projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
udp: split sk_hash into two u16 hashes
[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 || udp_sk(sk2)->udp_port_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(udp_sk(sk2)->udp_port_hash >> log,
148 bitmap);
149 else
150 return 1;
151 }
152 return 0;
153 }
154
155 /**
156 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
157 *
158 * @sk: socket struct in question
159 * @snum: port number to look up
160 * @saddr_comp: AF-dependent comparison of bound local IP addresses
161 */
162 int udp_lib_get_port(struct sock *sk, unsigned short snum,
163 int (*saddr_comp)(const struct sock *sk1,
164 const struct sock *sk2))
165 {
166 struct udp_hslot *hslot;
167 struct udp_table *udptable = sk->sk_prot->h.udp_table;
168 int error = 1;
169 struct net *net = sock_net(sk);
170
171 if (!snum) {
172 int low, high, remaining;
173 unsigned rand;
174 unsigned short first, last;
175 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
176
177 inet_get_local_port_range(&low, &high);
178 remaining = (high - low) + 1;
179
180 rand = net_random();
181 first = (((u64)rand * remaining) >> 32) + low;
182 /*
183 * force rand to be an odd multiple of UDP_HTABLE_SIZE
184 */
185 rand = (rand | 1) * (udptable->mask + 1);
186 for (last = first + udptable->mask + 1;
187 first != last;
188 first++) {
189 hslot = udp_hashslot(udptable, net, first);
190 bitmap_zero(bitmap, PORTS_PER_CHAIN);
191 spin_lock_bh(&hslot->lock);
192 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
193 saddr_comp, udptable->log);
194
195 snum = first;
196 /*
197 * Iterate on all possible values of snum for this hash.
198 * Using steps of an odd multiple of UDP_HTABLE_SIZE
199 * give us randomization and full range coverage.
200 */
201 do {
202 if (low <= snum && snum <= high &&
203 !test_bit(snum >> udptable->log, bitmap))
204 goto found;
205 snum += rand;
206 } while (snum != first);
207 spin_unlock_bh(&hslot->lock);
208 }
209 goto fail;
210 } else {
211 hslot = udp_hashslot(udptable, net, snum);
212 spin_lock_bh(&hslot->lock);
213 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
214 saddr_comp, 0))
215 goto fail_unlock;
216 }
217 found:
218 inet_sk(sk)->inet_num = snum;
219 udp_sk(sk)->udp_port_hash = snum;
220 udp_sk(sk)->udp_portaddr_hash ^= snum;
221 if (sk_unhashed(sk)) {
222 sk_nulls_add_node_rcu(sk, &hslot->head);
223 hslot->count++;
224 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
225 }
226 error = 0;
227 fail_unlock:
228 spin_unlock_bh(&hslot->lock);
229 fail:
230 return error;
231 }
232 EXPORT_SYMBOL(udp_lib_get_port);
233
234 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
235 {
236 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
237
238 return (!ipv6_only_sock(sk2) &&
239 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
240 inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
241 }
242
243 static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
244 unsigned int port)
245 {
246 return jhash_1word(saddr, net_hash_mix(net)) ^ port;
247 }
248
249 int udp_v4_get_port(struct sock *sk, unsigned short snum)
250 {
251 /* precompute partial secondary hash */
252 udp_sk(sk)->udp_portaddr_hash =
253 udp4_portaddr_hash(sock_net(sk),
254 inet_sk(sk)->inet_rcv_saddr,
255 0);
256 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal);
257 }
258
259 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
260 unsigned short hnum,
261 __be16 sport, __be32 daddr, __be16 dport, int dif)
262 {
263 int score = -1;
264
265 if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
266 !ipv6_only_sock(sk)) {
267 struct inet_sock *inet = inet_sk(sk);
268
269 score = (sk->sk_family == PF_INET ? 1 : 0);
270 if (inet->inet_rcv_saddr) {
271 if (inet->inet_rcv_saddr != daddr)
272 return -1;
273 score += 2;
274 }
275 if (inet->inet_daddr) {
276 if (inet->inet_daddr != saddr)
277 return -1;
278 score += 2;
279 }
280 if (inet->inet_dport) {
281 if (inet->inet_dport != sport)
282 return -1;
283 score += 2;
284 }
285 if (sk->sk_bound_dev_if) {
286 if (sk->sk_bound_dev_if != dif)
287 return -1;
288 score += 2;
289 }
290 }
291 return score;
292 }
293
294 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
295 * harder than this. -DaveM
296 */
297 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
298 __be16 sport, __be32 daddr, __be16 dport,
299 int dif, struct udp_table *udptable)
300 {
301 struct sock *sk, *result;
302 struct hlist_nulls_node *node;
303 unsigned short hnum = ntohs(dport);
304 unsigned int hash = udp_hashfn(net, hnum, udptable->mask);
305 struct udp_hslot *hslot = &udptable->hash[hash];
306 int score, badness;
307
308 rcu_read_lock();
309 begin:
310 result = NULL;
311 badness = -1;
312 sk_nulls_for_each_rcu(sk, node, &hslot->head) {
313 score = compute_score(sk, net, saddr, hnum, sport,
314 daddr, dport, dif);
315 if (score > badness) {
316 result = sk;
317 badness = score;
318 }
319 }
320 /*
321 * if the nulls value we got at the end of this lookup is
322 * not the expected one, we must restart lookup.
323 * We probably met an item that was moved to another chain.
324 */
325 if (get_nulls_value(node) != hash)
326 goto begin;
327
328 if (result) {
329 if (unlikely(!atomic_inc_not_zero(&result->sk_refcnt)))
330 result = NULL;
331 else if (unlikely(compute_score(result, net, saddr, hnum, sport,
332 daddr, dport, dif) < badness)) {
333 sock_put(result);
334 goto begin;
335 }
336 }
337 rcu_read_unlock();
338 return result;
339 }
340
341 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
342 __be16 sport, __be16 dport,
343 struct udp_table *udptable)
344 {
345 struct sock *sk;
346 const struct iphdr *iph = ip_hdr(skb);
347
348 if (unlikely(sk = skb_steal_sock(skb)))
349 return sk;
350 else
351 return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
352 iph->daddr, dport, inet_iif(skb),
353 udptable);
354 }
355
356 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
357 __be32 daddr, __be16 dport, int dif)
358 {
359 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
360 }
361 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
362
363 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
364 __be16 loc_port, __be32 loc_addr,
365 __be16 rmt_port, __be32 rmt_addr,
366 int dif)
367 {
368 struct hlist_nulls_node *node;
369 struct sock *s = sk;
370 unsigned short hnum = ntohs(loc_port);
371
372 sk_nulls_for_each_from(s, node) {
373 struct inet_sock *inet = inet_sk(s);
374
375 if (!net_eq(sock_net(s), net) ||
376 udp_sk(s)->udp_port_hash != hnum ||
377 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
378 (inet->inet_dport != rmt_port && inet->inet_dport) ||
379 (inet->inet_rcv_saddr &&
380 inet->inet_rcv_saddr != loc_addr) ||
381 ipv6_only_sock(s) ||
382 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
383 continue;
384 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
385 continue;
386 goto found;
387 }
388 s = NULL;
389 found:
390 return s;
391 }
392
393 /*
394 * This routine is called by the ICMP module when it gets some
395 * sort of error condition. If err < 0 then the socket should
396 * be closed and the error returned to the user. If err > 0
397 * it's just the icmp type << 8 | icmp code.
