?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
fix a kmap leak in virtio_console
[linux/fpc-iii.git] / net / ipv4 / udp.c
blob77bd16fa9f34381a79e40f3b9c4b0adb4631f186
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/inet_hashtables.h>
107 #include <net/route.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <trace/events/udp.h>
111 #include <linux/static_key.h>
112 #include <trace/events/skb.h>
113 #include <net/busy_poll.h>
114 #include "udp_impl.h"
115
116 struct udp_table udp_table __read_mostly;
117 EXPORT_SYMBOL(udp_table);
118
119 long sysctl_udp_mem[3] __read_mostly;
120 EXPORT_SYMBOL(sysctl_udp_mem);
121
122 int sysctl_udp_rmem_min __read_mostly;
123 EXPORT_SYMBOL(sysctl_udp_rmem_min);
124
125 int sysctl_udp_wmem_min __read_mostly;
126 EXPORT_SYMBOL(sysctl_udp_wmem_min);
127
128 atomic_long_t udp_memory_allocated;
129 EXPORT_SYMBOL(udp_memory_allocated);
130
131 #define MAX_UDP_PORTS 65536
132 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
133
134 static int udp_lib_lport_inuse(struct net *net, __u16 num,
135 const struct udp_hslot *hslot,
136 unsigned long *bitmap,
137 struct sock *sk,
138 int (*saddr_comp)(const struct sock *sk1,
139 const struct sock *sk2),
140 unsigned int log)
141 {
142 struct sock *sk2;
143 struct hlist_nulls_node *node;
144 kuid_t uid = sock_i_uid(sk);
145
146 sk_nulls_for_each(sk2, node, &hslot->head)
147 if (net_eq(sock_net(sk2), net) &&
148 sk2 != sk &&
149 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
150 (!sk2->sk_reuse || !sk->sk_reuse) &&
151 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
152 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
153 (!sk2->sk_reuseport || !sk->sk_reuseport ||
154 !uid_eq(uid, sock_i_uid(sk2))) &&
155 (*saddr_comp)(sk, sk2)) {
156 if (bitmap)
157 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
158 bitmap);
159 else
160 return 1;
161 }
162 return 0;
163 }
164
165 /*
166 * Note: we still hold spinlock of primary hash chain, so no other writer
167 * can insert/delete a socket with local_port == num
168 */
169 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
170 struct udp_hslot *hslot2,
171 struct sock *sk,
172 int (*saddr_comp)(const struct sock *sk1,
173 const struct sock *sk2))
174 {
175 struct sock *sk2;
176 struct hlist_nulls_node *node;
177 kuid_t uid = sock_i_uid(sk);
178 int res = 0;
179
180 spin_lock(&hslot2->lock);
181 udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
182 if (net_eq(sock_net(sk2), net) &&
183 sk2 != sk &&
184 (udp_sk(sk2)->udp_port_hash == num) &&
185 (!sk2->sk_reuse || !sk->sk_reuse) &&
186 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
187 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
188 (!sk2->sk_reuseport || !sk->sk_reuseport ||
189 !uid_eq(uid, sock_i_uid(sk2))) &&
190 (*saddr_comp)(sk, sk2)) {
191 res = 1;
192 break;
193 }
194 spin_unlock(&hslot2->lock);
195 return res;
196 }
197
198 /**
199 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
200 *
201 * @sk: socket struct in question
202 * @snum: port number to look up
203 * @saddr_comp: AF-dependent comparison of bound local IP addresses
204 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
205 * with NULL address
206 */
207 int udp_lib_get_port(struct sock *sk, unsigned short snum,
208 int (*saddr_comp)(const struct sock *sk1,
209 const struct sock *sk2),
210 unsigned int hash2_nulladdr)
211 {
212 struct udp_hslot *hslot, *hslot2;
213 struct udp_table *udptable = sk->sk_prot->h.udp_table;
214 int error = 1;
215 struct net *net = sock_net(sk);
216
217 if (!snum) {
218 int low, high, remaining;
219 unsigned int rand;
220 unsigned short first, last;
221 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
222
223 inet_get_local_port_range(net, &low, &high);
224 remaining = (high - low) + 1;
225
226 rand = prandom_u32();
227 first = (((u64)rand * remaining) >> 32) + low;
228 /*
229 * force rand to be an odd multiple of UDP_HTABLE_SIZE
230 */
231 rand = (rand | 1) * (udptable->mask + 1);
232 last = first + udptable->mask + 1;
233 do {
234 hslot = udp_hashslot(udptable, net, first);
235 bitmap_zero(bitmap, PORTS_PER_CHAIN);
236 spin_lock_bh(&hslot->lock);
237 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
238 saddr_comp, udptable->log);
239
240 snum = first;
241 /*
242 * Iterate on all possible values of snum for this hash.
243 * Using steps of an odd multiple of UDP_HTABLE_SIZE
244 * give us randomization and full range coverage.
245 */
246 do {
247 if (low <= snum && snum <= high &&
248 !test_bit(snum >> udptable->log, bitmap) &&
249 !inet_is_reserved_local_port(snum))
250 goto found;
251 snum += rand;
252 } while (snum != first);
253 spin_unlock_bh(&hslot->lock);
254 } while (++first != last);
255 goto fail;
256 } else {
257 hslot = udp_hashslot(udptable, net, snum);
258 spin_lock_bh(&hslot->lock);
259 if (hslot->count > 10) {
260 int exist;
261 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
262
263 slot2 &= udptable->mask;
264 hash2_nulladdr &= udptable->mask;
265
266 hslot2 = udp_hashslot2(udptable, slot2);
267 if (hslot->count < hslot2->count)
268 goto scan_primary_hash;
269
270 exist = udp_lib_lport_inuse2(net, snum, hslot2,
271 sk, saddr_comp);
272 if (!exist && (hash2_nulladdr != slot2)) {
273 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
274 exist = udp_lib_lport_inuse2(net, snum, hslot2,
275 sk, saddr_comp);
276 }
277 if (exist)
278 goto fail_unlock;
279 else
280 goto found;
281 }
282 scan_primary_hash:
283 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
284 saddr_comp, 0))
285 goto fail_unlock;
286 }
287 found:
288 inet_sk(sk)->inet_num = snum;
289 udp_sk(sk)->udp_port_hash = snum;
290 udp_sk(sk)->udp_portaddr_hash ^= snum;
291 if (sk_unhashed(sk)) {
292 sk_nulls_add_node_rcu(sk, &hslot->head);
293 hslot->count++;
294 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
295
296 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
297 spin_lock(&hslot2->lock);
298 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
299 &hslot2->head);
300 hslot2->count++;
301 spin_unlock(&hslot2->lock);
302 }
303 error = 0;
304 fail_unlock:
305 spin_unlock_bh(&hslot->lock);
306 fail:
307 return error;
308 }
309 EXPORT_SYMBOL(udp_lib_get_port);
310
311 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
312 {
313 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
314
315 return (!ipv6_only_sock(sk2) &&
316 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
317 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
318 }
319
320 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
321 unsigned int port)
322 {
323 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
324 }
325
326 int udp_v4_get_port(struct sock *sk, unsigned short snum)
327 {
328 unsigned int hash2_nulladdr =
329 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
330 unsigned int hash2_partial =
331 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
332
333 /* precompute partial secondary hash */
334 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
335 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
336 }
337
338 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
339 unsigned short hnum,
340 __be16 sport, __be32 daddr, __be16 dport, int dif)
341 {
342 int score = -1;
343
344 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
345 !ipv6_only_sock(sk)) {
346 struct inet_sock *inet = inet_sk(sk);
347
348 score = (sk->sk_family == PF_INET ? 2 : 1);
349 if (inet->inet_rcv_saddr) {
350 if (inet->inet_rcv_saddr != daddr)
351 return -1;
352 score += 4;
353 }
354 if (inet->inet_daddr) {
355 if (inet->inet_daddr != saddr)
356 return -1;
357 score += 4;
358 }
359 if (inet->inet_dport) {
360 if (inet->inet_dport != sport)
361 return -1;
362 score += 4;
363 }
364 if (sk->sk_bound_dev_if) {
365 if (sk->sk_bound_dev_if != dif)
366 return -1;
367 score += 4;
368 }
369 }
370 return score;
371 }
372
373 /*
374 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
375 */
376 static inline int compute_score2(struct sock *sk, struct net *net,
377 __be32 saddr, __be16 sport,
378 __be32 daddr, unsigned int hnum, int dif)
379 {
380 int score = -1;
381
382 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
383 struct inet_sock *inet = inet_sk(sk);
384
385 if (inet->inet_rcv_saddr != daddr)
386 return -1;
387 if (inet->inet_num != hnum)
388 return -1;
389
390 score = (sk->sk_family == PF_INET ? 2 : 1);
391 if (inet->inet_daddr) {
392 if (inet->inet_daddr != saddr)
393 return -1;
394 score += 4;
395 }
396 if (inet->inet_dport) {
397 if (inet->inet_dport != sport)
398 return -1;
399 score += 4;
400 }
401 if (sk->sk_bound_dev_if) {
402 if (sk->sk_bound_dev_if != dif)
403 return -1;
404 score += 4;
405 }
406 }
407 return score;
408 }
409
410 static unsigned int udp_ehashfn(struct net *net, const __be32 laddr,
411 const __u16 lport, const __be32 faddr,
412 const __be16 fport)
413 {
414 static u32 udp_ehash_secret __read_mostly;
415
416 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
417
418 return __inet_ehashfn(laddr, lport, faddr, fport,
419 udp_ehash_secret + net_hash_mix(net));
420 }
421
422
423 /* called with read_rcu_lock() */
424 static struct sock *udp4_lib_lookup2(struct net *net,
425 __be32 saddr, __be16 sport,
426 __be32 daddr, unsigned int hnum, int dif,
427 struct udp_hslot *hslot2, unsigned int slot2)
428 {
429 struct sock *sk, *result;
430 struct hlist_nulls_node *node;
431 int score, badness, matches = 0, reuseport = 0;
432 u32 hash = 0;
433
434 begin:
435 result = NULL;
436 badness = 0;
437 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
438 score = compute_score2(sk, net, saddr, sport,
439 daddr, hnum, dif);
440 if (score > badness) {
441 result = sk;
442 badness = score;
443 reuseport = sk->sk_reuseport;
444 if (reuseport) {
445 hash = udp_ehashfn(net, daddr, hnum,
446 saddr, sport);
447 matches = 1;
448 }
449 } else if (score == badness && reuseport) {
450 matches++;
451 if (((u64)hash * matches) >> 32 == 0)
452 result = sk;
453 hash = next_pseudo_random32(hash);
454 }
455 }
456 /*
457 * if the nulls value we got at the end of this lookup is
458 * not the expected one, we must restart lookup.
