projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
net: sk_drops consolidation
[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)->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->rcv_saddr || !inet2->rcv_saddr ||
237 inet1->rcv_saddr == inet2->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->rcv_saddr) {
257 if (inet->rcv_saddr != daddr)
258 return -1;
259 score += 2;
260 }
261 if (inet->daddr) {
262 if (inet->daddr != saddr)
263 return -1;
264 score += 2;
265 }
266 if (inet->dport) {
267 if (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->daddr && inet->daddr != rmt_addr) ||
364 (inet->dport != rmt_port && inet->dport) ||
365 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
366 ipv6_only_sock(s) ||
367 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
368 continue;
369 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
370 continue;
371 goto found;
372 }
373 s = NULL;
374 found:
375 return s;
376 }
377
378 /*
379 * This routine is called by the ICMP module when it gets some
380 * sort of error condition. If err < 0 then the socket should
381 * be closed and the error returned to the user. If err > 0
382 * it's just the icmp type << 8 | icmp code.
383 * Header points to the ip header of the error packet. We move
384 * on past this. Then (as it used to claim before adjustment)
385 * header points to the first 8 bytes of the udp header. We need
386 * to find the appropriate port.
387 */
388
389 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
390 {
391 struct inet_sock *inet;
392 struct iphdr *iph = (struct iphdr *)skb->data;
393 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
394 const int type = icmp_hdr(skb)->type;
395 const int code = icmp_hdr(skb)->code;
396 struct sock *sk;
397 int harderr;
398 int err;
399 struct net *net = dev_net(skb->dev);
400
401 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
402 iph->saddr, uh->source, skb->dev->ifindex, udptable);
403 if (sk == NULL) {
404 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
405 return; /* No socket for error */
406 }
407
408 err = 0;
409 harderr = 0;
410 inet = inet_sk(sk);
411
412 switch (type) {
413 default:
414 case ICMP_TIME_EXCEEDED:
415 err = EHOSTUNREACH;
416 break;
417 case ICMP_SOURCE_QUENCH:
418 goto out;
419 case ICMP_PARAMETERPROB:
420 err = EPROTO;
421 harderr = 1;
422 break;
423 case ICMP_DEST_UNREACH:
424 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
425 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
426 err = EMSGSIZE;
427 harderr = 1;
428 break;
429 }
430 goto out;
431 }
432 err = EHOSTUNREACH;
433 if (code <= NR_ICMP_UNREACH) {
434 harderr = icmp_err_convert[code].fatal;
435 err = icmp_err_convert[code].errno;
436 }
437 break;
438 }
439
440 /*
441 * RFC1122: OK. Passes ICMP errors back to application, as per
442 * 4.1.3.3.
443 */
444 if (!inet->recverr) {
445 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
446 goto out;
447 } else {
448 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
449 }
450 sk->sk_err = err;
451 sk->sk_error_report(sk);
452 out:
453 sock_put(sk);
454 }
455
456 void udp_err(struct sk_buff *skb, u32 info)
457 {
458 __udp4_lib_err(skb, info, &udp_table);
459 }
460
461 /*
462 * Throw away all pending data and cancel the corking. Socket is locked.
463 */
464 void udp_flush_pending_frames(struct sock *sk)
465 {
466 struct udp_sock *up = udp_sk(sk);
467
468 if (up->pending) {
469 up->len = 0;
470 up->pending = 0;
471 ip_flush_pending_frames(sk);
472 }
473 }
474 EXPORT_SYMBOL(udp_flush_pending_frames);
475
476 /**
477 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
478 * @sk: socket we are sending on
479 * @skb: sk_buff containing the filled-in UDP header
480 * (checksum field must be zeroed out)
481 */
482 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
483 __be32 src, __be32 dst, int len)
484 {
485 unsigned int offset;
486 struct udphdr *uh = udp_hdr(skb);
487 __wsum csum = 0;
488
489 if (skb_queue_len(&sk->sk_write_queue) == 1) {
490 /*
491 * Only one fragment on the socket.
492 */
493 skb->csum_start = skb_transport_header(skb) - skb->head;
494 skb->csum_offset = offsetof(struct udphdr, check);
495 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
496 } else {
497 /*
498 * HW-checksum won't work as there are two or more
499 * fragments on the socket so that all csums of sk_buffs
500 * should be together
501 */
502 offset = skb_transport_offset(skb);
503 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
504
505 skb->ip_summed = CHECKSUM_NONE;
506
507 skb_queue_walk(&sk->sk_write_queue, skb) {
508 csum = csum_add(csum, skb->csum);
509 }
510
511 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
512 if (uh->check == 0)
513 uh->check = CSUM_MANGLED_0;
514 }
515 }
516
517 /*
518 * Push out all pending data as one UDP datagram. Socket is locked.
519 */
520 static int udp_push_pending_frames(struct sock *sk)
521 {
522 struct udp_sock *up = udp_sk(sk);
523 struct inet_sock *inet = inet_sk(sk);
524 struct flowi *fl = &inet->cork.fl;
525 struct sk_buff *skb;
526 struct udphdr *uh;
527 int err = 0;
528 int is_udplite = IS_UDPLITE(sk);
529 __wsum csum = 0;
530
531 /* Grab the skbuff where UDP header space exists. */
532 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
533 goto out;
534
535 /*
536 * Create a UDP header
537 */
538 uh = udp_hdr(skb);
539 uh->source = fl->fl_ip_sport;
540 uh->dest = fl->fl_ip_dport;
541 uh->len = htons(up->len);
542 uh->check = 0;
543
544 if (is_udplite) /* UDP-Lite */
545 csum = udplite_csum_outgoing(sk, skb);
546
547 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
548
549 skb->ip_summed = CHECKSUM_NONE;
550 goto send;
551
552 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
553
554 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src, fl->fl4_dst, up->len);
555 goto send;
556
557 } else /* `normal' UDP */
558 csum = udp_csum_outgoing(sk, skb);
559
560 /* add protocol-dependent pseudo-header */
561 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
562 sk->sk_protocol, csum);
563 if (uh->check == 0)
564 uh->check = CSUM_MANGLED_0;
565
566 send:
567 err = ip_push_pending_frames(sk);
568 if (err) {
569 if (err == -ENOBUFS && !inet->recverr) {
570 UDP_INC_STATS_USER(sock_net(sk),
571 UDP_MIB_SNDBUFERRORS, is_udplite);
572 err = 0;
573 }
574 } else
575 UDP_INC_STATS_USER(sock_net(sk),
576 UDP_MIB_OUTDATAGRAMS, is_udplite);
577 out:
578 up->len = 0;
579 up->pending = 0;
580 return err;
581 }
582
583 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
584 size_t len)
585 {
586 struct inet_sock *inet = inet_sk(sk);
587 struct udp_sock *up = udp_sk(sk);
588 int ulen = len;
589 struct ipcm_cookie ipc;
590 struct rtable *rt = NULL;
591 int free = 0;
592 int connected = 0;
593 __be32 daddr, faddr, saddr;
594 __be16 dport;
595 u8 tos;
596 int err, is_udplite = IS_UDPLITE(sk);
597 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
598 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
599
600 if (len > 0xFFFF)
601 return -EMSGSIZE;
602
603 /*
604 * Check the flags.
