?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'renesas-boards-for-v3.13' of git://git.kernel.org/pub/scm/linux/kernel...
[linux-2.6.git] / net / ipv4 / udp.c
blob74d2c95db57f3768d62ee450ed6bab64ca09b26d
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 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 EXPORT_SYMBOL_GPL(udp4_hwcsum);
744
745 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
746 {
747 struct sock *sk = skb->sk;
748 struct inet_sock *inet = inet_sk(sk);
749 struct udphdr *uh;
750 int err = 0;
751 int is_udplite = IS_UDPLITE(sk);
752 int offset = skb_transport_offset(skb);
753 int len = skb->len - offset;
754 __wsum csum = 0;
755
756 /*
757 * Create a UDP header
758 */
759 uh = udp_hdr(skb);
760 uh->source = inet->inet_sport;
761 uh->dest = fl4->fl4_dport;
762 uh->len = htons(len);
763 uh->check = 0;
764
765 if (is_udplite) /* UDP-Lite */
766 csum = udplite_csum(skb);
767
768 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
769
770 skb->ip_summed = CHECKSUM_NONE;
771 goto send;
772
773 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
774
775 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
776 goto send;
777
778 } else
779 csum = udp_csum(skb);
780
781 /* add protocol-dependent pseudo-header */
782 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
783 sk->sk_protocol, csum);
784 if (uh->check == 0)
785 uh->check = CSUM_MANGLED_0;
786
787 send:
788 err = ip_send_skb(sock_net(sk), skb);
789 if (err) {
790 if (err == -ENOBUFS && !inet->recverr) {
791 UDP_INC_STATS_USER(sock_net(sk),
792 UDP_MIB_SNDBUFERRORS, is_udplite);
793 err = 0;
794 }
795 } else
796 UDP_INC_STATS_USER(sock_net(sk),
797 UDP_MIB_OUTDATAGRAMS, is_udplite);
798 return err;
799 }
800
801 /*
802 * Push out all pending data as one UDP datagram. Socket is locked.
803 */
804 int udp_push_pending_frames(struct sock *sk)
805 {
806 struct udp_sock *up = udp_sk(sk);
807 struct inet_sock *inet = inet_sk(sk);
808 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
809 struct sk_buff *skb;
810 int err = 0;
811
812 skb = ip_finish_skb(sk, fl4);
813 if (!skb)
814 goto out;
815
816 err = udp_send_skb(skb, fl4);
817
818 out:
819 up->len = 0;
820 up->pending = 0;
821 return err;
822 }
823 EXPORT_SYMBOL(udp_push_pending_frames);
824
825 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
826 size_t len)
827 {
828 struct inet_sock *inet = inet_sk(sk);
829 struct udp_sock *up = udp_sk(sk);
830 struct flowi4 fl4_stack;
831 struct flowi4 *fl4;
832 int ulen = len;
833 struct ipcm_cookie ipc;
834 struct rtable *rt = NULL;
835 int free = 0;
836 int connected = 0;
837 __be32 daddr, faddr, saddr;
838 __be16 dport;
839 u8 tos;
840 int err, is_udplite = IS_UDPLITE(sk);
841 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
842 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
843 struct sk_buff *skb;
844 struct ip_options_data opt_copy;
845
846 if (len > 0xFFFF)
847 return -EMSGSIZE;
848
849 /*
850 * Check the flags.
851 */
852
853 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
854 return -EOPNOTSUPP;
855
856 ipc.opt = NULL;
857 ipc.tx_flags = 0;
858
859 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
860
861 fl4 = &inet->cork.fl.u.ip4;
862 if (up->pending) {
863 /*
864 * There are pending frames.
865 * The socket lock must be held while it's corked.
866 */
867 lock_sock(sk);
868 if (likely(up->pending)) {
869 if (unlikely(up->pending != AF_INET)) {
870 release_sock(sk);
871 return -EINVAL;
872 }
873 goto do_append_data;
874 }
875 release_sock(sk);
876 }
877 ulen += sizeof(struct udphdr);
878
879 /*
880 * Get and verify the address.
881 */
882 if (msg->msg_name) {
883 struct sockaddr_in *usin = (struct sockaddr_in *)msg->msg_name;
884 if (msg->msg_namelen < sizeof(*usin))
885 return -EINVAL;
886 if (usin->sin_family != AF_INET) {
887 if (usin->sin_family != AF_UNSPEC)
888 return -EAFNOSUPPORT;
889 }
890
891 daddr = usin->sin_addr.s_addr;
892 dport = usin->sin_port;
893 if (dport == 0)
894 return -EINVAL;
895 } else {
896 if (sk->sk_state != TCP_ESTABLISHED)
897 return -EDESTADDRREQ;
898 daddr = inet->inet_daddr;
899 dport = inet->inet_dport;
900 /* Open fast path for connected socket.
