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 * @udptable: hash list table, must be of UDP_HTABLE_SIZE
151 * @saddr_comp: AF-dependent comparison of bound local IP addresses
153 int __udp_lib_get_port(struct sock *sk, unsigned short snum,
154 struct hlist_head udptable[],
155 int (*saddr_comp)(const struct sock *sk1,
156 const struct sock *sk2 ) )
158 struct hlist_node *node;
159 struct hlist_head *head;
162 struct net *net = sk->sk_net;
164 write_lock_bh(&udp_hash_lock);
167 int i, low, high, remaining;
168 unsigned rover, best, best_size_so_far;
170 inet_get_local_port_range(&low, &high);
171 remaining = (high - low) + 1;
173 best_size_so_far = UINT_MAX;
174 best = rover = net_random() % remaining + low;
176 /* 1st pass: look for empty (or shortest) hash chain */
177 for (i = 0; i < UDP_HTABLE_SIZE; i++) {
180 head = &udptable[rover & (UDP_HTABLE_SIZE - 1)];
181 if (hlist_empty(head))
184 sk_for_each(sk2, node, head) {
185 if (++size >= best_size_so_far)
188 best_size_so_far = size;
191 /* fold back if end of range */
193 rover = low + ((rover - low)
194 & (UDP_HTABLE_SIZE - 1));
199 /* 2nd pass: find hole in shortest hash chain */
201 for (i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++) {
202 if (! __udp_lib_lport_inuse(net, rover, udptable))
204 rover += UDP_HTABLE_SIZE;
206 rover = low + ((rover - low)
207 & (UDP_HTABLE_SIZE - 1));
211 /* All ports in use! */
217 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
219 sk_for_each(sk2, node, head)
220 if (sk2->sk_hash == snum &&
222 sk2->sk_net == net &&
223 (!sk2->sk_reuse || !sk->sk_reuse) &&
224 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
225 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
226 (*saddr_comp)(sk, sk2) )
230 inet_sk(sk)->num = snum;
232 if (sk_unhashed(sk)) {
233 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
234 sk_add_node(sk, head);
235 sock_prot_inuse_add(sk->sk_prot, 1);
239 write_unlock_bh(&udp_hash_lock);
243 int udp_get_port(struct sock *sk, unsigned short snum,
244 int (*scmp)(const struct sock *, const struct sock *))
246 return __udp_lib_get_port(sk, snum, udp_hash, scmp);
249 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
251 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
253 return ( !ipv6_only_sock(sk2) &&
254 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
255 inet1->rcv_saddr == inet2->rcv_saddr ));
258 static inline int udp_v4_get_port(struct sock *sk, unsigned short snum)
260 return udp_get_port(sk, snum, ipv4_rcv_saddr_equal);
263 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
264 * harder than this. -DaveM
266 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
267 __be16 sport, __be32 daddr, __be16 dport,
268 int dif, struct hlist_head udptable[])
270 struct sock *sk, *result = NULL;
271 struct hlist_node *node;
272 unsigned short hnum = ntohs(dport);
275 read_lock(&udp_hash_lock);
276 sk_for_each(sk, node, &udptable[hnum & (UDP_HTABLE_SIZE - 1)]) {
277 struct inet_sock *inet = inet_sk(sk);
279 if (sk->sk_net == net && sk->sk_hash == hnum &&
280 !ipv6_only_sock(sk)) {
281 int score = (sk->sk_family == PF_INET ? 1 : 0);
282 if (inet->rcv_saddr) {
283 if (inet->rcv_saddr != daddr)
288 if (inet->daddr != saddr)
293 if (inet->dport != sport)
297 if (sk->sk_bound_dev_if) {
298 if (sk->sk_bound_dev_if != dif)
305 } else if (score > badness) {
313 read_unlock(&udp_hash_lock);
317 static inline struct sock *udp_v4_mcast_next(struct sock *sk,
318 __be16 loc_port, __be32 loc_addr,
319 __be16 rmt_port, __be32 rmt_addr,
322 struct hlist_node *node;
324 unsigned short hnum = ntohs(loc_port);
326 sk_for_each_from(s, node) {
327 struct inet_sock *inet = inet_sk(s);
329 if (s->sk_hash != hnum ||
330 (inet->daddr && inet->daddr != rmt_addr) ||
331 (inet->dport != rmt_port && inet->dport) ||
332 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
334 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
336 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
346 * This routine is called by the ICMP module when it gets some
347 * sort of error condition. If err < 0 then the socket should
348 * be closed and the error returned to the user. If err > 0
349 * it's just the icmp type << 8 | icmp code.
