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.
6 * The User Datagram Protocol (UDP).
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
11 * Alan Cox, <Alan.Cox@linux.org>
12 * Hirokazu Takahashi, <taka@valinux.co.jp>
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
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
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
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.
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.
80 #include <asm/system.h>
81 #include <asm/uaccess.h>
82 #include <asm/ioctls.h>
83 #include <linux/bootmem.h>
84 #include <linux/types.h>
85 #include <linux/fcntl.h>
86 #include <linux/module.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/igmp.h>
91 #include <linux/errno.h>
92 #include <linux/timer.h>
94 #include <linux/inet.h>
95 #include <linux/netdevice.h>
96 #include <net/tcp_states.h>
97 #include <linux/skbuff.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <net/net_namespace.h>
101 #include <net/icmp.h>
102 #include <net/route.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include "udp_impl.h"
108 * Snmp MIB for the UDP layer
111 DEFINE_SNMP_STAT(struct udp_mib, udp_statistics) __read_mostly;
112 EXPORT_SYMBOL(udp_statistics);
114 DEFINE_SNMP_STAT(struct udp_mib, udp_stats_in6) __read_mostly;
115 EXPORT_SYMBOL(udp_stats_in6);
117 struct hlist_head udp_hash[UDP_HTABLE_SIZE];
118 DEFINE_RWLOCK(udp_hash_lock);
120 int sysctl_udp_mem[3] __read_mostly;
121 int sysctl_udp_rmem_min __read_mostly;
122 int sysctl_udp_wmem_min __read_mostly;
124 EXPORT_SYMBOL(sysctl_udp_mem);
125 EXPORT_SYMBOL(sysctl_udp_rmem_min);
126 EXPORT_SYMBOL(sysctl_udp_wmem_min);
128 atomic_t udp_memory_allocated;
129 EXPORT_SYMBOL(udp_memory_allocated);
131 static inline int __udp_lib_lport_inuse(struct net *net, __u16 num,
132 const struct hlist_head udptable[])
135 struct hlist_node *node;
137 sk_for_each(sk, node, &udptable[udp_hashfn(net, num)])
138 if (net_eq(sock_net(sk), net) && sk->sk_hash == num)
144 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
146 * @sk: socket struct in question
147 * @snum: port number to look up
148 * @saddr_comp: AF-dependent comparison of bound local IP addresses
150 int udp_lib_get_port(struct sock *sk, unsigned short snum,
151 int (*saddr_comp)(const struct sock *sk1,
152 const struct sock *sk2 ) )
154 struct hlist_head *udptable = sk->sk_prot->h.udp_hash;
155 struct hlist_node *node;
156 struct hlist_head *head;
159 struct net *net = sock_net(sk);
161 write_lock_bh(&udp_hash_lock);
164 int i, low, high, remaining;
165 unsigned rover, best, best_size_so_far;
167 inet_get_local_port_range(&low, &high);
168 remaining = (high - low) + 1;
170 best_size_so_far = UINT_MAX;
171 best = rover = net_random() % remaining + low;
173 /* 1st pass: look for empty (or shortest) hash chain */
174 for (i = 0; i < UDP_HTABLE_SIZE; i++) {
177 head = &udptable[udp_hashfn(net, rover)];
178 if (hlist_empty(head))
181 sk_for_each(sk2, node, head) {
182 if (++size >= best_size_so_far)
185 best_size_so_far = size;
188 /* fold back if end of range */
190 rover = low + ((rover - low)
191 & (UDP_HTABLE_SIZE - 1));
196 /* 2nd pass: find hole in shortest hash chain */
198 for (i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++) {
199 if (! __udp_lib_lport_inuse(net, rover, udptable))
201 rover += UDP_HTABLE_SIZE;
203 rover = low + ((rover - low)
204 & (UDP_HTABLE_SIZE - 1));
208 /* All ports in use! */
214 head = &udptable[udp_hashfn(net, snum)];
216 sk_for_each(sk2, node, head)
217 if (sk2->sk_hash == snum &&
219 net_eq(sock_net(sk2), net) &&
220 (!sk2->sk_reuse || !sk->sk_reuse) &&
221 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
222 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
223 (*saddr_comp)(sk, sk2) )
227 inet_sk(sk)->num = snum;
229 if (sk_unhashed(sk)) {
230 head = &udptable[udp_hashfn(net, snum)];
231 sk_add_node(sk, head);
232 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
236 write_unlock_bh(&udp_hash_lock);
240 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
242 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
244 return ( !ipv6_only_sock(sk2) &&
245 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
246 inet1->rcv_saddr == inet2->rcv_saddr ));
249 int udp_v4_get_port(struct sock *sk, unsigned short snum)
251 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal);
254 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
