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_stats_in6) __read_mostly;
112 EXPORT_SYMBOL(udp_stats_in6);
114 struct hlist_head udp_hash[UDP_HTABLE_SIZE];
115 DEFINE_RWLOCK(udp_hash_lock);
117 int sysctl_udp_mem[3] __read_mostly;
118 int sysctl_udp_rmem_min __read_mostly;
119 int sysctl_udp_wmem_min __read_mostly;
121 EXPORT_SYMBOL(sysctl_udp_mem);
122 EXPORT_SYMBOL(sysctl_udp_rmem_min);
123 EXPORT_SYMBOL(sysctl_udp_wmem_min);
125 atomic_t udp_memory_allocated;
126 EXPORT_SYMBOL(udp_memory_allocated);
128 static inline int __udp_lib_lport_inuse(struct net *net, __u16 num,
129 const struct hlist_head udptable[])
132 struct hlist_node *node;
134 sk_for_each(sk, node, &udptable[udp_hashfn(net, num)])
135 if (net_eq(sock_net(sk), net) && sk->sk_hash == num)
141 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
143 * @sk: socket struct in question
144 * @snum: port number to look up
145 * @saddr_comp: AF-dependent comparison of bound local IP addresses
147 int udp_lib_get_port(struct sock *sk, unsigned short snum,
148 int (*saddr_comp)(const struct sock *sk1,
149 const struct sock *sk2 ) )
151 struct hlist_head *udptable = sk->sk_prot->h.udp_hash;
152 struct hlist_node *node;
153 struct hlist_head *head;
156 struct net *net = sock_net(sk);
158 write_lock_bh(&udp_hash_lock);
161 int i, low, high, remaining;
162 unsigned rover, best, best_size_so_far;
164 inet_get_local_port_range(&low, &high);
165 remaining = (high - low) + 1;
167 best_size_so_far = UINT_MAX;
168 best = rover = net_random() % remaining + low;
170 /* 1st pass: look for empty (or shortest) hash chain */
171 for (i = 0; i < UDP_HTABLE_SIZE; i++) {
174 head = &udptable[udp_hashfn(net, rover)];
175 if (hlist_empty(head))
178 sk_for_each(sk2, node, head) {
179 if (++size >= best_size_so_far)
182 best_size_so_far = size;
185 /* fold back if end of range */
187 rover = low + ((rover - low)
188 & (UDP_HTABLE_SIZE - 1));
193 /* 2nd pass: find hole in shortest hash chain */
195 for (i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++) {
196 if (! __udp_lib_lport_inuse(net, rover, udptable))
198 rover += UDP_HTABLE_SIZE;
200 rover = low + ((rover - low)
201 & (UDP_HTABLE_SIZE - 1));
205 /* All ports in use! */
211 head = &udptable[udp_hashfn(net, snum)];
213 sk_for_each(sk2, node, head)
214 if (sk2->sk_hash == snum &&
216 net_eq(sock_net(sk2), net) &&
217 (!sk2->sk_reuse || !sk->sk_reuse) &&
218 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
219 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
220 (*saddr_comp)(sk, sk2) )
224 inet_sk(sk)->num = snum;
226 if (sk_unhashed(sk)) {
227 head = &udptable[udp_hashfn(net, snum)];
228 sk_add_node(sk, head);
229 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
233 write_unlock_bh(&udp_hash_lock);
237 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
239 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
241 return ( !ipv6_only_sock(sk2) &&
242 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
243 inet1->rcv_saddr == inet2->rcv_saddr ));
246 int udp_v4_get_port(struct sock *sk, unsigned short snum)
248 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal);
251 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
252 * harder than this. -DaveM
254 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
255 __be16 sport, __be32 daddr, __be16 dport,
256 int dif, struct hlist_head udptable[])
258 struct sock *sk, *result = NULL;
259 struct hlist_node *node;
260 unsigned short hnum = ntohs(dport);
263 read_lock(&udp_hash_lock);
264 sk_for_each(sk, node, &udptable[udp_hashfn(net, hnum)]) {
265 struct inet_sock *inet = inet_sk(sk);
267 if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum &&
268 !ipv6_only_sock(sk)) {
269 int score = (sk->sk_family == PF_INET ? 1 : 0);
270 if (inet->rcv_saddr) {
271 if (inet->rcv_saddr != daddr)
276 if (inet->daddr != saddr)
281 if (inet->dport != sport)
285 if (sk->sk_bound_dev_if) {
286 if (sk->sk_bound_dev_if != dif)
293 } else if (score > badness) {
301 read_unlock(&udp_hash_lock);
305 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
306 __be32 daddr, __be16 dport, int dif)
308 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, udp_hash);
310 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
312 static inline struct sock *udp_v4_mcast_next(struct sock *sk,
313 __be16 loc_port, __be32 loc_addr,
314 __be16 rmt_port, __be32 rmt_addr,
317 struct hlist_node *node;
319 unsigned short hnum = ntohs(loc_port);
321 sk_for_each_from(s, node) {
322 struct inet_sock *inet = inet_sk(s);
324 if (s->sk_hash != hnum ||
325 (inet->daddr && inet->daddr != rmt_addr) ||
326 (inet->dport != rmt_port && inet->dport) ||
327 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
329 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
331 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
341 * This routine is called by the ICMP module when it gets some
342 * sort of error condition. If err < 0 then the socket should
343 * be closed and the error returned to the user. If err > 0
344 * it's just the icmp type << 8 | icmp code.