350 * Header points to the ip header of the error packet. We move
351 * on past this. Then (as it used to claim before adjustment)
352 * header points to the first 8 bytes of the udp header. We need
353 * to find the appropriate port.
356 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[])
358 struct inet_sock *inet;
359 struct iphdr *iph = (struct iphdr*)skb->data;
360 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
361 const int type = icmp_hdr(skb)->type;
362 const int code = icmp_hdr(skb)->code;
367 sk = __udp4_lib_lookup(skb->dev->nd_net, iph->daddr, uh->dest,
368 iph->saddr, uh->source, skb->dev->ifindex, udptable);
370 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
371 return; /* No socket for error */
380 case ICMP_TIME_EXCEEDED:
383 case ICMP_SOURCE_QUENCH:
385 case ICMP_PARAMETERPROB:
389 case ICMP_DEST_UNREACH:
390 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
391 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
399 if (code <= NR_ICMP_UNREACH) {
400 harderr = icmp_err_convert[code].fatal;
401 err = icmp_err_convert[code].errno;
407 * RFC1122: OK. Passes ICMP errors back to application, as per
410 if (!inet->recverr) {
411 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
414 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
417 sk->sk_error_report(sk);
422 void udp_err(struct sk_buff *skb, u32 info)
424 __udp4_lib_err(skb, info, udp_hash);
428 * Throw away all pending data and cancel the corking. Socket is locked.
430 static void udp_flush_pending_frames(struct sock *sk)
432 struct udp_sock *up = udp_sk(sk);
437 ip_flush_pending_frames(sk);
442 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
443 * @sk: socket we are sending on
444 * @skb: sk_buff containing the filled-in UDP header
445 * (checksum field must be zeroed out)
447 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
448 __be32 src, __be32 dst, int len )
451 struct udphdr *uh = udp_hdr(skb);
454 if (skb_queue_len(&sk->sk_write_queue) == 1) {
456 * Only one fragment on the socket.
458 skb->csum_start = skb_transport_header(skb) - skb->head;
459 skb->csum_offset = offsetof(struct udphdr, check);
460 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
463 * HW-checksum won't work as there are two or more
464 * fragments on the socket so that all csums of sk_buffs
467 offset = skb_transport_offset(skb);
468 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
470 skb->ip_summed = CHECKSUM_NONE;
472 skb_queue_walk(&sk->sk_write_queue, skb) {
473 csum = csum_add(csum, skb->csum);
476 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
478 uh->check = CSUM_MANGLED_0;
483 * Push out all pending data as one UDP datagram. Socket is locked.
485 static int udp_push_pending_frames(struct sock *sk)
487 struct udp_sock *up = udp_sk(sk);
488 struct inet_sock *inet = inet_sk(sk);
489 struct flowi *fl = &inet->cork.fl;
493 int is_udplite = IS_UDPLITE(sk);
496 /* Grab the skbuff where UDP header space exists. */
497 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
501 * Create a UDP header
504 uh->source = fl->fl_ip_sport;
505 uh->dest = fl->fl_ip_dport;
506 uh->len = htons(up->len);
509 if (is_udplite) /* UDP-Lite */
510 csum = udplite_csum_outgoing(sk, skb);
512 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
514 skb->ip_summed = CHECKSUM_NONE;
517 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
519 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
522 } else /* `normal' UDP */
523 csum = udp_csum_outgoing(sk, skb);
525 /* add protocol-dependent pseudo-header */
526 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
527 sk->sk_protocol, csum );
529 uh->check = CSUM_MANGLED_0;
532 err = ip_push_pending_frames(sk);
537 UDP_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, is_udplite);
541 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
544 struct inet_sock *inet = inet_sk(sk);
545 struct udp_sock *up = udp_sk(sk);
547 struct ipcm_cookie ipc;
548 struct rtable *rt = NULL;
551 __be32 daddr, faddr, saddr;
554 int err, is_udplite = IS_UDPLITE(sk);
555 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
556 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
565 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
572 * There are pending frames.
