projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ipv4: Use flowi4 in public route lookup interfaces.
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / net / ipv4 / udp.c
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * The User Datagram Protocol (UDP).
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
11 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
12 * Hirokazu Takahashi, <taka@valinux.co.jp>
13 *
14 * Fixes:
15 * Alan Cox : verify_area() calls
16 * Alan Cox : stopped close while in use off icmp
17 * messages. Not a fix but a botch that
18 * for udp at least is 'valid'.
19 * Alan Cox : Fixed icmp handling properly
20 * Alan Cox : Correct error for oversized datagrams
21 * Alan Cox : Tidied select() semantics.
22 * Alan Cox : udp_err() fixed properly, also now
23 * select and read wake correctly on errors
24 * Alan Cox : udp_send verify_area moved to avoid mem leak
25 * Alan Cox : UDP can count its memory
26 * Alan Cox : send to an unknown connection causes
27 * an ECONNREFUSED off the icmp, but
28 * does NOT close.
29 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
31 * bug no longer crashes it.
32 * Fred Van Kempen : Net2e support for sk->broadcast.
33 * Alan Cox : Uses skb_free_datagram
34 * Alan Cox : Added get/set sockopt support.
35 * Alan Cox : Broadcasting without option set returns EACCES.
36 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
37 * Alan Cox : Use ip_tos and ip_ttl
38 * Alan Cox : SNMP Mibs
39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
40 * Matt Dillon : UDP length checks.
41 * Alan Cox : Smarter af_inet used properly.
42 * Alan Cox : Use new kernel side addressing.
43 * Alan Cox : Incorrect return on truncated datagram receive.
44 * Arnt Gulbrandsen : New udp_send and stuff
45 * Alan Cox : Cache last socket
46 * Alan Cox : Route cache
47 * Jon Peatfield : Minor efficiency fix to sendto().
48 * Mike Shaver : RFC1122 checks.
49 * Alan Cox : Nonblocking error fix.
50 * Willy Konynenberg : Transparent proxying support.
51 * Mike McLagan : Routing by source
52 * David S. Miller : New socket lookup architecture.
53 * Last socket cache retained as it
54 * does have a high hit rate.
55 * Olaf Kirch : Don't linearise iovec on sendmsg.
56 * Andi Kleen : Some cleanups, cache destination entry
57 * for connect.
58 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
59 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
60 * return ENOTCONN for unconnected sockets (POSIX)
61 * Janos Farkas : don't deliver multi/broadcasts to a different
62 * bound-to-device socket
63 * Hirokazu Takahashi : HW checksumming for outgoing UDP
64 * datagrams.
65 * Hirokazu Takahashi : sendfile() on UDP works now.
66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
69 * a single port at the same time.
70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
71 * James Chapman : Add L2TP encapsulation type.
72 *
73 *
74 * This program is free software; you can redistribute it and/or
75 * modify it under the terms of the GNU General Public License
76 * as published by the Free Software Foundation; either version
77 * 2 of the License, or (at your option) any later version.
78 */
79
80 #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-dependant 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 struct iphdr *iph = (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 flowi *fl = &inet->cork.fl;
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, fl->fl4_dst, fl->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 int ulen = len;
795 struct ipcm_cookie ipc;
796 struct rtable *rt = NULL;
797 int free = 0;
798 int connected = 0;
799 __be32 daddr, faddr, saddr;
800 __be16 dport;
801 u8 tos;
802 int err, is_udplite = IS_UDPLITE(sk);
803 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
804 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
805 struct sk_buff *skb;
806
807 if (len > 0xFFFF)
808 return -EMSGSIZE;
809
810 /*
811 * Check the flags.
812 */
813
814 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
815 return -EOPNOTSUPP;
816
817 ipc.opt = NULL;
818 ipc.tx_flags = 0;
819
820 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
821
822 if (up->pending) {
823 /*
824 * There are pending frames.
825 * The socket lock must be held while it's corked.
826 */
827 lock_sock(sk);
828 if (likely(up->pending)) {
829 if (unlikely(up->pending != AF_INET)) {
830 release_sock(sk);
831 return -EINVAL;
832 }
833 goto do_append_data;
834 }
835 release_sock(sk);
836 }
837 ulen += sizeof(struct udphdr);
838
839 /*
840 * Get and verify the address.
841 */
842 if (msg->msg_name) {
843 struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name;
844 if (msg->msg_namelen < sizeof(*usin))
845 return -EINVAL;
846 if (usin->sin_family != AF_INET) {
847 if (usin->sin_family != AF_UNSPEC)
848 return -EAFNOSUPPORT;
849 }
850
851 daddr = usin->sin_addr.s_addr;
852 dport = usin->sin_port;
853 if (dport == 0)
854 return -EINVAL;
855 } else {
856 if (sk->sk_state != TCP_ESTABLISHED)
857 return -EDESTADDRREQ;
858 daddr = inet->inet_daddr;
859 dport = inet->inet_dport;
860 /* Open fast path for connected socket.
