projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
import PULS_20180308
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / net / ipv4 / udp.c
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 "udp_impl.h"
113
114 struct udp_table udp_table __read_mostly;
115 EXPORT_SYMBOL(udp_table);
116
117 long sysctl_udp_mem[3] __read_mostly;
118 EXPORT_SYMBOL(sysctl_udp_mem);
119
120 int sysctl_udp_rmem_min __read_mostly;
121 EXPORT_SYMBOL(sysctl_udp_rmem_min);
122
123 int sysctl_udp_wmem_min __read_mostly;
124 EXPORT_SYMBOL(sysctl_udp_wmem_min);
125
126 #ifdef UDP_SKT_WIFI
127 #include <linux/kallsyms.h>
128 #include <linux/ftrace_event.h>
129 int sysctl_udp_met_port __read_mostly = -1;
130 EXPORT_SYMBOL(sysctl_udp_met_port);
131 int sysctl_met_is_enable __read_mostly = -1;
132 EXPORT_SYMBOL(sysctl_met_is_enable);
133 #ifdef CONFIG_TRACING
134 unsigned long __read_mostly udp_tracing_mark_write_addr = 0;
135 #endif
136 #endif
137
138 atomic_long_t udp_memory_allocated;
139 EXPORT_SYMBOL(udp_memory_allocated);
140
141 #define MAX_UDP_PORTS 65536
142 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
143
144 static int udp_lib_lport_inuse(struct net *net, __u16 num,
145 const struct udp_hslot *hslot,
146 unsigned long *bitmap,
147 struct sock *sk,
148 int (*saddr_comp)(const struct sock *sk1,
149 const struct sock *sk2),
150 unsigned int log)
151 {
152 struct sock *sk2;
153 struct hlist_nulls_node *node;
154 kuid_t uid = sock_i_uid(sk);
155
156 sk_nulls_for_each(sk2, node, &hslot->head)
157 if (net_eq(sock_net(sk2), net) &&
158 sk2 != sk &&
159 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
160 (!sk2->sk_reuse || !sk->sk_reuse) &&
161 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
162 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
163 (!sk2->sk_reuseport || !sk->sk_reuseport ||
164 !uid_eq(uid, sock_i_uid(sk2))) &&
165 (*saddr_comp)(sk, sk2)) {
166 if (bitmap)
167 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
168 bitmap);
169 else
170 return 1;
171 }
172 return 0;
173 }
174
175 /*
176 * Note: we still hold spinlock of primary hash chain, so no other writer
177 * can insert/delete a socket with local_port == num
178 */
179 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
180 struct udp_hslot *hslot2,
181 struct sock *sk,
182 int (*saddr_comp)(const struct sock *sk1,
183 const struct sock *sk2))
184 {
185 struct sock *sk2;
186 struct hlist_nulls_node *node;
187 kuid_t uid = sock_i_uid(sk);
188 int res = 0;
189
190 spin_lock(&hslot2->lock);
191 udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
192 if (net_eq(sock_net(sk2), net) &&
193 sk2 != sk &&
194 (udp_sk(sk2)->udp_port_hash == num) &&
195 (!sk2->sk_reuse || !sk->sk_reuse) &&
196 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
197 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
198 (!sk2->sk_reuseport || !sk->sk_reuseport ||
199 !uid_eq(uid, sock_i_uid(sk2))) &&
200 (*saddr_comp)(sk, sk2)) {
201 res = 1;
202 break;
203 }
204 spin_unlock(&hslot2->lock);
205 return res;
206 }
207
208 /**
209 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
210 *
211 * @sk: socket struct in question
212 * @snum: port number to look up
213 * @saddr_comp: AF-dependent comparison of bound local IP addresses
214 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
215 * with NULL address
216 */
217 int udp_lib_get_port(struct sock *sk, unsigned short snum,
218 int (*saddr_comp)(const struct sock *sk1,
219 const struct sock *sk2),
220 unsigned int hash2_nulladdr)
221 {
222 struct udp_hslot *hslot, *hslot2;
223 struct udp_table *udptable = sk->sk_prot->h.udp_table;
224 int error = 1;
225 struct net *net = sock_net(sk);
226
227 if (!snum) {
228 int low, high, remaining;
229 unsigned int rand;
230 unsigned short first, last;
231 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
232
233 inet_get_local_port_range(&low, &high);
234 remaining = (high - low) + 1;
235
236 rand = net_random();
237 first = (((u64)rand * remaining) >> 32) + low;
238 /*
239 * force rand to be an odd multiple of UDP_HTABLE_SIZE
240 */
241 rand = (rand | 1) * (udptable->mask + 1);
242 last = first + udptable->mask + 1;
243 do {
244 hslot = udp_hashslot(udptable, net, first);
245 bitmap_zero(bitmap, PORTS_PER_CHAIN);
246 spin_lock_bh(&hslot->lock);
247 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
248 saddr_comp, udptable->log);
249
250 snum = first;
251 /*
252 * Iterate on all possible values of snum for this hash.
253 * Using steps of an odd multiple of UDP_HTABLE_SIZE
254 * give us randomization and full range coverage.
255 */
256 do {
257 if (low <= snum && snum <= high &&
258 !test_bit(snum >> udptable->log, bitmap) &&
259 !inet_is_reserved_local_port(snum))
260 goto found;
261 snum += rand;
262 } while (snum != first);
263 spin_unlock_bh(&hslot->lock);
264 } while (++first != last);
265 goto fail;
266 } else {
267 hslot = udp_hashslot(udptable, net, snum);
268 spin_lock_bh(&hslot->lock);
269 if (hslot->count > 10) {
270 int exist;
271 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
272
273 slot2 &= udptable->mask;
274 hash2_nulladdr &= udptable->mask;
275
276 hslot2 = udp_hashslot2(udptable, slot2);
277 if (hslot->count < hslot2->count)
278 goto scan_primary_hash;
279
280 exist = udp_lib_lport_inuse2(net, snum, hslot2,
281 sk, saddr_comp);
282 if (!exist && (hash2_nulladdr != slot2)) {
283 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
284 exist = udp_lib_lport_inuse2(net, snum, hslot2,
285 sk, saddr_comp);
286 }
287 if (exist)
288 goto fail_unlock;
289 else
290 goto found;
291 }
292 scan_primary_hash:
293 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
294 saddr_comp, 0))
295 goto fail_unlock;
296 }
297 found:
298 inet_sk(sk)->inet_num = snum;
299 udp_sk(sk)->udp_port_hash = snum;
300 udp_sk(sk)->udp_portaddr_hash ^= snum;
301 if (sk_unhashed(sk)) {
302 sk_nulls_add_node_rcu(sk, &hslot->head);
303 hslot->count++;
304 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
305
306 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
307 spin_lock(&hslot2->lock);
308 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
309 &hslot2->head);
310 hslot2->count++;
311 spin_unlock(&hslot2->lock);
312 }
313 error = 0;
314 fail_unlock:
315 spin_unlock_bh(&hslot->lock);
316 fail:
317 return error;
318 }
319 EXPORT_SYMBOL(udp_lib_get_port);
320
321 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
322 {
323 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
324
325 return (!ipv6_only_sock(sk2) &&
326 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
327 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
328 }
329
330 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
331 unsigned int port)
332 {
333 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
334 }
335
336 int udp_v4_get_port(struct sock *sk, unsigned short snum)
337 {
338 unsigned int hash2_nulladdr =
339 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
340 unsigned int hash2_partial =
341 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
342
343 /* precompute partial secondary hash */
344 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
345 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
346 }
347
348 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
349 unsigned short hnum,
350 __be16 sport, __be32 daddr, __be16 dport, int dif)
351 {
352 int score = -1;
353
354 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
355 !ipv6_only_sock(sk)) {
356 struct inet_sock *inet = inet_sk(sk);
357
358 score = (sk->sk_family == PF_INET ? 2 : 1);
359 if (inet->inet_rcv_saddr) {
360 if (inet->inet_rcv_saddr != daddr)
361 return -1;
362 score += 4;
363 }
364 if (inet->inet_daddr) {
365 if (inet->inet_daddr != saddr)
366 return -1;
367 score += 4;
368 }
369 if (inet->inet_dport) {
370 if (inet->inet_dport != sport)
371 return -1;
372 score += 4;
373 }
374 if (sk->sk_bound_dev_if) {
375 if (sk->sk_bound_dev_if != dif)
376 return -1;
377 score += 4;
378 }
379 }
380 return score;
381 }
382
383 /*
384 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
385 */
386 static inline int compute_score2(struct sock *sk, struct net *net,
387 __be32 saddr, __be16 sport,
388 __be32 daddr, unsigned int hnum, int dif)
389 {
390 int score = -1;
391
392 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
393 struct inet_sock *inet = inet_sk(sk);
394
395 if (inet->inet_rcv_saddr != daddr)
396 return -1;
397 if (inet->inet_num != hnum)
398 return -1;
399
400 score = (sk->sk_family == PF_INET ? 2 : 1);
401 if (inet->inet_daddr) {
402 if (inet->inet_daddr != saddr)
403 return -1;
404 score += 4;
405 }
406 if (inet->inet_dport) {
407 if (inet->inet_dport != sport)
408 return -1;
409 score += 4;
410 }
411 if (sk->sk_bound_dev_if) {
412 if (sk->sk_bound_dev_if != dif)
413 return -1;
414 score += 4;
415 }
416 }
417 return score;
418 }
419
420
421 /* called with read_rcu_lock() */
422 static struct sock *udp4_lib_lookup2(struct net *net,
423 __be32 saddr, __be16 sport,
424 __be32 daddr, unsigned int hnum, int dif,
425 struct udp_hslot *hslot2, unsigned int slot2)
426 {
427 struct sock *sk, *result;
428 struct hlist_nulls_node *node;
429 int score, badness, matches = 0, reuseport = 0;
430 u32 hash = 0;
431
432 begin:
433 result = NULL;
434 badness = 0;
435 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
436 score = compute_score2(sk, net, saddr, sport,
437 daddr, hnum, dif);
438 if (score > badness) {
439 result = sk;
440 badness = score;
441 reuseport = sk->sk_reuseport;
442 if (reuseport) {
443 hash = inet_ehashfn(net, daddr, hnum,
444 saddr, htons(sport));
445 matches = 1;
446 }
447 } else if (score == badness && reuseport) {
448 matches++;
449 if (((u64)hash * matches) >> 32 == 0)
450 result = sk;
451 hash = next_pseudo_random32(hash);
452 }
453 }
454 /*
455 * if the nulls value we got at the end of this lookup is
456 * not the expected one, we must restart lookup.
