projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
inet: constify ip headers and in6_addr
[deliverable/linux.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);
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;
795 int ulen = len;
796 struct ipcm_cookie ipc;
797 struct rtable *rt = NULL;
798 int free = 0;
799 int connected = 0;
800 __be32 daddr, faddr, saddr;
801 __be16 dport;
802 u8 tos;
803 int err, is_udplite = IS_UDPLITE(sk);
804 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
805 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
806 struct sk_buff *skb;
807
808 if (len > 0xFFFF)
809 return -EMSGSIZE;
810
811 /*
812 * Check the flags.
813 */
814
815 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
816 return -EOPNOTSUPP;
817
818 ipc.opt = NULL;
819 ipc.tx_flags = 0;
820
821 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
822
823 if (up->pending) {
824 /*
825 * There are pending frames.
826 * The socket lock must be held while it's corked.
827 */
828 lock_sock(sk);
829 if (likely(up->pending)) {
830 if (unlikely(up->pending != AF_INET)) {
831 release_sock(sk);
832 return -EINVAL;
833 }
834 goto do_append_data;
835 }
836 release_sock(sk);
837 }
838 ulen += sizeof(struct udphdr);
839
840 /*
841 * Get and verify the address.
842 */
843 if (msg->msg_name) {
844 struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name;
845 if (msg->msg_namelen < sizeof(*usin))
846 return -EINVAL;
847 if (usin->sin_family != AF_INET) {
848 if (usin->sin_family != AF_UNSPEC)
849 return -EAFNOSUPPORT;
850 }
851
852 daddr = usin->sin_addr.s_addr;
853 dport = usin->sin_port;
854 if (dport == 0)
855 return -EINVAL;
856 } else {
857 if (sk->sk_state != TCP_ESTABLISHED)
858 return -EDESTADDRREQ;
859 daddr = inet->inet_daddr;
860 dport = inet->inet_dport;
861 /* Open fast path for connected socket.
862 Route will not be used, if at least one option is set.
863 */
864 connected = 1;
865 }
866 ipc.addr = inet->inet_saddr;
867
868 ipc.oif = sk->sk_bound_dev_if;
869 err = sock_tx_timestamp(sk, &ipc.tx_flags);
870 if (err)
871 return err;
872 if (msg->msg_controllen) {
873 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
874 if (err)
875 return err;
876 if (ipc.opt)
877 free = 1;
878 connected = 0;
879 }
880 if (!ipc.opt)
881 ipc.opt = inet->opt;
882
883 saddr = ipc.addr;
884 ipc.addr = faddr = daddr;
885
886 if (ipc.opt && ipc.opt->srr) {
887 if (!daddr)
888 return -EINVAL;
889 faddr = ipc.opt->faddr;
890 connected = 0;
891 }
892 tos = RT_TOS(inet->tos);
893 if (sock_flag(sk, SOCK_LOCALROUTE) ||
894 (msg->msg_flags & MSG_DONTROUTE) ||
895 (ipc.opt && ipc.opt->is_strictroute)) {
896 tos |= RTO_ONLINK;
897 connected = 0;
898 }
899
900 if (ipv4_is_multicast(daddr)) {
901 if (!ipc.oif)
902 ipc.oif = inet->mc_index;
903 if (!saddr)
904 saddr = inet->mc_addr;
905 connected = 0;
906 }
907
908 if (connected)
909 rt = (struct rtable *)sk_dst_check(sk, 0);
910
911 if (rt == NULL) {
912 struct flowi4 fl4;
913 struct net *net = sock_net(sk);
914
915 flowi4_init_output(&fl4, ipc.oif, sk->sk_mark, tos,
916 RT_SCOPE_UNIVERSE, sk->sk_protocol,
917 inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP,
918 faddr, saddr, dport, inet->inet_sport);
919
920 security_sk_classify_flow(sk, flowi4_to_flowi(&fl4));
921 rt = ip_route_output_flow(net, &fl4, sk);
922 if (IS_ERR(rt)) {
923 err = PTR_ERR(rt);
924 rt = NULL;
925 if (err == -ENETUNREACH)
926 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
927 goto out;
928 }
929
930 err = -EACCES;
931 if ((rt->rt_flags & RTCF_BROADCAST) &&
932 !sock_flag(sk, SOCK_BROADCAST))
933 goto out;
934 if (connected)
935 sk_dst_set(sk, dst_clone(&rt->dst));
936 }
937
938 if (msg->msg_flags&MSG_CONFIRM)
939 goto do_confirm;
940 back_from_confirm:
941
942 saddr = rt->rt_src;
943 if (!ipc.addr)
944 daddr = ipc.addr = rt->rt_dst;
945
946 /* Lockless fast path for the non-corking case. */
947 if (!corkreq) {
948 skb = ip_make_skb(sk, getfrag, msg->msg_iov, ulen,
949 sizeof(struct udphdr), &ipc, &rt,
950 msg->msg_flags);
951 err = PTR_ERR(skb);
952 if (skb && !IS_ERR(skb))
953 err = udp_send_skb(skb, daddr, dport);
954 goto out;
955 }
956
957 lock_sock(sk);
958 if (unlikely(up->pending)) {
959 /* The socket is already corked while preparing it. */
960 /* ... which is an evident application bug. --ANK */
961 release_sock(sk);
962
963 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
964 err = -EINVAL;
965 goto out;
966 }
967 /*
968 * Now cork the socket to pend data.