861 Route will not be used, if at least one option is set.
862 */
863 connected = 1;
864 }
865 ipc.addr = inet->inet_saddr;
866
867 ipc.oif = sk->sk_bound_dev_if;
868 err = sock_tx_timestamp(sk, &ipc.tx_flags);
869 if (err)
870 return err;
871 if (msg->msg_controllen) {
872 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
873 if (err)
874 return err;
875 if (ipc.opt)
876 free = 1;
877 connected = 0;
878 }
879 if (!ipc.opt)
880 ipc.opt = inet->opt;
881
882 saddr = ipc.addr;
883 ipc.addr = faddr = daddr;
884
885 if (ipc.opt && ipc.opt->srr) {
886 if (!daddr)
887 return -EINVAL;
888 faddr = ipc.opt->faddr;
889 connected = 0;
890 }
891 tos = RT_TOS(inet->tos);
892 if (sock_flag(sk, SOCK_LOCALROUTE) ||
893 (msg->msg_flags & MSG_DONTROUTE) ||
894 (ipc.opt && ipc.opt->is_strictroute)) {
895 tos |= RTO_ONLINK;
896 connected = 0;
897 }
898
899 if (ipv4_is_multicast(daddr)) {
900 if (!ipc.oif)
901 ipc.oif = inet->mc_index;
902 if (!saddr)
903 saddr = inet->mc_addr;
904 connected = 0;
905 }
906
907 if (connected)
908 rt = (struct rtable *)sk_dst_check(sk, 0);
909
910 if (rt == NULL) {
911 struct flowi4 fl4 = {
912 .flowi4_oif = ipc.oif,
913 .flowi4_mark = sk->sk_mark,
914 .daddr = faddr,
915 .saddr = saddr,
916 .flowi4_tos = tos,
917 .flowi4_proto = sk->sk_protocol,
918 .flowi4_flags = (inet_sk_flowi_flags(sk) |
919 FLOWI_FLAG_CAN_SLEEP),
920 .uli.ports.sport = inet->inet_sport,
921 .uli.ports.dport = dport,
922 };
923 struct net *net = sock_net(sk);
924
925 security_sk_classify_flow(sk, flowi4_to_flowi(&fl4));
926 rt = ip_route_output_flow(net, &fl4, sk);
927 if (IS_ERR(rt)) {
928 err = PTR_ERR(rt);
929 rt = NULL;
930 if (err == -ENETUNREACH)
931 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
932 goto out;
933 }
934
935 err = -EACCES;
936 if ((rt->rt_flags & RTCF_BROADCAST) &&
937 !sock_flag(sk, SOCK_BROADCAST))
938 goto out;
939 if (connected)
940 sk_dst_set(sk, dst_clone(&rt->dst));
941 }
942
943 if (msg->msg_flags&MSG_CONFIRM)
944 goto do_confirm;
945 back_from_confirm:
946
947 saddr = rt->rt_src;
948 if (!ipc.addr)
949 daddr = ipc.addr = rt->rt_dst;
950
951 /* Lockless fast path for the non-corking case. */
952 if (!corkreq) {
953 skb = ip_make_skb(sk, getfrag, msg->msg_iov, ulen,
954 sizeof(struct udphdr), &ipc, &rt,
955 msg->msg_flags);
956 err = PTR_ERR(skb);
957 if (skb && !IS_ERR(skb))
958 err = udp_send_skb(skb, daddr, dport);
959 goto out;
960 }
961
962 lock_sock(sk);
963 if (unlikely(up->pending)) {
964 /* The socket is already corked while preparing it. */
965 /* ... which is an evident application bug. --ANK */
966 release_sock(sk);
967
968 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
969 err = -EINVAL;
970 goto out;
971 }
972 /*
973 * Now cork the socket to pend data.
974 */
975 inet->cork.fl.fl4_dst = daddr;
976 inet->cork.fl.fl4_dport = dport;
977 inet->cork.fl.fl4_src = saddr;
978 inet->cork.fl.fl4_sport = inet->inet_sport;
979 up->pending = AF_INET;
980
981 do_append_data:
982 up->len += ulen;
983 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
984 sizeof(struct udphdr), &ipc, &rt,
985 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
986 if (err)
987 udp_flush_pending_frames(sk);
988 else if (!corkreq)
989 err = udp_push_pending_frames(sk);
990 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
991 up->pending = 0;
992 release_sock(sk);
993
994 out:
995 ip_rt_put(rt);
996 if (free)
997 kfree(ipc.opt);
998 if (!err)
999 return len;
1000 /*
1001 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1002 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1003 * we don't have a good statistic (IpOutDiscards but it can be too many
1004 * things). We could add another new stat but at least for now that
1005 * seems like overkill.
1006 */
1007 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1008 UDP_INC_STATS_USER(sock_net(sk),
1009 UDP_MIB_SNDBUFERRORS, is_udplite);
1010 }
1011 return err;
1012
1013 do_confirm:
1014 dst_confirm(&rt->dst);
1015 if (!(msg->msg_flags&MSG_PROBE) || len)
1016 goto back_from_confirm;
1017 err = 0;
1018 goto out;
1019 }
1020 EXPORT_SYMBOL(udp_sendmsg);
1021
1022 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1023 size_t size, int flags)
1024 {
1025 struct udp_sock *up = udp_sk(sk);
1026 int ret;
1027
1028 if (!up->pending) {
1029 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
1030
1031 /* Call udp_sendmsg to specify destination address which
1032 * sendpage interface can't pass.
1033 * This will succeed only when the socket is connected.
1034 */
1035 ret = udp_sendmsg(NULL, sk, &msg, 0);
1036 if (ret < 0)
1037 return ret;
1038 }
1039
1040 lock_sock(sk);
1041
1042 if (unlikely(!up->pending)) {
1043 release_sock(sk);
1044
1045 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
1046 return -EINVAL;
1047 }
1048
1049 ret = ip_append_page(sk, page, offset, size, flags);
1050 if (ret == -EOPNOTSUPP) {
1051 release_sock(sk);
1052 return sock_no_sendpage(sk->sk_socket, page, offset,
1053 size, flags);
1054 }
1055 if (ret < 0) {
1056 udp_flush_pending_frames(sk);
1057 goto out;
1058 }
1059
1060 up->len += size;
1061 if (!(up->corkflag || (flags&MSG_MORE)))
1062 ret = udp_push_pending_frames(sk);
1063 if (!ret)
1064 ret = size;
1065 out:
1066 release_sock(sk);
1067 return ret;
1068 }
1069
1070
1071 /**
1072 * first_packet_length - return length of first packet in receive queue
1073 * @sk: socket
1074 *
1075 * Drops all bad checksum frames, until a valid one is found.
1076 * Returns the length of found skb, or 0 if none is found.