255 * harder than this. -DaveM
257 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
258 __be16 sport, __be32 daddr, __be16 dport,
259 int dif, struct hlist_head udptable[])
261 struct sock *sk, *result = NULL;
262 struct hlist_node *node;
263 unsigned short hnum = ntohs(dport);
266 read_lock(&udp_hash_lock);
267 sk_for_each(sk, node, &udptable[udp_hashfn(net, hnum)]) {
268 struct inet_sock *inet = inet_sk(sk);
270 if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum &&
271 !ipv6_only_sock(sk)) {
272 int score = (sk->sk_family == PF_INET ? 1 : 0);
273 if (inet->rcv_saddr) {
274 if (inet->rcv_saddr != daddr)
279 if (inet->daddr != saddr)
284 if (inet->dport != sport)
288 if (sk->sk_bound_dev_if) {
289 if (sk->sk_bound_dev_if != dif)
296 } else if (score > badness) {
304 read_unlock(&udp_hash_lock);
308 static inline struct sock *udp_v4_mcast_next(struct sock *sk,
309 __be16 loc_port, __be32 loc_addr,
310 __be16 rmt_port, __be32 rmt_addr,
313 struct hlist_node *node;
315 unsigned short hnum = ntohs(loc_port);
317 sk_for_each_from(s, node) {
318 struct inet_sock *inet = inet_sk(s);
320 if (s->sk_hash != hnum ||
321 (inet->daddr && inet->daddr != rmt_addr) ||
322 (inet->dport != rmt_port && inet->dport) ||
323 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
325 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
327 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
337 * This routine is called by the ICMP module when it gets some
338 * sort of error condition. If err < 0 then the socket should
339 * be closed and the error returned to the user. If err > 0
340 * it's just the icmp type << 8 | icmp code.
341 * Header points to the ip header of the error packet. We move
342 * on past this. Then (as it used to claim before adjustment)
343 * header points to the first 8 bytes of the udp header. We need
344 * to find the appropriate port.
347 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[])
349 struct inet_sock *inet;
350 struct iphdr *iph = (struct iphdr*)skb->data;
351 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
352 const int type = icmp_hdr(skb)->type;
353 const int code = icmp_hdr(skb)->code;
357 struct net *net = dev_net(skb->dev);
359 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
360 iph->saddr, uh->source, skb->dev->ifindex, udptable);
362 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
363 return; /* No socket for error */
372 case ICMP_TIME_EXCEEDED:
375 case ICMP_SOURCE_QUENCH:
377 case ICMP_PARAMETERPROB:
381 case ICMP_DEST_UNREACH:
382 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
383 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
391 if (code <= NR_ICMP_UNREACH) {
392 harderr = icmp_err_convert[code].fatal;
393 err = icmp_err_convert[code].errno;
399 * RFC1122: OK. Passes ICMP errors back to application, as per
402 if (!inet->recverr) {
403 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
406 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
409 sk->sk_error_report(sk);
414 void udp_err(struct sk_buff *skb, u32 info)
416 __udp4_lib_err(skb, info, udp_hash);
420 * Throw away all pending data and cancel the corking. Socket is locked.
422 void udp_flush_pending_frames(struct sock *sk)
424 struct udp_sock *up = udp_sk(sk);
429 ip_flush_pending_frames(sk);
432 EXPORT_SYMBOL(udp_flush_pending_frames);
435 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
436 * @sk: socket we are sending on
437 * @skb: sk_buff containing the filled-in UDP header
438 * (checksum field must be zeroed out)
440 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
441 __be32 src, __be32 dst, int len )
444 struct udphdr *uh = udp_hdr(skb);
447 if (skb_queue_len(&sk->sk_write_queue) == 1) {
449 * Only one fragment on the socket.
451 skb->csum_start = skb_transport_header(skb) - skb->head;
452 skb->csum_offset = offsetof(struct udphdr, check);
453 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
456 * HW-checksum won't work as there are two or more
457 * fragments on the socket so that all csums of sk_buffs
460 offset = skb_transport_offset(skb);
461 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
463 skb->ip_summed = CHECKSUM_NONE;
465 skb_queue_walk(&sk->sk_write_queue, skb) {
466 csum = csum_add(csum, skb->csum);
469 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
471 uh->check = CSUM_MANGLED_0;
476 * Push out all pending data as one UDP datagram. Socket is locked.