459 * We probably met an item that was moved to another chain.
460 */
461 if (get_nulls_value(node) != slot2)
462 goto begin;
463 if (result) {
464 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
465 result = NULL;
466 else if (unlikely(compute_score2(result, net, saddr, sport,
467 daddr, hnum, dif) < badness)) {
468 sock_put(result);
469 goto begin;
470 }
471 }
472 return result;
473 }
474
475 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
476 * harder than this. -DaveM
477 */
478 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
479 __be16 sport, __be32 daddr, __be16 dport,
480 int dif, struct udp_table *udptable)
481 {
482 struct sock *sk, *result;
483 struct hlist_nulls_node *node;
484 unsigned short hnum = ntohs(dport);
485 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
486 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
487 int score, badness, matches = 0, reuseport = 0;
488 u32 hash = 0;
489
490 rcu_read_lock();
491 if (hslot->count > 10) {
492 hash2 = udp4_portaddr_hash(net, daddr, hnum);
493 slot2 = hash2 & udptable->mask;
494 hslot2 = &udptable->hash2[slot2];
495 if (hslot->count < hslot2->count)
496 goto begin;
497
498 result = udp4_lib_lookup2(net, saddr, sport,
499 daddr, hnum, dif,
500 hslot2, slot2);
501 if (!result) {
502 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
503 slot2 = hash2 & udptable->mask;
504 hslot2 = &udptable->hash2[slot2];
505 if (hslot->count < hslot2->count)
506 goto begin;
507
508 result = udp4_lib_lookup2(net, saddr, sport,
509 htonl(INADDR_ANY), hnum, dif,
510 hslot2, slot2);
511 }
512 rcu_read_unlock();
513 return result;
514 }
515 begin:
516 result = NULL;
517 badness = 0;
518 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
519 score = compute_score(sk, net, saddr, hnum, sport,
520 daddr, dport, dif);
521 if (score > badness) {
522 result = sk;
523 badness = score;
524 reuseport = sk->sk_reuseport;
525 if (reuseport) {
526 hash = udp_ehashfn(net, daddr, hnum,
527 saddr, sport);
528 matches = 1;
529 }
530 } else if (score == badness && reuseport) {
531 matches++;
532 if (((u64)hash * matches) >> 32 == 0)
533 result = sk;
534 hash = next_pseudo_random32(hash);
535 }
536 }
537 /*
538 * if the nulls value we got at the end of this lookup is
539 * not the expected one, we must restart lookup.
540 * We probably met an item that was moved to another chain.
541 */
542 if (get_nulls_value(node) != slot)
543 goto begin;
544
545 if (result) {
546 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
547 result = NULL;
548 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
549 daddr, dport, dif) < badness)) {
550 sock_put(result);
551 goto begin;
552 }
553 }
554 rcu_read_unlock();
555 return result;
556 }
557 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
558
559 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
560 __be16 sport, __be16 dport,
561 struct udp_table *udptable)
562 {
563 const struct iphdr *iph = ip_hdr(skb);
564
565 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
566 iph->daddr, dport, inet_iif(skb),
567 udptable);
568 }
569
570 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
571 __be32 daddr, __be16 dport, int dif)
572 {
573 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
574 }
575 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
576
577 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
578 __be16 loc_port, __be32 loc_addr,
579 __be16 rmt_port, __be32 rmt_addr,
580 int dif, unsigned short hnum)
581 {
582 struct inet_sock *inet = inet_sk(sk);
583
584 if (!net_eq(sock_net(sk), net) ||
585 udp_sk(sk)->udp_port_hash != hnum ||
586 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
587 (inet->inet_dport != rmt_port && inet->inet_dport) ||
588 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
589 ipv6_only_sock(sk) ||
590 (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
591 return false;
592 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif))
593 return false;
594 return true;
595 }
596
597 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
598 __be16 loc_port, __be32 loc_addr,
599 __be16 rmt_port, __be32 rmt_addr,
600 int dif)
601 {
602 struct hlist_nulls_node *node;
603 struct sock *s = sk;
604 unsigned short hnum = ntohs(loc_port);
605
606 sk_nulls_for_each_from(s, node) {
607 if (__udp_is_mcast_sock(net, s,
608 loc_port, loc_addr,
609 rmt_port, rmt_addr,
610 dif, hnum))
611 goto found;
612 }
613 s = NULL;
614 found:
615 return s;
616 }
617
618 /*
619 * This routine is called by the ICMP module when it gets some
620 * sort of error condition. If err < 0 then the socket should
621 * be closed and the error returned to the user. If err > 0
622 * it's just the icmp type << 8 | icmp code.
623 * Header points to the ip header of the error packet. We move
624 * on past this. Then (as it used to claim before adjustment)
625 * header points to the first 8 bytes of the udp header. We need
626 * to find the appropriate port.
627 */
628
629 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
630 {
631 struct inet_sock *inet;
632 const struct iphdr *iph = (const struct iphdr *)skb->data;
633 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
634 const int type = icmp_hdr(skb)->type;
635 const int code = icmp_hdr(skb)->code;
636 struct sock *sk;
637 int harderr;
638 int err;
639 struct net *net = dev_net(skb->dev);
640
641 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
642 iph->saddr, uh->source, skb->dev->ifindex, udptable);
643 if (sk == NULL) {
644 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
645 return; /* No socket for error */
646 }
647
648 err = 0;
649 harderr = 0;
650 inet = inet_sk(sk);
651
652 switch (type) {
653 default:
654 case ICMP_TIME_EXCEEDED:
655 err = EHOSTUNREACH;
656 break;
657 case ICMP_SOURCE_QUENCH:
658 goto out;
659 case ICMP_PARAMETERPROB:
660 err = EPROTO;
661 harderr = 1;
662 break;
663 case ICMP_DEST_UNREACH:
664 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
665 ipv4_sk_update_pmtu(skb, sk, info);
666 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
667 err = EMSGSIZE;
668 harderr = 1;
669 break;
670 }
671 goto out;
672 }
673 err = EHOSTUNREACH;
674 if (code <= NR_ICMP_UNREACH) {
675 harderr = icmp_err_convert[code].fatal;
676 err = icmp_err_convert[code].errno;
677 }
678 break;
679 case ICMP_REDIRECT:
680 ipv4_sk_redirect(skb, sk);
681 goto out;
682 }
683
684 /*
685 * RFC1122: OK. Passes ICMP errors back to application, as per
686 * 4.1.3.3.
687 */
688 if (!inet->recverr) {
689 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
690 goto out;
691 } else
692 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
693
694 sk->sk_err = err;
695 sk->sk_error_report(sk);
696 out:
697 sock_put(sk);
698 }
699
700 void udp_err(struct sk_buff *skb, u32 info)
701 {
702 __udp4_lib_err(skb, info, &udp_table);
703 }
704
705 /*
706 * Throw away all pending data and cancel the corking. Socket is locked.
707 */
708 void udp_flush_pending_frames(struct sock *sk)
709 {
710 struct udp_sock *up = udp_sk(sk);
711
712 if (up->pending) {
713 up->len = 0;
714 up->pending = 0;
715 ip_flush_pending_frames(sk);
716 }
717 }
718 EXPORT_SYMBOL(udp_flush_pending_frames);
719
720 /**
721 * udp4_hwcsum - handle outgoing HW checksumming
722 * @skb: sk_buff containing the filled-in UDP header
723 * (checksum field must be zeroed out)
724 * @src: source IP address
725 * @dst: destination IP address
726 */
727 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
728 {
729 struct udphdr *uh = udp_hdr(skb);
730 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
731 int offset = skb_transport_offset(skb);
732 int len = skb->len - offset;
733 int hlen = len;
734 __wsum csum = 0;
735
736 if (!frags) {
737 /*
738 * Only one fragment on the socket.
