?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'for-3.11' of git://linux-nfs.org/~bfields/linux
[linux-2.6.git] / net / ipv4 / udp.c
blob766e6bab911362cb0caa44511b10bb82418e7980
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/in.h>
94 #include <linux/errno.h>
95 #include <linux/timer.h>
96 #include <linux/mm.h>
97 #include <linux/inet.h>
98 #include <linux/netdevice.h>
99 #include <linux/slab.h>
100 #include <net/tcp_states.h>
101 #include <linux/skbuff.h>
102 #include <linux/proc_fs.h>
103 #include <linux/seq_file.h>
104 #include <net/net_namespace.h>
105 #include <net/icmp.h>
106 #include <net/route.h>
107 #include <net/checksum.h>
108 #include <net/xfrm.h>
109 #include <trace/events/udp.h>
110 #include <linux/static_key.h>
111 #include <trace/events/skb.h>
112 #include <net/busy_poll.h>
113 #include "udp_impl.h"
114
115 struct udp_table udp_table __read_mostly;
116 EXPORT_SYMBOL(udp_table);
117
118 long sysctl_udp_mem[3] __read_mostly;
119 EXPORT_SYMBOL(sysctl_udp_mem);
120
121 int sysctl_udp_rmem_min __read_mostly;
122 EXPORT_SYMBOL(sysctl_udp_rmem_min);
123
124 int sysctl_udp_wmem_min __read_mostly;
125 EXPORT_SYMBOL(sysctl_udp_wmem_min);
126
127 atomic_long_t udp_memory_allocated;
128 EXPORT_SYMBOL(udp_memory_allocated);
129
130 #define MAX_UDP_PORTS 65536
131 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
132
133 static int udp_lib_lport_inuse(struct net *net, __u16 num,
134 const struct udp_hslot *hslot,
135 unsigned long *bitmap,
136 struct sock *sk,
137 int (*saddr_comp)(const struct sock *sk1,
138 const struct sock *sk2),
139 unsigned int log)
140 {
141 struct sock *sk2;
142 struct hlist_nulls_node *node;
143 kuid_t uid = sock_i_uid(sk);
144
145 sk_nulls_for_each(sk2, node, &hslot->head)
146 if (net_eq(sock_net(sk2), net) &&
147 sk2 != sk &&
148 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
149 (!sk2->sk_reuse || !sk->sk_reuse) &&
150 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
151 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
152 (!sk2->sk_reuseport || !sk->sk_reuseport ||
153 !uid_eq(uid, sock_i_uid(sk2))) &&
154 (*saddr_comp)(sk, sk2)) {
155 if (bitmap)
156 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
157 bitmap);
158 else
159 return 1;
160 }
161 return 0;
162 }
163
164 /*
165 * Note: we still hold spinlock of primary hash chain, so no other writer
166 * can insert/delete a socket with local_port == num
167 */
168 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
169 struct udp_hslot *hslot2,
170 struct sock *sk,
171 int (*saddr_comp)(const struct sock *sk1,
172 const struct sock *sk2))
173 {
174 struct sock *sk2;
175 struct hlist_nulls_node *node;
176 kuid_t uid = sock_i_uid(sk);
177 int res = 0;
178
179 spin_lock(&hslot2->lock);
180 udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
181 if (net_eq(sock_net(sk2), net) &&
182 sk2 != sk &&
183 (udp_sk(sk2)->udp_port_hash == num) &&
184 (!sk2->sk_reuse || !sk->sk_reuse) &&
185 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
186 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
187 (!sk2->sk_reuseport || !sk->sk_reuseport ||
188 !uid_eq(uid, sock_i_uid(sk2))) &&
189 (*saddr_comp)(sk, sk2)) {
190 res = 1;
191 break;
192 }
193 spin_unlock(&hslot2->lock);
194 return res;
195 }
196
197 /**
198 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
199 *
200 * @sk: socket struct in question
201 * @snum: port number to look up
202 * @saddr_comp: AF-dependent comparison of bound local IP addresses
203 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
204 * with NULL address
205 */
206 int udp_lib_get_port(struct sock *sk, unsigned short snum,
207 int (*saddr_comp)(const struct sock *sk1,
208 const struct sock *sk2),
209 unsigned int hash2_nulladdr)
210 {
211 struct udp_hslot *hslot, *hslot2;
212 struct udp_table *udptable = sk->sk_prot->h.udp_table;
213 int error = 1;
214 struct net *net = sock_net(sk);
215
216 if (!snum) {
217 int low, high, remaining;
218 unsigned int rand;
219 unsigned short first, last;
220 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
221
222 inet_get_local_port_range(&low, &high);
223 remaining = (high - low) + 1;
224
225 rand = net_random();
226 first = (((u64)rand * remaining) >> 32) + low;
227 /*
228 * force rand to be an odd multiple of UDP_HTABLE_SIZE
229 */
230 rand = (rand | 1) * (udptable->mask + 1);
231 last = first + udptable->mask + 1;
232 do {
233 hslot = udp_hashslot(udptable, net, first);
234 bitmap_zero(bitmap, PORTS_PER_CHAIN);
235 spin_lock_bh(&hslot->lock);
236 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
237 saddr_comp, udptable->log);
238
239 snum = first;
240 /*
241 * Iterate on all possible values of snum for this hash.
242 * Using steps of an odd multiple of UDP_HTABLE_SIZE
243 * give us randomization and full range coverage.
244 */
245 do {
246 if (low <= snum && snum <= high &&
247 !test_bit(snum >> udptable->log, bitmap) &&
248 !inet_is_reserved_local_port(snum))
249 goto found;
250 snum += rand;
251 } while (snum != first);
252 spin_unlock_bh(&hslot->lock);
253 } while (++first != last);
254 goto fail;
255 } else {
256 hslot = udp_hashslot(udptable, net, snum);
257 spin_lock_bh(&hslot->lock);
258 if (hslot->count > 10) {
259 int exist;
260 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
261
262 slot2 &= udptable->mask;
263 hash2_nulladdr &= udptable->mask;
264
265 hslot2 = udp_hashslot2(udptable, slot2);
266 if (hslot->count < hslot2->count)
267 goto scan_primary_hash;
268
269 exist = udp_lib_lport_inuse2(net, snum, hslot2,
270 sk, saddr_comp);
271 if (!exist && (hash2_nulladdr != slot2)) {
272 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
273 exist = udp_lib_lport_inuse2(net, snum, hslot2,
274 sk, saddr_comp);
275 }
276 if (exist)
277 goto fail_unlock;
278 else
279 goto found;
280 }
281 scan_primary_hash:
282 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
283 saddr_comp, 0))
284 goto fail_unlock;
285 }
286 found:
287 inet_sk(sk)->inet_num = snum;
288 udp_sk(sk)->udp_port_hash = snum;
289 udp_sk(sk)->udp_portaddr_hash ^= snum;
290 if (sk_unhashed(sk)) {
291 sk_nulls_add_node_rcu(sk, &hslot->head);
292 hslot->count++;
293 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
294
295 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
296 spin_lock(&hslot2->lock);
297 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
298 &hslot2->head);
299 hslot2->count++;
300 spin_unlock(&hslot2->lock);
301 }
302 error = 0;
303 fail_unlock:
304 spin_unlock_bh(&hslot->lock);
305 fail:
306 return error;
307 }
308 EXPORT_SYMBOL(udp_lib_get_port);
309
310 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
311 {
312 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
313
314 return (!ipv6_only_sock(sk2) &&
315 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
316 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
317 }
318
319 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
320 unsigned int port)
321 {
322 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
323 }
324
325 int udp_v4_get_port(struct sock *sk, unsigned short snum)
326 {
327 unsigned int hash2_nulladdr =
328 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
329 unsigned int hash2_partial =
330 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
331
332 /* precompute partial secondary hash */
333 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
334 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
335 }
336
337 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
338 unsigned short hnum,
339 __be16 sport, __be32 daddr, __be16 dport, int dif)
340 {
341 int score = -1;
342
343 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
344 !ipv6_only_sock(sk)) {
345 struct inet_sock *inet = inet_sk(sk);
346
347 score = (sk->sk_family == PF_INET ? 2 : 1);
348 if (inet->inet_rcv_saddr) {
349 if (inet->inet_rcv_saddr != daddr)
350 return -1;
351 score += 4;
352 }
353 if (inet->inet_daddr) {
354 if (inet->inet_daddr != saddr)
355 return -1;
356 score += 4;
357 }
358 if (inet->inet_dport) {
359 if (inet->inet_dport != sport)
360 return -1;
361 score += 4;
362 }
363 if (sk->sk_bound_dev_if) {
364 if (sk->sk_bound_dev_if != dif)
365 return -1;
366 score += 4;
367 }
368 }
369 return score;
370 }
371
372 /*
373 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
374 */
375 static inline int compute_score2(struct sock *sk, struct net *net,
376 __be32 saddr, __be16 sport,
377 __be32 daddr, unsigned int hnum, int dif)
378 {
379 int score = -1;
380
381 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
382 struct inet_sock *inet = inet_sk(sk);
383
384 if (inet->inet_rcv_saddr != daddr)
385 return -1;
386 if (inet->inet_num != hnum)
387 return -1;
388
389 score = (sk->sk_family == PF_INET ? 2 : 1);
390 if (inet->inet_daddr) {
391 if (inet->inet_daddr != saddr)
392 return -1;
393 score += 4;
394 }
395 if (inet->inet_dport) {
396 if (inet->inet_dport != sport)
397 return -1;
398 score += 4;
399 }
400 if (sk->sk_bound_dev_if) {
401 if (sk->sk_bound_dev_if != dif)
402 return -1;
403 score += 4;
404 }
405 }
406 return score;
407 }
408
409
410 /* called with read_rcu_lock() */
411 static struct sock *udp4_lib_lookup2(struct net *net,
412 __be32 saddr, __be16 sport,
413 __be32 daddr, unsigned int hnum, int dif,
414 struct udp_hslot *hslot2, unsigned int slot2)
415 {
416 struct sock *sk, *result;
417 struct hlist_nulls_node *node;
418 int score, badness, matches = 0, reuseport = 0;
419 u32 hash = 0;
420
421 begin:
422 result = NULL;
423 badness = 0;
424 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
425 score = compute_score2(sk, net, saddr, sport,
426 daddr, hnum, dif);
427 if (score > badness) {
428 result = sk;
429 badness = score;
430 reuseport = sk->sk_reuseport;
431 if (reuseport) {
432 hash = inet_ehashfn(net, daddr, hnum,
433 saddr, sport);
434 matches = 1;
435 }
436 } else if (score == badness && reuseport) {
437 matches++;
438 if (((u64)hash * matches) >> 32 == 0)
439 result = sk;
440 hash = next_pseudo_random32(hash);
441 }
442 }
443 /*
444 * if the nulls value we got at the end of this lookup is
445 * not the expected one, we must restart lookup.