901 Route will not be used, if at least one option is set.
902 */
903 connected = 1;
904 }
905 ipc.addr = inet->inet_saddr;
906
907 ipc.oif = sk->sk_bound_dev_if;
908
909 sock_tx_timestamp(sk, &ipc.tx_flags);
910
911 if (msg->msg_controllen) {
912 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
913 if (err)
914 return err;
915 if (ipc.opt)
916 free = 1;
917 connected = 0;
918 }
919 if (!ipc.opt) {
920 struct ip_options_rcu *inet_opt;
921
922 rcu_read_lock();
923 inet_opt = rcu_dereference(inet->inet_opt);
924 if (inet_opt) {
925 memcpy(&opt_copy, inet_opt,
926 sizeof(*inet_opt) + inet_opt->opt.optlen);
927 ipc.opt = &opt_copy.opt;
928 }
929 rcu_read_unlock();
930 }
931
932 saddr = ipc.addr;
933 ipc.addr = faddr = daddr;
934
935 if (ipc.opt && ipc.opt->opt.srr) {
936 if (!daddr)
937 return -EINVAL;
938 faddr = ipc.opt->opt.faddr;
939 connected = 0;
940 }
941 tos = RT_TOS(inet->tos);
942 if (sock_flag(sk, SOCK_LOCALROUTE) ||
943 (msg->msg_flags & MSG_DONTROUTE) ||
944 (ipc.opt && ipc.opt->opt.is_strictroute)) {
945 tos |= RTO_ONLINK;
946 connected = 0;
947 }
948
949 if (ipv4_is_multicast(daddr)) {
950 if (!ipc.oif)
951 ipc.oif = inet->mc_index;
952 if (!saddr)
953 saddr = inet->mc_addr;
954 connected = 0;
955 } else if (!ipc.oif)
956 ipc.oif = inet->uc_index;
957
958 if (connected)
959 rt = (struct rtable *)sk_dst_check(sk, 0);
960
961 if (rt == NULL) {
962 struct net *net = sock_net(sk);
963
964 fl4 = &fl4_stack;
965 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
966 RT_SCOPE_UNIVERSE, sk->sk_protocol,
967 inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP,
968 faddr, saddr, dport, inet->inet_sport);
969
970 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
971 rt = ip_route_output_flow(net, fl4, sk);
972 if (IS_ERR(rt)) {
973 err = PTR_ERR(rt);
974 rt = NULL;
975 if (err == -ENETUNREACH)
976 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
977 goto out;
978 }
979
980 err = -EACCES;
981 if ((rt->rt_flags & RTCF_BROADCAST) &&
982 !sock_flag(sk, SOCK_BROADCAST))
983 goto out;
984 if (connected)
985 sk_dst_set(sk, dst_clone(&rt->dst));
986 }
987
988 if (msg->msg_flags&MSG_CONFIRM)
989 goto do_confirm;
990 back_from_confirm:
991
992 saddr = fl4->saddr;
993 if (!ipc.addr)
994 daddr = ipc.addr = fl4->daddr;
995
996 /* Lockless fast path for the non-corking case. */
997 if (!corkreq) {
998 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen,
999 sizeof(struct udphdr), &ipc, &rt,
1000 msg->msg_flags);
1001 err = PTR_ERR(skb);
1002 if (!IS_ERR_OR_NULL(skb))
1003 err = udp_send_skb(skb, fl4);
1004 goto out;
1005 }
1006
1007 lock_sock(sk);
1008 if (unlikely(up->pending)) {
1009 /* The socket is already corked while preparing it. */
1010 /* ... which is an evident application bug. --ANK */
1011 release_sock(sk);
1012
1013 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("cork app bug 2\n"));
1014 err = -EINVAL;
1015 goto out;
1016 }
1017 /*
1018 * Now cork the socket to pend data.
1019 */
1020 fl4 = &inet->cork.fl.u.ip4;
1021 fl4->daddr = daddr;
1022 fl4->saddr = saddr;
1023 fl4->fl4_dport = dport;
1024 fl4->fl4_sport = inet->inet_sport;
1025 up->pending = AF_INET;
1026
1027 do_append_data:
1028 up->len += ulen;
1029 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen,
1030 sizeof(struct udphdr), &ipc, &rt,
1031 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1032 if (err)
1033 udp_flush_pending_frames(sk);
1034 else if (!corkreq)
1035 err = udp_push_pending_frames(sk);
1036 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1037 up->pending = 0;
1038 release_sock(sk);
1039
1040 out:
1041 ip_rt_put(rt);
1042 if (free)
1043 kfree(ipc.opt);
1044 if (!err)
1045 return len;
1046 /*
1047 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1048 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1049 * we don't have a good statistic (IpOutDiscards but it can be too many
1050 * things). We could add another new stat but at least for now that
1051 * seems like overkill.
1052 */
1053 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1054 UDP_INC_STATS_USER(sock_net(sk),
1055 UDP_MIB_SNDBUFERRORS, is_udplite);
1056 }
1057 return err;
1058
1059 do_confirm:
1060 dst_confirm(&rt->dst);
1061 if (!(msg->msg_flags&MSG_PROBE) || len)
1062 goto back_from_confirm;
1063 err = 0;
1064 goto out;
1065 }
1066 EXPORT_SYMBOL(udp_sendmsg);
1067
1068 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1069 size_t size, int flags)
1070 {
1071 struct inet_sock *inet = inet_sk(sk);
1072 struct udp_sock *up = udp_sk(sk);
1073 int ret;
1074
1075 if (!up->pending) {
1076 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1077
1078 /* Call udp_sendmsg to specify destination address which
1079 * sendpage interface can't pass.
1080 * This will succeed only when the socket is connected.
1081 */
1082 ret = udp_sendmsg(NULL, sk, &msg, 0);
1083 if (ret < 0)
1084 return ret;
1085 }
1086
1087 lock_sock(sk);
1088
1089 if (unlikely(!up->pending)) {
1090 release_sock(sk);
1091
1092 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("udp cork app bug 3\n"));
1093 return -EINVAL;
1094 }
1095
1096 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1097 page, offset, size, flags);
1098 if (ret == -EOPNOTSUPP) {
1099 release_sock(sk);
1100 return sock_no_sendpage(sk->sk_socket, page, offset,
1101 size, flags);
1102 }
1103 if (ret < 0) {
1104 udp_flush_pending_frames(sk);
1105 goto out;
1106 }
1107
1108 up->len += size;
1109 if (!(up->corkflag || (flags&MSG_MORE)))
1110 ret = udp_push_pending_frames(sk);
1111 if (!ret)
1112 ret = size;
1113 out:
1114 release_sock(sk);
1115 return ret;
1116 }
1117
1118
1119 /**
1120 * first_packet_length - return length of first packet in receive queue
1121 * @sk: socket
1122 *
1123 * Drops all bad checksum frames, until a valid one is found.
1124 * Returns the length of found skb, or 0 if none is found.
1125 */
1126 static unsigned int first_packet_length(struct sock *sk)
1127 {
1128 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1129 struct sk_buff *skb;
1130 unsigned int res;
1131
1132 __skb_queue_head_init(&list_kill);
1133
1134 spin_lock_bh(&rcvq->lock);
1135 while ((skb = skb_peek(rcvq)) != NULL &&
1136 udp_lib_checksum_complete(skb)) {
1137 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS,
1138 IS_UDPLITE(sk));
1139 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1140 IS_UDPLITE(sk));
1141 atomic_inc(&sk->sk_drops);
1142 __skb_unlink(skb, rcvq);
1143 __skb_queue_tail(&list_kill, skb);
1144 }
1145 res = skb ? skb->len : 0;
1146 spin_unlock_bh(&rcvq->lock);
1147
1148 if (!skb_queue_empty(&list_kill)) {
1149 bool slow = lock_sock_fast(sk);
1150
1151 __skb_queue_purge(&list_kill);
1152 sk_mem_reclaim_partial(sk);
1153 unlock_sock_fast(sk, slow);
1154 }
1155 return res;
1156 }
1157
1158 /*
1159 * IOCTL requests applicable to the UDP protocol
1160 */
1161
1162 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1163 {
1164 switch (cmd) {
1165 case SIOCOUTQ:
1166 {
1167 int amount = sk_wmem_alloc_get(sk);
1168
1169 return put_user(amount, (int __user *)arg);
1170 }
1171
1172 case SIOCINQ:
1173 {
1174 unsigned int amount = first_packet_length(sk);
1175
1176 if (amount)
1177 /*
1178 * We will only return the amount
1179 * of this packet since that is all
1180 * that will be read.