398 * Header points to the ip header of the error packet. We move
399 * on past this. Then (as it used to claim before adjustment)
400 * header points to the first 8 bytes of the udp header. We need
401 * to find the appropriate port.
402 */
403
404 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
405 {
406 struct inet_sock *inet;
407 struct iphdr *iph = (struct iphdr *)skb->data;
408 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
409 const int type = icmp_hdr(skb)->type;
410 const int code = icmp_hdr(skb)->code;
411 struct sock *sk;
412 int harderr;
413 int err;
414 struct net *net = dev_net(skb->dev);
415
416 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
417 iph->saddr, uh->source, skb->dev->ifindex, udptable);
418 if (sk == NULL) {
419 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
420 return; /* No socket for error */
421 }
422
423 err = 0;
424 harderr = 0;
425 inet = inet_sk(sk);
426
427 switch (type) {
428 default:
429 case ICMP_TIME_EXCEEDED:
430 err = EHOSTUNREACH;
431 break;
432 case ICMP_SOURCE_QUENCH:
433 goto out;
434 case ICMP_PARAMETERPROB:
435 err = EPROTO;
436 harderr = 1;
437 break;
438 case ICMP_DEST_UNREACH:
439 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
440 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
441 err = EMSGSIZE;
442 harderr = 1;
443 break;
444 }
445 goto out;
446 }
447 err = EHOSTUNREACH;
448 if (code <= NR_ICMP_UNREACH) {
449 harderr = icmp_err_convert[code].fatal;
450 err = icmp_err_convert[code].errno;
451 }
452 break;
453 }
454
455 /*
456 * RFC1122: OK. Passes ICMP errors back to application, as per
457 * 4.1.3.3.
458 */
459 if (!inet->recverr) {
460 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
461 goto out;
462 } else {
463 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
464 }
465 sk->sk_err = err;
466 sk->sk_error_report(sk);
467 out:
468 sock_put(sk);
469 }
470
471 void udp_err(struct sk_buff *skb, u32 info)
472 {
473 __udp4_lib_err(skb, info, &udp_table);
474 }
475
476 /*
477 * Throw away all pending data and cancel the corking. Socket is locked.
478 */
479 void udp_flush_pending_frames(struct sock *sk)
480 {
481 struct udp_sock *up = udp_sk(sk);
482
483 if (up->pending) {
484 up->len = 0;
485 up->pending = 0;
486 ip_flush_pending_frames(sk);
487 }
488 }
489 EXPORT_SYMBOL(udp_flush_pending_frames);
490
491 /**
492 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
493 * @sk: socket we are sending on
494 * @skb: sk_buff containing the filled-in UDP header
495 * (checksum field must be zeroed out)
496 */
497 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
498 __be32 src, __be32 dst, int len)
499 {
500 unsigned int offset;
501 struct udphdr *uh = udp_hdr(skb);
502 __wsum csum = 0;
503
504 if (skb_queue_len(&sk->sk_write_queue) == 1) {
505 /*
506 * Only one fragment on the socket.
507 */
508 skb->csum_start = skb_transport_header(skb) - skb->head;
509 skb->csum_offset = offsetof(struct udphdr, check);
510 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
511 } else {
512 /*
513 * HW-checksum won't work as there are two or more
514 * fragments on the socket so that all csums of sk_buffs
515 * should be together
516 */
517 offset = skb_transport_offset(skb);
518 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
519
520 skb->ip_summed = CHECKSUM_NONE;
521
522 skb_queue_walk(&sk->sk_write_queue, skb) {
523 csum = csum_add(csum, skb->csum);
524 }
525
526 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
527 if (uh->check == 0)
528 uh->check = CSUM_MANGLED_0;
529 }
530 }
531
532 /*
533 * Push out all pending data as one UDP datagram. Socket is locked.
534 */
535 static int udp_push_pending_frames(struct sock *sk)
536 {
537 struct udp_sock *up = udp_sk(sk);
538 struct inet_sock *inet = inet_sk(sk);
539 struct flowi *fl = &inet->cork.fl;
540 struct sk_buff *skb;
541 struct udphdr *uh;
542 int err = 0;
543 int is_udplite = IS_UDPLITE(sk);
544 __wsum csum = 0;
545
546 /* Grab the skbuff where UDP header space exists. */
547 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
548 goto out;
549
550 /*
551 * Create a UDP header
552 */
553 uh = udp_hdr(skb);
554 uh->source = fl->fl_ip_sport;
555 uh->dest = fl->fl_ip_dport;
556 uh->len = htons(up->len);
557 uh->check = 0;
558
559 if (is_udplite) /* UDP-Lite */
560 csum = udplite_csum_outgoing(sk, skb);
561
562 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
563
564 skb->ip_summed = CHECKSUM_NONE;
565 goto send;
566
567 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
568
569 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src, fl->fl4_dst, up->len);
570 goto send;
571
572 } else /* `normal' UDP */
573 csum = udp_csum_outgoing(sk, skb);
574
575 /* add protocol-dependent pseudo-header */
576 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
577 sk->sk_protocol, csum);
578 if (uh->check == 0)
579 uh->check = CSUM_MANGLED_0;
580
581 send:
582 err = ip_push_pending_frames(sk);
583 if (err) {
584 if (err == -ENOBUFS && !inet->recverr) {
585 UDP_INC_STATS_USER(sock_net(sk),
586 UDP_MIB_SNDBUFERRORS, is_udplite);
587 err = 0;
588 }
589 } else
590 UDP_INC_STATS_USER(sock_net(sk),
591 UDP_MIB_OUTDATAGRAMS, is_udplite);
592 out:
593 up->len = 0;
594 up->pending = 0;
595 return err;
596 }
597
598 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
599 size_t len)
600 {
601 struct inet_sock *inet = inet_sk(sk);
602 struct udp_sock *up = udp_sk(sk);
603 int ulen = len;
604 struct ipcm_cookie ipc;
605 struct rtable *rt = NULL;
606 int free = 0;
607 int connected = 0;
608 __be32 daddr, faddr, saddr;
609 __be16 dport;
610 u8 tos;
611 int err, is_udplite = IS_UDPLITE(sk);
612 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
613 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
614
615 if (len > 0xFFFF)
616 return -EMSGSIZE;
617
618 /*
619 * Check the flags.