1077 */
1078 static unsigned int first_packet_length(struct sock *sk)
1079 {
1080 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1081 struct sk_buff *skb;
1082 unsigned int res;
1083
1084 __skb_queue_head_init(&list_kill);
1085
1086 spin_lock_bh(&rcvq->lock);
1087 while ((skb = skb_peek(rcvq)) != NULL &&
1088 udp_lib_checksum_complete(skb)) {
1089 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1090 IS_UDPLITE(sk));
1091 atomic_inc(&sk->sk_drops);
1092 __skb_unlink(skb, rcvq);
1093 __skb_queue_tail(&list_kill, skb);
1094 }
1095 res = skb ? skb->len : 0;
1096 spin_unlock_bh(&rcvq->lock);
1097
1098 if (!skb_queue_empty(&list_kill)) {
1099 bool slow = lock_sock_fast(sk);
1100
1101 __skb_queue_purge(&list_kill);
1102 sk_mem_reclaim_partial(sk);
1103 unlock_sock_fast(sk, slow);
1104 }
1105 return res;
1106 }
1107
1108 /*
1109 * IOCTL requests applicable to the UDP protocol
1110 */
1111
1112 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1113 {
1114 switch (cmd) {
1115 case SIOCOUTQ:
1116 {
1117 int amount = sk_wmem_alloc_get(sk);
1118
1119 return put_user(amount, (int __user *)arg);
1120 }
1121
1122 case SIOCINQ:
1123 {
1124 unsigned int amount = first_packet_length(sk);
1125
1126 if (amount)
1127 /*
1128 * We will only return the amount
1129 * of this packet since that is all
1130 * that will be read.
1131 */
1132 amount -= sizeof(struct udphdr);
1133
1134 return put_user(amount, (int __user *)arg);
1135 }
1136
1137 default:
1138 return -ENOIOCTLCMD;
1139 }
1140
1141 return 0;
1142 }
1143 EXPORT_SYMBOL(udp_ioctl);
1144
1145 /*
1146 * This should be easy, if there is something there we
1147 * return it, otherwise we block.
1148 */
1149
1150 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1151 size_t len, int noblock, int flags, int *addr_len)
1152 {
1153 struct inet_sock *inet = inet_sk(sk);
1154 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1155 struct sk_buff *skb;
1156 unsigned int ulen;
1157 int peeked;
1158 int err;
1159 int is_udplite = IS_UDPLITE(sk);
1160 bool slow;
1161
1162 /*
1163 * Check any passed addresses
1164 */
1165 if (addr_len)
1166 *addr_len = sizeof(*sin);
1167
1168 if (flags & MSG_ERRQUEUE)
1169 return ip_recv_error(sk, msg, len);
1170
1171 try_again:
1172 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1173 &peeked, &err);
1174 if (!skb)
1175 goto out;
1176
1177 ulen = skb->len - sizeof(struct udphdr);
1178 if (len > ulen)
1179 len = ulen;
1180 else if (len < ulen)
1181 msg->msg_flags |= MSG_TRUNC;
1182
1183 /*
1184 * If checksum is needed at all, try to do it while copying the
1185 * data. If the data is truncated, or if we only want a partial
1186 * coverage checksum (UDP-Lite), do it before the copy.
1187 */
1188
1189 if (len < ulen || UDP_SKB_CB(skb)->partial_cov) {
1190 if (udp_lib_checksum_complete(skb))
1191 goto csum_copy_err;
1192 }
1193
1194 if (skb_csum_unnecessary(skb))
1195 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1196 msg->msg_iov, len);
1197 else {
1198 err = skb_copy_and_csum_datagram_iovec(skb,
1199 sizeof(struct udphdr),
1200 msg->msg_iov);
1201
1202 if (err == -EINVAL)
1203 goto csum_copy_err;
1204 }
1205
1206 if (err)
1207 goto out_free;
1208
1209 if (!peeked)
1210 UDP_INC_STATS_USER(sock_net(sk),
1211 UDP_MIB_INDATAGRAMS, is_udplite);
1212
1213 sock_recv_ts_and_drops(msg, sk, skb);
1214
1215 /* Copy the address. */
1216 if (sin) {
1217 sin->sin_family = AF_INET;
1218 sin->sin_port = udp_hdr(skb)->source;
1219 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1220 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1221 }
1222 if (inet->cmsg_flags)
1223 ip_cmsg_recv(msg, skb);
1224
1225 err = len;
1226 if (flags & MSG_TRUNC)
1227 err = ulen;
1228
1229 out_free:
1230 skb_free_datagram_locked(sk, skb);
1231 out:
1232 return err;
1233
1234 csum_copy_err:
1235 slow = lock_sock_fast(sk);
1236 if (!skb_kill_datagram(sk, skb, flags))
1237 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1238 unlock_sock_fast(sk, slow);
1239
1240 if (noblock)
1241 return -EAGAIN;
1242 goto try_again;
1243 }
1244
1245
1246 int udp_disconnect(struct sock *sk, int flags)
1247 {
1248 struct inet_sock *inet = inet_sk(sk);
1249 /*
1250 * 1003.1g - break association.
