projects / GitHub / LineageOS / android_kernel_samsung_universal7580.git / blob
? search:


006e2ccd6cc2c48e0702717117772ab0df6fe121
[GitHub/LineageOS/android_kernel_samsung_universal7580.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 #include <asm/system.h>
81 #include <asm/uaccess.h>
82 #include <asm/ioctls.h>
83 #include <linux/bootmem.h>
84 #include <linux/highmem.h>
85 #include <linux/swap.h>
86 #include <linux/types.h>
87 #include <linux/fcntl.h>
88 #include <linux/module.h>
89 #include <linux/socket.h>
90 #include <linux/sockios.h>
91 #include <linux/igmp.h>
92 #include <linux/in.h>
93 #include <linux/errno.h>
94 #include <linux/timer.h>
95 #include <linux/mm.h>
96 #include <linux/inet.h>
97 #include <linux/netdevice.h>
98 #include <linux/slab.h>
99 #include <net/tcp_states.h>
100 #include <linux/skbuff.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <net/net_namespace.h>
104 #include <net/icmp.h>
105 #include <net/route.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include "udp_impl.h"
109
110 struct udp_table udp_table __read_mostly;
111 EXPORT_SYMBOL(udp_table);
112
113 long sysctl_udp_mem[3] __read_mostly;
114 EXPORT_SYMBOL(sysctl_udp_mem);
115
116 int sysctl_udp_rmem_min __read_mostly;
117 EXPORT_SYMBOL(sysctl_udp_rmem_min);
118
119 int sysctl_udp_wmem_min __read_mostly;
120 EXPORT_SYMBOL(sysctl_udp_wmem_min);
121
122 atomic_long_t udp_memory_allocated;
123 EXPORT_SYMBOL(udp_memory_allocated);
124
125 #define MAX_UDP_PORTS 65536
126 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
127
128 static int udp_lib_lport_inuse(struct net *net, __u16 num,
129 const struct udp_hslot *hslot,
130 unsigned long *bitmap,
131 struct sock *sk,
132 int (*saddr_comp)(const struct sock *sk1,
133 const struct sock *sk2),
134 unsigned int log)
135 {
136 struct sock *sk2;
137 struct hlist_nulls_node *node;
138
139 sk_nulls_for_each(sk2, node, &hslot->head)
140 if (net_eq(sock_net(sk2), net) &&
141 sk2 != sk &&
142 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
143 (!sk2->sk_reuse || !sk->sk_reuse) &&
144 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
145 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
146 (*saddr_comp)(sk, sk2)) {
147 if (bitmap)
148 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
149 bitmap);
150 else
151 return 1;
152 }
153 return 0;
154 }
155
156 /*
157 * Note: we still hold spinlock of primary hash chain, so no other writer
158 * can insert/delete a socket with local_port == num
159 */
160 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
161 struct udp_hslot *hslot2,
162 struct sock *sk,
163 int (*saddr_comp)(const struct sock *sk1,
164 const struct sock *sk2))
165 {
166 struct sock *sk2;
167 struct hlist_nulls_node *node;
168 int res = 0;
169
170 spin_lock(&hslot2->lock);
171 udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
172 if (net_eq(sock_net(sk2), net) &&
173 sk2 != sk &&
174 (udp_sk(sk2)->udp_port_hash == num) &&
175 (!sk2->sk_reuse || !sk->sk_reuse) &&
176 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
177 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
178 (*saddr_comp)(sk, sk2)) {
179 res = 1;
180 break;
181 }
182 spin_unlock(&hslot2->lock);
183 return res;
184 }
185
186 /**
187 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
188 *
189 * @sk: socket struct in question
190 * @snum: port number to look up
191 * @saddr_comp: AF-dependent comparison of bound local IP addresses
192 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
193 * with NULL address
194 */
195 int udp_lib_get_port(struct sock *sk, unsigned short snum,
196 int (*saddr_comp)(const struct sock *sk1,
197 const struct sock *sk2),
198 unsigned int hash2_nulladdr)
199 {
200 struct udp_hslot *hslot, *hslot2;
201 struct udp_table *udptable = sk->sk_prot->h.udp_table;
202 int error = 1;
203 struct net *net = sock_net(sk);
204
205 if (!snum) {
206 int low, high, remaining;
207 unsigned rand;
208 unsigned short first, last;
209 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
210
211 inet_get_local_port_range(&low, &high);
212 remaining = (high - low) + 1;
213
214 rand = net_random();
215 first = (((u64)rand * remaining) >> 32) + low;
216 /*
217 * force rand to be an odd multiple of UDP_HTABLE_SIZE
218 */
219 rand = (rand | 1) * (udptable->mask + 1);
220 last = first + udptable->mask + 1;
221 do {
222 hslot = udp_hashslot(udptable, net, first);
223 bitmap_zero(bitmap, PORTS_PER_CHAIN);
224 spin_lock_bh(&hslot->lock);
225 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
226 saddr_comp, udptable->log);
227
228 snum = first;
229 /*
230 * Iterate on all possible values of snum for this hash.
231 * Using steps of an odd multiple of UDP_HTABLE_SIZE
232 * give us randomization and full range coverage.
233 */
234 do {
235 if (low <= snum && snum <= high &&
236 !test_bit(snum >> udptable->log, bitmap) &&
237 !inet_is_reserved_local_port(snum))
238 goto found;
239 snum += rand;
240 } while (snum != first);
241 spin_unlock_bh(&hslot->lock);
242 } while (++first != last);
243 goto fail;
244 } else {
245 hslot = udp_hashslot(udptable, net, snum);
246 spin_lock_bh(&hslot->lock);
247 if (hslot->count > 10) {
248 int exist;
249 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
250
251 slot2 &= udptable->mask;
252 hash2_nulladdr &= udptable->mask;
253
254 hslot2 = udp_hashslot2(udptable, slot2);
255 if (hslot->count < hslot2->count)
256 goto scan_primary_hash;
257
258 exist = udp_lib_lport_inuse2(net, snum, hslot2,
259 sk, saddr_comp);
260 if (!exist && (hash2_nulladdr != slot2)) {
261 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
262 exist = udp_lib_lport_inuse2(net, snum, hslot2,
263 sk, saddr_comp);
264 }
265 if (exist)
266 goto fail_unlock;
267 else
268 goto found;
269 }
270 scan_primary_hash:
271 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
272 saddr_comp, 0))
273 goto fail_unlock;
274 }
275 found:
276 inet_sk(sk)->inet_num = snum;
277 udp_sk(sk)->udp_port_hash = snum;
278 udp_sk(sk)->udp_portaddr_hash ^= snum;
279 if (sk_unhashed(sk)) {
280 sk_nulls_add_node_rcu(sk, &hslot->head);
281 hslot->count++;
282 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
283
284 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
285 spin_lock(&hslot2->lock);
286 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
287 &hslot2->head);
288 hslot2->count++;
289 spin_unlock(&hslot2->lock);
290 }
291 error = 0;
292 fail_unlock:
293 spin_unlock_bh(&hslot->lock);
294 fail:
295 return error;
296 }
297 EXPORT_SYMBOL(udp_lib_get_port);
298
299 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
300 {
301 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
302
303 return (!ipv6_only_sock(sk2) &&
304 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
305 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
306 }
307
308 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
309 unsigned int port)
310 {
311 return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
312 }
313
314 int udp_v4_get_port(struct sock *sk, unsigned short snum)
315 {
316 unsigned int hash2_nulladdr =
317 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
318 unsigned int hash2_partial =
319 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
320
321 /* precompute partial secondary hash */
322 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
323 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
324 }
325
326 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
327 unsigned short hnum,
328 __be16 sport, __be32 daddr, __be16 dport, int dif)
329 {
330 int score = -1;
331
332 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
333 !ipv6_only_sock(sk)) {
334 struct inet_sock *inet = inet_sk(sk);
335
336 score = (sk->sk_family == PF_INET ? 1 : 0);
337 if (inet->inet_rcv_saddr) {
338 if (inet->inet_rcv_saddr != daddr)
339 return -1;
340 score += 2;
341 }
342 if (inet->inet_daddr) {
343 if (inet->inet_daddr != saddr)
344 return -1;
345 score += 2;
346 }
347 if (inet->inet_dport) {
348 if (inet->inet_dport != sport)
349 return -1;
350 score += 2;
351 }
352 if (sk->sk_bound_dev_if) {
353 if (sk->sk_bound_dev_if != dif)
354 return -1;
355 score += 2;
356 }
357 }
358 return score;
359 }
360
361 /*
362 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
363 */
364 #define SCORE2_MAX (1 + 2 + 2 + 2)
365 static inline int compute_score2(struct sock *sk, struct net *net,
366 __be32 saddr, __be16 sport,
367 __be32 daddr, unsigned int hnum, int dif)
368 {
369 int score = -1;
370
371 if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
372 struct inet_sock *inet = inet_sk(sk);
373
374 if (inet->inet_rcv_saddr != daddr)
375 return -1;
376 if (inet->inet_num != hnum)
377 return -1;
378
379 score = (sk->sk_family == PF_INET ? 1 : 0);
380 if (inet->inet_daddr) {
381 if (inet->inet_daddr != saddr)
382 return -1;
383 score += 2;
384 }
385 if (inet->inet_dport) {
386 if (inet->inet_dport != sport)
387 return -1;
388 score += 2;
389 }
390 if (sk->sk_bound_dev_if) {
391 if (sk->sk_bound_dev_if != dif)
392 return -1;
393 score += 2;
394 }
395 }
396 return score;
397 }
398
399
400 /* called with read_rcu_lock() */
401 static struct sock *udp4_lib_lookup2(struct net *net,
402 __be32 saddr, __be16 sport,
403 __be32 daddr, unsigned int hnum, int dif,
404 struct udp_hslot *hslot2, unsigned int slot2)
405 {
406 struct sock *sk, *result;
407 struct hlist_nulls_node *node;
408 int score, badness;
409
410 begin:
411 result = NULL;
412 badness = -1;
413 udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
414 score = compute_score2(sk, net, saddr, sport,
415 daddr, hnum, dif);
416 if (score > badness) {
417 result = sk;
418 badness = score;
419 if (score == SCORE2_MAX)
420 goto exact_match;
421 }
422 }
423 /*
424 * if the nulls value we got at the end of this lookup is
425 * not the expected one, we must restart lookup.