605 */
606
607 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
608 return -EOPNOTSUPP;
609
610 ipc.opt = NULL;
611 ipc.shtx.flags = 0;
612
613 if (up->pending) {
614 /*
615 * There are pending frames.
616 * The socket lock must be held while it's corked.
617 */
618 lock_sock(sk);
619 if (likely(up->pending)) {
620 if (unlikely(up->pending != AF_INET)) {
621 release_sock(sk);
622 return -EINVAL;
623 }
624 goto do_append_data;
625 }
626 release_sock(sk);
627 }
628 ulen += sizeof(struct udphdr);
629
630 /*
631 * Get and verify the address.
632 */
633 if (msg->msg_name) {
634 struct sockaddr_in * usin = (struct sockaddr_in *)msg->msg_name;
635 if (msg->msg_namelen < sizeof(*usin))
636 return -EINVAL;
637 if (usin->sin_family != AF_INET) {
638 if (usin->sin_family != AF_UNSPEC)
639 return -EAFNOSUPPORT;
640 }
641
642 daddr = usin->sin_addr.s_addr;
643 dport = usin->sin_port;
644 if (dport == 0)
645 return -EINVAL;
646 } else {
647 if (sk->sk_state != TCP_ESTABLISHED)
648 return -EDESTADDRREQ;
649 daddr = inet->daddr;
650 dport = inet->dport;
651 /* Open fast path for connected socket.
652 Route will not be used, if at least one option is set.
653 */
654 connected = 1;
655 }
656 ipc.addr = inet->saddr;
657
658 ipc.oif = sk->sk_bound_dev_if;
659 err = sock_tx_timestamp(msg, sk, &ipc.shtx);
660 if (err)
661 return err;
662 if (msg->msg_controllen) {
663 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
664 if (err)
665 return err;
666 if (ipc.opt)
667 free = 1;
668 connected = 0;
669 }
670 if (!ipc.opt)
671 ipc.opt = inet->opt;
672
673 saddr = ipc.addr;
674 ipc.addr = faddr = daddr;
675
676 if (ipc.opt && ipc.opt->srr) {
677 if (!daddr)
678 return -EINVAL;
679 faddr = ipc.opt->faddr;
680 connected = 0;
681 }
682 tos = RT_TOS(inet->tos);
683 if (sock_flag(sk, SOCK_LOCALROUTE) ||
684 (msg->msg_flags & MSG_DONTROUTE) ||
685 (ipc.opt && ipc.opt->is_strictroute)) {
686 tos |= RTO_ONLINK;
687 connected = 0;
688 }
689
690 if (ipv4_is_multicast(daddr)) {
691 if (!ipc.oif)
692 ipc.oif = inet->mc_index;
693 if (!saddr)
694 saddr = inet->mc_addr;
695 connected = 0;
696 }
697
698 if (connected)
699 rt = (struct rtable *)sk_dst_check(sk, 0);
700
701 if (rt == NULL) {
702 struct flowi fl = { .oif = ipc.oif,
703 .mark = sk->sk_mark,
704 .nl_u = { .ip4_u =
705 { .daddr = faddr,
706 .saddr = saddr,
707 .tos = tos } },
708 .proto = sk->sk_protocol,
709 .flags = inet_sk_flowi_flags(sk),
710 .uli_u = { .ports =
711 { .sport = inet->sport,
712 .dport = dport } } };
713 struct net *net = sock_net(sk);
714
715 security_sk_classify_flow(sk, &fl);
716 err = ip_route_output_flow(net, &rt, &fl, sk, 1);
717 if (err) {
718 if (err == -ENETUNREACH)
719 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
720 goto out;
721 }
722
723 err = -EACCES;
724 if ((rt->rt_flags & RTCF_BROADCAST) &&
725 !sock_flag(sk, SOCK_BROADCAST))
726 goto out;
727 if (connected)
728 sk_dst_set(sk, dst_clone(&rt->u.dst));
729 }
730
731 if (msg->msg_flags&MSG_CONFIRM)
732 goto do_confirm;
733 back_from_confirm:
734
735 saddr = rt->rt_src;
736 if (!ipc.addr)
737 daddr = ipc.addr = rt->rt_dst;
738
739 lock_sock(sk);
740 if (unlikely(up->pending)) {
741 /* The socket is already corked while preparing it. */
742 /* ... which is an evident application bug. --ANK */
743 release_sock(sk);
744
745 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
746 err = -EINVAL;
747 goto out;
748 }
749 /*
750 * Now cork the socket to pend data.