573 * The socket lock must be held while it's corked.
576 if (likely(up->pending)) {
577 if (unlikely(up->pending != AF_INET)) {
585 ulen += sizeof(struct udphdr);
588 * Get and verify the address.
591 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
592 if (msg->msg_namelen < sizeof(*usin))
594 if (usin->sin_family != AF_INET) {
595 if (usin->sin_family != AF_UNSPEC)
596 return -EAFNOSUPPORT;
599 daddr = usin->sin_addr.s_addr;
600 dport = usin->sin_port;
604 if (sk->sk_state != TCP_ESTABLISHED)
605 return -EDESTADDRREQ;
608 /* Open fast path for connected socket.
609 Route will not be used, if at least one option is set.
613 ipc.addr = inet->saddr;
615 ipc.oif = sk->sk_bound_dev_if;
616 if (msg->msg_controllen) {
617 err = ip_cmsg_send(msg, &ipc);
628 ipc.addr = faddr = daddr;
630 if (ipc.opt && ipc.opt->srr) {
633 faddr = ipc.opt->faddr;
636 tos = RT_TOS(inet->tos);
637 if (sock_flag(sk, SOCK_LOCALROUTE) ||
638 (msg->msg_flags & MSG_DONTROUTE) ||
639 (ipc.opt && ipc.opt->is_strictroute)) {
644 if (ipv4_is_multicast(daddr)) {
646 ipc.oif = inet->mc_index;
648 saddr = inet->mc_addr;
653 rt = (struct rtable*)sk_dst_check(sk, 0);
656 struct flowi fl = { .oif = ipc.oif,
661 .proto = sk->sk_protocol,
663 { .sport = inet->sport,
664 .dport = dport } } };
665 security_sk_classify_flow(sk, &fl);
666 err = ip_route_output_flow(&init_net, &rt, &fl, sk, 1);
668 if (err == -ENETUNREACH)
669 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
674 if ((rt->rt_flags & RTCF_BROADCAST) &&
675 !sock_flag(sk, SOCK_BROADCAST))
678 sk_dst_set(sk, dst_clone(&rt->u.dst));
681 if (msg->msg_flags&MSG_CONFIRM)
687 daddr = ipc.addr = rt->rt_dst;
690 if (unlikely(up->pending)) {
691 /* The socket is already corked while preparing it. */
692 /* ... which is an evident application bug. --ANK */
695 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
700 * Now cork the socket to pend data.
702 inet->cork.fl.fl4_dst = daddr;
703 inet->cork.fl.fl_ip_dport = dport;
704 inet->cork.fl.fl4_src = saddr;
705 inet->cork.fl.fl_ip_sport = inet->sport;
706 up->pending = AF_INET;
710 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
711 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
712 sizeof(struct udphdr), &ipc, rt,
713 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
715 udp_flush_pending_frames(sk);
717 err = udp_push_pending_frames(sk);
718 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
729 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
730 * ENOBUFS might not be good (it's not tunable per se), but otherwise
731 * we don't have a good statistic (IpOutDiscards but it can be too many
732 * things). We could add another new stat but at least for now that
733 * seems like overkill.
735 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
736 UDP_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite);
741 dst_confirm(&rt->u.dst);
742 if (!(msg->msg_flags&MSG_PROBE) || len)
743 goto back_from_confirm;
748 int udp_sendpage(struct sock *sk, struct page *page, int offset,
749 size_t size, int flags)
751 struct udp_sock *up = udp_sk(sk);
755 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
757 /* Call udp_sendmsg to specify destination address which
758 * sendpage interface can't pass.
759 * This will succeed only when the socket is connected.