457 * We probably met an item that was moved to another chain.
458 */
459 if (get_nulls_value(node) != slot2)
460 goto begin;
461 if (result) {
462 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
463 result = NULL;
464 else if (unlikely(compute_score2(result, net, saddr, sport,
465 daddr, hnum, dif) < badness)) {
466 sock_put(result);
467 goto begin;
468 }
469 }
470 return result;
471 }
472
473 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
474 * harder than this. -DaveM
475 */
476 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
477 __be16 sport, __be32 daddr, __be16 dport,
478 int dif, struct udp_table *udptable)
479 {
480 struct sock *sk, *result;
481 struct hlist_nulls_node *node;
482 unsigned short hnum = ntohs(dport);
483 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
484 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
485 int score, badness, matches = 0, reuseport = 0;
486 u32 hash = 0;
487
488 rcu_read_lock();
489 if (hslot->count > 10) {
490 hash2 = udp4_portaddr_hash(net, daddr, hnum);
491 slot2 = hash2 & udptable->mask;
492 hslot2 = &udptable->hash2[slot2];
493 if (hslot->count < hslot2->count)
494 goto begin;
495
496 result = udp4_lib_lookup2(net, saddr, sport,
497 daddr, hnum, dif,
498 hslot2, slot2);
499 if (!result) {
500 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
501 slot2 = hash2 & udptable->mask;
502 hslot2 = &udptable->hash2[slot2];
503 if (hslot->count < hslot2->count)
504 goto begin;
505
506 result = udp4_lib_lookup2(net, saddr, sport,
507 htonl(INADDR_ANY), hnum, dif,
508 hslot2, slot2);
509 }
510 rcu_read_unlock();
511 return result;
512 }
513 begin:
514 result = NULL;
515 badness = 0;
516 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
517 score = compute_score(sk, net, saddr, hnum, sport,
518 daddr, dport, dif);
519 if (score > badness) {
520 result = sk;
521 badness = score;
522 reuseport = sk->sk_reuseport;
523 if (reuseport) {
524 hash = inet_ehashfn(net, daddr, hnum,
525 saddr, htons(sport));
526 matches = 1;
527 }
528 } else if (score == badness && reuseport) {
529 matches++;
530 if (((u64)hash * matches) >> 32 == 0)
531 result = sk;
532 hash = next_pseudo_random32(hash);
533 }
534 }
535 /*
536 * if the nulls value we got at the end of this lookup is
537 * not the expected one, we must restart lookup.
538 * We probably met an item that was moved to another chain.
539 */
540 if (get_nulls_value(node) != slot)
541 goto begin;
542
543 if (result) {
544 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
545 result = NULL;
546 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
547 daddr, dport, dif) < badness)) {
548 sock_put(result);
549 goto begin;
550 }
551 }
552 rcu_read_unlock();
553 return result;
554 }
555 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
556
557 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
558 __be16 sport, __be16 dport,
559 struct udp_table *udptable)
560 {
561 struct sock *sk;
562 const struct iphdr *iph = ip_hdr(skb);
563
564 if (unlikely(sk = skb_steal_sock(skb)))
565 return sk;
566 else
567 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
568 iph->daddr, dport, inet_iif(skb),
569 udptable);
570 }
571
572 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
573 __be32 daddr, __be16 dport, int dif)
574 {
575 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
576 }
577 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
578
579 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
580 __be16 loc_port, __be32 loc_addr,
581 __be16 rmt_port, __be32 rmt_addr,
582 int dif)
583 {
584 struct hlist_nulls_node *node;
585 struct sock *s = sk;
586 unsigned short hnum = ntohs(loc_port);
587
588 sk_nulls_for_each_from(s, node) {
589 struct inet_sock *inet = inet_sk(s);
590
591 if (!net_eq(sock_net(s), net) ||
592 udp_sk(s)->udp_port_hash != hnum ||
593 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
594 (inet->inet_dport != rmt_port && inet->inet_dport) ||
595 (inet->inet_rcv_saddr &&
596 inet->inet_rcv_saddr != loc_addr) ||
597 ipv6_only_sock(s) ||
598 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
599 continue;
600 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
601 continue;
602 goto found;
603 }
604 s = NULL;
605 found:
606 return s;
607 }
608
609 /*
610 * This routine is called by the ICMP module when it gets some
611 * sort of error condition. If err < 0 then the socket should
612 * be closed and the error returned to the user. If err > 0
613 * it's just the icmp type << 8 | icmp code.
614 * Header points to the ip header of the error packet. We move
615 * on past this. Then (as it used to claim before adjustment)
616 * header points to the first 8 bytes of the udp header. We need
617 * to find the appropriate port.
618 */
619
620 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
621 {
622 struct inet_sock *inet;
623 const struct iphdr *iph = (const struct iphdr *)skb->data;
624 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
625 const int type = icmp_hdr(skb)->type;
626 const int code = icmp_hdr(skb)->code;
627 struct sock *sk;
628 int harderr;
629 int err;
630 struct net *net = dev_net(skb->dev);
631
632 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
633 iph->saddr, uh->source, skb->dev->ifindex, udptable);
634 if (sk == NULL) {
635 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
636 return; /* No socket for error */
637 }
638
639 err = 0;
640 harderr = 0;
641 inet = inet_sk(sk);
642
643 switch (type) {
644 default:
645 case ICMP_TIME_EXCEEDED:
646 err = EHOSTUNREACH;
647 break;
648 case ICMP_SOURCE_QUENCH:
649 goto out;
650 case ICMP_PARAMETERPROB:
651 err = EPROTO;
652 harderr = 1;
653 break;
654 case ICMP_DEST_UNREACH:
655 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
656 ipv4_sk_update_pmtu(skb, sk, info);
657 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
658 err = EMSGSIZE;
659 harderr = 1;
660 break;
661 }
662 goto out;
663 }
664 err = EHOSTUNREACH;
665 if (code <= NR_ICMP_UNREACH) {
666 harderr = icmp_err_convert[code].fatal;
667 err = icmp_err_convert[code].errno;
668 }
669 break;
670 case ICMP_REDIRECT:
671 ipv4_sk_redirect(skb, sk);
672 break;
673 }
674
675 /*
676 * RFC1122: OK. Passes ICMP errors back to application, as per
677 * 4.1.3.3.
678 */
679 if (!inet->recverr) {
680 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
681 goto out;
682 } else
683 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
684
685 sk->sk_err = err;
686 sk->sk_error_report(sk);
687 out:
688 sock_put(sk);
689 }
690
691 void udp_err(struct sk_buff *skb, u32 info)
692 {
693 __udp4_lib_err(skb, info, &udp_table);
694 }
695
696 /*
697 * Throw away all pending data and cancel the corking. Socket is locked.
698 */
699 void udp_flush_pending_frames(struct sock *sk)
700 {
701 struct udp_sock *up = udp_sk(sk);
702
703 if (up->pending) {
704 up->len = 0;
705 up->pending = 0;
706 ip_flush_pending_frames(sk);
707 }
708 }
709 EXPORT_SYMBOL(udp_flush_pending_frames);
710
711 /**
712 * udp4_hwcsum - handle outgoing HW checksumming
713 * @skb: sk_buff containing the filled-in UDP header
714 * (checksum field must be zeroed out)
715 * @src: source IP address
716 * @dst: destination IP address
717 */
718 static void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
719 {
720 struct udphdr *uh = udp_hdr(skb);
721 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
722 int offset = skb_transport_offset(skb);
723 int len = skb->len - offset;
724 int hlen = len;
725 __wsum csum = 0;
726
727 if (!frags) {
728 /*
729 * Only one fragment on the socket.