751 */
752 inet->cork.fl.fl4_dst = daddr;
753 inet->cork.fl.fl_ip_dport = dport;
754 inet->cork.fl.fl4_src = saddr;
755 inet->cork.fl.fl_ip_sport = inet->sport;
756 up->pending = AF_INET;
757
758 do_append_data:
759 up->len += ulen;
760 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
761 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
762 sizeof(struct udphdr), &ipc, &rt,
763 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
764 if (err)
765 udp_flush_pending_frames(sk);
766 else if (!corkreq)
767 err = udp_push_pending_frames(sk);
768 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
769 up->pending = 0;
770 release_sock(sk);
771
772 out:
773 ip_rt_put(rt);
774 if (free)
775 kfree(ipc.opt);
776 if (!err)
777 return len;
778 /*
779 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
780 * ENOBUFS might not be good (it's not tunable per se), but otherwise
781 * we don't have a good statistic (IpOutDiscards but it can be too many
782 * things). We could add another new stat but at least for now that
783 * seems like overkill.
784 */
785 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
786 UDP_INC_STATS_USER(sock_net(sk),
787 UDP_MIB_SNDBUFERRORS, is_udplite);
788 }
789 return err;
790
791 do_confirm:
792 dst_confirm(&rt->u.dst);
793 if (!(msg->msg_flags&MSG_PROBE) || len)
794 goto back_from_confirm;
795 err = 0;
796 goto out;
797 }
798 EXPORT_SYMBOL(udp_sendmsg);
799
800 int udp_sendpage(struct sock *sk, struct page *page, int offset,
801 size_t size, int flags)
802 {
803 struct udp_sock *up = udp_sk(sk);
804 int ret;
805
806 if (!up->pending) {
807 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
808
809 /* Call udp_sendmsg to specify destination address which
810 * sendpage interface can't pass.
811 * This will succeed only when the socket is connected.
812 */
813 ret = udp_sendmsg(NULL, sk, &msg, 0);
814 if (ret < 0)
815 return ret;
816 }
817
818 lock_sock(sk);
819
820 if (unlikely(!up->pending)) {
821 release_sock(sk);
822
823 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
824 return -EINVAL;
825 }
826
827 ret = ip_append_page(sk, page, offset, size, flags);
828 if (ret == -EOPNOTSUPP) {
829 release_sock(sk);
830 return sock_no_sendpage(sk->sk_socket, page, offset,
831 size, flags);
832 }
833 if (ret < 0) {
834 udp_flush_pending_frames(sk);
835 goto out;
836 }
837
838 up->len += size;
839 if (!(up->corkflag || (flags&MSG_MORE)))
840 ret = udp_push_pending_frames(sk);
841 if (!ret)
842 ret = size;
843 out:
844 release_sock(sk);
845 return ret;
846 }
847
848
849 /**
850 * first_packet_length - return length of first packet in receive queue
851 * @sk: socket
852 *
853 * Drops all bad checksum frames, until a valid one is found.
854 * Returns the length of found skb, or 0 if none is found.
855 */
856 static unsigned int first_packet_length(struct sock *sk)
857 {
858 struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
859 struct sk_buff *skb;
860 unsigned int res;
861
862 __skb_queue_head_init(&list_kill);
863
864 spin_lock_bh(&rcvq->lock);
865 while ((skb = skb_peek(rcvq)) != NULL &&
866 udp_lib_checksum_complete(skb)) {
867 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
868 IS_UDPLITE(sk));
869 __skb_unlink(skb, rcvq);
870 __skb_queue_tail(&list_kill, skb);
871 }
872 res = skb ? skb->len : 0;
873 spin_unlock_bh(&rcvq->lock);
874
875 if (!skb_queue_empty(&list_kill)) {
876 lock_sock(sk);
877 __skb_queue_purge(&list_kill);
878 sk_mem_reclaim_partial(sk);
879 release_sock(sk);
880 }
881 return res;
882 }
883
884 /*
885 * IOCTL requests applicable to the UDP protocol
886 */
887
888 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
889 {
890 switch (cmd) {
891 case SIOCOUTQ:
892 {
893 int amount = sk_wmem_alloc_get(sk);
894
895 return put_user(amount, (int __user *)arg);
896 }
897
898 case SIOCINQ:
899 {
900 unsigned int amount = first_packet_length(sk);
901
902 if (amount)
903 /*
904 * We will only return the amount
905 * of this packet since that is all
906 * that will be read.
907 */
908 amount -= sizeof(struct udphdr);
909
910 return put_user(amount, (int __user *)arg);
911 }
912
913 default:
914 return -ENOIOCTLCMD;
915 }
916
917 return 0;
918 }
919 EXPORT_SYMBOL(udp_ioctl);
920
921 /*
922 * This should be easy, if there is something there we
923 * return it, otherwise we block.
924 */
925
926 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
927 size_t len, int noblock, int flags, int *addr_len)
928 {
929 struct inet_sock *inet = inet_sk(sk);
930 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
931 struct sk_buff *skb;
932 unsigned int ulen, copied;
933 int peeked;
934 int err;
935 int is_udplite = IS_UDPLITE(sk);
936
937 /*
938 * Check any passed addresses
939 */
940 if (addr_len)
941 *addr_len = sizeof(*sin);
942
943 if (flags & MSG_ERRQUEUE)
944 return ip_recv_error(sk, msg, len);
945
946 try_again:
947 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
948 &peeked, &err);
949 if (!skb)
950 goto out;
951
952 ulen = skb->len - sizeof(struct udphdr);
953 copied = len;
954 if (copied > ulen)
955 copied = ulen;
956 else if (copied < ulen)
957 msg->msg_flags |= MSG_TRUNC;
958
959 /*
960 * If checksum is needed at all, try to do it while copying the
961 * data. If the data is truncated, or if we only want a partial
962 * coverage checksum (UDP-Lite), do it before the copy.