761 ret = udp_sendmsg(NULL, sk, &msg, 0);
768 if (unlikely(!up->pending)) {
771 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
775 ret = ip_append_page(sk, page, offset, size, flags);
776 if (ret == -EOPNOTSUPP) {
778 return sock_no_sendpage(sk->sk_socket, page, offset,
782 udp_flush_pending_frames(sk);
787 if (!(up->corkflag || (flags&MSG_MORE)))
788 ret = udp_push_pending_frames(sk);
797 * IOCTL requests applicable to the UDP protocol
800 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
805 int amount = atomic_read(&sk->sk_wmem_alloc);
806 return put_user(amount, (int __user *)arg);
812 unsigned long amount;
815 spin_lock_bh(&sk->sk_receive_queue.lock);
816 skb = skb_peek(&sk->sk_receive_queue);
819 * We will only return the amount
820 * of this packet since that is all
823 amount = skb->len - sizeof(struct udphdr);
825 spin_unlock_bh(&sk->sk_receive_queue.lock);
826 return put_user(amount, (int __user *)arg);
837 * This should be easy, if there is something there we
838 * return it, otherwise we block.
841 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
842 size_t len, int noblock, int flags, int *addr_len)
844 struct inet_sock *inet = inet_sk(sk);
845 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
847 unsigned int ulen, copied;
850 int is_udplite = IS_UDPLITE(sk);
853 * Check any passed addresses
856 *addr_len=sizeof(*sin);
858 if (flags & MSG_ERRQUEUE)
859 return ip_recv_error(sk, msg, len);
862 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
867 ulen = skb->len - sizeof(struct udphdr);
871 else if (copied < ulen)
872 msg->msg_flags |= MSG_TRUNC;
875 * If checksum is needed at all, try to do it while copying the
876 * data. If the data is truncated, or if we only want a partial
877 * coverage checksum (UDP-Lite), do it before the copy.
880 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
881 if (udp_lib_checksum_complete(skb))
885 if (skb_csum_unnecessary(skb))
886 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
887 msg->msg_iov, copied );
889 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
899 UDP_INC_STATS_USER(UDP_MIB_INDATAGRAMS, is_udplite);
901 sock_recv_timestamp(msg, sk, skb);
903 /* Copy the address. */
906 sin->sin_family = AF_INET;
907 sin->sin_port = udp_hdr(skb)->source;
908 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
909 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
911 if (inet->cmsg_flags)
912 ip_cmsg_recv(msg, skb);
915 if (flags & MSG_TRUNC)
920 skb_free_datagram(sk, skb);
927 if (!skb_kill_datagram(sk, skb, flags))
928 UDP_INC_STATS_USER(UDP_MIB_INERRORS, is_udplite);
937 int udp_disconnect(struct sock *sk, int flags)
939 struct inet_sock *inet = inet_sk(sk);
941 * 1003.1g - break association.
944 sk->sk_state = TCP_CLOSE;
947 sk->sk_bound_dev_if = 0;
948 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
949 inet_reset_saddr(sk);
951 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
952 sk->sk_prot->unhash(sk);
962 * >0: "udp encap" protocol resubmission
964 * Note that in the success and error cases, the skb is assumed to
965 * have either been requeued or freed.
967 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
969 struct udp_sock *up = udp_sk(sk);
971 int is_udplite = IS_UDPLITE(sk);
974 * Charge it to the socket, dropping if the queue is full.
976 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
980 if (up->encap_type) {
982 * This is an encapsulation socket so pass the skb to
983 * the socket's udp_encap_rcv() hook. Otherwise, just
984 * fall through and pass this up the UDP socket.
985 * up->encap_rcv() returns the following value:
986 * =0 if skb was successfully passed to the encap
987 * handler or was discarded by it.
988 * >0 if skb should be passed on to UDP.
989 * <0 if skb should be resubmitted as proto -N
992 /* if we're overly short, let UDP handle it */
993 if (skb->len > sizeof(struct udphdr) &&
994 up->encap_rcv != NULL) {
997 ret = (*up->encap_rcv)(sk, skb);
999 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS,
1005 /* FALLTHROUGH -- it's a UDP Packet */
1009 * UDP-Lite specific tests, ignored on UDP sockets
1011 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1014 * MIB statistics other than incrementing the error count are
1015 * disabled for the following two types of errors: these depend
1016 * on the application settings, not on the functioning of the
1017 * protocol stack as such.