426 * We probably met an item that was moved to another chain.
427 */
428 if (get_nulls_value(node) != slot2)
429 goto begin;
430
431 if (result) {
432 exact_match:
433 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
434 result = NULL;
435 else if (unlikely(compute_score2(result, net, saddr, sport,
436 daddr, hnum, dif) < badness)) {
437 sock_put(result);
438 goto begin;
439 }
440 }
441 return result;
442 }
443
444 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
445 * harder than this. -DaveM
446 */
447 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
448 __be16 sport, __be32 daddr, __be16 dport,
449 int dif, struct udp_table *udptable)
450 {
451 struct sock *sk, *result;
452 struct hlist_nulls_node *node;
453 unsigned short hnum = ntohs(dport);
454 unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
455 struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
456 int score, badness;
457
458 rcu_read_lock();
459 if (hslot->count > 10) {
460 hash2 = udp4_portaddr_hash(net, daddr, hnum);
461 slot2 = hash2 & udptable->mask;
462 hslot2 = &udptable->hash2[slot2];
463 if (hslot->count < hslot2->count)
464 goto begin;
465
466 result = udp4_lib_lookup2(net, saddr, sport,
467 daddr, hnum, dif,
468 hslot2, slot2);
469 if (!result) {
470 hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
471 slot2 = hash2 & udptable->mask;
472 hslot2 = &udptable->hash2[slot2];
473 if (hslot->count < hslot2->count)
474 goto begin;
475
476 result = udp4_lib_lookup2(net, saddr, sport,
477 htonl(INADDR_ANY), hnum, dif,
478 hslot2, slot2);
479 }
480 rcu_read_unlock();
481 return result;
482 }
483 begin:
484 result = NULL;
485 badness = -1;
486 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
487 score = compute_score(sk, net, saddr, hnum, sport,
488 daddr, dport, dif);
489 if (score > badness) {
490 result = sk;
491 badness = score;
492 }
493 }
494 /*
495 * if the nulls value we got at the end of this lookup is
496 * not the expected one, we must restart lookup.
497 * We probably met an item that was moved to another chain.
498 */
499 if (get_nulls_value(node) != slot)
500 goto begin;
501
502 if (result) {
503 if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
504 result = NULL;
505 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
506 daddr, dport, dif) < badness)) {
507 sock_put(result);
508 goto begin;
509 }
510 }
511 rcu_read_unlock();
512 return result;
513 }
514
515 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
516 __be16 sport, __be16 dport,
517 struct udp_table *udptable)
518 {
519 struct sock *sk;
520 const struct iphdr *iph = ip_hdr(skb);
521
522 if (unlikely(sk = skb_steal_sock(skb)))
523 return sk;
524 else
525 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
526 iph->daddr, dport, inet_iif(skb),
527 udptable);
528 }
529
530 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
531 __be32 daddr, __be16 dport, int dif)
532 {
533 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
534 }
535 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
536
537 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
538 __be16 loc_port, __be32 loc_addr,
539 __be16 rmt_port, __be32 rmt_addr,
540 int dif)
541 {
542 struct hlist_nulls_node *node;
543 struct sock *s = sk;
544 unsigned short hnum = ntohs(loc_port);
545
546 sk_nulls_for_each_from(s, node) {
547 struct inet_sock *inet = inet_sk(s);
548
549 if (!net_eq(sock_net(s), net) ||
550 udp_sk(s)->udp_port_hash != hnum ||
551 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
552 (inet->inet_dport != rmt_port && inet->inet_dport) ||
553 (inet->inet_rcv_saddr &&
554 inet->inet_rcv_saddr != loc_addr) ||
555 ipv6_only_sock(s) ||
556 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
557 continue;
558 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
559 continue;
560 goto found;
561 }
562 s = NULL;
563 found:
564 return s;
565 }
566
567 /*
568 * This routine is called by the ICMP module when it gets some
569 * sort of error condition. If err < 0 then the socket should
570 * be closed and the error returned to the user. If err > 0
571 * it's just the icmp type << 8 | icmp code.
572 * Header points to the ip header of the error packet. We move
573 * on past this. Then (as it used to claim before adjustment)
574 * header points to the first 8 bytes of the udp header. We need
575 * to find the appropriate port.
576 */
577
578 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
579 {
580 struct inet_sock *inet;
581 const struct iphdr *iph = (const struct iphdr *)skb->data;
582 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
583 const int type = icmp_hdr(skb)->type;
584 const int code = icmp_hdr(skb)->code;
585 struct sock *sk;
586 int harderr;
587 int err;
588 struct net *net = dev_net(skb->dev);
589
590 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
591 iph->saddr, uh->source, skb->dev->ifindex, udptable);
592 if (sk == NULL) {
593 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
594 return; /* No socket for error */
595 }
596
597 err = 0;
598 harderr = 0;
599 inet = inet_sk(sk);
600
601 switch (type) {
602 default:
603 case ICMP_TIME_EXCEEDED:
604 err = EHOSTUNREACH;
605 break;
606 case ICMP_SOURCE_QUENCH:
607 goto out;
608 case ICMP_PARAMETERPROB:
609 err = EPROTO;
610 harderr = 1;
611 break;
612 case ICMP_DEST_UNREACH:
613 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
614 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
615 err = EMSGSIZE;
616 harderr = 1;
617 break;
618 }
619 goto out;
620 }
621 err = EHOSTUNREACH;
622 if (code <= NR_ICMP_UNREACH) {
623 harderr = icmp_err_convert[code].fatal;
624 err = icmp_err_convert[code].errno;
625 }
626 break;
627 }
628
629 /*
630 * RFC1122: OK. Passes ICMP errors back to application, as per
631 * 4.1.3.3.
632 */
633 if (!inet->recverr) {
634 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
635 goto out;
636 } else
637 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
638
639 sk->sk_err = err;
640 sk->sk_error_report(sk);
641 out:
642 sock_put(sk);
643 }
644
645 void udp_err(struct sk_buff *skb, u32 info)
646 {
647 __udp4_lib_err(skb, info, &udp_table);
648 }
649
650 /*
651 * Throw away all pending data and cancel the corking. Socket is locked.