963 */
964
965 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
966 if (udp_lib_checksum_complete(skb))
967 goto csum_copy_err;
968 }
969
970 if (skb_csum_unnecessary(skb))
971 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
972 msg->msg_iov, copied);
973 else {
974 err = skb_copy_and_csum_datagram_iovec(skb,
975 sizeof(struct udphdr),
976 msg->msg_iov);
977
978 if (err == -EINVAL)
979 goto csum_copy_err;
980 }
981
982 if (err)
983 goto out_free;
984
985 if (!peeked)
986 UDP_INC_STATS_USER(sock_net(sk),
987 UDP_MIB_INDATAGRAMS, is_udplite);
988
989 sock_recv_ts_and_drops(msg, sk, skb);
990
991 /* Copy the address. */
992 if (sin) {
993 sin->sin_family = AF_INET;
994 sin->sin_port = udp_hdr(skb)->source;
995 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
996 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
997 }
998 if (inet->cmsg_flags)
999 ip_cmsg_recv(msg, skb);
1000
1001 err = copied;
1002 if (flags & MSG_TRUNC)
1003 err = ulen;
1004
1005 out_free:
1006 lock_sock(sk);
1007 skb_free_datagram(sk, skb);
1008 release_sock(sk);
1009 out:
1010 return err;
1011
1012 csum_copy_err:
1013 lock_sock(sk);
1014 if (!skb_kill_datagram(sk, skb, flags))
1015 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1016 release_sock(sk);
1017
1018 if (noblock)
1019 return -EAGAIN;
1020 goto try_again;
1021 }
1022
1023
1024 int udp_disconnect(struct sock *sk, int flags)
1025 {
1026 struct inet_sock *inet = inet_sk(sk);
1027 /*
1028 * 1003.1g - break association.
1029 */
1030
1031 sk->sk_state = TCP_CLOSE;
1032 inet->daddr = 0;
1033 inet->dport = 0;
1034 sk->sk_bound_dev_if = 0;
1035 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1036 inet_reset_saddr(sk);
1037
1038 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1039 sk->sk_prot->unhash(sk);
1040 inet->sport = 0;
1041 }
1042 sk_dst_reset(sk);
1043 return 0;
1044 }
1045 EXPORT_SYMBOL(udp_disconnect);
1046
1047 void udp_lib_unhash(struct sock *sk)
1048 {
1049 if (sk_hashed(sk)) {
1050 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1051 struct udp_hslot *hslot = udp_hashslot(udptable, sock_net(sk),
1052 sk->sk_hash);
1053
1054 spin_lock_bh(&hslot->lock);
1055 if (sk_nulls_del_node_init_rcu(sk)) {
1056 inet_sk(sk)->num = 0;
1057 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1058 }
1059 spin_unlock_bh(&hslot->lock);
1060 }
1061 }
1062 EXPORT_SYMBOL(udp_lib_unhash);
1063
1064 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1065 {
1066 int rc = sock_queue_rcv_skb(sk, skb);
1067
1068 if (rc < 0) {
1069 int is_udplite = IS_UDPLITE(sk);
1070
1071 /* Note that an ENOMEM error is charged twice */
1072 if (rc == -ENOMEM)
1073 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1074 is_udplite);
1075 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1076 kfree_skb(skb);
1077 return -1;
1078 }
1079
1080 return 0;
1081
1082 }
1083
1084 /* returns:
1085 * -1: error
1086 * 0: success
1087 * >0: "udp encap" protocol resubmission
1088 *
1089 * Note that in the success and error cases, the skb is assumed to
1090 * have either been requeued or freed.
1091 */
1092 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1093 {
1094 struct udp_sock *up = udp_sk(sk);
1095 int rc;
1096 int is_udplite = IS_UDPLITE(sk);
1097
1098 /*
1099 * Charge it to the socket, dropping if the queue is full.
1100 */
1101 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1102 goto drop;
1103 nf_reset(skb);
1104
1105 if (up->encap_type) {
1106 /*
1107 * This is an encapsulation socket so pass the skb to
1108 * the socket's udp_encap_rcv() hook. Otherwise, just
1109 * fall through and pass this up the UDP socket.
1110 * up->encap_rcv() returns the following value:
1111 * =0 if skb was successfully passed to the encap
1112 * handler or was discarded by it.
1113 * >0 if skb should be passed on to UDP.
1114 * <0 if skb should be resubmitted as proto -N
1115 */
1116
1117 /* if we're overly short, let UDP handle it */
1118 if (skb->len > sizeof(struct udphdr) &&
1119 up->encap_rcv != NULL) {
1120 int ret;
1121
1122 ret = (*up->encap_rcv)(sk, skb);
1123 if (ret <= 0) {
1124 UDP_INC_STATS_BH(sock_net(sk),
1125 UDP_MIB_INDATAGRAMS,
1126 is_udplite);
1127 return -ret;
1128 }
1129 }
1130
1131 /* FALLTHROUGH -- it's a UDP Packet */
1132 }
1133
1134 /*
1135 * UDP-Lite specific tests, ignored on UDP sockets
1136 */
1137 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1138
1139 /*
1140 * MIB statistics other than incrementing the error count are
1141 * disabled for the following two types of errors: these depend
1142 * on the application settings, not on the functioning of the
1143 * protocol stack as such.
1144 *
1145 * RFC 3828 here recommends (sec 3.3): "There should also be a
1146 * way ... to ... at least let the receiving application block
1147 * delivery of packets with coverage values less than a value
1148 * provided by the application."
1149 */
1150 if (up->pcrlen == 0) { /* full coverage was set */
1151 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1152 "%d while full coverage %d requested\n",
1153 UDP_SKB_CB(skb)->cscov, skb->len);
1154 goto drop;
1155 }
1156 /* The next case involves violating the min. coverage requested
1157 * by the receiver. This is subtle: if receiver wants x and x is
1158 * greater than the buffersize/MTU then receiver will complain
1159 * that it wants x while sender emits packets of smaller size y.
1160 * Therefore the above ...()->partial_cov statement is essential.
1161 */
1162 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1163 LIMIT_NETDEBUG(KERN_WARNING
1164 "UDPLITE: coverage %d too small, need min %d\n",
1165 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1166 goto drop;
1167 }
1168 }
1169
1170 if (sk->sk_filter) {
1171 if (udp_lib_checksum_complete(skb))
1172 goto drop;
1173 }
1174
1175 rc = 0;
1176
1177 bh_lock_sock(sk);
1178 if (!sock_owned_by_user(sk))
1179 rc = __udp_queue_rcv_skb(sk, skb);
1180 else
1181 sk_add_backlog(sk, skb);
1182 bh_unlock_sock(sk);
1183
1184 return rc;
1185
1186 drop:
1187 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1188 kfree_skb(skb);
1189 return -1;
1190 }
1191
1192 /*
1193 * Multicasts and broadcasts go to each listener.