620 */
621
622 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
623 return -EOPNOTSUPP;
624
625 ipc.opt = NULL;
626 ipc.shtx.flags = 0;
627
628 if (up->pending) {
629 /*
630 * There are pending frames.
631 * The socket lock must be held while it's corked.
632 */
633 lock_sock(sk);
634 if (likely(up->pending)) {
635 if (unlikely(up->pending != AF_INET)) {
636 release_sock(sk);
637 return -EINVAL;
638 }
639 goto do_append_data;
640 }
641 release_sock(sk);
642 }
643 ulen += sizeof(struct udphdr);
644
645 /*
646 * Get and verify the address.
647 */
648 if (msg->msg_name) {
649 struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name;
650 if (msg->msg_namelen < sizeof(*usin))
651 return -EINVAL;
652 if (usin->sin_family != AF_INET) {
653 if (usin->sin_family != AF_UNSPEC)
654 return -EAFNOSUPPORT;
655 }
656
657 daddr = usin->sin_addr.s_addr;
658 dport = usin->sin_port;
659 if (dport == 0)
660 return -EINVAL;
661 } else {
662 if (sk->sk_state != TCP_ESTABLISHED)
663 return -EDESTADDRREQ;
664 daddr = inet->inet_daddr;
665 dport = inet->inet_dport;
666 /* Open fast path for connected socket.
667 Route will not be used, if at least one option is set.
668 */
669 connected = 1;
670 }
671 ipc.addr = inet->inet_saddr;
672
673 ipc.oif = sk->sk_bound_dev_if;
674 err = sock_tx_timestamp(msg, sk, &ipc.shtx);
675 if (err)
676 return err;
677 if (msg->msg_controllen) {
678 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
679 if (err)
680 return err;
681 if (ipc.opt)
682 free = 1;
683 connected = 0;
684 }
685 if (!ipc.opt)
686 ipc.opt = inet->opt;
687
688 saddr = ipc.addr;
689 ipc.addr = faddr = daddr;
690
691 if (ipc.opt && ipc.opt->srr) {
692 if (!daddr)
693 return -EINVAL;
694 faddr = ipc.opt->faddr;
695 connected = 0;
696 }
697 tos = RT_TOS(inet->tos);
698 if (sock_flag(sk, SOCK_LOCALROUTE) ||
699 (msg->msg_flags & MSG_DONTROUTE) ||
700 (ipc.opt && ipc.opt->is_strictroute)) {
701 tos |= RTO_ONLINK;
702 connected = 0;
703 }
704
705 if (ipv4_is_multicast(daddr)) {
706 if (!ipc.oif)
707 ipc.oif = inet->mc_index;
708 if (!saddr)
709 saddr = inet->mc_addr;
710 connected = 0;
711 }
712
713 if (connected)
714 rt = (struct rtable *)sk_dst_check(sk, 0);
715
716 if (rt == NULL) {
717 struct flowi fl = { .oif = ipc.oif,
718 .mark = sk->sk_mark,
719 .nl_u = { .ip4_u =
720 { .daddr = faddr,
721 .saddr = saddr,
722 .tos = tos } },
723 .proto = sk->sk_protocol,
724 .flags = inet_sk_flowi_flags(sk),
725 .uli_u = { .ports =
726 { .sport = inet->inet_sport,
727 .dport = dport } } };
728 struct net *net = sock_net(sk);
729
730 security_sk_classify_flow(sk, &fl);
731 err = ip_route_output_flow(net, &rt, &fl, sk, 1);
732 if (err) {
733 if (err == -ENETUNREACH)
734 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
735 goto out;
736 }
737
738 err = -EACCES;
739 if ((rt->rt_flags & RTCF_BROADCAST) &&
740 !sock_flag(sk, SOCK_BROADCAST))
741 goto out;
742 if (connected)
743 sk_dst_set(sk, dst_clone(&rt->u.dst));
744 }
745
746 if (msg->msg_flags&MSG_CONFIRM)
747 goto do_confirm;
748 back_from_confirm:
749
750 saddr = rt->rt_src;
751 if (!ipc.addr)
752 daddr = ipc.addr = rt->rt_dst;
753
754 lock_sock(sk);
755 if (unlikely(up->pending)) {
756 /* The socket is already corked while preparing it. */
757 /* ... which is an evident application bug. --ANK */
758 release_sock(sk);
759
760 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
761 err = -EINVAL;
762 goto out;
763 }
764 /*
765 * Now cork the socket to pend data.
766 */
767 inet->cork.fl.fl4_dst = daddr;
768 inet->cork.fl.fl_ip_dport = dport;
769 inet->cork.fl.fl4_src = saddr;
770 inet->cork.fl.fl_ip_sport = inet->inet_sport;
771 up->pending = AF_INET;
772
773 do_append_data:
774 up->len += ulen;
775 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
776 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
777 sizeof(struct udphdr), &ipc, &rt,
778 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
779 if (err)
780 udp_flush_pending_frames(sk);
781 else if (!corkreq)
782 err = udp_push_pending_frames(sk);
783 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
784 up->pending = 0;
785 release_sock(sk);
786
787 out:
788 ip_rt_put(rt);
789 if (free)
790 kfree(ipc.opt);
791 if (!err)
792 return len;
793 /*
794 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
795 * ENOBUFS might not be good (it's not tunable per se), but otherwise
796 * we don't have a good statistic (IpOutDiscards but it can be too many
797 * things). We could add another new stat but at least for now that
798 * seems like overkill.
799 */
800 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
801 UDP_INC_STATS_USER(sock_net(sk),
802 UDP_MIB_SNDBUFERRORS, is_udplite);
803 }
804 return err;
805
806 do_confirm:
807 dst_confirm(&rt->u.dst);
808 if (!(msg->msg_flags&MSG_PROBE) || len)
809 goto back_from_confirm;
810 err = 0;
811 goto out;
812 }
813 EXPORT_SYMBOL(udp_sendmsg);
814
815 int udp_sendpage(struct sock *sk, struct page *page, int offset,
816 size_t size, int flags)
817 {
818 struct udp_sock *up = udp_sk(sk);
819 int ret;
820
821 if (!up->pending) {
822 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
823
824 /* Call udp_sendmsg to specify destination address which
825 * sendpage interface can't pass.
826 * This will succeed only when the socket is connected.