969 */
970 fl4 = &inet->cork.fl.u.ip4;
971 fl4->daddr = daddr;
972 fl4->saddr = saddr;
973 fl4->fl4_dport = dport;
974 fl4->fl4_sport = inet->inet_sport;
975 up->pending = AF_INET;
976
977 do_append_data:
978 up->len += ulen;
979 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
980 sizeof(struct udphdr), &ipc, &rt,
981 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
982 if (err)
983 udp_flush_pending_frames(sk);
984 else if (!corkreq)
985 err = udp_push_pending_frames(sk);
986 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
987 up->pending = 0;
988 release_sock(sk);
989
990 out:
991 ip_rt_put(rt);
992 if (free)
993 kfree(ipc.opt);
994 if (!err)
995 return len;
996 /*
997 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
998 * ENOBUFS might not be good (it's not tunable per se), but otherwise
999 * we don't have a good statistic (IpOutDiscards but it can be too many
1000 * things). We could add another new stat but at least for now that
1001 * seems like overkill.
1002 */
1003 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1004 UDP_INC_STATS_USER(sock_net(sk),
1005 UDP_MIB_SNDBUFERRORS, is_udplite);
1006 }
1007 return err;
1008
1009 do_confirm:
1010 dst_confirm(&rt->dst);
1011 if (!(msg->msg_flags&MSG_PROBE) || len)
1012 goto back_from_confirm;
1013 err = 0;
1014 goto out;
1015 }
1016 EXPORT_SYMBOL(udp_sendmsg);
1017
1018 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1019 size_t size, int flags)
1020 {
1021 struct udp_sock *up = udp_sk(sk);
1022 int ret;
1023
1024 if (!up->pending) {
1025 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1026
1027 /* Call udp_sendmsg to specify destination address which
1028 * sendpage interface can't pass.
1029 * This will succeed only when the socket is connected.
1030 */
1031 ret = udp_sendmsg(NULL, sk, &msg, 0);
1032 if (ret < 0)
1033 return ret;
1034 }
1035
1036 lock_sock(sk);
1037
1038 if (unlikely(!up->pending)) {
1039 release_sock(sk);
1040
1041 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
1042 return -EINVAL;
1043 }
1044
1045 ret = ip_append_page(sk, page, offset, size, flags);
1046 if (ret == -EOPNOTSUPP) {
1047 release_sock(sk);
1048 return sock_no_sendpage(sk->sk_socket, page, offset,
1049 size, flags);
1050 }
1051 if (ret < 0) {
1052 udp_flush_pending_frames(sk);
1053 goto out;
1054 }
1055
1056 up->len += size;
1057 if (!(up->corkflag || (flags&MSG_MORE)))
1058 ret = udp_push_pending_frames(sk);
1059 if (!ret)
1060 ret = size;
1061 out:
1062 release_sock(sk);
1063 return ret;
1064 }
1065
1066
1067 /**
1068 * first_packet_length - return length of first packet in receive queue
1069 * @sk: socket
1070 *
1071 * Drops all bad checksum frames, until a valid one is found.
1072 * Returns the length of found skb, or 0 if none is found.
1073 */
1074 static unsigned int first_packet_length(struct sock *sk)
1075 {
1076 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1077 struct sk_buff *skb;
1078 unsigned int res;
1079
1080 __skb_queue_head_init(&list_kill);
1081
1082 spin_lock_bh(&rcvq->lock);
1083 while ((skb = skb_peek(rcvq)) != NULL &&
1084 udp_lib_checksum_complete(skb)) {
1085 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1086 IS_UDPLITE(sk));
1087 atomic_inc(&sk->sk_drops);
1088 __skb_unlink(skb, rcvq);
1089 __skb_queue_tail(&list_kill, skb);
1090 }
1091 res = skb ? skb->len : 0;
1092 spin_unlock_bh(&rcvq->lock);
1093
1094 if (!skb_queue_empty(&list_kill)) {
1095 bool slow = lock_sock_fast(sk);
1096
1097 __skb_queue_purge(&list_kill);
1098 sk_mem_reclaim_partial(sk);
1099 unlock_sock_fast(sk, slow);
1100 }
1101 return res;
1102 }
1103
1104 /*
1105 * IOCTL requests applicable to the UDP protocol
1106 */
1107
1108 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1109 {
1110 switch (cmd) {
1111 case SIOCOUTQ:
1112 {
1113 int amount = sk_wmem_alloc_get(sk);
1114
1115 return put_user(amount, (int __user *)arg);
1116 }
1117
1118 case SIOCINQ:
1119 {
1120 unsigned int amount = first_packet_length(sk);
1121
1122 if (amount)
1123 /*
1124 * We will only return the amount
1125 * of this packet since that is all
1126 * that will be read.
1127 */
1128 amount -= sizeof(struct udphdr);
1129
1130 return put_user(amount, (int __user *)arg);
1131 }
1132
1133 default:
1134 return -ENOIOCTLCMD;
1135 }
1136
1137 return 0;
1138 }
1139 EXPORT_SYMBOL(udp_ioctl);
1140
1141 /*
1142 * This should be easy, if there is something there we
1143 * return it, otherwise we block.
1144 */
1145
1146 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1147 size_t len, int noblock, int flags, int *addr_len)
1148 {
1149 struct inet_sock *inet = inet_sk(sk);
1150 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1151 struct sk_buff *skb;
1152 unsigned int ulen;
1153 int peeked;
1154 int err;
1155 int is_udplite = IS_UDPLITE(sk);
1156 bool slow;
1157
1158 /*
1159 * Check any passed addresses
1160 */
1161 if (addr_len)
1162 *addr_len = sizeof(*sin);
1163
1164 if (flags & MSG_ERRQUEUE)
1165 return ip_recv_error(sk, msg, len);
1166
1167 try_again:
1168 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1169 &peeked, &err);
1170 if (!skb)
1171 goto out;
1172
1173 ulen = skb->len - sizeof(struct udphdr);
1174 if (len > ulen)
1175 len = ulen;
1176 else if (len < ulen)
1177 msg->msg_flags |= MSG_TRUNC;
1178
1179 /*
1180 * If checksum is needed at all, try to do it while copying the
1181 * data. If the data is truncated, or if we only want a partial
1182 * coverage checksum (UDP-Lite), do it before the copy.