652 */
653 void udp_flush_pending_frames(struct sock *sk)
654 {
655 struct udp_sock *up = udp_sk(sk);
656
657 if (up->pending) {
658 up->len = 0;
659 up->pending = 0;
660 ip_flush_pending_frames(sk);
661 }
662 }
663 EXPORT_SYMBOL(udp_flush_pending_frames);
664
665 /**
666 * udp4_hwcsum - handle outgoing HW checksumming
667 * @skb: sk_buff containing the filled-in UDP header
668 * (checksum field must be zeroed out)
669 * @src: source IP address
670 * @dst: destination IP address
671 */
672 static void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
673 {
674 struct udphdr *uh = udp_hdr(skb);
675 struct sk_buff *frags = skb_shinfo(skb)->frag_list;
676 int offset = skb_transport_offset(skb);
677 int len = skb->len - offset;
678 int hlen = len;
679 __wsum csum = 0;
680
681 if (!frags) {
682 /*
683 * Only one fragment on the socket.
684 */
685 skb->csum_start = skb_transport_header(skb) - skb->head;
686 skb->csum_offset = offsetof(struct udphdr, check);
687 uh->check = ~csum_tcpudp_magic(src, dst, len,
688 IPPROTO_UDP, 0);
689 } else {
690 /*
691 * HW-checksum won't work as there are two or more
692 * fragments on the socket so that all csums of sk_buffs
693 * should be together
694 */
695 do {
696 csum = csum_add(csum, frags->csum);
697 hlen -= frags->len;
698 } while ((frags = frags->next));
699
700 csum = skb_checksum(skb, offset, hlen, csum);
701 skb->ip_summed = CHECKSUM_NONE;
702
703 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
704 if (uh->check == 0)
705 uh->check = CSUM_MANGLED_0;
706 }
707 }
708
709 static int udp_send_skb(struct sk_buff *skb, __be32 daddr, __be32 dport)
710 {
711 struct sock *sk = skb->sk;
712 struct inet_sock *inet = inet_sk(sk);
713 struct udphdr *uh;
714 struct rtable *rt = (struct rtable *)skb_dst(skb);
715 int err = 0;
716 int is_udplite = IS_UDPLITE(sk);
717 int offset = skb_transport_offset(skb);
718 int len = skb->len - offset;
719 __wsum csum = 0;
720
721 /*
722 * Create a UDP header
723 */
724 uh = udp_hdr(skb);
725 uh->source = inet->inet_sport;
726 uh->dest = dport;
727 uh->len = htons(len);
728 uh->check = 0;
729
730 if (is_udplite) /* UDP-Lite */
731 csum = udplite_csum(skb);
732
733 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
734
735 skb->ip_summed = CHECKSUM_NONE;
736 goto send;
737
738 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
739
740 udp4_hwcsum(skb, rt->rt_src, daddr);
741 goto send;
742
743 } else
744 csum = udp_csum(skb);
745
746 /* add protocol-dependent pseudo-header */
747 uh->check = csum_tcpudp_magic(rt->rt_src, daddr, len,
748 sk->sk_protocol, csum);
749 if (uh->check == 0)
750 uh->check = CSUM_MANGLED_0;
751
752 send:
753 err = ip_send_skb(skb);
754 if (err) {
755 if (err == -ENOBUFS && !inet->recverr) {
756 UDP_INC_STATS_USER(sock_net(sk),
757 UDP_MIB_SNDBUFERRORS, is_udplite);
758 err = 0;
759 }
760 } else
761 UDP_INC_STATS_USER(sock_net(sk),
762 UDP_MIB_OUTDATAGRAMS, is_udplite);
763 return err;
764 }
765
766 /*
767 * Push out all pending data as one UDP datagram. Socket is locked.
768 */
769 static int udp_push_pending_frames(struct sock *sk)
770 {
771 struct udp_sock *up = udp_sk(sk);
772 struct inet_sock *inet = inet_sk(sk);
773 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
774 struct sk_buff *skb;
775 int err = 0;
776
777 skb = ip_finish_skb(sk, fl4);
778 if (!skb)
779 goto out;
780
781 err = udp_send_skb(skb, fl4->daddr, fl4->fl4_dport);
782
783 out:
784 up->len = 0;
785 up->pending = 0;
786 return err;
787 }
788
789 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
790 size_t len)
791 {
792 struct inet_sock *inet = inet_sk(sk);
793 struct udp_sock *up = udp_sk(sk);
794 struct flowi4 fl4_stack;
795 struct flowi4 *fl4;
796 int ulen = len;
797 struct ipcm_cookie ipc;
798 struct rtable *rt = NULL;
799 int free = 0;
800 int connected = 0;
801 __be32 daddr, faddr, saddr;
802 __be16 dport;
803 u8 tos;
804 int err, is_udplite = IS_UDPLITE(sk);
805 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
806 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
807 struct sk_buff *skb;
808 struct ip_options_data opt_copy;
809
810 if (len > 0xFFFF)
811 return -EMSGSIZE;
812
813 /*
814 * Check the flags.
815 */
816
817 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
818 return -EOPNOTSUPP;
819
820 ipc.opt = NULL;
821 ipc.tx_flags = 0;
822
823 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
824
825 if (up->pending) {
826 /*
827 * There are pending frames.
828 * The socket lock must be held while it's corked.
829 */
830 lock_sock(sk);
831 if (likely(up->pending)) {
832 if (unlikely(up->pending != AF_INET)) {
833 release_sock(sk);
834 return -EINVAL;
835 }
836 goto do_append_data;
837 }
838 release_sock(sk);
839 }
840 ulen += sizeof(struct udphdr);
841
842 /*
843 * Get and verify the address.
844 */
845 if (msg->msg_name) {
846 struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name;
847 if (msg->msg_namelen < sizeof(*usin))
848 return -EINVAL;
849 if (usin->sin_family != AF_INET) {
850 if (usin->sin_family != AF_UNSPEC)
851 return -EAFNOSUPPORT;
852 }
853
854 daddr = usin->sin_addr.s_addr;
855 dport = usin->sin_port;
856 if (dport == 0)
857 return -EINVAL;
858 } else {
859 if (sk->sk_state != TCP_ESTABLISHED)
860 return -EDESTADDRREQ;
861 daddr = inet->inet_daddr;
862 dport = inet->inet_dport;
863 /* Open fast path for connected socket.
864 Route will not be used, if at least one option is set.
865 */
866 connected = 1;
867 }
868 ipc.addr = inet->inet_saddr;
869
870 ipc.oif = sk->sk_bound_dev_if;
871 err = sock_tx_timestamp(sk, &ipc.tx_flags);
872 if (err)
873 return err;
874 if (msg->msg_controllen) {
875 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
876 if (err)
877 return err;
878 if (ipc.opt)
879 free = 1;
880 connected = 0;
881 }
882 if (!ipc.opt) {
883 struct ip_options_rcu *inet_opt;
884
885 rcu_read_lock();
886 inet_opt = rcu_dereference(inet->inet_opt);
887 if (inet_opt) {
888 memcpy(&opt_copy, inet_opt,
889 sizeof(*inet_opt) + inet_opt->opt.optlen);
890 ipc.opt = &opt_copy.opt;
891 }
892 rcu_read_unlock();
893 }
894
895 saddr = ipc.addr;
896 ipc.addr = faddr = daddr;
897
898 if (ipc.opt && ipc.opt->opt.srr) {
899 if (!daddr)
900 return -EINVAL;
901 faddr = ipc.opt->opt.faddr;
902 connected = 0;
903 }
904 tos = RT_TOS(inet->tos);
905 if (sock_flag(sk, SOCK_LOCALROUTE) ||
906 (msg->msg_flags & MSG_DONTROUTE) ||
907 (ipc.opt && ipc.opt->opt.is_strictroute)) {
908 tos |= RTO_ONLINK;
909 connected = 0;
910 }
911
912 if (ipv4_is_multicast(daddr)) {
913 if (!ipc.oif)
914 ipc.oif = inet->mc_index;
915 if (!saddr)
916 saddr = inet->mc_addr;
917 connected = 0;
918 }
919
920 if (connected)
921 rt = (struct rtable *)sk_dst_check(sk, 0);
922
923 fl4 = &inet->cork.fl.u.ip4;
924 if (rt == NULL) {
925 struct net *net = sock_net(sk);
926
927 fl4 = &fl4_stack;
928 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
929 RT_SCOPE_UNIVERSE, sk->sk_protocol,
930 inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP,
931 faddr, saddr, dport, inet->inet_sport);
932
933 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
934 rt = ip_route_output_flow(net, fl4, sk);
935 if (IS_ERR(rt)) {
936 err = PTR_ERR(rt);
937 rt = NULL;
938 if (err == -ENETUNREACH)
939 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
940 goto out;
941 }
942
943 err = -EACCES;
944 if ((rt->rt_flags & RTCF_BROADCAST) &&
945 !sock_flag(sk, SOCK_BROADCAST))
946 goto out;
947 if (connected)
948 sk_dst_set(sk, dst_clone(&rt->dst));
949 }
950
951 if (msg->msg_flags&MSG_CONFIRM)
952 goto do_confirm;
953 back_from_confirm:
954
955 saddr = fl4->saddr;
956 if (!ipc.addr)
957 daddr = ipc.addr = fl4->daddr;
958
959 /* Lockless fast path for the non-corking case. */
960 if (!corkreq) {
961 skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen,
962 sizeof(struct udphdr), &ipc, &rt,
963 msg->msg_flags);
964 err = PTR_ERR(skb);
965 if (skb && !IS_ERR(skb))
966 err = udp_send_skb(skb, daddr, dport);
967 goto out;
968 }
969
970 lock_sock(sk);
971 if (unlikely(up->pending)) {
972 /* The socket is already corked while preparing it. */
973 /* ... which is an evident application bug. --ANK */
974 release_sock(sk);
975
976 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
977 err = -EINVAL;
978 goto out;
979 }
980 /*
981 * Now cork the socket to pend data.