827 */
828 ret = udp_sendmsg(NULL, sk, &msg, 0);
829 if (ret < 0)
830 return ret;
831 }
832
833 lock_sock(sk);
834
835 if (unlikely(!up->pending)) {
836 release_sock(sk);
837
838 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
839 return -EINVAL;
840 }
841
842 ret = ip_append_page(sk, page, offset, size, flags);
843 if (ret == -EOPNOTSUPP) {
844 release_sock(sk);
845 return sock_no_sendpage(sk->sk_socket, page, offset,
846 size, flags);
847 }
848 if (ret < 0) {
849 udp_flush_pending_frames(sk);
850 goto out;
851 }
852
853 up->len += size;
854 if (!(up->corkflag || (flags&MSG_MORE)))
855 ret = udp_push_pending_frames(sk);
856 if (!ret)
857 ret = size;
858 out:
859 release_sock(sk);
860 return ret;
861 }
862
863
864 /**
865 * first_packet_length - return length of first packet in receive queue
866 * @sk: socket
867 *
868 * Drops all bad checksum frames, until a valid one is found.
869 * Returns the length of found skb, or 0 if none is found.
870 */
871 static unsigned int first_packet_length(struct sock *sk)
872 {
873 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
874 struct sk_buff *skb;
875 unsigned int res;
876
877 __skb_queue_head_init(&list_kill);
878
879 spin_lock_bh(&rcvq->lock);
880 while ((skb = skb_peek(rcvq)) != NULL &&
881 udp_lib_checksum_complete(skb)) {
882 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
883 IS_UDPLITE(sk));
884 atomic_inc(&sk->sk_drops);
885 __skb_unlink(skb, rcvq);
886 __skb_queue_tail(&list_kill, skb);
887 }
888 res = skb ? skb->len : 0;
889 spin_unlock_bh(&rcvq->lock);
890
891 if (!skb_queue_empty(&list_kill)) {
892 lock_sock(sk);
893 __skb_queue_purge(&list_kill);
894 sk_mem_reclaim_partial(sk);
895 release_sock(sk);
896 }
897 return res;
898 }
899
900 /*
901 * IOCTL requests applicable to the UDP protocol
902 */
903
904 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
905 {
906 switch (cmd) {
907 case SIOCOUTQ:
908 {
909 int amount = sk_wmem_alloc_get(sk);
910
911 return put_user(amount, (int __user *)arg);
912 }
913
914 case SIOCINQ:
915 {
916 unsigned int amount = first_packet_length(sk);
917
918 if (amount)
919 /*
920 * We will only return the amount
921 * of this packet since that is all
922 * that will be read.
923 */
924 amount -= sizeof(struct udphdr);
925
926 return put_user(amount, (int __user *)arg);
927 }
928
929 default:
930 return -ENOIOCTLCMD;
931 }
932
933 return 0;
934 }
935 EXPORT_SYMBOL(udp_ioctl);
936
937 /*
938 * This should be easy, if there is something there we
939 * return it, otherwise we block.
940 */
941
942 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
943 size_t len, int noblock, int flags, int *addr_len)
944 {
945 struct inet_sock *inet = inet_sk(sk);
946 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
947 struct sk_buff *skb;
948 unsigned int ulen, copied;
949 int peeked;
950 int err;
951 int is_udplite = IS_UDPLITE(sk);
952
953 /*
954 * Check any passed addresses
955 */
956 if (addr_len)
957 *addr_len = sizeof(*sin);
958
959 if (flags & MSG_ERRQUEUE)
960 return ip_recv_error(sk, msg, len);
961
962 try_again:
963 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
964 &peeked, &err);
965 if (!skb)
966 goto out;
967
968 ulen = skb->len - sizeof(struct udphdr);
969 copied = len;
970 if (copied > ulen)
971 copied = ulen;
972 else if (copied < ulen)
973 msg->msg_flags |= MSG_TRUNC;
974
975 /*
976 * If checksum is needed at all, try to do it while copying the
977 * data. If the data is truncated, or if we only want a partial
978 * coverage checksum (UDP-Lite), do it before the copy.
979 */
980
981 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
982 if (udp_lib_checksum_complete(skb))
983 goto csum_copy_err;
984 }
985
986 if (skb_csum_unnecessary(skb))
987 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
988 msg->msg_iov, copied);
989 else {
990 err = skb_copy_and_csum_datagram_iovec(skb,
991 sizeof(struct udphdr),
992 msg->msg_iov);
993
994 if (err == -EINVAL)
995 goto csum_copy_err;
996 }
997
998 if (err)
999 goto out_free;
1000
1001 if (!peeked)
1002 UDP_INC_STATS_USER(sock_net(sk),
1003 UDP_MIB_INDATAGRAMS, is_udplite);
1004
1005 sock_recv_ts_and_drops(msg, sk, skb);
1006
1007 /* Copy the address. */
1008 if (sin) {
1009 sin->sin_family = AF_INET;
1010 sin->sin_port = udp_hdr(skb)->source;
1011 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1012 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1013 }
1014 if (inet->cmsg_flags)
1015 ip_cmsg_recv(msg, skb);
1016
1017 err = copied;
1018 if (flags & MSG_TRUNC)
1019 err = ulen;
1020
1021 out_free:
1022 skb_free_datagram_locked(sk, skb);
1023 out:
1024 return err;
1025
1026 csum_copy_err:
1027 lock_sock(sk);
1028 if (!skb_kill_datagram(sk, skb, flags))
1029 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1030 release_sock(sk);
1031
1032 if (noblock)
1033 return -EAGAIN;
1034 goto try_again;
1035 }
1036
1037
1038 int udp_disconnect(struct sock *sk, int flags)
1039 {
1040 struct inet_sock *inet = inet_sk(sk);
1041 /*
1042 * 1003.1g - break association.
1043 */
1044
1045 sk->sk_state = TCP_CLOSE;
1046 inet->inet_daddr = 0;
1047 inet->inet_dport = 0;
1048 sk->sk_bound_dev_if = 0;
1049 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1050 inet_reset_saddr(sk);
1051
1052 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1053 sk->sk_prot->unhash(sk);
1054 inet->inet_sport = 0;
1055 }
1056 sk_dst_reset(sk);
1057 return 0;
1058 }
1059 EXPORT_SYMBOL(udp_disconnect);
1060
1061 void udp_lib_unhash(struct sock *sk)
1062 {
1063 if (sk_hashed(sk)) {
1064 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1065 struct udp_hslot *hslot = udp_hashslot(udptable, sock_net(sk),
1066 udp_sk(sk)->udp_port_hash);
1067
1068 spin_lock_bh(&hslot->lock);
1069 if (sk_nulls_del_node_init_rcu(sk)) {
1070 hslot->count--;
1071 inet_sk(sk)->inet_num = 0;
1072 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1073 }
1074 spin_unlock_bh(&hslot->lock);
1075 }
1076 }
1077 EXPORT_SYMBOL(udp_lib_unhash);
1078
1079 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1080 {
1081 int rc = sock_queue_rcv_skb(sk, skb);
1082
1083 if (rc < 0) {
1084 int is_udplite = IS_UDPLITE(sk);
1085
1086 /* Note that an ENOMEM error is charged twice */
1087 if (rc == -ENOMEM)
1088 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1089 is_udplite);
1090 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1091 kfree_skb(skb);
1092 return -1;
1093 }
1094
1095 return 0;
1096
1097 }
1098
1099 /* returns:
1100 * -1: error
1101 * 0: success
1102 * >0: "udp encap" protocol resubmission
1103 *
1104 * Note that in the success and error cases, the skb is assumed to
1105 * have either been requeued or freed.