1181 */
1182 amount -= sizeof(struct udphdr);
1183
1184 return put_user(amount, (int __user *)arg);
1185 }
1186
1187 default:
1188 return -ENOIOCTLCMD;
1189 }
1190
1191 return 0;
1192 }
1193 EXPORT_SYMBOL(udp_ioctl);
1194
1195 /*
1196 * This should be easy, if there is something there we
1197 * return it, otherwise we block.
1198 */
1199
1200 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1201 size_t len, int noblock, int flags, int *addr_len)
1202 {
1203 struct inet_sock *inet = inet_sk(sk);
1204 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1205 struct sk_buff *skb;
1206 unsigned int ulen, copied;
1207 int peeked, off = 0;
1208 int err;
1209 int is_udplite = IS_UDPLITE(sk);
1210 bool slow;
1211
1212 /*
1213 * Check any passed addresses
1214 */
1215 if (addr_len)
1216 *addr_len = sizeof(*sin);
1217
1218 if (flags & MSG_ERRQUEUE)
1219 return ip_recv_error(sk, msg, len);
1220
1221 try_again:
1222 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1223 &peeked, &off, &err);
1224 if (!skb)
1225 goto out;
1226
1227 ulen = skb->len - sizeof(struct udphdr);
1228 copied = len;
1229 if (copied > ulen)
1230 copied = ulen;
1231 else if (copied < ulen)
1232 msg->msg_flags |= MSG_TRUNC;
1233
1234 /*
1235 * If checksum is needed at all, try to do it while copying the
1236 * data. If the data is truncated, or if we only want a partial
1237 * coverage checksum (UDP-Lite), do it before the copy.
1238 */
1239
1240 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1241 if (udp_lib_checksum_complete(skb))
1242 goto csum_copy_err;
1243 }
1244
1245 if (skb_csum_unnecessary(skb))
1246 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1247 msg->msg_iov, copied);
1248 else {
1249 err = skb_copy_and_csum_datagram_iovec(skb,
1250 sizeof(struct udphdr),
1251 msg->msg_iov);
1252
1253 if (err == -EINVAL)
1254 goto csum_copy_err;
1255 }
1256
1257 if (unlikely(err)) {
1258 trace_kfree_skb(skb, udp_recvmsg);
1259 if (!peeked) {
1260 atomic_inc(&sk->sk_drops);
1261 UDP_INC_STATS_USER(sock_net(sk),
1262 UDP_MIB_INERRORS, is_udplite);
1263 }
1264 goto out_free;
1265 }
1266
1267 if (!peeked)
1268 UDP_INC_STATS_USER(sock_net(sk),
1269 UDP_MIB_INDATAGRAMS, is_udplite);
1270
1271 sock_recv_ts_and_drops(msg, sk, skb);
1272
1273 /* Copy the address. */
1274 if (sin) {
1275 sin->sin_family = AF_INET;
1276 sin->sin_port = udp_hdr(skb)->source;
1277 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1278 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1279 }
1280 if (inet->cmsg_flags)
1281 ip_cmsg_recv(msg, skb);
1282
1283 err = copied;
1284 if (flags & MSG_TRUNC)
1285 err = ulen;
1286
1287 out_free:
1288 skb_free_datagram_locked(sk, skb);
1289 out:
1290 return err;
1291
1292 csum_copy_err:
1293 slow = lock_sock_fast(sk);
1294 if (!skb_kill_datagram(sk, skb, flags)) {
1295 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1296 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1297 }
1298 unlock_sock_fast(sk, slow);
1299
1300 if (noblock)
1301 return -EAGAIN;
1302
1303 /* starting over for a new packet */
1304 msg->msg_flags &= ~MSG_TRUNC;
1305 goto try_again;
1306 }
1307
1308
1309 int udp_disconnect(struct sock *sk, int flags)
1310 {
1311 struct inet_sock *inet = inet_sk(sk);
1312 /*
1313 * 1003.1g - break association.