982 */
983 fl4 = &inet->cork.fl.u.ip4;
984 fl4->daddr = daddr;
985 fl4->saddr = saddr;
986 fl4->fl4_dport = dport;
987 fl4->fl4_sport = inet->inet_sport;
988 up->pending = AF_INET;
989
990 do_append_data:
991 up->len += ulen;
992 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
993 sizeof(struct udphdr), &ipc, &rt,
994 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
995 if (err)
996 udp_flush_pending_frames(sk);
997 else if (!corkreq)
998 err = udp_push_pending_frames(sk);
999 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1000 up->pending = 0;
1001 release_sock(sk);
1002
1003 out:
1004 ip_rt_put(rt);
1005 if (free)
1006 kfree(ipc.opt);
1007 if (!err)
1008 return len;
1009 /*
1010 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1011 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1012 * we don't have a good statistic (IpOutDiscards but it can be too many
1013 * things). We could add another new stat but at least for now that
1014 * seems like overkill.
1015 */
1016 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1017 UDP_INC_STATS_USER(sock_net(sk),
1018 UDP_MIB_SNDBUFERRORS, is_udplite);
1019 }
1020 return err;
1021
1022 do_confirm:
1023 dst_confirm(&rt->dst);
1024 if (!(msg->msg_flags&MSG_PROBE) || len)
1025 goto back_from_confirm;
1026 err = 0;
1027 goto out;
1028 }
1029 EXPORT_SYMBOL(udp_sendmsg);
1030
1031 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1032 size_t size, int flags)
1033 {
1034 struct udp_sock *up = udp_sk(sk);
1035 int ret;
1036
1037 if (!up->pending) {
1038 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1039
1040 /* Call udp_sendmsg to specify destination address which
1041 * sendpage interface can't pass.
1042 * This will succeed only when the socket is connected.
1043 */
1044 ret = udp_sendmsg(NULL, sk, &msg, 0);
1045 if (ret < 0)
1046 return ret;
1047 }
1048
1049 lock_sock(sk);
1050
1051 if (unlikely(!up->pending)) {
1052 release_sock(sk);
1053
1054 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
1055 return -EINVAL;
1056 }
1057
1058 ret = ip_append_page(sk, page, offset, size, flags);
1059 if (ret == -EOPNOTSUPP) {
1060 release_sock(sk);
1061 return sock_no_sendpage(sk->sk_socket, page, offset,
1062 size, flags);
1063 }
1064 if (ret < 0) {
1065 udp_flush_pending_frames(sk);
1066 goto out;
1067 }
1068
1069 up->len += size;
1070 if (!(up->corkflag || (flags&MSG_MORE)))
1071 ret = udp_push_pending_frames(sk);
1072 if (!ret)
1073 ret = size;
1074 out:
1075 release_sock(sk);
1076 return ret;
1077 }
1078
1079
1080 /**
1081 * first_packet_length - return length of first packet in receive queue
1082 * @sk: socket
1083 *
1084 * Drops all bad checksum frames, until a valid one is found.
1085 * Returns the length of found skb, or 0 if none is found.
1086 */
1087 static unsigned int first_packet_length(struct sock *sk)
1088 {
1089 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1090 struct sk_buff *skb;
1091 unsigned int res;
1092
1093 __skb_queue_head_init(&list_kill);
1094
1095 spin_lock_bh(&rcvq->lock);
1096 while ((skb = skb_peek(rcvq)) != NULL &&
1097 udp_lib_checksum_complete(skb)) {
1098 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1099 IS_UDPLITE(sk));
1100 atomic_inc(&sk->sk_drops);
1101 __skb_unlink(skb, rcvq);
1102 __skb_queue_tail(&list_kill, skb);
1103 }
1104 res = skb ? skb->len : 0;
1105 spin_unlock_bh(&rcvq->lock);
1106
1107 if (!skb_queue_empty(&list_kill)) {
1108 bool slow = lock_sock_fast(sk);
1109
1110 __skb_queue_purge(&list_kill);
1111 sk_mem_reclaim_partial(sk);
1112 unlock_sock_fast(sk, slow);
1113 }
1114 return res;
1115 }
1116
1117 /*
1118 * IOCTL requests applicable to the UDP protocol
1119 */
1120
1121 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1122 {
1123 switch (cmd) {
1124 case SIOCOUTQ:
1125 {
1126 int amount = sk_wmem_alloc_get(sk);
1127
1128 return put_user(amount, (int __user *)arg);
1129 }
1130
1131 case SIOCINQ:
1132 {
1133 unsigned int amount = first_packet_length(sk);
1134
1135 if (amount)
1136 /*
1137 * We will only return the amount
1138 * of this packet since that is all
1139 * that will be read.
1140 */
1141 amount -= sizeof(struct udphdr);
1142
1143 return put_user(amount, (int __user *)arg);
1144 }
1145
1146 default:
1147 return -ENOIOCTLCMD;
1148 }
1149
1150 return 0;
1151 }
1152 EXPORT_SYMBOL(udp_ioctl);
1153
1154 /*
1155 * This should be easy, if there is something there we
1156 * return it, otherwise we block.
1157 */
1158
1159 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1160 size_t len, int noblock, int flags, int *addr_len)
1161 {
1162 struct inet_sock *inet = inet_sk(sk);
1163 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1164 struct sk_buff *skb;
1165 unsigned int ulen;
1166 int peeked;
1167 int err;
1168 int is_udplite = IS_UDPLITE(sk);
1169 bool slow;
1170
1171 /*
1172 * Check any passed addresses
1173 */
1174 if (addr_len)
1175 *addr_len = sizeof(*sin);
1176
1177 if (flags & MSG_ERRQUEUE)
1178 return ip_recv_error(sk, msg, len);
1179
1180 try_again:
1181 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1182 &peeked, &err);
1183 if (!skb)
1184 goto out;
1185
1186 ulen = skb->len - sizeof(struct udphdr);
1187 if (len > ulen)
1188 len = ulen;
1189 else if (len < ulen)
1190 msg->msg_flags |= MSG_TRUNC;
1191
1192 /*
1193 * If checksum is needed at all, try to do it while copying the
1194 * data. If the data is truncated, or if we only want a partial
1195 * coverage checksum (UDP-Lite), do it before the copy.
