *)sk_dst_check(sk, 0);
1088
1089 if (!rt) {
1090 struct net *net = sock_net(sk);
1091 __u8 flow_flags = inet_sk_flowi_flags(sk);
1092
1093 fl4 = &fl4_stack;
1094
1095 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1096 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1097 flow_flags,
1098 faddr, saddr, dport, inet->inet_sport);
1099
1100 if (!saddr && ipc.oif) {
1101 err = l3mdev_get_saddr(net, ipc.oif, fl4);
1102 if (err < 0)
1103 goto out;
1104 }
1105
1106 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1107 rt = ip_route_output_flow(net, fl4, sk);
1108 if (IS_ERR(rt)) {
1109 err = PTR_ERR(rt);
1110 rt = NULL;
1111 if (err == -ENETUNREACH)
1112 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1113 goto out;
1114 }
1115
1116 err = -EACCES;
1117 if ((rt->rt_flags & RTCF_BROADCAST) &&
1118 !sock_flag(sk, SOCK_BROADCAST))
1119 goto out;
1120 if (connected)
1121 sk_dst_set(sk, dst_clone(&rt->dst));
1122 }
1123
1124 if (msg->msg_flags&MSG_CONFIRM)
1125 goto do_confirm;
1126 back_from_confirm:
1127
1128 saddr = fl4->saddr;
1129 if (!ipc.addr)
1130 daddr = ipc.addr = fl4->daddr;
1131
1132 /* Lockless fast path for the non-corking case. */
1133 if (!corkreq) {
1134 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1135 sizeof(struct udphdr), &ipc, &rt,
1136 msg->msg_flags);
1137 err = PTR_ERR(skb);
1138 if (!IS_ERR_OR_NULL(skb))
1139 err = udp_send_skb(skb, fl4);
1140 goto out;
1141 }
1142
1143 lock_sock(sk);
1144 if (unlikely(up->pending)) {
1145 /* The socket is already corked while preparing it. */
1146 /* ... which is an evident application bug. --ANK */
1147 release_sock(sk);
1148
1149 net_dbg_ratelimited("cork app bug 2\n");
1150 err = -EINVAL;
1151 goto out;
1152 }
1153 /*
1154 * Now cork the socket to pend data.
1155 */
1156 fl4 = &inet->cork.fl.u.ip4;
1157 fl4->daddr = daddr;
1158 fl4->saddr = saddr;
1159 fl4->fl4_dport = dport;
1160 fl4->fl4_sport = inet->inet_sport;
1161 up->pending = AF_INET;
1162
1163 do_append_data:
1164 up->len += ulen;
1165 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1166 sizeof(struct udphdr), &ipc, &rt,
1167 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1168 if (err)
1169 udp_flush_pending_frames(sk);
1170 else if (!corkreq)
1171 err = udp_push_pending_frames(sk);
1172 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1173 up->pending = 0;
1174 release_sock(sk);
1175
1176 out:
1177 ip_rt_put(rt);
1178 if (free)
1179 kfree(ipc.opt);
1180 if (!err)
1181 return len;
1182 /*
1183 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1184 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1185 * we don't have a good statistic (IpOutDiscards but it can be too many
1186 * things). We could add another new stat but at least for now that
1187 * seems like overkill.
1188 */
1189 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1190 UDP_INC_STATS_USER(sock_net(sk),
1191 UDP_MIB_SNDBUFERRORS, is_udplite);
1192 }
1193 return err;
1194
1195 do_confirm:
1196 dst_confirm(&rt->dst);
1197 if (!(msg->msg_flags&MSG_PROBE) || len)
1198 goto back_from_confirm;
1199 err = 0;
1200 goto out;
1201 }
1202 EXPORT_SYMBOL(udp_sendmsg);
1203
1204 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1205 size_t size, int flags)
1206 {
1207 struct inet_sock *inet = inet_sk(sk);
1208 struct udp_sock *up = udp_sk(sk);
1209 int ret;
1210
1211 if (flags & MSG_SENDPAGE_NOTLAST)
1212 flags |= MSG_MORE;
1213
1214 if (!up->pending) {
1215 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1216
1217 /* Call udp_sendmsg to specify destination address which
1218 * sendpage interface can't pass.
1219 * This will succeed only when the socket is connected.
1220 */
1221 ret = udp_sendmsg(sk, &msg, 0);
1222 if (ret < 0)
1223 return ret;
1224 }
1225
1226 lock_sock(sk);
1227
1228 if (unlikely(!up->pending)) {
1229 release_sock(sk);
1230
1231 net_dbg_ratelimited("udp cork app bug 3\n");
1232 return -EINVAL;
1233 }
1234
1235 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1236 page, offset, size, flags);
1237 if (ret == -EOPNOTSUPP) {
1238 release_sock(sk);
1239 return sock_no_sendpage(sk->sk_socket, page, offset,
1240 size, flags);
1241 }
1242 if (ret < 0) {
1243 udp_flush_pending_frames(sk);
1244 goto out;
1245 }
1246
1247 up->len += size;
1248 if (!(up->corkflag || (flags&MSG_MORE)))
1249 ret = udp_push_pending_frames(sk);
1250 if (!ret)
1251 ret = size;
1252 out:
1253 release_sock(sk);
1254 return ret;
1255 }
1256
1257 /**
1258 * first_packet_length - return length of first packet in receive queue
1259 * @sk: socket
1260 *
1261 * Drops all bad checksum frames, until a valid one is found.
1262 * Returns the length of found skb, or 0 if none is found.