730 */
731 skb->csum_start = skb_transport_header(skb) - skb->head;
732 skb->csum_offset = offsetof(struct udphdr, check);
733 uh->check = ~csum_tcpudp_magic(src, dst, len,
734 IPPROTO_UDP, 0);
735 } else {
736 /*
737 * HW-checksum won't work as there are two or more
738 * fragments on the socket so that all csums of sk_buffs
739 * should be together
740 */
741 do {
742 csum = csum_add(csum, frags->csum);
743 hlen -= frags->len;
744 } while ((frags = frags->next));
745
746 csum = skb_checksum(skb, offset, hlen, csum);
747 skb->ip_summed = CHECKSUM_NONE;
748
749 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
750 if (uh->check == 0)
751 uh->check = CSUM_MANGLED_0;
752 }
753 }
754
755 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
756 {
757 struct sock *sk = skb->sk;
758 struct inet_sock *inet = inet_sk(sk);
759 struct udphdr *uh;
760 int err = 0;
761 int is_udplite = IS_UDPLITE(sk);
762 int offset = skb_transport_offset(skb);
763 int len = skb->len - offset;
764 __wsum csum = 0;
765
766 /*
767 * Create a UDP header
768 */
769 uh = udp_hdr(skb);
770 uh->source = inet->inet_sport;
771 uh->dest = fl4->fl4_dport;
772 uh->len = htons(len);
773 uh->check = 0;
774
775 if (is_udplite) /* UDP-Lite */
776 csum = udplite_csum(skb);
777
778 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
779
780 skb->ip_summed = CHECKSUM_NONE;
781 goto send;
782
783 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
784
785 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
786 goto send;
787
788 } else
789 csum = udp_csum(skb);
790
791 /* add protocol-dependent pseudo-header */
792 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
793 sk->sk_protocol, csum);
794 if (uh->check == 0)
795 uh->check = CSUM_MANGLED_0;
796
797 send:
798 err = ip_send_skb(sock_net(sk), skb);
799 if (err) {
800 if (err == -ENOBUFS && !inet->recverr) {
801 UDP_INC_STATS_USER(sock_net(sk),
802 UDP_MIB_SNDBUFERRORS, is_udplite);
803 err = 0;
804 }
805 } else
806 UDP_INC_STATS_USER(sock_net(sk),
807 UDP_MIB_OUTDATAGRAMS, is_udplite);
808 return err;
809 }
810
811 /*
812 * Push out all pending data as one UDP datagram. Socket is locked.
813 */
814 int udp_push_pending_frames(struct sock *sk)
815 {
816 struct udp_sock *up = udp_sk(sk);
817 struct inet_sock *inet = inet_sk(sk);
818 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
819 struct sk_buff *skb;
820 int err = 0;
821
822 skb = ip_finish_skb(sk, fl4);
823 if (!skb)
824 goto out;
825
826 err = udp_send_skb(skb, fl4);
827
828 out:
829 up->len = 0;
830 up->pending = 0;
831 return err;
832 }
833 EXPORT_SYMBOL(udp_push_pending_frames);
834
835 #ifdef UDP_SKT_WIFI
836 void udp_event_trace_printk(const char * fmt, int pid, __u16 port)
837 {
838
839 #ifdef CONFIG_TRACING
840 if(unlikely(0 == udp_tracing_mark_write_addr)) {
841 udp_tracing_mark_write_addr = kallsyms_lookup_name("tracing_mark_write");
842 }
843 event_trace_printk(udp_tracing_mark_write_addr, fmt, pid, MET_SOCKET_LATENCY_NAME, ntohs(port));
844 #endif
845 }
846 EXPORT_SYMBOL(udp_event_trace_printk);
847 #endif
848
849 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
850 size_t len)
851 {
852 struct inet_sock *inet = inet_sk(sk);
853 struct udp_sock *up = udp_sk(sk);
854 struct flowi4 fl4_stack;
855 struct flowi4 *fl4;
856 int ulen = len;
857 struct ipcm_cookie ipc;
858 struct rtable *rt = NULL;
859 int free = 0;
860 int connected = 0;
861 __be32 daddr, faddr, saddr;
862 __be16 dport;
863 u8 tos;
864 int err, is_udplite = IS_UDPLITE(sk);
865 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
866 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
867 struct sk_buff *skb;
868 struct ip_options_data opt_copy;
869
870 if (len > 0xFFFF)
871 return -EMSGSIZE;
872
873 /*
874 * Check the flags.
875 */
876
877 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
878 return -EOPNOTSUPP;
879
880 ipc.opt = NULL;
881 ipc.tx_flags = 0;
882
883 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
884
885 fl4 = &inet->cork.fl.u.ip4;
886 if (up->pending) {
887 /*
888 * There are pending frames.
889 * The socket lock must be held while it's corked.
890 */
891 lock_sock(sk);
892 if (likely(up->pending)) {
893 if (unlikely(up->pending != AF_INET)) {
894 release_sock(sk);
895 return -EINVAL;
896 }
897 goto do_append_data;
898 }
899 release_sock(sk);
900 }
901 ulen += sizeof(struct udphdr);
902
903 /*
904 * Get and verify the address.
905 */
906 if (msg->msg_name) {
907 struct sockaddr_in *usin = (struct sockaddr_in *)msg->msg_name;
908 if (msg->msg_namelen < sizeof(*usin))
909 return -EINVAL;
910 if (usin->sin_family != AF_INET) {
911 if (usin->sin_family != AF_UNSPEC)
912 return -EAFNOSUPPORT;
913 }
914
915 daddr = usin->sin_addr.s_addr;
916 dport = usin->sin_port;
917 if (dport == 0)
918 return -EINVAL;
919 } else {
920 if (sk->sk_state != TCP_ESTABLISHED)
921 return -EDESTADDRREQ;
922 daddr = inet->inet_daddr;
923 dport = inet->inet_dport;
924 /* Open fast path for connected socket.
925 Route will not be used, if at least one option is set.
926 */
927 connected = 1;
928 }
929 ipc.addr = inet->inet_saddr;
930
931 ipc.oif = sk->sk_bound_dev_if;
932
933 #ifdef UDP_SKT_WIFI
934
935 if (unlikely((sysctl_met_is_enable == 1) && (sysctl_udp_met_port > 0))) {
936
937 if ((ntohs(inet->inet_sport) == sysctl_udp_met_port) && (len >= 4)) {
938 __u16 * seq_id = (__u16 *)((char *)msg->msg_iov->iov_base + 2);
939 udp_event_trace_printk("S|%d|%s|%d\n", current->pid, *seq_id);
940
941 }
942 }
943 #endif
944
945
946 sock_tx_timestamp(sk, &ipc.tx_flags);
947
948 if (msg->msg_controllen) {
949 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
950 if (err)
951 return err;
952 if (ipc.opt)
953 free = 1;
954 connected = 0;
955 }
956 if (!ipc.opt) {
957 struct ip_options_rcu *inet_opt;
958
959 rcu_read_lock();
960 inet_opt = rcu_dereference(inet->inet_opt);
961 if (inet_opt) {
962 memcpy(&opt_copy, inet_opt,
963 sizeof(*inet_opt) + inet_opt->opt.optlen);
964 ipc.opt = &opt_copy.opt;
965 }
966 rcu_read_unlock();
967 }
968
969 saddr = ipc.addr;
970 ipc.addr = faddr = daddr;
971
972 if (ipc.opt && ipc.opt->opt.srr) {
973 if (!daddr)
974 return -EINVAL;
975 faddr = ipc.opt->opt.faddr;
976 connected = 0;
977 }
978 tos = RT_TOS(inet->tos);
979 if (sock_flag(sk, SOCK_LOCALROUTE) ||
980 (msg->msg_flags & MSG_DONTROUTE) ||
981 (ipc.opt && ipc.opt->opt.is_strictroute)) {
982 tos |= RTO_ONLINK;
983 connected = 0;
984 }
985
986 if (ipv4_is_multicast(daddr)) {
987 if (!ipc.oif)
988 ipc.oif = inet->mc_index;
989 if (!saddr)
990 saddr = inet->mc_addr;
991 connected = 0;
992 } else if (!ipc.oif)
993 ipc.oif = inet->uc_index;
994
995 if (connected)
996 rt = (struct rtable *)sk_dst_check(sk, 0);
997
998 if (rt == NULL) {
999 struct net *net = sock_net(sk);
1000
1001 fl4 = &fl4_stack;
1002 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1003 RT_SCOPE_UNIVERSE, sk->sk_protocol,
1004 inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP,
1005 faddr, saddr, dport, inet->inet_sport,
1006 sock_i_uid(sk));
1007
1008 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1009 rt = ip_route_output_flow(net, fl4, sk);
1010 if (IS_ERR(rt)) {
1011 err = PTR_ERR(rt);
1012 rt = NULL;
1013 if (err == -ENETUNREACH)
1014 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1015 goto out;
1016 }
1017
1018 err = -EACCES;
1019 if ((rt->rt_flags & RTCF_BROADCAST) &&
1020 !sock_flag(sk, SOCK_BROADCAST))
1021 goto out;
1022 if (connected)
1023 sk_dst_set(sk, dst_clone(&rt->dst));
1024 }
1025
1026 if (msg->msg_flags&MSG_CONFIRM)
1027 goto do_confirm;
1028 back_from_confirm:
1029
1030 saddr = fl4->saddr;
1031 if (!ipc.addr)
1032 daddr = ipc.addr = fl4->daddr;
1033
1034 /* Lockless fast path for the non-corking case. */
1035 if (!corkreq) {
1036 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen,
1037 sizeof(struct udphdr), &ipc, &rt,
1038 msg->msg_flags);
1039 err = PTR_ERR(skb);
1040 if (!IS_ERR_OR_NULL(skb))
1041 err = udp_send_skb(skb, fl4);
1042 goto out;
1043 }
1044
1045 lock_sock(sk);
1046 if (unlikely(up->pending)) {
1047 /* The socket is already corked while preparing it. */
1048 /* ... which is an evident application bug. --ANK */
1049 release_sock(sk);
1050
1051 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("cork app bug 2\n"));
1052 err = -EINVAL;
1053 goto out;
1054 }
1055 /*
1056 * Now cork the socket to pend data.
