projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[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 <net/tcp_states.h>
99 #include <linux/skbuff.h>
100 #include <linux/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <net/net_namespace.h>
103 #include <net/icmp.h>
104 #include <net/route.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include "udp_impl.h"
108
109 struct udp_table udp_table __read_mostly;
110 EXPORT_SYMBOL(udp_table);
111
112 int sysctl_udp_mem[3] __read_mostly;
113 EXPORT_SYMBOL(sysctl_udp_mem);
114
115 int sysctl_udp_rmem_min __read_mostly;
116 EXPORT_SYMBOL(sysctl_udp_rmem_min);
117
118 int sysctl_udp_wmem_min __read_mostly;
119 EXPORT_SYMBOL(sysctl_udp_wmem_min);
120
121 atomic_t udp_memory_allocated;
122 EXPORT_SYMBOL(udp_memory_allocated);
123
124 #define MAX_UDP_PORTS 65536
125 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
126
127 static int udp_lib_lport_inuse(struct net *net, __u16 num,
128 const struct udp_hslot *hslot,
129 unsigned long *bitmap,
130 struct sock *sk,
131 int (*saddr_comp)(const struct sock *sk1,
132 const struct sock *sk2),
133 unsigned int log)
134 {
135 struct sock *sk2;
136 struct hlist_nulls_node *node;
137
138 sk_nulls_for_each(sk2, node, &hslot->head)
139 if (net_eq(sock_net(sk2), net) &&
140 sk2 != sk &&
141 (bitmap || sk2->sk_hash == num) &&
142 (!sk2->sk_reuse || !sk->sk_reuse) &&
143 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
144 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
145 (*saddr_comp)(sk, sk2)) {
146 if (bitmap)
147 __set_bit(sk2->sk_hash >> log, bitmap);
148 else
149 return 1;
150 }
151 return 0;
152 }
153
154 /**
155 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
156 *
157 * @sk: socket struct in question
158 * @snum: port number to look up
159 * @saddr_comp: AF-dependent comparison of bound local IP addresses
160 */
161 int udp_lib_get_port(struct sock *sk, unsigned short snum,
162 int (*saddr_comp)(const struct sock *sk1,
163 const struct sock *sk2))
164 {
165 struct udp_hslot *hslot;
166 struct udp_table *udptable = sk->sk_prot->h.udp_table;
167 int error = 1;
168 struct net *net = sock_net(sk);
169
170 if (!snum) {
171 int low, high, remaining;
172 unsigned rand;
173 unsigned short first, last;
174 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
175
176 inet_get_local_port_range(&low, &high);
177 remaining = (high - low) + 1;
178
179 rand = net_random();
180 first = (((u64)rand * remaining) >> 32) + low;
181 /*
182 * force rand to be an odd multiple of UDP_HTABLE_SIZE
183 */
184 rand = (rand | 1) * (udptable->mask + 1);
185 for (last = first + udptable->mask + 1;
186 first != last;
187 first++) {
188 hslot = udp_hashslot(udptable, net, first);
189 bitmap_zero(bitmap, PORTS_PER_CHAIN);
190 spin_lock_bh(&hslot->lock);
191 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
192 saddr_comp, udptable->log);
193
194 snum = first;
195 /*
196 * Iterate on all possible values of snum for this hash.
197 * Using steps of an odd multiple of UDP_HTABLE_SIZE
198 * give us randomization and full range coverage.
199 */
200 do {
201 if (low <= snum && snum <= high &&
202 !test_bit(snum >> udptable->log, bitmap))
203 goto found;
204 snum += rand;
205 } while (snum != first);
206 spin_unlock_bh(&hslot->lock);
207 }
208 goto fail;
209 } else {
210 hslot = udp_hashslot(udptable, net, snum);
211 spin_lock_bh(&hslot->lock);
212 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
213 saddr_comp, 0))
214 goto fail_unlock;
215 }
216 found:
217 inet_sk(sk)->inet_num = snum;
218 sk->sk_hash = snum;
219 if (sk_unhashed(sk)) {
220 sk_nulls_add_node_rcu(sk, &hslot->head);
221 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
222 }
223 error = 0;
224 fail_unlock:
225 spin_unlock_bh(&hslot->lock);
226 fail:
227 return error;
228 }
229 EXPORT_SYMBOL(udp_lib_get_port);
230
231 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
232 {
233 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
234
235 return (!ipv6_only_sock(sk2) &&
236 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
237 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
238 }
239
240 int udp_v4_get_port(struct sock *sk, unsigned short snum)
241 {
242 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal);
243 }
244
245 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
246 unsigned short hnum,
247 __be16 sport, __be32 daddr, __be16 dport, int dif)
248 {
249 int score = -1;
250
251 if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum &&
252 !ipv6_only_sock(sk)) {
253 struct inet_sock *inet = inet_sk(sk);
254
255 score = (sk->sk_family == PF_INET ? 1 : 0);
256 if (inet->inet_rcv_saddr) {
257 if (inet->inet_rcv_saddr != daddr)
258 return -1;
259 score += 2;
260 }
261 if (inet->inet_daddr) {
262 if (inet->inet_daddr != saddr)
263 return -1;
264 score += 2;
265 }
266 if (inet->inet_dport) {
267 if (inet->inet_dport != sport)
268 return -1;
269 score += 2;
270 }
271 if (sk->sk_bound_dev_if) {
272 if (sk->sk_bound_dev_if != dif)
273 return -1;
274 score += 2;
275 }
276 }
277 return score;
278 }
279
280 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
281 * harder than this. -DaveM
282 */
283 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
284 __be16 sport, __be32 daddr, __be16 dport,
285 int dif, struct udp_table *udptable)
286 {
287 struct sock *sk, *result;
288 struct hlist_nulls_node *node;
289 unsigned short hnum = ntohs(dport);
290 unsigned int hash = udp_hashfn(net, hnum, udptable->mask);
291 struct udp_hslot *hslot = &udptable->hash[hash];
292 int score, badness;
293
294 rcu_read_lock();
295 begin:
296 result = NULL;
297 badness = -1;
298 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
299 score = compute_score(sk, net, saddr, hnum, sport,
300 daddr, dport, dif);
301 if (score > badness) {
302 result = sk;
303 badness = score;
304 }
305 }
306 /*
307 * if the nulls value we got at the end of this lookup is
308 * not the expected one, we must restart lookup.
309 * We probably met an item that was moved to another chain.