1106 */
1107 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1108 {
1109 struct udp_sock *up = udp_sk(sk);
1110 int rc;
1111 int is_udplite = IS_UDPLITE(sk);
1112
1113 /*
1114 * Charge it to the socket, dropping if the queue is full.
1115 */
1116 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1117 goto drop;
1118 nf_reset(skb);
1119
1120 if (up->encap_type) {
1121 /*
1122 * This is an encapsulation socket so pass the skb to
1123 * the socket's udp_encap_rcv() hook. Otherwise, just
1124 * fall through and pass this up the UDP socket.
1125 * up->encap_rcv() returns the following value:
1126 * =0 if skb was successfully passed to the encap
1127 * handler or was discarded by it.
1128 * >0 if skb should be passed on to UDP.
1129 * <0 if skb should be resubmitted as proto -N
1130 */
1131
1132 /* if we're overly short, let UDP handle it */
1133 if (skb->len > sizeof(struct udphdr) &&
1134 up->encap_rcv != NULL) {
1135 int ret;
1136
1137 ret = (*up->encap_rcv)(sk, skb);
1138 if (ret <= 0) {
1139 UDP_INC_STATS_BH(sock_net(sk),
1140 UDP_MIB_INDATAGRAMS,
1141 is_udplite);
1142 return -ret;
1143 }
1144 }
1145
1146 /* FALLTHROUGH -- it's a UDP Packet */
1147 }
1148
1149 /*
1150 * UDP-Lite specific tests, ignored on UDP sockets
1151 */
1152 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1153
1154 /*
1155 * MIB statistics other than incrementing the error count are
1156 * disabled for the following two types of errors: these depend
1157 * on the application settings, not on the functioning of the
1158 * protocol stack as such.
1159 *
1160 * RFC 3828 here recommends (sec 3.3): "There should also be a
1161 * way ... to ... at least let the receiving application block
1162 * delivery of packets with coverage values less than a value
1163 * provided by the application."
1164 */
1165 if (up->pcrlen == 0) { /* full coverage was set */
1166 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1167 "%d while full coverage %d requested\n",
1168 UDP_SKB_CB(skb)->cscov, skb->len);
1169 goto drop;
1170 }
1171 /* The next case involves violating the min. coverage requested
1172 * by the receiver. This is subtle: if receiver wants x and x is
1173 * greater than the buffersize/MTU then receiver will complain
1174 * that it wants x while sender emits packets of smaller size y.
1175 * Therefore the above ...()->partial_cov statement is essential.
1176 */
1177 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1178 LIMIT_NETDEBUG(KERN_WARNING
1179 "UDPLITE: coverage %d too small, need min %d\n",
1180 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1181 goto drop;
1182 }
1183 }
1184
1185 if (sk->sk_filter) {
1186 if (udp_lib_checksum_complete(skb))
1187 goto drop;
1188 }
1189
1190 rc = 0;
1191
1192 bh_lock_sock(sk);
1193 if (!sock_owned_by_user(sk))
1194 rc = __udp_queue_rcv_skb(sk, skb);
1195 else
1196 sk_add_backlog(sk, skb);
1197 bh_unlock_sock(sk);
1198
1199 return rc;
1200
1201 drop:
1202 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1203 atomic_inc(&sk->sk_drops);
1204 kfree_skb(skb);
1205 return -1;
1206 }
1207
1208 /*
1209 * Multicasts and broadcasts go to each listener.
1210 *
1211 * Note: called only from the BH handler context,
1212 * so we don't need to lock the hashes.
1213 */
1214 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1215 struct udphdr *uh,
1216 __be32 saddr, __be32 daddr,
1217 struct udp_table *udptable)
1218 {
1219 struct sock *sk;
1220 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1221 int dif;
1222
1223 spin_lock(&hslot->lock);
1224 sk = sk_nulls_head(&hslot->head);
1225 dif = skb->dev->ifindex;
1226 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1227 if (sk) {
1228 struct sock *sknext = NULL;
1229
1230 do {
1231 struct sk_buff *skb1 = skb;
1232
1233 sknext = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1234 daddr, uh->source, saddr,
1235 dif);
1236 if (sknext)
1237 skb1 = skb_clone(skb, GFP_ATOMIC);
1238
1239 if (skb1) {
1240 int ret = udp_queue_rcv_skb(sk, skb1);
1241 if (ret > 0)
1242 /* we should probably re-process instead
1243 * of dropping packets here. */
1244 kfree_skb(skb1);
1245 }
1246 sk = sknext;
1247 } while (sknext);
1248 } else
1249 consume_skb(skb);
1250 spin_unlock(&hslot->lock);
1251 return 0;
1252 }
1253
1254 /* Initialize UDP checksum. If exited with zero value (success),
1255 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1256 * Otherwise, csum completion requires chacksumming packet body,
1257 * including udp header and folding it to skb->csum.
1258 */
1259 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1260 int proto)
1261 {
1262 const struct iphdr *iph;
1263 int err;
1264
1265 UDP_SKB_CB(skb)->partial_cov = 0;
1266 UDP_SKB_CB(skb)->cscov = skb->len;
1267
1268 if (proto == IPPROTO_UDPLITE) {
1269 err = udplite_checksum_init(skb, uh);
1270 if (err)
1271 return err;
1272 }
1273
1274 iph = ip_hdr(skb);
1275 if (uh->check == 0) {
1276 skb->ip_summed = CHECKSUM_UNNECESSARY;
1277 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1278 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1279 proto, skb->csum))
1280 skb->ip_summed = CHECKSUM_UNNECESSARY;
1281 }
1282 if (!skb_csum_unnecessary(skb))
1283 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1284 skb->len, proto, 0);
1285 /* Probably, we should checksum udp header (it should be in cache
1286 * in any case) and data in tiny packets (< rx copybreak).
1287 */
1288
1289 return 0;
1290 }
1291
1292 /*
1293 * All we need to do is get the socket, and then do a checksum.
1294 */
1295
1296 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1297 int proto)
1298 {
1299 struct sock *sk;
1300 struct udphdr *uh;
1301 unsigned short ulen;
1302 struct rtable *rt = skb_rtable(skb);
1303 __be32 saddr, daddr;
1304 struct net *net = dev_net(skb->dev);
1305
1306 /*
1307 * Validate the packet.