1057 */
1058 fl4 = &inet->cork.fl.u.ip4;
1059 fl4->daddr = daddr;
1060 fl4->saddr = saddr;
1061 fl4->fl4_dport = dport;
1062 fl4->fl4_sport = inet->inet_sport;
1063 up->pending = AF_INET;
1064
1065 do_append_data:
1066 up->len += ulen;
1067 err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen,
1068 sizeof(struct udphdr), &ipc, &rt,
1069 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1070 if (err)
1071 udp_flush_pending_frames(sk);
1072 else if (!corkreq)
1073 err = udp_push_pending_frames(sk);
1074 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1075 up->pending = 0;
1076 release_sock(sk);
1077
1078 out:
1079 ip_rt_put(rt);
1080 if (free)
1081 kfree(ipc.opt);
1082 if (!err)
1083 return len;
1084 /*
1085 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1086 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1087 * we don't have a good statistic (IpOutDiscards but it can be too many
1088 * things). We could add another new stat but at least for now that
1089 * seems like overkill.
1090 */
1091 /* MTK_NET */
1092 if (err == -ENOBUFS || (sk->sk_socket && test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) ) {
1093 UDP_INC_STATS_USER(sock_net(sk),
1094 UDP_MIB_SNDBUFERRORS, is_udplite);
1095 }
1096 return err;
1097
1098 do_confirm:
1099 dst_confirm(&rt->dst);
1100 if (!(msg->msg_flags&MSG_PROBE) || len)
1101 goto back_from_confirm;
1102 err = 0;
1103 goto out;
1104 }
1105 EXPORT_SYMBOL(udp_sendmsg);
1106
1107 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1108 size_t size, int flags)
1109 {
1110 struct inet_sock *inet = inet_sk(sk);
1111 struct udp_sock *up = udp_sk(sk);
1112 int ret;
1113
1114 if (flags & MSG_SENDPAGE_NOTLAST)
1115 flags |= MSG_MORE;
1116
1117 if (!up->pending) {
1118 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1119
1120 /* Call udp_sendmsg to specify destination address which
1121 * sendpage interface can't pass.
1122 * This will succeed only when the socket is connected.
1123 */
1124 ret = udp_sendmsg(NULL, sk, &msg, 0);
1125 if (ret < 0)
1126 return ret;
1127 }
1128
1129 lock_sock(sk);
1130
1131 if (unlikely(!up->pending)) {
1132 release_sock(sk);
1133
1134 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("udp cork app bug 3\n"));
1135 return -EINVAL;
1136 }
1137
1138 ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1139 page, offset, size, flags);
1140 if (ret == -EOPNOTSUPP) {
1141 release_sock(sk);
1142 return sock_no_sendpage(sk->sk_socket, page, offset,
1143 size, flags);
1144 }
1145 if (ret < 0) {
1146 udp_flush_pending_frames(sk);
1147 goto out;
1148 }
1149
1150 up->len += size;
1151 if (!(up->corkflag || (flags&MSG_MORE)))
1152 ret = udp_push_pending_frames(sk);
1153 if (!ret)
1154 ret = size;
1155 out:
1156 release_sock(sk);
1157 return ret;
1158 }
1159
1160
1161 /**
1162 * first_packet_length - return length of first packet in receive queue
1163 * @sk: socket
1164 *
1165 * Drops all bad checksum frames, until a valid one is found.
1166 * Returns the length of found skb, or 0 if none is found.
1167 */
1168 static unsigned int first_packet_length(struct sock *sk)
1169 {
1170 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1171 struct sk_buff *skb;
1172 unsigned int res;
1173
1174 __skb_queue_head_init(&list_kill);
1175
1176 spin_lock_bh(&rcvq->lock);
1177 while ((skb = skb_peek(rcvq)) != NULL &&
1178 udp_lib_checksum_complete(skb)) {
1179 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS,
1180 IS_UDPLITE(sk));
1181 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1182 IS_UDPLITE(sk));
1183 atomic_inc(&sk->sk_drops);
1184 __skb_unlink(skb, rcvq);
1185 __skb_queue_tail(&list_kill, skb);
1186 }
1187 res = skb ? skb->len : 0;
1188 spin_unlock_bh(&rcvq->lock);
1189
1190 if (!skb_queue_empty(&list_kill)) {
1191 bool slow = lock_sock_fast(sk);
1192
1193 __skb_queue_purge(&list_kill);
1194 sk_mem_reclaim_partial(sk);
1195 unlock_sock_fast(sk, slow);
1196 }
1197 return res;
1198 }
1199
1200 /*
1201 * IOCTL requests applicable to the UDP protocol
1202 */
1203
1204 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1205 {
1206 switch (cmd) {
1207 case SIOCOUTQ:
1208 {
1209 int amount = sk_wmem_alloc_get(sk);
1210
1211 return put_user(amount, (int __user *)arg);
1212 }
1213
1214 case SIOCINQ:
1215 {
1216 unsigned int amount = first_packet_length(sk);
1217
1218 if (amount)
1219 /*
1220 * We will only return the amount
1221 * of this packet since that is all
1222 * that will be read.
1223 */
1224 amount -= sizeof(struct udphdr);
1225
1226 return put_user(amount, (int __user *)arg);
1227 }
1228
1229 default:
1230 return -ENOIOCTLCMD;
1231 }
1232
1233 return 0;
1234 }
1235 EXPORT_SYMBOL(udp_ioctl);
1236
1237 /*
1238 * This should be easy, if there is something there we
1239 * return it, otherwise we block.
1240 */
1241
1242 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1243 size_t len, int noblock, int flags, int *addr_len)
1244 {
1245 struct inet_sock *inet = inet_sk(sk);
1246 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1247 struct sk_buff *skb;
1248 unsigned int ulen, copied;
1249 int peeked, off = 0;
1250 int err;
1251 int is_udplite = IS_UDPLITE(sk);
1252 bool checksum_valid = false;
1253 bool slow;
1254
1255 if (flags & MSG_ERRQUEUE)
1256 return ip_recv_error(sk, msg, len, addr_len);
1257
1258 try_again:
1259 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1260 &peeked, &off, &err);
1261 if (!skb)
1262 goto out;
1263
1264 ulen = skb->len - sizeof(struct udphdr);
1265 copied = len;
1266 if (copied > ulen)
1267 copied = ulen;
1268 else if (copied < ulen)
1269 msg->msg_flags |= MSG_TRUNC;
1270
1271 /*
1272 * If checksum is needed at all, try to do it while copying the
1273 * data. If the data is truncated, or if we only want a partial
1274 * coverage checksum (UDP-Lite), do it before the copy.