446 * We probably met an item that was moved to another chain.
447 */
448 if (get_nulls_value(node) != slot2)
449 goto begin;
450 if (result) {
451 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
452 result = NULL;
453 else if (unlikely(compute_score2(result, net, saddr, sport,
454 daddr, hnum, dif) < badness)) {
455 sock_put(result);
456 goto begin;
457 }
458 }
459 return result;
460 }
461
462 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
463 * harder than this. -DaveM
464 */
465 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
466 __be16 sport, __be32 daddr, __be16 dport,
467 int dif, struct udp_table *udptable)
468 {
469 struct sock *sk, *result;
470 struct hlist_nulls_node *node;
471 unsigned short hnum = ntohs(dport);
472 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
473 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
474 int score, badness, matches = 0, reuseport = 0;
475 u32 hash = 0;
476
477 rcu_read_lock();
478 if (hslot->count > 10) {
479 hash2 = udp4_portaddr_hash(net, daddr, hnum);
480 slot2 = hash2 & udptable->mask;
481 hslot2 = &udptable->hash2[slot2];
482 if (hslot->count < hslot2->count)
483 goto begin;
484
485 result = udp4_lib_lookup2(net, saddr, sport,
486 daddr, hnum, dif,
487 hslot2, slot2);
488 if (!result) {
489 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
490 slot2 = hash2 & udptable->mask;
491 hslot2 = &udptable->hash2[slot2];
492 if (hslot->count < hslot2->count)
493 goto begin;
494
495 result = udp4_lib_lookup2(net, saddr, sport,
496 htonl(INADDR_ANY), hnum, dif,
497 hslot2, slot2);
498 }
499 rcu_read_unlock();
500 return result;
501 }
502 begin:
503 result = NULL;
504 badness = 0;
505 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
506 score = compute_score(sk, net, saddr, hnum, sport,
507 daddr, dport, dif);
508 if (score > badness) {
509 result = sk;
510 badness = score;
511 reuseport = sk->sk_reuseport;
512 if (reuseport) {
513 hash = inet_ehashfn(net, daddr, hnum,
514 saddr, sport);
515 matches = 1;
516 }
517 } else if (score == badness && reuseport) {
518 matches++;
519 if (((u64)hash * matches) >> 32 == 0)
520 result = sk;
521 hash = next_pseudo_random32(hash);
522 }
523 }
524 /*
525 * if the nulls value we got at the end of this lookup is
526 * not the expected one, we must restart lookup.
527 * We probably met an item that was moved to another chain.
528 */
529 if (get_nulls_value(node) != slot)
530 goto begin;
531
532 if (result) {
533 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
534 result = NULL;
535 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
536 daddr, dport, dif) < badness)) {
537 sock_put(result);
538 goto begin;
539 }
540 }
541 rcu_read_unlock();
542 return result;
543 }
544 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
545
546 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
547 __be16 sport, __be16 dport,
548 struct udp_table *udptable)
549 {
550 struct sock *sk;
551 const struct iphdr *iph = ip_hdr(skb);
552
553 if (unlikely(sk = skb_steal_sock(skb)))
554 return sk;
555 else
556 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
557 iph->daddr, dport, inet_iif(skb),
558 udptable);
559 }
560
561 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
562 __be32 daddr, __be16 dport, int dif)
563 {
564 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
565 }
566 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
567
568 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
569 __be16 loc_port, __be32 loc_addr,
570 __be16 rmt_port, __be32 rmt_addr,
571 int dif)
572 {
573 struct hlist_nulls_node *node;
574 struct sock *s = sk;
575 unsigned short hnum = ntohs(loc_port);
576
577 sk_nulls_for_each_from(s, node) {
578 struct inet_sock *inet = inet_sk(s);
579
580 if (!net_eq(sock_net(s), net) ||
581 udp_sk(s)->udp_port_hash != hnum ||
582 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
583 (inet->inet_dport != rmt_port && inet->inet_dport) ||
584 (inet->inet_rcv_saddr &&
585 inet->inet_rcv_saddr != loc_addr) ||
586 ipv6_only_sock(s) ||
587 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
588 continue;
589 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
590 continue;
591 goto found;
592 }
593 s = NULL;
594 found:
595 return s;
596 }
597
598 /*
599 * This routine is called by the ICMP module when it gets some
600 * sort of error condition. If err < 0 then the socket should
601 * be closed and the error returned to the user. If err > 0
602 * it's just the icmp type << 8 | icmp code.
603 * Header points to the ip header of the error packet. We move
604 * on past this. Then (as it used to claim before adjustment)
605 * header points to the first 8 bytes of the udp header. We need
606 * to find the appropriate port.
607 */
608
609 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
610 {
611 struct inet_sock *inet;
612 const struct iphdr *iph = (const struct iphdr *)skb->data;
613 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
614 const int type = icmp_hdr(skb)->type;
615 const int code = icmp_hdr(skb)->code;
616 struct sock *sk;
617 int harderr;
618 int err;
619 struct net *net = dev_net(skb->dev);
620
621 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
622 iph->saddr, uh->source, skb->dev->ifindex, udptable);
623 if (sk == NULL) {
624 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
625 return; /* No socket for error */
626 }
627
628 err = 0;
629 harderr = 0;
630 inet = inet_sk(sk);
631
632 switch (type) {
633 default:
634 case ICMP_TIME_EXCEEDED:
635 err = EHOSTUNREACH;
636 break;
637 case ICMP_SOURCE_QUENCH:
638 goto out;
639 case ICMP_PARAMETERPROB:
640 err = EPROTO;
641 harderr = 1;
642 break;
643 case ICMP_DEST_UNREACH:
644 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
645 ipv4_sk_update_pmtu(skb, sk, info);
646 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
647 err = EMSGSIZE;
648 harderr = 1;
649 break;
650 }
651 goto out;
652 }
653 err = EHOSTUNREACH;
654 if (code <= NR_ICMP_UNREACH) {
655 harderr = icmp_err_convert[code].fatal;
656 err = icmp_err_convert[code].errno;
657 }
658 break;
659 case ICMP_REDIRECT:
660 ipv4_sk_redirect(skb, sk);
661 break;
662 }
663
664 /*
665 * RFC1122: OK. Passes ICMP errors back to application, as per
666 * 4.1.3.3.