1308 */
1309 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1310 goto drop; /* No space for header. */
1311
1312 uh = udp_hdr(skb);
1313 ulen = ntohs(uh->len);
1314 if (ulen > skb->len)
1315 goto short_packet;
1316
1317 if (proto == IPPROTO_UDP) {
1318 /* UDP validates ulen. */
1319 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1320 goto short_packet;
1321 uh = udp_hdr(skb);
1322 }
1323
1324 if (udp4_csum_init(skb, uh, proto))
1325 goto csum_error;
1326
1327 saddr = ip_hdr(skb)->saddr;
1328 daddr = ip_hdr(skb)->daddr;
1329
1330 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1331 return __udp4_lib_mcast_deliver(net, skb, uh,
1332 saddr, daddr, udptable);
1333
1334 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1335
1336 if (sk != NULL) {
1337 int ret = udp_queue_rcv_skb(sk, skb);
1338 sock_put(sk);
1339
1340 /* a return value > 0 means to resubmit the input, but
1341 * it wants the return to be -protocol, or 0
1342 */
1343 if (ret > 0)
1344 return -ret;
1345 return 0;
1346 }
1347
1348 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1349 goto drop;
1350 nf_reset(skb);
1351
1352 /* No socket. Drop packet silently, if checksum is wrong */
1353 if (udp_lib_checksum_complete(skb))
1354 goto csum_error;
1355
1356 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1357 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1358
1359 /*
1360 * Hmm. We got an UDP packet to a port to which we
1361 * don't wanna listen. Ignore it.
1362 */
1363 kfree_skb(skb);
1364 return 0;
1365
1366 short_packet:
1367 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1368 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1369 &saddr,
1370 ntohs(uh->source),
1371 ulen,
1372 skb->len,
1373 &daddr,
1374 ntohs(uh->dest));
1375 goto drop;
1376
1377 csum_error:
1378 /*
1379 * RFC1122: OK. Discards the bad packet silently (as far as
1380 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1381 */
1382 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1383 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1384 &saddr,
1385 ntohs(uh->source),
1386 &daddr,
1387 ntohs(uh->dest),
1388 ulen);
1389 drop:
1390 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1391 kfree_skb(skb);
1392 return 0;
1393 }
1394
1395 int udp_rcv(struct sk_buff *skb)
1396 {
1397 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1398 }
1399
1400 void udp_destroy_sock(struct sock *sk)
1401 {
1402 lock_sock(sk);
1403 udp_flush_pending_frames(sk);
1404 release_sock(sk);
1405 }
1406
1407 /*
1408 * Socket option code for UDP
1409 */
1410 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1411 char __user *optval, unsigned int optlen,
1412 int (*push_pending_frames)(struct sock *))
1413 {
1414 struct udp_sock *up = udp_sk(sk);
1415 int val;
1416 int err = 0;
1417 int is_udplite = IS_UDPLITE(sk);
1418
1419 if (optlen < sizeof(int))
1420 return -EINVAL;
1421
1422 if (get_user(val, (int __user *)optval))
1423 return -EFAULT;
1424
1425 switch (optname) {
1426 case UDP_CORK:
1427 if (val != 0) {
1428 up->corkflag = 1;
1429 } else {
1430 up->corkflag = 0;
1431 lock_sock(sk);
1432 (*push_pending_frames)(sk);
1433 release_sock(sk);
1434 }
1435 break;
1436
1437 case UDP_ENCAP:
1438 switch (val) {
1439 case 0:
1440 case UDP_ENCAP_ESPINUDP:
1441 case UDP_ENCAP_ESPINUDP_NON_IKE:
1442 up->encap_rcv = xfrm4_udp_encap_rcv;
1443 /* FALLTHROUGH */
1444 case UDP_ENCAP_L2TPINUDP:
1445 up->encap_type = val;
1446 break;
1447 default:
1448 err = -ENOPROTOOPT;
1449 break;
1450 }
1451 break;
1452
1453 /*
1454 * UDP-Lite's partial checksum coverage (RFC 3828).
1455 */
1456 /* The sender sets actual checksum coverage length via this option.
1457 * The case coverage > packet length is handled by send module. */
1458 case UDPLITE_SEND_CSCOV:
1459 if (!is_udplite) /* Disable the option on UDP sockets */
1460 return -ENOPROTOOPT;
1461 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1462 val = 8;
1463 else if (val > USHORT_MAX)
1464 val = USHORT_MAX;
1465 up->pcslen = val;
1466 up->pcflag |= UDPLITE_SEND_CC;
1467 break;
1468
1469 /* The receiver specifies a minimum checksum coverage value. To make
1470 * sense, this should be set to at least 8 (as done below). If zero is
1471 * used, this again means full checksum coverage. */
1472 case UDPLITE_RECV_CSCOV:
1473 if (!is_udplite) /* Disable the option on UDP sockets */
1474 return -ENOPROTOOPT;
1475 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1476 val = 8;
1477 else if (val > USHORT_MAX)
1478 val = USHORT_MAX;
1479 up->pcrlen = val;
1480 up->pcflag |= UDPLITE_RECV_CC;
1481 break;
1482
1483 default:
1484 err = -ENOPROTOOPT;
1485 break;
1486 }
1487
1488 return err;
1489 }
1490 EXPORT_SYMBOL(udp_lib_setsockopt);
1491
1492 int udp_setsockopt(struct sock *sk, int level, int optname,
1493 char __user *optval, unsigned int optlen)
1494 {
1495 if (level == SOL_UDP || level == SOL_UDPLITE)
1496 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1497 udp_push_pending_frames);
1498 return ip_setsockopt(sk, level, optname, optval, optlen);
1499 }
1500
1501 #ifdef CONFIG_COMPAT
1502 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1503 char __user *optval, unsigned int optlen)
1504 {
1505 if (level == SOL_UDP || level == SOL_UDPLITE)
1506 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1507 udp_push_pending_frames);
1508 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1509 }
1510 #endif
1511
1512 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1513 char __user *optval, int __user *optlen)
1514 {
1515 struct udp_sock *up = udp_sk(sk);
1516 int val, len;
1517
1518 if (get_user(len, optlen))
1519 return -EFAULT;
1520
1521 len = min_t(unsigned int, len, sizeof(int));
1522
1523 if (len < 0)
1524 return -EINVAL;
1525
1526 switch (optname) {
1527 case UDP_CORK:
1528 val = up->corkflag;
1529 break;
1530
1531 case UDP_ENCAP:
1532 val = up->encap_type;
1533 break;
1534
1535 /* The following two cannot be changed on UDP sockets, the return is
1536 * always 0 (which corresponds to the full checksum coverage of UDP). */
1537 case UDPLITE_SEND_CSCOV:
1538 val = up->pcslen;
1539 break;
1540
1541 case UDPLITE_RECV_CSCOV:
1542 val = up->pcrlen;
1543 break;
1544
1545 default:
1546 return -ENOPROTOOPT;
1547 }
1548
1549 if (put_user(len, optlen))
1550 return -EFAULT;
1551 if (copy_to_user(optval, &val, len))
1552 return -EFAULT;
1553 return 0;
1554 }
1555 EXPORT_SYMBOL(udp_lib_getsockopt);
1556
1557 int udp_getsockopt(struct sock *sk, int level, int optname,
1558 char __user *optval, int __user *optlen)
1559 {
1560 if (level == SOL_UDP || level == SOL_UDPLITE)
1561 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1562 return ip_getsockopt(sk, level, optname, optval, optlen);
1563 }
1564
1565 #ifdef CONFIG_COMPAT
1566 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1567 char __user *optval, int __user *optlen)
1568 {
1569 if (level == SOL_UDP || level == SOL_UDPLITE)
1570 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1571 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1572 }
1573 #endif
1574 /**
1575 * udp_poll - wait for a UDP event.