1263 */
1264 static unsigned int first_packet_length(struct sock *sk)
1265 {
1266 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1267 struct sk_buff *skb;
1268 unsigned int res;
1269
1270 __skb_queue_head_init(&list_kill);
1271
1272 spin_lock_bh(&rcvq->lock);
1273 while ((skb = skb_peek(rcvq)) != NULL &&
1274 udp_lib_checksum_complete(skb)) {
1275 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS,
1276 IS_UDPLITE(sk));
1277 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1278 IS_UDPLITE(sk));
1279 atomic_inc(&sk->sk_drops);
1280 __skb_unlink(skb, rcvq);
1281 __skb_queue_tail(&list_kill, skb);
1282 }
1283 res = skb ? skb->len : 0;
1284 spin_unlock_bh(&rcvq->lock);
1285
1286 if (!skb_queue_empty(&list_kill)) {
1287 bool slow = lock_sock_fast(sk);
1288
1289 __skb_queue_purge(&list_kill);
1290 sk_mem_reclaim_partial(sk);
1291 unlock_sock_fast(sk, slow);
1292 }
1293 return res;
1294 }
1295
1296 /*
1297 * IOCTL requests applicable to the UDP protocol
1298 */
1299
1300 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1301 {
1302 switch (cmd) {
1303 case SIOCOUTQ:
1304 {
1305 int amount = sk_wmem_alloc_get(sk);
1306
1307 return put_user(amount, (int __user *)arg);
1308 }
1309
1310 case SIOCINQ:
1311 {
1312 unsigned int amount = first_packet_length(sk);
1313
1314 if (amount)
1315 /*
1316 * We will only return the amount
1317 * of this packet since that is all
1318 * that will be read.
1319 */
1320 amount -= sizeof(struct udphdr);
1321
1322 return put_user(amount, (int __user *)arg);
1323 }
1324
1325 default:
1326 return -ENOIOCTLCMD;
1327 }
1328
1329 return 0;
1330 }
1331 EXPORT_SYMBOL(udp_ioctl);
1332
1333 /*
1334 * This should be easy, if there is something there we
1335 * return it, otherwise we block.
1336 */
1337
1338 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1339 int flags, int *addr_len)
1340 {
1341 struct inet_sock *inet = inet_sk(sk);
1342 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1343 struct sk_buff *skb;
1344 unsigned int ulen, copied;
1345 int peeked, off = 0;
1346 int err;
1347 int is_udplite = IS_UDPLITE(sk);
1348 bool checksum_valid = false;
1349 bool slow;
1350
1351 if (flags & MSG_ERRQUEUE)
1352 return ip_recv_error(sk, msg, len, addr_len);
1353
1354 try_again:
1355 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1356 &peeked, &off, &err);
1357 if (!skb)
1358 goto out;
1359
1360 ulen = skb->len - sizeof(struct udphdr);
1361 copied = len;
1362 if (copied > ulen)
1363 copied = ulen;
1364 else if (copied < ulen)
1365 msg->msg_flags |= MSG_TRUNC;
1366
1367 /*
1368 * If checksum is needed at all, try to do it while copying the
1369 * data. If the data is truncated, or if we only want a partial
1370 * coverage checksum (UDP-Lite), do it before the copy.
1371 */
1372
1373 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1374 checksum_valid = !udp_lib_checksum_complete(skb);
1375 if (!checksum_valid)
1376 goto csum_copy_err;
1377 }
1378
1379 if (checksum_valid || skb_csum_unnecessary(skb))
1380 err = skb_copy_datagram_msg(skb, sizeof(struct udphdr),
1381 msg, copied);
1382 else {
1383 err = skb_copy_and_csum_datagram_msg(skb, sizeof(struct udphdr),
1384 msg);
1385
1386 if (err == -EINVAL)
1387 goto csum_copy_err;
1388 }
1389
1390 if (unlikely(err)) {
1391 trace_kfree_skb(skb, udp_recvmsg);
1392 if (!peeked) {
1393 atomic_inc(&sk->sk_drops);
1394 UDP_INC_STATS_USER(sock_net(sk),
1395 UDP_MIB_INERRORS, is_udplite);
1396 }
1397 goto out_free;
1398 }
1399
1400 if (!peeked)
1401 UDP_INC_STATS_USER(sock_net(sk),
1402 UDP_MIB_INDATAGRAMS, is_udplite);
1403
1404 sock_recv_ts_and_drops(msg, sk, skb);
1405
1406 /* Copy the address. */
1407 if (sin) {
1408 sin->sin_family = AF_INET;
1409 sin->sin_port = udp_hdr(skb)->source;
1410 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1411 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1412 *addr_len = sizeof(*sin);
1413 }
1414 if (inet->cmsg_flags)
1415 ip_cmsg_recv_offset(msg, skb, sizeof(struct udphdr));
1416
1417 err = copied;
1418 if (flags & MSG_TRUNC)
1419 err = ulen;
1420
1421 out_free:
1422 skb_free_datagram_locked(sk, skb);
1423 out:
1424 return err;
1425
1426 csum_copy_err:
1427 slow = lock_sock_fast(sk);
1428 if (!skb_kill_datagram(sk, skb, flags)) {
1429 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1430 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1431 }
1432 unlock_sock_fast(sk, slow);
1433
1434 /* starting over for a new packet, but check if we need to yield */
1435 cond_resched();
1436 msg->msg_flags &= ~MSG_TRUNC;
1437 goto try_again;
1438 }
1439
1440 int udp_disconnect(struct sock *sk, int flags)
1441 {
1442 struct inet_sock *inet = inet_sk(sk);
1443 /*
1444 * 1003.1g - break association.