1275 */
1276
1277 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1278 checksum_valid = !udp_lib_checksum_complete(skb);
1279 if (!checksum_valid)
1280 goto csum_copy_err;
1281 }
1282
1283 if (checksum_valid || skb_csum_unnecessary(skb))
1284 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1285 msg->msg_iov, copied);
1286 else {
1287 err = skb_copy_and_csum_datagram_iovec(skb,
1288 sizeof(struct udphdr),
1289 msg->msg_iov);
1290
1291 if (err == -EINVAL)
1292 goto csum_copy_err;
1293 }
1294
1295 if (unlikely(err)) {
1296 trace_kfree_skb(skb, udp_recvmsg);
1297 if (!peeked) {
1298 atomic_inc(&sk->sk_drops);
1299 UDP_INC_STATS_USER(sock_net(sk),
1300 UDP_MIB_INERRORS, is_udplite);
1301 }
1302 goto out_free;
1303 }
1304
1305 if (!peeked)
1306 UDP_INC_STATS_USER(sock_net(sk),
1307 UDP_MIB_INDATAGRAMS, is_udplite);
1308
1309 sock_recv_ts_and_drops(msg, sk, skb);
1310
1311 /* Copy the address. */
1312 if (sin) {
1313 sin->sin_family = AF_INET;
1314 sin->sin_port = udp_hdr(skb)->source;
1315 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1316 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1317 *addr_len = sizeof(*sin);
1318 }
1319 if (inet->cmsg_flags)
1320 ip_cmsg_recv(msg, skb);
1321
1322 err = copied;
1323 if (flags & MSG_TRUNC)
1324 err = ulen;
1325
1326 out_free:
1327 skb_free_datagram_locked(sk, skb);
1328 out:
1329 return err;
1330
1331 csum_copy_err:
1332 slow = lock_sock_fast(sk);
1333 if (!skb_kill_datagram(sk, skb, flags)) {
1334 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1335 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1336 }
1337 unlock_sock_fast(sk, slow);
1338
1339 /* starting over for a new packet, but check if we need to yield */
1340 cond_resched();
1341 msg->msg_flags &= ~MSG_TRUNC;
1342 goto try_again;
1343 }
1344
1345
1346 int udp_disconnect(struct sock *sk, int flags)
1347 {
1348 struct inet_sock *inet = inet_sk(sk);
1349 /*
1350 * 1003.1g - break association.
1351 */
1352
1353 sk->sk_state = TCP_CLOSE;
1354 inet->inet_daddr = 0;
1355 inet->inet_dport = 0;
1356 sock_rps_reset_rxhash(sk);
1357 sk->sk_bound_dev_if = 0;
1358 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1359 inet_reset_saddr(sk);
1360
1361 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1362 sk->sk_prot->unhash(sk);
1363 inet->inet_sport = 0;
1364 }
1365 sk_dst_reset(sk);
1366 return 0;
1367 }
1368 EXPORT_SYMBOL(udp_disconnect);
1369
1370 void udp_lib_unhash(struct sock *sk)
1371 {
1372 if (sk_hashed(sk)) {
1373 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1374 struct udp_hslot *hslot, *hslot2;
1375
1376 hslot = udp_hashslot(udptable, sock_net(sk),
1377 udp_sk(sk)->udp_port_hash);
1378 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1379
1380 spin_lock_bh(&hslot->lock);
1381 if (sk_nulls_del_node_init_rcu(sk)) {
1382 hslot->count--;
1383 inet_sk(sk)->inet_num = 0;
1384 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1385
1386 spin_lock(&hslot2->lock);
1387 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1388 hslot2->count--;
1389 spin_unlock(&hslot2->lock);
1390 }
1391 spin_unlock_bh(&hslot->lock);
1392 }
1393 }
1394 EXPORT_SYMBOL(udp_lib_unhash);
1395
1396 /*
1397 * inet_rcv_saddr was changed, we must rehash secondary hash
1398 */
1399 void udp_lib_rehash(struct sock *sk, u16 newhash)
1400 {
1401 if (sk_hashed(sk)) {
1402 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1403 struct udp_hslot *hslot, *hslot2, *nhslot2;
1404
1405 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1406 nhslot2 = udp_hashslot2(udptable, newhash);
1407 udp_sk(sk)->udp_portaddr_hash = newhash;
1408 if (hslot2 != nhslot2) {
1409 hslot = udp_hashslot(udptable, sock_net(sk),
1410 udp_sk(sk)->udp_port_hash);
1411 /* we must lock primary chain too */
1412 spin_lock_bh(&hslot->lock);
1413
1414 spin_lock(&hslot2->lock);
1415 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1416 hslot2->count--;
1417 spin_unlock(&hslot2->lock);
1418
1419 spin_lock(&nhslot2->lock);
1420 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1421 &nhslot2->head);
1422 nhslot2->count++;
1423 spin_unlock(&nhslot2->lock);
1424
1425 spin_unlock_bh(&hslot->lock);
1426 }
1427 }
1428 }
1429 EXPORT_SYMBOL(udp_lib_rehash);
1430
1431 static void udp_v4_rehash(struct sock *sk)
1432 {
1433 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1434 inet_sk(sk)->inet_rcv_saddr,
1435 inet_sk(sk)->inet_num);
1436 udp_lib_rehash(sk, new_hash);
1437 }
1438
1439 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1440 {
1441 int rc;
1442
1443 if (inet_sk(sk)->inet_daddr)
1444 sock_rps_save_rxhash(sk, skb);
1445
1446 rc = sock_queue_rcv_skb(sk, skb);
1447 if (rc < 0) {
1448 int is_udplite = IS_UDPLITE(sk);
1449
1450 /* Note that an ENOMEM error is charged twice */
1451 if (rc == -ENOMEM)
1452 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1453 is_udplite);
1454 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1455 kfree_skb(skb);
1456 trace_udp_fail_queue_rcv_skb(rc, sk);
1457 return -1;
1458 }
1459
1460 return 0;
1461
1462 }
1463
1464 static struct static_key udp_encap_needed __read_mostly;
1465 void udp_encap_enable(void)
1466 {
1467 if (!static_key_enabled(&udp_encap_needed))
1468 static_key_slow_inc(&udp_encap_needed);
1469 }
1470 EXPORT_SYMBOL(udp_encap_enable);
1471
1472 /* returns:
1473 * -1: error
1474 * 0: success
1475 * >0: "udp encap" protocol resubmission
1476 *
1477 * Note that in the success and error cases, the skb is assumed to
1478 * have either been requeued or freed.
1479 */
1480 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1481 {
1482 struct udp_sock *up = udp_sk(sk);
1483 int rc;
1484 int is_udplite = IS_UDPLITE(sk);
1485
1486 /*
1487 * Charge it to the socket, dropping if the queue is full.
1488 */
1489 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1490 goto drop;
1491 nf_reset(skb);
1492
1493 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1494 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1495
1496 /*
1497 * This is an encapsulation socket so pass the skb to
1498 * the socket's udp_encap_rcv() hook. Otherwise, just
1499 * fall through and pass this up the UDP socket.
1500 * up->encap_rcv() returns the following value:
1501 * =0 if skb was successfully passed to the encap
1502 * handler or was discarded by it.
1503 * >0 if skb should be passed on to UDP.
1504 * <0 if skb should be resubmitted as proto -N
1505 */
1506
1507 /* if we're overly short, let UDP handle it */
1508 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1509 if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
1510 int ret;
1511
1512 ret = encap_rcv(sk, skb);
1513 if (ret <= 0) {
1514 UDP_INC_STATS_BH(sock_net(sk),
1515 UDP_MIB_INDATAGRAMS,
1516 is_udplite);
1517 return -ret;
1518 }
1519 }
1520
1521 /* FALLTHROUGH -- it's a UDP Packet */
1522 }
1523
1524 /*
1525 * UDP-Lite specific tests, ignored on UDP sockets
1526 */
1527 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1528
1529 /*
1530 * MIB statistics other than incrementing the error count are
1531 * disabled for the following two types of errors: these depend
1532 * on the application settings, not on the functioning of the
1533 * protocol stack as such.
1534 *
1535 * RFC 3828 here recommends (sec 3.3): "There should also be a
1536 * way ... to ... at least let the receiving application block
1537 * delivery of packets with coverage values less than a value
1538 * provided by the application."
1539 */
1540 if (up->pcrlen == 0) { /* full coverage was set */
1541 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: partial coverage %d while full coverage %d requested\n",
1542 UDP_SKB_CB(skb)->cscov, skb->len);
1543 goto drop;
1544 }
1545 /* The next case involves violating the min. coverage requested
1546 * by the receiver. This is subtle: if receiver wants x and x is
1547 * greater than the buffersize/MTU then receiver will complain
1548 * that it wants x while sender emits packets of smaller size y.
1549 * Therefore the above ...()->partial_cov statement is essential.
1550 */
1551 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1552 LIMIT_NETDEBUG(KERN_WARNING "UDPLite: coverage %d too small, need min %d\n",
1553 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1554 goto drop;
1555 }
1556 }
1557
1558 if (rcu_access_pointer(sk->sk_filter) &&
1559 udp_lib_checksum_complete(skb))
1560 goto csum_error;
1561
1562
1563 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
1564 goto drop;
1565
1566 rc = 0;
1567
1568 ipv4_pktinfo_prepare(skb);
1569 bh_lock_sock(sk);
1570 if (!sock_owned_by_user(sk))
1571 rc = __udp_queue_rcv_skb(sk, skb);
1572 else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1573 bh_unlock_sock(sk);
1574 goto drop;
1575 }
1576 bh_unlock_sock(sk);
1577
1578 return rc;
1579
1580 csum_error:
1581 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1582 drop:
1583 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1584 atomic_inc(&sk->sk_drops);
1585 kfree_skb(skb);
1586 return -1;
1587 }
1588
1589
1590 static void flush_stack(struct sock **stack, unsigned int count,
1591 struct sk_buff *skb, unsigned int final)
1592 {
1593 unsigned int i;
1594 struct sk_buff *skb1 = NULL;
1595 struct sock *sk;
1596
1597 for (i = 0; i < count; i++) {
1598 sk = stack[i];
1599 if (likely(skb1 == NULL))
1600 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1601
1602 if (!skb1) {
1603 atomic_inc(&sk->sk_drops);
1604 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1605 IS_UDPLITE(sk));
1606 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1607 IS_UDPLITE(sk));
1608 }
1609
1610 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1611 skb1 = NULL;
1612 }
1613 if (unlikely(skb1))
1614 kfree_skb(skb1);
1615 }
1616
1617 /*
1618 * Multicasts and broadcasts go to each listener.