1576 * @file - file struct
1577 * @sock - socket
1578 * @wait - poll table
1579 *
1580 * This is same as datagram poll, except for the special case of
1581 * blocking sockets. If application is using a blocking fd
1582 * and a packet with checksum error is in the queue;
1583 * then it could get return from select indicating data available
1584 * but then block when reading it. Add special case code
1585 * to work around these arguably broken applications.
1586 */
1587 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1588 {
1589 unsigned int mask = datagram_poll(file, sock, wait);
1590 struct sock *sk = sock->sk;
1591
1592 /* Check for false positives due to checksum errors */
1593 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1594 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1595 mask &= ~(POLLIN | POLLRDNORM);
1596
1597 return mask;
1598
1599 }
1600 EXPORT_SYMBOL(udp_poll);
1601
1602 struct proto udp_prot = {
1603 .name = "UDP",
1604 .owner = THIS_MODULE,
1605 .close = udp_lib_close,
1606 .connect = ip4_datagram_connect,
1607 .disconnect = udp_disconnect,
1608 .ioctl = udp_ioctl,
1609 .destroy = udp_destroy_sock,
1610 .setsockopt = udp_setsockopt,
1611 .getsockopt = udp_getsockopt,
1612 .sendmsg = udp_sendmsg,
1613 .recvmsg = udp_recvmsg,
1614 .sendpage = udp_sendpage,
1615 .backlog_rcv = __udp_queue_rcv_skb,
1616 .hash = udp_lib_hash,
1617 .unhash = udp_lib_unhash,
1618 .get_port = udp_v4_get_port,
1619 .memory_allocated = &udp_memory_allocated,
1620 .sysctl_mem = sysctl_udp_mem,
1621 .sysctl_wmem = &sysctl_udp_wmem_min,
1622 .sysctl_rmem = &sysctl_udp_rmem_min,
1623 .obj_size = sizeof(struct udp_sock),
1624 .slab_flags = SLAB_DESTROY_BY_RCU,
1625 .h.udp_table = &udp_table,
1626 #ifdef CONFIG_COMPAT
1627 .compat_setsockopt = compat_udp_setsockopt,
1628 .compat_getsockopt = compat_udp_getsockopt,
1629 #endif
1630 };
1631 EXPORT_SYMBOL(udp_prot);
1632
1633 /* ------------------------------------------------------------------------ */
1634 #ifdef CONFIG_PROC_FS
1635
1636 static struct sock *udp_get_first(struct seq_file *seq, int start)
1637 {
1638 struct sock *sk;
1639 struct udp_iter_state *state = seq->private;
1640 struct net *net = seq_file_net(seq);
1641
1642 for (state->bucket = start; state->bucket <= state->udp_table->mask;
1643 ++state->bucket) {
1644 struct hlist_nulls_node *node;
1645 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
1646
1647 if (hlist_nulls_empty(&hslot->head))
1648 continue;
1649
1650 spin_lock_bh(&hslot->lock);
1651 sk_nulls_for_each(sk, node, &hslot->head) {
1652 if (!net_eq(sock_net(sk), net))
1653 continue;
1654 if (sk->sk_family == state->family)
1655 goto found;
1656 }
1657 spin_unlock_bh(&hslot->lock);
1658 }
1659 sk = NULL;
1660 found:
1661 return sk;
1662 }
1663
1664 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1665 {
1666 struct udp_iter_state *state = seq->private;
1667 struct net *net = seq_file_net(seq);
1668
1669 do {
1670 sk = sk_nulls_next(sk);
1671 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1672
1673 if (!sk) {
1674 if (state->bucket <= state->udp_table->mask)
1675 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
1676 return udp_get_first(seq, state->bucket + 1);
1677 }
1678 return sk;
1679 }
1680
1681 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1682 {
1683 struct sock *sk = udp_get_first(seq, 0);
1684
1685 if (sk)
1686 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1687 --pos;
1688 return pos ? NULL : sk;
1689 }
1690
1691 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1692 {
1693 struct udp_iter_state *state = seq->private;
1694 state->bucket = MAX_UDP_PORTS;
1695
1696 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
1697 }
1698
1699 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1700 {
1701 struct sock *sk;
1702
1703 if (v == SEQ_START_TOKEN)
1704 sk = udp_get_idx(seq, 0);
1705 else
1706 sk = udp_get_next(seq, v);
1707
1708 ++*pos;
1709 return sk;
1710 }
1711
1712 static void udp_seq_stop(struct seq_file *seq, void *v)
1713 {
1714 struct udp_iter_state *state = seq->private;
1715
1716 if (state->bucket <= state->udp_table->mask)
1717 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
1718 }
1719
1720 static int udp_seq_open(struct inode *inode, struct file *file)
1721 {
1722 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1723 struct udp_iter_state *s;
1724 int err;
1725
1726 err = seq_open_net(inode, file, &afinfo->seq_ops,
1727 sizeof(struct udp_iter_state));
1728 if (err < 0)
1729 return err;
1730
1731 s = ((struct seq_file *)file->private_data)->private;
1732 s->family = afinfo->family;
1733 s->udp_table = afinfo->udp_table;
1734 return err;
1735 }
1736
1737 /* ------------------------------------------------------------------------ */
1738 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
1739 {
1740 struct proc_dir_entry *p;
1741 int rc = 0;
1742
1743 afinfo->seq_fops.open = udp_seq_open;
1744 afinfo->seq_fops.read = seq_read;
1745 afinfo->seq_fops.llseek = seq_lseek;
1746 afinfo->seq_fops.release = seq_release_net;
1747
1748 afinfo->seq_ops.start = udp_seq_start;
1749 afinfo->seq_ops.next = udp_seq_next;
1750 afinfo->seq_ops.stop = udp_seq_stop;
1751
1752 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
1753 &afinfo->seq_fops, afinfo);
1754 if (!p)
1755 rc = -ENOMEM;
1756 return rc;
1757 }
1758 EXPORT_SYMBOL(udp_proc_register);
1759
1760 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
1761 {
1762 proc_net_remove(net, afinfo->name);
1763 }
1764 EXPORT_SYMBOL(udp_proc_unregister);
1765
1766 /* ------------------------------------------------------------------------ */
1767 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
1768 int bucket, int *len)
1769 {
1770 struct inet_sock *inet = inet_sk(sp);
1771 __be32 dest = inet->inet_daddr;