1194 *
1195 * Note: called only from the BH handler context,
1196 * so we don't need to lock the hashes.
1197 */
1198 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1199 struct udphdr *uh,
1200 __be32 saddr, __be32 daddr,
1201 struct udp_table *udptable)
1202 {
1203 struct sock *sk;
1204 struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1205 int dif;
1206
1207 spin_lock(&hslot->lock);
1208 sk = sk_nulls_head(&hslot->head);
1209 dif = skb->dev->ifindex;
1210 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1211 if (sk) {
1212 struct sock *sknext = NULL;
1213
1214 do {
1215 struct sk_buff *skb1 = skb;
1216
1217 sknext = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1218 daddr, uh->source, saddr,
1219 dif);
1220 if (sknext)
1221 skb1 = skb_clone(skb, GFP_ATOMIC);
1222
1223 if (skb1) {
1224 int ret = udp_queue_rcv_skb(sk, skb1);
1225 if (ret > 0)
1226 /* we should probably re-process instead
1227 * of dropping packets here. */
1228 kfree_skb(skb1);
1229 }
1230 sk = sknext;
1231 } while (sknext);
1232 } else
1233 consume_skb(skb);
1234 spin_unlock(&hslot->lock);
1235 return 0;
1236 }
1237
1238 /* Initialize UDP checksum. If exited with zero value (success),
1239 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1240 * Otherwise, csum completion requires chacksumming packet body,
1241 * including udp header and folding it to skb->csum.
1242 */
1243 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1244 int proto)
1245 {
1246 const struct iphdr *iph;
1247 int err;
1248
1249 UDP_SKB_CB(skb)->partial_cov = 0;
1250 UDP_SKB_CB(skb)->cscov = skb->len;
1251
1252 if (proto == IPPROTO_UDPLITE) {
1253 err = udplite_checksum_init(skb, uh);
1254 if (err)
1255 return err;
1256 }
1257
1258 iph = ip_hdr(skb);
1259 if (uh->check == 0) {
1260 skb->ip_summed = CHECKSUM_UNNECESSARY;
1261 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1262 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1263 proto, skb->csum))
1264 skb->ip_summed = CHECKSUM_UNNECESSARY;
1265 }
1266 if (!skb_csum_unnecessary(skb))
1267 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1268 skb->len, proto, 0);
1269 /* Probably, we should checksum udp header (it should be in cache
1270 * in any case) and data in tiny packets (< rx copybreak).
1271 */
1272
1273 return 0;
1274 }
1275
1276 /*
1277 * All we need to do is get the socket, and then do a checksum.
1278 */
1279
1280 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1281 int proto)
1282 {
1283 struct sock *sk;
1284 struct udphdr *uh;
1285 unsigned short ulen;
1286 struct rtable *rt = skb_rtable(skb);
1287 __be32 saddr, daddr;
1288 struct net *net = dev_net(skb->dev);
1289
1290 /*
1291 * Validate the packet.
1292 */
1293 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1294 goto drop; /* No space for header. */
1295
1296 uh = udp_hdr(skb);
1297 ulen = ntohs(uh->len);
1298 if (ulen > skb->len)
1299 goto short_packet;
1300
1301 if (proto == IPPROTO_UDP) {
1302 /* UDP validates ulen. */
1303 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1304 goto short_packet;
1305 uh = udp_hdr(skb);
1306 }
1307
1308 if (udp4_csum_init(skb, uh, proto))
1309 goto csum_error;
1310
1311 saddr = ip_hdr(skb)->saddr;
1312 daddr = ip_hdr(skb)->daddr;
1313
1314 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1315 return __udp4_lib_mcast_deliver(net, skb, uh,
1316 saddr, daddr, udptable);
1317
1318 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1319
1320 if (sk != NULL) {
1321 int ret = udp_queue_rcv_skb(sk, skb);
1322 sock_put(sk);
1323
1324 /* a return value > 0 means to resubmit the input, but
1325 * it wants the return to be -protocol, or 0
1326 */
1327 if (ret > 0)
1328 return -ret;
1329 return 0;
1330 }
1331
1332 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1333 goto drop;
1334 nf_reset(skb);
1335
1336 /* No socket. Drop packet silently, if checksum is wrong */
1337 if (udp_lib_checksum_complete(skb))
1338 goto csum_error;
1339
1340 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1341 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1342
1343 /*
1344 * Hmm. We got an UDP packet to a port to which we
1345 * don't wanna listen. Ignore it.