1619 *
1620 * Note: called only from the BH handler context.
1621 */
1622 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1623 struct udphdr *uh,
1624 __be32 saddr, __be32 daddr,
1625 struct udp_table *udptable)
1626 {
1627 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1628 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1629 int dif;
1630 unsigned int i, count = 0;
1631
1632 spin_lock(&hslot->lock);
1633 sk = sk_nulls_head(&hslot->head);
1634 dif = skb->dev->ifindex;
1635 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1636 while (sk) {
1637 stack[count++] = sk;
1638 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1639 daddr, uh->source, saddr, dif);
1640 if (unlikely(count == ARRAY_SIZE(stack))) {
1641 if (!sk)
1642 break;
1643 flush_stack(stack, count, skb, ~0);
1644 count = 0;
1645 }
1646 }
1647 /*
1648 * before releasing chain lock, we must take a reference on sockets
1649 */
1650 for (i = 0; i < count; i++)
1651 sock_hold(stack[i]);
1652
1653 spin_unlock(&hslot->lock);
1654
1655 /*
1656 * do the slow work with no lock held
1657 */
1658 if (count) {
1659 flush_stack(stack, count, skb, count - 1);
1660
1661 for (i = 0; i < count; i++)
1662 sock_put(stack[i]);
1663 } else {
1664 kfree_skb(skb);
1665 }
1666 return 0;
1667 }
1668
1669 /* Initialize UDP checksum. If exited with zero value (success),
1670 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1671 * Otherwise, csum completion requires chacksumming packet body,
1672 * including udp header and folding it to skb->csum.
1673 */
1674 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1675 int proto)
1676 {
1677 const struct iphdr *iph;
1678 int err;
1679
1680 UDP_SKB_CB(skb)->partial_cov = 0;
1681 UDP_SKB_CB(skb)->cscov = skb->len;
1682
1683 if (proto == IPPROTO_UDPLITE) {
1684 err = udplite_checksum_init(skb, uh);
1685 if (err)
1686 return err;
1687 }
1688
1689 iph = ip_hdr(skb);
1690 if (uh->check == 0) {
1691 skb->ip_summed = CHECKSUM_UNNECESSARY;
1692 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1693 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1694 proto, skb->csum))
1695 skb->ip_summed = CHECKSUM_UNNECESSARY;
1696 }
1697 if (!skb_csum_unnecessary(skb))
1698 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1699 skb->len, proto, 0);
1700 /* Probably, we should checksum udp header (it should be in cache
1701 * in any case) and data in tiny packets (< rx copybreak).
1702 */
1703
1704 return 0;
1705 }
1706
1707 /*
1708 * All we need to do is get the socket, and then do a checksum.
1709 */
1710
1711 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1712 int proto)
1713 {
1714 struct sock *sk;
1715 struct udphdr *uh;
1716 unsigned short ulen;
1717 struct rtable *rt = skb_rtable(skb);
1718 __be32 saddr, daddr;
1719 struct net *net = dev_net(skb->dev);
1720
1721 /*
1722 * Validate the packet.
1723 */
1724 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1725 goto drop; /* No space for header. */
1726
1727 uh = udp_hdr(skb);
1728 ulen = ntohs(uh->len);
1729 saddr = ip_hdr(skb)->saddr;
1730 daddr = ip_hdr(skb)->daddr;
1731
1732 if (ulen > skb->len)
1733 goto short_packet;
1734
1735 if (proto == IPPROTO_UDP) {
1736 /* UDP validates ulen. */
1737 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1738 goto short_packet;
1739 uh = udp_hdr(skb);
1740 }
1741
1742 if (udp4_csum_init(skb, uh, proto))
1743 goto csum_error;
1744
1745 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1746 return __udp4_lib_mcast_deliver(net, skb, uh,
1747 saddr, daddr, udptable);
1748
1749 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1750
1751 if (sk != NULL) {
1752 int ret = udp_queue_rcv_skb(sk, skb);
1753 sock_put(sk);
1754
1755 /* a return value > 0 means to resubmit the input, but
1756 * it wants the return to be -protocol, or 0
1757 */
1758 if (ret > 0)
1759 return -ret;
1760 return 0;
1761 }
1762
1763 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1764 goto drop;
1765 nf_reset(skb);
1766
1767 /* No socket. Drop packet silently, if checksum is wrong */
1768 if (udp_lib_checksum_complete(skb))
1769 goto csum_error;
1770
1771 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1772 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1773
1774 /*
1775 * Hmm. We got an UDP packet to a port to which we
1776 * don't wanna listen. Ignore it.
1777 */
1778 kfree_skb(skb);
1779 return 0;
1780
1781 short_packet:
1782 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1783 proto == IPPROTO_UDPLITE ? "Lite" : "",
1784 &saddr, ntohs(uh->source),
1785 ulen, skb->len,
1786 &daddr, ntohs(uh->dest));
1787 goto drop;
1788
1789 csum_error:
1790 /*
1791 * RFC1122: OK. Discards the bad packet silently (as far as
1792 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1793 */
1794 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1795 proto == IPPROTO_UDPLITE ? "Lite" : "",
1796 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1797 ulen);
1798 UDP_INC_STATS_BH(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1799 drop:
1800 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1801 kfree_skb(skb);
1802 return 0;
1803 }
1804
1805 int udp_rcv(struct sk_buff *skb)
1806 {
1807 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1808 }
1809
1810 void udp_destroy_sock(struct sock *sk)
1811 {
1812 struct udp_sock *up = udp_sk(sk);
1813 bool slow = lock_sock_fast(sk);
1814 udp_flush_pending_frames(sk);
1815 unlock_sock_fast(sk, slow);
1816 if (static_key_false(&udp_encap_needed) && up->encap_type) {
1817 void (*encap_destroy)(struct sock *sk);
1818 encap_destroy = ACCESS_ONCE(up->encap_destroy);
1819 if (encap_destroy)
1820 encap_destroy(sk);
1821 }
1822 }
1823
1824 /*
1825 * Socket option code for UDP
1826 */
1827 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1828 char __user *optval, unsigned int optlen,
1829 int (*push_pending_frames)(struct sock *))
1830 {
1831 struct udp_sock *up = udp_sk(sk);
1832 int val;
1833 int err = 0;
1834 int is_udplite = IS_UDPLITE(sk);
1835
1836 if (optlen < sizeof(int))
1837 return -EINVAL;
1838
1839 if (get_user(val, (int __user *)optval))
1840 return -EFAULT;
1841
1842 switch (optname) {
1843 case UDP_CORK:
1844 if (val != 0) {
1845 up->corkflag = 1;
1846 } else {
1847 up->corkflag = 0;
1848 lock_sock(sk);
1849 (*push_pending_frames)(sk);
1850 release_sock(sk);
1851 }
1852 break;
1853
1854 case UDP_ENCAP:
1855 switch (val) {
1856 case 0:
1857 case UDP_ENCAP_ESPINUDP:
1858 case UDP_ENCAP_ESPINUDP_NON_IKE:
1859 up->encap_rcv = xfrm4_udp_encap_rcv;
1860 /* FALLTHROUGH */
1861 case UDP_ENCAP_L2TPINUDP:
1862 up->encap_type = val;
1863 udp_encap_enable();
1864 break;
1865 default:
1866 err = -ENOPROTOOPT;
1867 break;
1868 }
1869 break;
1870
1871 /*
1872 * UDP-Lite's partial checksum coverage (RFC 3828).
1873 */
1874 /* The sender sets actual checksum coverage length via this option.