1314 */
1315
1316 sk->sk_state = TCP_CLOSE;
1317 inet->inet_daddr = 0;
1318 inet->inet_dport = 0;
1319 sock_rps_reset_rxhash(sk);
1320 sk->sk_bound_dev_if = 0;
1321 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1322 inet_reset_saddr(sk);
1323
1324 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1325 sk->sk_prot->unhash(sk);
1326 inet->inet_sport = 0;
1327 }
1328 sk_dst_reset(sk);
1329 return 0;
1330 }
1331 EXPORT_SYMBOL(udp_disconnect);
1332
1333 void udp_lib_unhash(struct sock *sk)
1334 {
1335 if (sk_hashed(sk)) {
1336 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1337 struct udp_hslot *hslot, *hslot2;
1338
1339 hslot = udp_hashslot(udptable, sock_net(sk),
1340 udp_sk(sk)->udp_port_hash);
1341 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1342
1343 spin_lock_bh(&hslot->lock);
1344 if (sk_nulls_del_node_init_rcu(sk)) {
1345 hslot->count--;
1346 inet_sk(sk)->inet_num = 0;
1347 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1348
1349 spin_lock(&hslot2->lock);
1350 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1351 hslot2->count--;
1352 spin_unlock(&hslot2->lock);
1353 }
1354 spin_unlock_bh(&hslot->lock);
1355 }
1356 }
1357 EXPORT_SYMBOL(udp_lib_unhash);
1358
1359 /*
1360 * inet_rcv_saddr was changed, we must rehash secondary hash
1361 */
1362 void udp_lib_rehash(struct sock *sk, u16 newhash)
1363 {
1364 if (sk_hashed(sk)) {
1365 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1366 struct udp_hslot *hslot, *hslot2, *nhslot2;
1367
1368 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1369 nhslot2 = udp_hashslot2(udptable, newhash);
1370 udp_sk(sk)->udp_portaddr_hash = newhash;
1371 if (hslot2 != nhslot2) {
1372 hslot = udp_hashslot(udptable, sock_net(sk),
1373 udp_sk(sk)->udp_port_hash);
1374 /* we must lock primary chain too */
1375 spin_lock_bh(&hslot->lock);
1376
1377 spin_lock(&hslot2->lock);
1378 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1379 hslot2->count--;
1380 spin_unlock(&hslot2->lock);
1381
1382 spin_lock(&nhslot2->lock);
1383 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1384 &nhslot2->head);
1385 nhslot2->count++;
1386 spin_unlock(&nhslot2->lock);
1387
1388 spin_unlock_bh(&hslot->lock);
1389 }
1390 }
1391 }
1392 EXPORT_SYMBOL(udp_lib_rehash);
1393
1394 static void udp_v4_rehash(struct sock *sk)
1395 {
1396 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1397 inet_sk(sk)->inet_rcv_saddr,
1398 inet_sk(sk)->inet_num);
1399 udp_lib_rehash(sk, new_hash);
1400 }
1401
1402 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1403 {
1404 int rc;
1405
1406 if (inet_sk(sk)->inet_daddr)
1407 sock_rps_save_rxhash(sk, skb);
1408
1409 rc = sock_queue_rcv_skb(sk, skb);
1410 if (rc < 0) {
1411 int is_udplite = IS_UDPLITE(sk);
1412
1413 /* Note that an ENOMEM error is charged twice */
1414 if (rc == -ENOMEM)
1415 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1416 is_udplite);
1417 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1418 kfree_skb(skb);
1419 trace_udp_fail_queue_rcv_skb(rc, sk);
1420 return -1;
1421 }
1422
1423 return 0;
1424
1425 }
1426
1427 static struct static_key udp_encap_needed __read_mostly;
1428 void udp_encap_enable(void)
1429 {
1430 if (!static_key_enabled(&udp_encap_needed))
1431 static_key_slow_inc(&udp_encap_needed);
1432 }
1433 EXPORT_SYMBOL(udp_encap_enable);
1434
1435 /* returns:
1436 * -1: error
1437 * 0: success
1438 * >0: "udp encap" protocol resubmission
1439 *
1440 * Note that in the success and error cases, the skb is assumed to
1441 * have either been requeued or freed.
1442 */
1443 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1444 {
1445 struct udp_sock *up = udp_sk(sk);
1446 int rc;
1447 int is_udplite = IS_UDPLITE(sk);
1448
1449 /*
1450 * Charge it to the socket, dropping if the queue is full.
1451 */
1452 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1453 goto drop;
1454 nf_reset(skb);
1455
1456 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1457 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1458
1459 /*
1460 * This is an encapsulation socket so pass the skb to
1461 * the socket's udp_encap_rcv() hook. Otherwise, just
1462 * fall through and pass this up the UDP socket.
1463 * up->encap_rcv() returns the following value:
1464 * =0 if skb was successfully passed to the encap
1465 * handler or was discarded by it.
1466 * >0 if skb should be passed on to UDP.
1467 * <0 if skb should be resubmitted as proto -N
1468 */
1469
1470 /* if we're overly short, let UDP handle it */
1471 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1472 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
1473 int ret;
1474
1475 ret = encap_rcv(sk, skb);
1476 if (ret <= 0) {
1477 UDP_INC_STATS_BH(sock_net(sk),
1478 UDP_MIB_INDATAGRAMS,
1479 is_udplite);
1480 return -ret;
1481 }
1482 }
1483
1484 /* FALLTHROUGH -- it's a UDP Packet */
1485 }
1486
1487 /*
1488 * UDP-Lite specific tests, ignored on UDP sockets
1489 */
1490 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1491
1492 /*
1493 * MIB statistics other than incrementing the error count are
1494 * disabled for the following two types of errors: these depend
1495 * on the application settings, not on the functioning of the
1496 * protocol stack as such.
1497 *
1498 * RFC 3828 here recommends (sec 3.3): "There should also be a
1499 * way ... to ... at least let the receiving application block
1500 * delivery of packets with coverage values less than a value
1501 * provided by the application."
1502 */
1503 if (up->pcrlen == 0) { /* full coverage was set */
1504 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: partial coverage %d while full coverage %d requested\n",
1505 UDP_SKB_CB(skb)->cscov, skb->len);
1506 goto drop;
1507 }
1508 /* The next case involves violating the min. coverage requested
1509 * by the receiver. This is subtle: if receiver wants x and x is
1510 * greater than the buffersize/MTU then receiver will complain
1511 * that it wants x while sender emits packets of smaller size y.
1512 * Therefore the above ...()->partial_cov statement is essential.
1513 */
1514 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1515 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: coverage %d too small, need min %d\n",
1516 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1517 goto drop;
1518 }
1519 }
1520
1521 if (rcu_access_pointer(sk->sk_filter) &&
1522 udp_lib_checksum_complete(skb))
1523 goto csum_error;
1524
1525
1526 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
1527 goto drop;
1528
1529 rc = 0;
1530
1531 ipv4_pktinfo_prepare(skb);
1532 bh_lock_sock(sk);
1533 if (!sock_owned_by_user(sk))
1534 rc = __udp_queue_rcv_skb(sk, skb);
1535 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1536 bh_unlock_sock(sk);
1537 goto drop;
1538 }
1539 bh_unlock_sock(sk);
1540
1541 return rc;
1542
1543 csum_error:
1544 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1545 drop:
1546 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1547 atomic_inc(&sk->sk_drops);
1548 kfree_skb(skb);
1549 return -1;
1550 }
1551
1552
1553 static void flush_stack(struct sock **stack, unsigned int count,
1554 struct sk_buff *skb, unsigned int final)
1555 {
1556 unsigned int i;
1557 struct sk_buff *skb1 = NULL;
1558 struct sock *sk;
1559
1560 for (i = 0; i < count; i++) {
1561 sk = stack[i];
1562 if (likely(skb1 == NULL))
1563 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1564
1565 if (!skb1) {
1566 atomic_inc(&sk->sk_drops);
1567 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1568 IS_UDPLITE(sk));
1569 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1570 IS_UDPLITE(sk));
1571 }
1572
1573 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1574 skb1 = NULL;
1575 }
1576 if (unlikely(skb1))
1577 kfree_skb(skb1);
1578 }
1579
1580 /*
1581 * Multicasts and broadcasts go to each listener.