310 */
311 if (get_nulls_value(node) != hash)
312 goto begin;
313
314 if (result) {
315 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
316 result = NULL;
317 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
318 daddr, dport, dif) < badness)) {
319 sock_put(result);
320 goto begin;
321 }
322 }
323 rcu_read_unlock();
324 return result;
325 }
326
327 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
328 __be16 sport, __be16 dport,
329 struct udp_table *udptable)
330 {
331 struct sock *sk;
332 const struct iphdr *iph = ip_hdr(skb);
333
334 if (unlikely(sk = skb_steal_sock(skb)))
335 return sk;
336 else
337 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
338 iph->daddr, dport, inet_iif(skb),
339 udptable);
340 }
341
342 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
343 __be32 daddr, __be16 dport, int dif)
344 {
345 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
346 }
347 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
348
349 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
350 __be16 loc_port, __be32 loc_addr,
351 __be16 rmt_port, __be32 rmt_addr,
352 int dif)
353 {
354 struct hlist_nulls_node *node;
355 struct sock *s = sk;
356 unsigned short hnum = ntohs(loc_port);
357
358 sk_nulls_for_each_from(s, node) {
359 struct inet_sock *inet = inet_sk(s);
360
361 if (!net_eq(sock_net(s), net) ||
362 s->sk_hash != hnum ||
363 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
364 (inet->inet_dport != rmt_port && inet->inet_dport) ||
365 (inet->inet_rcv_saddr &&
366 inet->inet_rcv_saddr != loc_addr) ||
367 ipv6_only_sock(s) ||
368 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
369 continue;
370 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
371 continue;
372 goto found;
373 }
374 s = NULL;
375 found:
376 return s;
377 }
378
379 /*
380 * This routine is called by the ICMP module when it gets some
381 * sort of error condition. If err < 0 then the socket should
382 * be closed and the error returned to the user. If err > 0
383 * it's just the icmp type << 8 | icmp code.
384 * Header points to the ip header of the error packet. We move
385 * on past this. Then (as it used to claim before adjustment)
386 * header points to the first 8 bytes of the udp header. We need
387 * to find the appropriate port.
388 */
389
390 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
391 {
392 struct inet_sock *inet;
393 struct iphdr *iph = (struct iphdr *)skb->data;
394 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
395 const int type = icmp_hdr(skb)->type;
396 const int code = icmp_hdr(skb)->code;
397 struct sock *sk;
398 int harderr;
399 int err;
400 struct net *net = dev_net(skb->dev);
401
402 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
403 iph->saddr, uh->source, skb->dev->ifindex, udptable);
404 if (sk == NULL) {
405 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
406 return; /* No socket for error */
407 }
408
409 err = 0;
410 harderr = 0;
411 inet = inet_sk(sk);
412
413 switch (type) {
414 default:
415 case ICMP_TIME_EXCEEDED:
416 err = EHOSTUNREACH;
417 break;
418 case ICMP_SOURCE_QUENCH:
419 goto out;
420 case ICMP_PARAMETERPROB:
421 err = EPROTO;
422 harderr = 1;
423 break;
424 case ICMP_DEST_UNREACH:
425 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
426 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
427 err = EMSGSIZE;
428 harderr = 1;
429 break;
430 }
431 goto out;
432 }
433 err = EHOSTUNREACH;
434 if (code <= NR_ICMP_UNREACH) {
435 harderr = icmp_err_convert[code].fatal;
436 err = icmp_err_convert[code].errno;
437 }
438 break;
439 }
440
441 /*
442 * RFC1122: OK. Passes ICMP errors back to application, as per
443 * 4.1.3.3.
444 */
445 if (!inet->recverr) {
446 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
447 goto out;
448 } else {
449 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
450 }
451 sk->sk_err = err;
452 sk->sk_error_report(sk);
453 out:
454 sock_put(sk);
455 }
456
457 void udp_err(struct sk_buff *skb, u32 info)
458 {
459 __udp4_lib_err(skb, info, &udp_table);
460 }
461
462 /*
463 * Throw away all pending data and cancel the corking. Socket is locked.
464 */
465 void udp_flush_pending_frames(struct sock *sk)
466 {
467 struct udp_sock *up = udp_sk(sk);
468
469 if (up->pending) {
470 up->len = 0;
471 up->pending = 0;
472 ip_flush_pending_frames(sk);
473 }
474 }
475 EXPORT_SYMBOL(udp_flush_pending_frames);
476
477 /**
478 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
479 * @sk: socket we are sending on
480 * @skb: sk_buff containing the filled-in UDP header
481 * (checksum field must be zeroed out)
482 */
483 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
484 __be32 src, __be32 dst, int len)
485 {
486 unsigned int offset;
487 struct udphdr *uh = udp_hdr(skb);
488 __wsum csum = 0;
489
490 if (skb_queue_len(&sk->sk_write_queue) == 1) {
491 /*
492 * Only one fragment on the socket.
493 */
494 skb->csum_start = skb_transport_header(skb) - skb->head;
495 skb->csum_offset = offsetof(struct udphdr, check);
496 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
497 } else {
498 /*
499 * HW-checksum won't work as there are two or more
500 * fragments on the socket so that all csums of sk_buffs
501 * should be together
502 */
503 offset = skb_transport_offset(skb);
504 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
505
506 skb->ip_summed = CHECKSUM_NONE;
507
508 skb_queue_walk(&sk->sk_write_queue, skb) {
509 csum = csum_add(csum, skb->csum);
510 }
511
512 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
513 if (uh->check == 0)
514 uh->check = CSUM_MANGLED_0;
515 }
516 }
517
518 /*
519 * Push out all pending data as one UDP datagram. Socket is locked.
520 */
521 static int udp_push_pending_frames(struct sock *sk)
522 {
523 struct udp_sock *up = udp_sk(sk);
524 struct inet_sock *inet = inet_sk(sk);
525 struct flowi *fl = &inet->cork.fl;
526 struct sk_buff *skb;
527 struct udphdr *uh;
528 int err = 0;
529 int is_udplite = IS_UDPLITE(sk);
530 __wsum csum = 0;
531
532 /* Grab the skbuff where UDP header space exists. */
533 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
534 goto out;
535
536 /*
537 * Create a UDP header
538 */
539 uh = udp_hdr(skb);
540 uh->source = fl->fl_ip_sport;
541 uh->dest = fl->fl_ip_dport;
542 uh->len = htons(up->len);
543 uh->check = 0;
544
545 if (is_udplite) /* UDP-Lite */
546 csum = udplite_csum_outgoing(sk, skb);
547
548 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
549
550 skb->ip_summed = CHECKSUM_NONE;
551 goto send;
552
553 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
554
555 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src, fl->fl4_dst, up->len);
556 goto send;
557
558 } else /* `normal' UDP */
559 csum = udp_csum_outgoing(sk, skb);
560
561 /* add protocol-dependent pseudo-header */
562 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
563 sk->sk_protocol, csum);
564 if (uh->check == 0)
565 uh->check = CSUM_MANGLED_0;
566
567 send:
568 err = ip_push_pending_frames(sk);
569 if (err) {
570 if (err == -ENOBUFS && !inet->recverr) {
571 UDP_INC_STATS_USER(sock_net(sk),
572 UDP_MIB_SNDBUFERRORS, is_udplite);
573 err = 0;
574 }
575 } else
576 UDP_INC_STATS_USER(sock_net(sk),
577 UDP_MIB_OUTDATAGRAMS, is_udplite);
578 out:
579 up->len = 0;
580 up->pending = 0;
581 return err;
582 }
583
584 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
585 size_t len)
586 {
587 struct inet_sock *inet = inet_sk(sk);
588 struct udp_sock *up = udp_sk(sk);
589 int ulen = len;
590 struct ipcm_cookie ipc;
591 struct rtable *rt = NULL;
592 int free = 0;
593 int connected = 0;
594 __be32 daddr, faddr, saddr;
595 __be16 dport;
596 u8 tos;
597 int err, is_udplite = IS_UDPLITE(sk);
598 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
599 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
600
601 if (len > 0xFFFF)
602 return -EMSGSIZE;
603
604 /*
605 * Check the flags.