1875 * The case coverage > packet length is handled by send module. */
1876 case UDPLITE_SEND_CSCOV:
1877 if (!is_udplite) /* Disable the option on UDP sockets */
1878 return -ENOPROTOOPT;
1879 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1880 val = 8;
1881 else if (val > USHRT_MAX)
1882 val = USHRT_MAX;
1883 up->pcslen = val;
1884 up->pcflag |= UDPLITE_SEND_CC;
1885 break;
1886
1887 /* The receiver specifies a minimum checksum coverage value. To make
1888 * sense, this should be set to at least 8 (as done below). If zero is
1889 * used, this again means full checksum coverage. */
1890 case UDPLITE_RECV_CSCOV:
1891 if (!is_udplite) /* Disable the option on UDP sockets */
1892 return -ENOPROTOOPT;
1893 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1894 val = 8;
1895 else if (val > USHRT_MAX)
1896 val = USHRT_MAX;
1897 up->pcrlen = val;
1898 up->pcflag |= UDPLITE_RECV_CC;
1899 break;
1900
1901 default:
1902 err = -ENOPROTOOPT;
1903 break;
1904 }
1905
1906 return err;
1907 }
1908 EXPORT_SYMBOL(udp_lib_setsockopt);
1909
1910 int udp_setsockopt(struct sock *sk, int level, int optname,
1911 char __user *optval, unsigned int optlen)
1912 {
1913 if (level == SOL_UDP || level == SOL_UDPLITE)
1914 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1915 udp_push_pending_frames);
1916 return ip_setsockopt(sk, level, optname, optval, optlen);
1917 }
1918
1919 #ifdef CONFIG_COMPAT
1920 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1921 char __user *optval, unsigned int optlen)
1922 {
1923 if (level == SOL_UDP || level == SOL_UDPLITE)
1924 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1925 udp_push_pending_frames);
1926 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1927 }
1928 #endif
1929
1930 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1931 char __user *optval, int __user *optlen)
1932 {
1933 struct udp_sock *up = udp_sk(sk);
1934 int val, len;
1935
1936 if (get_user(len, optlen))
1937 return -EFAULT;
1938
1939 len = min_t(unsigned int, len, sizeof(int));
1940
1941 if (len < 0)
1942 return -EINVAL;
1943
1944 switch (optname) {
1945 case UDP_CORK:
1946 val = up->corkflag;
1947 break;
1948
1949 case UDP_ENCAP:
1950 val = up->encap_type;
1951 break;
1952
1953 /* The following two cannot be changed on UDP sockets, the return is
1954 * always 0 (which corresponds to the full checksum coverage of UDP). */
1955 case UDPLITE_SEND_CSCOV:
1956 val = up->pcslen;
1957 break;
1958
1959 case UDPLITE_RECV_CSCOV:
1960 val = up->pcrlen;
1961 break;
1962
1963 default:
1964 return -ENOPROTOOPT;
1965 }
1966
1967 if (put_user(len, optlen))
1968 return -EFAULT;
1969 if (copy_to_user(optval, &val, len))
1970 return -EFAULT;
1971 return 0;
1972 }
1973 EXPORT_SYMBOL(udp_lib_getsockopt);
1974
1975 int udp_getsockopt(struct sock *sk, int level, int optname,
1976 char __user *optval, int __user *optlen)
1977 {
1978 if (level == SOL_UDP || level == SOL_UDPLITE)
1979 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1980 return ip_getsockopt(sk, level, optname, optval, optlen);
1981 }
1982
1983 #ifdef CONFIG_COMPAT
1984 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1985 char __user *optval, int __user *optlen)
1986 {
1987 if (level == SOL_UDP || level == SOL_UDPLITE)
1988 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1989 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1990 }
1991 #endif
1992 /**
1993 * udp_poll - wait for a UDP event.
1994 * @file - file struct
1995 * @sock - socket
1996 * @wait - poll table
1997 *
1998 * This is same as datagram poll, except for the special case of
1999 * blocking sockets. If application is using a blocking fd
2000 * and a packet with checksum error is in the queue;
2001 * then it could get return from select indicating data available
2002 * but then block when reading it. Add special case code
2003 * to work around these arguably broken applications.
2004 */
2005 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2006 {
2007 unsigned int mask = datagram_poll(file, sock, wait);
2008 struct sock *sk = sock->sk;
2009
2010 /* Check for false positives due to checksum errors */
2011 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2012 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
2013 mask &= ~(POLLIN | POLLRDNORM);
2014
2015 return mask;
2016
2017 }
2018 EXPORT_SYMBOL(udp_poll);
2019
2020 struct proto udp_prot = {
2021 .name = "UDP",
2022 .owner = THIS_MODULE,
2023 .close = udp_lib_close,
2024 .connect = ip4_datagram_connect,
2025 .disconnect = udp_disconnect,
2026 .ioctl = udp_ioctl,
2027 .destroy = udp_destroy_sock,
2028 .setsockopt = udp_setsockopt,
2029 .getsockopt = udp_getsockopt,
2030 .sendmsg = udp_sendmsg,
2031 .recvmsg = udp_recvmsg,
2032 .sendpage = udp_sendpage,
2033 .backlog_rcv = __udp_queue_rcv_skb,
2034 .release_cb = ip4_datagram_release_cb,
2035 .hash = udp_lib_hash,
2036 .unhash = udp_lib_unhash,
2037 .rehash = udp_v4_rehash,
2038 .get_port = udp_v4_get_port,
2039 .memory_allocated = &udp_memory_allocated,
2040 .sysctl_mem = sysctl_udp_mem,
2041 .sysctl_wmem = &sysctl_udp_wmem_min,
2042 .sysctl_rmem = &sysctl_udp_rmem_min,
2043 .obj_size = sizeof(struct udp_sock),
2044 .slab_flags = SLAB_DESTROY_BY_RCU,
2045 .h.udp_table = &udp_table,
2046 #ifdef CONFIG_COMPAT
2047 .compat_setsockopt = compat_udp_setsockopt,
2048 .compat_getsockopt = compat_udp_getsockopt,
2049 #endif
2050 .clear_sk = sk_prot_clear_portaddr_nulls,
2051 };
2052 EXPORT_SYMBOL(udp_prot);
2053
2054 /* ------------------------------------------------------------------------ */
2055 #ifdef CONFIG_PROC_FS
2056
2057 static struct sock *udp_get_first(struct seq_file *seq, int start)
2058 {
2059 struct sock *sk;
2060 struct udp_iter_state *state = seq->private;
2061 struct net *net = seq_file_net(seq);
2062
2063 for (state->bucket = start; state->bucket <= state->udp_table->mask;
2064 ++state->bucket) {
2065 struct hlist_nulls_node *node;
2066 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2067
2068 if (hlist_nulls_empty(&hslot->head))
2069 continue;
2070
2071 spin_lock_bh(&hslot->lock);
2072 sk_nulls_for_each(sk, node, &hslot->head) {
2073 if (!net_eq(sock_net(sk), net))
2074 continue;
2075 if (sk->sk_family == state->family)
2076 goto found;
2077 }
2078 spin_unlock_bh(&hslot->lock);
2079 }
2080 sk = NULL;
2081 found:
2082 return sk;
2083 }
2084
2085 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2086 {
2087 struct udp_iter_state *state = seq->private;
2088 struct net *net = seq_file_net(seq);
2089
2090 do {
2091 sk = sk_nulls_next(sk);
2092 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2093
2094 if (!sk) {
2095 if (state->bucket <= state->udp_table->mask)
2096 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2097 return udp_get_first(seq, state->bucket + 1);
2098 }
2099 return sk;
2100 }
2101
2102 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2103 {
2104 struct sock *sk = udp_get_first(seq, 0);
2105
2106 if (sk)
2107 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2108 --pos;
2109 return pos ? NULL : sk;
2110 }
2111
2112 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2113 {
2114 struct udp_iter_state *state = seq->private;
2115 state->bucket = MAX_UDP_PORTS;
2116
2117 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2118 }
2119
2120 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2121 {
2122 struct sock *sk;
2123
2124 if (v == SEQ_START_TOKEN)
2125 sk = udp_get_idx(seq, 0);
2126 else
2127 sk = udp_get_next(seq, v);
2128
2129 ++*pos;
2130 return sk;
2131 }
2132
2133 static void udp_seq_stop(struct seq_file *seq, void *v)
2134 {
2135 struct udp_iter_state *state = seq->private;
2136
2137 if (state->bucket <= state->udp_table->mask)
2138 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2139 }
2140
2141 int udp_seq_open(struct inode *inode, struct file *file)
2142 {
2143 struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2144 struct udp_iter_state *s;
2145 int err;
2146
2147 err = seq_open_net(inode, file, &afinfo->seq_ops,
2148 sizeof(struct udp_iter_state));
2149 if (err < 0)
2150 return err;
2151
2152 s = ((struct seq_file *)file->private_data)->private;
2153 s->family = afinfo->family;
2154 s->udp_table = afinfo->udp_table;
2155 return err;
2156 }
2157 EXPORT_SYMBOL(udp_seq_open);
2158
2159 /* ------------------------------------------------------------------------ */
2160 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2161 {
2162 struct proc_dir_entry *p;
2163 int rc = 0;
2164
2165 afinfo->seq_ops.start = udp_seq_start;