739 */
740 skb->csum_start = skb_transport_header(skb) - skb->head;
741 skb->csum_offset = offsetof(struct udphdr, check);
742 uh->check = ~csum_tcpudp_magic(src, dst, len,
743 IPPROTO_UDP, 0);
744 } else {
745 /*
746 * HW-checksum won't work as there are two or more
747 * fragments on the socket so that all csums of sk_buffs
748 * should be together
749 */
750 do {
751 csum = csum_add(csum, frags->csum);
752 hlen -= frags->len;
753 } while ((frags = frags->next));
754
755 csum = skb_checksum(skb, offset, hlen, csum);
756 skb->ip_summed = CHECKSUM_NONE;
757
758 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
759 if (uh->check == 0)
760 uh->check = CSUM_MANGLED_0;
761 }
762 }
763 EXPORT_SYMBOL_GPL(udp4_hwcsum);
764
765 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
766 {
767 struct sock *sk = skb->sk;
768 struct inet_sock *inet = inet_sk(sk);
769 struct udphdr *uh;
770 int err = 0;
771 int is_udplite = IS_UDPLITE(sk);
772 int offset = skb_transport_offset(skb);
773 int len = skb->len - offset;
774 __wsum csum = 0;
775
776 /*
777 * Create a UDP header
778 */
779 uh = udp_hdr(skb);
780 uh->source = inet->inet_sport;
781 uh->dest = fl4->fl4_dport;
782 uh->len = htons(len);
783 uh->check = 0;
784
785 if (is_udplite) /* UDP-Lite */
786 csum = udplite_csum(skb);
787
788 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
789
790 skb->ip_summed = CHECKSUM_NONE;
791 goto send;
792
793 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
794
795 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
796 goto send;
797
798 } else
799 csum = udp_csum(skb);
800
801 /* add protocol-dependent pseudo-header */
802 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
803 sk->sk_protocol, csum);
804 if (uh->check == 0)
805 uh->check = CSUM_MANGLED_0;
806
807 send:
808 err = ip_send_skb(sock_net(sk), skb);
809 if (err) {
810 if (err == -ENOBUFS && !inet->recverr) {
811 UDP_INC_STATS_USER(sock_net(sk),
812 UDP_MIB_SNDBUFERRORS, is_udplite);
813 err = 0;
814 }
815 } else
816 UDP_INC_STATS_USER(sock_net(sk),
817 UDP_MIB_OUTDATAGRAMS, is_udplite);
818 return err;
819 }
820
821 /*
822 * Push out all pending data as one UDP datagram. Socket is locked.
823 */
824 int udp_push_pending_frames(struct sock *sk)
825 {
826 struct udp_sock *up = udp_sk(sk);
827 struct inet_sock *inet = inet_sk(sk);
828 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
829 struct sk_buff *skb;
830 int err = 0;
831
832 skb = ip_finish_skb(sk, fl4);
833 if (!skb)
834 goto out;
835
836 err = udp_send_skb(skb, fl4);
837
838 out:
839 up->len = 0;
840 up->pending = 0;
841 return err;
842 }
843 EXPORT_SYMBOL(udp_push_pending_frames);
844
845 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
846 size_t len)
847 {
848 struct inet_sock *inet = inet_sk(sk);
849 struct udp_sock *up = udp_sk(sk);
850 struct flowi4 fl4_stack;
851 struct flowi4 *fl4;
852 int ulen = len;
853 struct ipcm_cookie ipc;
854 struct rtable *rt = NULL;
855 int free = 0;
856 int connected = 0;
857 __be32 daddr, faddr, saddr;
858 __be16 dport;
859 u8 tos;
860 int err, is_udplite = IS_UDPLITE(sk);
861 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
862 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
863 struct sk_buff *skb;
864 struct ip_options_data opt_copy;
865
866 if (len > 0xFFFF)
867 return -EMSGSIZE;
868
869 /*
870 * Check the flags.
871 */
872
873 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
874 return -EOPNOTSUPP;
875
876 ipc.opt = NULL;
877 ipc.tx_flags = 0;
878 ipc.ttl = 0;
879 ipc.tos = -1;
880
881 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
882
883 fl4 = &inet->cork.fl.u.ip4;
884 if (up->pending) {
885 /*
886 * There are pending frames.
887 * The socket lock must be held while it's corked.
888 */
889 lock_sock(sk);
890 if (likely(up->pending)) {
891 if (unlikely(up->pending != AF_INET)) {
892 release_sock(sk);
893 return -EINVAL;
894 }
895 goto do_append_data;
896 }
897 release_sock(sk);
898 }
899 ulen += sizeof(struct udphdr);
900
901 /*
902 * Get and verify the address.
903 */
904 if (msg->msg_name) {
905 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
906 if (msg->msg_namelen < sizeof(*usin))
907 return -EINVAL;
908 if (usin->sin_family != AF_INET) {
909 if (usin->sin_family != AF_UNSPEC)
910 return -EAFNOSUPPORT;
911 }
912
913 daddr = usin->sin_addr.s_addr;
914 dport = usin->sin_port;
915 if (dport == 0)
916 return -EINVAL;
917 } else {
918 if (sk->sk_state != TCP_ESTABLISHED)
919 return -EDESTADDRREQ;
920 daddr = inet->inet_daddr;
921 dport = inet->inet_dport;
922 /* Open fast path for connected socket.
923 Route will not be used, if at least one option is set.
924 */
925 connected = 1;
926 }
927 ipc.addr = inet->inet_saddr;
928
929 ipc.oif = sk->sk_bound_dev_if;
930
931 sock_tx_timestamp(sk, &ipc.tx_flags);
932
933 if (msg->msg_controllen) {
934 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
935 if (err)
936 return err;
937 if (ipc.opt)
938 free = 1;
939 connected = 0;
940 }
941 if (!ipc.opt) {
942 struct ip_options_rcu *inet_opt;
943
944 rcu_read_lock();
945 inet_opt = rcu_dereference(inet->inet_opt);
946 if (inet_opt) {
947 memcpy(&opt_copy, inet_opt,
948 sizeof(*inet_opt) + inet_opt->opt.optlen);
949 ipc.opt = &opt_copy.opt;
950 }
951 rcu_read_unlock();
952 }
953
954 saddr = ipc.addr;
955 ipc.addr = faddr = daddr;
956
957 if (ipc.opt && ipc.opt->opt.srr) {
958 if (!daddr)
959 return -EINVAL;
960 faddr = ipc.opt->opt.faddr;
961 connected = 0;
962 }
963 tos = get_rttos(&ipc, inet);
964 if (sock_flag(sk, SOCK_LOCALROUTE) ||
965 (msg->msg_flags & MSG_DONTROUTE) ||
966 (ipc.opt && ipc.opt->opt.is_strictroute)) {
967 tos |= RTO_ONLINK;
968 connected = 0;
969 }
970
971 if (ipv4_is_multicast(daddr)) {
972 if (!ipc.oif)
973 ipc.oif = inet->mc_index;
974 if (!saddr)
975 saddr = inet->mc_addr;
976 connected = 0;
977 } else if (!ipc.oif)
978 ipc.oif = inet->uc_index;
979
980 if (connected)
981 rt = (struct rtable *)sk_dst_check(sk, 0);
982
983 if (rt == NULL) {
984 struct net *net = sock_net(sk);
985
986 fl4 = &fl4_stack;
987 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
988 RT_SCOPE_UNIVERSE, sk->sk_protocol,
989 inet_sk_flowi_flags(sk),
990 faddr, saddr, dport, inet->inet_sport);
991
992 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
993 rt = ip_route_output_flow(net, fl4, sk);
994 if (IS_ERR(rt)) {
995 err = PTR_ERR(rt);
996 rt = NULL;
997 if (err == -ENETUNREACH)
998 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
999 goto out;
1000 }
1001
1002 err = -EACCES;
1003 if ((rt->rt_flags & RTCF_BROADCAST) &&
1004 !sock_flag(sk, SOCK_BROADCAST))
1005 goto out;
1006 if (connected)
1007 sk_dst_set(sk, dst_clone(&rt->dst));
1008 }
1009
1010 if (msg->msg_flags&MSG_CONFIRM)
1011 goto do_confirm;
1012 back_from_confirm:
1013
1014 saddr = fl4->saddr;
1015 if (!ipc.addr)
1016 daddr = ipc.addr = fl4->daddr;
1017
1018 /* Lockless fast path for the non-corking case. */
1019 if (!corkreq) {
1020 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen,
1021 sizeof(struct udphdr), &ipc, &rt,
1022 msg->msg_flags);
1023 err = PTR_ERR(skb);
1024 if (!IS_ERR_OR_NULL(skb))
1025 err = udp_send_skb(skb, fl4);
1026 goto out;
1027 }
1028
1029 lock_sock(sk);
1030 if (unlikely(up->pending)) {
1031 /* The socket is already corked while preparing it. */
1032 /* ... which is an evident application bug. --ANK */
1033 release_sock(sk);
1034
1035 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("cork app bug 2\n"));
1036 err = -EINVAL;
1037 goto out;
1038 }
1039 /*
1040 * Now cork the socket to pend data.