1183 */
1184
1185 if (len < ulen || UDP_SKB_CB(skb)->partial_cov) {
1186 if (udp_lib_checksum_complete(skb))
1187 goto csum_copy_err;
1188 }
1189
1190 if (skb_csum_unnecessary(skb))
1191 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1192 msg->msg_iov, len);
1193 else {
1194 err = skb_copy_and_csum_datagram_iovec(skb,
1195 sizeof(struct udphdr),
1196 msg->msg_iov);
1197
1198 if (err == -EINVAL)
1199 goto csum_copy_err;
1200 }
1201
1202 if (err)
1203 goto out_free;
1204
1205 if (!peeked)
1206 UDP_INC_STATS_USER(sock_net(sk),
1207 UDP_MIB_INDATAGRAMS, is_udplite);
1208
1209 sock_recv_ts_and_drops(msg, sk, skb);
1210
1211 /* Copy the address. */
1212 if (sin) {
1213 sin->sin_family = AF_INET;
1214 sin->sin_port = udp_hdr(skb)->source;
1215 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1216 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1217 }
1218 if (inet->cmsg_flags)
1219 ip_cmsg_recv(msg, skb);
1220
1221 err = len;
1222 if (flags & MSG_TRUNC)
1223 err = ulen;
1224
1225 out_free:
1226 skb_free_datagram_locked(sk, skb);
1227 out:
1228 return err;
1229
1230 csum_copy_err:
1231 slow = lock_sock_fast(sk);
1232 if (!skb_kill_datagram(sk, skb, flags))
1233 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1234 unlock_sock_fast(sk, slow);
1235
1236 if (noblock)
1237 return -EAGAIN;
1238 goto try_again;
1239 }
1240
1241
1242 int udp_disconnect(struct sock *sk, int flags)
1243 {
1244 struct inet_sock *inet = inet_sk(sk);
1245 /*
1246 * 1003.1g - break association.
1247 */
1248
1249 sk->sk_state = TCP_CLOSE;
1250 inet->inet_daddr = 0;
1251 inet->inet_dport = 0;
1252 sock_rps_save_rxhash(sk, 0);
1253 sk->sk_bound_dev_if = 0;
1254 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1255 inet_reset_saddr(sk);
1256
1257 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1258 sk->sk_prot->unhash(sk);
1259 inet->inet_sport = 0;
1260 }
1261 sk_dst_reset(sk);
1262 return 0;
1263 }
1264 EXPORT_SYMBOL(udp_disconnect);
1265
1266 void udp_lib_unhash(struct sock *sk)
1267 {
1268 if (sk_hashed(sk)) {
1269 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1270 struct udp_hslot *hslot, *hslot2;
1271
1272 hslot = udp_hashslot(udptable, sock_net(sk),
1273 udp_sk(sk)->udp_port_hash);
1274 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1275
1276 spin_lock_bh(&hslot->lock);
1277 if (sk_nulls_del_node_init_rcu(sk)) {
1278 hslot->count--;
1279 inet_sk(sk)->inet_num = 0;
1280 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1281
1282 spin_lock(&hslot2->lock);
1283 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1284 hslot2->count--;
1285 spin_unlock(&hslot2->lock);
1286 }
1287 spin_unlock_bh(&hslot->lock);
1288 }
1289 }
1290 EXPORT_SYMBOL(udp_lib_unhash);
1291
1292 /*
1293 * inet_rcv_saddr was changed, we must rehash secondary hash
1294 */
1295 void udp_lib_rehash(struct sock *sk, u16 newhash)
1296 {
1297 if (sk_hashed(sk)) {
1298 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1299 struct udp_hslot *hslot, *hslot2, *nhslot2;
1300
1301 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1302 nhslot2 = udp_hashslot2(udptable, newhash);
1303 udp_sk(sk)->udp_portaddr_hash = newhash;
1304 if (hslot2 != nhslot2) {
1305 hslot = udp_hashslot(udptable, sock_net(sk),
1306 udp_sk(sk)->udp_port_hash);
1307 /* we must lock primary chain too */
1308 spin_lock_bh(&hslot->lock);
1309
1310 spin_lock(&hslot2->lock);
1311 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1312 hslot2->count--;
1313 spin_unlock(&hslot2->lock);
1314
1315 spin_lock(&nhslot2->lock);
1316 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1317 &nhslot2->head);
1318 nhslot2->count++;
1319 spin_unlock(&nhslot2->lock);
1320
1321 spin_unlock_bh(&hslot->lock);
1322 }
1323 }
1324 }
1325 EXPORT_SYMBOL(udp_lib_rehash);
1326
1327 static void udp_v4_rehash(struct sock *sk)
1328 {
1329 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1330 inet_sk(sk)->inet_rcv_saddr,
1331 inet_sk(sk)->inet_num);
1332 udp_lib_rehash(sk, new_hash);
1333 }
1334
1335 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1336 {
1337 int rc;
1338
1339 if (inet_sk(sk)->inet_daddr)
1340 sock_rps_save_rxhash(sk, skb->rxhash);
1341
1342 rc = ip_queue_rcv_skb(sk, skb);
1343 if (rc < 0) {
1344 int is_udplite = IS_UDPLITE(sk);
1345
1346 /* Note that an ENOMEM error is charged twice */
1347 if (rc == -ENOMEM)
1348 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1349 is_udplite);
1350 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1351 kfree_skb(skb);
1352 return -1;
1353 }
1354
1355 return 0;
1356
1357 }
1358
1359 /* returns:
1360 * -1: error
1361 * 0: success
1362 * >0: "udp encap" protocol resubmission
1363 *
1364 * Note that in the success and error cases, the skb is assumed to
1365 * have either been requeued or freed.
