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).
8 * Version: $Id: udp.c,v 1.102 2002/02/01 22:01:04 davem Exp $
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
13 * Alan Cox, <Alan.Cox@linux.org>
14 * Hirokazu Takahashi, <taka@valinux.co.jp>
17 * Alan Cox : verify_area() calls
18 * Alan Cox : stopped close while in use off icmp
19 * messages. Not a fix but a botch that
20 * for udp at least is 'valid'.
21 * Alan Cox : Fixed icmp handling properly
22 * Alan Cox : Correct error for oversized datagrams
23 * Alan Cox : Tidied select() semantics.
24 * Alan Cox : udp_err() fixed properly, also now
25 * select and read wake correctly on errors
26 * Alan Cox : udp_send verify_area moved to avoid mem leak
27 * Alan Cox : UDP can count its memory
28 * Alan Cox : send to an unknown connection causes
29 * an ECONNREFUSED off the icmp, but
31 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
32 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
33 * bug no longer crashes it.
34 * Fred Van Kempen : Net2e support for sk->broadcast.
35 * Alan Cox : Uses skb_free_datagram
36 * Alan Cox : Added get/set sockopt support.
37 * Alan Cox : Broadcasting without option set returns EACCES.
38 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
39 * Alan Cox : Use ip_tos and ip_ttl
40 * Alan Cox : SNMP Mibs
41 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
42 * Matt Dillon : UDP length checks.
43 * Alan Cox : Smarter af_inet used properly.
44 * Alan Cox : Use new kernel side addressing.
45 * Alan Cox : Incorrect return on truncated datagram receive.
46 * Arnt Gulbrandsen : New udp_send and stuff
47 * Alan Cox : Cache last socket
48 * Alan Cox : Route cache
49 * Jon Peatfield : Minor efficiency fix to sendto().
50 * Mike Shaver : RFC1122 checks.
51 * Alan Cox : Nonblocking error fix.
52 * Willy Konynenberg : Transparent proxying support.
53 * Mike McLagan : Routing by source
54 * David S. Miller : New socket lookup architecture.
55 * Last socket cache retained as it
56 * does have a high hit rate.
57 * Olaf Kirch : Don't linearise iovec on sendmsg.
58 * Andi Kleen : Some cleanups, cache destination entry
60 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
61 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
62 * return ENOTCONN for unconnected sockets (POSIX)
63 * Janos Farkas : don't deliver multi/broadcasts to a different
64 * bound-to-device socket
65 * Hirokazu Takahashi : HW checksumming for outgoing UDP
67 * Hirokazu Takahashi : sendfile() on UDP works now.
68 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
69 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
70 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
71 * a single port at the same time.
72 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
73 * James Chapman : Add L2TP encapsulation type.
76 * This program is free software; you can redistribute it and/or
77 * modify it under the terms of the GNU General Public License
78 * as published by the Free Software Foundation; either version
79 * 2 of the License, or (at your option) any later version.
82 #include <asm/system.h>
83 #include <asm/uaccess.h>
84 #include <asm/ioctls.h>
85 #include <linux/bootmem.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>
93 #include <linux/errno.h>
94 #include <linux/timer.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"
110 * Snmp MIB for the UDP layer
113 DEFINE_SNMP_STAT(struct udp_mib, udp_statistics) __read_mostly;
114 EXPORT_SYMBOL(udp_statistics);
116 DEFINE_SNMP_STAT(struct udp_mib, udp_stats_in6) __read_mostly;
117 EXPORT_SYMBOL(udp_stats_in6);
119 struct hlist_head udp_hash[UDP_HTABLE_SIZE];
120 DEFINE_RWLOCK(udp_hash_lock);
122 int sysctl_udp_mem[3] __read_mostly;
123 int sysctl_udp_rmem_min __read_mostly;
124 int sysctl_udp_wmem_min __read_mostly;
126 EXPORT_SYMBOL(sysctl_udp_mem);
127 EXPORT_SYMBOL(sysctl_udp_rmem_min);
128 EXPORT_SYMBOL(sysctl_udp_wmem_min);
130 atomic_t udp_memory_allocated;
131 EXPORT_SYMBOL(udp_memory_allocated);
133 static inline int __udp_lib_lport_inuse(struct net *net, __u16 num,
134 const struct hlist_head udptable[])
137 struct hlist_node *node;
139 sk_for_each(sk, node, &udptable[num & (UDP_HTABLE_SIZE - 1)])
140 if (sk->sk_net == net && sk->sk_hash == num)