1019 * RFC 3828 here recommends (sec 3.3): "There should also be a
1020 * way ... to ... at least let the receiving application block
1021 * delivery of packets with coverage values less than a value
1022 * provided by the application."
1024 if (up->pcrlen == 0) { /* full coverage was set */
1025 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1026 "%d while full coverage %d requested\n",
1027 UDP_SKB_CB(skb)->cscov, skb->len);
1030 /* The next case involves violating the min. coverage requested
1031 * by the receiver. This is subtle: if receiver wants x and x is
1032 * greater than the buffersize/MTU then receiver will complain
1033 * that it wants x while sender emits packets of smaller size y.
1034 * Therefore the above ...()->partial_cov statement is essential.
1036 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1037 LIMIT_NETDEBUG(KERN_WARNING
1038 "UDPLITE: coverage %d too small, need min %d\n",
1039 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1044 if (sk->sk_filter) {
1045 if (udp_lib_checksum_complete(skb))
1049 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
1050 /* Note that an ENOMEM error is charged twice */
1052 UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, is_udplite);
1059 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite);
1065 * Multicasts and broadcasts go to each listener.
1067 * Note: called only from the BH handler context,
1068 * so we don't need to lock the hashes.
1070 static int __udp4_lib_mcast_deliver(struct sk_buff *skb,
1072 __be32 saddr, __be32 daddr,
1073 struct hlist_head udptable[])
1078 read_lock(&udp_hash_lock);
1079 sk = sk_head(&udptable[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]);
1080 dif = skb->dev->ifindex;
1081 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
1083 struct sock *sknext = NULL;
1086 struct sk_buff *skb1 = skb;
1088 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr,
1089 uh->source, saddr, dif);
1091 skb1 = skb_clone(skb, GFP_ATOMIC);
1096 bh_lock_sock_nested(sk);
1097 if (!sock_owned_by_user(sk))
1098 ret = udp_queue_rcv_skb(sk, skb1);
1100 sk_add_backlog(sk, skb1);
1104 /* we should probably re-process instead
1105 * of dropping packets here. */
1112 read_unlock(&udp_hash_lock);
1116 /* Initialize UDP checksum. If exited with zero value (success),
1117 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1118 * Otherwise, csum completion requires chacksumming packet body,
1119 * including udp header and folding it to skb->csum.
1121 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1124 const struct iphdr *iph;
1127 UDP_SKB_CB(skb)->partial_cov = 0;
1128 UDP_SKB_CB(skb)->cscov = skb->len;
1130 if (proto == IPPROTO_UDPLITE) {
1131 err = udplite_checksum_init(skb, uh);
1137 if (uh->check == 0) {
1138 skb->ip_summed = CHECKSUM_UNNECESSARY;
1139 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1140 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1142 skb->ip_summed = CHECKSUM_UNNECESSARY;
1144 if (!skb_csum_unnecessary(skb))
1145 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1146 skb->len, proto, 0);
1147 /* Probably, we should checksum udp header (it should be in cache
1148 * in any case) and data in tiny packets (< rx copybreak).
1155 * All we need to do is get the socket, and then do a checksum.
1158 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
1162 struct udphdr *uh = udp_hdr(skb);
1163 unsigned short ulen;
1164 struct rtable *rt = (struct rtable*)skb->dst;
1165 __be32 saddr = ip_hdr(skb)->saddr;
1166 __be32 daddr = ip_hdr(skb)->daddr;
1169 * Validate the packet.
1171 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1172 goto drop; /* No space for header. */
1174 ulen = ntohs(uh->len);
1175 if (ulen > skb->len)
1178 if (proto == IPPROTO_UDP) {
1179 /* UDP validates ulen. */
1180 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1185 if (udp4_csum_init(skb, uh, proto))
1188 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1189 return __udp4_lib_mcast_deliver(skb, uh, saddr, daddr, udptable);
1191 sk = __udp4_lib_lookup(skb->dev->nd_net, saddr, uh->source, daddr,
1192 uh->dest, inet_iif(skb), udptable);
1196 bh_lock_sock_nested(sk);
1197 if (!sock_owned_by_user(sk))
1198 ret = udp_queue_rcv_skb(sk, skb);
1200 sk_add_backlog(sk, skb);
1204 /* a return value > 0 means to resubmit the input, but
1205 * it wants the return to be -protocol, or 0
1212 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1216 /* No socket. Drop packet silently, if checksum is wrong */
1217 if (udp_lib_checksum_complete(skb))
1220 UDP_INC_STATS_BH(UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1221 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1224 * Hmm. We got an UDP packet to a port to which we
1225 * don't wanna listen. Ignore it.