1445 */
1446
1447 sk->sk_state = TCP_CLOSE;
1448 inet->inet_daddr = 0;
1449 inet->inet_dport = 0;
1450 sock_rps_reset_rxhash(sk);
1451 sk->sk_bound_dev_if = 0;
1452 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1453 inet_reset_saddr(sk);
1454
1455 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1456 sk->sk_prot->unhash(sk);
1457 inet->inet_sport = 0;
1458 }
1459 sk_dst_reset(sk);
1460 return 0;
1461 }
1462 EXPORT_SYMBOL(udp_disconnect);
1463
1464 void udp_lib_unhash(struct sock *sk)
1465 {
1466 if (sk_hashed(sk)) {
1467 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1468 struct udp_hslot *hslot, *hslot2;
1469
1470 hslot = udp_hashslot(udptable, sock_net(sk),
1471 udp_sk(sk)->udp_port_hash);
1472 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1473
1474 spin_lock_bh(&hslot->lock);
1475 if (rcu_access_pointer(sk->sk_reuseport_cb))
1476 reuseport_detach_sock(sk);
1477 if (sk_nulls_del_node_init_rcu(sk)) {
1478 hslot->count--;
1479 inet_sk(sk)->inet_num = 0;
1480 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1481
1482 spin_lock(&hslot2->lock);
1483 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1484 hslot2->count--;
1485 spin_unlock(&hslot2->lock);
1486 }
1487 spin_unlock_bh(&hslot->lock);
1488 }
1489 }
1490 EXPORT_SYMBOL(udp_lib_unhash);
1491
1492 /*
1493 * inet_rcv_saddr was changed, we must rehash secondary hash
1494 */
1495 void udp_lib_rehash(struct sock *sk, u16 newhash)
1496 {
1497 if (sk_hashed(sk)) {
1498 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1499 struct udp_hslot *hslot, *hslot2, *nhslot2;
1500
1501 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1502 nhslot2 = udp_hashslot2(udptable, newhash);
1503 udp_sk(sk)->udp_portaddr_hash = newhash;
1504
1505 if (hslot2 != nhslot2 ||
1506 rcu_access_pointer(sk->sk_reuseport_cb)) {
1507 hslot = udp_hashslot(udptable, sock_net(sk),
1508 udp_sk(sk)->udp_port_hash);
1509 /* we must lock primary chain too */
1510 spin_lock_bh(&hslot->lock);
1511 if (rcu_access_pointer(sk->sk_reuseport_cb))
1512 reuseport_detach_sock(sk);
1513
1514 if (hslot2 != nhslot2) {
1515 spin_lock(&hslot2->lock);
1516 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1517 hslot2->count--;
1518 spin_unlock(&hslot2->lock);
1519
1520 spin_lock(&nhslot2->lock);
1521 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1522 &nhslot2->head);
1523 nhslot2->count++;
1524 spin_unlock(&nhslot2->lock);
1525 }
1526
1527 spin_unlock_bh(&hslot->lock);
1528 }
1529 }
1530 }
1531 EXPORT_SYMBOL(udp_lib_rehash);
1532
1533 static void udp_v4_rehash(struct sock *sk)
1534 {
1535 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1536 inet_sk(sk)->inet_rcv_saddr,
1537 inet_sk(sk)->inet_num);
1538 udp_lib_rehash(sk, new_hash);
1539 }
1540
1541 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1542 {
1543 int rc;
1544
1545 if (inet_sk(sk)->inet_daddr) {
1546 sock_rps_save_rxhash(sk, skb);
1547 sk_mark_napi_id(sk, skb);
1548 sk_incoming_cpu_update(sk);
1549 }
1550
1551 rc = sock_queue_rcv_skb(sk, skb);
1552 if (rc < 0) {
1553 int is_udplite = IS_UDPLITE(sk);
1554
1555 /* Note that an ENOMEM error is charged twice */
1556 if (rc == -ENOMEM)
1557 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1558 is_udplite);
1559 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1560 kfree_skb(skb);
1561 trace_udp_fail_queue_rcv_skb(rc, sk);
1562 return -1;
1563 }
1564
1565 return 0;
1566
1567 }
1568
1569 static struct static_key udp_encap_needed __read_mostly;
1570 void udp_encap_enable(void)
1571 {
1572 if (!static_key_enabled(&udp_encap_needed))
1573 static_key_slow_inc(&udp_encap_needed);
1574 }
1575 EXPORT_SYMBOL(udp_encap_enable);
1576
1577 /* returns:
1578 * -1: error
1579 * 0: success
1580 * >0: "udp encap" protocol resubmission
1581 *
1582 * Note that in the success and error cases, the skb is assumed to
1583 * have either been requeued or freed.
1584 */
1585 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1586 {
1587 struct udp_sock *up = udp_sk(sk);
1588 int rc;
1589 int is_udplite = IS_UDPLITE(sk);
1590
1591 /*
1592 * Charge it to the socket, dropping if the queue is full.
1593 */
1594 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1595 goto drop;
1596 nf_reset(skb);
1597
1598 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1599 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1600
1601 /*
1602 * This is an encapsulation socket so pass the skb to
1603 * the socket's udp_encap_rcv() hook. Otherwise, just
1604 * fall through and pass this up the UDP socket.
1605 * up->encap_rcv() returns the following value:
1606 * =0 if skb was successfully passed to the encap
1607 * handler or was discarded by it.
1608 * >0 if skb should be passed on to UDP.
1609 * <0 if skb should be resubmitted as proto -N
1610 */
1611
1612 /* if we're overly short, let UDP handle it */
1613 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1614 if (skb->len > sizeof(struct udphdr) && encap_rcv) {
1615 int ret;
1616
1617 /* Verify checksum before giving to encap */
1618 if (udp_lib_checksum_complete(skb))
1619 goto csum_error;
1620
1621 ret = encap_rcv(sk, skb);
1622 if (ret <= 0) {
1623 UDP_INC_STATS_BH(sock_net(sk),
1624 UDP_MIB_INDATAGRAMS,
1625 is_udplite);
1626 return -ret;
1627 }
1628 }
1629
1630 /* FALLTHROUGH -- it's a UDP Packet */
1631 }
1632
1633 /*
1634 * UDP-Lite specific tests, ignored on UDP sockets
1635 */
1636 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1637
1638 /*
1639 * MIB statistics other than incrementing the error count are
1640 * disabled for the following two types of errors: these depend
1641 * on the application settings, not on the functioning of the
1642 * protocol stack as such.
1643 *
1644 * RFC 3828 here recommends (sec 3.3): "There should also be a
1645 * way ... to ... at least let the receiving application block
1646 * delivery of packets with coverage values less than a value
1647 * provided by the application."
1648 */
1649 if (up->pcrlen == 0) { /* full coverage was set */
1650 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1651 UDP_SKB_CB(skb)->cscov, skb->len);
1652 goto drop;
1653 }
1654 /* The next case involves violating the min. coverage requested
1655 * by the receiver. This is subtle: if receiver wants x and x is
1656 * greater than the buffersize/MTU then receiver will complain
1657 * that it wants x while sender emits packets of smaller size y.
1658 * Therefore the above ...()->partial_cov statement is essential.