478 static int udp_push_pending_frames(struct sock *sk)
480 struct udp_sock *up = udp_sk(sk);
481 struct inet_sock *inet = inet_sk(sk);
482 struct flowi *fl = &inet->cork.fl;
486 int is_udplite = IS_UDPLITE(sk);
489 /* Grab the skbuff where UDP header space exists. */
490 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
494 * Create a UDP header
497 uh->source = fl->fl_ip_sport;
498 uh->dest = fl->fl_ip_dport;
499 uh->len = htons(up->len);
502 if (is_udplite) /* UDP-Lite */
503 csum = udplite_csum_outgoing(sk, skb);
505 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
507 skb->ip_summed = CHECKSUM_NONE;
510 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
512 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
515 } else /* `normal' UDP */
516 csum = udp_csum_outgoing(sk, skb);
518 /* add protocol-dependent pseudo-header */
519 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
520 sk->sk_protocol, csum );
522 uh->check = CSUM_MANGLED_0;
525 err = ip_push_pending_frames(sk);
530 UDP_INC_STATS_USER(sock_net(sk),
531 UDP_MIB_OUTDATAGRAMS, is_udplite);
535 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
538 struct inet_sock *inet = inet_sk(sk);
539 struct udp_sock *up = udp_sk(sk);
541 struct ipcm_cookie ipc;
542 struct rtable *rt = NULL;
545 __be32 daddr, faddr, saddr;
548 int err, is_udplite = IS_UDPLITE(sk);
549 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
550 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
559 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
566 * There are pending frames.
567 * The socket lock must be held while it's corked.
570 if (likely(up->pending)) {
571 if (unlikely(up->pending != AF_INET)) {
579 ulen += sizeof(struct udphdr);
582 * Get and verify the address.
585 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
586 if (msg->msg_namelen < sizeof(*usin))
588 if (usin->sin_family != AF_INET) {
589 if (usin->sin_family != AF_UNSPEC)
590 return -EAFNOSUPPORT;
593 daddr = usin->sin_addr.s_addr;
594 dport = usin->sin_port;
598 if (sk->sk_state != TCP_ESTABLISHED)
599 return -EDESTADDRREQ;
602 /* Open fast path for connected socket.
603 Route will not be used, if at least one option is set.
607 ipc.addr = inet->saddr;
609 ipc.oif = sk->sk_bound_dev_if;
610 if (msg->msg_controllen) {
611 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
622 ipc.addr = faddr = daddr;
624 if (ipc.opt && ipc.opt->srr) {
627 faddr = ipc.opt->faddr;
630 tos = RT_TOS(inet->tos);
631 if (sock_flag(sk, SOCK_LOCALROUTE) ||
632 (msg->msg_flags & MSG_DONTROUTE) ||
633 (ipc.opt && ipc.opt->is_strictroute)) {
638 if (ipv4_is_multicast(daddr)) {
640 ipc.oif = inet->mc_index;
642 saddr = inet->mc_addr;
647 rt = (struct rtable*)sk_dst_check(sk, 0);
650 struct flowi fl = { .oif = ipc.oif,
655 .proto = sk->sk_protocol,
657 { .sport = inet->sport,
658 .dport = dport } } };
659 struct net *net = sock_net(sk);
661 security_sk_classify_flow(sk, &fl);
662 err = ip_route_output_flow(net, &rt, &fl, sk, 1);
664 if (err == -ENETUNREACH)
665 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
670 if ((rt->rt_flags & RTCF_BROADCAST) &&
671 !sock_flag(sk, SOCK_BROADCAST))
674 sk_dst_set(sk, dst_clone(&rt->u.dst));
677 if (msg->msg_flags&MSG_CONFIRM)
683 daddr = ipc.addr = rt->rt_dst;
686 if (unlikely(up->pending)) {
687 /* The socket is already corked while preparing it. */
688 /* ... which is an evident application bug. --ANK */
691 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
696 * Now cork the socket to pend data.
698 inet->cork.fl.fl4_dst = daddr;
699 inet->cork.fl.fl_ip_dport = dport;
700 inet->cork.fl.fl4_src = saddr;
701 inet->cork.fl.fl_ip_sport = inet->sport;
702 up->pending = AF_INET;
706 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
707 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
708 sizeof(struct udphdr), &ipc, rt,
709 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
711 udp_flush_pending_frames(sk);
713 err = udp_push_pending_frames(sk);
714 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
725 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
726 * ENOBUFS might not be good (it's not tunable per se), but otherwise
727 * we don't have a good statistic (IpOutDiscards but it can be too many
728 * things). We could add another new stat but at least for now that
729 * seems like overkill.