1196 */
1197
1198 if (len < ulen || UDP_SKB_CB(skb)->partial_cov) {
1199 if (udp_lib_checksum_complete(skb))
1200 goto csum_copy_err;
1201 }
1202
1203 if (skb_csum_unnecessary(skb))
1204 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1205 msg->msg_iov, len);
1206 else {
1207 err = skb_copy_and_csum_datagram_iovec(skb,
1208 sizeof(struct udphdr),
1209 msg->msg_iov);
1210
1211 if (err == -EINVAL)
1212 goto csum_copy_err;
1213 }
1214
1215 if (err)
1216 goto out_free;
1217
1218 if (!peeked)
1219 UDP_INC_STATS_USER(sock_net(sk),
1220 UDP_MIB_INDATAGRAMS, is_udplite);
1221
1222 sock_recv_ts_and_drops(msg, sk, skb);
1223
1224 /* Copy the address. */
1225 if (sin) {
1226 sin->sin_family = AF_INET;
1227 sin->sin_port = udp_hdr(skb)->source;
1228 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1229 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1230 }
1231 if (inet->cmsg_flags)
1232 ip_cmsg_recv(msg, skb);
1233
1234 err = len;
1235 if (flags & MSG_TRUNC)
1236 err = ulen;
1237
1238 out_free:
1239 skb_free_datagram_locked(sk, skb);
1240 out:
1241 return err;
1242
1243 csum_copy_err:
1244 slow = lock_sock_fast(sk);
1245 if (!skb_kill_datagram(sk, skb, flags))
1246 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1247 unlock_sock_fast(sk, slow);
1248
1249 if (noblock)
1250 return -EAGAIN;
1251 goto try_again;
1252 }
1253
1254
1255 int udp_disconnect(struct sock *sk, int flags)
1256 {
1257 struct inet_sock *inet = inet_sk(sk);
1258 /*
1259 * 1003.1g - break association.
1260 */
1261
1262 sk->sk_state = TCP_CLOSE;
1263 inet->inet_daddr = 0;
1264 inet->inet_dport = 0;
1265 sock_rps_save_rxhash(sk, 0);
1266 sk->sk_bound_dev_if = 0;
1267 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1268 inet_reset_saddr(sk);
1269
1270 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1271 sk->sk_prot->unhash(sk);
1272 inet->inet_sport = 0;
1273 }
1274 sk_dst_reset(sk);
1275 return 0;
1276 }
1277 EXPORT_SYMBOL(udp_disconnect);
1278
1279 void udp_lib_unhash(struct sock *sk)
1280 {
1281 if (sk_hashed(sk)) {
1282 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1283 struct udp_hslot *hslot, *hslot2;
1284
1285 hslot = udp_hashslot(udptable, sock_net(sk),
1286 udp_sk(sk)->udp_port_hash);
1287 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1288
1289 spin_lock_bh(&hslot->lock);
1290 if (sk_nulls_del_node_init_rcu(sk)) {
1291 hslot->count--;
1292 inet_sk(sk)->inet_num = 0;
1293 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1294
1295 spin_lock(&hslot2->lock);
1296 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1297 hslot2->count--;
1298 spin_unlock(&hslot2->lock);
1299 }
1300 spin_unlock_bh(&hslot->lock);
1301 }
1302 }
1303 EXPORT_SYMBOL(udp_lib_unhash);
1304
1305 /*
1306 * inet_rcv_saddr was changed, we must rehash secondary hash
1307 */
1308 void udp_lib_rehash(struct sock *sk, u16 newhash)
1309 {
1310 if (sk_hashed(sk)) {
1311 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1312 struct udp_hslot *hslot, *hslot2, *nhslot2;
1313
1314 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1315 nhslot2 = udp_hashslot2(udptable, newhash);
1316 udp_sk(sk)->udp_portaddr_hash = newhash;
1317 if (hslot2 != nhslot2) {
1318 hslot = udp_hashslot(udptable, sock_net(sk),
1319 udp_sk(sk)->udp_port_hash);
1320 /* we must lock primary chain too */
1321 spin_lock_bh(&hslot->lock);
1322
1323 spin_lock(&hslot2->lock);
1324 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1325 hslot2->count--;
1326 spin_unlock(&hslot2->lock);
1327
1328 spin_lock(&nhslot2->lock);
1329 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1330 &nhslot2->head);
1331 nhslot2->count++;
1332 spin_unlock(&nhslot2->lock);
1333
1334 spin_unlock_bh(&hslot->lock);
1335 }
1336 }
1337 }
1338 EXPORT_SYMBOL(udp_lib_rehash);
1339
1340 static void udp_v4_rehash(struct sock *sk)
1341 {
1342 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1343 inet_sk(sk)->inet_rcv_saddr,
1344 inet_sk(sk)->inet_num);
1345 udp_lib_rehash(sk, new_hash);
1346 }
1347
1348 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1349 {
1350 int rc;
1351
1352 if (inet_sk(sk)->inet_daddr)
1353 sock_rps_save_rxhash(sk, skb->rxhash);
1354
1355 rc = ip_queue_rcv_skb(sk, skb);
1356 if (rc < 0) {
1357 int is_udplite = IS_UDPLITE(sk);
1358
1359 /* Note that an ENOMEM error is charged twice */
1360 if (rc == -ENOMEM)
1361 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1362 is_udplite);
1363 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1364 kfree_skb(skb);
1365 return -1;
1366 }
1367
1368 return 0;
1369
1370 }
1371
1372 /* returns:
1373 * -1: error
1374 * 0: success
1375 * >0: "udp encap" protocol resubmission
1376 *
1377 * Note that in the success and error cases, the skb is assumed to
1378 * have either been requeued or freed.
1379 */
1380 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1381 {
1382 struct udp_sock *up = udp_sk(sk);
1383 int rc;
1384 int is_udplite = IS_UDPLITE(sk);
1385
1386 /*
1387 * Charge it to the socket, dropping if the queue is full.
1388 */
1389 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1390 goto drop;
1391 nf_reset(skb);
1392
1393 if (up->encap_type) {
1394 /*
1395 * This is an encapsulation socket so pass the skb to
1396 * the socket's udp_encap_rcv() hook. Otherwise, just
1397 * fall through and pass this up the UDP socket.
1398 * up->encap_rcv() returns the following value:
1399 * =0 if skb was successfully passed to the encap
1400 * handler or was discarded by it.
1401 * >0 if skb should be passed on to UDP.
1402 * <0 if skb should be resubmitted as proto -N
1403 */
1404
1405 /* if we're overly short, let UDP handle it */
1406 if (skb->len > sizeof(struct udphdr) &&
1407 up->encap_rcv != NULL) {
1408 int ret;
1409
1410 ret = (*up->encap_rcv)(sk, skb);
1411 if (ret <= 0) {
1412 UDP_INC_STATS_BH(sock_net(sk),
1413 UDP_MIB_INDATAGRAMS,
1414 is_udplite);
1415 return -ret;
1416 }
1417 }
1418
1419 /* FALLTHROUGH -- it's a UDP Packet */
1420 }
1421
1422 /*
1423 * UDP-Lite specific tests, ignored on UDP sockets
1424 */
1425 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1426
1427 /*
1428 * MIB statistics other than incrementing the error count are
1429 * disabled for the following two types of errors: these depend
1430 * on the application settings, not on the functioning of the
1431 * protocol stack as such.
1432 *
1433 * RFC 3828 here recommends (sec 3.3): "There should also be a
1434 * way ... to ... at least let the receiving application block
1435 * delivery of packets with coverage values less than a value
1436 * provided by the application."
1437 */
1438 if (up->pcrlen == 0) { /* full coverage was set */
1439 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1440 "%d while full coverage %d requested\n",
1441 UDP_SKB_CB(skb)->cscov, skb->len);
1442 goto drop;
1443 }
1444 /* The next case involves violating the min. coverage requested
1445 * by the receiver. This is subtle: if receiver wants x and x is
1446 * greater than the buffersize/MTU then receiver will complain
1447 * that it wants x while sender emits packets of smaller size y.
1448 * Therefore the above ...()->partial_cov statement is essential.