1659 */
1660 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1661 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1662 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1663 goto drop;
1664 }
1665 }
1666
1667 if (rcu_access_pointer(sk->sk_filter) &&
1668 udp_lib_checksum_complete(skb))
1669 goto csum_error;
1670
1671 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
1672 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1673 is_udplite);
1674 goto drop;
1675 }
1676
1677 rc = 0;
1678
1679 ipv4_pktinfo_prepare(sk, skb);
1680 bh_lock_sock(sk);
1681 if (!sock_owned_by_user(sk))
1682 rc = __udp_queue_rcv_skb(sk, skb);
1683 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1684 bh_unlock_sock(sk);
1685 goto drop;
1686 }
1687 bh_unlock_sock(sk);
1688
1689 return rc;
1690
1691 csum_error:
1692 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1693 drop:
1694 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1695 atomic_inc(&sk->sk_drops);
1696 kfree_skb(skb);
1697 return -1;
1698 }
1699
1700 static void flush_stack(struct sock **stack, unsigned int count,
1701 struct sk_buff *skb, unsigned int final)
1702 {
1703 unsigned int i;
1704 struct sk_buff *skb1 = NULL;
1705 struct sock *sk;
1706
1707 for (i = 0; i < count; i++) {
1708 sk = stack[i];
1709 if (likely(!skb1))
1710 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1711
1712 if (!skb1) {
1713 atomic_inc(&sk->sk_drops);
1714 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1715 IS_UDPLITE(sk));
1716 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1717 IS_UDPLITE(sk));
1718 }
1719
1720 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1721 skb1 = NULL;
1722
1723 sock_put(sk);
1724 }
1725 if (unlikely(skb1))
1726 kfree_skb(skb1);
1727 }
1728
1729 /* For TCP sockets, sk_rx_dst is protected by socket lock
1730 * For UDP, we use xchg() to guard against concurrent changes.
1731 */
1732 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1733 {
1734 struct dst_entry *old;
1735
1736 dst_hold(dst);
1737 old = xchg(&sk->sk_rx_dst, dst);
1738 dst_release(old);
1739 }
1740
1741 /*
1742 * Multicasts and broadcasts go to each listener.
1743 *
1744 * Note: called only from the BH handler context.
1745 */
1746 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1747 struct udphdr *uh,
1748 __be32 saddr, __be32 daddr,
1749 struct udp_table *udptable,
1750 int proto)
1751 {
1752 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1753 struct hlist_nulls_node *node;
1754 unsigned short hnum = ntohs(uh->dest);
1755 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1756 int dif = skb->dev->ifindex;
1757 unsigned int count = 0, offset = offsetof(typeof(*sk), sk_nulls_node);
1758 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1759 bool inner_flushed = false;
1760
1761 if (use_hash2) {
1762 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1763 udp_table.mask;
1764 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udp_table.mask;
1765 start_lookup:
1766 hslot = &udp_table.hash2[hash2];
1767 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1768 }
1769
1770 spin_lock(&hslot->lock);
1771 sk_nulls_for_each_entry_offset(sk, node, &hslot->head, offset) {
1772 if (__udp_is_mcast_sock(net, sk,
1773 uh->dest, daddr,
1774 uh->source, saddr,
1775 dif, hnum)) {
1776 if (unlikely(count == ARRAY_SIZE(stack))) {
1777 flush_stack(stack, count, skb, ~0);
1778 inner_flushed = true;
1779 count = 0;
1780 }
1781 stack[count++] = sk;
1782 sock_hold(sk);
1783 }
1784 }
1785
1786 spin_unlock(&hslot->lock);
1787
1788 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
1789 if (use_hash2 && hash2 != hash2_any) {
1790 hash2 = hash2_any;
1791 goto start_lookup;
1792 }
1793
1794 /*
1795 * do the slow work with no lock held
1796 */
1797 if (count) {
1798 flush_stack(stack, count, skb, count - 1);
1799 } else {
1800 if (!inner_flushed)
1801 UDP_INC_STATS_BH(net, UDP_MIB_IGNOREDMULTI,
1802 proto == IPPROTO_UDPLITE);
1803 consume_skb(skb);
1804 }
1805 return 0;
1806 }
1807
1808 /* Initialize UDP checksum. If exited with zero value (success),
1809 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1810 * Otherwise, csum completion requires chacksumming packet body,
1811 * including udp header and folding it to skb->csum.
1812 */
1813 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1814 int proto)
1815 {
1816 int err;
1817
1818 UDP_SKB_CB(skb)->partial_cov = 0;
1819 UDP_SKB_CB(skb)->cscov = skb->len;
1820
1821 if (proto == IPPROTO_UDPLITE) {
1822 err = udplite_checksum_init(skb, uh);
1823 if (err)
1824 return err;
1825 }
1826
1827 return skb_checksum_init_zero_check(skb, proto, uh->check,
1828 inet_compute_pseudo);
1829 }
1830
1831 /*
1832 * All we need to do is get the socket, and then do a checksum.
1833 */
1834
1835 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1836 int proto)
1837 {
1838 struct sock *sk;
1839 struct udphdr *uh;
1840 unsigned short ulen;
1841 struct rtable *rt = skb_rtable(skb);
1842 __be32 saddr, daddr;
1843 struct net *net = dev_net(skb->dev);
1844
1845 /*
1846 * Validate the packet.