1251 */
1252
1253 sk->sk_state = TCP_CLOSE;
1254 inet->inet_daddr = 0;
1255 inet->inet_dport = 0;
1256 sock_rps_save_rxhash(sk, 0);
1257 sk->sk_bound_dev_if = 0;
1258 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1259 inet_reset_saddr(sk);
1260
1261 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1262 sk->sk_prot->unhash(sk);
1263 inet->inet_sport = 0;
1264 }
1265 sk_dst_reset(sk);
1266 return 0;
1267 }
1268 EXPORT_SYMBOL(udp_disconnect);
1269
1270 void udp_lib_unhash(struct sock *sk)
1271 {
1272 if (sk_hashed(sk)) {
1273 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1274 struct udp_hslot *hslot, *hslot2;
1275
1276 hslot = udp_hashslot(udptable, sock_net(sk),
1277 udp_sk(sk)->udp_port_hash);
1278 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1279
1280 spin_lock_bh(&hslot->lock);
1281 if (sk_nulls_del_node_init_rcu(sk)) {
1282 hslot->count--;
1283 inet_sk(sk)->inet_num = 0;
1284 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1285
1286 spin_lock(&hslot2->lock);
1287 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1288 hslot2->count--;
1289 spin_unlock(&hslot2->lock);
1290 }
1291 spin_unlock_bh(&hslot->lock);
1292 }
1293 }
1294 EXPORT_SYMBOL(udp_lib_unhash);
1295
1296 /*
1297 * inet_rcv_saddr was changed, we must rehash secondary hash
1298 */
1299 void udp_lib_rehash(struct sock *sk, u16 newhash)
1300 {
1301 if (sk_hashed(sk)) {
1302 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1303 struct udp_hslot *hslot, *hslot2, *nhslot2;
1304
1305 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1306 nhslot2 = udp_hashslot2(udptable, newhash);
1307 udp_sk(sk)->udp_portaddr_hash = newhash;
1308 if (hslot2 != nhslot2) {
1309 hslot = udp_hashslot(udptable, sock_net(sk),
1310 udp_sk(sk)->udp_port_hash);
1311 /* we must lock primary chain too */
1312 spin_lock_bh(&hslot->lock);
1313
1314 spin_lock(&hslot2->lock);
1315 hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1316 hslot2->count--;
1317 spin_unlock(&hslot2->lock);
1318
1319 spin_lock(&nhslot2->lock);
1320 hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1321 &nhslot2->head);
1322 nhslot2->count++;
1323 spin_unlock(&nhslot2->lock);
1324
1325 spin_unlock_bh(&hslot->lock);
1326 }
1327 }
1328 }
1329 EXPORT_SYMBOL(udp_lib_rehash);
1330
1331 static void udp_v4_rehash(struct sock *sk)
1332 {
1333 u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1334 inet_sk(sk)->inet_rcv_saddr,
1335 inet_sk(sk)->inet_num);
1336 udp_lib_rehash(sk, new_hash);
1337 }
1338
1339 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1340 {
1341 int rc;
1342
1343 if (inet_sk(sk)->inet_daddr)
1344 sock_rps_save_rxhash(sk, skb->rxhash);
1345
1346 rc = ip_queue_rcv_skb(sk, skb);
1347 if (rc < 0) {
1348 int is_udplite = IS_UDPLITE(sk);
1349
1350 /* Note that an ENOMEM error is charged twice */
1351 if (rc == -ENOMEM)
1352 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1353 is_udplite);
1354 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1355 kfree_skb(skb);
1356 return -1;
1357 }
1358
1359 return 0;
1360
1361 }
1362
1363 /* returns:
1364 * -1: error
1365 * 0: success
1366 * >0: "udp encap" protocol resubmission
1367 *
1368 * Note that in the success and error cases, the skb is assumed to
1369 * have either been requeued or freed.
1370 */
1371 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1372 {
1373 struct udp_sock *up = udp_sk(sk);
1374 int rc;
1375 int is_udplite = IS_UDPLITE(sk);
1376
1377 /*
1378 * Charge it to the socket, dropping if the queue is full.
1379 */
1380 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1381 goto drop;
1382 nf_reset(skb);
1383
1384 if (up->encap_type) {
1385 /*
1386 * This is an encapsulation socket so pass the skb to
1387 * the socket's udp_encap_rcv() hook. Otherwise, just
1388 * fall through and pass this up the UDP socket.
1389 * up->encap_rcv() returns the following value:
1390 * =0 if skb was successfully passed to the encap
1391 * handler or was discarded by it.
1392 * >0 if skb should be passed on to UDP.
1393 * <0 if skb should be resubmitted as proto -N
1394 */
1395
1396 /* if we're overly short, let UDP handle it */
1397 if (skb->len > sizeof(struct udphdr) &&
1398 up->encap_rcv != NULL) {
1399 int ret;
1400
1401 ret = (*up->encap_rcv)(sk, skb);
1402 if (ret <= 0) {
1403 UDP_INC_STATS_BH(sock_net(sk),
1404 UDP_MIB_INDATAGRAMS,
1405 is_udplite);
1406 return -ret;
1407 }
1408 }
1409
1410 /* FALLTHROUGH -- it's a UDP Packet */
1411 }
1412
1413 /*
1414 * UDP-Lite specific tests, ignored on UDP sockets
1415 */
1416 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1417
1418 /*
1419 * MIB statistics other than incrementing the error count are
1420 * disabled for the following two types of errors: these depend
1421 * on the application settings, not on the functioning of the
1422 * protocol stack as such.
1423 *
1424 * RFC 3828 here recommends (sec 3.3): "There should also be a
1425 * way ... to ... at least let the receiving application block
1426 * delivery of packets with coverage values less than a value
1427 * provided by the application."
1428 */
1429 if (up->pcrlen == 0) { /* full coverage was set */
1430 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1431 "%d while full coverage %d requested\n",
1432 UDP_SKB_CB(skb)->cscov, skb->len);
1433 goto drop;
1434 }
1435 /* The next case involves violating the min. coverage requested
1436 * by the receiver. This is subtle: if receiver wants x and x is
1437 * greater than the buffersize/MTU then receiver will complain
1438 * that it wants x while sender emits packets of smaller size y.
1439 * Therefore the above ...()->partial_cov statement is essential.
1440 */
1441 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1442 LIMIT_NETDEBUG(KERN_WARNING
1443 "UDPLITE: coverage %d too small, need min %d\n",
1444 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1445 goto drop;
1446 }
1447 }
1448
1449 if (rcu_dereference_raw(sk->sk_filter)) {
1450 if (udp_lib_checksum_complete(skb))
1451 goto drop;
1452 }
1453
1454
1455 if (sk_rcvqueues_full(sk, skb))
1456 goto drop;
1457
1458 rc = 0;
1459
1460 bh_lock_sock(sk);
1461 if (!sock_owned_by_user(sk))
1462 rc = __udp_queue_rcv_skb(sk, skb);
1463 else if (sk_add_backlog(sk, skb)) {
1464 bh_unlock_sock(sk);
1465 goto drop;
1466 }
1467 bh_unlock_sock(sk);
1468
1469 return rc;
1470
1471 drop:
1472 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1473 atomic_inc(&sk->sk_drops);
1474 kfree_skb(skb);
1475 return -1;
1476 }
1477
1478
1479 static void flush_stack(struct sock **stack, unsigned int count,
1480 struct sk_buff *skb, unsigned int final)
1481 {
1482 unsigned int i;
1483 struct sk_buff *skb1 = NULL;
1484 struct sock *sk;
1485
1486 for (i = 0; i < count; i++) {
1487 sk = stack[i];
1488 if (likely(skb1 == NULL))
1489 skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1490
1491 if (!skb1) {
1492 atomic_inc(&sk->sk_drops);
1493 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1494 IS_UDPLITE(sk));
1495 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1496 IS_UDPLITE(sk));
1497 }
1498
1499 if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1500 skb1 = NULL;
1501 }
1502 if (unlikely(skb1))
1503 kfree_skb(skb1);
1504 }
1505
1506 /*
1507 * Multicasts and broadcasts go to each listener.