606 */
607
608 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
609 return -EOPNOTSUPP;
610
611 ipc.opt = NULL;
612 ipc.shtx.flags = 0;
613
614 if (up->pending) {
615 /*
616 * There are pending frames.
617 * The socket lock must be held while it's corked.
618 */
619 lock_sock(sk);
620 if (likely(up->pending)) {
621 if (unlikely(up->pending != AF_INET)) {
622 release_sock(sk);
623 return -EINVAL;
624 }
625 goto do_append_data;
626 }
627 release_sock(sk);
628 }
629 ulen += sizeof(struct udphdr);
630
631 /*
632 * Get and verify the address.
633 */
634 if (msg->msg_name) {
635 struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name;
636 if (msg->msg_namelen < sizeof(*usin))
637 return -EINVAL;
638 if (usin->sin_family != AF_INET) {
639 if (usin->sin_family != AF_UNSPEC)
640 return -EAFNOSUPPORT;
641 }
642
643 daddr = usin->sin_addr.s_addr;
644 dport = usin->sin_port;
645 if (dport == 0)
646 return -EINVAL;
647 } else {
648 if (sk->sk_state != TCP_ESTABLISHED)
649 return -EDESTADDRREQ;
650 daddr = inet->inet_daddr;
651 dport = inet->inet_dport;
652 /* Open fast path for connected socket.
653 Route will not be used, if at least one option is set.
654 */
655 connected = 1;
656 }
657 ipc.addr = inet->inet_saddr;
658
659 ipc.oif = sk->sk_bound_dev_if;
660 err = sock_tx_timestamp(msg, sk, &ipc.shtx);
661 if (err)
662 return err;
663 if (msg->msg_controllen) {
664 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
665 if (err)
666 return err;
667 if (ipc.opt)
668 free = 1;
669 connected = 0;
670 }
671 if (!ipc.opt)
672 ipc.opt = inet->opt;
673
674 saddr = ipc.addr;
675 ipc.addr = faddr = daddr;
676
677 if (ipc.opt && ipc.opt->srr) {
678 if (!daddr)
679 return -EINVAL;
680 faddr = ipc.opt->faddr;
681 connected = 0;
682 }
683 tos = RT_TOS(inet->tos);
684 if (sock_flag(sk, SOCK_LOCALROUTE) ||
685 (msg->msg_flags & MSG_DONTROUTE) ||
686 (ipc.opt && ipc.opt->is_strictroute)) {
687 tos |= RTO_ONLINK;
688 connected = 0;
689 }
690
691 if (ipv4_is_multicast(daddr)) {
692 if (!ipc.oif)
693 ipc.oif = inet->mc_index;
694 if (!saddr)
695 saddr = inet->mc_addr;
696 connected = 0;
697 }
698
699 if (connected)
700 rt = (struct rtable *)sk_dst_check(sk, 0);
701
702 if (rt == NULL) {
703 struct flowi fl = { .oif = ipc.oif,
704 .mark = sk->sk_mark,
705 .nl_u = { .ip4_u =
706 { .daddr = faddr,
707 .saddr = saddr,
708 .tos = tos } },
709 .proto = sk->sk_protocol,
710 .flags = inet_sk_flowi_flags(sk),
711 .uli_u = { .ports =
712 { .sport = inet->inet_sport,
713 .dport = dport } } };
714 struct net *net = sock_net(sk);
715
716 security_sk_classify_flow(sk, &fl);
717 err = ip_route_output_flow(net, &rt, &fl, sk, 1);
718 if (err) {
719 if (err == -ENETUNREACH)
720 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
721 goto out;
722 }
723
724 err = -EACCES;
725 if ((rt->rt_flags & RTCF_BROADCAST) &&
726 !sock_flag(sk, SOCK_BROADCAST))
727 goto out;
728 if (connected)
729 sk_dst_set(sk, dst_clone(&rt->u.dst));
730 }
731
732 if (msg->msg_flags&MSG_CONFIRM)
733 goto do_confirm;
734 back_from_confirm:
735
736 saddr = rt->rt_src;
737 if (!ipc.addr)
738 daddr = ipc.addr = rt->rt_dst;
739
740 lock_sock(sk);
741 if (unlikely(up->pending)) {
742 /* The socket is already corked while preparing it. */
743 /* ... which is an evident application bug. --ANK */
744 release_sock(sk);
745
746 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
747 err = -EINVAL;
748 goto out;
749 }
750 /*
751 * Now cork the socket to pend data.
752 */
753 inet->cork.fl.fl4_dst = daddr;
754 inet->cork.fl.fl_ip_dport = dport;
755 inet->cork.fl.fl4_src = saddr;
756 inet->cork.fl.fl_ip_sport = inet->inet_sport;
757 up->pending = AF_INET;
758
759 do_append_data:
760 up->len += ulen;
761 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
762 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
763 sizeof(struct udphdr), &ipc, &rt,
764 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
765 if (err)
766 udp_flush_pending_frames(sk);
767 else if (!corkreq)
768 err = udp_push_pending_frames(sk);
769 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
770 up->pending = 0;
771 release_sock(sk);
772
773 out:
774 ip_rt_put(rt);
775 if (free)
776 kfree(ipc.opt);
777 if (!err)
778 return len;
779 /*
780 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
781 * ENOBUFS might not be good (it's not tunable per se), but otherwise
782 * we don't have a good statistic (IpOutDiscards but it can be too many
783 * things). We could add another new stat but at least for now that
784 * seems like overkill.
785 */
786 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
787 UDP_INC_STATS_USER(sock_net(sk),
788 UDP_MIB_SNDBUFERRORS, is_udplite);
789 }
790 return err;
791
792 do_confirm:
793 dst_confirm(&rt->u.dst);
794 if (!(msg->msg_flags&MSG_PROBE) || len)
795 goto back_from_confirm;
796 err = 0;
797 goto out;
798 }
799 EXPORT_SYMBOL(udp_sendmsg);
800
801 int udp_sendpage(struct sock *sk, struct page *page, int offset,
802 size_t size, int flags)
803 {
804 struct udp_sock *up = udp_sk(sk);
805 int ret;
806
807 if (!up->pending) {
808 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
809
810 /* Call udp_sendmsg to specify destination address which
811 * sendpage interface can't pass.