1847 */
1848 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1849 goto drop; /* No space for header. */
1850
1851 uh = udp_hdr(skb);
1852 ulen = ntohs(uh->len);
1853 saddr = ip_hdr(skb)->saddr;
1854 daddr = ip_hdr(skb)->daddr;
1855
1856 if (ulen > skb->len)
1857 goto short_packet;
1858
1859 if (proto == IPPROTO_UDP) {
1860 /* UDP validates ulen. */
1861 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1862 goto short_packet;
1863 uh = udp_hdr(skb);
1864 }
1865
1866 if (udp4_csum_init(skb, uh, proto))
1867 goto csum_error;
1868
1869 sk = skb_steal_sock(skb);
1870 if (sk) {
1871 struct dst_entry *dst = skb_dst(skb);
1872 int ret;
1873
1874 if (unlikely(sk->sk_rx_dst != dst))
1875 udp_sk_rx_dst_set(sk, dst);
1876
1877 ret = udp_queue_rcv_skb(sk, skb);
1878 sock_put(sk);
1879 /* a return value > 0 means to resubmit the input, but
1880 * it wants the return to be -protocol, or 0
1881 */
1882 if (ret > 0)
1883 return -ret;
1884 return 0;
1885 }
1886
1887 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1888 return __udp4_lib_mcast_deliver(net, skb, uh,
1889 saddr, daddr, udptable, proto);
1890
1891 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1892 if (sk) {
1893 int ret;
1894
1895 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
1896 skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
1897 inet_compute_pseudo);
1898
1899 ret = udp_queue_rcv_skb(sk, skb);
1900 sock_put(sk);
1901
1902 /* a return value > 0 means to resubmit the input, but
1903 * it wants the return to be -protocol, or 0
1904 */
1905 if (ret > 0)
1906 return -ret;
1907 return 0;
1908 }
1909
1910 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1911 goto drop;
1912 nf_reset(skb);
1913
1914 /* No socket. Drop packet silently, if checksum is wrong */
1915 if (udp_lib_checksum_complete(skb))
1916 goto csum_error;
1917
1918 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1919 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1920
1921 /*
1922 * Hmm. We got an UDP packet to a port to which we
1923 * don't wanna listen. Ignore it.
1924 */
1925 kfree_skb(skb);
1926 return 0;
1927
1928 short_packet:
1929 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1930 proto == IPPROTO_UDPLITE ? "Lite" : "",
1931 &saddr, ntohs(uh->source),
1932 ulen, skb->len,
1933 &daddr, ntohs(uh->dest));
1934 goto drop;
1935
1936 csum_error:
1937 /*
1938 * RFC1122: OK. Discards the bad packet silently (as far as
1939 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1940 */
1941 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1942 proto == IPPROTO_UDPLITE ? "Lite" : "",
1943 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1944 ulen);
1945 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1946 drop:
1947 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1948 kfree_skb(skb);
1949 return 0;
1950 }
1951
1952 /* We can only early demux multicast if there is a single matching socket.
1953 * If more than one socket found returns NULL
1954 */
1955 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
1956 __be16 loc_port, __be32 loc_addr,
1957 __be16 rmt_port, __be32 rmt_addr,
1958 int dif)
1959 {
1960 struct sock *sk, *result;
1961 struct hlist_nulls_node *node;
1962 unsigned short hnum = ntohs(loc_port);
1963 unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask);
1964 struct udp_hslot *hslot = &udp_table.hash[slot];
1965
1966 /* Do not bother scanning a too big list */
1967 if (hslot->count > 10)
1968 return NULL;
1969
1970 rcu_read_lock();
1971 begin:
1972 count = 0;
1973 result = NULL;
1974 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
1975 if (__udp_is_mcast_sock(net, sk,
1976 loc_port, loc_addr,
1977 rmt_port, rmt_addr,
1978 dif, hnum)) {
1979 result = sk;
1980 ++count;
1981 }
1982 }
1983 /*
1984 * if the nulls value we got at the end of this lookup is
1985 * not the expected one, we must restart lookup.
1986 * We probably met an item that was moved to another chain.
1987 */
1988 if (get_nulls_value(node) != slot)
1989 goto begin;
1990
1991 if (result) {
1992 if (count != 1 ||
1993 unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
1994 result = NULL;
1995 else if (unlikely(!__udp_is_mcast_sock(net, result,
1996 loc_port, loc_addr,
1997 rmt_port, rmt_addr,
1998 dif, hnum))) {
1999 sock_put(result);
2000 result = NULL;
2001 }
2002 }
2003 rcu_read_unlock();
2004 return result;
2005 }
2006
2007 /* For unicast we should only early demux connected sockets or we can
2008 * break forwarding setups. The chains here can be long so only check
2009 * if the first socket is an exact match and if not move on.
2010 */
2011 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2012 __be16 loc_port, __be32 loc_addr,
2013 __be16 rmt_port, __be32 rmt_addr,
2014 int dif)
2015 {
2016 struct sock *sk, *result;
2017 struct hlist_nulls_node *node;
2018 unsigned short hnum = ntohs(loc_port);
2019 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
2020 unsigned int slot2 = hash2 & udp_table.mask;
2021 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2022 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2023 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2024
2025 rcu_read_lock();
2026 result = NULL;
2027 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
2028 if (INET_MATCH(sk, net, acookie,
2029 rmt_addr, loc_addr, ports, dif))
2030 result = sk;
2031 /* Only check first socket in chain */
2032 break;
2033 }
2034
2035 if (result) {
2036 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
2037 result = NULL;
2038 else if (unlikely(!INET_MATCH(sk, net, acookie,
2039 rmt_addr, loc_addr,
2040 ports, dif))) {
2041 sock_put(result);
2042 result = NULL;
2043 }
2044 }
2045 rcu_read_unlock();
2046 return result;
2047 }
2048
2049 void udp_v4_early_demux(struct sk_buff *skb)
2050 {
2051 struct net *net = dev_net(skb->dev);
2052 const struct iphdr *iph;
2053 const struct udphdr *uh;
2054 struct sock *sk;
2055 struct dst_entry *dst;
2056 int dif = skb->dev->ifindex;
2057 int ours;
2058
2059 /* validate the packet */
2060 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2061 return;
2062
2063 iph = ip_hdr(skb);
2064 uh = udp_hdr(skb);
2065
2066 if (skb->pkt_type == PACKET_BROADCAST ||
2067 skb->pkt_type == PACKET_MULTICAST) {
2068 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
2069
2070 if (!in_dev)
2071 return;
2072
2073 /* we are supposed to accept bcast packets */