667 */
668 if (!inet->recverr) {
669 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
670 goto out;
671 } else
672 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
673
674 sk->sk_err = err;
675 sk->sk_error_report(sk);
676 out:
677 sock_put(sk);
678 }
679
680 void udp_err(struct sk_buff *skb, u32 info)
681 {
682 __udp4_lib_err(skb, info, &udp_table);
683 }
684
685 /*
686 * Throw away all pending data and cancel the corking. Socket is locked.
687 */
688 void udp_flush_pending_frames(struct sock *sk)
689 {
690 struct udp_sock *up = udp_sk(sk);
691
692 if (up->pending) {
693 up->len = 0;
694 up->pending = 0;
695 ip_flush_pending_frames(sk);
696 }
697 }
698 EXPORT_SYMBOL(udp_flush_pending_frames);
699
700 /**
701 * udp4_hwcsum - handle outgoing HW checksumming
702 * @skb: sk_buff containing the filled-in UDP header
703 * (checksum field must be zeroed out)
704 * @src: source IP address
705 * @dst: destination IP address
706 */
707 static void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
708 {
709 struct udphdr *uh = udp_hdr(skb);
710 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
711 int offset = skb_transport_offset(skb);
712 int len = skb->len - offset;
713 int hlen = len;
714 __wsum csum = 0;
715
716 if (!frags) {
717 /*
718 * Only one fragment on the socket.
719 */
720 skb->csum_start = skb_transport_header(skb) - skb->head;
721 skb->csum_offset = offsetof(struct udphdr, check);
722 uh->check = ~csum_tcpudp_magic(src, dst, len,
723 IPPROTO_UDP, 0);
724 } else {
725 /*
726 * HW-checksum won't work as there are two or more
727 * fragments on the socket so that all csums of sk_buffs
728 * should be together
729 */
730 do {
731 csum = csum_add(csum, frags->csum);
732 hlen -= frags->len;
733 } while ((frags = frags->next));
734
735 csum = skb_checksum(skb, offset, hlen, csum);
736 skb->ip_summed = CHECKSUM_NONE;
737
738 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
739 if (uh->check == 0)
740 uh->check = CSUM_MANGLED_0;
741 }
742 }
743
744 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
745 {
746 struct sock *sk = skb->sk;
747 struct inet_sock *inet = inet_sk(sk);
748 struct udphdr *uh;
749 int err = 0;
750 int is_udplite = IS_UDPLITE(sk);
751 int offset = skb_transport_offset(skb);
752 int len = skb->len - offset;
753 __wsum csum = 0;
754
755 /*
756 * Create a UDP header
757 */
758 uh = udp_hdr(skb);
759 uh->source = inet->inet_sport;
760 uh->dest = fl4->fl4_dport;
761 uh->len = htons(len);
762 uh->check = 0;
763
764 if (is_udplite) /* UDP-Lite */
765 csum = udplite_csum(skb);
766
767 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
768
769 skb->ip_summed = CHECKSUM_NONE;
770 goto send;
771
772 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
773
774 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
775 goto send;
776
777 } else
778 csum = udp_csum(skb);
779
780 /* add protocol-dependent pseudo-header */
781 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
782 sk->sk_protocol, csum);
783 if (uh->check == 0)
784 uh->check = CSUM_MANGLED_0;
785
786 send:
787 err = ip_send_skb(sock_net(sk), skb);
788 if (err) {
789 if (err == -ENOBUFS && !inet->recverr) {
790 UDP_INC_STATS_USER(sock_net(sk),
791 UDP_MIB_SNDBUFERRORS, is_udplite);
792 err = 0;
793 }
794 } else
795 UDP_INC_STATS_USER(sock_net(sk),
796 UDP_MIB_OUTDATAGRAMS, is_udplite);
797 return err;
798 }
799
800 /*
801 * Push out all pending data as one UDP datagram. Socket is locked.
802 */
803 int udp_push_pending_frames(struct sock *sk)
804 {
805 struct udp_sock *up = udp_sk(sk);
806 struct inet_sock *inet = inet_sk(sk);
807 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
808 struct sk_buff *skb;
809 int err = 0;
810
811 skb = ip_finish_skb(sk, fl4);
812 if (!skb)
813 goto out;
814
815 err = udp_send_skb(skb, fl4);
816
817 out:
818 up->len = 0;
819 up->pending = 0;
820 return err;
821 }
822 EXPORT_SYMBOL(udp_push_pending_frames);
823
824 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
825 size_t len)
826 {
827 struct inet_sock *inet = inet_sk(sk);
828 struct udp_sock *up = udp_sk(sk);
829 struct flowi4 fl4_stack;
830 struct flowi4 *fl4;
831 int ulen = len;
832 struct ipcm_cookie ipc;
833 struct rtable *rt = NULL;
834 int free = 0;
835 int connected = 0;
836 __be32 daddr, faddr, saddr;
837 __be16 dport;
838 u8 tos;
839 int err, is_udplite = IS_UDPLITE(sk);
840 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
841 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
842 struct sk_buff *skb;
843 struct ip_options_data opt_copy;
844
845 if (len > 0xFFFF)
846 return -EMSGSIZE;
847
848 /*
849 * Check the flags.
850 */
851
852 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
853 return -EOPNOTSUPP;
854
855 ipc.opt = NULL;
856 ipc.tx_flags = 0;
857
858 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
859
860 fl4 = &inet->cork.fl.u.ip4;
861 if (up->pending) {
862 /*
863 * There are pending frames.
864 * The socket lock must be held while it's corked.
865 */
866 lock_sock(sk);
867 if (likely(up->pending)) {
868 if (unlikely(up->pending != AF_INET)) {
869 release_sock(sk);
870 return -EINVAL;
871 }
872 goto do_append_data;
873 }
874 release_sock(sk);
875 }
876 ulen += sizeof(struct udphdr);
877
878 /*
879 * Get and verify the address.
880 */
881 if (msg->msg_name) {
882 struct sockaddr_in *usin = (struct sockaddr_in *)msg->msg_name;
883 if (msg->msg_namelen < sizeof(*usin))
884 return -EINVAL;
885 if (usin->sin_family != AF_INET) {
886 if (usin->sin_family != AF_UNSPEC)
887 return -EAFNOSUPPORT;
888 }
889
890 daddr = usin->sin_addr.s_addr;
891 dport = usin->sin_port;
892 if (dport == 0)
893 return -EINVAL;
894 } else {
895 if (sk->sk_state != TCP_ESTABLISHED)
896 return -EDESTADDRREQ;
897 daddr = inet->inet_daddr;
898 dport = inet->inet_dport;
899 /* Open fast path for connected socket.
900 Route will not be used, if at least one option is set.
901 */
902 connected = 1;
903 }
904 ipc.addr = inet->inet_saddr;
905
906 ipc.oif = sk->sk_bound_dev_if;
907
908 sock_tx_timestamp(sk, &ipc.tx_flags);
909
910 if (msg->msg_controllen) {
911 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
912 if (err)
913 return err;
914 if (ipc.opt)
915 free = 1;
916 connected = 0;
917 }
918 if (!ipc.opt) {
919 struct ip_options_rcu *inet_opt;
920
921 rcu_read_lock();
922 inet_opt = rcu_dereference(inet->inet_opt);
923 if (inet_opt) {
924 memcpy(&opt_copy, inet_opt,
925 sizeof(*inet_opt) + inet_opt->opt.optlen);
926 ipc.opt = &opt_copy.opt;
927 }
928 rcu_read_unlock();
929 }
930
931 saddr = ipc.addr;
932 ipc.addr = faddr = daddr;
933
934 if (ipc.opt && ipc.opt->opt.srr) {
935 if (!daddr)
936 return -EINVAL;
937 faddr = ipc.opt->opt.faddr;
938 connected = 0;
939 }
940 tos = RT_TOS(inet->tos);
941 if (sock_flag(sk, SOCK_LOCALROUTE) ||
942 (msg->msg_flags & MSG_DONTROUTE) ||
943 (ipc.opt && ipc.opt->opt.is_strictroute)) {
944 tos |= RTO_ONLINK;
945 connected = 0;
946 }
947
948 if (ipv4_is_multicast(daddr)) {
949 if (!ipc.oif)
950 ipc.oif = inet->mc_index;
951 if (!saddr)
952 saddr = inet->mc_addr;
953 connected = 0;
954 } else if (!ipc.oif)
955 ipc.oif = inet->uc_index;
956
957 if (connected)
958 rt = (struct rtable *)sk_dst_check(sk, 0);
959
960 if (rt == NULL) {
961 struct net *net = sock_net(sk);
962
963 fl4 = &fl4_stack;
964 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
965 RT_SCOPE_UNIVERSE, sk->sk_protocol,
966 inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP,
967 faddr, saddr, dport, inet->inet_sport);
968
969 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
970 rt = ip_route_output_flow(net, fl4, sk);
971 if (IS_ERR(rt)) {
972 err = PTR_ERR(rt);
973 rt = NULL;
974 if (err == -ENETUNREACH)
975 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
976 goto out;
977 }
978
979 err = -EACCES;
980 if ((rt->rt_flags & RTCF_BROADCAST) &&
981 !sock_flag(sk, SOCK_BROADCAST))
982 goto out;
983 if (connected)
984 sk_dst_set(sk, dst_clone(&rt->dst));
985 }
986
987 if (msg->msg_flags&MSG_CONFIRM)
988 goto do_confirm;
989 back_from_confirm:
990
991 saddr = fl4->saddr;
992 if (!ipc.addr)
993 daddr = ipc.addr = fl4->daddr;
994
995 /* Lockless fast path for the non-corking case. */
996 if (!corkreq) {
997 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen,
998 sizeof(struct udphdr), &ipc, &rt,
999 msg->msg_flags);
1000 err = PTR_ERR(skb);
1001 if (!IS_ERR_OR_NULL(skb))
1002 err = udp_send_skb(skb, fl4);
1003 goto out;
1004 }
1005
1006 lock_sock(sk);
1007 if (unlikely(up->pending)) {
1008 /* The socket is already corked while preparing it. */
1009 /* ... which is an evident application bug. --ANK */
1010 release_sock(sk);
1011
1012 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("cork app bug 2\n"));
1013 err = -EINVAL;
1014 goto out;
1015 }
1016 /*
1017 * Now cork the socket to pend data.