1346 */
1347 kfree_skb(skb);
1348 return 0;
1349
1350 short_packet:
1351 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1352 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1353 &saddr,
1354 ntohs(uh->source),
1355 ulen,
1356 skb->len,
1357 &daddr,
1358 ntohs(uh->dest));
1359 goto drop;
1360
1361 csum_error:
1362 /*
1363 * RFC1122: OK. Discards the bad packet silently (as far as
1364 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1365 */
1366 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1367 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1368 &saddr,
1369 ntohs(uh->source),
1370 &daddr,
1371 ntohs(uh->dest),
1372 ulen);
1373 drop:
1374 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1375 kfree_skb(skb);
1376 return 0;
1377 }
1378
1379 int udp_rcv(struct sk_buff *skb)
1380 {
1381 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1382 }
1383
1384 void udp_destroy_sock(struct sock *sk)
1385 {
1386 lock_sock(sk);
1387 udp_flush_pending_frames(sk);
1388 release_sock(sk);
1389 }
1390
1391 /*
1392 * Socket option code for UDP
1393 */
1394 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1395 char __user *optval, unsigned int optlen,
1396 int (*push_pending_frames)(struct sock *))
1397 {
1398 struct udp_sock *up = udp_sk(sk);
1399 int val;
1400 int err = 0;
1401 int is_udplite = IS_UDPLITE(sk);
1402
1403 if (optlen < sizeof(int))
1404 return -EINVAL;
1405
1406 if (get_user(val, (int __user *)optval))
1407 return -EFAULT;
1408
1409 switch (optname) {
1410 case UDP_CORK:
1411 if (val != 0) {
1412 up->corkflag = 1;
1413 } else {
1414 up->corkflag = 0;
1415 lock_sock(sk);
1416 (*push_pending_frames)(sk);
1417 release_sock(sk);
1418 }
1419 break;
1420
1421 case UDP_ENCAP:
1422 switch (val) {
1423 case 0:
1424 case UDP_ENCAP_ESPINUDP:
1425 case UDP_ENCAP_ESPINUDP_NON_IKE:
1426 up->encap_rcv = xfrm4_udp_encap_rcv;
1427 /* FALLTHROUGH */
1428 case UDP_ENCAP_L2TPINUDP:
1429 up->encap_type = val;
1430 break;
1431 default:
1432 err = -ENOPROTOOPT;
1433 break;
1434 }
1435 break;
1436
1437 /*
1438 * UDP-Lite's partial checksum coverage (RFC 3828).
1439 */
1440 /* The sender sets actual checksum coverage length via this option.
1441 * The case coverage > packet length is handled by send module. */
1442 case UDPLITE_SEND_CSCOV:
1443 if (!is_udplite) /* Disable the option on UDP sockets */
1444 return -ENOPROTOOPT;
1445 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1446 val = 8;
1447 else if (val > USHORT_MAX)
1448 val = USHORT_MAX;
1449 up->pcslen = val;
1450 up->pcflag |= UDPLITE_SEND_CC;
1451 break;
1452
1453 /* The receiver specifies a minimum checksum coverage value. To make
1454 * sense, this should be set to at least 8 (as done below). If zero is
1455 * used, this again means full checksum coverage. */
1456 case UDPLITE_RECV_CSCOV:
1457 if (!is_udplite) /* Disable the option on UDP sockets */
1458 return -ENOPROTOOPT;
1459 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1460 val = 8;
1461 else if (val > USHORT_MAX)
1462 val = USHORT_MAX;
1463 up->pcrlen = val;
1464 up->pcflag |= UDPLITE_RECV_CC;
1465 break;
1466
1467 default:
1468 err = -ENOPROTOOPT;
1469 break;
1470 }
1471
1472 return err;
1473 }
1474 EXPORT_SYMBOL(udp_lib_setsockopt);
1475
1476 int udp_setsockopt(struct sock *sk, int level, int optname,
1477 char __user *optval, unsigned int optlen)
1478 {
1479 if (level == SOL_UDP || level == SOL_UDPLITE)
1480 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1481 udp_push_pending_frames);
1482 return ip_setsockopt(sk, level, optname, optval, optlen);
1483 }
1484
1485 #ifdef CONFIG_COMPAT
1486 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1487 char __user *optval, unsigned int optlen)
1488 {
1489 if (level == SOL_UDP || level == SOL_UDPLITE)
1490 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1491 udp_push_pending_frames);
1492 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1493 }
1494 #endif
1495
1496 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1497 char __user *optval, int __user *optlen)
1498 {
1499 struct udp_sock *up = udp_sk(sk);
1500 int val, len;
1501
1502 if (get_user(len, optlen))
1503 return -EFAULT;
1504
1505 len = min_t(unsigned int, len, sizeof(int));
1506
1507 if (len < 0)
1508 return -EINVAL;
1509
1510 switch (optname) {
1511 case UDP_CORK:
1512 val = up->corkflag;
1513 break;
1514
1515 case UDP_ENCAP:
1516 val = up->encap_type;
1517 break;
1518
1519 /* The following two cannot be changed on UDP sockets, the return is
1520 * always 0 (which corresponds to the full checksum coverage of UDP). */
1521 case UDPLITE_SEND_CSCOV:
1522 val = up->pcslen;
1523 break;
1524
1525 case UDPLITE_RECV_CSCOV:
1526 val = up->pcrlen;
1527 break;
1528
1529 default:
1530 return -ENOPROTOOPT;
1531 }
1532
1533 if (put_user(len, optlen))
1534 return -EFAULT;
1535 if (copy_to_user(optval, &val, len))
1536 return -EFAULT;
1537 return 0;
1538 }
1539 EXPORT_SYMBOL(udp_lib_getsockopt);
1540
1541 int udp_getsockopt(struct sock *sk, int level, int optname,
1542 char __user *optval, int __user *optlen)
1543 {
1544 if (level == SOL_UDP || level == SOL_UDPLITE)
1545 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1546 return ip_getsockopt(sk, level, optname, optval, optlen);
1547 }
1548
1549 #ifdef CONFIG_COMPAT
1550 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1551 char __user *optval, int __user *optlen)
1552 {
1553 if (level == SOL_UDP || level == SOL_UDPLITE)
1554 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1555 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1556 }
1557 #endif
1558 /**
1559 * udp_poll - wait for a UDP event.
1560 * @file - file struct
1561 * @sock - socket
1562 * @wait - poll table
1563 *
1564 * This is same as datagram poll, except for the special case of
1565 * blocking sockets. If application is using a blocking fd
1566 * and a packet with checksum error is in the queue;
1567 * then it could get return from select indicating data available
1568 * but then block when reading it. Add special case code
1569 * to work around these arguably broken applications.