146 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
148 * @sk: socket struct in question
149 * @snum: port number to look up
150 * @saddr_comp: AF-dependent comparison of bound local IP addresses
152 int udp_lib_get_port(struct sock *sk, unsigned short snum,
153 int (*saddr_comp)(const struct sock *sk1,
154 const struct sock *sk2 ) )
156 struct hlist_head *udptable = sk->sk_prot->h.udp_hash;
157 struct hlist_node *node;
158 struct hlist_head *head;
161 struct net *net = sk->sk_net;
163 write_lock_bh(&udp_hash_lock);
166 int i, low, high, remaining;
167 unsigned rover, best, best_size_so_far;
169 inet_get_local_port_range(&low, &high);
170 remaining = (high - low) + 1;
172 best_size_so_far = UINT_MAX;
173 best = rover = net_random() % remaining + low;
175 /* 1st pass: look for empty (or shortest) hash chain */
176 for (i = 0; i < UDP_HTABLE_SIZE; i++) {
179 head = &udptable[rover & (UDP_HTABLE_SIZE - 1)];
180 if (hlist_empty(head))
183 sk_for_each(sk2, node, head) {
184 if (++size >= best_size_so_far)
187 best_size_so_far = size;
190 /* fold back if end of range */
192 rover = low + ((rover - low)
193 & (UDP_HTABLE_SIZE - 1));
198 /* 2nd pass: find hole in shortest hash chain */
200 for (i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++) {
201 if (! __udp_lib_lport_inuse(net, rover, udptable))
203 rover += UDP_HTABLE_SIZE;
205 rover = low + ((rover - low)
206 & (UDP_HTABLE_SIZE - 1));
210 /* All ports in use! */
216 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
218 sk_for_each(sk2, node, head)
219 if (sk2->sk_hash == snum &&
221 sk2->sk_net == net &&
222 (!sk2->sk_reuse || !sk->sk_reuse) &&
223 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
224 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
225 (*saddr_comp)(sk, sk2) )
229 inet_sk(sk)->num = snum;
231 if (sk_unhashed(sk)) {
232 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
233 sk_add_node(sk, head);
234 sock_prot_inuse_add(sk->sk_prot, 1);
238 write_unlock_bh(&udp_hash_lock);
242 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
244 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
246 return ( !ipv6_only_sock(sk2) &&
247 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
248 inet1->rcv_saddr == inet2->rcv_saddr ));
251 int udp_v4_get_port(struct sock *sk, unsigned short snum)
253 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal);
256 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
257 * harder than this. -DaveM
259 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
260 __be16 sport, __be32 daddr, __be16 dport,
261 int dif, struct hlist_head udptable[])
263 struct sock *sk, *result = NULL;
264 struct hlist_node *node;
265 unsigned short hnum = ntohs(dport);
268 read_lock(&udp_hash_lock);
269 sk_for_each(sk, node, &udptable[hnum & (UDP_HTABLE_SIZE - 1)]) {
270 struct inet_sock *inet = inet_sk(sk);
272 if (sk->sk_net == net && sk->sk_hash == hnum &&
273 !ipv6_only_sock(sk)) {
274 int score = (sk->sk_family == PF_INET ? 1 : 0);
275 if (inet->rcv_saddr) {
276 if (inet->rcv_saddr != daddr)
281 if (inet->daddr != saddr)
286 if (inet->dport != sport)
290 if (sk->sk_bound_dev_if) {
291 if (sk->sk_bound_dev_if != dif)
298 } else if (score > badness) {
306 read_unlock(&udp_hash_lock);
310 static inline struct sock *udp_v4_mcast_next(struct sock *sk,
311 __be16 loc_port, __be32 loc_addr,
312 __be16 rmt_port, __be32 rmt_addr,
315 struct hlist_node *node;
317 unsigned short hnum = ntohs(loc_port);
319 sk_for_each_from(s, node) {
320 struct inet_sock *inet = inet_sk(s);
322 if (s->sk_hash != hnum ||
323 (inet->daddr && inet->daddr != rmt_addr) ||
324 (inet->dport != rmt_port && inet->dport) ||
325 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
327 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
329 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
339 * This routine is called by the ICMP module when it gets some
340 * sort of error condition. If err < 0 then the socket should
341 * be closed and the error returned to the user. If err > 0
342 * it's just the icmp type << 8 | icmp code.
343 * Header points to the ip header of the error packet. We move
344 * on past this. Then (as it used to claim before adjustment)
345 * header points to the first 8 bytes of the udp header. We need
346 * to find the appropriate port.