1231 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %u.%u.%u.%u:%u %d/%d to %u.%u.%u.%u:%u\n",
1232 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1243 * RFC1122: OK. Discards the bad packet silently (as far as
1244 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1246 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %d.%d.%d.%d:%d to %d.%d.%d.%d:%d ulen %d\n",
1247 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1254 UDP_INC_STATS_BH(UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1259 int udp_rcv(struct sk_buff *skb)
1261 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP);
1264 int udp_destroy_sock(struct sock *sk)
1267 udp_flush_pending_frames(sk);
1273 * Socket option code for UDP
1275 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1276 char __user *optval, int optlen,
1277 int (*push_pending_frames)(struct sock *))
1279 struct udp_sock *up = udp_sk(sk);
1282 int is_udplite = IS_UDPLITE(sk);
1284 if (optlen<sizeof(int))
1287 if (get_user(val, (int __user *)optval))
1297 (*push_pending_frames)(sk);
1305 case UDP_ENCAP_ESPINUDP:
1306 case UDP_ENCAP_ESPINUDP_NON_IKE:
1307 up->encap_rcv = xfrm4_udp_encap_rcv;
1309 case UDP_ENCAP_L2TPINUDP:
1310 up->encap_type = val;
1319 * UDP-Lite's partial checksum coverage (RFC 3828).
1321 /* The sender sets actual checksum coverage length via this option.
1322 * The case coverage > packet length is handled by send module. */
1323 case UDPLITE_SEND_CSCOV:
1324 if (!is_udplite) /* Disable the option on UDP sockets */
1325 return -ENOPROTOOPT;
1326 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1329 up->pcflag |= UDPLITE_SEND_CC;
1332 /* The receiver specifies a minimum checksum coverage value. To make
1333 * sense, this should be set to at least 8 (as done below). If zero is
1334 * used, this again means full checksum coverage. */
1335 case UDPLITE_RECV_CSCOV:
1336 if (!is_udplite) /* Disable the option on UDP sockets */
1337 return -ENOPROTOOPT;
1338 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1341 up->pcflag |= UDPLITE_RECV_CC;
1352 int udp_setsockopt(struct sock *sk, int level, int optname,
1353 char __user *optval, int optlen)
1355 if (level == SOL_UDP || level == SOL_UDPLITE)
1356 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1357 udp_push_pending_frames);
1358 return ip_setsockopt(sk, level, optname, optval, optlen);
1361 #ifdef CONFIG_COMPAT
1362 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1363 char __user *optval, int optlen)
1365 if (level == SOL_UDP || level == SOL_UDPLITE)
1366 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1367 udp_push_pending_frames);
1368 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1372 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1373 char __user *optval, int __user *optlen)
1375 struct udp_sock *up = udp_sk(sk);
1378 if (get_user(len,optlen))
1381 len = min_t(unsigned int, len, sizeof(int));
1392 val = up->encap_type;
1395 /* The following two cannot be changed on UDP sockets, the return is
1396 * always 0 (which corresponds to the full checksum coverage of UDP). */
1397 case UDPLITE_SEND_CSCOV:
1401 case UDPLITE_RECV_CSCOV:
1406 return -ENOPROTOOPT;
1409 if (put_user(len, optlen))
1411 if (copy_to_user(optval, &val,len))
1416 int udp_getsockopt(struct sock *sk, int level, int optname,
1417 char __user *optval, int __user *optlen)
1419 if (level == SOL_UDP || level == SOL_UDPLITE)
1420 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1421 return ip_getsockopt(sk, level, optname, optval, optlen);
1424 #ifdef CONFIG_COMPAT
1425 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1426 char __user *optval, int __user *optlen)
1428 if (level == SOL_UDP || level == SOL_UDPLITE)
1429 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1430 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1434 * udp_poll - wait for a UDP event.