1366 */
1367 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1368 {
1369 struct udp_sock *up = udp_sk(sk);
1370 int rc;
1371 int is_udplite = IS_UDPLITE(sk);
1372
1373 /*
1374 * Charge it to the socket, dropping if the queue is full.
1375 */
1376 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1377 goto drop;
1378 nf_reset(skb);
1379
1380 if (up->encap_type) {
1381 /*
1382 * This is an encapsulation socket so pass the skb to
1383 * the socket's udp_encap_rcv() hook. Otherwise, just
1384 * fall through and pass this up the UDP socket.
1385 * up->encap_rcv() returns the following value:
1386 * =0 if skb was successfully passed to the encap
1387 * handler or was discarded by it.
1388 * >0 if skb should be passed on to UDP.
1389 * <0 if skb should be resubmitted as proto -N
1390 */
1391
1392 /* if we're overly short, let UDP handle it */
1393 if (skb->len > sizeof(struct udphdr) &&
1394 up->encap_rcv != NULL) {
1395 int ret;
1396
1397 ret = (*up->encap_rcv)(sk, skb);
1398 if (ret <= 0) {
1399 UDP_INC_STATS_BH(sock_net(sk),
1400 UDP_MIB_INDATAGRAMS,
1401 is_udplite);
1402 return -ret;
1403 }
1404 }
1405
1406 /* FALLTHROUGH -- it's a UDP Packet */
1407 }
1408
1409 /*
1410 * UDP-Lite specific tests, ignored on UDP sockets
1411 */
1412 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1413
1414 /*
1415 * MIB statistics other than incrementing the error count are
1416 * disabled for the following two types of errors: these depend
1417 * on the application settings, not on the functioning of the
1418 * protocol stack as such.
1419 *
1420 * RFC 3828 here recommends (sec 3.3): "There should also be a
1421 * way ... to ... at least let the receiving application block
1422 * delivery of packets with coverage values less than a value
1423 * provided by the application."
1424 */
1425 if (up->pcrlen == 0) { /* full coverage was set */
1426 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1427 "%d while full coverage %d requested\n",
1428 UDP_SKB_CB(skb)->cscov, skb->len);
1429 goto drop;
1430 }
1431 /* The next case involves violating the min. coverage requested
1432 * by the receiver. This is subtle: if receiver wants x and x is
1433 * greater than the buffersize/MTU then receiver will complain
1434 * that it wants x while sender emits packets of smaller size y.
1435 * Therefore the above ...()->partial_cov statement is essential.
1436 */
1437 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1438 LIMIT_NETDEBUG(KERN_WARNING
1439 "UDPLITE: coverage %d too small, need min %d\n",
1440 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1441 goto drop;
1442 }
1443 }
1444
1445 if (rcu_dereference_raw(sk->sk_filter)) {
1446 if (udp_lib_checksum_complete(skb))
1447 goto drop;
1448 }
1449
1450
1451 if (sk_rcvqueues_full(sk, skb))
1452 goto drop;
1453
1454 rc = 0;
1455
1456 bh_lock_sock(sk);
1457 if (!sock_owned_by_user(sk))
1458 rc = __udp_queue_rcv_skb(sk, skb);
1459 else if (sk_add_backlog(sk, skb)) {
1460 bh_unlock_sock(sk);
1461 goto drop;
1462 }
1463 bh_unlock_sock(sk);
1464
1465 return rc;
1466
1467 drop:
1468 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1469 atomic_inc(&sk->sk_drops);
1470 kfree_skb(skb);
1471 return -1;
1472 }
1473
1474
1475 static void flush_stack(struct sock **stack, unsigned int count,
1476 struct sk_buff *skb, unsigned int final)
1477 {
1478 unsigned int i;
1479 struct sk_buff *skb1 = NULL;
1480 struct sock *sk;
1481
1482 for (i = 0; i < count; i++) {
1483 sk = stack[i];
1484 if (likely(skb1 == NULL))
1485 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1486
1487 if (!skb1) {
1488 atomic_inc(&sk->sk_drops);
1489 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1490 IS_UDPLITE(sk));
1491 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1492 IS_UDPLITE(sk));
1493 }
1494
1495 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1496 skb1 = NULL;
1497 }
1498 if (unlikely(skb1))
1499 kfree_skb(skb1);
1500 }
1501
1502 /*
1503 * Multicasts and broadcasts go to each listener.
1504 *
1505 * Note: called only from the BH handler context.
1506 */
1507 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1508 struct udphdr *uh,
1509 __be32 saddr, __be32 daddr,
1510 struct udp_table *udptable)
1511 {
1512 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1513 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1514 int dif;
1515 unsigned int i, count = 0;
1516
1517 spin_lock(&hslot->lock);
1518 sk = sk_nulls_head(&hslot->head);
1519 dif = skb->dev->ifindex;
1520 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1521 while (sk) {
1522 stack[count++] = sk;
1523 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1524 daddr, uh->source, saddr, dif);
1525 if (unlikely(count == ARRAY_SIZE(stack))) {
1526 if (!sk)
1527 break;
1528 flush_stack(stack, count, skb, ~0);
1529 count = 0;
1530 }
1531 }
1532 /*
1533 * before releasing chain lock, we must take a reference on sockets
1534 */
1535 for (i = 0; i < count; i++)
1536 sock_hold(stack[i]);
1537
1538 spin_unlock(&hslot->lock);
1539
1540 /*
1541 * do the slow work with no lock held
1542 */
1543 if (count) {
1544 flush_stack(stack, count, skb, count - 1);
1545
1546 for (i = 0; i < count; i++)
1547 sock_put(stack[i]);
1548 } else {
1549 kfree_skb(skb);
1550 }
1551 return 0;
1552 }
1553
1554 /* Initialize UDP checksum. If exited with zero value (success),
1555 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1556 * Otherwise, csum completion requires chacksumming packet body,
1557 * including udp header and folding it to skb->csum.
