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