2074 if (skb->pkt_type == PACKET_MULTICAST) {
2075 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2076 iph->protocol);
2077 if (!ours)
2078 return;
2079 }
2080
2081 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2082 uh->source, iph->saddr, dif);
2083 } else if (skb->pkt_type == PACKET_HOST) {
2084 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2085 uh->source, iph->saddr, dif);
2086 } else {
2087 return;
2088 }
2089
2090 if (!sk)
2091 return;
2092
2093 skb->sk = sk;
2094 skb->destructor = sock_efree;
2095 dst = READ_ONCE(sk->sk_rx_dst);
2096
2097 if (dst)
2098 dst = dst_check(dst, 0);
2099 if (dst) {
2100 /* DST_NOCACHE can not be used without taking a reference */
2101 if (dst->flags & DST_NOCACHE) {
2102 if (likely(atomic_inc_not_zero(&dst->__refcnt)))
2103 skb_dst_set(skb, dst);
2104 } else {
2105 skb_dst_set_noref(skb, dst);
2106 }
2107 }
2108 }
2109
2110 int udp_rcv(struct sk_buff *skb)
2111 {
2112 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2113 }
2114
2115 void udp_destroy_sock(struct sock *sk)
2116 {
2117 struct udp_sock *up = udp_sk(sk);
2118 bool slow = lock_sock_fast(sk);
2119 udp_flush_pending_frames(sk);
2120 unlock_sock_fast(sk, slow);
2121 if (static_key_false(&udp_encap_needed) && up->encap_type) {
2122 void (*encap_destroy)(struct sock *sk);
2123 encap_destroy = ACCESS_ONCE(up->encap_destroy);
2124 if (encap_destroy)
2125 encap_destroy(sk);
2126 }
2127 }
2128
2129 /*
2130 * Socket option code for UDP
2131 */
2132 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2133 char __user *optval, unsigned int optlen,
2134 int (*push_pending_frames)(struct sock *))
2135 {
2136 struct udp_sock *up = udp_sk(sk);
2137 int val, valbool;
2138 int err = 0;
2139 int is_udplite = IS_UDPLITE(sk);
2140
2141 if (optlen < sizeof(int))
2142 return -EINVAL;
2143
2144 if (get_user(val, (int __user *)optval))
2145 return -EFAULT;
2146
2147 valbool = val ? 1 : 0;
2148
2149 switch (optname) {
2150 case UDP_CORK:
2151 if (val != 0) {
2152 up->corkflag = 1;
2153 } else {
2154 up->corkflag = 0;
2155 lock_sock(sk);
2156 push_pending_frames(sk);
2157 release_sock(sk);
2158 }
2159 break;
2160
2161 case UDP_ENCAP:
2162 switch (val) {
2163 case 0:
2164 case UDP_ENCAP_ESPINUDP:
2165 case UDP_ENCAP_ESPINUDP_NON_IKE:
2166 up->encap_rcv = xfrm4_udp_encap_rcv;
2167 /* FALLTHROUGH */
2168 case UDP_ENCAP_L2TPINUDP:
2169 up->encap_type = val;
2170 udp_encap_enable();
2171 break;
2172 default:
2173 err = -ENOPROTOOPT;
2174 break;
2175 }
2176 break;
2177
2178 case UDP_NO_CHECK6_TX:
2179 up->no_check6_tx = valbool;
2180 break;
2181
2182 case UDP_NO_CHECK6_RX:
2183 up->no_check6_rx = valbool;
2184 break;
2185
2186 /*
2187 * UDP-Lite's partial checksum coverage (RFC 3828).
2188 */
2189 /* The sender sets actual checksum coverage length via this option.
2190 * The case coverage > packet length is handled by send module. */
2191 case UDPLITE_SEND_CSCOV:
2192 if (!is_udplite) /* Disable the option on UDP sockets */
2193 return -ENOPROTOOPT;
2194 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2195 val = 8;
2196 else if (val > USHRT_MAX)
2197 val = USHRT_MAX;
2198 up->pcslen = val;
2199 up->pcflag |= UDPLITE_SEND_CC;
2200 break;
2201
2202 /* The receiver specifies a minimum checksum coverage value. To make
2203 * sense, this should be set to at least 8 (as done below). If zero is
2204 * used, this again means full checksum coverage. */
2205 case UDPLITE_RECV_CSCOV:
2206 if (!is_udplite) /* Disable the option on UDP sockets */
2207 return -ENOPROTOOPT;
2208 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2209 val = 8;
2210 else if (val > USHRT_MAX)
2211 val = USHRT_MAX;
2212 up->pcrlen = val;
2213 up->pcflag |= UDPLITE_RECV_CC;
2214 break;
2215
2216 default:
2217 err = -ENOPROTOOPT;
2218 break;
2219 }
2220
2221 return err;
2222 }
2223 EXPORT_SYMBOL(udp_lib_setsockopt);
2224
2225 int udp_setsockopt(struct sock *sk, int level, int optname,
2226 char __user *optval, unsigned int optlen)
2227 {
2228 if (level == SOL_UDP || level == SOL_UDPLITE)
2229 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2230 udp_push_pending_frames);
2231 return ip_setsockopt(sk, level, optname, optval, optlen);
2232 }
2233
2234 #ifdef CONFIG_COMPAT
2235 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2236 char __user *optval, unsigned int optlen)
2237 {
2238 if (level == SOL_UDP || level == SOL_UDPLITE)
2239 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2240 udp_push_pending_frames);
2241 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2242 }
2243 #endif
2244
2245 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2246 char __user *optval, int __user *optlen)
2247 {
2248 struct udp_sock *up = udp_sk(sk);
2249 int val, len;
2250
2251 if (get_user(len, optlen))
2252 return -EFAULT;
2253
2254 len = min_t(unsigned int, len, sizeof(int));
2255
2256 if (len < 0)
2257 return -EINVAL;
2258
2259 switch (optname) {
2260 case UDP_CORK:
2261 val = up->corkflag;
2262 break;
2263
2264 case UDP_ENCAP:
2265 val = up->encap_type;
2266 break;
2267
2268 case UDP_NO_CHECK6_TX:
2269 val = up->no_check6_tx;
2270 break;
2271
2272 case UDP_NO_CHECK6_RX:
2273 val = up->no_check6_rx;
2274 break;
2275
2276 /* The following two cannot be changed on UDP sockets, the return is
2277 * always 0 (which corresponds to the full checksum coverage of UDP). */
2278 case UDPLITE_SEND_CSCOV:
2279 val = up->pcslen;
2280 break;
2281
2282 case UDPLITE_RECV_CSCOV:
2283 val = up->pcrlen;
2284 break;
2285
2286 default:
2287 return -ENOPROTOOPT;
2288 }
2289
2290 if (put_user(len, optlen))
2291 return -EFAULT;
2292 if (copy_to_user(optval, &val, len))
2293 return -EFAULT;
2294 return 0;
2295 }
2296 EXPORT_SYMBOL(udp_lib_getsockopt);
2297
2298 int udp_getsockopt(struct sock *sk, int level, int optname,
2299 char __user *optval, int __user *optlen)
2300 {
2301 if (level == SOL_UDP || level == SOL_UDPLITE)
2302 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2303 return ip_getsockopt(sk, level, optname, optval, optlen);
2304 }
2305
2306 #ifdef CONFIG_COMPAT
2307 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2308 char __user *optval, int __user *optlen)
2309 {
2310 if (level == SOL_UDP || level == SOL_UDPLITE)
2311 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2312 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2313 }
2314 #endif
2315 /**
2316 * udp_poll - wait for a UDP event.