1508 *
1509 * Note: called only from the BH handler context.
1510 */
1511 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1512 struct udphdr *uh,
1513 __be32 saddr, __be32 daddr,
1514 struct udp_table *udptable)
1515 {
1516 struct sock *sk, *stack[256 / sizeof(struct sock *)];
1517 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1518 int dif;
1519 unsigned int i, count = 0;
1520
1521 spin_lock(&hslot->lock);
1522 sk = sk_nulls_head(&hslot->head);
1523 dif = skb->dev->ifindex;
1524 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1525 while (sk) {
1526 stack[count++] = sk;
1527 sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1528 daddr, uh->source, saddr, dif);
1529 if (unlikely(count == ARRAY_SIZE(stack))) {
1530 if (!sk)
1531 break;
1532 flush_stack(stack, count, skb, ~0);
1533 count = 0;
1534 }
1535 }
1536 /*
1537 * before releasing chain lock, we must take a reference on sockets
1538 */
1539 for (i = 0; i < count; i++)
1540 sock_hold(stack[i]);
1541
1542 spin_unlock(&hslot->lock);
1543
1544 /*
1545 * do the slow work with no lock held
1546 */
1547 if (count) {
1548 flush_stack(stack, count, skb, count - 1);
1549
1550 for (i = 0; i < count; i++)
1551 sock_put(stack[i]);
1552 } else {
1553 kfree_skb(skb);
1554 }
1555 return 0;
1556 }
1557
1558 /* Initialize UDP checksum. If exited with zero value (success),
1559 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1560 * Otherwise, csum completion requires chacksumming packet body,
1561 * including udp header and folding it to skb->csum.
1562 */
1563 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1564 int proto)
1565 {
1566 const struct iphdr *iph;
1567 int err;
1568
1569 UDP_SKB_CB(skb)->partial_cov = 0;
1570 UDP_SKB_CB(skb)->cscov = skb->len;
1571
1572 if (proto == IPPROTO_UDPLITE) {
1573 err = udplite_checksum_init(skb, uh);
1574 if (err)
1575 return err;
1576 }
1577
1578 iph = ip_hdr(skb);
1579 if (uh->check == 0) {
1580 skb->ip_summed = CHECKSUM_UNNECESSARY;
1581 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1582 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1583 proto, skb->csum))
1584 skb->ip_summed = CHECKSUM_UNNECESSARY;
1585 }
1586 if (!skb_csum_unnecessary(skb))
1587 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1588 skb->len, proto, 0);
1589 /* Probably, we should checksum udp header (it should be in cache
1590 * in any case) and data in tiny packets (< rx copybreak).
1591 */
1592
1593 return 0;
1594 }
1595
1596 /*
1597 * All we need to do is get the socket, and then do a checksum.
1598 */
1599
1600 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1601 int proto)
1602 {
1603 struct sock *sk;
1604 struct udphdr *uh;
1605 unsigned short ulen;
1606 struct rtable *rt = skb_rtable(skb);
1607 __be32 saddr, daddr;
1608 struct net *net = dev_net(skb->dev);
1609
1610 /*
1611 * Validate the packet.
1612 */
1613 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1614 goto drop; /* No space for header. */
1615
1616 uh = udp_hdr(skb);
1617 ulen = ntohs(uh->len);
1618 saddr = ip_hdr(skb)->saddr;
1619 daddr = ip_hdr(skb)->daddr;
1620
1621 if (ulen > skb->len)
1622 goto short_packet;
1623
1624 if (proto == IPPROTO_UDP) {
1625 /* UDP validates ulen. */
1626 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1627 goto short_packet;
1628 uh = udp_hdr(skb);
1629 }
1630
1631 if (udp4_csum_init(skb, uh, proto))
1632 goto csum_error;
1633
1634 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1635 return __udp4_lib_mcast_deliver(net, skb, uh,
1636 saddr, daddr, udptable);
1637
1638 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1639
1640 if (sk != NULL) {
1641 int ret = udp_queue_rcv_skb(sk, skb);
1642 sock_put(sk);
1643
1644 /* a return value > 0 means to resubmit the input, but
1645 * it wants the return to be -protocol, or 0
1646 */
1647 if (ret > 0)
1648 return -ret;
1649 return 0;
1650 }
1651
1652 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1653 goto drop;
1654 nf_reset(skb);
1655
1656 /* No socket. Drop packet silently, if checksum is wrong */
1657 if (udp_lib_checksum_complete(skb))
1658 goto csum_error;
1659
1660 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1661 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1662
1663 /*
1664 * Hmm. We got an UDP packet to a port to which we
1665 * don't wanna listen. Ignore it.