1018 */
1019 fl4 = &inet->cork.fl.u.ip4;
1020 fl4->daddr = daddr;
1021 fl4->saddr = saddr;
1022 fl4->fl4_dport = dport;
1023 fl4->fl4_sport = inet->inet_sport;
1024 up->pending = AF_INET;
1025
1026 do_append_data:
1027 up->len += ulen;
1028 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen,
1029 sizeof(struct udphdr), &ipc, &rt,
1030 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1031 if (err)
1032 udp_flush_pending_frames(sk);
1033 else if (!corkreq)
1034 err = udp_push_pending_frames(sk);
1035 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1036 up->pending = 0;
1037 release_sock(sk);
1038
1039 out:
1040 ip_rt_put(rt);
1041 if (free)
1042 kfree(ipc.opt);
1043 if (!err)
1044 return len;
1045 /*
1046 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1047 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1048 * we don't have a good statistic (IpOutDiscards but it can be too many
1049 * things). We could add another new stat but at least for now that
1050 * seems like overkill.
1051 */
1052 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1053 UDP_INC_STATS_USER(sock_net(sk),
1054 UDP_MIB_SNDBUFERRORS, is_udplite);
1055 }
1056 return err;
1057
1058 do_confirm:
1059 dst_confirm(&rt->dst);
1060 if (!(msg->msg_flags&MSG_PROBE) || len)
1061 goto back_from_confirm;
1062 err = 0;
1063 goto out;
1064 }
1065 EXPORT_SYMBOL(udp_sendmsg);
1066
1067 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1068 size_t size, int flags)
1069 {
1070 struct inet_sock *inet = inet_sk(sk);
1071 struct udp_sock *up = udp_sk(sk);
1072 int ret;
1073
1074 if (!up->pending) {
1075 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1076
1077 /* Call udp_sendmsg to specify destination address which
1078 * sendpage interface can't pass.
1079 * This will succeed only when the socket is connected.
1080 */
1081 ret = udp_sendmsg(NULL, sk, &msg, 0);
1082 if (ret < 0)
1083 return ret;
1084 }
1085
1086 lock_sock(sk);
1087
1088 if (unlikely(!up->pending)) {
1089 release_sock(sk);
1090
1091 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("udp cork app bug 3\n"));
1092 return -EINVAL;
1093 }
1094
1095 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1096 page, offset, size, flags);
1097 if (ret == -EOPNOTSUPP) {
1098 release_sock(sk);
1099 return sock_no_sendpage(sk->sk_socket, page, offset,
1100 size, flags);
1101 }
1102 if (ret < 0) {
1103 udp_flush_pending_frames(sk);
1104 goto out;
1105 }
1106
1107 up->len += size;
1108 if (!(up->corkflag || (flags&MSG_MORE)))
1109 ret = udp_push_pending_frames(sk);
1110 if (!ret)
1111 ret = size;
1112 out:
1113 release_sock(sk);
1114 return ret;
1115 }
1116
1117
1118 /**
1119 * first_packet_length - return length of first packet in receive queue
1120 * @sk: socket
1121 *
1122 * Drops all bad checksum frames, until a valid one is found.
1123 * Returns the length of found skb, or 0 if none is found.
1124 */
1125 static unsigned int first_packet_length(struct sock *sk)
1126 {
1127 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1128 struct sk_buff *skb;
1129 unsigned int res;
1130
1131 __skb_queue_head_init(&list_kill);
1132
1133 spin_lock_bh(&rcvq->lock);
1134 while ((skb = skb_peek(rcvq)) != NULL &&
1135 udp_lib_checksum_complete(skb)) {
1136 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS,
1137 IS_UDPLITE(sk));
1138 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1139 IS_UDPLITE(sk));
1140 atomic_inc(&sk->sk_drops);
1141 __skb_unlink(skb, rcvq);
1142 __skb_queue_tail(&list_kill, skb);
1143 }
1144 res = skb ? skb->len : 0;
1145 spin_unlock_bh(&rcvq->lock);
1146
1147 if (!skb_queue_empty(&list_kill)) {
1148 bool slow = lock_sock_fast(sk);
1149
1150 __skb_queue_purge(&list_kill);
1151 sk_mem_reclaim_partial(sk);
1152 unlock_sock_fast(sk, slow);
1153 }
1154 return res;
1155 }
1156
1157 /*
1158 * IOCTL requests applicable to the UDP protocol
1159 */
1160
1161 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1162 {
1163 switch (cmd) {
1164 case SIOCOUTQ:
1165 {
1166 int amount = sk_wmem_alloc_get(sk);
1167
1168 return put_user(amount, (int __user *)arg);
1169 }
1170
1171 case SIOCINQ:
1172 {
1173 unsigned int amount = first_packet_length(sk);
1174
1175 if (amount)
1176 /*
1177 * We will only return the amount
1178 * of this packet since that is all
1179 * that will be read.
1180 */
1181 amount -= sizeof(struct udphdr);
1182
1183 return put_user(amount, (int __user *)arg);
1184 }
1185
1186 default:
1187 return -ENOIOCTLCMD;
1188 }
1189
1190 return 0;
1191 }
1192 EXPORT_SYMBOL(udp_ioctl);
1193
1194 /*
1195 * This should be easy, if there is something there we
1196 * return it, otherwise we block.
1197 */
1198
1199 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1200 size_t len, int noblock, int flags, int *addr_len)
1201 {
1202 struct inet_sock *inet = inet_sk(sk);
1203 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1204 struct sk_buff *skb;
1205 unsigned int ulen, copied;
1206 int peeked, off = 0;
1207 int err;
1208 int is_udplite = IS_UDPLITE(sk);
1209 bool slow;
1210
1211 /*
1212 * Check any passed addresses
1213 */
1214 if (addr_len)
1215 *addr_len = sizeof(*sin);
1216
1217 if (flags & MSG_ERRQUEUE)
1218 return ip_recv_error(sk, msg, len);
1219
1220 try_again:
1221 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1222 &peeked, &off, &err);
1223 if (!skb)
1224 goto out;
1225
1226 ulen = skb->len - sizeof(struct udphdr);
1227 copied = len;
1228 if (copied > ulen)
1229 copied = ulen;
1230 else if (copied < ulen)
1231 msg->msg_flags |= MSG_TRUNC;
1232
1233 /*
1234 * If checksum is needed at all, try to do it while copying the
1235 * data. If the data is truncated, or if we only want a partial
1236 * coverage checksum (UDP-Lite), do it before the copy.