731 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
732 UDP_INC_STATS_USER(sock_net(sk),
733 UDP_MIB_SNDBUFERRORS, is_udplite);
738 dst_confirm(&rt->u.dst);
739 if (!(msg->msg_flags&MSG_PROBE) || len)
740 goto back_from_confirm;
745 int udp_sendpage(struct sock *sk, struct page *page, int offset,
746 size_t size, int flags)
748 struct udp_sock *up = udp_sk(sk);
752 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
754 /* Call udp_sendmsg to specify destination address which
755 * sendpage interface can't pass.
756 * This will succeed only when the socket is connected.
758 ret = udp_sendmsg(NULL, sk, &msg, 0);
765 if (unlikely(!up->pending)) {
768 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
772 ret = ip_append_page(sk, page, offset, size, flags);
773 if (ret == -EOPNOTSUPP) {
775 return sock_no_sendpage(sk->sk_socket, page, offset,
779 udp_flush_pending_frames(sk);
784 if (!(up->corkflag || (flags&MSG_MORE)))
785 ret = udp_push_pending_frames(sk);
794 * IOCTL requests applicable to the UDP protocol
797 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
802 int amount = atomic_read(&sk->sk_wmem_alloc);
803 return put_user(amount, (int __user *)arg);
809 unsigned long amount;
812 spin_lock_bh(&sk->sk_receive_queue.lock);
813 skb = skb_peek(&sk->sk_receive_queue);
816 * We will only return the amount
817 * of this packet since that is all
820 amount = skb->len - sizeof(struct udphdr);
822 spin_unlock_bh(&sk->sk_receive_queue.lock);
823 return put_user(amount, (int __user *)arg);
834 * This should be easy, if there is something there we
835 * return it, otherwise we block.
838 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
839 size_t len, int noblock, int flags, int *addr_len)
841 struct inet_sock *inet = inet_sk(sk);
842 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
844 unsigned int ulen, copied;
847 int is_udplite = IS_UDPLITE(sk);
850 * Check any passed addresses
853 *addr_len=sizeof(*sin);
855 if (flags & MSG_ERRQUEUE)
856 return ip_recv_error(sk, msg, len);
859 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
864 ulen = skb->len - sizeof(struct udphdr);
868 else if (copied < ulen)
869 msg->msg_flags |= MSG_TRUNC;
872 * If checksum is needed at all, try to do it while copying the
873 * data. If the data is truncated, or if we only want a partial
874 * coverage checksum (UDP-Lite), do it before the copy.
877 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
878 if (udp_lib_checksum_complete(skb))
882 if (skb_csum_unnecessary(skb))
883 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
884 msg->msg_iov, copied );
886 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
896 UDP_INC_STATS_USER(sock_net(sk),
897 UDP_MIB_INDATAGRAMS, is_udplite);
899 sock_recv_timestamp(msg, sk, skb);
901 /* Copy the address. */
904 sin->sin_family = AF_INET;
905 sin->sin_port = udp_hdr(skb)->source;
906 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
907 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
909 if (inet->cmsg_flags)
910 ip_cmsg_recv(msg, skb);
913 if (flags & MSG_TRUNC)
918 skb_free_datagram(sk, skb);
925 if (!skb_kill_datagram(sk, skb, flags))
926 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
935 int udp_disconnect(struct sock *sk, int flags)
937 struct inet_sock *inet = inet_sk(sk);
939 * 1003.1g - break association.
942 sk->sk_state = TCP_CLOSE;
945 sk->sk_bound_dev_if = 0;
946 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
947 inet_reset_saddr(sk);
949 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
950 sk->sk_prot->unhash(sk);
960 * >0: "udp encap" protocol resubmission
962 * Note that in the success and error cases, the skb is assumed to
963 * have either been requeued or freed.
965 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
967 struct udp_sock *up = udp_sk(sk);
969 int is_udplite = IS_UDPLITE(sk);
972 * Charge it to the socket, dropping if the queue is full.
974 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
978 if (up->encap_type) {
980 * This is an encapsulation socket so pass the skb to
981 * the socket's udp_encap_rcv() hook. Otherwise, just
982 * fall through and pass this up the UDP socket.