1582 *
1583 * Note: called only from the BH handler context.
1584 */
1585 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1586 struct udphdr *uh,
1587 __be32 saddr, __be32 daddr,
1588 struct udp_table *udptable)
1589 {
1590 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1591 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1592 int dif;
1593 unsigned int i, count = 0;
1594
1595 spin_lock(&hslot->lock);
1596 sk = sk_nulls_head(&hslot->head);
1597 dif = skb->dev->ifindex;
1598 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1599 while (sk) {
1600 stack[count++] = sk;
1601 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1602 daddr, uh->source, saddr, dif);
1603 if (unlikely(count == ARRAY_SIZE(stack))) {
1604 if (!sk)
1605 break;
1606 flush_stack(stack, count, skb, ~0);
1607 count = 0;
1608 }
1609 }
1610 /*
1611 * before releasing chain lock, we must take a reference on sockets
1612 */
1613 for (i = 0; i < count; i++)
1614 sock_hold(stack[i]);
1615
1616 spin_unlock(&hslot->lock);
1617
1618 /*
1619 * do the slow work with no lock held
1620 */
1621 if (count) {
1622 flush_stack(stack, count, skb, count - 1);
1623
1624 for (i = 0; i < count; i++)
1625 sock_put(stack[i]);
1626 } else {
1627 kfree_skb(skb);
1628 }
1629 return 0;
1630 }
1631
1632 /* Initialize UDP checksum. If exited with zero value (success),
1633 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1634 * Otherwise, csum completion requires chacksumming packet body,
1635 * including udp header and folding it to skb->csum.
1636 */
1637 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1638 int proto)
1639 {
1640 const struct iphdr *iph;
1641 int err;
1642
1643 UDP_SKB_CB(skb)->partial_cov = 0;
1644 UDP_SKB_CB(skb)->cscov = skb->len;
1645
1646 if (proto == IPPROTO_UDPLITE) {
1647 err = udplite_checksum_init(skb, uh);
1648 if (err)
1649 return err;
1650 }
1651
1652 iph = ip_hdr(skb);
1653 if (uh->check == 0) {
1654 skb->ip_summed = CHECKSUM_UNNECESSARY;
1655 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1656 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1657 proto, skb->csum))
1658 skb->ip_summed = CHECKSUM_UNNECESSARY;
1659 }
1660 if (!skb_csum_unnecessary(skb))
1661 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1662 skb->len, proto, 0);
1663 /* Probably, we should checksum udp header (it should be in cache
1664 * in any case) and data in tiny packets (< rx copybreak).
1665 */
1666
1667 return 0;
1668 }
1669
1670 /*
1671 * All we need to do is get the socket, and then do a checksum.
1672 */
1673
1674 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1675 int proto)
1676 {
1677 struct sock *sk;
1678 struct udphdr *uh;
1679 unsigned short ulen;
1680 struct rtable *rt = skb_rtable(skb);
1681 __be32 saddr, daddr;
1682 struct net *net = dev_net(skb->dev);
1683
1684 /*
1685 * Validate the packet.
1686 */
1687 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1688 goto drop; /* No space for header. */
1689
1690 uh = udp_hdr(skb);
1691 ulen = ntohs(uh->len);
1692 saddr = ip_hdr(skb)->saddr;
1693 daddr = ip_hdr(skb)->daddr;
1694
1695 if (ulen > skb->len)
1696 goto short_packet;
1697
1698 if (proto == IPPROTO_UDP) {
1699 /* UDP validates ulen. */
1700 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1701 goto short_packet;
1702 uh = udp_hdr(skb);
1703 }
1704
1705 if (udp4_csum_init(skb, uh, proto))
1706 goto csum_error;
1707
1708 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1709 return __udp4_lib_mcast_deliver(net, skb, uh,
1710 saddr, daddr, udptable);
1711
1712 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1713
1714 if (sk != NULL) {
1715 int ret;
1716
1717 sk_mark_napi_id(sk, skb);
1718 ret = udp_queue_rcv_skb(sk, skb);
1719 sock_put(sk);
1720
1721 /* a return value > 0 means to resubmit the input, but
1722 * it wants the return to be -protocol, or 0
1723 */
1724 if (ret > 0)
1725 return -ret;
1726 return 0;
1727 }
1728
1729 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1730 goto drop;
1731 nf_reset(skb);
1732
1733 /* No socket. Drop packet silently, if checksum is wrong */
1734 if (udp_lib_checksum_complete(skb))
1735 goto csum_error;
1736
1737 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1738 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1739
1740 /*
1741 * Hmm. We got an UDP packet to a port to which we
1742 * don't wanna listen. Ignore it.