1041 */
1042 fl4 = &inet->cork.fl.u.ip4;
1043 fl4->daddr = daddr;
1044 fl4->saddr = saddr;
1045 fl4->fl4_dport = dport;
1046 fl4->fl4_sport = inet->inet_sport;
1047 up->pending = AF_INET;
1048
1049 do_append_data:
1050 up->len += ulen;
1051 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen,
1052 sizeof(struct udphdr), &ipc, &rt,
1053 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1054 if (err)
1055 udp_flush_pending_frames(sk);
1056 else if (!corkreq)
1057 err = udp_push_pending_frames(sk);
1058 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1059 up->pending = 0;
1060 release_sock(sk);
1061
1062 out:
1063 ip_rt_put(rt);
1064 if (free)
1065 kfree(ipc.opt);
1066 if (!err)
1067 return len;
1068 /*
1069 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1070 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1071 * we don't have a good statistic (IpOutDiscards but it can be too many
1072 * things). We could add another new stat but at least for now that
1073 * seems like overkill.
1074 */
1075 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1076 UDP_INC_STATS_USER(sock_net(sk),
1077 UDP_MIB_SNDBUFERRORS, is_udplite);
1078 }
1079 return err;
1080
1081 do_confirm:
1082 dst_confirm(&rt->dst);
1083 if (!(msg->msg_flags&MSG_PROBE) || len)
1084 goto back_from_confirm;
1085 err = 0;
1086 goto out;
1087 }
1088 EXPORT_SYMBOL(udp_sendmsg);
1089
1090 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1091 size_t size, int flags)
1092 {
1093 struct inet_sock *inet = inet_sk(sk);
1094 struct udp_sock *up = udp_sk(sk);
1095 int ret;
1096
1097 if (flags & MSG_SENDPAGE_NOTLAST)
1098 flags |= MSG_MORE;
1099
1100 if (!up->pending) {
1101 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1102
1103 /* Call udp_sendmsg to specify destination address which
1104 * sendpage interface can't pass.
1105 * This will succeed only when the socket is connected.
1106 */
1107 ret = udp_sendmsg(NULL, sk, &msg, 0);
1108 if (ret < 0)
1109 return ret;
1110 }
1111
1112 lock_sock(sk);
1113
1114 if (unlikely(!up->pending)) {
1115 release_sock(sk);
1116
1117 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("udp cork app bug 3\n"));
1118 return -EINVAL;
1119 }
1120
1121 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1122 page, offset, size, flags);
1123 if (ret == -EOPNOTSUPP) {
1124 release_sock(sk);
1125 return sock_no_sendpage(sk->sk_socket, page, offset,
1126 size, flags);
1127 }
1128 if (ret < 0) {
1129 udp_flush_pending_frames(sk);
1130 goto out;
1131 }
1132
1133 up->len += size;
1134 if (!(up->corkflag || (flags&MSG_MORE)))
1135 ret = udp_push_pending_frames(sk);
1136 if (!ret)
1137 ret = size;
1138 out:
1139 release_sock(sk);
1140 return ret;
1141 }
1142
1143
1144 /**
1145 * first_packet_length - return length of first packet in receive queue
1146 * @sk: socket
1147 *
1148 * Drops all bad checksum frames, until a valid one is found.
1149 * Returns the length of found skb, or 0 if none is found.
1150 */
1151 static unsigned int first_packet_length(struct sock *sk)
1152 {
1153 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1154 struct sk_buff *skb;
1155 unsigned int res;
1156
1157 __skb_queue_head_init(&list_kill);
1158
1159 spin_lock_bh(&rcvq->lock);
1160 while ((skb = skb_peek(rcvq)) != NULL &&
1161 udp_lib_checksum_complete(skb)) {
1162 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS,
1163 IS_UDPLITE(sk));
1164 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1165 IS_UDPLITE(sk));
1166 atomic_inc(&sk->sk_drops);
1167 __skb_unlink(skb, rcvq);
1168 __skb_queue_tail(&list_kill, skb);
1169 }
1170 res = skb ? skb->len : 0;
1171 spin_unlock_bh(&rcvq->lock);
1172
1173 if (!skb_queue_empty(&list_kill)) {
1174 bool slow = lock_sock_fast(sk);
1175
1176 __skb_queue_purge(&list_kill);
1177 sk_mem_reclaim_partial(sk);
1178 unlock_sock_fast(sk, slow);
1179 }
1180 return res;
1181 }
1182
1183 /*
1184 * IOCTL requests applicable to the UDP protocol
1185 */
1186
1187 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1188 {
1189 switch (cmd) {
1190 case SIOCOUTQ:
1191 {
1192 int amount = sk_wmem_alloc_get(sk);
1193
1194 return put_user(amount, (int __user *)arg);
1195 }
1196
1197 case SIOCINQ:
1198 {
1199 unsigned int amount = first_packet_length(sk);
1200
1201 if (amount)
1202 /*
1203 * We will only return the amount
1204 * of this packet since that is all
1205 * that will be read.
1206 */
1207 amount -= sizeof(struct udphdr);
1208
1209 return put_user(amount, (int __user *)arg);
1210 }
1211
1212 default:
1213 return -ENOIOCTLCMD;
1214 }
1215
1216 return 0;
1217 }
1218 EXPORT_SYMBOL(udp_ioctl);
1219
1220 /*
1221 * This should be easy, if there is something there we
1222 * return it, otherwise we block.
1223 */
1224
1225 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1226 size_t len, int noblock, int flags, int *addr_len)
1227 {
1228 struct inet_sock *inet = inet_sk(sk);
1229 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1230 struct sk_buff *skb;
1231 unsigned int ulen, copied;
1232 int peeked, off = 0;
1233 int err;
1234 int is_udplite = IS_UDPLITE(sk);
1235 bool slow;
1236
1237 if (flags & MSG_ERRQUEUE)
1238 return ip_recv_error(sk, msg, len, addr_len);
1239
1240 try_again:
1241 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1242 &peeked, &off, &err);
1243 if (!skb)
1244 goto out;
1245
1246 ulen = skb->len - sizeof(struct udphdr);
1247 copied = len;
1248 if (copied > ulen)
1249 copied = ulen;
1250 else if (copied < ulen)
1251 msg->msg_flags |= MSG_TRUNC;
1252
1253 /*
1254 * If checksum is needed at all, try to do it while copying the
1255 * data. If the data is truncated, or if we only want a partial
1256 * coverage checksum (UDP-Lite), do it before the copy.
1257 */
1258
1259 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1260 if (udp_lib_checksum_complete(skb))
1261 goto csum_copy_err;
1262 }
1263
1264 if (skb_csum_unnecessary(skb))
1265 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1266 msg->msg_iov, copied);
1267 else {
1268 err = skb_copy_and_csum_datagram_iovec(skb,
1269 sizeof(struct udphdr),
1270 msg->msg_iov);
1271
1272 if (err == -EINVAL)
1273 goto csum_copy_err;
1274 }
1275
1276 if (unlikely(err)) {
1277 trace_kfree_skb(skb, udp_recvmsg);
1278 if (!peeked) {
1279 atomic_inc(&sk->sk_drops);
1280 UDP_INC_STATS_USER(sock_net(sk),
1281 UDP_MIB_INERRORS, is_udplite);
1282 }
1283 goto out_free;
1284 }
1285
1286 if (!peeked)
1287 UDP_INC_STATS_USER(sock_net(sk),
1288 UDP_MIB_INDATAGRAMS, is_udplite);
1289
1290 sock_recv_ts_and_drops(msg, sk, skb);
1291
1292 /* Copy the address. */
1293 if (sin) {
1294 sin->sin_family = AF_INET;
1295 sin->sin_port = udp_hdr(skb)->source;
1296 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1297 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1298 *addr_len = sizeof(*sin);
1299 }
1300 if (inet->cmsg_flags)
1301 ip_cmsg_recv(msg, skb);
1302
1303 err = copied;
1304 if (flags & MSG_TRUNC)
1305 err = ulen;
1306
1307 out_free:
1308 skb_free_datagram_locked(sk, skb);
1309 out:
1310 return err;
1311
1312 csum_copy_err:
1313 slow = lock_sock_fast(sk);
1314 if (!skb_kill_datagram(sk, skb, flags)) {
1315 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1316 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1317 }
1318 unlock_sock_fast(sk, slow);
1319
1320 if (noblock)
1321 return -EAGAIN;
1322
1323 /* starting over for a new packet */
1324 msg->msg_flags &= ~MSG_TRUNC;
1325 goto try_again;
1326 }
1327
1328
1329 int udp_disconnect(struct sock *sk, int flags)
1330 {
1331 struct inet_sock *inet = inet_sk(sk);
1332 /*
1333 * 1003.1g - break association.