1558 */
1559 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1560 int proto)
1561 {
1562 const struct iphdr *iph;
1563 int err;
1564
1565 UDP_SKB_CB(skb)->partial_cov = 0;
1566 UDP_SKB_CB(skb)->cscov = skb->len;
1567
1568 if (proto == IPPROTO_UDPLITE) {
1569 err = udplite_checksum_init(skb, uh);
1570 if (err)
1571 return err;
1572 }
1573
1574 iph = ip_hdr(skb);
1575 if (uh->check == 0) {
1576 skb->ip_summed = CHECKSUM_UNNECESSARY;
1577 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1578 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1579 proto, skb->csum))
1580 skb->ip_summed = CHECKSUM_UNNECESSARY;
1581 }
1582 if (!skb_csum_unnecessary(skb))
1583 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1584 skb->len, proto, 0);
1585 /* Probably, we should checksum udp header (it should be in cache
1586 * in any case) and data in tiny packets (< rx copybreak).
1587 */
1588
1589 return 0;
1590 }
1591
1592 /*
1593 * All we need to do is get the socket, and then do a checksum.
1594 */
1595
1596 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1597 int proto)
1598 {
1599 struct sock *sk;
1600 struct udphdr *uh;
1601 unsigned short ulen;
1602 struct rtable *rt = skb_rtable(skb);
1603 __be32 saddr, daddr;
1604 struct net *net = dev_net(skb->dev);
1605
1606 /*
1607 * Validate the packet.
1608 */
1609 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1610 goto drop; /* No space for header. */
1611
1612 uh = udp_hdr(skb);
1613 ulen = ntohs(uh->len);
1614 saddr = ip_hdr(skb)->saddr;
1615 daddr = ip_hdr(skb)->daddr;
1616
1617 if (ulen > skb->len)
1618 goto short_packet;
1619
1620 if (proto == IPPROTO_UDP) {
1621 /* UDP validates ulen. */
1622 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1623 goto short_packet;
1624 uh = udp_hdr(skb);
1625 }
1626
1627 if (udp4_csum_init(skb, uh, proto))
1628 goto csum_error;
1629
1630 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1631 return __udp4_lib_mcast_deliver(net, skb, uh,
1632 saddr, daddr, udptable);
1633
1634 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1635
1636 if (sk != NULL) {
1637 int ret = udp_queue_rcv_skb(sk, skb);
1638 sock_put(sk);
1639
1640 /* a return value > 0 means to resubmit the input, but
1641 * it wants the return to be -protocol, or 0
1642 */
1643 if (ret > 0)
1644 return -ret;
1645 return 0;
1646 }
1647
1648 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1649 goto drop;
1650 nf_reset(skb);
1651
1652 /* No socket. Drop packet silently, if checksum is wrong */
1653 if (udp_lib_checksum_complete(skb))
1654 goto csum_error;
1655
1656 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1657 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1658
1659 /*
1660 * Hmm. We got an UDP packet to a port to which we
1661 * don't wanna listen. Ignore it.
1662 */
1663 kfree_skb(skb);
1664 return 0;
1665
1666 short_packet:
1667 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1668 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1669 &saddr,
1670 ntohs(uh->source),
1671 ulen,
1672 skb->len,
1673 &daddr,
1674 ntohs(uh->dest));
1675 goto drop;
1676
1677 csum_error:
1678 /*
1679 * RFC1122: OK. Discards the bad packet silently (as far as
1680 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1681 */
1682 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1683 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1684 &saddr,
1685 ntohs(uh->source),
1686 &daddr,
1687 ntohs(uh->dest),
1688 ulen);
1689 drop:
1690 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1691 kfree_skb(skb);
1692 return 0;
1693 }
1694
1695 int udp_rcv(struct sk_buff *skb)
1696 {
1697 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1698 }
1699
1700 void udp_destroy_sock(struct sock *sk)
1701 {
1702 bool slow = lock_sock_fast(sk);
1703 udp_flush_pending_frames(sk);
1704 unlock_sock_fast(sk, slow);
1705 }
1706
1707 /*
1708 * Socket option code for UDP
1709 */
1710 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1711 char __user *optval, unsigned int optlen,
1712 int (*push_pending_frames)(struct sock *))
1713 {
1714 struct udp_sock *up = udp_sk(sk);
1715 int val;
1716 int err = 0;
1717 int is_udplite = IS_UDPLITE(sk);
1718
1719 if (optlen < sizeof(int))
1720 return -EINVAL;
1721
1722 if (get_user(val, (int __user *)optval))
1723 return -EFAULT;
1724
1725 switch (optname) {
1726 case UDP_CORK:
1727 if (val != 0) {
1728 up->corkflag = 1;
1729 } else {
1730 up->corkflag = 0;
1731 lock_sock(sk);
1732 (*push_pending_frames)(sk);
1733 release_sock(sk);
1734 }
1735 break;
1736
1737 case UDP_ENCAP:
1738 switch (val) {
1739 case 0:
1740 case UDP_ENCAP_ESPINUDP:
1741 case UDP_ENCAP_ESPINUDP_NON_IKE:
1742 up->encap_rcv = xfrm4_udp_encap_rcv;
1743 /* FALLTHROUGH */
1744 case UDP_ENCAP_L2TPINUDP:
1745 up->encap_type = val;
1746 break;
1747 default:
1748 err = -ENOPROTOOPT;
1749 break;
1750 }
1751 break;
1752
1753 /*
1754 * UDP-Lite's partial checksum coverage (RFC 3828).