812 * This will succeed only when the socket is connected.
813 */
814 ret = udp_sendmsg(NULL, sk, &msg, 0);
815 if (ret < 0)
816 return ret;
817 }
818
819 lock_sock(sk);
820
821 if (unlikely(!up->pending)) {
822 release_sock(sk);
823
824 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
825 return -EINVAL;
826 }
827
828 ret = ip_append_page(sk, page, offset, size, flags);
829 if (ret == -EOPNOTSUPP) {
830 release_sock(sk);
831 return sock_no_sendpage(sk->sk_socket, page, offset,
832 size, flags);
833 }
834 if (ret < 0) {
835 udp_flush_pending_frames(sk);
836 goto out;
837 }
838
839 up->len += size;
840 if (!(up->corkflag || (flags&MSG_MORE)))
841 ret = udp_push_pending_frames(sk);
842 if (!ret)
843 ret = size;
844 out:
845 release_sock(sk);
846 return ret;
847 }
848
849
850 /**
851 * first_packet_length - return length of first packet in receive queue
852 * @sk: socket
853 *
854 * Drops all bad checksum frames, until a valid one is found.
855 * Returns the length of found skb, or 0 if none is found.
856 */
857 static unsigned int first_packet_length(struct sock *sk)
858 {
859 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
860 struct sk_buff *skb;
861 unsigned int res;
862
863 __skb_queue_head_init(&list_kill);
864
865 spin_lock_bh(&rcvq->lock);
866 while ((skb = skb_peek(rcvq)) != NULL &&
867 udp_lib_checksum_complete(skb)) {
868 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
869 IS_UDPLITE(sk));
870 atomic_inc(&sk->sk_drops);
871 __skb_unlink(skb, rcvq);
872 __skb_queue_tail(&list_kill, skb);
873 }
874 res = skb ? skb->len : 0;
875 spin_unlock_bh(&rcvq->lock);
876
877 if (!skb_queue_empty(&list_kill)) {
878 lock_sock(sk);
879 __skb_queue_purge(&list_kill);
880 sk_mem_reclaim_partial(sk);
881 release_sock(sk);
882 }
883 return res;
884 }
885
886 /*
887 * IOCTL requests applicable to the UDP protocol
888 */
889
890 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
891 {
892 switch (cmd) {
893 case SIOCOUTQ:
894 {
895 int amount = sk_wmem_alloc_get(sk);
896
897 return put_user(amount, (int __user *)arg);
898 }
899
900 case SIOCINQ:
901 {
902 unsigned int amount = first_packet_length(sk);
903
904 if (amount)
905 /*
906 * We will only return the amount
907 * of this packet since that is all
908 * that will be read.
909 */
910 amount -= sizeof(struct udphdr);
911
912 return put_user(amount, (int __user *)arg);
913 }
914
915 default:
916 return -ENOIOCTLCMD;
917 }
918
919 return 0;
920 }
921 EXPORT_SYMBOL(udp_ioctl);
922
923 /*
924 * This should be easy, if there is something there we
925 * return it, otherwise we block.
926 */
927
928 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
929 size_t len, int noblock, int flags, int *addr_len)
930 {
931 struct inet_sock *inet = inet_sk(sk);
932 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
933 struct sk_buff *skb;
934 unsigned int ulen, copied;
935 int peeked;
936 int err;
937 int is_udplite = IS_UDPLITE(sk);
938
939 /*
940 * Check any passed addresses
941 */
942 if (addr_len)
943 *addr_len = sizeof(*sin);
944
945 if (flags & MSG_ERRQUEUE)
946 return ip_recv_error(sk, msg, len);
947
948 try_again:
949 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
950 &peeked, &err);
951 if (!skb)
952 goto out;
953
954 ulen = skb->len - sizeof(struct udphdr);
955 copied = len;
956 if (copied > ulen)
957 copied = ulen;
958 else if (copied < ulen)
959 msg->msg_flags |= MSG_TRUNC;
960
961 /*
962 * If checksum is needed at all, try to do it while copying the
963 * data. If the data is truncated, or if we only want a partial
964 * coverage checksum (UDP-Lite), do it before the copy.
965 */
966
967 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
968 if (udp_lib_checksum_complete(skb))
969 goto csum_copy_err;
970 }
971
972 if (skb_csum_unnecessary(skb))
973 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
974 msg->msg_iov, copied);
975 else {
976 err = skb_copy_and_csum_datagram_iovec(skb,
977 sizeof(struct udphdr),
978 msg->msg_iov);
979
980 if (err == -EINVAL)
981 goto csum_copy_err;
982 }
983
984 if (err)
985 goto out_free;
986
987 if (!peeked)
988 UDP_INC_STATS_USER(sock_net(sk),
989 UDP_MIB_INDATAGRAMS, is_udplite);
990
991 sock_recv_ts_and_drops(msg, sk, skb);
992
993 /* Copy the address. */
994 if (sin) {
995 sin->sin_family = AF_INET;
996 sin->sin_port = udp_hdr(skb)->source;
997 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
998 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
999 }
1000 if (inet->cmsg_flags)
1001 ip_cmsg_recv(msg, skb);
1002
1003 err = copied;
1004 if (flags & MSG_TRUNC)
1005 err = ulen;
1006
1007 out_free:
1008 lock_sock(sk);
1009 skb_free_datagram(sk, skb);
1010 release_sock(sk);
1011 out:
1012 return err;
1013
1014 csum_copy_err:
1015 lock_sock(sk);
1016 if (!skb_kill_datagram(sk, skb, flags))
1017 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1018 release_sock(sk);
1019
1020 if (noblock)
1021 return -EAGAIN;
1022 goto try_again;
1023 }
1024
1025
1026 int udp_disconnect(struct sock *sk, int flags)
1027 {
1028 struct inet_sock *inet = inet_sk(sk);
1029 /*
1030 * 1003.1g - break association.
1031 */
1032
1033 sk->sk_state = TCP_CLOSE;
1034 inet->inet_daddr = 0;
1035 inet->inet_dport = 0;
1036 sk->sk_bound_dev_if = 0;
1037 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1038 inet_reset_saddr(sk);
1039
1040 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1041 sk->sk_prot->unhash(sk);
1042 inet->inet_sport = 0;
1043 }
1044 sk_dst_reset(sk);
1045 return 0;
1046 }
1047 EXPORT_SYMBOL(udp_disconnect);
1048
1049 void udp_lib_unhash(struct sock *sk)
1050 {
1051 if (sk_hashed(sk)) {
1052 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1053 struct udp_hslot *hslot = udp_hashslot(udptable, sock_net(sk),
1054 sk->sk_hash);
1055
1056 spin_lock_bh(&hslot->lock);
1057 if (sk_nulls_del_node_init_rcu(sk)) {
1058 inet_sk(sk)->inet_num = 0;
1059 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1060 }
1061 spin_unlock_bh(&hslot->lock);
1062 }
1063 }
1064 EXPORT_SYMBOL(udp_lib_unhash);
1065
1066 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1067 {
1068 int rc = sock_queue_rcv_skb(sk, skb);
1069
1070 if (rc < 0) {
1071 int is_udplite = IS_UDPLITE(sk);
1072
1073 /* Note that an ENOMEM error is charged twice */
1074 if (rc == -ENOMEM)
1075 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1076 is_udplite);
1077 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1078 kfree_skb(skb);
1079 return -1;
1080 }
1081
1082 return 0;
1083
1084 }
1085
1086 /* returns:
1087 * -1: error
1088 * 0: success
1089 * >0: "udp encap" protocol resubmission
1090 *
1091 * Note that in the success and error cases, the skb is assumed to
1092 * have either been requeued or freed.