2317 * @file - file struct
2318 * @sock - socket
2319 * @wait - poll table
2320 *
2321 * This is same as datagram poll, except for the special case of
2322 * blocking sockets. If application is using a blocking fd
2323 * and a packet with checksum error is in the queue;
2324 * then it could get return from select indicating data available
2325 * but then block when reading it. Add special case code
2326 * to work around these arguably broken applications.
2327 */
2328 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2329 {
2330 unsigned int mask = datagram_poll(file, sock, wait);
2331 struct sock *sk = sock->sk;
2332
2333 sock_rps_record_flow(sk);
2334
2335 /* Check for false positives due to checksum errors */
2336 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2337 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
2338 mask &= ~(POLLIN | POLLRDNORM);
2339
2340 return mask;
2341
2342 }
2343 EXPORT_SYMBOL(udp_poll);
2344
2345 struct proto udp_prot = {
2346 .name = "UDP",
2347 .owner = THIS_MODULE,
2348 .close = udp_lib_close,
2349 .connect = ip4_datagram_connect,
2350 .disconnect = udp_disconnect,
2351 .ioctl = udp_ioctl,
2352 .destroy = udp_destroy_sock,
2353 .setsockopt = udp_setsockopt,
2354 .getsockopt = udp_getsockopt,
2355 .sendmsg = udp_sendmsg,
2356 .recvmsg = udp_recvmsg,
2357 .sendpage = udp_sendpage,
2358 .backlog_rcv = __udp_queue_rcv_skb,
2359 .release_cb = ip4_datagram_release_cb,
2360 .hash = udp_lib_hash,
2361 .unhash = udp_lib_unhash,
2362 .rehash = udp_v4_rehash,
2363 .get_port = udp_v4_get_port,
2364 .memory_allocated = &udp_memory_allocated,
2365 .sysctl_mem = sysctl_udp_mem,
2366 .sysctl_wmem = &sysctl_udp_wmem_min,
2367 .sysctl_rmem = &sysctl_udp_rmem_min,
2368 .obj_size = sizeof(struct udp_sock),
2369 .slab_flags = SLAB_DESTROY_BY_RCU,
2370 .h.udp_table = &udp_table,
2371 #ifdef CONFIG_COMPAT
2372 .compat_setsockopt = compat_udp_setsockopt,
2373 .compat_getsockopt = compat_udp_getsockopt,
2374 #endif
2375 .clear_sk = sk_prot_clear_portaddr_nulls,
2376 };
2377 EXPORT_SYMBOL(udp_prot);
2378
2379 /* ------------------------------------------------------------------------ */
2380 #ifdef CONFIG_PROC_FS
2381
2382 static struct sock *udp_get_first(struct seq_file *seq, int start)
2383 {
2384 struct sock *sk;
2385 struct udp_iter_state *state = seq->private;
2386 struct net *net = seq_file_net(seq);
2387
2388 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2389 ++state->bucket) {
2390 struct hlist_nulls_node *node;
2391 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2392
2393 if (hlist_nulls_empty(&hslot->head))
2394 continue;
2395
2396 spin_lock_bh(&hslot->lock);
2397 sk_nulls_for_each(sk, node, &hslot->head) {
2398 if (!net_eq(sock_net(sk), net))
2399 continue;
2400 if (sk->sk_family == state->family)
2401 goto found;
2402 }
2403 spin_unlock_bh(&hslot->lock);
2404 }
2405 sk = NULL;
2406 found:
2407 return sk;
2408 }
2409
2410 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2411 {
2412 struct udp_iter_state *state = seq->private;
2413 struct net *net = seq_file_net(seq);
2414
2415 do {
2416 sk = sk_nulls_next(sk);
2417 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2418
2419 if (!sk) {
2420 if (state->bucket <= state->udp_table->mask)
2421 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2422 return udp_get_first(seq, state->bucket + 1);
2423 }
2424 return sk;
2425 }
2426
2427 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2428 {
2429 struct sock *sk = udp_get_first(seq, 0);
2430
2431 if (sk)
2432 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2433 --pos;
2434 return pos ? NULL : sk;
2435 }
2436
2437 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2438 {
2439 struct udp_iter_state *state = seq->private;
2440 state->bucket = MAX_UDP_PORTS;
2441
2442 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2443 }
2444
2445 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2446 {
2447 struct sock *sk;
2448
2449 if (v == SEQ_START_TOKEN)
2450 sk = udp_get_idx(seq, 0);
2451 else
2452 sk = udp_get_next(seq, v);
2453
2454 ++*pos;
2455 return sk;
2456 }
2457
2458 static void udp_seq_stop(struct seq_file *seq, void *v)
2459 {
2460 struct udp_iter_state *state = seq->private;
2461
2462 if (state->bucket <= state->udp_table->mask)
2463 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2464 }
2465
2466 int udp_seq_open(struct inode *inode, struct file *file)
2467 {
2468 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2469 struct udp_iter_state *s;
2470 int err;
2471
2472 err = seq_open_net(inode, file, &afinfo->seq_ops,
2473 sizeof(struct udp_iter_state));
2474 if (err < 0)
2475 return err;
2476
2477 s = ((struct seq_file *)file->private_data)->private;
2478 s->family = afinfo->family;
2479 s->udp_table = afinfo->udp_table;
2480 return err;
2481 }
2482 EXPORT_SYMBOL(udp_seq_open);
2483
2484 /* ------------------------------------------------------------------------ */