1334 */
1335
1336 sk->sk_state = TCP_CLOSE;
1337 inet->inet_daddr = 0;
1338 inet->inet_dport = 0;
1339 sock_rps_reset_rxhash(sk);
1340 sk->sk_bound_dev_if = 0;
1341 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1342 inet_reset_saddr(sk);
1343
1344 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1345 sk->sk_prot->unhash(sk);
1346 inet->inet_sport = 0;
1347 }
1348 sk_dst_reset(sk);
1349 return 0;
1350 }
1351 EXPORT_SYMBOL(udp_disconnect);
1352
1353 void udp_lib_unhash(struct sock *sk)
1354 {
1355 if (sk_hashed(sk)) {
1356 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1357 struct udp_hslot *hslot, *hslot2;
1358
1359 hslot = udp_hashslot(udptable, sock_net(sk),
1360 udp_sk(sk)->udp_port_hash);
1361 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1362
1363 spin_lock_bh(&hslot->lock);
1364 if (sk_nulls_del_node_init_rcu(sk)) {
1365 hslot->count--;
1366 inet_sk(sk)->inet_num = 0;
1367 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1368
1369 spin_lock(&hslot2->lock);
1370 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1371 hslot2->count--;
1372 spin_unlock(&hslot2->lock);
1373 }
1374 spin_unlock_bh(&hslot->lock);
1375 }
1376 }
1377 EXPORT_SYMBOL(udp_lib_unhash);
1378
1379 /*
1380 * inet_rcv_saddr was changed, we must rehash secondary hash
1381 */
1382 void udp_lib_rehash(struct sock *sk, u16 newhash)
1383 {
1384 if (sk_hashed(sk)) {
1385 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1386 struct udp_hslot *hslot, *hslot2, *nhslot2;
1387
1388 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1389 nhslot2 = udp_hashslot2(udptable, newhash);
1390 udp_sk(sk)->udp_portaddr_hash = newhash;
1391 if (hslot2 != nhslot2) {
1392 hslot = udp_hashslot(udptable, sock_net(sk),
1393 udp_sk(sk)->udp_port_hash);
1394 /* we must lock primary chain too */
1395 spin_lock_bh(&hslot->lock);
1396
1397 spin_lock(&hslot2->lock);
1398 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1399 hslot2->count--;
1400 spin_unlock(&hslot2->lock);
1401
1402 spin_lock(&nhslot2->lock);
1403 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1404 &nhslot2->head);
1405 nhslot2->count++;
1406 spin_unlock(&nhslot2->lock);
1407
1408 spin_unlock_bh(&hslot->lock);
1409 }
1410 }
1411 }
1412 EXPORT_SYMBOL(udp_lib_rehash);
1413
1414 static void udp_v4_rehash(struct sock *sk)
1415 {
1416 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1417 inet_sk(sk)->inet_rcv_saddr,
1418 inet_sk(sk)->inet_num);
1419 udp_lib_rehash(sk, new_hash);
1420 }
1421
1422 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1423 {
1424 int rc;
1425
1426 if (inet_sk(sk)->inet_daddr) {
1427 sock_rps_save_rxhash(sk, skb);
1428 sk_mark_napi_id(sk, skb);
1429 }
1430
1431 rc = sock_queue_rcv_skb(sk, skb);
1432 if (rc < 0) {
1433 int is_udplite = IS_UDPLITE(sk);
1434
1435 /* Note that an ENOMEM error is charged twice */
1436 if (rc == -ENOMEM)
1437 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1438 is_udplite);
1439 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1440 kfree_skb(skb);
1441 trace_udp_fail_queue_rcv_skb(rc, sk);
1442 return -1;
1443 }
1444
1445 return 0;
1446
1447 }
1448
1449 static struct static_key udp_encap_needed __read_mostly;
1450 void udp_encap_enable(void)
1451 {
1452 if (!static_key_enabled(&udp_encap_needed))
1453 static_key_slow_inc(&udp_encap_needed);
1454 }
1455 EXPORT_SYMBOL(udp_encap_enable);
1456
1457 /* returns:
1458 * -1: error
1459 * 0: success
1460 * >0: "udp encap" protocol resubmission
1461 *
1462 * Note that in the success and error cases, the skb is assumed to
1463 * have either been requeued or freed.
1464 */
1465 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1466 {
1467 struct udp_sock *up = udp_sk(sk);
1468 int rc;
1469 int is_udplite = IS_UDPLITE(sk);
1470
1471 /*
1472 * Charge it to the socket, dropping if the queue is full.
1473 */
1474 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1475 goto drop;
1476 nf_reset(skb);
1477
1478 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1479 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1480
1481 /*
1482 * This is an encapsulation socket so pass the skb to
1483 * the socket's udp_encap_rcv() hook. Otherwise, just
1484 * fall through and pass this up the UDP socket.
1485 * up->encap_rcv() returns the following value:
1486 * =0 if skb was successfully passed to the encap
1487 * handler or was discarded by it.
1488 * >0 if skb should be passed on to UDP.
1489 * <0 if skb should be resubmitted as proto -N
1490 */
1491
1492 /* if we're overly short, let UDP handle it */
1493 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1494 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
1495 int ret;
1496
1497 ret = encap_rcv(sk, skb);
1498 if (ret <= 0) {
1499 UDP_INC_STATS_BH(sock_net(sk),
1500 UDP_MIB_INDATAGRAMS,
1501 is_udplite);
1502 return -ret;
1503 }
1504 }
1505
1506 /* FALLTHROUGH -- it's a UDP Packet */
1507 }
1508
1509 /*
1510 * UDP-Lite specific tests, ignored on UDP sockets
1511 */
1512 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1513
1514 /*
1515 * MIB statistics other than incrementing the error count are
1516 * disabled for the following two types of errors: these depend
1517 * on the application settings, not on the functioning of the
1518 * protocol stack as such.
1519 *
1520 * RFC 3828 here recommends (sec 3.3): "There should also be a
1521 * way ... to ... at least let the receiving application block
1522 * delivery of packets with coverage values less than a value
1523 * provided by the application."
1524 */
1525 if (up->pcrlen == 0) { /* full coverage was set */
1526 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: partial coverage %d while full coverage %d requested\n",
1527 UDP_SKB_CB(skb)->cscov, skb->len);
1528 goto drop;
1529 }
1530 /* The next case involves violating the min. coverage requested
1531 * by the receiver. This is subtle: if receiver wants x and x is
1532 * greater than the buffersize/MTU then receiver will complain
1533 * that it wants x while sender emits packets of smaller size y.
1534 * Therefore the above ...()->partial_cov statement is essential.
1535 */
1536 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1537 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: coverage %d too small, need min %d\n",
1538 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1539 goto drop;
1540 }
1541 }
1542
1543 if (rcu_access_pointer(sk->sk_filter) &&
1544 udp_lib_checksum_complete(skb))
1545 goto csum_error;
1546
1547
1548 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
1549 goto drop;
1550
1551 rc = 0;
1552
1553 ipv4_pktinfo_prepare(sk, skb);
1554 bh_lock_sock(sk);
1555 if (!sock_owned_by_user(sk))
1556 rc = __udp_queue_rcv_skb(sk, skb);
1557 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1558 bh_unlock_sock(sk);
1559 goto drop;
1560 }
1561 bh_unlock_sock(sk);
1562
1563 return rc;
1564
1565 csum_error:
1566 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1567 drop:
1568 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1569 atomic_inc(&sk->sk_drops);
1570 kfree_skb(skb);
1571 return -1;
1572 }
1573
1574
1575 static void flush_stack(struct sock **stack, unsigned int count,
1576 struct sk_buff *skb, unsigned int final)
1577 {
1578 unsigned int i;
1579 struct sk_buff *skb1 = NULL;
1580 struct sock *sk;
1581
1582 for (i = 0; i < count; i++) {
1583 sk = stack[i];
1584 if (likely(skb1 == NULL))
1585 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1586
1587 if (!skb1) {
1588 atomic_inc(&sk->sk_drops);
1589 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1590 IS_UDPLITE(sk));
1591 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1592 IS_UDPLITE(sk));
1593 }
1594
1595 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1596 skb1 = NULL;
1597 }
1598 if (unlikely(skb1))
1599 kfree_skb(skb1);
1600 }
1601
1602 /* For TCP sockets, sk_rx_dst is protected by socket lock
1603 * For UDP, we use xchg() to guard against concurrent changes.
1604 */
1605 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1606 {
1607 struct dst_entry *old;
1608
1609 dst_hold(dst);
1610 old = xchg(&sk->sk_rx_dst, dst);
1611 dst_release(old);
1612 }
1613
1614 /*
1615 * Multicasts and broadcasts go to each listener.
1616 *
1617 * Note: called only from the BH handler context.
1618 */
1619 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1620 struct udphdr *uh,
1621 __be32 saddr, __be32 daddr,
1622 struct udp_table *udptable)
1623 {
1624 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1625 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1626 int dif;
1627 unsigned int i, count = 0;
1628
1629 spin_lock(&hslot->lock);
1630 sk = sk_nulls_head(&hslot->head);
1631 dif = skb->dev->ifindex;
1632 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1633 while (sk) {
1634 stack[count++] = sk;
1635 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1636 daddr, uh->source, saddr, dif);
1637 if (unlikely(count == ARRAY_SIZE(stack))) {
1638 if (!sk)
1639 break;
1640 flush_stack(stack, count, skb, ~0);
1641 count = 0;
1642 }
1643 }
1644 /*
1645 * before releasing chain lock, we must take a reference on sockets
1646 */
1647 for (i = 0; i < count; i++)
1648 sock_hold(stack[i]);
1649
1650 spin_unlock(&hslot->lock);
1651
1652 /*
1653 * do the slow work with no lock held
1654 */
1655 if (count) {
1656 flush_stack(stack, count, skb, count - 1);
1657
1658 for (i = 0; i < count; i++)
1659 sock_put(stack[i]);
1660 } else {
1661 kfree_skb(skb);
1662 }
1663 return 0;
1664 }
1665
1666 /* Initialize UDP checksum. If exited with zero value (success),
1667 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1668 * Otherwise, csum completion requires chacksumming packet body,
1669 * including udp header and folding it to skb->csum.