1093 */
1094 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1095 {
1096 struct udp_sock *up = udp_sk(sk);
1097 int rc;
1098 int is_udplite = IS_UDPLITE(sk);
1099
1100 /*
1101 * Charge it to the socket, dropping if the queue is full.
1102 */
1103 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1104 goto drop;
1105 nf_reset(skb);
1106
1107 if (up->encap_type) {
1108 /*
1109 * This is an encapsulation socket so pass the skb to
1110 * the socket's udp_encap_rcv() hook. Otherwise, just
1111 * fall through and pass this up the UDP socket.
1112 * up->encap_rcv() returns the following value:
1113 * =0 if skb was successfully passed to the encap
1114 * handler or was discarded by it.
1115 * >0 if skb should be passed on to UDP.
1116 * <0 if skb should be resubmitted as proto -N
1117 */
1118
1119 /* if we're overly short, let UDP handle it */
1120 if (skb->len > sizeof(struct udphdr) &&
1121 up->encap_rcv != NULL) {
1122 int ret;
1123
1124 ret = (*up->encap_rcv)(sk, skb);
1125 if (ret <= 0) {
1126 UDP_INC_STATS_BH(sock_net(sk),
1127 UDP_MIB_INDATAGRAMS,
1128 is_udplite);
1129 return -ret;
1130 }
1131 }
1132
1133 /* FALLTHROUGH -- it's a UDP Packet */
1134 }
1135
1136 /*
1137 * UDP-Lite specific tests, ignored on UDP sockets
1138 */
1139 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1140
1141 /*
1142 * MIB statistics other than incrementing the error count are
1143 * disabled for the following two types of errors: these depend
1144 * on the application settings, not on the functioning of the
1145 * protocol stack as such.
1146 *
1147 * RFC 3828 here recommends (sec 3.3): "There should also be a
1148 * way ... to ... at least let the receiving application block
1149 * delivery of packets with coverage values less than a value
1150 * provided by the application."
1151 */
1152 if (up->pcrlen == 0) { /* full coverage was set */
1153 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1154 "%d while full coverage %d requested\n",
1155 UDP_SKB_CB(skb)->cscov, skb->len);
1156 goto drop;
1157 }
1158 /* The next case involves violating the min. coverage requested
1159 * by the receiver. This is subtle: if receiver wants x and x is
1160 * greater than the buffersize/MTU then receiver will complain
1161 * that it wants x while sender emits packets of smaller size y.
1162 * Therefore the above ...()->partial_cov statement is essential.
1163 */
1164 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1165 LIMIT_NETDEBUG(KERN_WARNING
1166 "UDPLITE: coverage %d too small, need min %d\n",
1167 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1168 goto drop;
1169 }
1170 }
1171
1172 if (sk->sk_filter) {
1173 if (udp_lib_checksum_complete(skb))
1174 goto drop;
1175 }
1176
1177 rc = 0;
1178
1179 bh_lock_sock(sk);
1180 if (!sock_owned_by_user(sk))
1181 rc = __udp_queue_rcv_skb(sk, skb);
1182 else
1183 sk_add_backlog(sk, skb);
1184 bh_unlock_sock(sk);
1185
1186 return rc;
1187
1188 drop:
1189 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1190 atomic_inc(&sk->sk_drops);
1191 kfree_skb(skb);
1192 return -1;
1193 }
1194
1195 /*
1196 * Multicasts and broadcasts go to each listener.
1197 *
1198 * Note: called only from the BH handler context,
1199 * so we don't need to lock the hashes.
1200 */
1201 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1202 struct udphdr *uh,
1203 __be32 saddr, __be32 daddr,
1204 struct udp_table *udptable)
1205 {
1206 struct sock *sk;
1207 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1208 int dif;
1209
1210 spin_lock(&hslot->lock);
1211 sk = sk_nulls_head(&hslot->head);
1212 dif = skb->dev->ifindex;
1213 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1214 if (sk) {
1215 struct sock *sknext = NULL;
1216
1217 do {
1218 struct sk_buff *skb1 = skb;
1219
1220 sknext = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1221 daddr, uh->source, saddr,
1222 dif);
1223 if (sknext)
1224 skb1 = skb_clone(skb, GFP_ATOMIC);
1225
1226 if (skb1) {
1227 int ret = udp_queue_rcv_skb(sk, skb1);
1228 if (ret > 0)
1229 /* we should probably re-process instead
1230 * of dropping packets here. */
1231 kfree_skb(skb1);
1232 }
1233 sk = sknext;
1234 } while (sknext);
1235 } else
1236 consume_skb(skb);
1237 spin_unlock(&hslot->lock);
1238 return 0;
1239 }
1240
1241 /* Initialize UDP checksum. If exited with zero value (success),
1242 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1243 * Otherwise, csum completion requires chacksumming packet body,
1244 * including udp header and folding it to skb->csum.
1245 */
1246 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1247 int proto)
1248 {
1249 const struct iphdr *iph;
1250 int err;
1251
1252 UDP_SKB_CB(skb)->partial_cov = 0;
1253 UDP_SKB_CB(skb)->cscov = skb->len;
1254
1255 if (proto == IPPROTO_UDPLITE) {
1256 err = udplite_checksum_init(skb, uh);
1257 if (err)
1258 return err;
1259 }
1260
1261 iph = ip_hdr(skb);
1262 if (uh->check == 0) {
1263 skb->ip_summed = CHECKSUM_UNNECESSARY;
1264 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1265 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1266 proto, skb->csum))
1267 skb->ip_summed = CHECKSUM_UNNECESSARY;
1268 }
1269 if (!skb_csum_unnecessary(skb))
1270 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1271 skb->len, proto, 0);
1272 /* Probably, we should checksum udp header (it should be in cache
1273 * in any case) and data in tiny packets (< rx copybreak).
1274 */
1275
1276 return 0;
1277 }
1278
1279 /*
1280 * All we need to do is get the socket, and then do a checksum.
1281 */
1282
1283 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1284 int proto)
1285 {
1286 struct sock *sk;
1287 struct udphdr *uh;
1288 unsigned short ulen;
1289 struct rtable *rt = skb_rtable(skb);
1290 __be32 saddr, daddr;
1291 struct net *net = dev_net(skb->dev);
1292
1293 /*
1294 * Validate the packet.