349 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[])
351 struct inet_sock *inet;
352 struct iphdr *iph = (struct iphdr*)skb->data;
353 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
354 const int type = icmp_hdr(skb)->type;
355 const int code = icmp_hdr(skb)->code;
360 sk = __udp4_lib_lookup(skb->dev->nd_net, iph->daddr, uh->dest,
361 iph->saddr, uh->source, skb->dev->ifindex, udptable);
363 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
364 return; /* No socket for error */
373 case ICMP_TIME_EXCEEDED:
376 case ICMP_SOURCE_QUENCH:
378 case ICMP_PARAMETERPROB:
382 case ICMP_DEST_UNREACH:
383 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
384 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
392 if (code <= NR_ICMP_UNREACH) {
393 harderr = icmp_err_convert[code].fatal;
394 err = icmp_err_convert[code].errno;
400 * RFC1122: OK. Passes ICMP errors back to application, as per
403 if (!inet->recverr) {
404 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
407 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
410 sk->sk_error_report(sk);
415 void udp_err(struct sk_buff *skb, u32 info)
417 __udp4_lib_err(skb, info, udp_hash);
421 * Throw away all pending data and cancel the corking. Socket is locked.
423 static void udp_flush_pending_frames(struct sock *sk)
425 struct udp_sock *up = udp_sk(sk);
430 ip_flush_pending_frames(sk);
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(UDP_MIB_OUTDATAGRAMS, is_udplite);
534 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
537 struct inet_sock *inet = inet_sk(sk);
538 struct udp_sock *up = udp_sk(sk);
540 struct ipcm_cookie ipc;
541 struct rtable *rt = NULL;
544 __be32 daddr, faddr, saddr;
547 int err, is_udplite = IS_UDPLITE(sk);
548 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
549 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
558 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
565 * There are pending frames.
566 * The socket lock must be held while it's corked.
569 if (likely(up->pending)) {
570 if (unlikely(up->pending != AF_INET)) {
578 ulen += sizeof(struct udphdr);
581 * Get and verify the address.
584 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
585 if (msg->msg_namelen < sizeof(*usin))
587 if (usin->sin_family != AF_INET) {
588 if (usin->sin_family != AF_UNSPEC)
589 return -EAFNOSUPPORT;
592 daddr = usin->sin_addr.s_addr;
593 dport = usin->sin_port;
597 if (sk->sk_state != TCP_ESTABLISHED)
598 return -EDESTADDRREQ;
601 /* Open fast path for connected socket.
602 Route will not be used, if at least one option is set.
606 ipc.addr = inet->saddr;
608 ipc.oif = sk->sk_bound_dev_if;
609 if (msg->msg_controllen) {
610 err = ip_cmsg_send(sk->sk_net, msg, &ipc);
621 ipc.addr = faddr = daddr;
623 if (ipc.opt && ipc.opt->srr) {
626 faddr = ipc.opt->faddr;
629 tos = RT_TOS(inet->tos);
630 if (sock_flag(sk, SOCK_LOCALROUTE) ||
631 (msg->msg_flags & MSG_DONTROUTE) ||
632 (ipc.opt && ipc.opt->is_strictroute)) {
637 if (ipv4_is_multicast(daddr)) {
639 ipc.oif = inet->mc_index;
641 saddr = inet->mc_addr;
646 rt = (struct rtable*)sk_dst_check(sk, 0);
649 struct flowi fl = { .oif = ipc.oif,
654 .proto = sk->sk_protocol,
656 { .sport = inet->sport,
657 .dport = dport } } };
658 security_sk_classify_flow(sk, &fl);
659 err = ip_route_output_flow(sk->sk_net, &rt, &fl, sk, 1);
661 if (err == -ENETUNREACH)
662 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
667 if ((rt->rt_flags & RTCF_BROADCAST) &&
668 !sock_flag(sk, SOCK_BROADCAST))
671 sk_dst_set(sk, dst_clone(&rt->u.dst));
674 if (msg->msg_flags&MSG_CONFIRM)
680 daddr = ipc.addr = rt->rt_dst;
683 if (unlikely(up->pending)) {
684 /* The socket is already corked while preparing it. */
685 /* ... which is an evident application bug. --ANK */
688 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
693 * Now cork the socket to pend data.