1743 */
1744 kfree_skb(skb);
1745 return 0;
1746
1747 short_packet:
1748 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1749 proto == IPPROTO_UDPLITE ? "Lite" : "",
1750 &saddr, ntohs(uh->source),
1751 ulen, skb->len,
1752 &daddr, ntohs(uh->dest));
1753 goto drop;
1754
1755 csum_error:
1756 /*
1757 * RFC1122: OK. Discards the bad packet silently (as far as
1758 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1759 */
1760 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1761 proto == IPPROTO_UDPLITE ? "Lite" : "",
1762 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1763 ulen);
1764 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1765 drop:
1766 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1767 kfree_skb(skb);
1768 return 0;
1769 }
1770
1771 int udp_rcv(struct sk_buff *skb)
1772 {
1773 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1774 }
1775
1776 void udp_destroy_sock(struct sock *sk)
1777 {
1778 struct udp_sock *up = udp_sk(sk);
1779 bool slow = lock_sock_fast(sk);
1780 udp_flush_pending_frames(sk);
1781 unlock_sock_fast(sk, slow);
1782 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1783 void (*encap_destroy)(struct sock *sk);
1784 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1785 if (encap_destroy)
1786 encap_destroy(sk);
1787 }
1788 }
1789
1790 /*
1791 * Socket option code for UDP
1792 */
1793 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1794 char __user *optval, unsigned int optlen,
1795 int (*push_pending_frames)(struct sock *))
1796 {
1797 struct udp_sock *up = udp_sk(sk);
1798 int val;
1799 int err = 0;
1800 int is_udplite = IS_UDPLITE(sk);
1801
1802 if (optlen < sizeof(int))
1803 return -EINVAL;
1804
1805 if (get_user(val, (int __user *)optval))
1806 return -EFAULT;
1807
1808 switch (optname) {
1809 case UDP_CORK:
1810 if (val != 0) {
1811 up->corkflag = 1;
1812 } else {
1813 up->corkflag = 0;
1814 lock_sock(sk);
1815 (*push_pending_frames)(sk);
1816 release_sock(sk);
1817 }
1818 break;
1819
1820 case UDP_ENCAP:
1821 switch (val) {
1822 case 0:
1823 case UDP_ENCAP_ESPINUDP:
1824 case UDP_ENCAP_ESPINUDP_NON_IKE:
1825 up->encap_rcv = xfrm4_udp_encap_rcv;
1826 /* FALLTHROUGH */
1827 case UDP_ENCAP_L2TPINUDP:
1828 up->encap_type = val;
1829 udp_encap_enable();
1830 break;
1831 default:
1832 err = -ENOPROTOOPT;
1833 break;
1834 }
1835 break;
1836
1837 /*
1838 * UDP-Lite's partial checksum coverage (RFC 3828).
1839 */
1840 /* The sender sets actual checksum coverage length via this option.
1841 * The case coverage > packet length is handled by send module. */
1842 case UDPLITE_SEND_CSCOV:
1843 if (!is_udplite) /* Disable the option on UDP sockets */
1844 return -ENOPROTOOPT;
1845 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1846 val = 8;
1847 else if (val > USHRT_MAX)
1848 val = USHRT_MAX;
1849 up->pcslen = val;
1850 up->pcflag |= UDPLITE_SEND_CC;
1851 break;
1852
1853 /* The receiver specifies a minimum checksum coverage value. To make
1854 * sense, this should be set to at least 8 (as done below). If zero is
1855 * used, this again means full checksum coverage. */
1856 case UDPLITE_RECV_CSCOV:
1857 if (!is_udplite) /* Disable the option on UDP sockets */
1858 return -ENOPROTOOPT;
1859 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1860 val = 8;
1861 else if (val > USHRT_MAX)
1862 val = USHRT_MAX;
1863 up->pcrlen = val;
1864 up->pcflag |= UDPLITE_RECV_CC;
1865 break;
1866
1867 default:
1868 err = -ENOPROTOOPT;
1869 break;
1870 }
1871
1872 return err;
1873 }
1874 EXPORT_SYMBOL(udp_lib_setsockopt);
1875
1876 int udp_setsockopt(struct sock *sk, int level, int optname,
1877 char __user *optval, unsigned int optlen)
1878 {
1879 if (level == SOL_UDP || level == SOL_UDPLITE)
1880 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1881 udp_push_pending_frames);
1882 return ip_setsockopt(sk, level, optname, optval, optlen);
1883 }
1884
1885 #ifdef CONFIG_COMPAT
1886 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1887 char __user *optval, unsigned int optlen)
1888 {
1889 if (level == SOL_UDP || level == SOL_UDPLITE)
1890 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1891 udp_push_pending_frames);
1892 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1893 }
1894 #endif
1895
1896 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1897 char __user *optval, int __user *optlen)
1898 {
1899 struct udp_sock *up = udp_sk(sk);
1900 int val, len;
1901
1902 if (get_user(len, optlen))
1903 return -EFAULT;
1904
1905 len = min_t(unsigned int, len, sizeof(int));
1906
1907 if (len < 0)
1908 return -EINVAL;
1909
1910 switch (optname) {
1911 case UDP_CORK:
1912 val = up->corkflag;
1913 break;
1914
1915 case UDP_ENCAP:
1916 val = up->encap_type;
1917 break;
1918
1919 /* The following two cannot be changed on UDP sockets, the return is
1920 * always 0 (which corresponds to the full checksum coverage of UDP). */
1921 case UDPLITE_SEND_CSCOV:
1922 val = up->pcslen;
1923 break;
1924
1925 case UDPLITE_RECV_CSCOV:
1926 val = up->pcrlen;
1927 break;
1928
1929 default:
1930 return -ENOPROTOOPT;
1931 }
1932
1933 if (put_user(len, optlen))
1934 return -EFAULT;
1935 if (copy_to_user(optval, &val, len))
1936 return -EFAULT;
1937 return 0;
1938 }
1939 EXPORT_SYMBOL(udp_lib_getsockopt);
1940
1941 int udp_getsockopt(struct sock *sk, int level, int optname,
1942 char __user *optval, int __user *optlen)
1943 {
1944 if (level == SOL_UDP || level == SOL_UDPLITE)
1945 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1946 return ip_getsockopt(sk, level, optname, optval, optlen);
1947 }
1948
1949 #ifdef CONFIG_COMPAT
1950 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1951 char __user *optval, int __user *optlen)
1952 {
1953 if (level == SOL_UDP || level == SOL_UDPLITE)
1954 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1955 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1956 }
1957 #endif
1958 /**
1959 * udp_poll - wait for a UDP event.
1960 * @file - file struct
1961 * @sock - socket
1962 * @wait - poll table
1963 *
1964 * This is same as datagram poll, except for the special case of
1965 * blocking sockets. If application is using a blocking fd
1966 * and a packet with checksum error is in the queue;
1967 * then it could get return from select indicating data available
1968 * but then block when reading it. Add special case code
1969 * to work around these arguably broken applications.