1670 */
1671 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1672 int proto)
1673 {
1674 const struct iphdr *iph;
1675 int err;
1676
1677 UDP_SKB_CB(skb)->partial_cov = 0;
1678 UDP_SKB_CB(skb)->cscov = skb->len;
1679
1680 if (proto == IPPROTO_UDPLITE) {
1681 err = udplite_checksum_init(skb, uh);
1682 if (err)
1683 return err;
1684 }
1685
1686 iph = ip_hdr(skb);
1687 if (uh->check == 0) {
1688 skb->ip_summed = CHECKSUM_UNNECESSARY;
1689 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1690 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1691 proto, skb->csum))
1692 skb->ip_summed = CHECKSUM_UNNECESSARY;
1693 }
1694 if (!skb_csum_unnecessary(skb))
1695 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1696 skb->len, proto, 0);
1697 /* Probably, we should checksum udp header (it should be in cache
1698 * in any case) and data in tiny packets (< rx copybreak).
1699 */
1700
1701 return 0;
1702 }
1703
1704 /*
1705 * All we need to do is get the socket, and then do a checksum.
1706 */
1707
1708 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1709 int proto)
1710 {
1711 struct sock *sk;
1712 struct udphdr *uh;
1713 unsigned short ulen;
1714 struct rtable *rt = skb_rtable(skb);
1715 __be32 saddr, daddr;
1716 struct net *net = dev_net(skb->dev);
1717
1718 /*
1719 * Validate the packet.
1720 */
1721 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1722 goto drop; /* No space for header. */
1723
1724 uh = udp_hdr(skb);
1725 ulen = ntohs(uh->len);
1726 saddr = ip_hdr(skb)->saddr;
1727 daddr = ip_hdr(skb)->daddr;
1728
1729 if (ulen > skb->len)
1730 goto short_packet;
1731
1732 if (proto == IPPROTO_UDP) {
1733 /* UDP validates ulen. */
1734 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1735 goto short_packet;
1736 uh = udp_hdr(skb);
1737 }
1738
1739 if (udp4_csum_init(skb, uh, proto))
1740 goto csum_error;
1741
1742 sk = skb_steal_sock(skb);
1743 if (sk) {
1744 struct dst_entry *dst = skb_dst(skb);
1745 int ret;
1746
1747 if (unlikely(sk->sk_rx_dst != dst))
1748 udp_sk_rx_dst_set(sk, dst);
1749
1750 ret = udp_queue_rcv_skb(sk, skb);
1751 sock_put(sk);
1752 /* a return value > 0 means to resubmit the input, but
1753 * it wants the return to be -protocol, or 0
1754 */
1755 if (ret > 0)
1756 return -ret;
1757 return 0;
1758 } else {
1759 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1760 return __udp4_lib_mcast_deliver(net, skb, uh,
1761 saddr, daddr, udptable);
1762
1763 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1764 }
1765
1766 if (sk != NULL) {
1767 int ret;
1768
1769 ret = udp_queue_rcv_skb(sk, skb);
1770 sock_put(sk);
1771
1772 /* a return value > 0 means to resubmit the input, but
1773 * it wants the return to be -protocol, or 0
1774 */
1775 if (ret > 0)
1776 return -ret;
1777 return 0;
1778 }
1779
1780 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1781 goto drop;
1782 nf_reset(skb);
1783
1784 /* No socket. Drop packet silently, if checksum is wrong */
1785 if (udp_lib_checksum_complete(skb))
1786 goto csum_error;
1787
1788 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1789 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1790
1791 /*
1792 * Hmm. We got an UDP packet to a port to which we
1793 * don't wanna listen. Ignore it.
1794 */
1795 kfree_skb(skb);
1796 return 0;
1797
1798 short_packet:
1799 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1800 proto == IPPROTO_UDPLITE ? "Lite" : "",
1801 &saddr, ntohs(uh->source),
1802 ulen, skb->len,
1803 &daddr, ntohs(uh->dest));
1804 goto drop;
1805
1806 csum_error:
1807 /*
1808 * RFC1122: OK. Discards the bad packet silently (as far as
1809 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1810 */
1811 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1812 proto == IPPROTO_UDPLITE ? "Lite" : "",
1813 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1814 ulen);
1815 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1816 drop:
1817 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1818 kfree_skb(skb);
1819 return 0;
1820 }
1821
1822 /* We can only early demux multicast if there is a single matching socket.
1823 * If more than one socket found returns NULL
1824 */
1825 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
1826 __be16 loc_port, __be32 loc_addr,
1827 __be16 rmt_port, __be32 rmt_addr,
1828 int dif)
1829 {
1830 struct sock *sk, *result;
1831 struct hlist_nulls_node *node;
1832 unsigned short hnum = ntohs(loc_port);
1833 unsigned int count, slot = udp_hashfn(net, hnum, udp_table.mask);
1834 struct udp_hslot *hslot = &udp_table.hash[slot];
1835
1836 rcu_read_lock();
1837 begin:
1838 count = 0;
1839 result = NULL;
1840 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
1841 if (__udp_is_mcast_sock(net, sk,
1842 loc_port, loc_addr,
1843 rmt_port, rmt_addr,
1844 dif, hnum)) {
1845 result = sk;
1846 ++count;
1847 }
1848 }
1849 /*
1850 * if the nulls value we got at the end of this lookup is
1851 * not the expected one, we must restart lookup.
1852 * We probably met an item that was moved to another chain.
1853 */
1854 if (get_nulls_value(node) != slot)
1855 goto begin;
1856
1857 if (result) {
1858 if (count != 1 ||
1859 unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
1860 result = NULL;
1861 else if (unlikely(!__udp_is_mcast_sock(net, result,
1862 loc_port, loc_addr,
1863 rmt_port, rmt_addr,
1864 dif, hnum))) {
1865 sock_put(result);
1866 result = NULL;
1867 }
1868 }
1869 rcu_read_unlock();
1870 return result;
1871 }
1872
1873 /* For unicast we should only early demux connected sockets or we can
1874 * break forwarding setups. The chains here can be long so only check
1875 * if the first socket is an exact match and if not move on.
1876 */
1877 static struct sock *__udp4_lib_demux_lookup(struct net *net,
1878 __be16 loc_port, __be32 loc_addr,
1879 __be16 rmt_port, __be32 rmt_addr,
1880 int dif)
1881 {
1882 struct sock *sk, *result;
1883 struct hlist_nulls_node *node;
1884 unsigned short hnum = ntohs(loc_port);
1885 unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
1886 unsigned int slot2 = hash2 & udp_table.mask;
1887 struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
1888 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr)
1889 const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
1890
1891 rcu_read_lock();
1892 result = NULL;
1893 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
1894 if (INET_MATCH(sk, net, acookie,
1895 rmt_addr, loc_addr, ports, dif))
1896 result = sk;
1897 /* Only check first socket in chain */
1898 break;
1899 }
1900
1901 if (result) {
1902 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
1903 result = NULL;
1904 else if (unlikely(!INET_MATCH(sk, net, acookie,
1905 rmt_addr, loc_addr,
1906 ports, dif))) {
1907 sock_put(result);
1908 result = NULL;
1909 }
1910 }
1911 rcu_read_unlock();
1912 return result;
1913 }
1914
1915 void udp_v4_early_demux(struct sk_buff *skb)
1916 {
1917 struct net *net = dev_net(skb->dev);
1918 const struct iphdr *iph;
1919 const struct udphdr *uh;
1920 struct sock *sk;
1921 struct dst_entry *dst;
1922 int dif = skb->dev->ifindex;
1923
1924 /* validate the packet */
1925 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
1926 return;
1927
1928 iph = ip_hdr(skb);
1929 uh = udp_hdr(skb);
1930
1931 if (skb->pkt_type == PACKET_BROADCAST ||
1932 skb->pkt_type == PACKET_MULTICAST)
1933 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
1934 uh->source, iph->saddr, dif);
1935 else if (skb->pkt_type == PACKET_HOST)
1936 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
1937 uh->source, iph->saddr, dif);
1938 else
1939 return;
1940
1941 if (!sk)
1942 return;
1943
1944 skb->sk = sk;
1945 skb->destructor = sock_edemux;
1946 dst = sk->sk_rx_dst;
1947
1948 if (dst)
1949 dst = dst_check(dst, 0);
1950 if (dst)
1951 skb_dst_set_noref(skb, dst);
1952 }
1953
1954 int udp_rcv(struct sk_buff *skb)
1955 {
1956 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1957 }
1958
1959 void udp_destroy_sock(struct sock *sk)
1960 {
1961 struct udp_sock *up = udp_sk(sk);
1962 bool slow = lock_sock_fast(sk);
1963 udp_flush_pending_frames(sk);
1964 unlock_sock_fast(sk, slow);
1965 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1966 void (*encap_destroy)(struct sock *sk);
1967 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1968 if (encap_destroy)
1969 encap_destroy(sk);
1970 }
1971 }
1972
1973 /*
1974 * Socket option code for UDP
1975 */
1976 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1977 char __user *optval, unsigned int optlen,
1978 int (*push_pending_frames)(struct sock *))
1979 {
1980 struct udp_sock *up = udp_sk(sk);
1981 int val;
1982 int err = 0;
1983 int is_udplite = IS_UDPLITE(sk);
1984
1985 if (optlen < sizeof(int))
1986 return -EINVAL;
1987
1988 if (get_user(val, (int __user *)optval))
1989 return -EFAULT;
1990
1991 switch (optname) {
1992 case UDP_CORK:
1993 if (val != 0) {
1994 up->corkflag = 1;
1995 } else {
1996 up->corkflag = 0;
1997 lock_sock(sk);
1998 (*push_pending_frames)(sk);
1999 release_sock(sk);
2000 }
2001 break;
2002
2003 case UDP_ENCAP:
2004 switch (val) {
2005 case 0:
2006 case UDP_ENCAP_ESPINUDP:
2007 case UDP_ENCAP_ESPINUDP_NON_IKE:
2008 up->encap_rcv = xfrm4_udp_encap_rcv;
2009 /* FALLTHROUGH */
2010 case UDP_ENCAP_L2TPINUDP:
2011 up->encap_type = val;
2012 udp_encap_enable();
2013 break;
2014 default:
2015 err = -ENOPROTOOPT;
2016 break;
2017 }
2018 break;
2019
2020 /*
2021 * UDP-Lite's partial checksum coverage (RFC 3828).