695 inet->cork.fl.fl4_dst = daddr;
696 inet->cork.fl.fl_ip_dport = dport;
697 inet->cork.fl.fl4_src = saddr;
698 inet->cork.fl.fl_ip_sport = inet->sport;
699 up->pending = AF_INET;
703 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
704 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
705 sizeof(struct udphdr), &ipc, rt,
706 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
708 udp_flush_pending_frames(sk);
710 err = udp_push_pending_frames(sk);
711 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
722 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
723 * ENOBUFS might not be good (it's not tunable per se), but otherwise
724 * we don't have a good statistic (IpOutDiscards but it can be too many
725 * things). We could add another new stat but at least for now that
726 * seems like overkill.
728 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
729 UDP_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite);
734 dst_confirm(&rt->u.dst);
735 if (!(msg->msg_flags&MSG_PROBE) || len)
736 goto back_from_confirm;
741 int udp_sendpage(struct sock *sk, struct page *page, int offset,
742 size_t size, int flags)
744 struct udp_sock *up = udp_sk(sk);
748 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
750 /* Call udp_sendmsg to specify destination address which
751 * sendpage interface can't pass.
752 * This will succeed only when the socket is connected.
754 ret = udp_sendmsg(NULL, sk, &msg, 0);
761 if (unlikely(!up->pending)) {
764 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
768 ret = ip_append_page(sk, page, offset, size, flags);
769 if (ret == -EOPNOTSUPP) {
771 return sock_no_sendpage(sk->sk_socket, page, offset,
775 udp_flush_pending_frames(sk);
780 if (!(up->corkflag || (flags&MSG_MORE)))
781 ret = udp_push_pending_frames(sk);
790 * IOCTL requests applicable to the UDP protocol
793 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
798 int amount = atomic_read(&sk->sk_wmem_alloc);
799 return put_user(amount, (int __user *)arg);
805 unsigned long amount;
808 spin_lock_bh(&sk->sk_receive_queue.lock);
809 skb = skb_peek(&sk->sk_receive_queue);
812 * We will only return the amount
813 * of this packet since that is all
816 amount = skb->len - sizeof(struct udphdr);
818 spin_unlock_bh(&sk->sk_receive_queue.lock);
819 return put_user(amount, (int __user *)arg);
830 * This should be easy, if there is something there we
831 * return it, otherwise we block.
834 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
835 size_t len, int noblock, int flags, int *addr_len)
837 struct inet_sock *inet = inet_sk(sk);
838 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
840 unsigned int ulen, copied;
843 int is_udplite = IS_UDPLITE(sk);
846 * Check any passed addresses
849 *addr_len=sizeof(*sin);
851 if (flags & MSG_ERRQUEUE)
852 return ip_recv_error(sk, msg, len);
855 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
860 ulen = skb->len - sizeof(struct udphdr);
864 else if (copied < ulen)
865 msg->msg_flags |= MSG_TRUNC;
868 * If checksum is needed at all, try to do it while copying the
869 * data. If the data is truncated, or if we only want a partial
870 * coverage checksum (UDP-Lite), do it before the copy.
873 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
874 if (udp_lib_checksum_complete(skb))
878 if (skb_csum_unnecessary(skb))
879 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
880 msg->msg_iov, copied );
882 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
892 UDP_INC_STATS_USER(UDP_MIB_INDATAGRAMS, is_udplite);
894 sock_recv_timestamp(msg, sk, skb);
896 /* Copy the address. */
899 sin->sin_family = AF_INET;
900 sin->sin_port = udp_hdr(skb)->source;
901 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
902 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
904 if (inet->cmsg_flags)
905 ip_cmsg_recv(msg, skb);
908 if (flags & MSG_TRUNC)
913 skb_free_datagram(sk, skb);
920 if (!skb_kill_datagram(sk, skb, flags))
921 UDP_INC_STATS_USER(UDP_MIB_INERRORS, is_udplite);
930 int udp_disconnect(struct sock *sk, int flags)
932 struct inet_sock *inet = inet_sk(sk);
934 * 1003.1g - break association.
937 sk->sk_state = TCP_CLOSE;
940 sk->sk_bound_dev_if = 0;
941 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
942 inet_reset_saddr(sk);
944 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
945 sk->sk_prot->unhash(sk);
955 * >0: "udp encap" protocol resubmission
957 * Note that in the success and error cases, the skb is assumed to
958 * have either been requeued or freed.
960 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
962 struct udp_sock *up = udp_sk(sk);
964 int is_udplite = IS_UDPLITE(sk);
967 * Charge it to the socket, dropping if the queue is full.
969 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
973 if (up->encap_type) {
975 * This is an encapsulation socket so pass the skb to
976 * the socket's udp_encap_rcv() hook. Otherwise, just
977 * fall through and pass this up the UDP socket.