1970 */
1971 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1972 {
1973 unsigned int mask = datagram_poll(file, sock, wait);
1974 struct sock *sk = sock->sk;
1975
1976 sock_rps_record_flow(sk);
1977
1978 /* Check for false positives due to checksum errors */
1979 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1980 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1981 mask &= ~(POLLIN | POLLRDNORM);
1982
1983 return mask;
1984
1985 }
1986 EXPORT_SYMBOL(udp_poll);
1987
1988 struct proto udp_prot = {
1989 .name = "UDP",
1990 .owner = THIS_MODULE,
1991 .close = udp_lib_close,
1992 .connect = ip4_datagram_connect,
1993 .disconnect = udp_disconnect,
1994 .ioctl = udp_ioctl,
1995 .destroy = udp_destroy_sock,
1996 .setsockopt = udp_setsockopt,
1997 .getsockopt = udp_getsockopt,
1998 .sendmsg = udp_sendmsg,
1999 .recvmsg = udp_recvmsg,
2000 .sendpage = udp_sendpage,
2001 .backlog_rcv = __udp_queue_rcv_skb,
2002 .release_cb = ip4_datagram_release_cb,
2003 .hash = udp_lib_hash,
2004 .unhash = udp_lib_unhash,
2005 .rehash = udp_v4_rehash,
2006 .get_port = udp_v4_get_port,
2007 .memory_allocated = &udp_memory_allocated,
2008 .sysctl_mem = sysctl_udp_mem,
2009 .sysctl_wmem = &sysctl_udp_wmem_min,
2010 .sysctl_rmem = &sysctl_udp_rmem_min,
2011 .obj_size = sizeof(struct udp_sock),
2012 .slab_flags = SLAB_DESTROY_BY_RCU,
2013 .h.udp_table = &udp_table,
2014 #ifdef CONFIG_COMPAT
2015 .compat_setsockopt = compat_udp_setsockopt,
2016 .compat_getsockopt = compat_udp_getsockopt,
2017 #endif
2018 .clear_sk = sk_prot_clear_portaddr_nulls,
2019 };
2020 EXPORT_SYMBOL(udp_prot);
2021
2022 /* ------------------------------------------------------------------------ */
2023 #ifdef CONFIG_PROC_FS
2024
2025 static struct sock *udp_get_first(struct seq_file *seq, int start)
2026 {
2027 struct sock *sk;
2028 struct udp_iter_state *state = seq->private;
2029 struct net *net = seq_file_net(seq);
2030
2031 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2032 ++state->bucket) {
2033 struct hlist_nulls_node *node;
2034 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2035
2036 if (hlist_nulls_empty(&hslot->head))
2037 continue;
2038
2039 spin_lock_bh(&hslot->lock);
2040 sk_nulls_for_each(sk, node, &hslot->head) {
2041 if (!net_eq(sock_net(sk), net))
2042 continue;
2043 if (sk->sk_family == state->family)
2044 goto found;
2045 }
2046 spin_unlock_bh(&hslot->lock);
2047 }
2048 sk = NULL;
2049 found:
2050 return sk;
2051 }
2052
2053 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2054 {
2055 struct udp_iter_state *state = seq->private;
2056 struct net *net = seq_file_net(seq);
2057
2058 do {
2059 sk = sk_nulls_next(sk);
2060 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2061
2062 if (!sk) {
2063 if (state->bucket <= state->udp_table->mask)
2064 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2065 return udp_get_first(seq, state->bucket + 1);
2066 }
2067 return sk;
2068 }
2069
2070 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2071 {
2072 struct sock *sk = udp_get_first(seq, 0);
2073
2074 if (sk)
2075 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2076 --pos;
2077 return pos ? NULL : sk;
2078 }
2079
2080 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2081 {
2082 struct udp_iter_state *state = seq->private;
2083 state->bucket = MAX_UDP_PORTS;
2084
2085 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2086 }
2087
2088 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2089 {
2090 struct sock *sk;
2091
2092 if (v == SEQ_START_TOKEN)
2093 sk = udp_get_idx(seq, 0);
2094 else
2095 sk = udp_get_next(seq, v);
2096
2097 ++*pos;
2098 return sk;
2099 }
2100
2101 static void udp_seq_stop(struct seq_file *seq, void *v)
2102 {
2103 struct udp_iter_state *state = seq->private;
2104
2105 if (state->bucket <= state->udp_table->mask)
2106 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2107 }
2108
2109 int udp_seq_open(struct inode *inode, struct file *file)
2110 {
2111 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2112 struct udp_iter_state *s;
2113 int err;
2114
2115 err = seq_open_net(inode, file, &afinfo->seq_ops,
2116 sizeof(struct udp_iter_state));
2117 if (err < 0)
2118 return err;
2119
2120 s = ((struct seq_file *)file->private_data)->private;
2121 s->family = afinfo->family;
2122 s->udp_table = afinfo->udp_table;
2123 return err;
2124 }
2125 EXPORT_SYMBOL(udp_seq_open);
2126
2127 /* ------------------------------------------------------------------------ */
2128 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2129 {
2130 struct proc_dir_entry *p;
2131 int rc = 0;
2132
2133 afinfo->seq_ops.start = udp_seq_start;
2134 afinfo->seq_ops.next = udp_seq_next;
2135 afinfo->seq_ops.stop = udp_seq_stop;
2136
2137 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2138 afinfo->seq_fops, afinfo);
2139 if (!p)
2140 rc = -ENOMEM;
2141 return rc;
2142 }
2143 EXPORT_SYMBOL(udp_proc_register);
2144
2145 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2146 {
2147 remove_proc_entry(afinfo->name, net->proc_net);
2148 }
2149 EXPORT_SYMBOL(udp_proc_unregister);
2150
2151 /* ------------------------------------------------------------------------ */
2152 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2153 int bucket, int *len)
2154 {
2155 struct inet_sock *inet = inet_sk(sp);
2156 __be32 dest = inet->inet_daddr;
2157 __be32 src = inet->inet_rcv_saddr;
2158 __u16 destp = ntohs(inet->inet_dport);
2159 __u16 srcp = ntohs(inet->inet_sport);
2160
2161 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2162 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d%n",
2163 bucket, src, srcp, dest, destp, sp->sk_state,
2164 sk_wmem_alloc_get(sp),
2165 sk_rmem_alloc_get(sp),
2166 0, 0L, 0,