1449 */
1450 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1451 LIMIT_NETDEBUG(KERN_WARNING
1452 "UDPLITE: coverage %d too small, need min %d\n",
1453 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1454 goto drop;
1455 }
1456 }
1457
1458 if (rcu_dereference_raw(sk->sk_filter)) {
1459 if (udp_lib_checksum_complete(skb))
1460 goto drop;
1461 }
1462
1463
1464 if (sk_rcvqueues_full(sk, skb))
1465 goto drop;
1466
1467 rc = 0;
1468
1469 bh_lock_sock(sk);
1470 if (!sock_owned_by_user(sk))
1471 rc = __udp_queue_rcv_skb(sk, skb);
1472 else if (sk_add_backlog(sk, skb)) {
1473 bh_unlock_sock(sk);
1474 goto drop;
1475 }
1476 bh_unlock_sock(sk);
1477
1478 return rc;
1479
1480 drop:
1481 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1482 atomic_inc(&sk->sk_drops);
1483 kfree_skb(skb);
1484 return -1;
1485 }
1486
1487
1488 static void flush_stack(struct sock **stack, unsigned int count,
1489 struct sk_buff *skb, unsigned int final)
1490 {
1491 unsigned int i;
1492 struct sk_buff *skb1 = NULL;
1493 struct sock *sk;
1494
1495 for (i = 0; i < count; i++) {
1496 sk = stack[i];
1497 if (likely(skb1 == NULL))
1498 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1499
1500 if (!skb1) {
1501 atomic_inc(&sk->sk_drops);
1502 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1503 IS_UDPLITE(sk));
1504 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1505 IS_UDPLITE(sk));
1506 }
1507
1508 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1509 skb1 = NULL;
1510 }
1511 if (unlikely(skb1))
1512 kfree_skb(skb1);
1513 }
1514
1515 /*
1516 * Multicasts and broadcasts go to each listener.
1517 *
1518 * Note: called only from the BH handler context.
1519 */
1520 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1521 struct udphdr *uh,
1522 __be32 saddr, __be32 daddr,
1523 struct udp_table *udptable)
1524 {
1525 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1526 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1527 int dif;
1528 unsigned int i, count = 0;
1529
1530 spin_lock(&hslot->lock);
1531 sk = sk_nulls_head(&hslot->head);
1532 dif = skb->dev->ifindex;
1533 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1534 while (sk) {
1535 stack[count++] = sk;
1536 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1537 daddr, uh->source, saddr, dif);
1538 if (unlikely(count == ARRAY_SIZE(stack))) {
1539 if (!sk)
1540 break;
1541 flush_stack(stack, count, skb, ~0);
1542 count = 0;
1543 }
1544 }
1545 /*
1546 * before releasing chain lock, we must take a reference on sockets
1547 */
1548 for (i = 0; i < count; i++)
1549 sock_hold(stack[i]);
1550
1551 spin_unlock(&hslot->lock);
1552
1553 /*
1554 * do the slow work with no lock held
1555 */
1556 if (count) {
1557 flush_stack(stack, count, skb, count - 1);
1558
1559 for (i = 0; i < count; i++)
1560 sock_put(stack[i]);
1561 } else {
1562 kfree_skb(skb);
1563 }
1564 return 0;
1565 }
1566
1567 /* Initialize UDP checksum. If exited with zero value (success),
1568 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1569 * Otherwise, csum completion requires chacksumming packet body,
1570 * including udp header and folding it to skb->csum.
1571 */
1572 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1573 int proto)
1574 {
1575 const struct iphdr *iph;
1576 int err;
1577
1578 UDP_SKB_CB(skb)->partial_cov = 0;
1579 UDP_SKB_CB(skb)->cscov = skb->len;
1580
1581 if (proto == IPPROTO_UDPLITE) {
1582 err = udplite_checksum_init(skb, uh);
1583 if (err)
1584 return err;
1585 }
1586
1587 iph = ip_hdr(skb);
1588 if (uh->check == 0) {
1589 skb->ip_summed = CHECKSUM_UNNECESSARY;
1590 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1591 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1592 proto, skb->csum))
1593 skb->ip_summed = CHECKSUM_UNNECESSARY;
1594 }
1595 if (!skb_csum_unnecessary(skb))
1596 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1597 skb->len, proto, 0);
1598 /* Probably, we should checksum udp header (it should be in cache
1599 * in any case) and data in tiny packets (< rx copybreak).
1600 */
1601
1602 return 0;
1603 }
1604
1605 /*
1606 * All we need to do is get the socket, and then do a checksum.
1607 */
1608
1609 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1610 int proto)
1611 {
1612 struct sock *sk;
1613 struct udphdr *uh;
1614 unsigned short ulen;
1615 struct rtable *rt = skb_rtable(skb);
1616 __be32 saddr, daddr;
1617 struct net *net = dev_net(skb->dev);
1618
1619 /*
1620 * Validate the packet.
1621 */
1622 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1623 goto drop; /* No space for header. */
1624
1625 uh = udp_hdr(skb);
1626 ulen = ntohs(uh->len);
1627 saddr = ip_hdr(skb)->saddr;
1628 daddr = ip_hdr(skb)->daddr;
1629
1630 if (ulen > skb->len)
1631 goto short_packet;
1632
1633 if (proto == IPPROTO_UDP) {
1634 /* UDP validates ulen. */
1635 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1636 goto short_packet;
1637 uh = udp_hdr(skb);
1638 }
1639
1640 if (udp4_csum_init(skb, uh, proto))
1641 goto csum_error;
1642
1643 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1644 return __udp4_lib_mcast_deliver(net, skb, uh,
1645 saddr, daddr, udptable);
1646
1647 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1648
1649 if (sk != NULL) {
1650 int ret = udp_queue_rcv_skb(sk, skb);
1651 sock_put(sk);
1652
1653 /* a return value > 0 means to resubmit the input, but
1654 * it wants the return to be -protocol, or 0
1655 */
1656 if (ret > 0)
1657 return -ret;
1658 return 0;
1659 }
1660
1661 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1662 goto drop;
1663 nf_reset(skb);
1664
1665 /* No socket. Drop packet silently, if checksum is wrong */
1666 if (udp_lib_checksum_complete(skb))
1667 goto csum_error;
1668
1669 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1670 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1671
1672 /*
1673 * Hmm. We got an UDP packet to a port to which we
1674 * don't wanna listen. Ignore it.
1675 */
1676 kfree_skb(skb);
1677 return 0;
1678
1679 short_packet:
1680 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1681 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1682 &saddr,
1683 ntohs(uh->source),
1684 ulen,
1685 skb->len,
1686 &daddr,
1687 ntohs(uh->dest));
1688 goto drop;
1689
1690 csum_error:
1691 /*
1692 * RFC1122: OK. Discards the bad packet silently (as far as
1693 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1694 */
1695 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1696 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1697 &saddr,
1698 ntohs(uh->source),
1699 &daddr,
1700 ntohs(uh->dest),
1701 ulen);
1702 drop:
1703 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1704 kfree_skb(skb);
1705 return 0;
1706 }
1707
1708 int udp_rcv(struct sk_buff *skb)
1709 {
1710 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1711 }
1712
1713 void udp_destroy_sock(struct sock *sk)
1714 {
1715 bool slow = lock_sock_fast(sk);
1716 udp_flush_pending_frames(sk);
1717 unlock_sock_fast(sk, slow);
1718 }
1719
1720 /*
1721 * Socket option code for UDP
1722 */
1723 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1724 char __user *optval, unsigned int optlen,
1725 int (*push_pending_frames)(struct sock *))
1726 {
1727 struct udp_sock *up = udp_sk(sk);
1728 int val;
1729 int err = 0;
1730 int is_udplite = IS_UDPLITE(sk);
1731
1732 if (optlen < sizeof(int))
1733 return -EINVAL;
1734
1735 if (get_user(val, (int __user *)optval))
1736 return -EFAULT;
1737
1738 switch (optname) {
1739 case UDP_CORK:
1740 if (val != 0) {
1741 up->corkflag = 1;
1742 } else {
1743 up->corkflag = 0;
1744 lock_sock(sk);
1745 (*push_pending_frames)(sk);
1746 release_sock(sk);
1747 }
1748 break;
1749
1750 case UDP_ENCAP:
1751 switch (val) {
1752 case 0:
1753 case UDP_ENCAP_ESPINUDP:
1754 case UDP_ENCAP_ESPINUDP_NON_IKE:
1755 up->encap_rcv = xfrm4_udp_encap_rcv;
1756 /* FALLTHROUGH */
1757 case UDP_ENCAP_L2TPINUDP:
1758 up->encap_type = val;
1759 break;
1760 default:
1761 err = -ENOPROTOOPT;
1762 break;
1763 }
1764 break;
1765
1766 /*
1767 * UDP-Lite's partial checksum coverage (RFC 3828).
1768 */
1769 /* The sender sets actual checksum coverage length via this option.