978 * up->encap_rcv() returns the following value:
979 * =0 if skb was successfully passed to the encap
980 * handler or was discarded by it.
981 * >0 if skb should be passed on to UDP.
982 * <0 if skb should be resubmitted as proto -N
985 /* if we're overly short, let UDP handle it */
986 if (skb->len > sizeof(struct udphdr) &&
987 up->encap_rcv != NULL) {
990 ret = (*up->encap_rcv)(sk, skb);
992 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS,
998 /* FALLTHROUGH -- it's a UDP Packet */
1002 * UDP-Lite specific tests, ignored on UDP sockets
1004 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1007 * MIB statistics other than incrementing the error count are
1008 * disabled for the following two types of errors: these depend
1009 * on the application settings, not on the functioning of the
1010 * protocol stack as such.
1012 * RFC 3828 here recommends (sec 3.3): "There should also be a
1013 * way ... to ... at least let the receiving application block
1014 * delivery of packets with coverage values less than a value
1015 * provided by the application."
1017 if (up->pcrlen == 0) { /* full coverage was set */
1018 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1019 "%d while full coverage %d requested\n",
1020 UDP_SKB_CB(skb)->cscov, skb->len);
1023 /* The next case involves violating the min. coverage requested
1024 * by the receiver. This is subtle: if receiver wants x and x is
1025 * greater than the buffersize/MTU then receiver will complain
1026 * that it wants x while sender emits packets of smaller size y.
1027 * Therefore the above ...()->partial_cov statement is essential.
1029 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1030 LIMIT_NETDEBUG(KERN_WARNING
1031 "UDPLITE: coverage %d too small, need min %d\n",
1032 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1037 if (sk->sk_filter) {
1038 if (udp_lib_checksum_complete(skb))
1042 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
1043 /* Note that an ENOMEM error is charged twice */
1045 UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, is_udplite);
1052 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite);
1058 * Multicasts and broadcasts go to each listener.
1060 * Note: called only from the BH handler context,
1061 * so we don't need to lock the hashes.
1063 static int __udp4_lib_mcast_deliver(struct sk_buff *skb,
1065 __be32 saddr, __be32 daddr,
1066 struct hlist_head udptable[])
1071 read_lock(&udp_hash_lock);
1072 sk = sk_head(&udptable[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]);
1073 dif = skb->dev->ifindex;
1074 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
1076 struct sock *sknext = NULL;
1079 struct sk_buff *skb1 = skb;
1081 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr,
1082 uh->source, saddr, dif);
1084 skb1 = skb_clone(skb, GFP_ATOMIC);
1089 bh_lock_sock_nested(sk);
1090 if (!sock_owned_by_user(sk))
1091 ret = udp_queue_rcv_skb(sk, skb1);
1093 sk_add_backlog(sk, skb1);
1097 /* we should probably re-process instead
1098 * of dropping packets here. */
1105 read_unlock(&udp_hash_lock);
1109 /* Initialize UDP checksum. If exited with zero value (success),
1110 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1111 * Otherwise, csum completion requires chacksumming packet body,
1112 * including udp header and folding it to skb->csum.
1114 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1117 const struct iphdr *iph;
1120 UDP_SKB_CB(skb)->partial_cov = 0;
1121 UDP_SKB_CB(skb)->cscov = skb->len;
1123 if (proto == IPPROTO_UDPLITE) {
1124 err = udplite_checksum_init(skb, uh);
1130 if (uh->check == 0) {
1131 skb->ip_summed = CHECKSUM_UNNECESSARY;
1132 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1133 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1135 skb->ip_summed = CHECKSUM_UNNECESSARY;
1137 if (!skb_csum_unnecessary(skb))
1138 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1139 skb->len, proto, 0);
1140 /* Probably, we should checksum udp header (it should be in cache
1141 * in any case) and data in tiny packets (< rx copybreak).
1148 * All we need to do is get the socket, and then do a checksum.
1151 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
1155 struct udphdr *uh = udp_hdr(skb);
1156 unsigned short ulen;
1157 struct rtable *rt = (struct rtable*)skb->dst;
1158 __be32 saddr = ip_hdr(skb)->saddr;
1159 __be32 daddr = ip_hdr(skb)->daddr;
1162 * Validate the packet.