1755 */
1756 /* The sender sets actual checksum coverage length via this option.
1757 * The case coverage > packet length is handled by send module. */
1758 case UDPLITE_SEND_CSCOV:
1759 if (!is_udplite) /* Disable the option on UDP sockets */
1760 return -ENOPROTOOPT;
1761 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1762 val = 8;
1763 else if (val > USHRT_MAX)
1764 val = USHRT_MAX;
1765 up->pcslen = val;
1766 up->pcflag |= UDPLITE_SEND_CC;
1767 break;
1768
1769 /* The receiver specifies a minimum checksum coverage value. To make
1770 * sense, this should be set to at least 8 (as done below). If zero is
1771 * used, this again means full checksum coverage. */
1772 case UDPLITE_RECV_CSCOV:
1773 if (!is_udplite) /* Disable the option on UDP sockets */
1774 return -ENOPROTOOPT;
1775 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1776 val = 8;
1777 else if (val > USHRT_MAX)
1778 val = USHRT_MAX;
1779 up->pcrlen = val;
1780 up->pcflag |= UDPLITE_RECV_CC;
1781 break;
1782
1783 default:
1784 err = -ENOPROTOOPT;
1785 break;
1786 }
1787
1788 return err;
1789 }
1790 EXPORT_SYMBOL(udp_lib_setsockopt);
1791
1792 int udp_setsockopt(struct sock *sk, int level, int optname,
1793 char __user *optval, unsigned int optlen)
1794 {
1795 if (level == SOL_UDP || level == SOL_UDPLITE)
1796 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1797 udp_push_pending_frames);
1798 return ip_setsockopt(sk, level, optname, optval, optlen);
1799 }
1800
1801 #ifdef CONFIG_COMPAT
1802 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1803 char __user *optval, unsigned int optlen)
1804 {
1805 if (level == SOL_UDP || level == SOL_UDPLITE)
1806 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1807 udp_push_pending_frames);
1808 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1809 }
1810 #endif
1811
1812 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1813 char __user *optval, int __user *optlen)
1814 {
1815 struct udp_sock *up = udp_sk(sk);
1816 int val, len;
1817
1818 if (get_user(len, optlen))
1819 return -EFAULT;
1820
1821 len = min_t(unsigned int, len, sizeof(int));
1822
1823 if (len < 0)
1824 return -EINVAL;
1825
1826 switch (optname) {
1827 case UDP_CORK:
1828 val = up->corkflag;
1829 break;
1830
1831 case UDP_ENCAP:
1832 val = up->encap_type;
1833 break;
1834
1835 /* The following two cannot be changed on UDP sockets, the return is
1836 * always 0 (which corresponds to the full checksum coverage of UDP). */
1837 case UDPLITE_SEND_CSCOV:
1838 val = up->pcslen;
1839 break;
1840
1841 case UDPLITE_RECV_CSCOV:
1842 val = up->pcrlen;
1843 break;
1844
1845 default:
1846 return -ENOPROTOOPT;
1847 }
1848
1849 if (put_user(len, optlen))
1850 return -EFAULT;
1851 if (copy_to_user(optval, &val, len))
1852 return -EFAULT;
1853 return 0;
1854 }
1855 EXPORT_SYMBOL(udp_lib_getsockopt);
1856
1857 int udp_getsockopt(struct sock *sk, int level, int optname,
1858 char __user *optval, int __user *optlen)
1859 {
1860 if (level == SOL_UDP || level == SOL_UDPLITE)
1861 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1862 return ip_getsockopt(sk, level, optname, optval, optlen);
1863 }
1864
1865 #ifdef CONFIG_COMPAT
1866 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1867 char __user *optval, int __user *optlen)
1868 {
1869 if (level == SOL_UDP || level == SOL_UDPLITE)
1870 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1871 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1872 }
1873 #endif
1874 /**
1875 * udp_poll - wait for a UDP event.
1876 * @file - file struct
1877 * @sock - socket
1878 * @wait - poll table
1879 *
1880 * This is same as datagram poll, except for the special case of
1881 * blocking sockets. If application is using a blocking fd
1882 * and a packet with checksum error is in the queue;
1883 * then it could get return from select indicating data available
1884 * but then block when reading it. Add special case code
1885 * to work around these arguably broken applications.