345 * Header points to the ip header of the error packet. We move
346 * on past this. Then (as it used to claim before adjustment)
347 * header points to the first 8 bytes of the udp header. We need
348 * to find the appropriate port.
351 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[])
353 struct inet_sock *inet;
354 struct iphdr *iph = (struct iphdr*)skb->data;
355 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
356 const int type = icmp_hdr(skb)->type;
357 const int code = icmp_hdr(skb)->code;
361 struct net *net = dev_net(skb->dev);
363 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
364 iph->saddr, uh->source, skb->dev->ifindex, udptable);
366 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
367 return; /* No socket for error */
376 case ICMP_TIME_EXCEEDED:
379 case ICMP_SOURCE_QUENCH:
381 case ICMP_PARAMETERPROB:
385 case ICMP_DEST_UNREACH:
386 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
387 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
395 if (code <= NR_ICMP_UNREACH) {
396 harderr = icmp_err_convert[code].fatal;
397 err = icmp_err_convert[code].errno;
403 * RFC1122: OK. Passes ICMP errors back to application, as per
406 if (!inet->recverr) {
407 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
410 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
413 sk->sk_error_report(sk);
418 void udp_err(struct sk_buff *skb, u32 info)
420 __udp4_lib_err(skb, info, udp_hash);
424 * Throw away all pending data and cancel the corking. Socket is locked.
426 void udp_flush_pending_frames(struct sock *sk)
428 struct udp_sock *up = udp_sk(sk);
433 ip_flush_pending_frames(sk);
436 EXPORT_SYMBOL(udp_flush_pending_frames);
439 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
440 * @sk: socket we are sending on
441 * @skb: sk_buff containing the filled-in UDP header
442 * (checksum field must be zeroed out)
444 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
445 __be32 src, __be32 dst, int len )
448 struct udphdr *uh = udp_hdr(skb);
451 if (skb_queue_len(&sk->sk_write_queue) == 1) {
453 * Only one fragment on the socket.
455 skb->csum_start = skb_transport_header(skb) - skb->head;
456 skb->csum_offset = offsetof(struct udphdr, check);
457 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
460 * HW-checksum won't work as there are two or more
461 * fragments on the socket so that all csums of sk_buffs
464 offset = skb_transport_offset(skb);
465 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
467 skb->ip_summed = CHECKSUM_NONE;
469 skb_queue_walk(&sk->sk_write_queue, skb) {
470 csum = csum_add(csum, skb->csum);
473 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
475 uh->check = CSUM_MANGLED_0;
480 * Push out all pending data as one UDP datagram. Socket is locked.
482 static int udp_push_pending_frames(struct sock *sk)
484 struct udp_sock *up = udp_sk(sk);
485 struct inet_sock *inet = inet_sk(sk);
486 struct flowi *fl = &inet->cork.fl;
490 int is_udplite = IS_UDPLITE(sk);
493 /* Grab the skbuff where UDP header space exists. */
494 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
498 * Create a UDP header
501 uh->source = fl->fl_ip_sport;
502 uh->dest = fl->fl_ip_dport;
503 uh->len = htons(up->len);
506 if (is_udplite) /* UDP-Lite */
507 csum = udplite_csum_outgoing(sk, skb);
509 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
511 skb->ip_summed = CHECKSUM_NONE;
514 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
516 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
519 } else /* `normal' UDP */
520 csum = udp_csum_outgoing(sk, skb);
522 /* add protocol-dependent pseudo-header */
523 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
524 sk->sk_protocol, csum );
526 uh->check = CSUM_MANGLED_0;
529 err = ip_push_pending_frames(sk);
534 UDP_INC_STATS_USER(sock_net(sk),
535 UDP_MIB_OUTDATAGRAMS, is_udplite);
539 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
542 struct inet_sock *inet = inet_sk(sk);
543 struct udp_sock *up = udp_sk(sk);
545 struct ipcm_cookie ipc;
546 struct rtable *rt = NULL;
549 __be32 daddr, faddr, saddr;
552 int err, is_udplite = IS_UDPLITE(sk);
553 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
554 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
563 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
570 * There are pending frames.