1237 */
1238
1239 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1240 if (udp_lib_checksum_complete(skb))
1241 goto csum_copy_err;
1242 }
1243
1244 if (skb_csum_unnecessary(skb))
1245 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1246 msg->msg_iov, copied);
1247 else {
1248 err = skb_copy_and_csum_datagram_iovec(skb,
1249 sizeof(struct udphdr),
1250 msg->msg_iov);
1251
1252 if (err == -EINVAL)
1253 goto csum_copy_err;
1254 }
1255
1256 if (unlikely(err)) {
1257 trace_kfree_skb(skb, udp_recvmsg);
1258 if (!peeked) {
1259 atomic_inc(&sk->sk_drops);
1260 UDP_INC_STATS_USER(sock_net(sk),
1261 UDP_MIB_INERRORS, is_udplite);
1262 }
1263 goto out_free;
1264 }
1265
1266 if (!peeked)
1267 UDP_INC_STATS_USER(sock_net(sk),
1268 UDP_MIB_INDATAGRAMS, is_udplite);
1269
1270 sock_recv_ts_and_drops(msg, sk, skb);
1271
1272 /* Copy the address. */
1273 if (sin) {
1274 sin->sin_family = AF_INET;
1275 sin->sin_port = udp_hdr(skb)->source;
1276 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1277 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1278 }
1279 if (inet->cmsg_flags)
1280 ip_cmsg_recv(msg, skb);
1281
1282 err = copied;
1283 if (flags & MSG_TRUNC)
1284 err = ulen;
1285
1286 out_free:
1287 skb_free_datagram_locked(sk, skb);
1288 out:
1289 return err;
1290
1291 csum_copy_err:
1292 slow = lock_sock_fast(sk);
1293 if (!skb_kill_datagram(sk, skb, flags)) {
1294 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1295 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1296 }
1297 unlock_sock_fast(sk, slow);
1298
1299 if (noblock)
1300 return -EAGAIN;
1301
1302 /* starting over for a new packet */
1303 msg->msg_flags &= ~MSG_TRUNC;
1304 goto try_again;
1305 }
1306
1307
1308 int udp_disconnect(struct sock *sk, int flags)
1309 {
1310 struct inet_sock *inet = inet_sk(sk);
1311 /*
1312 * 1003.1g - break association.
1313 */
1314
1315 sk->sk_state = TCP_CLOSE;
1316 inet->inet_daddr = 0;
1317 inet->inet_dport = 0;
1318 sock_rps_reset_rxhash(sk);
1319 sk->sk_bound_dev_if = 0;
1320 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1321 inet_reset_saddr(sk);
1322
1323 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1324 sk->sk_prot->unhash(sk);
1325 inet->inet_sport = 0;
1326 }
1327 sk_dst_reset(sk);
1328 return 0;
1329 }
1330 EXPORT_SYMBOL(udp_disconnect);
1331
1332 void udp_lib_unhash(struct sock *sk)
1333 {
1334 if (sk_hashed(sk)) {
1335 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1336 struct udp_hslot *hslot, *hslot2;
1337
1338 hslot = udp_hashslot(udptable, sock_net(sk),
1339 udp_sk(sk)->udp_port_hash);
1340 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1341
1342 spin_lock_bh(&hslot->lock);
1343 if (sk_nulls_del_node_init_rcu(sk)) {
1344 hslot->count--;
1345 inet_sk(sk)->inet_num = 0;
1346 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1347
1348 spin_lock(&hslot2->lock);
1349 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1350 hslot2->count--;
1351 spin_unlock(&hslot2->lock);
1352 }
1353 spin_unlock_bh(&hslot->lock);
1354 }
1355 }
1356 EXPORT_SYMBOL(udp_lib_unhash);
1357
1358 /*
1359 * inet_rcv_saddr was changed, we must rehash secondary hash
1360 */
1361 void udp_lib_rehash(struct sock *sk, u16 newhash)
1362 {
1363 if (sk_hashed(sk)) {
1364 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1365 struct udp_hslot *hslot, *hslot2, *nhslot2;
1366
1367 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1368 nhslot2 = udp_hashslot2(udptable, newhash);
1369 udp_sk(sk)->udp_portaddr_hash = newhash;
1370 if (hslot2 != nhslot2) {
1371 hslot = udp_hashslot(udptable, sock_net(sk),
1372 udp_sk(sk)->udp_port_hash);
1373 /* we must lock primary chain too */
1374 spin_lock_bh(&hslot->lock);
1375
1376 spin_lock(&hslot2->lock);
1377 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1378 hslot2->count--;
1379 spin_unlock(&hslot2->lock);
1380
1381 spin_lock(&nhslot2->lock);
1382 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1383 &nhslot2->head);
1384 nhslot2->count++;
1385 spin_unlock(&nhslot2->lock);
1386
1387 spin_unlock_bh(&hslot->lock);
1388 }
1389 }
1390 }
1391 EXPORT_SYMBOL(udp_lib_rehash);
1392
1393 static void udp_v4_rehash(struct sock *sk)
1394 {
1395 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1396 inet_sk(sk)->inet_rcv_saddr,
1397 inet_sk(sk)->inet_num);
1398 udp_lib_rehash(sk, new_hash);
1399 }
1400
1401 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1402 {
1403 int rc;
1404
1405 if (inet_sk(sk)->inet_daddr)
1406 sock_rps_save_rxhash(sk, skb);
1407
1408 rc = sock_queue_rcv_skb(sk, skb);
1409 if (rc < 0) {
1410 int is_udplite = IS_UDPLITE(sk);
1411
1412 /* Note that an ENOMEM error is charged twice */
1413 if (rc == -ENOMEM)
1414 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1415 is_udplite);
1416 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1417 kfree_skb(skb);
1418 trace_udp_fail_queue_rcv_skb(rc, sk);
1419 return -1;
1420 }
1421
1422 return 0;
1423
1424 }
1425
1426 static struct static_key udp_encap_needed __read_mostly;
1427 void udp_encap_enable(void)
1428 {
1429 if (!static_key_enabled(&udp_encap_needed))
1430 static_key_slow_inc(&udp_encap_needed);
1431 }
1432 EXPORT_SYMBOL(udp_encap_enable);
1433
1434 /* returns:
1435 * -1: error
1436 * 0: success
1437 * >0: "udp encap" protocol resubmission
1438 *
1439 * Note that in the success and error cases, the skb is assumed to
1440 * have either been requeued or freed.
1441 */
1442 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1443 {
1444 struct udp_sock *up = udp_sk(sk);
1445 int rc;
1446 int is_udplite = IS_UDPLITE(sk);
1447
1448 /*
1449 * Charge it to the socket, dropping if the queue is full.
1450 */
1451 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1452 goto drop;
1453 nf_reset(skb);
1454
1455 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1456 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1457
1458 /*
1459 * This is an encapsulation socket so pass the skb to
1460 * the socket's udp_encap_rcv() hook. Otherwise, just
1461 * fall through and pass this up the UDP socket.
1462 * up->encap_rcv() returns the following value:
1463 * =0 if skb was successfully passed to the encap
1464 * handler or was discarded by it.
1465 * >0 if skb should be passed on to UDP.
1466 * <0 if skb should be resubmitted as proto -N
1467 */
1468
1469 /* if we're overly short, let UDP handle it */
1470 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1471 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
1472 int ret;
1473
1474 ret = encap_rcv(sk, skb);
1475 if (ret <= 0) {
1476 UDP_INC_STATS_BH(sock_net(sk),
1477 UDP_MIB_INDATAGRAMS,
1478 is_udplite);
1479 return -ret;
1480 }
1481 }
1482
1483 /* FALLTHROUGH -- it's a UDP Packet */
1484 }
1485
1486 /*
1487 * UDP-Lite specific tests, ignored on UDP sockets
1488 */
1489 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1490
1491 /*
1492 * MIB statistics other than incrementing the error count are
1493 * disabled for the following two types of errors: these depend
1494 * on the application settings, not on the functioning of the
1495 * protocol stack as such.
1496 *
1497 * RFC 3828 here recommends (sec 3.3): "There should also be a
1498 * way ... to ... at least let the receiving application block
1499 * delivery of packets with coverage values less than a value
1500 * provided by the application."
1501 */
1502 if (up->pcrlen == 0) { /* full coverage was set */
1503 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: partial coverage %d while full coverage %d requested\n",
1504 UDP_SKB_CB(skb)->cscov, skb->len);
1505 goto drop;
1506 }
1507 /* The next case involves violating the min. coverage requested
1508 * by the receiver. This is subtle: if receiver wants x and x is
1509 * greater than the buffersize/MTU then receiver will complain
1510 * that it wants x while sender emits packets of smaller size y.
1511 * Therefore the above ...()->partial_cov statement is essential.