2166 afinfo->seq_ops.next = udp_seq_next;
2167 afinfo->seq_ops.stop = udp_seq_stop;
2168
2169 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2170 afinfo->seq_fops, afinfo);
2171 if (!p)
2172 rc = -ENOMEM;
2173 return rc;
2174 }
2175 EXPORT_SYMBOL(udp_proc_register);
2176
2177 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2178 {
2179 remove_proc_entry(afinfo->name, net->proc_net);
2180 }
2181 EXPORT_SYMBOL(udp_proc_unregister);
2182
2183 /* ------------------------------------------------------------------------ */
2184 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2185 int bucket, int *len)
2186 {
2187 struct inet_sock *inet = inet_sk(sp);
2188 __be32 dest = inet->inet_daddr;
2189 __be32 src = inet->inet_rcv_saddr;
2190 __u16 destp = ntohs(inet->inet_dport);
2191 __u16 srcp = ntohs(inet->inet_sport);
2192
2193 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2194 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d%n",
2195 bucket, src, srcp, dest, destp, sp->sk_state,
2196 sk_wmem_alloc_get(sp),
2197 sk_rmem_alloc_get(sp),
2198 0, 0L, 0,
2199 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2200 0, sock_i_ino(sp),
2201 atomic_read(&sp->sk_refcnt), sp,
2202 atomic_read(&sp->sk_drops), len);
2203 }
2204
2205 int udp4_seq_show(struct seq_file *seq, void *v)
2206 {
2207 if (v == SEQ_START_TOKEN)
2208 seq_printf(seq, "%-127s\n",
2209 " sl local_address rem_address st tx_queue "
2210 "rx_queue tr tm->when retrnsmt uid timeout "
2211 "inode ref pointer drops");
2212 else {
2213 struct udp_iter_state *state = seq->private;
2214 int len;
2215
2216 udp4_format_sock(v, seq, state->bucket, &len);
2217 seq_printf(seq, "%*s\n", 127 - len, "");
2218 }
2219 return 0;
2220 }
2221
2222 static const struct file_operations udp_afinfo_seq_fops = {
2223 .owner = THIS_MODULE,
2224 .open = udp_seq_open,
2225 .read = seq_read,
2226 .llseek = seq_lseek,
2227 .release = seq_release_net
2228 };
2229
2230 /* ------------------------------------------------------------------------ */
2231 static struct udp_seq_afinfo udp4_seq_afinfo = {
2232 .name = "udp",
2233 .family = AF_INET,
2234 .udp_table = &udp_table,
2235 .seq_fops = &udp_afinfo_seq_fops,
2236 .seq_ops = {
2237 .show = udp4_seq_show,
2238 },
2239 };
2240
2241 static int __net_init udp4_proc_init_net(struct net *net)
2242 {
2243 return udp_proc_register(net, &udp4_seq_afinfo);
2244 }
2245
2246 static void __net_exit udp4_proc_exit_net(struct net *net)
2247 {
2248 udp_proc_unregister(net, &udp4_seq_afinfo);
2249 }
2250
2251 static struct pernet_operations udp4_net_ops = {
2252 .init = udp4_proc_init_net,
2253 .exit = udp4_proc_exit_net,
2254 };
2255
2256 int __init udp4_proc_init(void)
2257 {
2258 return register_pernet_subsys(&udp4_net_ops);
2259 }
2260
2261 void udp4_proc_exit(void)
2262 {
2263 unregister_pernet_subsys(&udp4_net_ops);
2264 }
2265 #endif /* CONFIG_PROC_FS */
2266
2267 static __initdata unsigned long uhash_entries;
2268 static int __init set_uhash_entries(char *str)
2269 {
2270 ssize_t ret;
2271
2272 if (!str)
2273 return 0;
2274
2275 ret = kstrtoul(str, 0, &uhash_entries);
2276 if (ret)
2277 return 0;
2278
2279 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2280 uhash_entries = UDP_HTABLE_SIZE_MIN;
2281 return 1;
2282 }
2283 __setup("uhash_entries=", set_uhash_entries);
2284
2285 void __init udp_table_init(struct udp_table *table, const char *name)
2286 {
2287 unsigned int i;
2288
2289 table->hash = alloc_large_system_hash(name,
2290 2 * sizeof(struct udp_hslot),
2291 uhash_entries,
2292 21, /* one slot per 2 MB */
2293 0,
2294 &table->log,
2295 &table->mask,
2296 UDP_HTABLE_SIZE_MIN,
2297 64 * 1024);
2298
2299 table->hash2 = table->hash + (table->mask + 1);
2300 for (i = 0; i <= table->mask; i++) {
2301 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2302 table->hash[i].count = 0;
2303 spin_lock_init(&table->hash[i].lock);
2304 }
2305 for (i = 0; i <= table->mask; i++) {
2306 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2307 table->hash2[i].count = 0;
2308 spin_lock_init(&table->hash2[i].lock);
2309 }
2310 }
2311
2312 void __init udp_init(void)
2313 {
2314 unsigned long limit;
2315
2316 udp_table_init(&udp_table, "UDP");
2317 limit = nr_free_buffer_pages() / 8;
2318 limit = max(limit, 128UL);
2319 sysctl_udp_mem[0] = limit / 4 * 3;
2320 sysctl_udp_mem[1] = limit;
2321 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2322
2323 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2324 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2325 }
2326
2327 int udp4_ufo_send_check(struct sk_buff *skb)
2328 {
2329 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2330 return -EINVAL;
2331
2332 if (likely(!skb->encapsulation)) {
2333 const struct iphdr *iph;
2334 struct udphdr *uh;
2335
2336 iph = ip_hdr(skb);
2337 uh = udp_hdr(skb);
2338
2339 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
2340 IPPROTO_UDP, 0);
2341 skb->csum_start = skb_transport_header(skb) - skb->head;
2342 skb->csum_offset = offsetof(struct udphdr, check);
2343 skb->ip_summed = CHECKSUM_PARTIAL;
2344 }
2345 return 0;
2346 }
2347
2348 static struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
2349 netdev_features_t features)
2350 {
2351 struct sk_buff *segs = ERR_PTR(-EINVAL);
2352 int mac_len = skb->mac_len;
2353 int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
2354 __be16 protocol = skb->protocol;
2355 netdev_features_t enc_features;
2356 int outer_hlen;
2357
2358 if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
2359 goto out;
2360
2361 skb->encapsulation = 0;
2362 __skb_pull(skb, tnl_hlen);
2363 skb_reset_mac_header(skb);
2364 skb_set_network_header(skb, skb_inner_network_offset(skb));
2365 skb->mac_len = skb_inner_network_offset(skb);
2366 skb->protocol = htons(ETH_P_TEB);
2367
2368 /* segment inner packet. */
2369 enc_features = skb->dev->hw_enc_features & netif_skb_features(skb);
2370 segs = skb_mac_gso_segment(skb, enc_features);
2371 if (!segs || IS_ERR(segs))
2372 goto out;
2373
2374 outer_hlen = skb_tnl_header_len(skb);
2375 skb = segs;
2376 do {
2377 struct udphdr *uh;
2378 int udp_offset = outer_hlen - tnl_hlen;
2379
2380 skb->mac_len = mac_len;
2381
2382 skb_push(skb, outer_hlen);
2383 skb_reset_mac_header(skb);
2384 skb_set_network_header(skb, mac_len);
2385 skb_set_transport_header(skb, udp_offset);
2386 uh = udp_hdr(skb);
2387 uh->len = htons(skb->len - udp_offset);
2388
2389 /* csum segment if tunnel sets skb with csum. */
2390 if (unlikely(uh->check)) {
2391 struct iphdr *iph = ip_hdr(skb);
2392
2393 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
2394 skb->len - udp_offset,
2395 IPPROTO_UDP, 0);
2396 uh->check = csum_fold(skb_checksum(skb, udp_offset,
2397 skb->len - udp_offset, 0));
2398 if (uh->check == 0)
2399 uh->check = CSUM_MANGLED_0;
2400
2401 }
2402 skb->ip_summed = CHECKSUM_NONE;
2403 skb->protocol = protocol;
2404 } while ((skb = skb->next));
2405 out:
2406 return segs;
2407 }
2408
2409 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
2410 netdev_features_t features)
2411 {
2412 struct sk_buff *segs = ERR_PTR(-EINVAL);
2413 unsigned int mss;
2414 mss = skb_shinfo(skb)->gso_size;
2415 if (unlikely(skb->len <= mss))
2416 goto out;
2417
2418 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2419 /* Packet is from an untrusted source, reset gso_segs. */
2420 int type = skb_shinfo(skb)->gso_type;
2421
2422 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY |
2423 SKB_GSO_UDP_TUNNEL |
2424 SKB_GSO_GRE) ||
2425 !(type & (SKB_GSO_UDP))))
2426 goto out;
2427
2428 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
2429
2430 segs = NULL;
2431 goto out;
2432 }
2433
2434 /* Fragment the skb. IP headers of the fragments are updated in
2435 * inet_gso_segment()
2436 */
2437 if (skb->encapsulation && skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL)
2438 segs = skb_udp_tunnel_segment(skb, features);
2439 else {
2440 int offset;
2441 __wsum csum;
2442
2443 /* Do software UFO. Complete and fill in the UDP checksum as
2444 * HW cannot do checksum of UDP packets sent as multiple
2445 * IP fragments.
2446 */
2447 offset = skb_checksum_start_offset(skb);
2448 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2449 offset += skb->csum_offset;
2450 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2451 skb->ip_summed = CHECKSUM_NONE;
2452
2453 segs = skb_segment(skb, features);
2454 }
2455 out:
2456 return segs;
2457 }
kernel source for telekom puls (alcatel/mediatek)