2167 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2168 0, sock_i_ino(sp),
2169 atomic_read(&sp->sk_refcnt), sp,
2170 atomic_read(&sp->sk_drops), len);
2171 }
2172
2173 int udp4_seq_show(struct seq_file *seq, void *v)
2174 {
2175 if (v == SEQ_START_TOKEN)
2176 seq_printf(seq, "%-127s\n",
2177 " sl local_address rem_address st tx_queue "
2178 "rx_queue tr tm->when retrnsmt uid timeout "
2179 "inode ref pointer drops");
2180 else {
2181 struct udp_iter_state *state = seq->private;
2182 int len;
2183
2184 udp4_format_sock(v, seq, state->bucket, &len);
2185 seq_printf(seq, "%*s\n", 127 - len, "");
2186 }
2187 return 0;
2188 }
2189
2190 static const struct file_operations udp_afinfo_seq_fops = {
2191 .owner = THIS_MODULE,
2192 .open = udp_seq_open,
2193 .read = seq_read,
2194 .llseek = seq_lseek,
2195 .release = seq_release_net
2196 };
2197
2198 /* ------------------------------------------------------------------------ */
2199 static struct udp_seq_afinfo udp4_seq_afinfo = {
2200 .name = "udp",
2201 .family = AF_INET,
2202 .udp_table = &udp_table,
2203 .seq_fops = &udp_afinfo_seq_fops,
2204 .seq_ops = {
2205 .show = udp4_seq_show,
2206 },
2207 };
2208
2209 static int __net_init udp4_proc_init_net(struct net *net)
2210 {
2211 return udp_proc_register(net, &udp4_seq_afinfo);
2212 }
2213
2214 static void __net_exit udp4_proc_exit_net(struct net *net)
2215 {
2216 udp_proc_unregister(net, &udp4_seq_afinfo);
2217 }
2218
2219 static struct pernet_operations udp4_net_ops = {
2220 .init = udp4_proc_init_net,
2221 .exit = udp4_proc_exit_net,
2222 };
2223
2224 int __init udp4_proc_init(void)
2225 {
2226 return register_pernet_subsys(&udp4_net_ops);
2227 }
2228
2229 void udp4_proc_exit(void)
2230 {
2231 unregister_pernet_subsys(&udp4_net_ops);
2232 }
2233 #endif /* CONFIG_PROC_FS */
2234
2235 static __initdata unsigned long uhash_entries;
2236 static int __init set_uhash_entries(char *str)
2237 {
2238 ssize_t ret;
2239
2240 if (!str)
2241 return 0;
2242
2243 ret = kstrtoul(str, 0, &uhash_entries);
2244 if (ret)
2245 return 0;
2246
2247 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2248 uhash_entries = UDP_HTABLE_SIZE_MIN;
2249 return 1;
2250 }
2251 __setup("uhash_entries=", set_uhash_entries);
2252
2253 void __init udp_table_init(struct udp_table *table, const char *name)
2254 {
2255 unsigned int i;
2256
2257 table->hash = alloc_large_system_hash(name,
2258 2 * sizeof(struct udp_hslot),
2259 uhash_entries,
2260 21, /* one slot per 2 MB */
2261 0,
2262 &table->log,
2263 &table->mask,
2264 UDP_HTABLE_SIZE_MIN,
2265 64 * 1024);
2266
2267 table->hash2 = table->hash + (table->mask + 1);
2268 for (i = 0; i <= table->mask; i++) {
2269 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2270 table->hash[i].count = 0;
2271 spin_lock_init(&table->hash[i].lock);
2272 }
2273 for (i = 0; i <= table->mask; i++) {
2274 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2275 table->hash2[i].count = 0;
2276 spin_lock_init(&table->hash2[i].lock);
2277 }
2278 }
2279
2280 void __init udp_init(void)
2281 {
2282 unsigned long limit;
2283
2284 udp_table_init(&udp_table, "UDP");
2285 limit = nr_free_buffer_pages() / 8;
2286 limit = max(limit, 128UL);
2287 sysctl_udp_mem[0] = limit / 4 * 3;
2288 sysctl_udp_mem[1] = limit;
2289 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2290
2291 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2292 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2293 }
2294
2295 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
2296 netdev_features_t features)
2297 {
2298 struct sk_buff *segs = ERR_PTR(-EINVAL);
2299 int mac_len = skb->mac_len;
2300 int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
2301 __be16 protocol = skb->protocol;
2302 netdev_features_t enc_features;
2303 int outer_hlen;
2304
2305 if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
2306 goto out;
2307
2308 skb->encapsulation = 0;
2309 __skb_pull(skb, tnl_hlen);
2310 skb_reset_mac_header(skb);
2311 skb_set_network_header(skb, skb_inner_network_offset(skb));
2312 skb->mac_len = skb_inner_network_offset(skb);
2313 skb->protocol = htons(ETH_P_TEB);
2314
2315 /* segment inner packet. */
2316 enc_features = skb->dev->hw_enc_features & netif_skb_features(skb);
2317 segs = skb_mac_gso_segment(skb, enc_features);
2318 if (!segs || IS_ERR(segs))
2319 goto out;
2320
2321 outer_hlen = skb_tnl_header_len(skb);
2322 skb = segs;
2323 do {
2324 struct udphdr *uh;
2325 int udp_offset = outer_hlen - tnl_hlen;
2326
2327 skb_reset_inner_headers(skb);
2328 skb->encapsulation = 1;
2329
2330 skb->mac_len = mac_len;
2331
2332 skb_push(skb, outer_hlen);
2333 skb_reset_mac_header(skb);
2334 skb_set_network_header(skb, mac_len);
2335 skb_set_transport_header(skb, udp_offset);
2336 uh = udp_hdr(skb);
2337 uh->len = htons(skb->len - udp_offset);
2338
2339 /* csum segment if tunnel sets skb with csum. */
2340 if (protocol == htons(ETH_P_IP) && unlikely(uh->check)) {
2341 struct iphdr *iph = ip_hdr(skb);
2342
2343 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
2344 skb->len - udp_offset,
2345 IPPROTO_UDP, 0);
2346 uh->check = csum_fold(skb_checksum(skb, udp_offset,
2347 skb->len - udp_offset, 0));
2348 if (uh->check == 0)
2349 uh->check = CSUM_MANGLED_0;
2350
2351 } else if (protocol == htons(ETH_P_IPV6)) {
2352 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
2353 u32 len = skb->len - udp_offset;
2354
2355 uh->check = ~csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
2356 len, IPPROTO_UDP, 0);
2357 uh->check = csum_fold(skb_checksum(skb, udp_offset, len, 0));
2358 if (uh->check == 0)
2359 uh->check = CSUM_MANGLED_0;
2360 skb->ip_summed = CHECKSUM_NONE;
2361 }
2362
2363 skb->protocol = protocol;
2364 } while ((skb = skb->next));
2365 out:
2366 return segs;
2367 }
Linus' kernel tree