1666 */
1667 kfree_skb(skb);
1668 return 0;
1669
1670 short_packet:
1671 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1672 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1673 &saddr,
1674 ntohs(uh->source),
1675 ulen,
1676 skb->len,
1677 &daddr,
1678 ntohs(uh->dest));
1679 goto drop;
1680
1681 csum_error:
1682 /*
1683 * RFC1122: OK. Discards the bad packet silently (as far as
1684 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1685 */
1686 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1687 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1688 &saddr,
1689 ntohs(uh->source),
1690 &daddr,
1691 ntohs(uh->dest),
1692 ulen);
1693 drop:
1694 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1695 kfree_skb(skb);
1696 return 0;
1697 }
1698
1699 int udp_rcv(struct sk_buff *skb)
1700 {
1701 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1702 }
1703
1704 void udp_destroy_sock(struct sock *sk)
1705 {
1706 bool slow = lock_sock_fast(sk);
1707 udp_flush_pending_frames(sk);
1708 unlock_sock_fast(sk, slow);
1709 }
1710
1711 /*
1712 * Socket option code for UDP
1713 */
1714 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1715 char __user *optval, unsigned int optlen,
1716 int (*push_pending_frames)(struct sock *))
1717 {
1718 struct udp_sock *up = udp_sk(sk);
1719 int val;
1720 int err = 0;
1721 int is_udplite = IS_UDPLITE(sk);
1722
1723 if (optlen < sizeof(int))
1724 return -EINVAL;
1725
1726 if (get_user(val, (int __user *)optval))
1727 return -EFAULT;
1728
1729 switch (optname) {
1730 case UDP_CORK:
1731 if (val != 0) {
1732 up->corkflag = 1;
1733 } else {
1734 up->corkflag = 0;
1735 lock_sock(sk);
1736 (*push_pending_frames)(sk);
1737 release_sock(sk);
1738 }
1739 break;
1740
1741 case UDP_ENCAP:
1742 switch (val) {
1743 case 0:
1744 case UDP_ENCAP_ESPINUDP:
1745 case UDP_ENCAP_ESPINUDP_NON_IKE:
1746 up->encap_rcv = xfrm4_udp_encap_rcv;
1747 /* FALLTHROUGH */
1748 case UDP_ENCAP_L2TPINUDP:
1749 up->encap_type = val;
1750 break;
1751 default:
1752 err = -ENOPROTOOPT;
1753 break;
1754 }
1755 break;
1756
1757 /*
1758 * UDP-Lite's partial checksum coverage (RFC 3828).
1759 */
1760 /* The sender sets actual checksum coverage length via this option.
1761 * The case coverage > packet length is handled by send module. */
1762 case UDPLITE_SEND_CSCOV:
1763 if (!is_udplite) /* Disable the option on UDP sockets */
1764 return -ENOPROTOOPT;
1765 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1766 val = 8;
1767 else if (val > USHRT_MAX)
1768 val = USHRT_MAX;
1769 up->pcslen = val;
1770 up->pcflag |= UDPLITE_SEND_CC;
1771 break;
1772
1773 /* The receiver specifies a minimum checksum coverage value. To make
1774 * sense, this should be set to at least 8 (as done below). If zero is
1775 * used, this again means full checksum coverage. */
1776 case UDPLITE_RECV_CSCOV:
1777 if (!is_udplite) /* Disable the option on UDP sockets */
1778 return -ENOPROTOOPT;
1779 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1780 val = 8;
1781 else if (val > USHRT_MAX)
1782 val = USHRT_MAX;
1783 up->pcrlen = val;
1784 up->pcflag |= UDPLITE_RECV_CC;
1785 break;
1786
1787 default:
1788 err = -ENOPROTOOPT;
1789 break;
1790 }
1791
1792 return err;
1793 }
1794 EXPORT_SYMBOL(udp_lib_setsockopt);
1795
1796 int udp_setsockopt(struct sock *sk, int level, int optname,
1797 char __user *optval, unsigned int optlen)
1798 {
1799 if (level == SOL_UDP || level == SOL_UDPLITE)
1800 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1801 udp_push_pending_frames);
1802 return ip_setsockopt(sk, level, optname, optval, optlen);
1803 }
1804
1805 #ifdef CONFIG_COMPAT
1806 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1807 char __user *optval, unsigned int optlen)
1808 {
1809 if (level == SOL_UDP || level == SOL_UDPLITE)
1810 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1811 udp_push_pending_frames);
1812 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1813 }
1814 #endif
1815
1816 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1817 char __user *optval, int __user *optlen)
1818 {
1819 struct udp_sock *up = udp_sk(sk);
1820 int val, len;
1821
1822 if (get_user(len, optlen))
1823 return -EFAULT;
1824
1825 len = min_t(unsigned int, len, sizeof(int));
1826
1827 if (len < 0)
1828 return -EINVAL;
1829
1830 switch (optname) {
1831 case UDP_CORK:
1832 val = up->corkflag;
1833 break;
1834
1835 case UDP_ENCAP:
1836 val = up->encap_type;
1837 break;
1838
1839 /* The following two cannot be changed on UDP sockets, the return is
1840 * always 0 (which corresponds to the full checksum coverage of UDP). */
1841 case UDPLITE_SEND_CSCOV:
1842 val = up->pcslen;
1843 break;
1844
1845 case UDPLITE_RECV_CSCOV:
1846 val = up->pcrlen;
1847 break;
1848
1849 default:
1850 return -ENOPROTOOPT;
1851 }
1852
1853 if (put_user(len, optlen))
1854 return -EFAULT;
1855 if (copy_to_user(optval, &val, len))
1856 return -EFAULT;
1857 return 0;
1858 }
1859 EXPORT_SYMBOL(udp_lib_getsockopt);
1860
1861 int udp_getsockopt(struct sock *sk, int level, int optname,
1862 char __user *optval, int __user *optlen)
1863 {
1864 if (level == SOL_UDP || level == SOL_UDPLITE)
1865 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1866 return ip_getsockopt(sk, level, optname, optval, optlen);
1867 }
1868
1869 #ifdef CONFIG_COMPAT
1870 int compat_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 compat_ip_getsockopt(sk, level, optname, optval, optlen);
1876 }
1877 #endif
1878 /**
1879 * udp_poll - wait for a UDP event.
1880 * @file - file struct
1881 * @sock - socket
1882 * @wait - poll table
1883 *
1884 * This is same as datagram poll, except for the special case of
1885 * blocking sockets. If application is using a blocking fd
1886 * and a packet with checksum error is in the queue;
1887 * then it could get return from select indicating data available
1888 * but then block when reading it. Add special case code
1889 * to work around these arguably broken applications.