1164 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1165 goto drop; /* No space for header. */
1167 ulen = ntohs(uh->len);
1168 if (ulen > skb->len)
1171 if (proto == IPPROTO_UDP) {
1172 /* UDP validates ulen. */
1173 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1178 if (udp4_csum_init(skb, uh, proto))
1181 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1182 return __udp4_lib_mcast_deliver(skb, uh, saddr, daddr, udptable);
1184 sk = __udp4_lib_lookup(skb->dev->nd_net, saddr, uh->source, daddr,
1185 uh->dest, inet_iif(skb), udptable);
1189 bh_lock_sock_nested(sk);
1190 if (!sock_owned_by_user(sk))
1191 ret = udp_queue_rcv_skb(sk, skb);
1193 sk_add_backlog(sk, skb);
1197 /* a return value > 0 means to resubmit the input, but
1198 * it wants the return to be -protocol, or 0
1205 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1209 /* No socket. Drop packet silently, if checksum is wrong */
1210 if (udp_lib_checksum_complete(skb))
1213 UDP_INC_STATS_BH(UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1214 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1217 * Hmm. We got an UDP packet to a port to which we
1218 * don't wanna listen. Ignore it.
1224 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %u.%u.%u.%u:%u %d/%d to %u.%u.%u.%u:%u\n",
1225 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1236 * RFC1122: OK. Discards the bad packet silently (as far as
1237 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1239 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %d.%d.%d.%d:%d to %d.%d.%d.%d:%d ulen %d\n",
1240 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1247 UDP_INC_STATS_BH(UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1252 int udp_rcv(struct sk_buff *skb)
1254 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP);
1257 int udp_destroy_sock(struct sock *sk)
1260 udp_flush_pending_frames(sk);
1266 * Socket option code for UDP
1268 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1269 char __user *optval, int optlen,
1270 int (*push_pending_frames)(struct sock *))
1272 struct udp_sock *up = udp_sk(sk);
1275 int is_udplite = IS_UDPLITE(sk);
1277 if (optlen<sizeof(int))
1280 if (get_user(val, (int __user *)optval))
1290 (*push_pending_frames)(sk);
1298 case UDP_ENCAP_ESPINUDP:
1299 case UDP_ENCAP_ESPINUDP_NON_IKE:
1300 up->encap_rcv = xfrm4_udp_encap_rcv;
1302 case UDP_ENCAP_L2TPINUDP:
1303 up->encap_type = val;
1312 * UDP-Lite's partial checksum coverage (RFC 3828).
1314 /* The sender sets actual checksum coverage length via this option.
1315 * The case coverage > packet length is handled by send module. */
1316 case UDPLITE_SEND_CSCOV:
1317 if (!is_udplite) /* Disable the option on UDP sockets */
1318 return -ENOPROTOOPT;
1319 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1322 up->pcflag |= UDPLITE_SEND_CC;
1325 /* The receiver specifies a minimum checksum coverage value. To make
1326 * sense, this should be set to at least 8 (as done below). If zero is
1327 * used, this again means full checksum coverage. */
1328 case UDPLITE_RECV_CSCOV:
1329 if (!is_udplite) /* Disable the option on UDP sockets */
1330 return -ENOPROTOOPT;
1331 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1334 up->pcflag |= UDPLITE_RECV_CC;
1345 int udp_setsockopt(struct sock *sk, int level, int optname,
1346 char __user *optval, int optlen)
1348 if (level == SOL_UDP || level == SOL_UDPLITE)
1349 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1350 udp_push_pending_frames);
1351 return ip_setsockopt(sk, level, optname, optval, optlen);
1354 #ifdef CONFIG_COMPAT
1355 int compat_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 compat_ip_setsockopt(sk, level, optname, optval, optlen);
1365 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1366 char __user *optval, int __user *optlen)
1368 struct udp_sock *up = udp_sk(sk);
1371 if (get_user(len,optlen))
1374 len = min_t(unsigned int, len, sizeof(int));
1385 val = up->encap_type;
1388 /* The following two cannot be changed on UDP sockets, the return is
1389 * always 0 (which corresponds to the full checksum coverage of UDP). */
1390 case UDPLITE_SEND_CSCOV:
1394 case UDPLITE_RECV_CSCOV:
1399 return -ENOPROTOOPT;
1402 if (put_user(len, optlen))
1404 if (copy_to_user(optval, &val,len))
1409 int udp_getsockopt(struct sock *sk, int level, int optname,
1410 char __user *optval, int __user *optlen)
1412 if (level == SOL_UDP || level == SOL_UDPLITE)
1413 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1414 return ip_getsockopt(sk, level, optname, optval, optlen);
1417 #ifdef CONFIG_COMPAT
1418 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1419 char __user *optval, int __user *optlen)
1421 if (level == SOL_UDP || level == SOL_UDPLITE)
1422 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1423 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1427 * udp_poll - wait for a UDP event.