983 * up->encap_rcv() returns the following value:
984 * =0 if skb was successfully passed to the encap
985 * handler or was discarded by it.
986 * >0 if skb should be passed on to UDP.
987 * <0 if skb should be resubmitted as proto -N
990 /* if we're overly short, let UDP handle it */
991 if (skb->len > sizeof(struct udphdr) &&
992 up->encap_rcv != NULL) {
995 ret = (*up->encap_rcv)(sk, skb);
997 UDP_INC_STATS_BH(sock_net(sk),
1004 /* FALLTHROUGH -- it's a UDP Packet */
1008 * UDP-Lite specific tests, ignored on UDP sockets
1010 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1013 * MIB statistics other than incrementing the error count are
1014 * disabled for the following two types of errors: these depend
1015 * on the application settings, not on the functioning of the
1016 * protocol stack as such.
1018 * RFC 3828 here recommends (sec 3.3): "There should also be a
1019 * way ... to ... at least let the receiving application block
1020 * delivery of packets with coverage values less than a value
1021 * provided by the application."
1023 if (up->pcrlen == 0) { /* full coverage was set */
1024 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1025 "%d while full coverage %d requested\n",
1026 UDP_SKB_CB(skb)->cscov, skb->len);
1029 /* The next case involves violating the min. coverage requested
1030 * by the receiver. This is subtle: if receiver wants x and x is
1031 * greater than the buffersize/MTU then receiver will complain
1032 * that it wants x while sender emits packets of smaller size y.
1033 * Therefore the above ...()->partial_cov statement is essential.
1035 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1036 LIMIT_NETDEBUG(KERN_WARNING
1037 "UDPLITE: coverage %d too small, need min %d\n",
1038 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1043 if (sk->sk_filter) {
1044 if (udp_lib_checksum_complete(skb))
1048 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
1049 /* Note that an ENOMEM error is charged twice */
1050 if (rc == -ENOMEM) {
1051 UDP_INC_STATS_BH(sock_net(sk),
1052 UDP_MIB_RCVBUFERRORS, is_udplite);
1053 atomic_inc(&sk->sk_drops);
1061 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1067 * Multicasts and broadcasts go to each listener.
1069 * Note: called only from the BH handler context,
1070 * so we don't need to lock the hashes.
1072 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1074 __be32 saddr, __be32 daddr,
1075 struct hlist_head udptable[])
1080 read_lock(&udp_hash_lock);
1081 sk = sk_head(&udptable[udp_hashfn(net, ntohs(uh->dest))]);
1082 dif = skb->dev->ifindex;
1083 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
1085 struct sock *sknext = NULL;
1088 struct sk_buff *skb1 = skb;
1090 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr,
1091 uh->source, saddr, dif);
1093 skb1 = skb_clone(skb, GFP_ATOMIC);
1098 bh_lock_sock_nested(sk);
1099 if (!sock_owned_by_user(sk))
1100 ret = udp_queue_rcv_skb(sk, skb1);
1102 sk_add_backlog(sk, skb1);
1106 /* we should probably re-process instead
1107 * of dropping packets here. */
1114 read_unlock(&udp_hash_lock);
1118 /* Initialize UDP checksum. If exited with zero value (success),
1119 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1120 * Otherwise, csum completion requires chacksumming packet body,
1121 * including udp header and folding it to skb->csum.
1123 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1126 const struct iphdr *iph;
1129 UDP_SKB_CB(skb)->partial_cov = 0;
1130 UDP_SKB_CB(skb)->cscov = skb->len;
1132 if (proto == IPPROTO_UDPLITE) {
1133 err = udplite_checksum_init(skb, uh);
1139 if (uh->check == 0) {
1140 skb->ip_summed = CHECKSUM_UNNECESSARY;
1141 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1142 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1144 skb->ip_summed = CHECKSUM_UNNECESSARY;
1146 if (!skb_csum_unnecessary(skb))
1147 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1148 skb->len, proto, 0);
1149 /* Probably, we should checksum udp header (it should be in cache
1150 * in any case) and data in tiny packets (< rx copybreak).
1157 * All we need to do is get the socket, and then do a checksum.
1160 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
1164 struct udphdr *uh = udp_hdr(skb);
1165 unsigned short ulen;
1166 struct rtable *rt = (struct rtable*)skb->dst;
1167 __be32 saddr = ip_hdr(skb)->saddr;
1168 __be32 daddr = ip_hdr(skb)->daddr;
1169 struct net *net = dev_net(skb->dev);
1172 * Validate the packet.