1890 */
1891 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1892 {
1893 unsigned int mask = datagram_poll(file, sock, wait);
1894 struct sock *sk = sock->sk;
1895
1896 /* Check for false positives due to checksum errors */
1897 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1898 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1899 mask &= ~(POLLIN | POLLRDNORM);
1900
1901 return mask;
1902
1903 }
1904 EXPORT_SYMBOL(udp_poll);
1905
1906 struct proto udp_prot = {
1907 .name = "UDP",
1908 .owner = THIS_MODULE,
1909 .close = udp_lib_close,
1910 .connect = ip4_datagram_connect,
1911 .disconnect = udp_disconnect,
1912 .ioctl = udp_ioctl,
1913 .destroy = udp_destroy_sock,
1914 .setsockopt = udp_setsockopt,
1915 .getsockopt = udp_getsockopt,
1916 .sendmsg = udp_sendmsg,
1917 .recvmsg = udp_recvmsg,
1918 .sendpage = udp_sendpage,
1919 .backlog_rcv = __udp_queue_rcv_skb,
1920 .hash = udp_lib_hash,
1921 .unhash = udp_lib_unhash,
1922 .rehash = udp_v4_rehash,
1923 .get_port = udp_v4_get_port,
1924 .memory_allocated = &udp_memory_allocated,
1925 .sysctl_mem = sysctl_udp_mem,
1926 .sysctl_wmem = &sysctl_udp_wmem_min,
1927 .sysctl_rmem = &sysctl_udp_rmem_min,
1928 .obj_size = sizeof(struct udp_sock),
1929 .slab_flags = SLAB_DESTROY_BY_RCU,
1930 .h.udp_table = &udp_table,
1931 #ifdef CONFIG_COMPAT
1932 .compat_setsockopt = compat_udp_setsockopt,
1933 .compat_getsockopt = compat_udp_getsockopt,
1934 #endif
1935 .clear_sk = sk_prot_clear_portaddr_nulls,
1936 };
1937 EXPORT_SYMBOL(udp_prot);
1938
1939 /* ------------------------------------------------------------------------ */
1940 #ifdef CONFIG_PROC_FS
1941
1942 static struct sock *udp_get_first(struct seq_file *seq, int start)
1943 {
1944 struct sock *sk;
1945 struct udp_iter_state *state = seq->private;
1946 struct net *net = seq_file_net(seq);
1947
1948 for (state->bucket = start; state->bucket <= state->udp_table->mask;
1949 ++state->bucket) {
1950 struct hlist_nulls_node *node;
1951 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
1952
1953 if (hlist_nulls_empty(&hslot->head))
1954 continue;
1955
1956 spin_lock_bh(&hslot->lock);
1957 sk_nulls_for_each(sk, node, &hslot->head) {
1958 if (!net_eq(sock_net(sk), net))
1959 continue;
1960 if (sk->sk_family == state->family)
1961 goto found;
1962 }
1963 spin_unlock_bh(&hslot->lock);
1964 }
1965 sk = NULL;
1966 found:
1967 return sk;
1968 }
1969
1970 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1971 {
1972 struct udp_iter_state *state = seq->private;
1973 struct net *net = seq_file_net(seq);
1974
1975 do {
1976 sk = sk_nulls_next(sk);
1977 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1978
1979 if (!sk) {
1980 if (state->bucket <= state->udp_table->mask)
1981 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
1982 return udp_get_first(seq, state->bucket + 1);
1983 }
1984 return sk;
1985 }
1986
1987 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1988 {
1989 struct sock *sk = udp_get_first(seq, 0);
1990
1991 if (sk)
1992 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1993 --pos;
1994 return pos ? NULL : sk;
1995 }
1996
1997 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1998 {
1999 struct udp_iter_state *state = seq->private;
2000 state->bucket = MAX_UDP_PORTS;
2001
2002 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2003 }
2004
2005 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2006 {
2007 struct sock *sk;
2008
2009 if (v == SEQ_START_TOKEN)
2010 sk = udp_get_idx(seq, 0);
2011 else
2012 sk = udp_get_next(seq, v);
2013
2014 ++*pos;
2015 return sk;
2016 }
2017
2018 static void udp_seq_stop(struct seq_file *seq, void *v)
2019 {
2020 struct udp_iter_state *state = seq->private;
2021
2022 if (state->bucket <= state->udp_table->mask)
2023 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2024 }
2025
2026 static int udp_seq_open(struct inode *inode, struct file *file)
2027 {
2028 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
2029 struct udp_iter_state *s;
2030 int err;
2031
2032 err = seq_open_net(inode, file, &afinfo->seq_ops,
2033 sizeof(struct udp_iter_state));
2034 if (err < 0)
2035 return err;
2036
2037 s = ((struct seq_file *)file->private_data)->private;
2038 s->family = afinfo->family;
2039 s->udp_table = afinfo->udp_table;
2040 return err;
2041 }
2042
2043 /* ------------------------------------------------------------------------ */
2044 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2045 {
2046 struct proc_dir_entry *p;
2047 int rc = 0;
2048
2049 afinfo->seq_fops.open = udp_seq_open;
2050 afinfo->seq_fops.read = seq_read;
2051 afinfo->seq_fops.llseek = seq_lseek;
2052 afinfo->seq_fops.release = seq_release_net;
2053
2054 afinfo->seq_ops.start = udp_seq_start;
2055 afinfo->seq_ops.next = udp_seq_next;
2056 afinfo->seq_ops.stop = udp_seq_stop;
2057
2058 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2059 &afinfo->seq_fops, afinfo);
2060 if (!p)
2061 rc = -ENOMEM;
2062 return rc;
2063 }
2064 EXPORT_SYMBOL(udp_proc_register);
2065
2066 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2067 {
2068 proc_net_remove(net, afinfo->name);
2069 }
2070 EXPORT_SYMBOL(udp_proc_unregister);
2071
2072 /* ------------------------------------------------------------------------ */
2073 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2074 int bucket, int *len)
2075 {
2076 struct inet_sock *inet = inet_sk(sp);
2077 __be32 dest = inet->inet_daddr;
2078 __be32 src = inet->inet_rcv_saddr;
2079 __u16 destp = ntohs(inet->inet_dport);
2080 __u16 srcp = ntohs(inet->inet_sport);
2081
2082 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2083 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
2084 bucket, src, srcp, dest, destp, sp->sk_state,