1512 */
1513 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1514 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: coverage %d too small, need min %d\n",
1515 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1516 goto drop;
1517 }
1518 }
1519
1520 if (rcu_access_pointer(sk->sk_filter) &&
1521 udp_lib_checksum_complete(skb))
1522 goto csum_error;
1523
1524
1525 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
1526 goto drop;
1527
1528 rc = 0;
1529
1530 ipv4_pktinfo_prepare(skb);
1531 bh_lock_sock(sk);
1532 if (!sock_owned_by_user(sk))
1533 rc = __udp_queue_rcv_skb(sk, skb);
1534 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1535 bh_unlock_sock(sk);
1536 goto drop;
1537 }
1538 bh_unlock_sock(sk);
1539
1540 return rc;
1541
1542 csum_error:
1543 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1544 drop:
1545 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1546 atomic_inc(&sk->sk_drops);
1547 kfree_skb(skb);
1548 return -1;
1549 }
1550
1551
1552 static void flush_stack(struct sock **stack, unsigned int count,
1553 struct sk_buff *skb, unsigned int final)
1554 {
1555 unsigned int i;
1556 struct sk_buff *skb1 = NULL;
1557 struct sock *sk;
1558
1559 for (i = 0; i < count; i++) {
1560 sk = stack[i];
1561 if (likely(skb1 == NULL))
1562 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1563
1564 if (!skb1) {
1565 atomic_inc(&sk->sk_drops);
1566 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1567 IS_UDPLITE(sk));
1568 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1569 IS_UDPLITE(sk));
1570 }
1571
1572 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1573 skb1 = NULL;
1574 }
1575 if (unlikely(skb1))
1576 kfree_skb(skb1);
1577 }
1578
1579 /*
1580 * Multicasts and broadcasts go to each listener.
1581 *
1582 * Note: called only from the BH handler context.
1583 */
1584 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1585 struct udphdr *uh,
1586 __be32 saddr, __be32 daddr,
1587 struct udp_table *udptable)
1588 {
1589 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1590 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1591 int dif;
1592 unsigned int i, count = 0;
1593
1594 spin_lock(&hslot->lock);
1595 sk = sk_nulls_head(&hslot->head);
1596 dif = skb->dev->ifindex;
1597 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1598 while (sk) {
1599 stack[count++] = sk;
1600 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1601 daddr, uh->source, saddr, dif);
1602 if (unlikely(count == ARRAY_SIZE(stack))) {
1603 if (!sk)
1604 break;
1605 flush_stack(stack, count, skb, ~0);
1606 count = 0;
1607 }
1608 }
1609 /*
1610 * before releasing chain lock, we must take a reference on sockets
1611 */
1612 for (i = 0; i < count; i++)
1613 sock_hold(stack[i]);
1614
1615 spin_unlock(&hslot->lock);
1616
1617 /*
1618 * do the slow work with no lock held
1619 */
1620 if (count) {
1621 flush_stack(stack, count, skb, count - 1);
1622
1623 for (i = 0; i < count; i++)
1624 sock_put(stack[i]);
1625 } else {
1626 kfree_skb(skb);
1627 }
1628 return 0;
1629 }
1630
1631 /* Initialize UDP checksum. If exited with zero value (success),
1632 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1633 * Otherwise, csum completion requires chacksumming packet body,
1634 * including udp header and folding it to skb->csum.
1635 */
1636 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1637 int proto)
1638 {
1639 const struct iphdr *iph;
1640 int err;
1641
1642 UDP_SKB_CB(skb)->partial_cov = 0;
1643 UDP_SKB_CB(skb)->cscov = skb->len;
1644
1645 if (proto == IPPROTO_UDPLITE) {
1646 err = udplite_checksum_init(skb, uh);
1647 if (err)
1648 return err;
1649 }
1650
1651 iph = ip_hdr(skb);
1652 if (uh->check == 0) {
1653 skb->ip_summed = CHECKSUM_UNNECESSARY;
1654 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1655 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1656 proto, skb->csum))
1657 skb->ip_summed = CHECKSUM_UNNECESSARY;
1658 }
1659 if (!skb_csum_unnecessary(skb))
1660 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1661 skb->len, proto, 0);
1662 /* Probably, we should checksum udp header (it should be in cache
1663 * in any case) and data in tiny packets (< rx copybreak).
1664 */
1665
1666 return 0;
1667 }
1668
1669 /*
1670 * All we need to do is get the socket, and then do a checksum.
1671 */
1672
1673 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1674 int proto)
1675 {
1676 struct sock *sk;
1677 struct udphdr *uh;
1678 unsigned short ulen;
1679 struct rtable *rt = skb_rtable(skb);
1680 __be32 saddr, daddr;
1681 struct net *net = dev_net(skb->dev);
1682
1683 /*
1684 * Validate the packet.
1685 */
1686 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1687 goto drop; /* No space for header. */
1688
1689 uh = udp_hdr(skb);
1690 ulen = ntohs(uh->len);
1691 saddr = ip_hdr(skb)->saddr;
1692 daddr = ip_hdr(skb)->daddr;
1693
1694 if (ulen > skb->len)
1695 goto short_packet;
1696
1697 if (proto == IPPROTO_UDP) {
1698 /* UDP validates ulen. */
1699 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1700 goto short_packet;
1701 uh = udp_hdr(skb);
1702 }
1703
1704 if (udp4_csum_init(skb, uh, proto))
1705 goto csum_error;
1706
1707 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1708 return __udp4_lib_mcast_deliver(net, skb, uh,
1709 saddr, daddr, udptable);
1710
1711 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1712
1713 if (sk != NULL) {
1714 int ret;
1715
1716 sk_mark_napi_id(sk, skb);
1717 ret = udp_queue_rcv_skb(sk, skb);
1718 sock_put(sk);
1719
1720 /* a return value > 0 means to resubmit the input, but
1721 * it wants the return to be -protocol, or 0
1722 */
1723 if (ret > 0)
1724 return -ret;
1725 return 0;
1726 }
1727
1728 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1729 goto drop;
1730 nf_reset(skb);
1731
1732 /* No socket. Drop packet silently, if checksum is wrong */
1733 if (udp_lib_checksum_complete(skb))
1734 goto csum_error;
1735
1736 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1737 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1738
1739 /*
1740 * Hmm. We got an UDP packet to a port to which we
1741 * don't wanna listen. Ignore it.
1742 */
1743 kfree_skb(skb);
1744 return 0;
1745
1746 short_packet:
1747 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1748 proto == IPPROTO_UDPLITE ? "Lite" : "",
1749 &saddr, ntohs(uh->source),
1750 ulen, skb->len,
1751 &daddr, ntohs(uh->dest));
1752 goto drop;
1753
1754 csum_error:
1755 /*
1756 * RFC1122: OK. Discards the bad packet silently (as far as
1757 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1758 */
1759 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1760 proto == IPPROTO_UDPLITE ? "Lite" : "",
1761 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1762 ulen);
1763 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1764 drop:
1765 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1766 kfree_skb(skb);
1767 return 0;
1768 }
1769
1770 int udp_rcv(struct sk_buff *skb)
1771 {
1772 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1773 }
1774
1775 void udp_destroy_sock(struct sock *sk)
1776 {
1777 struct udp_sock *up = udp_sk(sk);
1778 bool slow = lock_sock_fast(sk);
1779 udp_flush_pending_frames(sk);
1780 unlock_sock_fast(sk, slow);
1781 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1782 void (*encap_destroy)(struct sock *sk);
1783 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1784 if (encap_destroy)
1785 encap_destroy(sk);
1786 }
1787 }
1788
1789 /*
1790 * Socket option code for UDP
1791 */
1792 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1793 char __user *optval, unsigned int optlen,
1794 int (*push_pending_frames)(struct sock *))
1795 {
1796 struct udp_sock *up = udp_sk(sk);
1797 int val;
1798 int err = 0;
1799 int is_udplite = IS_UDPLITE(sk);
1800
1801 if (optlen < sizeof(int))
1802 return -EINVAL;
1803
1804 if (get_user(val, (int __user *)optval))
1805 return -EFAULT;
1806
1807 switch (optname) {
1808 case UDP_CORK:
1809 if (val != 0) {
1810 up->corkflag = 1;
1811 } else {
1812 up->corkflag = 0;
1813 lock_sock(sk);
1814 (*push_pending_frames)(sk);
1815 release_sock(sk);
1816 }
1817 break;
1818
1819 case UDP_ENCAP:
1820 switch (val) {
1821 case 0:
1822 case UDP_ENCAP_ESPINUDP:
1823 case UDP_ENCAP_ESPINUDP_NON_IKE:
1824 up->encap_rcv = xfrm4_udp_encap_rcv;
1825 /* FALLTHROUGH */
1826 case UDP_ENCAP_L2TPINUDP:
1827 up->encap_type = val;
1828 udp_encap_enable();
1829 break;
1830 default:
1831 err = -ENOPROTOOPT;
1832 break;
1833 }
1834 break;
1835
1836 /*
1837 * UDP-Lite's partial checksum coverage (RFC 3828).