1886 */
1887 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1888 {
1889 unsigned int mask = datagram_poll(file, sock, wait);
1890 struct sock *sk = sock->sk;
1891
1892 /* Check for false positives due to checksum errors */
1893 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1894 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1895 mask &= ~(POLLIN | POLLRDNORM);
1896
1897 return mask;
1898
1899 }
1900 EXPORT_SYMBOL(udp_poll);
1901
1902 struct proto udp_prot = {
1903 .name = "UDP",
1904 .owner = THIS_MODULE,
1905 .close = udp_lib_close,
1906 .connect = ip4_datagram_connect,
1907 .disconnect = udp_disconnect,
1908 .ioctl = udp_ioctl,
1909 .destroy = udp_destroy_sock,
1910 .setsockopt = udp_setsockopt,
1911 .getsockopt = udp_getsockopt,
1912 .sendmsg = udp_sendmsg,
1913 .recvmsg = udp_recvmsg,
1914 .sendpage = udp_sendpage,
1915 .backlog_rcv = __udp_queue_rcv_skb,
1916 .hash = udp_lib_hash,
1917 .unhash = udp_lib_unhash,
1918 .rehash = udp_v4_rehash,
1919 .get_port = udp_v4_get_port,
1920 .memory_allocated = &udp_memory_allocated,
1921 .sysctl_mem = sysctl_udp_mem,
1922 .sysctl_wmem = &sysctl_udp_wmem_min,
1923 .sysctl_rmem = &sysctl_udp_rmem_min,
1924 .obj_size = sizeof(struct udp_sock),
1925 .slab_flags = SLAB_DESTROY_BY_RCU,
1926 .h.udp_table = &udp_table,
1927 #ifdef CONFIG_COMPAT
1928 .compat_setsockopt = compat_udp_setsockopt,
1929 .compat_getsockopt = compat_udp_getsockopt,
1930 #endif
1931 .clear_sk = sk_prot_clear_portaddr_nulls,
1932 };
1933 EXPORT_SYMBOL(udp_prot);
1934
1935 /* ------------------------------------------------------------------------ */
1936 #ifdef CONFIG_PROC_FS
1937
1938 static struct sock *udp_get_first(struct seq_file *seq, int start)
1939 {
1940 struct sock *sk;
1941 struct udp_iter_state *state = seq->private;
1942 struct net *net = seq_file_net(seq);
1943
1944 for (state->bucket = start; state->bucket <= state->udp_table->mask;
1945 ++state->bucket) {
1946 struct hlist_nulls_node *node;
1947 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
1948
1949 if (hlist_nulls_empty(&hslot->head))
1950 continue;
1951
1952 spin_lock_bh(&hslot->lock);
1953 sk_nulls_for_each(sk, node, &hslot->head) {
1954 if (!net_eq(sock_net(sk), net))
1955 continue;
1956 if (sk->sk_family == state->family)
1957 goto found;
1958 }
1959 spin_unlock_bh(&hslot->lock);
1960 }
1961 sk = NULL;
1962 found:
1963 return sk;
1964 }
1965
1966 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1967 {
1968 struct udp_iter_state *state = seq->private;
1969 struct net *net = seq_file_net(seq);
1970
1971 do {
1972 sk = sk_nulls_next(sk);
1973 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1974
1975 if (!sk) {
1976 if (state->bucket <= state->udp_table->mask)
1977 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
1978 return udp_get_first(seq, state->bucket + 1);
1979 }
1980 return sk;
1981 }
1982
1983 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1984 {
1985 struct sock *sk = udp_get_first(seq, 0);
1986
1987 if (sk)
1988 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1989 --pos;
1990 return pos ? NULL : sk;
1991 }
1992
1993 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1994 {
1995 struct udp_iter_state *state = seq->private;
1996 state->bucket = MAX_UDP_PORTS;
1997
1998 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
1999 }
2000
2001 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2002 {
2003 struct sock *sk;
2004
2005 if (v == SEQ_START_TOKEN)
2006 sk = udp_get_idx(seq, 0);
2007 else
2008 sk = udp_get_next(seq, v);
2009
2010 ++*pos;
2011 return sk;
2012 }
2013
2014 static void udp_seq_stop(struct seq_file *seq, void *v)
2015 {
2016 struct udp_iter_state *state = seq->private;
2017
2018 if (state->bucket <= state->udp_table->mask)
2019 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2020 }
2021
2022 static int udp_seq_open(struct inode *inode, struct file *file)
2023 {
2024 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
2025 struct udp_iter_state *s;
2026 int err;
2027
2028 err = seq_open_net(inode, file, &afinfo->seq_ops,
2029 sizeof(struct udp_iter_state));
2030 if (err < 0)
2031 return err;
2032
2033 s = ((struct seq_file *)file->private_data)->private;
2034 s->family = afinfo->family;
2035 s->udp_table = afinfo->udp_table;
2036 return err;
2037 }
2038
2039 /* ------------------------------------------------------------------------ */
2040 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2041 {
2042 struct proc_dir_entry *p;
2043 int rc = 0;
2044
2045 afinfo->seq_fops.open = udp_seq_open;
2046 afinfo->seq_fops.read = seq_read;
2047 afinfo->seq_fops.llseek = seq_lseek;
2048 afinfo->seq_fops.release = seq_release_net;
2049
2050 afinfo->seq_ops.start = udp_seq_start;
2051 afinfo->seq_ops.next = udp_seq_next;
2052 afinfo->seq_ops.stop = udp_seq_stop;
2053
2054 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2055 &afinfo->seq_fops, afinfo);
2056 if (!p)
2057 rc = -ENOMEM;
2058 return rc;
2059 }
2060 EXPORT_SYMBOL(udp_proc_register);
2061
2062 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2063 {
2064 proc_net_remove(net, afinfo->name);
2065 }
2066 EXPORT_SYMBOL(udp_proc_unregister);
2067
2068 /* ------------------------------------------------------------------------ */
2069 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2070 int bucket, int *len)
2071 {
2072 struct inet_sock *inet = inet_sk(sp);
2073 __be32 dest = inet->inet_daddr;
2074 __be32 src = inet->inet_rcv_saddr;
2075 __u16 destp = ntohs(inet->inet_dport);
2076 __u16 srcp = ntohs(inet->inet_sport);
2077
2078 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2079 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
2080 bucket, src, srcp, dest, destp, sp->sk_state,
2081 sk_wmem_alloc_get(sp),