1174 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1175 goto drop; /* No space for header. */
1177 ulen = ntohs(uh->len);
1178 if (ulen > skb->len)
1181 if (proto == IPPROTO_UDP) {
1182 /* UDP validates ulen. */
1183 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1188 if (udp4_csum_init(skb, uh, proto))
1191 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1192 return __udp4_lib_mcast_deliver(net, skb, uh,
1193 saddr, daddr, udptable);
1195 sk = __udp4_lib_lookup(net, saddr, uh->source, daddr,
1196 uh->dest, inet_iif(skb), udptable);
1200 bh_lock_sock_nested(sk);
1201 if (!sock_owned_by_user(sk))
1202 ret = udp_queue_rcv_skb(sk, skb);
1204 sk_add_backlog(sk, skb);
1208 /* a return value > 0 means to resubmit the input, but
1209 * it wants the return to be -protocol, or 0
1216 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1220 /* No socket. Drop packet silently, if checksum is wrong */
1221 if (udp_lib_checksum_complete(skb))
1224 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1225 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1228 * Hmm. We got an UDP packet to a port to which we
1229 * don't wanna listen. Ignore it.
1235 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From " NIPQUAD_FMT ":%u %d/%d to " NIPQUAD_FMT ":%u\n",
1236 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1247 * RFC1122: OK. Discards the bad packet silently (as far as
1248 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1250 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From " NIPQUAD_FMT ":%u to " NIPQUAD_FMT ":%u ulen %d\n",
1251 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1258 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1263 int udp_rcv(struct sk_buff *skb)
1265 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP);
1268 void udp_destroy_sock(struct sock *sk)
1271 udp_flush_pending_frames(sk);
1276 * Socket option code for UDP
1278 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1279 char __user *optval, int optlen,
1280 int (*push_pending_frames)(struct sock *))
1282 struct udp_sock *up = udp_sk(sk);
1285 int is_udplite = IS_UDPLITE(sk);
1287 if (optlen<sizeof(int))
1290 if (get_user(val, (int __user *)optval))
1300 (*push_pending_frames)(sk);
1308 case UDP_ENCAP_ESPINUDP:
1309 case UDP_ENCAP_ESPINUDP_NON_IKE:
1310 up->encap_rcv = xfrm4_udp_encap_rcv;
1312 case UDP_ENCAP_L2TPINUDP:
1313 up->encap_type = val;
1322 * UDP-Lite's partial checksum coverage (RFC 3828).
1324 /* The sender sets actual checksum coverage length via this option.
1325 * The case coverage > packet length is handled by send module. */
1326 case UDPLITE_SEND_CSCOV:
1327 if (!is_udplite) /* Disable the option on UDP sockets */
1328 return -ENOPROTOOPT;
1329 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1332 up->pcflag |= UDPLITE_SEND_CC;
1335 /* The receiver specifies a minimum checksum coverage value. To make
1336 * sense, this should be set to at least 8 (as done below). If zero is
1337 * used, this again means full checksum coverage. */
1338 case UDPLITE_RECV_CSCOV:
1339 if (!is_udplite) /* Disable the option on UDP sockets */
1340 return -ENOPROTOOPT;
1341 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1344 up->pcflag |= UDPLITE_RECV_CC;
1355 int udp_setsockopt(struct sock *sk, int level, int optname,
1356 char __user *optval, int optlen)
1358 if (level == SOL_UDP || level == SOL_UDPLITE)
1359 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1360 udp_push_pending_frames);
1361 return ip_setsockopt(sk, level, optname, optval, optlen);
1364 #ifdef CONFIG_COMPAT
1365 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1366 char __user *optval, int optlen)
1368 if (level == SOL_UDP || level == SOL_UDPLITE)
1369 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1370 udp_push_pending_frames);
1371 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1375 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1376 char __user *optval, int __user *optlen)
1378 struct udp_sock *up = udp_sk(sk);
1381 if (get_user(len,optlen))
1384 len = min_t(unsigned int, len, sizeof(int));
1395 val = up->encap_type;
1398 /* The following two cannot be changed on UDP sockets, the return is
1399 * always 0 (which corresponds to the full checksum coverage of UDP). */
1400 case UDPLITE_SEND_CSCOV:
1404 case UDPLITE_RECV_CSCOV:
1409 return -ENOPROTOOPT;
1412 if (put_user(len, optlen))
1414 if (copy_to_user(optval, &val,len))
1419 int udp_getsockopt(struct sock *sk, int level, int optname,
1420 char __user *optval, int __user *optlen)
1422 if (level == SOL_UDP || level == SOL_UDPLITE)
1423 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1424 return ip_getsockopt(sk, level, optname, optval, optlen);
1427 #ifdef CONFIG_COMPAT
1428 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1429 char __user *optval, int __user *optlen)
1431 if (level == SOL_UDP || level == SOL_UDPLITE)
1432 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1433 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1437 * udp_poll - wait for a UDP event.