1428 * @file - file struct
1430 * @wait - poll table
1432 * This is same as datagram poll, except for the special case of
1433 * blocking sockets. If application is using a blocking fd
1434 * and a packet with checksum error is in the queue;
1435 * then it could get return from select indicating data available
1436 * but then block when reading it. Add special case code
1437 * to work around these arguably broken applications.
1439 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1441 unsigned int mask = datagram_poll(file, sock, wait);
1442 struct sock *sk = sock->sk;
1443 int is_lite = IS_UDPLITE(sk);
1445 /* Check for false positives due to checksum errors */
1446 if ( (mask & POLLRDNORM) &&
1447 !(file->f_flags & O_NONBLOCK) &&
1448 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1449 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1450 struct sk_buff *skb;
1452 spin_lock_bh(&rcvq->lock);
1453 while ((skb = skb_peek(rcvq)) != NULL &&
1454 udp_lib_checksum_complete(skb)) {
1455 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_lite);
1456 __skb_unlink(skb, rcvq);
1459 spin_unlock_bh(&rcvq->lock);
1461 /* nothing to see, move along */
1463 mask &= ~(POLLIN | POLLRDNORM);
1470 DEFINE_PROTO_INUSE(udp)
1472 struct proto udp_prot = {
1474 .owner = THIS_MODULE,
1475 .close = udp_lib_close,
1476 .connect = ip4_datagram_connect,
1477 .disconnect = udp_disconnect,
1479 .destroy = udp_destroy_sock,
1480 .setsockopt = udp_setsockopt,
1481 .getsockopt = udp_getsockopt,
1482 .sendmsg = udp_sendmsg,
1483 .recvmsg = udp_recvmsg,
1484 .sendpage = udp_sendpage,
1485 .backlog_rcv = udp_queue_rcv_skb,
1486 .hash = udp_lib_hash,
1487 .unhash = udp_lib_unhash,
1488 .get_port = udp_v4_get_port,
1489 .memory_allocated = &udp_memory_allocated,
1490 .sysctl_mem = sysctl_udp_mem,
1491 .sysctl_wmem = &sysctl_udp_wmem_min,
1492 .sysctl_rmem = &sysctl_udp_rmem_min,
1493 .obj_size = sizeof(struct udp_sock),
1494 .h.udp_hash = udp_hash,
1495 #ifdef CONFIG_COMPAT
1496 .compat_setsockopt = compat_udp_setsockopt,
1497 .compat_getsockopt = compat_udp_getsockopt,
1499 REF_PROTO_INUSE(udp)
1502 /* ------------------------------------------------------------------------ */
1503 #ifdef CONFIG_PROC_FS
1505 static struct sock *udp_get_first(struct seq_file *seq)
1508 struct udp_iter_state *state = seq->private;
1509 struct net *net = state->net;
1511 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1512 struct hlist_node *node;
1513 sk_for_each(sk, node, state->hashtable + state->bucket) {
1514 if (sk->sk_net != net)
1516 if (sk->sk_family == state->family)
1525 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1527 struct udp_iter_state *state = seq->private;
1528 struct net *net = state->net;
1534 } while (sk && (sk->sk_net != net || sk->sk_family != state->family));
1536 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) {
1537 sk = sk_head(state->hashtable + state->bucket);
1543 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1545 struct sock *sk = udp_get_first(seq);
1548 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1550 return pos ? NULL : sk;
1553 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1554 __acquires(udp_hash_lock)
1556 read_lock(&udp_hash_lock);
1557 return *pos ? udp_get_idx(seq, *pos-1) : (void *)1;
1560 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1565 sk = udp_get_idx(seq, 0);
1567 sk = udp_get_next(seq, v);
1573 static void udp_seq_stop(struct seq_file *seq, void *v)
1574 __releases(udp_hash_lock)
1576 read_unlock(&udp_hash_lock);
1579 static int udp_seq_open(struct inode *inode, struct file *file)
1581 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1582 struct seq_file *seq;
1585 struct udp_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
1591 net = get_proc_net(inode);
1595 s->family = afinfo->family;
1596 s->hashtable = afinfo->hashtable;
1597 s->seq_ops.start = udp_seq_start;
1598 s->seq_ops.next = udp_seq_next;
1599 s->seq_ops.show = afinfo->seq_show;