1295 */
1296 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1297 goto drop; /* No space for header. */
1298
1299 uh = udp_hdr(skb);
1300 ulen = ntohs(uh->len);
1301 if (ulen > skb->len)
1302 goto short_packet;
1303
1304 if (proto == IPPROTO_UDP) {
1305 /* UDP validates ulen. */
1306 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1307 goto short_packet;
1308 uh = udp_hdr(skb);
1309 }
1310
1311 if (udp4_csum_init(skb, uh, proto))
1312 goto csum_error;
1313
1314 saddr = ip_hdr(skb)->saddr;
1315 daddr = ip_hdr(skb)->daddr;
1316
1317 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1318 return __udp4_lib_mcast_deliver(net, skb, uh,
1319 saddr, daddr, udptable);
1320
1321 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1322
1323 if (sk != NULL) {
1324 int ret = udp_queue_rcv_skb(sk, skb);
1325 sock_put(sk);
1326
1327 /* a return value > 0 means to resubmit the input, but
1328 * it wants the return to be -protocol, or 0
1329 */
1330 if (ret > 0)
1331 return -ret;
1332 return 0;
1333 }
1334
1335 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1336 goto drop;
1337 nf_reset(skb);
1338
1339 /* No socket. Drop packet silently, if checksum is wrong */
1340 if (udp_lib_checksum_complete(skb))
1341 goto csum_error;
1342
1343 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1344 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1345
1346 /*
1347 * Hmm. We got an UDP packet to a port to which we
1348 * don't wanna listen. Ignore it.
1349 */
1350 kfree_skb(skb);
1351 return 0;
1352
1353 short_packet:
1354 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1355 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1356 &saddr,
1357 ntohs(uh->source),
1358 ulen,
1359 skb->len,
1360 &daddr,
1361 ntohs(uh->dest));
1362 goto drop;
1363
1364 csum_error:
1365 /*
1366 * RFC1122: OK. Discards the bad packet silently (as far as
1367 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1368 */
1369 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1370 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1371 &saddr,
1372 ntohs(uh->source),
1373 &daddr,
1374 ntohs(uh->dest),
1375 ulen);
1376 drop:
1377 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1378 kfree_skb(skb);
1379 return 0;
1380 }
1381
1382 int udp_rcv(struct sk_buff *skb)
1383 {
1384 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1385 }
1386
1387 void udp_destroy_sock(struct sock *sk)
1388 {
1389 lock_sock(sk);
1390 udp_flush_pending_frames(sk);
1391 release_sock(sk);
1392 }
1393
1394 /*
1395 * Socket option code for UDP
1396 */
1397 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1398 char __user *optval, unsigned int optlen,
1399 int (*push_pending_frames)(struct sock *))
1400 {
1401 struct udp_sock *up = udp_sk(sk);
1402 int val;
1403 int err = 0;
1404 int is_udplite = IS_UDPLITE(sk);
1405
1406 if (optlen < sizeof(int))
1407 return -EINVAL;
1408
1409 if (get_user(val, (int __user *)optval))
1410 return -EFAULT;
1411
1412 switch (optname) {
1413 case UDP_CORK:
1414 if (val != 0) {
1415 up->corkflag = 1;
1416 } else {
1417 up->corkflag = 0;
1418 lock_sock(sk);
1419 (*push_pending_frames)(sk);
1420 release_sock(sk);
1421 }
1422 break;
1423
1424 case UDP_ENCAP:
1425 switch (val) {
1426 case 0:
1427 case UDP_ENCAP_ESPINUDP:
1428 case UDP_ENCAP_ESPINUDP_NON_IKE:
1429 up->encap_rcv = xfrm4_udp_encap_rcv;
1430 /* FALLTHROUGH */
1431 case UDP_ENCAP_L2TPINUDP:
1432 up->encap_type = val;
1433 break;
1434 default:
1435 err = -ENOPROTOOPT;
1436 break;
1437 }
1438 break;
1439
1440 /*
1441 * UDP-Lite's partial checksum coverage (RFC 3828).
1442 */
1443 /* The sender sets actual checksum coverage length via this option.
1444 * The case coverage > packet length is handled by send module. */
1445 case UDPLITE_SEND_CSCOV:
1446 if (!is_udplite) /* Disable the option on UDP sockets */
1447 return -ENOPROTOOPT;
1448 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1449 val = 8;
1450 else if (val > USHORT_MAX)
1451 val = USHORT_MAX;
1452 up->pcslen = val;
1453 up->pcflag |= UDPLITE_SEND_CC;
1454 break;
1455
1456 /* The receiver specifies a minimum checksum coverage value. To make
1457 * sense, this should be set to at least 8 (as done below). If zero is
1458 * used, this again means full checksum coverage. */
1459 case UDPLITE_RECV_CSCOV:
1460 if (!is_udplite) /* Disable the option on UDP sockets */
1461 return -ENOPROTOOPT;
1462 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1463 val = 8;
1464 else if (val > USHORT_MAX)
1465 val = USHORT_MAX;
1466 up->pcrlen = val;
1467 up->pcflag |= UDPLITE_RECV_CC;
1468 break;
1469
1470 default:
1471 err = -ENOPROTOOPT;
1472 break;
1473 }
1474
1475 return err;
1476 }
1477 EXPORT_SYMBOL(udp_lib_setsockopt);
1478
1479 int udp_setsockopt(struct sock *sk, int level, int optname,
1480 char __user *optval, unsigned int optlen)
1481 {
1482 if (level == SOL_UDP || level == SOL_UDPLITE)
1483 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1484 udp_push_pending_frames);
1485 return ip_setsockopt(sk, level, optname, optval, optlen);
1486 }
1487
1488 #ifdef CONFIG_COMPAT
1489 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1490 char __user *optval, unsigned int optlen)
1491 {
1492 if (level == SOL_UDP || level == SOL_UDPLITE)
1493 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1494 udp_push_pending_frames);
1495 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1496 }
1497 #endif
1498
1499 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1500 char __user *optval, int __user *optlen)
1501 {
1502 struct udp_sock *up = udp_sk(sk);
1503 int val, len;
1504
1505 if (get_user(len, optlen))
1506 return -EFAULT;
1507
1508 len = min_t(unsigned int, len, sizeof(int));
1509
1510 if (len < 0)
1511 return -EINVAL;
1512
1513 switch (optname) {
1514 case UDP_CORK:
1515 val = up->corkflag;
1516 break;
1517
1518 case UDP_ENCAP:
1519 val = up->encap_type;
1520 break;
1521
1522 /* The following two cannot be changed on UDP sockets, the return is
1523 * always 0 (which corresponds to the full checksum coverage of UDP). */
1524 case UDPLITE_SEND_CSCOV:
1525 val = up->pcslen;
1526 break;
1527
1528 case UDPLITE_RECV_CSCOV:
1529 val = up->pcrlen;
1530 break;
1531
1532 default:
1533 return -ENOPROTOOPT;
1534 }
1535
1536 if (put_user(len, optlen))
1537 return -EFAULT;
1538 if (copy_to_user(optval, &val, len))
1539 return -EFAULT;
1540 return 0;
1541 }
1542 EXPORT_SYMBOL(udp_lib_getsockopt);
1543
1544 int udp_getsockopt(struct sock *sk, int level, int optname,
1545 char __user *optval, int __user *optlen)
1546 {
1547 if (level == SOL_UDP || level == SOL_UDPLITE)
1548 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1549 return ip_getsockopt(sk, level, optname, optval, optlen);
1550 }
1551
1552 #ifdef CONFIG_COMPAT
1553 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1554 char __user *optval, int __user *optlen)
1555 {
1556 if (level == SOL_UDP || level == SOL_UDPLITE)
1557 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1558 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1559 }
1560 #endif
1561 /**
1562 * udp_poll - wait for a UDP event.