1435 * @file - file struct
1437 * @wait - poll table
1439 * This is same as datagram poll, except for the special case of
1440 * blocking sockets. If application is using a blocking fd
1441 * and a packet with checksum error is in the queue;
1442 * then it could get return from select indicating data available
1443 * but then block when reading it. Add special case code
1444 * to work around these arguably broken applications.
1446 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1448 unsigned int mask = datagram_poll(file, sock, wait);
1449 struct sock *sk = sock->sk;
1450 int is_lite = IS_UDPLITE(sk);
1452 /* Check for false positives due to checksum errors */
1453 if ( (mask & POLLRDNORM) &&
1454 !(file->f_flags & O_NONBLOCK) &&
1455 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1456 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1457 struct sk_buff *skb;
1459 spin_lock_bh(&rcvq->lock);
1460 while ((skb = skb_peek(rcvq)) != NULL &&
1461 udp_lib_checksum_complete(skb)) {
1462 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_lite);
1463 __skb_unlink(skb, rcvq);
1466 spin_unlock_bh(&rcvq->lock);
1468 /* nothing to see, move along */
1470 mask &= ~(POLLIN | POLLRDNORM);
1477 DEFINE_PROTO_INUSE(udp)
1479 struct proto udp_prot = {
1481 .owner = THIS_MODULE,
1482 .close = udp_lib_close,
1483 .connect = ip4_datagram_connect,
1484 .disconnect = udp_disconnect,
1486 .destroy = udp_destroy_sock,
1487 .setsockopt = udp_setsockopt,
1488 .getsockopt = udp_getsockopt,
1489 .sendmsg = udp_sendmsg,
1490 .recvmsg = udp_recvmsg,
1491 .sendpage = udp_sendpage,
1492 .backlog_rcv = udp_queue_rcv_skb,
1493 .hash = udp_lib_hash,
1494 .unhash = udp_lib_unhash,
1495 .get_port = udp_v4_get_port,
1496 .memory_allocated = &udp_memory_allocated,
1497 .sysctl_mem = sysctl_udp_mem,
1498 .sysctl_wmem = &sysctl_udp_wmem_min,
1499 .sysctl_rmem = &sysctl_udp_rmem_min,
1500 .obj_size = sizeof(struct udp_sock),
1501 #ifdef CONFIG_COMPAT
1502 .compat_setsockopt = compat_udp_setsockopt,
1503 .compat_getsockopt = compat_udp_getsockopt,
1505 REF_PROTO_INUSE(udp)
1508 /* ------------------------------------------------------------------------ */
1509 #ifdef CONFIG_PROC_FS
1511 static struct sock *udp_get_first(struct seq_file *seq)
1514 struct udp_iter_state *state = seq->private;
1516 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1517 struct hlist_node *node;
1518 sk_for_each(sk, node, state->hashtable + state->bucket) {
1519 if (sk->sk_family == state->family)
1528 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1530 struct udp_iter_state *state = seq->private;
1536 } while (sk && sk->sk_family != state->family);
1538 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) {
1539 sk = sk_head(state->hashtable + state->bucket);
1545 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1547 struct sock *sk = udp_get_first(seq);
1550 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1552 return pos ? NULL : sk;
1555 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1556 __acquires(udp_hash_lock)
1558 read_lock(&udp_hash_lock);
1559 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
1562 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1566 if (v == SEQ_START_TOKEN)
1567 sk = udp_get_idx(seq, 0);
1569 sk = udp_get_next(seq, v);
1575 static void udp_seq_stop(struct seq_file *seq, void *v)
1576 __releases(udp_hash_lock)
1578 read_unlock(&udp_hash_lock);
1581 static int udp_seq_open(struct inode *inode, struct file *file)
1583 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1584 struct seq_file *seq;
1586 struct udp_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