2485 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2486 {
2487 struct proc_dir_entry *p;
2488 int rc = 0;
2489
2490 afinfo->seq_ops.start = udp_seq_start;
2491 afinfo->seq_ops.next = udp_seq_next;
2492 afinfo->seq_ops.stop = udp_seq_stop;
2493
2494 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2495 afinfo->seq_fops, afinfo);
2496 if (!p)
2497 rc = -ENOMEM;
2498 return rc;
2499 }
2500 EXPORT_SYMBOL(udp_proc_register);
2501
2502 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2503 {
2504 remove_proc_entry(afinfo->name, net->proc_net);
2505 }
2506 EXPORT_SYMBOL(udp_proc_unregister);
2507
2508 /* ------------------------------------------------------------------------ */
2509 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2510 int bucket)
2511 {
2512 struct inet_sock *inet = inet_sk(sp);
2513 __be32 dest = inet->inet_daddr;
2514 __be32 src = inet->inet_rcv_saddr;
2515 __u16 destp = ntohs(inet->inet_dport);
2516 __u16 srcp = ntohs(inet->inet_sport);
2517
2518 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2519 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2520 bucket, src, srcp, dest, destp, sp->sk_state,
2521 sk_wmem_alloc_get(sp),
2522 sk_rmem_alloc_get(sp),
2523 0, 0L, 0,
2524 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2525 0, sock_i_ino(sp),
2526 atomic_read(&sp->sk_refcnt), sp,
2527 atomic_read(&sp->sk_drops));
2528 }
2529
2530 int udp4_seq_show(struct seq_file *seq, void *v)
2531 {
2532 seq_setwidth(seq, 127);
2533 if (v == SEQ_START_TOKEN)
2534 seq_puts(seq, " sl local_address rem_address st tx_queue "
2535 "rx_queue tr tm->when retrnsmt uid timeout "
2536 "inode ref pointer drops");
2537 else {
2538 struct udp_iter_state *state = seq->private;
2539
2540 udp4_format_sock(v, seq, state->bucket);
2541 }
2542 seq_pad(seq, '\n');
2543 return 0;
2544 }
2545
2546 static const struct file_operations udp_afinfo_seq_fops = {
2547 .owner = THIS_MODULE,
2548 .open = udp_seq_open,
2549 .read = seq_read,
2550 .llseek = seq_lseek,
2551 .release = seq_release_net
2552 };
2553
2554 /* ------------------------------------------------------------------------ */
2555 static struct udp_seq_afinfo udp4_seq_afinfo = {
2556 .name = "udp",
2557 .family = AF_INET,
2558 .udp_table = &udp_table,
2559 .seq_fops = &udp_afinfo_seq_fops,
2560 .seq_ops = {
2561 .show = udp4_seq_show,
2562 },
2563 };
2564
2565 static int __net_init udp4_proc_init_net(struct net *net)
2566 {
2567 return udp_proc_register(net, &udp4_seq_afinfo);
2568 }
2569
2570 static void __net_exit udp4_proc_exit_net(struct net *net)
2571 {
2572 udp_proc_unregister(net, &udp4_seq_afinfo);
2573 }
2574
2575 static struct pernet_operations udp4_net_ops = {
2576 .init = udp4_proc_init_net,
2577 .exit = udp4_proc_exit_net,
2578 };
2579
2580 int __init udp4_proc_init(void)
2581 {
2582 return register_pernet_subsys(&udp4_net_ops);
2583 }
2584
2585 void udp4_proc_exit(void)
2586 {
2587 unregister_pernet_subsys(&udp4_net_ops);
2588 }
2589 #endif /* CONFIG_PROC_FS */
2590
2591 static __initdata unsigned long uhash_entries;
2592 static int __init set_uhash_entries(char *str)
2593 {
2594 ssize_t ret;
2595
2596 if (!str)
2597 return 0;
2598
2599 ret = kstrtoul(str, 0, &uhash_entries);
2600 if (ret)
2601 return 0;
2602
2603 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2604 uhash_entries = UDP_HTABLE_SIZE_MIN;
2605 return 1;
2606 }
2607 __setup("uhash_entries=", set_uhash_entries);
2608
2609 void __init udp_table_init(struct udp_table *table, const char *name)
2610 {
2611 unsigned int i;
2612
2613 table->hash = alloc_large_system_hash(name,
2614 2 * sizeof(struct udp_hslot),
2615 uhash_entries,
2616 21, /* one slot per 2 MB */
2617 0,
2618 &table->log,
2619 &table->mask,
2620 UDP_HTABLE_SIZE_MIN,
2621 64 * 1024);
2622
2623 table->hash2 = table->hash + (table->mask + 1);
2624 for (i = 0; i <= table->mask; i++) {
2625 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2626 table->hash[i].count = 0;
2627 spin_lock_init(&table->hash[i].lock);
2628 }
2629 for (i = 0; i <= table->mask; i++) {
2630 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2631 table->hash2[i].count = 0;
2632 spin_lock_init(&table->hash2[i].lock);
2633 }
2634 }
2635
2636 u32 udp_flow_hashrnd(void)
2637 {
2638 static u32 hashrnd __read_mostly;
2639
2640 net_get_random_once(&hashrnd, sizeof(hashrnd));
2641
2642 return hashrnd;
2643 }
2644 EXPORT_SYMBOL(udp_flow_hashrnd);
2645
2646 void __init udp_init(void)
2647 {
2648 unsigned long limit;
2649
2650 udp_table_init(&udp_table, "UDP");
2651 limit = nr_free_buffer_pages() / 8;
2652 limit = max(limit, 128UL);
2653 sysctl_udp_mem[0] = limit / 4 * 3;
2654 sysctl_udp_mem[1] = limit;
2655 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2656
2657 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2658 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2659 }
CRIS linux kernel tree