1563 * @file - file struct
1564 * @sock - socket
1565 * @wait - poll table
1566 *
1567 * This is same as datagram poll, except for the special case of
1568 * blocking sockets. If application is using a blocking fd
1569 * and a packet with checksum error is in the queue;
1570 * then it could get return from select indicating data available
1571 * but then block when reading it. Add special case code
1572 * to work around these arguably broken applications.
1573 */
1574 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1575 {
1576 unsigned int mask = datagram_poll(file, sock, wait);
1577 struct sock *sk = sock->sk;
1578
1579 /* Check for false positives due to checksum errors */
1580 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1581 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1582 mask &= ~(POLLIN | POLLRDNORM);
1583
1584 return mask;
1585
1586 }
1587 EXPORT_SYMBOL(udp_poll);
1588
1589 struct proto udp_prot = {
1590 .name = "UDP",
1591 .owner = THIS_MODULE,
1592 .close = udp_lib_close,
1593 .connect = ip4_datagram_connect,
1594 .disconnect = udp_disconnect,
1595 .ioctl = udp_ioctl,
1596 .destroy = udp_destroy_sock,
1597 .setsockopt = udp_setsockopt,
1598 .getsockopt = udp_getsockopt,
1599 .sendmsg = udp_sendmsg,
1600 .recvmsg = udp_recvmsg,
1601 .sendpage = udp_sendpage,
1602 .backlog_rcv = __udp_queue_rcv_skb,
1603 .hash = udp_lib_hash,
1604 .unhash = udp_lib_unhash,
1605 .get_port = udp_v4_get_port,
1606 .memory_allocated = &udp_memory_allocated,
1607 .sysctl_mem = sysctl_udp_mem,
1608 .sysctl_wmem = &sysctl_udp_wmem_min,
1609 .sysctl_rmem = &sysctl_udp_rmem_min,
1610 .obj_size = sizeof(struct udp_sock),
1611 .slab_flags = SLAB_DESTROY_BY_RCU,
1612 .h.udp_table = &udp_table,
1613 #ifdef CONFIG_COMPAT
1614 .compat_setsockopt = compat_udp_setsockopt,
1615 .compat_getsockopt = compat_udp_getsockopt,
1616 #endif
1617 };
1618 EXPORT_SYMBOL(udp_prot);
1619
1620 /* ------------------------------------------------------------------------ */
1621 #ifdef CONFIG_PROC_FS
1622
1623 static struct sock *udp_get_first(struct seq_file *seq, int start)
1624 {
1625 struct sock *sk;
1626 struct udp_iter_state *state = seq->private;
1627 struct net *net = seq_file_net(seq);
1628
1629 for (state->bucket = start; state->bucket <= state->udp_table->mask;
1630 ++state->bucket) {
1631 struct hlist_nulls_node *node;
1632 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
1633
1634 if (hlist_nulls_empty(&hslot->head))
1635 continue;
1636
1637 spin_lock_bh(&hslot->lock);
1638 sk_nulls_for_each(sk, node, &hslot->head) {
1639 if (!net_eq(sock_net(sk), net))
1640 continue;
1641 if (sk->sk_family == state->family)
1642 goto found;
1643 }
1644 spin_unlock_bh(&hslot->lock);
1645 }
1646 sk = NULL;
1647 found:
1648 return sk;
1649 }
1650
1651 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1652 {
1653 struct udp_iter_state *state = seq->private;
1654 struct net *net = seq_file_net(seq);
1655
1656 do {
1657 sk = sk_nulls_next(sk);
1658 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1659
1660 if (!sk) {
1661 if (state->bucket <= state->udp_table->mask)
1662 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
1663 return udp_get_first(seq, state->bucket + 1);
1664 }
1665 return sk;
1666 }
1667
1668 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1669 {
1670 struct sock *sk = udp_get_first(seq, 0);
1671
1672 if (sk)
1673 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1674 --pos;
1675 return pos ? NULL : sk;
1676 }
1677
1678 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1679 {
1680 struct udp_iter_state *state = seq->private;
1681 state->bucket = MAX_UDP_PORTS;
1682
1683 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
1684 }
1685
1686 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1687 {
1688 struct sock *sk;
1689
1690 if (v == SEQ_START_TOKEN)
1691 sk = udp_get_idx(seq, 0);
1692 else
1693 sk = udp_get_next(seq, v);
1694
1695 ++*pos;
1696 return sk;
1697 }
1698
1699 static void udp_seq_stop(struct seq_file *seq, void *v)
1700 {
1701 struct udp_iter_state *state = seq->private;
1702
1703 if (state->bucket <= state->udp_table->mask)
1704 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
1705 }
1706
1707 static int udp_seq_open(struct inode *inode, struct file *file)
1708 {
1709 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1710 struct udp_iter_state *s;
1711 int err;
1712
1713 err = seq_open_net(inode, file, &afinfo->seq_ops,
1714 sizeof(struct udp_iter_state));
1715 if (err < 0)
1716 return err;
1717
1718 s = ((struct seq_file *)file->private_data)->private;
1719 s->family = afinfo->family;
1720 s->udp_table = afinfo->udp_table;
1721 return err;
1722 }
1723
1724 /* ------------------------------------------------------------------------ */
1725 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
1726 {
1727 struct proc_dir_entry *p;
1728 int rc = 0;
1729
1730 afinfo->seq_fops.open = udp_seq_open;
1731 afinfo->seq_fops.read = seq_read;
1732 afinfo->seq_fops.llseek = seq_lseek;
1733 afinfo->seq_fops.release = seq_release_net;
1734
1735 afinfo->seq_ops.start = udp_seq_start;
1736 afinfo->seq_ops.next = udp_seq_next;
1737 afinfo->seq_ops.stop = udp_seq_stop;
1738
1739 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
1740 &afinfo->seq_fops, afinfo);
1741 if (!p)
1742 rc = -ENOMEM;
1743 return rc;
1744 }
1745 EXPORT_SYMBOL(udp_proc_register);
1746
1747 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
1748 {
1749 proc_net_remove(net, afinfo->name);
1750 }
1751 EXPORT_SYMBOL(udp_proc_unregister);
1752
1753 /* ------------------------------------------------------------------------ */
1754 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
1755 int bucket, int *len)
1756 {
1757 struct inet_sock *inet = inet_sk(sp);
1758 __be32 dest = inet->inet_daddr;