1438 * @file - file struct
1440 * @wait - poll table
1442 * This is same as datagram poll, except for the special case of
1443 * blocking sockets. If application is using a blocking fd
1444 * and a packet with checksum error is in the queue;
1445 * then it could get return from select indicating data available
1446 * but then block when reading it. Add special case code
1447 * to work around these arguably broken applications.
1449 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1451 unsigned int mask = datagram_poll(file, sock, wait);
1452 struct sock *sk = sock->sk;
1453 int is_lite = IS_UDPLITE(sk);
1455 /* Check for false positives due to checksum errors */
1456 if ( (mask & POLLRDNORM) &&
1457 !(file->f_flags & O_NONBLOCK) &&
1458 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1459 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1460 struct sk_buff *skb;
1462 spin_lock_bh(&rcvq->lock);
1463 while ((skb = skb_peek(rcvq)) != NULL &&
1464 udp_lib_checksum_complete(skb)) {
1465 UDP_INC_STATS_BH(sock_net(sk),
1466 UDP_MIB_INERRORS, is_lite);
1467 __skb_unlink(skb, rcvq);
1470 spin_unlock_bh(&rcvq->lock);
1472 /* nothing to see, move along */
1474 mask &= ~(POLLIN | POLLRDNORM);
1481 struct proto udp_prot = {
1483 .owner = THIS_MODULE,
1484 .close = udp_lib_close,
1485 .connect = ip4_datagram_connect,
1486 .disconnect = udp_disconnect,
1488 .destroy = udp_destroy_sock,
1489 .setsockopt = udp_setsockopt,
1490 .getsockopt = udp_getsockopt,
1491 .sendmsg = udp_sendmsg,
1492 .recvmsg = udp_recvmsg,
1493 .sendpage = udp_sendpage,
1494 .backlog_rcv = udp_queue_rcv_skb,
1495 .hash = udp_lib_hash,
1496 .unhash = udp_lib_unhash,
1497 .get_port = udp_v4_get_port,
1498 .memory_allocated = &udp_memory_allocated,
1499 .sysctl_mem = sysctl_udp_mem,
1500 .sysctl_wmem = &sysctl_udp_wmem_min,
1501 .sysctl_rmem = &sysctl_udp_rmem_min,
1502 .obj_size = sizeof(struct udp_sock),
1503 .h.udp_hash = udp_hash,
1504 #ifdef CONFIG_COMPAT
1505 .compat_setsockopt = compat_udp_setsockopt,
1506 .compat_getsockopt = compat_udp_getsockopt,
1510 /* ------------------------------------------------------------------------ */
1511 #ifdef CONFIG_PROC_FS
1513 static struct sock *udp_get_first(struct seq_file *seq)
1516 struct udp_iter_state *state = seq->private;
1517 struct net *net = seq_file_net(seq);
1519 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1520 struct hlist_node *node;
1521 sk_for_each(sk, node, state->hashtable + state->bucket) {
1522 if (!net_eq(sock_net(sk), net))
1524 if (sk->sk_family == state->family)
1533 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1535 struct udp_iter_state *state = seq->private;
1536 struct net *net = seq_file_net(seq);
1542 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1544 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) {
1545 sk = sk_head(state->hashtable + state->bucket);
1551 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1553 struct sock *sk = udp_get_first(seq);
1556 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1558 return pos ? NULL : sk;
1561 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1562 __acquires(udp_hash_lock)
1564 read_lock(&udp_hash_lock);
1565 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
1568 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1572 if (v == SEQ_START_TOKEN)
1573 sk = udp_get_idx(seq, 0);
1575 sk = udp_get_next(seq, v);
1581 static void udp_seq_stop(struct seq_file *seq, void *v)
1582 __releases(udp_hash_lock)
1584 read_unlock(&udp_hash_lock);
1587 static int udp_seq_open(struct inode *inode, struct file *file)
1589 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1590 struct udp_iter_state *s;
1593 err = seq_open_net(inode, file, &afinfo->seq_ops,
1594 sizeof(struct udp_iter_state));
1598 s = ((struct seq_file *)file->private_data)->private;