2085 sk_wmem_alloc_get(sp),
2086 sk_rmem_alloc_get(sp),
2087 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
2088 atomic_read(&sp->sk_refcnt), sp,
2089 atomic_read(&sp->sk_drops), len);
2090 }
2091
2092 int udp4_seq_show(struct seq_file *seq, void *v)
2093 {
2094 if (v == SEQ_START_TOKEN)
2095 seq_printf(seq, "%-127s\n",
2096 " sl local_address rem_address st tx_queue "
2097 "rx_queue tr tm->when retrnsmt uid timeout "
2098 "inode ref pointer drops");
2099 else {
2100 struct udp_iter_state *state = seq->private;
2101 int len;
2102
2103 udp4_format_sock(v, seq, state->bucket, &len);
2104 seq_printf(seq, "%*s\n", 127 - len, "");
2105 }
2106 return 0;
2107 }
2108
2109 /* ------------------------------------------------------------------------ */
2110 static struct udp_seq_afinfo udp4_seq_afinfo = {
2111 .name = "udp",
2112 .family = AF_INET,
2113 .udp_table = &udp_table,
2114 .seq_fops = {
2115 .owner = THIS_MODULE,
2116 },
2117 .seq_ops = {
2118 .show = udp4_seq_show,
2119 },
2120 };
2121
2122 static int __net_init udp4_proc_init_net(struct net *net)
2123 {
2124 return udp_proc_register(net, &udp4_seq_afinfo);
2125 }
2126
2127 static void __net_exit udp4_proc_exit_net(struct net *net)
2128 {
2129 udp_proc_unregister(net, &udp4_seq_afinfo);
2130 }
2131
2132 static struct pernet_operations udp4_net_ops = {
2133 .init = udp4_proc_init_net,
2134 .exit = udp4_proc_exit_net,
2135 };
2136
2137 int __init udp4_proc_init(void)
2138 {
2139 return register_pernet_subsys(&udp4_net_ops);
2140 }
2141
2142 void udp4_proc_exit(void)
2143 {
2144 unregister_pernet_subsys(&udp4_net_ops);
2145 }
2146 #endif /* CONFIG_PROC_FS */
2147
2148 static __initdata unsigned long uhash_entries;
2149 static int __init set_uhash_entries(char *str)
2150 {
2151 if (!str)
2152 return 0;
2153 uhash_entries = simple_strtoul(str, &str, 0);
2154 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2155 uhash_entries = UDP_HTABLE_SIZE_MIN;
2156 return 1;
2157 }
2158 __setup("uhash_entries=", set_uhash_entries);
2159
2160 void __init udp_table_init(struct udp_table *table, const char *name)
2161 {
2162 unsigned int i;
2163
2164 if (!CONFIG_BASE_SMALL)
2165 table->hash = alloc_large_system_hash(name,
2166 2 * sizeof(struct udp_hslot),
2167 uhash_entries,
2168 21, /* one slot per 2 MB */
2169 0,
2170 &table->log,
2171 &table->mask,
2172 64 * 1024);
2173 /*
2174 * Make sure hash table has the minimum size
2175 */
2176 if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) {
2177 table->hash = kmalloc(UDP_HTABLE_SIZE_MIN *
2178 2 * sizeof(struct udp_hslot), GFP_KERNEL);
2179 if (!table->hash)
2180 panic(name);
2181 table->log = ilog2(UDP_HTABLE_SIZE_MIN);
2182 table->mask = UDP_HTABLE_SIZE_MIN - 1;
2183 }
2184 table->hash2 = table->hash + (table->mask + 1);
2185 for (i = 0; i <= table->mask; i++) {
2186 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2187 table->hash[i].count = 0;
2188 spin_lock_init(&table->hash[i].lock);
2189 }
2190 for (i = 0; i <= table->mask; i++) {
2191 INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2192 table->hash2[i].count = 0;
2193 spin_lock_init(&table->hash2[i].lock);
2194 }
2195 }
2196
2197 void __init udp_init(void)
2198 {
2199 unsigned long nr_pages, limit;
2200
2201 udp_table_init(&udp_table, "UDP");
2202 /* Set the pressure threshold up by the same strategy of TCP. It is a
2203 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
2204 * toward zero with the amount of memory, with a floor of 128 pages.
2205 */
2206 nr_pages = totalram_pages - totalhigh_pages;
2207 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
2208 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
2209 limit = max(limit, 128UL);
2210 sysctl_udp_mem[0] = limit / 4 * 3;
2211 sysctl_udp_mem[1] = limit;
2212 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2213
2214 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2215 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2216 }
2217
2218 int udp4_ufo_send_check(struct sk_buff *skb)
2219 {
2220 const struct iphdr *iph;
2221 struct udphdr *uh;
2222
2223 if (!pskb_may_pull(skb, sizeof(*uh)))
2224 return -EINVAL;
2225
2226 iph = ip_hdr(skb);
2227 uh = udp_hdr(skb);
2228
2229 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
2230 IPPROTO_UDP, 0);
2231 skb->csum_start = skb_transport_header(skb) - skb->head;
2232 skb->csum_offset = offsetof(struct udphdr, check);
2233 skb->ip_summed = CHECKSUM_PARTIAL;
2234 return 0;
2235 }
2236
2237 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, u32 features)
2238 {
2239 struct sk_buff *segs = ERR_PTR(-EINVAL);
2240 unsigned int mss;
2241 int offset;
2242 __wsum csum;
2243
2244 mss = skb_shinfo(skb)->gso_size;
2245 if (unlikely(skb->len <= mss))
2246 goto out;
2247
2248 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2249 /* Packet is from an untrusted source, reset gso_segs. */
2250 int type = skb_shinfo(skb)->gso_type;
2251
2252 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
2253 !(type & (SKB_GSO_UDP))))
2254 goto out;
2255
2256 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
2257
2258 segs = NULL;
2259 goto out;
2260 }
2261
2262 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
2263 * do checksum of UDP packets sent as multiple IP fragments.
2264 */
2265 offset = skb_checksum_start_offset(skb);
2266 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2267 offset += skb->csum_offset;
2268 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2269 skb->ip_summed = CHECKSUM_NONE;
2270
2271 /* Fragment the skb. IP headers of the fragments are updated in
2272 * inet_gso_segment()
2273 */
2274 segs = skb_segment(skb, features);
2275 out:
2276 return segs;
2277 }
2278
kernel source for telekom puls (alcatel/mediatek)