1759 __be32 src = inet->inet_rcv_saddr;
1760 __u16 destp = ntohs(inet->inet_dport);
1761 __u16 srcp = ntohs(inet->inet_sport);
1762
1763 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
1764 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
1765 bucket, src, srcp, dest, destp, sp->sk_state,
1766 sk_wmem_alloc_get(sp),
1767 sk_rmem_alloc_get(sp),
1768 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1769 atomic_read(&sp->sk_refcnt), sp,
1770 atomic_read(&sp->sk_drops), len);
1771 }
1772
1773 int udp4_seq_show(struct seq_file *seq, void *v)
1774 {
1775 if (v == SEQ_START_TOKEN)
1776 seq_printf(seq, "%-127s\n",
1777 " sl local_address rem_address st tx_queue "
1778 "rx_queue tr tm->when retrnsmt uid timeout "
1779 "inode ref pointer drops");
1780 else {
1781 struct udp_iter_state *state = seq->private;
1782 int len;
1783
1784 udp4_format_sock(v, seq, state->bucket, &len);
1785 seq_printf(seq, "%*s\n", 127 - len, "");
1786 }
1787 return 0;
1788 }
1789
1790 /* ------------------------------------------------------------------------ */
1791 static struct udp_seq_afinfo udp4_seq_afinfo = {
1792 .name = "udp",
1793 .family = AF_INET,
1794 .udp_table = &udp_table,
1795 .seq_fops = {
1796 .owner = THIS_MODULE,
1797 },
1798 .seq_ops = {
1799 .show = udp4_seq_show,
1800 },
1801 };
1802
1803 static int udp4_proc_init_net(struct net *net)
1804 {
1805 return udp_proc_register(net, &udp4_seq_afinfo);
1806 }
1807
1808 static void udp4_proc_exit_net(struct net *net)
1809 {
1810 udp_proc_unregister(net, &udp4_seq_afinfo);
1811 }
1812
1813 static struct pernet_operations udp4_net_ops = {
1814 .init = udp4_proc_init_net,
1815 .exit = udp4_proc_exit_net,
1816 };
1817
1818 int __init udp4_proc_init(void)
1819 {
1820 return register_pernet_subsys(&udp4_net_ops);
1821 }
1822
1823 void udp4_proc_exit(void)
1824 {
1825 unregister_pernet_subsys(&udp4_net_ops);
1826 }
1827 #endif /* CONFIG_PROC_FS */
1828
1829 static __initdata unsigned long uhash_entries;
1830 static int __init set_uhash_entries(char *str)
1831 {
1832 if (!str)
1833 return 0;
1834 uhash_entries = simple_strtoul(str, &str, 0);
1835 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
1836 uhash_entries = UDP_HTABLE_SIZE_MIN;
1837 return 1;
1838 }
1839 __setup("uhash_entries=", set_uhash_entries);
1840
1841 void __init udp_table_init(struct udp_table *table, const char *name)
1842 {
1843 unsigned int i;
1844
1845 if (!CONFIG_BASE_SMALL)
1846 table->hash = alloc_large_system_hash(name,
1847 sizeof(struct udp_hslot),
1848 uhash_entries,
1849 21, /* one slot per 2 MB */
1850 0,
1851 &table->log,
1852 &table->mask,
1853 64 * 1024);
1854 /*
1855 * Make sure hash table has the minimum size
1856 */
1857 if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) {
1858 table->hash = kmalloc(UDP_HTABLE_SIZE_MIN *
1859 sizeof(struct udp_hslot), GFP_KERNEL);
1860 if (!table->hash)
1861 panic(name);
1862 table->log = ilog2(UDP_HTABLE_SIZE_MIN);
1863 table->mask = UDP_HTABLE_SIZE_MIN - 1;
1864 }
1865 for (i = 0; i <= table->mask; i++) {
1866 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
1867 spin_lock_init(&table->hash[i].lock);
1868 }
1869 }
1870
1871 void __init udp_init(void)
1872 {
1873 unsigned long nr_pages, limit;
1874
1875 udp_table_init(&udp_table, "UDP");
1876 /* Set the pressure threshold up by the same strategy of TCP. It is a
1877 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
1878 * toward zero with the amount of memory, with a floor of 128 pages.
1879 */
1880 nr_pages = totalram_pages - totalhigh_pages;
1881 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
1882 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
1883 limit = max(limit, 128UL);
1884 sysctl_udp_mem[0] = limit / 4 * 3;
1885 sysctl_udp_mem[1] = limit;
1886 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
1887
1888 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
1889 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
1890 }
1891
1892 int udp4_ufo_send_check(struct sk_buff *skb)
1893 {
1894 const struct iphdr *iph;
1895 struct udphdr *uh;
1896
1897 if (!pskb_may_pull(skb, sizeof(*uh)))
1898 return -EINVAL;
1899
1900 iph = ip_hdr(skb);
1901 uh = udp_hdr(skb);
1902
1903 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1904 IPPROTO_UDP, 0);
1905 skb->csum_start = skb_transport_header(skb) - skb->head;
1906 skb->csum_offset = offsetof(struct udphdr, check);
1907 skb->ip_summed = CHECKSUM_PARTIAL;
1908 return 0;
1909 }
1910
1911 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, int features)
1912 {
1913 struct sk_buff *segs = ERR_PTR(-EINVAL);
1914 unsigned int mss;
1915 int offset;
1916 __wsum csum;
1917
1918 mss = skb_shinfo(skb)->gso_size;
1919 if (unlikely(skb->len <= mss))
1920 goto out;
1921
1922 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
1923 /* Packet is from an untrusted source, reset gso_segs. */
1924 int type = skb_shinfo(skb)->gso_type;
1925
1926 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
1927 !(type & (SKB_GSO_UDP))))
1928 goto out;
1929
1930 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
1931
1932 segs = NULL;
1933 goto out;
1934 }
1935
1936 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
1937 * do checksum of UDP packets sent as multiple IP fragments.
1938 */
1939 offset = skb->csum_start - skb_headroom(skb);
1940 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1941 offset += skb->csum_offset;
1942 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1943 skb->ip_summed = CHECKSUM_NONE;
1944
1945 /* Fragment the skb. IP headers of the fragments are updated in
1946 * inet_gso_segment()
1947 */
1948 segs = skb_segment(skb, features);
1949 out:
1950 return segs;
1951 }
1952
Linux kernel - Deliverables
This page took 0.09997 seconds and 6 git commands to generate.