1599 s->family = afinfo->family;
1600 s->hashtable = afinfo->hashtable;
1604 /* ------------------------------------------------------------------------ */
1605 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
1607 struct proc_dir_entry *p;
1610 afinfo->seq_fops.open = udp_seq_open;
1611 afinfo->seq_fops.read = seq_read;
1612 afinfo->seq_fops.llseek = seq_lseek;
1613 afinfo->seq_fops.release = seq_release_net;
1615 afinfo->seq_ops.start = udp_seq_start;
1616 afinfo->seq_ops.next = udp_seq_next;
1617 afinfo->seq_ops.stop = udp_seq_stop;
1619 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
1620 &afinfo->seq_fops, afinfo);
1626 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
1628 proc_net_remove(net, afinfo->name);
1631 /* ------------------------------------------------------------------------ */
1632 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
1633 int bucket, int *len)
1635 struct inet_sock *inet = inet_sk(sp);
1636 __be32 dest = inet->daddr;
1637 __be32 src = inet->rcv_saddr;
1638 __u16 destp = ntohs(inet->dport);
1639 __u16 srcp = ntohs(inet->sport);
1641 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
1642 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
1643 bucket, src, srcp, dest, destp, sp->sk_state,
1644 atomic_read(&sp->sk_wmem_alloc),
1645 atomic_read(&sp->sk_rmem_alloc),
1646 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1647 atomic_read(&sp->sk_refcnt), sp,
1648 atomic_read(&sp->sk_drops), len);
1651 int udp4_seq_show(struct seq_file *seq, void *v)
1653 if (v == SEQ_START_TOKEN)
1654 seq_printf(seq, "%-127s\n",
1655 " sl local_address rem_address st tx_queue "
1656 "rx_queue tr tm->when retrnsmt uid timeout "
1657 "inode ref pointer drops");
1659 struct udp_iter_state *state = seq->private;
1662 udp4_format_sock(v, seq, state->bucket, &len);
1663 seq_printf(seq, "%*s\n", 127 - len ,"");
1668 /* ------------------------------------------------------------------------ */
1669 static struct udp_seq_afinfo udp4_seq_afinfo = {
1672 .hashtable = udp_hash,
1674 .owner = THIS_MODULE,
1677 .show = udp4_seq_show,
1681 static int udp4_proc_init_net(struct net *net)
1683 return udp_proc_register(net, &udp4_seq_afinfo);
1686 static void udp4_proc_exit_net(struct net *net)
1688 udp_proc_unregister(net, &udp4_seq_afinfo);
1691 static struct pernet_operations udp4_net_ops = {
1692 .init = udp4_proc_init_net,
1693 .exit = udp4_proc_exit_net,
1696 int __init udp4_proc_init(void)
1698 return register_pernet_subsys(&udp4_net_ops);
1701 void udp4_proc_exit(void)
1703 unregister_pernet_subsys(&udp4_net_ops);
1705 #endif /* CONFIG_PROC_FS */
1707 void __init udp_init(void)
1709 unsigned long limit;
1711 /* Set the pressure threshold up by the same strategy of TCP. It is a
1712 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
1713 * toward zero with the amount of memory, with a floor of 128 pages.
1715 limit = min(nr_all_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
1716 limit = (limit * (nr_all_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
1717 limit = max(limit, 128UL);
1718 sysctl_udp_mem[0] = limit / 4 * 3;
1719 sysctl_udp_mem[1] = limit;
1720 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
1722 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
1723 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
1726 EXPORT_SYMBOL(udp_disconnect);
1727 EXPORT_SYMBOL(udp_hash);
1728 EXPORT_SYMBOL(udp_hash_lock);
1729 EXPORT_SYMBOL(udp_ioctl);
1730 EXPORT_SYMBOL(udp_prot);
1731 EXPORT_SYMBOL(udp_sendmsg);
1732 EXPORT_SYMBOL(udp_lib_getsockopt);
1733 EXPORT_SYMBOL(udp_lib_setsockopt);
1734 EXPORT_SYMBOL(udp_poll);
1735 EXPORT_SYMBOL(udp_lib_get_port);
1737 #ifdef CONFIG_PROC_FS
1738 EXPORT_SYMBOL(udp_proc_register);
1739 EXPORT_SYMBOL(udp_proc_unregister);