1770 * The case coverage > packet length is handled by send module. */
1771 case UDPLITE_SEND_CSCOV:
1772 if (!is_udplite) /* Disable the option on UDP sockets */
1773 return -ENOPROTOOPT;
1774 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1775 val = 8;
1776 else if (val > USHRT_MAX)
1777 val = USHRT_MAX;
1778 up->pcslen = val;
1779 up->pcflag |= UDPLITE_SEND_CC;
1780 break;
1781
1782 /* The receiver specifies a minimum checksum coverage value. To make
1783 * sense, this should be set to at least 8 (as done below). If zero is
1784 * used, this again means full checksum coverage. */
1785 case UDPLITE_RECV_CSCOV:
1786 if (!is_udplite) /* Disable the option on UDP sockets */
1787 return -ENOPROTOOPT;
1788 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1789 val = 8;
1790 else if (val > USHRT_MAX)
1791 val = USHRT_MAX;
1792 up->pcrlen = val;
1793 up->pcflag |= UDPLITE_RECV_CC;
1794 break;
1795
1796 default:
1797 err = -ENOPROTOOPT;
1798 break;
1799 }
1800
1801 return err;
1802 }
1803 EXPORT_SYMBOL(udp_lib_setsockopt);
1804
1805 int udp_setsockopt(struct sock *sk, int level, int optname,
1806 char __user *optval, unsigned int optlen)
1807 {
1808 if (level == SOL_UDP || level == SOL_UDPLITE)
1809 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1810 udp_push_pending_frames);
1811 return ip_setsockopt(sk, level, optname, optval, optlen);
1812 }
1813
1814 #ifdef CONFIG_COMPAT
1815 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1816 char __user *optval, unsigned int optlen)
1817 {
1818 if (level == SOL_UDP || level == SOL_UDPLITE)
1819 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1820 udp_push_pending_frames);
1821 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1822 }
1823 #endif
1824
1825 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1826 char __user *optval, int __user *optlen)
1827 {
1828 struct udp_sock *up = udp_sk(sk);
1829 int val, len;
1830
1831 if (get_user(len, optlen))
1832 return -EFAULT;
1833
1834 len = min_t(unsigned int, len, sizeof(int));
1835
1836 if (len < 0)
1837 return -EINVAL;
1838
1839 switch (optname) {
1840 case UDP_CORK:
1841 val = up->corkflag;
1842 break;
1843
1844 case UDP_ENCAP:
1845 val = up->encap_type;
1846 break;
1847
1848 /* The following two cannot be changed on UDP sockets, the return is
1849 * always 0 (which corresponds to the full checksum coverage of UDP). */
1850 case UDPLITE_SEND_CSCOV:
1851 val = up->pcslen;
1852 break;
1853
1854 case UDPLITE_RECV_CSCOV:
1855 val = up->pcrlen;
1856 break;
1857
1858 default:
1859 return -ENOPROTOOPT;
1860 }
1861
1862 if (put_user(len, optlen))
1863 return -EFAULT;
1864 if (copy_to_user(optval, &val, len))
1865 return -EFAULT;
1866 return 0;
1867 }
1868 EXPORT_SYMBOL(udp_lib_getsockopt);
1869
1870 int udp_getsockopt(struct sock *sk, int level, int optname,
1871 char __user *optval, int __user *optlen)
1872 {
1873 if (level == SOL_UDP || level == SOL_UDPLITE)
1874 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1875 return ip_getsockopt(sk, level, optname, optval, optlen);
1876 }
1877
1878 #ifdef CONFIG_COMPAT
1879 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1880 char __user *optval, int __user *optlen)
1881 {
1882 if (level == SOL_UDP || level == SOL_UDPLITE)
1883 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1884 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1885 }
1886 #endif
1887 /**
1888 * udp_poll - wait for a UDP event.
1889 * @file - file struct
1890 * @sock - socket
1891 * @wait - poll table
1892 *
1893 * This is same as datagram poll, except for the special case of
1894 * blocking sockets. If application is using a blocking fd
1895 * and a packet with checksum error is in the queue;
1896 * then it could get return from select indicating data available
1897 * but then block when reading it. Add special case code
1898 * to work around these arguably broken applications.
1899 */
1900 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1901 {
1902 unsigned int mask = datagram_poll(file, sock, wait);
1903 struct sock *sk = sock->sk;
1904
1905 /* Check for false positives due to checksum errors */
1906 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1907 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1908 mask &= ~(POLLIN | POLLRDNORM);
1909
1910 return mask;
1911
1912 }
1913 EXPORT_SYMBOL(udp_poll);
1914
1915 struct proto udp_prot = {
1916 .name = "UDP",
1917 .owner = THIS_MODULE,
1918 .close = udp_lib_close,
1919 .connect = ip4_datagram_connect,
1920 .disconnect = udp_disconnect,
1921 .ioctl = udp_ioctl,
1922 .destroy = udp_destroy_sock,
1923 .setsockopt = udp_setsockopt,
1924 .getsockopt = udp_getsockopt,
1925 .sendmsg = udp_sendmsg,
1926 .recvmsg = udp_recvmsg,
1927 .sendpage = udp_sendpage,
1928 .backlog_rcv = __udp_queue_rcv_skb,
1929 .hash = udp_lib_hash,
1930 .unhash = udp_lib_unhash,
1931 .rehash = udp_v4_rehash,
1932 .get_port = udp_v4_get_port,
1933 .memory_allocated = &udp_memory_allocated,
1934 .sysctl_mem = sysctl_udp_mem,
1935 .sysctl_wmem = &sysctl_udp_wmem_min,
1936 .sysctl_rmem = &sysctl_udp_rmem_min,
1937 .obj_size = sizeof(struct udp_sock),
1938 .slab_flags = SLAB_DESTROY_BY_RCU,
1939 .h.udp_table = &udp_table,
1940 #ifdef CONFIG_COMPAT
1941 .compat_setsockopt = compat_udp_setsockopt,
1942 .compat_getsockopt = compat_udp_getsockopt,
1943 #endif
1944 .clear_sk = sk_prot_clear_portaddr_nulls,
1945 };
1946 EXPORT_SYMBOL(udp_prot);
1947
1948 /* ------------------------------------------------------------------------ */
1949 #ifdef CONFIG_PROC_FS
1950
1951 static struct sock *udp_get_first(struct seq_file *seq, int start)
1952 {
1953 struct sock *sk;
1954 struct udp_iter_state *state = seq->private;
1955 struct net *net = seq_file_net(seq);
1956
1957 for (state->bucket = start; state->bucket <= state->udp_table->mask;
1958 ++state->bucket) {
1959 struct hlist_nulls_node *node;
1960 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
1961
1962 if (hlist_nulls_empty(&hslot->head))
1963 continue;
1964
1965 spin_lock_bh(&hslot->lock);
1966 sk_nulls_for_each(sk, node, &hslot->head) {
1967 if (!net_eq(sock_net(sk), net))
1968 continue;
1969 if (sk->sk_family == state->family)
1970 goto found;
1971 }
1972 spin_unlock_bh(&hslot->lock);
1973 }
1974 sk = NULL;
1975 found:
1976 return sk;
1977 }
1978
1979 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1980 {
1981 struct udp_iter_state *state = seq->private;
1982 struct net *net = seq_file_net(seq);
1983
1984 do {
1985 sk = sk_nulls_next(sk);
1986 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1987
1988 if (!sk) {
1989 if (state->bucket <= state->udp_table->mask)
1990 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
1991 return udp_get_first(seq, state->bucket + 1);
1992 }
1993 return sk;
1994 }
1995
1996 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1997 {
1998 struct sock *sk = udp_get_first(seq, 0);
1999
2000 if (sk)
2001 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2002 --pos;
2003 return pos ? NULL : sk;
2004 }
2005
2006 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2007 {
2008 struct udp_iter_state *state = seq->private;
2009 state->bucket = MAX_UDP_PORTS;
2010
2011 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2012 }
2013
2014 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2015 {
2016 struct sock *sk;
2017
2018 if (v == SEQ_START_TOKEN)
2019 sk = udp_get_idx(seq, 0);
2020 else
2021 sk = udp_get_next(seq, v);
2022
2023 ++*pos;
2024 return sk;
2025 }
2026
2027 static void udp_seq_stop(struct seq_file *seq, void *v)
2028 {
2029 struct udp_iter_state *state = seq->private;
2030
2031 if (state->bucket <= state->udp_table->mask)