2022 */
2023 /* The sender sets actual checksum coverage length via this option.
2024 * The case coverage > packet length is handled by send module. */
2025 case UDPLITE_SEND_CSCOV:
2026 if (!is_udplite) /* Disable the option on UDP sockets */
2027 return -ENOPROTOOPT;
2028 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2029 val = 8;
2030 else if (val > USHRT_MAX)
2031 val = USHRT_MAX;
2032 up->pcslen = val;
2033 up->pcflag |= UDPLITE_SEND_CC;
2034 break;
2035
2036 /* The receiver specifies a minimum checksum coverage value. To make
2037 * sense, this should be set to at least 8 (as done below). If zero is
2038 * used, this again means full checksum coverage. */
2039 case UDPLITE_RECV_CSCOV:
2040 if (!is_udplite) /* Disable the option on UDP sockets */
2041 return -ENOPROTOOPT;
2042 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2043 val = 8;
2044 else if (val > USHRT_MAX)
2045 val = USHRT_MAX;
2046 up->pcrlen = val;
2047 up->pcflag |= UDPLITE_RECV_CC;
2048 break;
2049
2050 default:
2051 err = -ENOPROTOOPT;
2052 break;
2053 }
2054
2055 return err;
2056 }
2057 EXPORT_SYMBOL(udp_lib_setsockopt);
2058
2059 int udp_setsockopt(struct sock *sk, int level, int optname,
2060 char __user *optval, unsigned int optlen)
2061 {
2062 if (level == SOL_UDP || level == SOL_UDPLITE)
2063 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2064 udp_push_pending_frames);
2065 return ip_setsockopt(sk, level, optname, optval, optlen);
2066 }
2067
2068 #ifdef CONFIG_COMPAT
2069 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2070 char __user *optval, unsigned int optlen)
2071 {
2072 if (level == SOL_UDP || level == SOL_UDPLITE)
2073 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2074 udp_push_pending_frames);
2075 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2076 }
2077 #endif
2078
2079 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2080 char __user *optval, int __user *optlen)
2081 {
2082 struct udp_sock *up = udp_sk(sk);
2083 int val, len;
2084
2085 if (get_user(len, optlen))
2086 return -EFAULT;
2087
2088 len = min_t(unsigned int, len, sizeof(int));
2089
2090 if (len < 0)
2091 return -EINVAL;
2092
2093 switch (optname) {
2094 case UDP_CORK:
2095 val = up->corkflag;
2096 break;
2097
2098 case UDP_ENCAP:
2099 val = up->encap_type;
2100 break;
2101
2102 /* The following two cannot be changed on UDP sockets, the return is
2103 * always 0 (which corresponds to the full checksum coverage of UDP). */
2104 case UDPLITE_SEND_CSCOV:
2105 val = up->pcslen;
2106 break;
2107
2108 case UDPLITE_RECV_CSCOV:
2109 val = up->pcrlen;
2110 break;
2111
2112 default:
2113 return -ENOPROTOOPT;
2114 }
2115
2116 if (put_user(len, optlen))
2117 return -EFAULT;
2118 if (copy_to_user(optval, &val, len))
2119 return -EFAULT;
2120 return 0;
2121 }
2122 EXPORT_SYMBOL(udp_lib_getsockopt);
2123
2124 int udp_getsockopt(struct sock *sk, int level, int optname,
2125 char __user *optval, int __user *optlen)
2126 {
2127 if (level == SOL_UDP || level == SOL_UDPLITE)
2128 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2129 return ip_getsockopt(sk, level, optname, optval, optlen);
2130 }
2131
2132 #ifdef CONFIG_COMPAT
2133 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2134 char __user *optval, int __user *optlen)
2135 {
2136 if (level == SOL_UDP || level == SOL_UDPLITE)
2137 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2138 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2139 }
2140 #endif
2141 /**
2142 * udp_poll - wait for a UDP event.
2143 * @file - file struct
2144 * @sock - socket
2145 * @wait - poll table
2146 *
2147 * This is same as datagram poll, except for the special case of
2148 * blocking sockets. If application is using a blocking fd
2149 * and a packet with checksum error is in the queue;
2150 * then it could get return from select indicating data available
2151 * but then block when reading it. Add special case code
2152 * to work around these arguably broken applications.
2153 */
2154 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2155 {
2156 unsigned int mask = datagram_poll(file, sock, wait);
2157 struct sock *sk = sock->sk;
2158
2159 sock_rps_record_flow(sk);
2160
2161 /* Check for false positives due to checksum errors */
2162 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2163 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
2164 mask &= ~(POLLIN | POLLRDNORM);
2165
2166 return mask;
2167
2168 }
2169 EXPORT_SYMBOL(udp_poll);
2170
2171 struct proto udp_prot = {
2172 .name = "UDP",
2173 .owner = THIS_MODULE,
2174 .close = udp_lib_close,
2175 .connect = ip4_datagram_connect,
2176 .disconnect = udp_disconnect,
2177 .ioctl = udp_ioctl,
2178 .destroy = udp_destroy_sock,
2179 .setsockopt = udp_setsockopt,
2180 .getsockopt = udp_getsockopt,
2181 .sendmsg = udp_sendmsg,
2182 .recvmsg = udp_recvmsg,
2183 .sendpage = udp_sendpage,
2184 .backlog_rcv = __udp_queue_rcv_skb,
2185 .release_cb = ip4_datagram_release_cb,
2186 .hash = udp_lib_hash,
2187 .unhash = udp_lib_unhash,
2188 .rehash = udp_v4_rehash,
2189 .get_port = udp_v4_get_port,
2190 .memory_allocated = &udp_memory_allocated,
2191 .sysctl_mem = sysctl_udp_mem,
2192 .sysctl_wmem = &sysctl_udp_wmem_min,
2193 .sysctl_rmem = &sysctl_udp_rmem_min,
2194 .obj_size = sizeof(struct udp_sock),
2195 .slab_flags = SLAB_DESTROY_BY_RCU,
2196 .h.udp_table = &udp_table,
2197 #ifdef CONFIG_COMPAT
2198 .compat_setsockopt = compat_udp_setsockopt,
2199 .compat_getsockopt = compat_udp_getsockopt,
2200 #endif
2201 .clear_sk = sk_prot_clear_portaddr_nulls,
2202 };
2203 EXPORT_SYMBOL(udp_prot);
2204
2205 /* ------------------------------------------------------------------------ */
2206 #ifdef CONFIG_PROC_FS
2207
2208 static struct sock *udp_get_first(struct seq_file *seq, int start)
2209 {
2210 struct sock *sk;
2211 struct udp_iter_state *state = seq->private;
2212 struct net *net = seq_file_net(seq);
2213
2214 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2215 ++state->bucket) {
2216 struct hlist_nulls_node *node;
2217 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2218
2219 if (hlist_nulls_empty(&hslot->head))
2220 continue;
2221
2222 spin_lock_bh(&hslot->lock);
2223 sk_nulls_for_each(sk, node, &hslot->head) {
2224 if (!net_eq(sock_net(sk), net))
2225 continue;
2226 if (sk->sk_family == state->family)
2227 goto found;
2228 }
2229 spin_unlock_bh(&hslot->lock);
2230 }
2231 sk = NULL;
2232 found:
2233 return sk;
2234 }
2235
2236 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2237 {
2238 struct udp_iter_state *state = seq->private;
2239 struct net *net = seq_file_net(seq);
2240
2241 do {
2242 sk = sk_nulls_next(sk);
2243 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2244
2245 if (!sk) {
2246 if (state->bucket <= state->udp_table->mask)
2247 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2248 return udp_get_first(seq, state->bucket + 1);
2249 }
2250 return sk;
2251 }
2252
2253 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2254 {
2255 struct sock *sk = udp_get_first(seq, 0);
2256
2257 if (sk)
2258 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2259 --pos;
2260 return pos ? NULL : sk;
2261 }
2262
2263 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2264 {
2265 struct udp_iter_state *state = seq->private;
2266 state->bucket = MAX_UDP_PORTS;
2267
2268 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2269 }
2270
2271 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2272 {
2273 struct sock *sk;
2274
2275 if (v == SEQ_START_TOKEN)
2276 sk = udp_get_idx(seq, 0);
2277 else
2278 sk = udp_get_next(seq, v);
2279
2280 ++*pos;
2281 return sk;
2282 }
2283
2284 static void udp_seq_stop(struct seq_file *seq, void *v)
2285 {
2286 struct udp_iter_state *state = seq->private;
2287
2288 if (state->bucket <= state->udp_table->mask)