2082 sk_rmem_alloc_get(sp),
2083 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
2084 atomic_read(&sp->sk_refcnt), sp,
2085 atomic_read(&sp->sk_drops), len);
2086 }
2087
2088 int udp4_seq_show(struct seq_file *seq, void *v)
2089 {
2090 if (v == SEQ_START_TOKEN)
2091 seq_printf(seq, "%-127s\n",
2092 " sl local_address rem_address st tx_queue "
2093 "rx_queue tr tm->when retrnsmt uid timeout "
2094 "inode ref pointer drops");
2095 else {
2096 struct udp_iter_state *state = seq->private;
2097 int len;
2098
2099 udp4_format_sock(v, seq, state->bucket, &len);
2100 seq_printf(seq, "%*s\n", 127 - len, "");
2101 }
2102 return 0;
2103 }
2104
2105 /* ------------------------------------------------------------------------ */
2106 static struct udp_seq_afinfo udp4_seq_afinfo = {
2107 .name = "udp",
2108 .family = AF_INET,
2109 .udp_table = &udp_table,
2110 .seq_fops = {
2111 .owner = THIS_MODULE,
2112 },
2113 .seq_ops = {
2114 .show = udp4_seq_show,
2115 },
2116 };
2117
2118 static int __net_init udp4_proc_init_net(struct net *net)
2119 {
2120 return udp_proc_register(net, &udp4_seq_afinfo);
2121 }
2122
2123 static void __net_exit udp4_proc_exit_net(struct net *net)
2124 {
2125 udp_proc_unregister(net, &udp4_seq_afinfo);
2126 }
2127
2128 static struct pernet_operations udp4_net_ops = {
2129 .init = udp4_proc_init_net,
2130 .exit = udp4_proc_exit_net,
2131 };
2132
2133 int __init udp4_proc_init(void)
2134 {
2135 return register_pernet_subsys(&udp4_net_ops);
2136 }
2137
2138 void udp4_proc_exit(void)
2139 {
2140 unregister_pernet_subsys(&udp4_net_ops);
2141 }
2142 #endif /* CONFIG_PROC_FS */
2143
2144 static __initdata unsigned long uhash_entries;
2145 static int __init set_uhash_entries(char *str)
2146 {
2147 if (!str)
2148 return 0;
2149 uhash_entries = simple_strtoul(str, &str, 0);
2150 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2151 uhash_entries = UDP_HTABLE_SIZE_MIN;
2152 return 1;
2153 }
2154 __setup("uhash_entries=", set_uhash_entries);
2155
2156 void __init udp_table_init(struct udp_table *table, const char *name)
2157 {
2158 unsigned int i;
2159
2160 if (!CONFIG_BASE_SMALL)
2161 table->hash = alloc_large_system_hash(name,
2162 2 * sizeof(struct udp_hslot),
2163 uhash_entries,
2164 21, /* one slot per 2 MB */
2165 0,
2166 &table->log,
2167 &table->mask,
2168 64 * 1024);
2169 /*
2170 * Make sure hash table has the minimum size
2171 */
2172 if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) {
2173 table->hash = kmalloc(UDP_HTABLE_SIZE_MIN *
2174 2 * sizeof(struct udp_hslot), GFP_KERNEL);
2175 if (!table->hash)
2176 panic(name);
2177 table->log = ilog2(UDP_HTABLE_SIZE_MIN);
2178 table->mask = UDP_HTABLE_SIZE_MIN - 1;
2179 }
2180 table->hash2 = table->hash + (table->mask + 1);
2181 for (i = 0; i <= table->mask; i++) {
2182 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2183 table->hash[i].count = 0;
2184 spin_lock_init(&table->hash[i].lock);
2185 }
2186 for (i = 0; i <= table->mask; i++) {
2187 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2188 table->hash2[i].count = 0;
2189 spin_lock_init(&table->hash2[i].lock);
2190 }
2191 }
2192
2193 void __init udp_init(void)
2194 {
2195 unsigned long nr_pages, limit;
2196
2197 udp_table_init(&udp_table, "UDP");
2198 /* Set the pressure threshold up by the same strategy of TCP. It is a
2199 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
2200 * toward zero with the amount of memory, with a floor of 128 pages.
2201 */
2202 nr_pages = totalram_pages - totalhigh_pages;
2203 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
2204 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
2205 limit = max(limit, 128UL);
2206 sysctl_udp_mem[0] = limit / 4 * 3;
2207 sysctl_udp_mem[1] = limit;
2208 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2209
2210 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2211 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2212 }
2213
2214 int udp4_ufo_send_check(struct sk_buff *skb)
2215 {
2216 const struct iphdr *iph;
2217 struct udphdr *uh;
2218
2219 if (!pskb_may_pull(skb, sizeof(*uh)))
2220 return -EINVAL;
2221
2222 iph = ip_hdr(skb);
2223 uh = udp_hdr(skb);
2224
2225 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
2226 IPPROTO_UDP, 0);
2227 skb->csum_start = skb_transport_header(skb) - skb->head;
2228 skb->csum_offset = offsetof(struct udphdr, check);
2229 skb->ip_summed = CHECKSUM_PARTIAL;
2230 return 0;
2231 }
2232
2233 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, u32 features)
2234 {
2235 struct sk_buff *segs = ERR_PTR(-EINVAL);
2236 unsigned int mss;
2237 int offset;
2238 __wsum csum;
2239
2240 mss = skb_shinfo(skb)->gso_size;
2241 if (unlikely(skb->len <= mss))
2242 goto out;
2243
2244 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2245 /* Packet is from an untrusted source, reset gso_segs. */
2246 int type = skb_shinfo(skb)->gso_type;
2247
2248 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
2249 !(type & (SKB_GSO_UDP))))
2250 goto out;
2251
2252 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
2253
2254 segs = NULL;
2255 goto out;
2256 }
2257
2258 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
2259 * do checksum of UDP packets sent as multiple IP fragments.
2260 */
2261 offset = skb_checksum_start_offset(skb);
2262 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2263 offset += skb->csum_offset;
2264 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2265 skb->ip_summed = CHECKSUM_NONE;
2266
2267 /* Fragment the skb. IP headers of the fragments are updated in
2268 * inet_gso_segment()
2269 */
2270 segs = skb_segment(skb, features);
2271 out:
2272 return segs;
2273 }
2274
Linux kernel - Deliverables
This page took 0.204981 seconds and 5 git commands to generate.