2032 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2033 }
2034
2035 static int udp_seq_open(struct inode *inode, struct file *file)
2036 {
2037 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
2038 struct udp_iter_state *s;
2039 int err;
2040
2041 err = seq_open_net(inode, file, &afinfo->seq_ops,
2042 sizeof(struct udp_iter_state));
2043 if (err < 0)
2044 return err;
2045
2046 s = ((struct seq_file *)file->private_data)->private;
2047 s->family = afinfo->family;
2048 s->udp_table = afinfo->udp_table;
2049 return err;
2050 }
2051
2052 /* ------------------------------------------------------------------------ */
2053 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2054 {
2055 struct proc_dir_entry *p;
2056 int rc = 0;
2057
2058 afinfo->seq_fops.open = udp_seq_open;
2059 afinfo->seq_fops.read = seq_read;
2060 afinfo->seq_fops.llseek = seq_lseek;
2061 afinfo->seq_fops.release = seq_release_net;
2062
2063 afinfo->seq_ops.start = udp_seq_start;
2064 afinfo->seq_ops.next = udp_seq_next;
2065 afinfo->seq_ops.stop = udp_seq_stop;
2066
2067 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2068 &afinfo->seq_fops, afinfo);
2069 if (!p)
2070 rc = -ENOMEM;
2071 return rc;
2072 }
2073 EXPORT_SYMBOL(udp_proc_register);
2074
2075 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2076 {
2077 proc_net_remove(net, afinfo->name);
2078 }
2079 EXPORT_SYMBOL(udp_proc_unregister);
2080
2081 /* ------------------------------------------------------------------------ */
2082 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2083 int bucket, int *len)
2084 {
2085 struct inet_sock *inet = inet_sk(sp);
2086 __be32 dest = inet->inet_daddr;
2087 __be32 src = inet->inet_rcv_saddr;
2088 __u16 destp = ntohs(inet->inet_dport);
2089 __u16 srcp = ntohs(inet->inet_sport);
2090
2091 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2092 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
2093 bucket, src, srcp, dest, destp, sp->sk_state,
2094 sk_wmem_alloc_get(sp),
2095 sk_rmem_alloc_get(sp),
2096 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
2097 atomic_read(&sp->sk_refcnt), sp,
2098 atomic_read(&sp->sk_drops), len);
2099 }
2100
2101 int udp4_seq_show(struct seq_file *seq, void *v)
2102 {
2103 if (v == SEQ_START_TOKEN)
2104 seq_printf(seq, "%-127s\n",
2105 " sl local_address rem_address st tx_queue "
2106 "rx_queue tr tm->when retrnsmt uid timeout "
2107 "inode ref pointer drops");
2108 else {
2109 struct udp_iter_state *state = seq->private;
2110 int len;
2111
2112 udp4_format_sock(v, seq, state->bucket, &len);
2113 seq_printf(seq, "%*s\n", 127 - len, "");
2114 }
2115 return 0;
2116 }
2117
2118 /* ------------------------------------------------------------------------ */
2119 static struct udp_seq_afinfo udp4_seq_afinfo = {
2120 .name = "udp",
2121 .family = AF_INET,
2122 .udp_table = &udp_table,
2123 .seq_fops = {
2124 .owner = THIS_MODULE,
2125 },
2126 .seq_ops = {
2127 .show = udp4_seq_show,
2128 },
2129 };
2130
2131 static int __net_init udp4_proc_init_net(struct net *net)
2132 {
2133 return udp_proc_register(net, &udp4_seq_afinfo);
2134 }
2135
2136 static void __net_exit udp4_proc_exit_net(struct net *net)
2137 {
2138 udp_proc_unregister(net, &udp4_seq_afinfo);
2139 }
2140
2141 static struct pernet_operations udp4_net_ops = {
2142 .init = udp4_proc_init_net,
2143 .exit = udp4_proc_exit_net,
2144 };
2145
2146 int __init udp4_proc_init(void)
2147 {
2148 return register_pernet_subsys(&udp4_net_ops);
2149 }
2150
2151 void udp4_proc_exit(void)
2152 {
2153 unregister_pernet_subsys(&udp4_net_ops);
2154 }
2155 #endif /* CONFIG_PROC_FS */
2156
2157 static __initdata unsigned long uhash_entries;
2158 static int __init set_uhash_entries(char *str)
2159 {
2160 if (!str)
2161 return 0;
2162 uhash_entries = simple_strtoul(str, &str, 0);
2163 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2164 uhash_entries = UDP_HTABLE_SIZE_MIN;
2165 return 1;
2166 }
2167 __setup("uhash_entries=", set_uhash_entries);
2168
2169 void __init udp_table_init(struct udp_table *table, const char *name)
2170 {
2171 unsigned int i;
2172
2173 if (!CONFIG_BASE_SMALL)
2174 table->hash = alloc_large_system_hash(name,
2175 2 * sizeof(struct udp_hslot),
2176 uhash_entries,
2177 21, /* one slot per 2 MB */
2178 0,
2179 &table->log,
2180 &table->mask,
2181 64 * 1024);
2182 /*
2183 * Make sure hash table has the minimum size
2184 */
2185 if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) {
2186 table->hash = kmalloc(UDP_HTABLE_SIZE_MIN *
2187 2 * sizeof(struct udp_hslot), GFP_KERNEL);
2188 if (!table->hash)
2189 panic(name);
2190 table->log = ilog2(UDP_HTABLE_SIZE_MIN);
2191 table->mask = UDP_HTABLE_SIZE_MIN - 1;
2192 }
2193 table->hash2 = table->hash + (table->mask + 1);
2194 for (i = 0; i <= table->mask; i++) {
2195 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2196 table->hash[i].count = 0;
2197 spin_lock_init(&table->hash[i].lock);
2198 }
2199 for (i = 0; i <= table->mask; i++) {
2200 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2201 table->hash2[i].count = 0;
2202 spin_lock_init(&table->hash2[i].lock);
2203 }
2204 }
2205
2206 void __init udp_init(void)
2207 {
2208 unsigned long nr_pages, limit;
2209
2210 udp_table_init(&udp_table, "UDP");
2211 /* Set the pressure threshold up by the same strategy of TCP. It is a
2212 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
2213 * toward zero with the amount of memory, with a floor of 128 pages.
2214 */
2215 nr_pages = totalram_pages - totalhigh_pages;
2216 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
2217 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
2218 limit = max(limit, 128UL);
2219 sysctl_udp_mem[0] = limit / 4 * 3;
2220 sysctl_udp_mem[1] = limit;
2221 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2222
2223 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2224 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2225 }
2226
2227 int udp4_ufo_send_check(struct sk_buff *skb)
2228 {
2229 const struct iphdr *iph;
2230 struct udphdr *uh;
2231
2232 if (!pskb_may_pull(skb, sizeof(*uh)))
2233 return -EINVAL;
2234
2235 iph = ip_hdr(skb);
2236 uh = udp_hdr(skb);
2237
2238 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
2239 IPPROTO_UDP, 0);
2240 skb->csum_start = skb_transport_header(skb) - skb->head;
2241 skb->csum_offset = offsetof(struct udphdr, check);
2242 skb->ip_summed = CHECKSUM_PARTIAL;
2243 return 0;
2244 }
2245
2246 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, u32 features)
2247 {
2248 struct sk_buff *segs = ERR_PTR(-EINVAL);
2249 unsigned int mss;
2250 int offset;
2251 __wsum csum;
2252
2253 mss = skb_shinfo(skb)->gso_size;
2254 if (unlikely(skb->len <= mss))
2255 goto out;
2256
2257 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2258 /* Packet is from an untrusted source, reset gso_segs. */
2259 int type = skb_shinfo(skb)->gso_type;
2260
2261 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
2262 !(type & (SKB_GSO_UDP))))
2263 goto out;
2264
2265 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
2266
2267 segs = NULL;
2268 goto out;
2269 }
2270
2271 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
2272 * do checksum of UDP packets sent as multiple IP fragments.
2273 */
2274 offset = skb_checksum_start_offset(skb);
2275 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2276 offset += skb->csum_offset;
2277 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2278 skb->ip_summed = CHECKSUM_NONE;
2279
2280 /* Fragment the skb. IP headers of the fragments are updated in
2281 * inet_gso_segment()
2282 */
2283 segs = skb_segment(skb, features);
2284 out:
2285 return segs;
2286 }
2287