1838 */
1839 /* The sender sets actual checksum coverage length via this option.
1840 * The case coverage > packet length is handled by send module. */
1841 case UDPLITE_SEND_CSCOV:
1842 if (!is_udplite) /* Disable the option on UDP sockets */
1843 return -ENOPROTOOPT;
1844 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1845 val = 8;
1846 else if (val > USHRT_MAX)
1847 val = USHRT_MAX;
1848 up->pcslen = val;
1849 up->pcflag |= UDPLITE_SEND_CC;
1850 break;
1851
1852 /* The receiver specifies a minimum checksum coverage value. To make
1853 * sense, this should be set to at least 8 (as done below). If zero is
1854 * used, this again means full checksum coverage. */
1855 case UDPLITE_RECV_CSCOV:
1856 if (!is_udplite) /* Disable the option on UDP sockets */
1857 return -ENOPROTOOPT;
1858 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1859 val = 8;
1860 else if (val > USHRT_MAX)
1861 val = USHRT_MAX;
1862 up->pcrlen = val;
1863 up->pcflag |= UDPLITE_RECV_CC;
1864 break;
1865
1866 default:
1867 err = -ENOPROTOOPT;
1868 break;
1869 }
1870
1871 return err;
1872 }
1873 EXPORT_SYMBOL(udp_lib_setsockopt);
1874
1875 int udp_setsockopt(struct sock *sk, int level, int optname,
1876 char __user *optval, unsigned int optlen)
1877 {
1878 if (level == SOL_UDP || level == SOL_UDPLITE)
1879 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1880 udp_push_pending_frames);
1881 return ip_setsockopt(sk, level, optname, optval, optlen);
1882 }
1883
1884 #ifdef CONFIG_COMPAT
1885 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1886 char __user *optval, unsigned int optlen)
1887 {
1888 if (level == SOL_UDP || level == SOL_UDPLITE)
1889 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1890 udp_push_pending_frames);
1891 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1892 }
1893 #endif
1894
1895 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1896 char __user *optval, int __user *optlen)
1897 {
1898 struct udp_sock *up = udp_sk(sk);
1899 int val, len;
1900
1901 if (get_user(len, optlen))
1902 return -EFAULT;
1903
1904 len = min_t(unsigned int, len, sizeof(int));
1905
1906 if (len < 0)
1907 return -EINVAL;
1908
1909 switch (optname) {
1910 case UDP_CORK:
1911 val = up->corkflag;
1912 break;
1913
1914 case UDP_ENCAP:
1915 val = up->encap_type;
1916 break;
1917
1918 /* The following two cannot be changed on UDP sockets, the return is
1919 * always 0 (which corresponds to the full checksum coverage of UDP). */
1920 case UDPLITE_SEND_CSCOV:
1921 val = up->pcslen;
1922 break;
1923
1924 case UDPLITE_RECV_CSCOV:
1925 val = up->pcrlen;
1926 break;
1927
1928 default:
1929 return -ENOPROTOOPT;
1930 }
1931
1932 if (put_user(len, optlen))
1933 return -EFAULT;
1934 if (copy_to_user(optval, &val, len))
1935 return -EFAULT;
1936 return 0;
1937 }
1938 EXPORT_SYMBOL(udp_lib_getsockopt);
1939
1940 int udp_getsockopt(struct sock *sk, int level, int optname,
1941 char __user *optval, int __user *optlen)
1942 {
1943 if (level == SOL_UDP || level == SOL_UDPLITE)
1944 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1945 return ip_getsockopt(sk, level, optname, optval, optlen);
1946 }
1947
1948 #ifdef CONFIG_COMPAT
1949 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1950 char __user *optval, int __user *optlen)
1951 {
1952 if (level == SOL_UDP || level == SOL_UDPLITE)
1953 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1954 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1955 }
1956 #endif
1957 /**
1958 * udp_poll - wait for a UDP event.
1959 * @file - file struct
1960 * @sock - socket
1961 * @wait - poll table
1962 *
1963 * This is same as datagram poll, except for the special case of
1964 * blocking sockets. If application is using a blocking fd
1965 * and a packet with checksum error is in the queue;
1966 * then it could get return from select indicating data available
1967 * but then block when reading it. Add special case code
1968 * to work around these arguably broken applications.
1969 */
1970 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1971 {
1972 unsigned int mask = datagram_poll(file, sock, wait);
1973 struct sock *sk = sock->sk;
1974
1975 sock_rps_record_flow(sk);
1976
1977 /* Check for false positives due to checksum errors */
1978 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1979 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1980 mask &= ~(POLLIN | POLLRDNORM);
1981
1982 return mask;
1983
1984 }
1985 EXPORT_SYMBOL(udp_poll);
1986
1987 struct proto udp_prot = {
1988 .name = "UDP",
1989 .owner = THIS_MODULE,
1990 .close = udp_lib_close,
1991 .connect = ip4_datagram_connect,
1992 .disconnect = udp_disconnect,
1993 .ioctl = udp_ioctl,
1994 .destroy = udp_destroy_sock,
1995 .setsockopt = udp_setsockopt,
1996 .getsockopt = udp_getsockopt,
1997 .sendmsg = udp_sendmsg,
1998 .recvmsg = udp_recvmsg,
1999 .sendpage = udp_sendpage,
2000 .backlog_rcv = __udp_queue_rcv_skb,
2001 .release_cb = ip4_datagram_release_cb,
2002 .hash = udp_lib_hash,
2003 .unhash = udp_lib_unhash,
2004 .rehash = udp_v4_rehash,
2005 .get_port = udp_v4_get_port,
2006 .memory_allocated = &udp_memory_allocated,
2007 .sysctl_mem = sysctl_udp_mem,
2008 .sysctl_wmem = &sysctl_udp_wmem_min,
2009 .sysctl_rmem = &sysctl_udp_rmem_min,
2010 .obj_size = sizeof(struct udp_sock),
2011 .slab_flags = SLAB_DESTROY_BY_RCU,
2012 .h.udp_table = &udp_table,
2013 #ifdef CONFIG_COMPAT
2014 .compat_setsockopt = compat_udp_setsockopt,
2015 .compat_getsockopt = compat_udp_getsockopt,
2016 #endif
2017 .clear_sk = sk_prot_clear_portaddr_nulls,
2018 };
2019 EXPORT_SYMBOL(udp_prot);
2020
2021 /* ------------------------------------------------------------------------ */
2022 #ifdef CONFIG_PROC_FS
2023
2024 static struct sock *udp_get_first(struct seq_file *seq, int start)
2025 {
2026 struct sock *sk;
2027 struct udp_iter_state *state = seq->private;
2028 struct net *net = seq_file_net(seq);
2029
2030 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2031 ++state->bucket) {
2032 struct hlist_nulls_node *node;
2033 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2034
2035 if (hlist_nulls_empty(&hslot->head))
2036 continue;
2037
2038 spin_lock_bh(&hslot->lock);
2039 sk_nulls_for_each(sk, node, &hslot->head) {
2040 if (!net_eq(sock_net(sk), net))
2041 continue;
2042 if (sk->sk_family == state->family)
2043 goto found;
2044 }
2045 spin_unlock_bh(&hslot->lock);
2046 }
2047 sk = NULL;
2048 found:
2049 return sk;
2050 }
2051
2052 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2053 {
2054 struct udp_iter_state *state = seq->private;
2055 struct net *net = seq_file_net(seq);
2056
2057 do {
2058 sk = sk_nulls_next(sk);
2059 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2060
2061 if (!sk) {
2062 if (state->bucket <= state->udp_table->mask)
2063 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2064 return udp_get_first(seq, state->bucket + 1);
2065 }
2066 return sk;
2067 }
2068
2069 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2070 {
2071 struct sock *sk = udp_get_first(seq, 0);
2072
2073 if (sk)
2074 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2075 --pos;
2076 return pos ? NULL : sk;
2077 }
2078
2079 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2080 {
2081 struct udp_iter_state *state = seq->private;
2082 state->bucket = MAX_UDP_PORTS;
2083
2084 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2085 }
2086
2087 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2088 {
2089 struct sock *sk;
2090
2091 if (v == SEQ_START_TOKEN)
2092 sk = udp_get_idx(seq, 0);