1570 */
1571 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1572 {
1573 unsigned int mask = datagram_poll(file, sock, wait);
1574 struct sock *sk = sock->sk;
1575
1576 /* Check for false positives due to checksum errors */
1577 if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1578 !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1579 mask &= ~(POLLIN | POLLRDNORM);
1580
1581 return mask;
1582
1583 }
1584 EXPORT_SYMBOL(udp_poll);
1585
1586 struct proto udp_prot = {
1587 .name = "UDP",
1588 .owner = THIS_MODULE,
1589 .close = udp_lib_close,
1590 .connect = ip4_datagram_connect,
1591 .disconnect = udp_disconnect,
1592 .ioctl = udp_ioctl,
1593 .destroy = udp_destroy_sock,
1594 .setsockopt = udp_setsockopt,
1595 .getsockopt = udp_getsockopt,
1596 .sendmsg = udp_sendmsg,
1597 .recvmsg = udp_recvmsg,
1598 .sendpage = udp_sendpage,
1599 .backlog_rcv = __udp_queue_rcv_skb,
1600 .hash = udp_lib_hash,
1601 .unhash = udp_lib_unhash,
1602 .get_port = udp_v4_get_port,
1603 .memory_allocated = &udp_memory_allocated,
1604 .sysctl_mem = sysctl_udp_mem,
1605 .sysctl_wmem = &sysctl_udp_wmem_min,
1606 .sysctl_rmem = &sysctl_udp_rmem_min,
1607 .obj_size = sizeof(struct udp_sock),
1608 .slab_flags = SLAB_DESTROY_BY_RCU,
1609 .h.udp_table = &udp_table,
1610 #ifdef CONFIG_COMPAT
1611 .compat_setsockopt = compat_udp_setsockopt,
1612 .compat_getsockopt = compat_udp_getsockopt,
1613 #endif
1614 };
1615 EXPORT_SYMBOL(udp_prot);
1616
1617 /* ------------------------------------------------------------------------ */
1618 #ifdef CONFIG_PROC_FS
1619
1620 static struct sock *udp_get_first(struct seq_file *seq, int start)
1621 {
1622 struct sock *sk;
1623 struct udp_iter_state *state = seq->private;
1624 struct net *net = seq_file_net(seq);
1625
1626 for (state->bucket = start; state->bucket <= state->udp_table->mask;
1627 ++state->bucket) {
1628 struct hlist_nulls_node *node;
1629 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
1630
1631 if (hlist_nulls_empty(&hslot->head))
1632 continue;
1633
1634 spin_lock_bh(&hslot->lock);
1635 sk_nulls_for_each(sk, node, &hslot->head) {
1636 if (!net_eq(sock_net(sk), net))
1637 continue;
1638 if (sk->sk_family == state->family)
1639 goto found;
1640 }
1641 spin_unlock_bh(&hslot->lock);
1642 }
1643 sk = NULL;
1644 found:
1645 return sk;
1646 }
1647
1648 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1649 {
1650 struct udp_iter_state *state = seq->private;
1651 struct net *net = seq_file_net(seq);
1652
1653 do {
1654 sk = sk_nulls_next(sk);
1655 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1656
1657 if (!sk) {
1658 if (state->bucket <= state->udp_table->mask)
1659 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
1660 return udp_get_first(seq, state->bucket + 1);
1661 }
1662 return sk;
1663 }
1664
1665 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1666 {
1667 struct sock *sk = udp_get_first(seq, 0);
1668
1669 if (sk)
1670 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1671 --pos;
1672 return pos ? NULL : sk;
1673 }
1674
1675 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1676 {
1677 struct udp_iter_state *state = seq->private;
1678 state->bucket = MAX_UDP_PORTS;
1679
1680 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
1681 }
1682
1683 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1684 {
1685 struct sock *sk;
1686
1687 if (v == SEQ_START_TOKEN)
1688 sk = udp_get_idx(seq, 0);
1689 else
1690 sk = udp_get_next(seq, v);
1691
1692 ++*pos;
1693 return sk;
1694 }
1695
1696 static void udp_seq_stop(struct seq_file *seq, void *v)
1697 {
1698 struct udp_iter_state *state = seq->private;
1699
1700 if (state->bucket <= state->udp_table->mask)
1701 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
1702 }
1703
1704 static int udp_seq_open(struct inode *inode, struct file *file)
1705 {
1706 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1707 struct udp_iter_state *s;
1708 int err;
1709
1710 err = seq_open_net(inode, file, &afinfo->seq_ops,
1711 sizeof(struct udp_iter_state));
1712 if (err < 0)
1713 return err;
1714
1715 s = ((struct seq_file *)file->private_data)->private;
1716 s->family = afinfo->family;
1717 s->udp_table = afinfo->udp_table;
1718 return err;
1719 }
1720
1721 /* ------------------------------------------------------------------------ */
1722 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
1723 {
1724 struct proc_dir_entry *p;
1725 int rc = 0;
1726
1727 afinfo->seq_fops.open = udp_seq_open;
1728 afinfo->seq_fops.read = seq_read;
1729 afinfo->seq_fops.llseek = seq_lseek;
1730 afinfo->seq_fops.release = seq_release_net;
1731
1732 afinfo->seq_ops.start = udp_seq_start;
1733 afinfo->seq_ops.next = udp_seq_next;
1734 afinfo->seq_ops.stop = udp_seq_stop;
1735
1736 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
1737 &afinfo->seq_fops, afinfo);
1738 if (!p)
1739 rc = -ENOMEM;
1740 return rc;
1741 }
1742 EXPORT_SYMBOL(udp_proc_register);
1743
1744 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
1745 {
1746 proc_net_remove(net, afinfo->name);
1747 }
1748 EXPORT_SYMBOL(udp_proc_unregister);
1749
1750 /* ------------------------------------------------------------------------ */
1751 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
1752 int bucket, int *len)
1753 {
1754 struct inet_sock *inet = inet_sk(sp);
1755 __be32 dest = inet->daddr;
1756 __be32 src = inet->rcv_saddr;
1757 __u16 destp = ntohs(inet->dport);
1758 __u16 srcp = ntohs(inet->sport);
1759
1760 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
1761 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