1600 s->seq_ops.stop = udp_seq_stop;
1603 rc = seq_open(file, &s->seq_ops);
1607 seq = file->private_data;
1618 static int udp_seq_release(struct inode *inode, struct file *file)
1620 struct seq_file *seq = file->private_data;
1621 struct udp_iter_state *s = seq->private;
1624 seq_release_private(inode, file);
1628 /* ------------------------------------------------------------------------ */
1629 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
1631 struct proc_dir_entry *p;
1636 afinfo->seq_fops->owner = afinfo->owner;
1637 afinfo->seq_fops->open = udp_seq_open;
1638 afinfo->seq_fops->read = seq_read;
1639 afinfo->seq_fops->llseek = seq_lseek;
1640 afinfo->seq_fops->release = udp_seq_release;
1642 p = proc_net_fops_create(net, afinfo->name, S_IRUGO, afinfo->seq_fops);
1650 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
1654 proc_net_remove(net, afinfo->name);
1655 memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops));
1658 /* ------------------------------------------------------------------------ */
1659 static void udp4_format_sock(struct sock *sp, char *tmpbuf, int bucket)
1661 struct inet_sock *inet = inet_sk(sp);
1662 __be32 dest = inet->daddr;
1663 __be32 src = inet->rcv_saddr;
1664 __u16 destp = ntohs(inet->dport);
1665 __u16 srcp = ntohs(inet->sport);
1667 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1668 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p",
1669 bucket, src, srcp, dest, destp, sp->sk_state,
1670 atomic_read(&sp->sk_wmem_alloc),
1671 atomic_read(&sp->sk_rmem_alloc),
1672 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1673 atomic_read(&sp->sk_refcnt), sp);
1676 int udp4_seq_show(struct seq_file *seq, void *v)
1678 if (v == SEQ_START_TOKEN)
1679 seq_printf(seq, "%-127s\n",
1680 " sl local_address rem_address st tx_queue "
1681 "rx_queue tr tm->when retrnsmt uid timeout "
1685 struct udp_iter_state *state = seq->private;
1687 udp4_format_sock(v, tmpbuf, state->bucket);
1688 seq_printf(seq, "%-127s\n", tmpbuf);
1693 /* ------------------------------------------------------------------------ */
1694 static struct file_operations udp4_seq_fops;
1695 static struct udp_seq_afinfo udp4_seq_afinfo = {
1696 .owner = THIS_MODULE,
1699 .hashtable = udp_hash,
1700 .seq_show = udp4_seq_show,
1701 .seq_fops = &udp4_seq_fops,
1704 static int udp4_proc_init_net(struct net *net)
1706 return udp_proc_register(net, &udp4_seq_afinfo);
1709 static void udp4_proc_exit_net(struct net *net)
1711 udp_proc_unregister(net, &udp4_seq_afinfo);
1714 static struct pernet_operations udp4_net_ops = {
1715 .init = udp4_proc_init_net,
1716 .exit = udp4_proc_exit_net,
1719 int __init udp4_proc_init(void)
1721 return register_pernet_subsys(&udp4_net_ops);
1724 void udp4_proc_exit(void)
1726 unregister_pernet_subsys(&udp4_net_ops);
1728 #endif /* CONFIG_PROC_FS */
1730 void __init udp_init(void)
1732 unsigned long limit;
1734 /* Set the pressure threshold up by the same strategy of TCP. It is a
1735 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
1736 * toward zero with the amount of memory, with a floor of 128 pages.
1738 limit = min(nr_all_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
1739 limit = (limit * (nr_all_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
1740 limit = max(limit, 128UL);
1741 sysctl_udp_mem[0] = limit / 4 * 3;
1742 sysctl_udp_mem[1] = limit;
1743 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
1745 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
1746 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
1749 EXPORT_SYMBOL(udp_disconnect);
1750 EXPORT_SYMBOL(udp_hash);
1751 EXPORT_SYMBOL(udp_hash_lock);
1752 EXPORT_SYMBOL(udp_ioctl);
1753 EXPORT_SYMBOL(udp_prot);
1754 EXPORT_SYMBOL(udp_sendmsg);
1755 EXPORT_SYMBOL(udp_lib_getsockopt);
1756 EXPORT_SYMBOL(udp_lib_setsockopt);
1757 EXPORT_SYMBOL(udp_poll);
1758 EXPORT_SYMBOL(udp_lib_get_port);
1760 #ifdef CONFIG_PROC_FS
1761 EXPORT_SYMBOL(udp_proc_register);
1762 EXPORT_SYMBOL(udp_proc_unregister);