2093 else
2094 sk = udp_get_next(seq, v);
2095
2096 ++*pos;
2097 return sk;
2098 }
2099
2100 static void udp_seq_stop(struct seq_file *seq, void *v)
2101 {
2102 struct udp_iter_state *state = seq->private;
2103
2104 if (state->bucket <= state->udp_table->mask)
2105 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2106 }
2107
2108 int udp_seq_open(struct inode *inode, struct file *file)
2109 {
2110 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2111 struct udp_iter_state *s;
2112 int err;
2113
2114 err = seq_open_net(inode, file, &afinfo->seq_ops,
2115 sizeof(struct udp_iter_state));
2116 if (err < 0)
2117 return err;
2118
2119 s = ((struct seq_file *)file->private_data)->private;
2120 s->family = afinfo->family;
2121 s->udp_table = afinfo->udp_table;
2122 return err;
2123 }
2124 EXPORT_SYMBOL(udp_seq_open);
2125
2126 /* ------------------------------------------------------------------------ */
2127 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2128 {
2129 struct proc_dir_entry *p;
2130 int rc = 0;
2131
2132 afinfo->seq_ops.start = udp_seq_start;
2133 afinfo->seq_ops.next = udp_seq_next;
2134 afinfo->seq_ops.stop = udp_seq_stop;
2135
2136 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2137 afinfo->seq_fops, afinfo);
2138 if (!p)
2139 rc = -ENOMEM;
2140 return rc;
2141 }
2142 EXPORT_SYMBOL(udp_proc_register);
2143
2144 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2145 {
2146 remove_proc_entry(afinfo->name, net->proc_net);
2147 }
2148 EXPORT_SYMBOL(udp_proc_unregister);
2149
2150 /* ------------------------------------------------------------------------ */
2151 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2152 int bucket, int *len)
2153 {
2154 struct inet_sock *inet = inet_sk(sp);
2155 __be32 dest = inet->inet_daddr;
2156 __be32 src = inet->inet_rcv_saddr;
2157 __u16 destp = ntohs(inet->inet_dport);
2158 __u16 srcp = ntohs(inet->inet_sport);
2159
2160 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2161 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d%n",
2162 bucket, src, srcp, dest, destp, sp->sk_state,
2163 sk_wmem_alloc_get(sp),
2164 sk_rmem_alloc_get(sp),
2165 0, 0L, 0,
2166 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2167 0, sock_i_ino(sp),
2168 atomic_read(&sp->sk_refcnt), sp,
2169 atomic_read(&sp->sk_drops), len);
2170 }
2171
2172 int udp4_seq_show(struct seq_file *seq, void *v)
2173 {
2174 if (v == SEQ_START_TOKEN)
2175 seq_printf(seq, "%-127s\n",
2176 " sl local_address rem_address st tx_queue "
2177 "rx_queue tr tm->when retrnsmt uid timeout "
2178 "inode ref pointer drops");
2179 else {
2180 struct udp_iter_state *state = seq->private;
2181 int len;
2182
2183 udp4_format_sock(v, seq, state->bucket, &len);
2184 seq_printf(seq, "%*s\n", 127 - len, "");
2185 }
2186 return 0;
2187 }
2188
2189 static const struct file_operations udp_afinfo_seq_fops = {
2190 .owner = THIS_MODULE,
2191 .open = udp_seq_open,
2192 .read = seq_read,
2193 .llseek = seq_lseek,
2194 .release = seq_release_net
2195 };
2196
2197 /* ------------------------------------------------------------------------ */
2198 static struct udp_seq_afinfo udp4_seq_afinfo = {
2199 .name = "udp",
2200 .family = AF_INET,
2201 .udp_table = &udp_table,
2202 .seq_fops = &udp_afinfo_seq_fops,
2203 .seq_ops = {
2204 .show = udp4_seq_show,
2205 },
2206 };
2207
2208 static int __net_init udp4_proc_init_net(struct net *net)
2209 {
2210 return udp_proc_register(net, &udp4_seq_afinfo);
2211 }
2212
2213 static void __net_exit udp4_proc_exit_net(struct net *net)
2214 {
2215 udp_proc_unregister(net, &udp4_seq_afinfo);
2216 }
2217
2218 static struct pernet_operations udp4_net_ops = {
2219 .init = udp4_proc_init_net,
2220 .exit = udp4_proc_exit_net,
2221 };
2222
2223 int __init udp4_proc_init(void)
2224 {
2225 return register_pernet_subsys(&udp4_net_ops);
2226 }
2227
2228 void udp4_proc_exit(void)
2229 {
2230 unregister_pernet_subsys(&udp4_net_ops);
2231 }
2232 #endif /* CONFIG_PROC_FS */
2233
2234 static __initdata unsigned long uhash_entries;
2235 static int __init set_uhash_entries(char *str)
2236 {
2237 ssize_t ret;
2238
2239 if (!str)
2240 return 0;
2241
2242 ret = kstrtoul(str, 0, &uhash_entries);
2243 if (ret)
2244 return 0;
2245
2246 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2247 uhash_entries = UDP_HTABLE_SIZE_MIN;
2248 return 1;
2249 }
2250 __setup("uhash_entries=", set_uhash_entries);
2251
2252 void __init udp_table_init(struct udp_table *table, const char *name)
2253 {
2254 unsigned int i;
2255
2256 table->hash = alloc_large_system_hash(name,
2257 2 * sizeof(struct udp_hslot),
2258 uhash_entries,
2259 21, /* one slot per 2 MB */
2260 0,
2261 &table->log,
2262 &table->mask,
2263 UDP_HTABLE_SIZE_MIN,
2264 64 * 1024);
2265
2266 table->hash2 = table->hash + (table->mask + 1);
2267 for (i = 0; i <= table->mask; i++) {
2268 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2269 table->hash[i].count = 0;
2270 spin_lock_init(&table->hash[i].lock);
2271 }
2272 for (i = 0; i <= table->mask; i++) {
2273 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2274 table->hash2[i].count = 0;
2275 spin_lock_init(&table->hash2[i].lock);
2276 }
2277 }
2278
2279 void __init udp_init(void)
2280 {
2281 unsigned long limit;
2282
2283 udp_table_init(&udp_table, "UDP");
2284 limit = nr_free_buffer_pages() / 8;
2285 limit = max(limit, 128UL);
2286 sysctl_udp_mem[0] = limit / 4 * 3;
2287 sysctl_udp_mem[1] = limit;
2288 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2289
2290 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2291 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2292 }
2293
2294 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
2295 netdev_features_t features)
2296 {
2297 struct sk_buff *segs = ERR_PTR(-EINVAL);
2298 int mac_len = skb->mac_len;
2299 int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
2300 __be16 protocol = skb->protocol;
2301 netdev_features_t enc_features;
2302 int outer_hlen;
2303
2304 if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
2305 goto out;
2306
2307 skb->encapsulation = 0;
2308 __skb_pull(skb, tnl_hlen);
2309 skb_reset_mac_header(skb);
2310 skb_set_network_header(skb, skb_inner_network_offset(skb));
2311 skb->mac_len = skb_inner_network_offset(skb);
2312 skb->protocol = htons(ETH_P_TEB);
2313
2314 /* segment inner packet. */
2315 enc_features = skb->dev->hw_enc_features & netif_skb_features(skb);
2316 segs = skb_mac_gso_segment(skb, enc_features);
2317 if (!segs || IS_ERR(segs))
2318 goto out;
2319
2320 outer_hlen = skb_tnl_header_len(skb);
2321 skb = segs;
2322 do {
2323 struct udphdr *uh;
2324 int udp_offset = outer_hlen - tnl_hlen;
2325
2326 skb_reset_inner_headers(skb);
2327 skb->encapsulation = 1;
2328
2329 skb->mac_len = mac_len;
2330
2331 skb_push(skb, outer_hlen);
2332 skb_reset_mac_header(skb);
2333 skb_set_network_header(skb, mac_len);
2334 skb_set_transport_header(skb, udp_offset);
2335 uh = udp_hdr(skb);
2336 uh->len = htons(skb->len - udp_offset);
2337
2338 /* csum segment if tunnel sets skb with csum. */
2339 if (unlikely(uh->check)) {
2340 struct iphdr *iph = ip_hdr(skb);
2341
2342 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
2343 skb->len - udp_offset,
2344 IPPROTO_UDP, 0);
2345 uh->check = csum_fold(skb_checksum(skb, udp_offset,
2346 skb->len - udp_offset, 0));
2347 if (uh->check == 0)
2348 uh->check = CSUM_MANGLED_0;
2349
2350 }
2351 skb->protocol = protocol;
2352 } while ((skb = skb->next));
2353 out:
2354 return segs;
2355 }
Linus' kernel tree