1590 s->family = afinfo->family;
1591 s->hashtable = afinfo->hashtable;
1592 s->seq_ops.start = udp_seq_start;
1593 s->seq_ops.next = udp_seq_next;
1594 s->seq_ops.show = afinfo->seq_show;
1595 s->seq_ops.stop = udp_seq_stop;
1597 rc = seq_open(file, &s->seq_ops);
1601 seq = file->private_data;
1610 /* ------------------------------------------------------------------------ */
1611 int udp_proc_register(struct udp_seq_afinfo *afinfo)
1613 struct proc_dir_entry *p;
1618 afinfo->seq_fops->owner = afinfo->owner;
1619 afinfo->seq_fops->open = udp_seq_open;
1620 afinfo->seq_fops->read = seq_read;
1621 afinfo->seq_fops->llseek = seq_lseek;
1622 afinfo->seq_fops->release = seq_release_private;
1624 p = proc_net_fops_create(&init_net, afinfo->name, S_IRUGO, afinfo->seq_fops);
1632 void udp_proc_unregister(struct udp_seq_afinfo *afinfo)
1636 proc_net_remove(&init_net, afinfo->name);
1637 memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops));
1640 /* ------------------------------------------------------------------------ */
1641 static void udp4_format_sock(struct sock *sp, char *tmpbuf, int bucket)
1643 struct inet_sock *inet = inet_sk(sp);
1644 __be32 dest = inet->daddr;
1645 __be32 src = inet->rcv_saddr;
1646 __u16 destp = ntohs(inet->dport);
1647 __u16 srcp = ntohs(inet->sport);
1649 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1650 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p",
1651 bucket, src, srcp, dest, destp, sp->sk_state,
1652 atomic_read(&sp->sk_wmem_alloc),
1653 atomic_read(&sp->sk_rmem_alloc),
1654 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1655 atomic_read(&sp->sk_refcnt), sp);
1658 int udp4_seq_show(struct seq_file *seq, void *v)
1660 if (v == SEQ_START_TOKEN)
1661 seq_printf(seq, "%-127s\n",
1662 " sl local_address rem_address st tx_queue "
1663 "rx_queue tr tm->when retrnsmt uid timeout "
1667 struct udp_iter_state *state = seq->private;
1669 udp4_format_sock(v, tmpbuf, state->bucket);
1670 seq_printf(seq, "%-127s\n", tmpbuf);
1675 /* ------------------------------------------------------------------------ */
1676 static struct file_operations udp4_seq_fops;
1677 static struct udp_seq_afinfo udp4_seq_afinfo = {
1678 .owner = THIS_MODULE,
1681 .hashtable = udp_hash,
1682 .seq_show = udp4_seq_show,
1683 .seq_fops = &udp4_seq_fops,
1686 int __init udp4_proc_init(void)
1688 return udp_proc_register(&udp4_seq_afinfo);
1691 void udp4_proc_exit(void)
1693 udp_proc_unregister(&udp4_seq_afinfo);
1695 #endif /* CONFIG_PROC_FS */
1697 void __init udp_init(void)
1699 unsigned long limit;
1701 /* Set the pressure threshold up by the same strategy of TCP. It is a
1702 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
1703 * toward zero with the amount of memory, with a floor of 128 pages.
1705 limit = min(nr_all_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
1706 limit = (limit * (nr_all_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
1707 limit = max(limit, 128UL);
1708 sysctl_udp_mem[0] = limit / 4 * 3;
1709 sysctl_udp_mem[1] = limit;
1710 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
1712 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
1713 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
1716 EXPORT_SYMBOL(udp_disconnect);
1717 EXPORT_SYMBOL(udp_hash);
1718 EXPORT_SYMBOL(udp_hash_lock);
1719 EXPORT_SYMBOL(udp_ioctl);
1720 EXPORT_SYMBOL(udp_get_port);
1721 EXPORT_SYMBOL(udp_prot);
1722 EXPORT_SYMBOL(udp_sendmsg);
1723 EXPORT_SYMBOL(udp_lib_getsockopt);
1724 EXPORT_SYMBOL(udp_lib_setsockopt);
1725 EXPORT_SYMBOL(udp_poll);
1727 #ifdef CONFIG_PROC_FS
1728 EXPORT_SYMBOL(udp_proc_register);
1729 EXPORT_SYMBOL(udp_proc_unregister);