1772 __be32 src = inet->inet_rcv_saddr;
1773 __u16 destp = ntohs(inet->inet_dport);
1774 __u16 srcp = ntohs(inet->inet_sport);
1775
1776 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
1777 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
1778 bucket, src, srcp, dest, destp, sp->sk_state,
1779 sk_wmem_alloc_get(sp),
1780 sk_rmem_alloc_get(sp),
1781 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1782 atomic_read(&sp->sk_refcnt), sp,
1783 atomic_read(&sp->sk_drops), len);
1784 }
1785
1786 int udp4_seq_show(struct seq_file *seq, void *v)
1787 {
1788 if (v == SEQ_START_TOKEN)
1789 seq_printf(seq, "%-127s\n",
1790 " sl local_address rem_address st tx_queue "
1791 "rx_queue tr tm->when retrnsmt uid timeout "
1792 "inode ref pointer drops");
1793 else {
1794 struct udp_iter_state *state = seq->private;
1795 int len;
1796
1797 udp4_format_sock(v, seq, state->bucket, &len);
1798 seq_printf(seq, "%*s\n", 127 - len, "");
1799 }
1800 return 0;
1801 }
1802
1803 /* ------------------------------------------------------------------------ */
1804 static struct udp_seq_afinfo udp4_seq_afinfo = {
1805 .name = "udp",
1806 .family = AF_INET,
1807 .udp_table = &udp_table,
1808 .seq_fops = {
1809 .owner = THIS_MODULE,
1810 },
1811 .seq_ops = {
1812 .show = udp4_seq_show,
1813 },
1814 };
1815
1816 static int udp4_proc_init_net(struct net *net)
1817 {
1818 return udp_proc_register(net, &udp4_seq_afinfo);
1819 }
1820
1821 static void udp4_proc_exit_net(struct net *net)
1822 {
1823 udp_proc_unregister(net, &udp4_seq_afinfo);
1824 }
1825
1826 static struct pernet_operations udp4_net_ops = {
1827 .init = udp4_proc_init_net,
1828 .exit = udp4_proc_exit_net,
1829 };
1830
1831 int __init udp4_proc_init(void)
1832 {
1833 return register_pernet_subsys(&udp4_net_ops);
1834 }
1835
1836 void udp4_proc_exit(void)
1837 {
1838 unregister_pernet_subsys(&udp4_net_ops);
1839 }
1840 #endif /* CONFIG_PROC_FS */
1841
1842 static __initdata unsigned long uhash_entries;
1843 static int __init set_uhash_entries(char *str)
1844 {
1845 if (!str)
1846 return 0;
1847 uhash_entries = simple_strtoul(str, &str, 0);
1848 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
1849 uhash_entries = UDP_HTABLE_SIZE_MIN;
1850 return 1;
1851 }
1852 __setup("uhash_entries=", set_uhash_entries);
1853
1854 void __init udp_table_init(struct udp_table *table, const char *name)
1855 {
1856 unsigned int i;
1857
1858 if (!CONFIG_BASE_SMALL)
1859 table->hash = alloc_large_system_hash(name,
1860 sizeof(struct udp_hslot),
1861 uhash_entries,
1862 21, /* one slot per 2 MB */
1863 0,
1864 &table->log,
1865 &table->mask,
1866 64 * 1024);
1867 /*
1868 * Make sure hash table has the minimum size
1869 */
1870 if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) {
1871 table->hash = kmalloc(UDP_HTABLE_SIZE_MIN *
1872 sizeof(struct udp_hslot), GFP_KERNEL);
1873 if (!table->hash)
1874 panic(name);
1875 table->log = ilog2(UDP_HTABLE_SIZE_MIN);
1876 table->mask = UDP_HTABLE_SIZE_MIN - 1;
1877 }
1878 for (i = 0; i <= table->mask; i++) {
1879 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
1880 table->hash[i].count = 0;
1881 spin_lock_init(&table->hash[i].lock);
1882 }
1883 }
1884
1885 void __init udp_init(void)
1886 {
1887 unsigned long nr_pages, limit;
1888
1889 udp_table_init(&udp_table, "UDP");
1890 /* Set the pressure threshold up by the same strategy of TCP. It is a
1891 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
1892 * toward zero with the amount of memory, with a floor of 128 pages.
1893 */
1894 nr_pages = totalram_pages - totalhigh_pages;
1895 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
1896 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
1897 limit = max(limit, 128UL);
1898 sysctl_udp_mem[0] = limit / 4 * 3;
1899 sysctl_udp_mem[1] = limit;
1900 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
1901
1902 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
1903 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
1904 }
1905
1906 int udp4_ufo_send_check(struct sk_buff *skb)
1907 {
1908 const struct iphdr *iph;
1909 struct udphdr *uh;
1910
1911 if (!pskb_may_pull(skb, sizeof(*uh)))
1912 return -EINVAL;
1913
1914 iph = ip_hdr(skb);
1915 uh = udp_hdr(skb);
1916
1917 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1918 IPPROTO_UDP, 0);
1919 skb->csum_start = skb_transport_header(skb) - skb->head;
1920 skb->csum_offset = offsetof(struct udphdr, check);
1921 skb->ip_summed = CHECKSUM_PARTIAL;
1922 return 0;
1923 }
1924
1925 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, int features)
1926 {
1927 struct sk_buff *segs = ERR_PTR(-EINVAL);
1928 unsigned int mss;
1929 int offset;
1930 __wsum csum;
1931
1932 mss = skb_shinfo(skb)->gso_size;
1933 if (unlikely(skb->len <= mss))
1934 goto out;
1935
1936 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
1937 /* Packet is from an untrusted source, reset gso_segs. */
1938 int type = skb_shinfo(skb)->gso_type;
1939
1940 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
1941 !(type & (SKB_GSO_UDP))))
1942 goto out;
1943
1944 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
1945
1946 segs = NULL;
1947 goto out;
1948 }
1949
1950 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
1951 * do checksum of UDP packets sent as multiple IP fragments.
1952 */
1953 offset = skb->csum_start - skb_headroom(skb);
1954 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1955 offset += skb->csum_offset;
1956 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1957 skb->ip_summed = CHECKSUM_NONE;
1958
1959 /* Fragment the skb. IP headers of the fragments are updated in
1960 * inet_gso_segment()
1961 */
1962 segs = skb_segment(skb, features);
1963 out:
1964 return segs;
1965 }
1966
Linux kernel - Deliverables
This page took 0.091443 seconds and 6 git commands to generate.