2289 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2290 }
2291
2292 int udp_seq_open(struct inode *inode, struct file *file)
2293 {
2294 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2295 struct udp_iter_state *s;
2296 int err;
2297
2298 err = seq_open_net(inode, file, &afinfo->seq_ops,
2299 sizeof(struct udp_iter_state));
2300 if (err < 0)
2301 return err;
2302
2303 s = ((struct seq_file *)file->private_data)->private;
2304 s->family = afinfo->family;
2305 s->udp_table = afinfo->udp_table;
2306 return err;
2307 }
2308 EXPORT_SYMBOL(udp_seq_open);
2309
2310 /* ------------------------------------------------------------------------ */
2311 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2312 {
2313 struct proc_dir_entry *p;
2314 int rc = 0;
2315
2316 afinfo->seq_ops.start = udp_seq_start;
2317 afinfo->seq_ops.next = udp_seq_next;
2318 afinfo->seq_ops.stop = udp_seq_stop;
2319
2320 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2321 afinfo->seq_fops, afinfo);
2322 if (!p)
2323 rc = -ENOMEM;
2324 return rc;
2325 }
2326 EXPORT_SYMBOL(udp_proc_register);
2327
2328 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2329 {
2330 remove_proc_entry(afinfo->name, net->proc_net);
2331 }
2332 EXPORT_SYMBOL(udp_proc_unregister);
2333
2334 /* ------------------------------------------------------------------------ */
2335 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2336 int bucket)
2337 {
2338 struct inet_sock *inet = inet_sk(sp);
2339 __be32 dest = inet->inet_daddr;
2340 __be32 src = inet->inet_rcv_saddr;
2341 __u16 destp = ntohs(inet->inet_dport);
2342 __u16 srcp = ntohs(inet->inet_sport);
2343
2344 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2345 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2346 bucket, src, srcp, dest, destp, sp->sk_state,
2347 sk_wmem_alloc_get(sp),
2348 sk_rmem_alloc_get(sp),
2349 0, 0L, 0,
2350 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2351 0, sock_i_ino(sp),
2352 atomic_read(&sp->sk_refcnt), sp,
2353 atomic_read(&sp->sk_drops));
2354 }
2355
2356 int udp4_seq_show(struct seq_file *seq, void *v)
2357 {
2358 seq_setwidth(seq, 127);
2359 if (v == SEQ_START_TOKEN)
2360 seq_puts(seq, " sl local_address rem_address st tx_queue "
2361 "rx_queue tr tm->when retrnsmt uid timeout "
2362 "inode ref pointer drops");
2363 else {
2364 struct udp_iter_state *state = seq->private;
2365
2366 udp4_format_sock(v, seq, state->bucket);
2367 }
2368 seq_pad(seq, '\n');
2369 return 0;
2370 }
2371
2372 static const struct file_operations udp_afinfo_seq_fops = {
2373 .owner = THIS_MODULE,
2374 .open = udp_seq_open,
2375 .read = seq_read,
2376 .llseek = seq_lseek,
2377 .release = seq_release_net
2378 };
2379
2380 /* ------------------------------------------------------------------------ */
2381 static struct udp_seq_afinfo udp4_seq_afinfo = {
2382 .name = "udp",
2383 .family = AF_INET,
2384 .udp_table = &udp_table,
2385 .seq_fops = &udp_afinfo_seq_fops,
2386 .seq_ops = {
2387 .show = udp4_seq_show,
2388 },
2389 };
2390
2391 static int __net_init udp4_proc_init_net(struct net *net)
2392 {
2393 return udp_proc_register(net, &udp4_seq_afinfo);
2394 }
2395
2396 static void __net_exit udp4_proc_exit_net(struct net *net)
2397 {
2398 udp_proc_unregister(net, &udp4_seq_afinfo);
2399 }
2400
2401 static struct pernet_operations udp4_net_ops = {
2402 .init = udp4_proc_init_net,
2403 .exit = udp4_proc_exit_net,
2404 };
2405
2406 int __init udp4_proc_init(void)
2407 {
2408 return register_pernet_subsys(&udp4_net_ops);
2409 }
2410
2411 void udp4_proc_exit(void)
2412 {
2413 unregister_pernet_subsys(&udp4_net_ops);
2414 }
2415 #endif /* CONFIG_PROC_FS */
2416
2417 static __initdata unsigned long uhash_entries;
2418 static int __init set_uhash_entries(char *str)
2419 {
2420 ssize_t ret;
2421
2422 if (!str)
2423 return 0;
2424
2425 ret = kstrtoul(str, 0, &uhash_entries);
2426 if (ret)
2427 return 0;
2428
2429 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2430 uhash_entries = UDP_HTABLE_SIZE_MIN;
2431 return 1;
2432 }
2433 __setup("uhash_entries=", set_uhash_entries);
2434
2435 void __init udp_table_init(struct udp_table *table, const char *name)
2436 {
2437 unsigned int i;
2438
2439 table->hash = alloc_large_system_hash(name,
2440 2 * sizeof(struct udp_hslot),
2441 uhash_entries,
2442 21, /* one slot per 2 MB */
2443 0,
2444 &table->log,
2445 &table->mask,
2446 UDP_HTABLE_SIZE_MIN,
2447 64 * 1024);
2448
2449 table->hash2 = table->hash + (table->mask + 1);
2450 for (i = 0; i <= table->mask; i++) {
2451 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2452 table->hash[i].count = 0;
2453 spin_lock_init(&table->hash[i].lock);
2454 }
2455 for (i = 0; i <= table->mask; i++) {
2456 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2457 table->hash2[i].count = 0;
2458 spin_lock_init(&table->hash2[i].lock);
2459 }
2460 }
2461
2462 void __init udp_init(void)
2463 {
2464 unsigned long limit;
2465
2466 udp_table_init(&udp_table, "UDP");
2467 limit = nr_free_buffer_pages() / 8;
2468 limit = max(limit, 128UL);
2469 sysctl_udp_mem[0] = limit / 4 * 3;
2470 sysctl_udp_mem[1] = limit;
2471 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2472
2473 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2474 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2475 }
2476
2477 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
2478 netdev_features_t features)
2479 {
2480 struct sk_buff *segs = ERR_PTR(-EINVAL);
2481 u16 mac_offset = skb->mac_header;
2482 int mac_len = skb->mac_len;
2483 int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
2484 __be16 protocol = skb->protocol;
2485 netdev_features_t enc_features;
2486 int outer_hlen;
2487
2488 if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
2489 goto out;
2490
2491 skb->encapsulation = 0;
2492 __skb_pull(skb, tnl_hlen);
2493 skb_reset_mac_header(skb);
2494 skb_set_network_header(skb, skb_inner_network_offset(skb));
2495 skb->mac_len = skb_inner_network_offset(skb);
2496 skb->protocol = htons(ETH_P_TEB);
2497
2498 /* segment inner packet. */
2499 enc_features = skb->dev->hw_enc_features & netif_skb_features(skb);
2500 segs = skb_mac_gso_segment(skb, enc_features);
2501 if (!segs || IS_ERR(segs)) {
2502 skb_gso_error_unwind(skb, protocol, tnl_hlen, mac_offset,
2503 mac_len);
2504 goto out;
2505 }
2506
2507 outer_hlen = skb_tnl_header_len(skb);
2508 skb = segs;
2509 do {
2510 struct udphdr *uh;
2511 int udp_offset = outer_hlen - tnl_hlen;
2512
2513 skb_reset_inner_headers(skb);
2514 skb->encapsulation = 1;
2515
2516 skb->mac_len = mac_len;
2517
2518 skb_push(skb, outer_hlen);
2519 skb_reset_mac_header(skb);
2520 skb_set_network_header(skb, mac_len);
2521 skb_set_transport_header(skb, udp_offset);
2522 uh = udp_hdr(skb);
2523 uh->len = htons(skb->len - udp_offset);
2524
2525 /* csum segment if tunnel sets skb with csum. */
2526 if (protocol == htons(ETH_P_IP) && unlikely(uh->check)) {
2527 struct iphdr *iph = ip_hdr(skb);
2528
2529 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
2530 skb->len - udp_offset,
2531 IPPROTO_UDP, 0);
2532 uh->check = csum_fold(skb_checksum(skb, udp_offset,
2533 skb->len - udp_offset, 0));
2534 if (uh->check == 0)
2535 uh->check = CSUM_MANGLED_0;
2536
2537 } else if (protocol == htons(ETH_P_IPV6)) {
2538 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
2539 u32 len = skb->len - udp_offset;
2540
2541 uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
2542 len, IPPROTO_UDP, 0);
2543 uh->check = csum_fold(skb_checksum(skb, udp_offset, len, 0));
2544 if (uh->check == 0)
2545 uh->check = CSUM_MANGLED_0;
2546 skb->ip_summed = CHECKSUM_NONE;
2547 }
2548
2549 skb->protocol = protocol;
2550 } while ((skb = skb->next));
2551 out:
2552 return segs;
2553 }
Linux with added FPC-III support