1762 bucket, src, srcp, dest, destp, sp->sk_state,
1763 sk_wmem_alloc_get(sp),
1764 sk_rmem_alloc_get(sp),
1765 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1766 atomic_read(&sp->sk_refcnt), sp,
1767 atomic_read(&sp->sk_drops), len);
1768 }
1769
1770 int udp4_seq_show(struct seq_file *seq, void *v)
1771 {
1772 if (v == SEQ_START_TOKEN)
1773 seq_printf(seq, "%-127s\n",
1774 " sl local_address rem_address st tx_queue "
1775 "rx_queue tr tm->when retrnsmt uid timeout "
1776 "inode ref pointer drops");
1777 else {
1778 struct udp_iter_state *state = seq->private;
1779 int len;
1780
1781 udp4_format_sock(v, seq, state->bucket, &len);
1782 seq_printf(seq, "%*s\n", 127 - len, "");
1783 }
1784 return 0;
1785 }
1786
1787 /* ------------------------------------------------------------------------ */
1788 static struct udp_seq_afinfo udp4_seq_afinfo = {
1789 .name = "udp",
1790 .family = AF_INET,
1791 .udp_table = &udp_table,
1792 .seq_fops = {
1793 .owner = THIS_MODULE,
1794 },
1795 .seq_ops = {
1796 .show = udp4_seq_show,
1797 },
1798 };
1799
1800 static int udp4_proc_init_net(struct net *net)
1801 {
1802 return udp_proc_register(net, &udp4_seq_afinfo);
1803 }
1804
1805 static void udp4_proc_exit_net(struct net *net)
1806 {
1807 udp_proc_unregister(net, &udp4_seq_afinfo);
1808 }
1809
1810 static struct pernet_operations udp4_net_ops = {
1811 .init = udp4_proc_init_net,
1812 .exit = udp4_proc_exit_net,
1813 };
1814
1815 int __init udp4_proc_init(void)
1816 {
1817 return register_pernet_subsys(&udp4_net_ops);
1818 }
1819
1820 void udp4_proc_exit(void)
1821 {
1822 unregister_pernet_subsys(&udp4_net_ops);
1823 }
1824 #endif /* CONFIG_PROC_FS */
1825
1826 static __initdata unsigned long uhash_entries;
1827 static int __init set_uhash_entries(char *str)
1828 {
1829 if (!str)
1830 return 0;
1831 uhash_entries = simple_strtoul(str, &str, 0);
1832 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
1833 uhash_entries = UDP_HTABLE_SIZE_MIN;
1834 return 1;
1835 }
1836 __setup("uhash_entries=", set_uhash_entries);
1837
1838 void __init udp_table_init(struct udp_table *table, const char *name)
1839 {
1840 unsigned int i;
1841
1842 if (!CONFIG_BASE_SMALL)
1843 table->hash = alloc_large_system_hash(name,
1844 sizeof(struct udp_hslot),
1845 uhash_entries,
1846 21, /* one slot per 2 MB */
1847 0,
1848 &table->log,
1849 &table->mask,
1850 64 * 1024);
1851 /*
1852 * Make sure hash table has the minimum size
1853 */
1854 if (CONFIG_BASE_SMALL || table->mask < UDP_HTABLE_SIZE_MIN - 1) {
1855 table->hash = kmalloc(UDP_HTABLE_SIZE_MIN *
1856 sizeof(struct udp_hslot), GFP_KERNEL);
1857 if (!table->hash)
1858 panic(name);
1859 table->log = ilog2(UDP_HTABLE_SIZE_MIN);
1860 table->mask = UDP_HTABLE_SIZE_MIN - 1;
1861 }
1862 for (i = 0; i <= table->mask; i++) {
1863 INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
1864 spin_lock_init(&table->hash[i].lock);
1865 }
1866 }
1867
1868 void __init udp_init(void)
1869 {
1870 unsigned long nr_pages, limit;
1871
1872 udp_table_init(&udp_table, "UDP");
1873 /* Set the pressure threshold up by the same strategy of TCP. It is a
1874 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
1875 * toward zero with the amount of memory, with a floor of 128 pages.
1876 */
1877 nr_pages = totalram_pages - totalhigh_pages;
1878 limit = min(nr_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
1879 limit = (limit * (nr_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
1880 limit = max(limit, 128UL);
1881 sysctl_udp_mem[0] = limit / 4 * 3;
1882 sysctl_udp_mem[1] = limit;
1883 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
1884
1885 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
1886 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
1887 }
1888
1889 int udp4_ufo_send_check(struct sk_buff *skb)
1890 {
1891 const struct iphdr *iph;
1892 struct udphdr *uh;
1893
1894 if (!pskb_may_pull(skb, sizeof(*uh)))
1895 return -EINVAL;
1896
1897 iph = ip_hdr(skb);
1898 uh = udp_hdr(skb);
1899
1900 uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1901 IPPROTO_UDP, 0);
1902 skb->csum_start = skb_transport_header(skb) - skb->head;
1903 skb->csum_offset = offsetof(struct udphdr, check);
1904 skb->ip_summed = CHECKSUM_PARTIAL;
1905 return 0;
1906 }
1907
1908 struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb, int features)
1909 {
1910 struct sk_buff *segs = ERR_PTR(-EINVAL);
1911 unsigned int mss;
1912 int offset;
1913 __wsum csum;
1914
1915 mss = skb_shinfo(skb)->gso_size;
1916 if (unlikely(skb->len <= mss))
1917 goto out;
1918
1919 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
1920 /* Packet is from an untrusted source, reset gso_segs. */
1921 int type = skb_shinfo(skb)->gso_type;
1922
1923 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
1924 !(type & (SKB_GSO_UDP))))
1925 goto out;
1926
1927 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
1928
1929 segs = NULL;
1930 goto out;
1931 }
1932
1933 /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
1934 * do checksum of UDP packets sent as multiple IP fragments.
1935 */
1936 offset = skb->csum_start - skb_headroom(skb);
1937 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1938 offset += skb->csum_offset;
1939 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1940 skb->ip_summed = CHECKSUM_NONE;
1941
1942 /* Fragment the skb. IP headers of the fragments are updated in
1943 * inet_gso_segment()
1944 */
1945 segs = skb_segment(skb, features);
1946 out:
1947 return segs;
1948 }
1949
Linux kernel - Deliverables
This page took 0.070588 seconds and 6 git commands to generate.