571 * The socket lock must be held while it's corked.
574 if (likely(up->pending)) {
575 if (unlikely(up->pending != AF_INET)) {
583 ulen += sizeof(struct udphdr);
586 * Get and verify the address.
589 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
590 if (msg->msg_namelen < sizeof(*usin))
592 if (usin->sin_family != AF_INET) {
593 if (usin->sin_family != AF_UNSPEC)
594 return -EAFNOSUPPORT;
597 daddr = usin->sin_addr.s_addr;
598 dport = usin->sin_port;
602 if (sk->sk_state != TCP_ESTABLISHED)
603 return -EDESTADDRREQ;
606 /* Open fast path for connected socket.
607 Route will not be used, if at least one option is set.
611 ipc.addr = inet->saddr;
613 ipc.oif = sk->sk_bound_dev_if;
614 if (msg->msg_controllen) {
615 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
626 ipc.addr = faddr = daddr;
628 if (ipc.opt && ipc.opt->srr) {
631 faddr = ipc.opt->faddr;
634 tos = RT_TOS(inet->tos);
635 if (sock_flag(sk, SOCK_LOCALROUTE) ||
636 (msg->msg_flags & MSG_DONTROUTE) ||
637 (ipc.opt && ipc.opt->is_strictroute)) {
642 if (ipv4_is_multicast(daddr)) {
644 ipc.oif = inet->mc_index;
646 saddr = inet->mc_addr;
651 rt = (struct rtable*)sk_dst_check(sk, 0);
654 struct flowi fl = { .oif = ipc.oif,
659 .proto = sk->sk_protocol,
661 { .sport = inet->sport,
662 .dport = dport } } };
663 struct net *net = sock_net(sk);
665 security_sk_classify_flow(sk, &fl);
666 err = ip_route_output_flow(net, &rt, &fl, sk, 1);
668 if (err == -ENETUNREACH)
669 IP_INC_STATS_BH(net, 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(sock_net(sk),
737 UDP_MIB_SNDBUFERRORS, is_udplite);
742 dst_confirm(&rt->u.dst);
743 if (!(msg->msg_flags&MSG_PROBE) || len)
744 goto back_from_confirm;
749 int udp_sendpage(struct sock *sk, struct page *page, int offset,
750 size_t size, int flags)
752 struct udp_sock *up = udp_sk(sk);
756 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
758 /* Call udp_sendmsg to specify destination address which
759 * sendpage interface can't pass.
760 * This will succeed only when the socket is connected.
762 ret = udp_sendmsg(NULL, sk, &msg, 0);
769 if (unlikely(!up->pending)) {
772 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
776 ret = ip_append_page(sk, page, offset, size, flags);
777 if (ret == -EOPNOTSUPP) {
779 return sock_no_sendpage(sk->sk_socket, page, offset,
783 udp_flush_pending_frames(sk);
788 if (!(up->corkflag || (flags&MSG_MORE)))
789 ret = udp_push_pending_frames(sk);
798 * IOCTL requests applicable to the UDP protocol
801 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
806 int amount = atomic_read(&sk->sk_wmem_alloc);
807 return put_user(amount, (int __user *)arg);
813 unsigned long amount;
816 spin_lock_bh(&sk->sk_receive_queue.lock);
817 skb = skb_peek(&sk->sk_receive_queue);
820 * We will only return the amount
821 * of this packet since that is all
824 amount = skb->len - sizeof(struct udphdr);
826 spin_unlock_bh(&sk->sk_receive_queue.lock);
827 return put_user(amount, (int __user *)arg);
838 * This should be easy, if there is something there we
839 * return it, otherwise we block.
842 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
843 size_t len, int noblock, int flags, int *addr_len)
845 struct inet_sock *inet = inet_sk(sk);
846 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
848 unsigned int ulen, copied;
851 int is_udplite = IS_UDPLITE(sk);
854 * Check any passed addresses
857 *addr_len=sizeof(*sin);
859 if (flags & MSG_ERRQUEUE)
860 return ip_recv_error(sk, msg, len);
863 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
868 ulen = skb->len - sizeof(struct udphdr);
872 else if (copied < ulen)
873 msg->msg_flags |= MSG_TRUNC;
876 * If checksum is needed at all, try to do it while copying the
877 * data. If the data is truncated, or if we only want a partial
878 * coverage checksum (UDP-Lite), do it before the copy.
881 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
882 if (udp_lib_checksum_complete(skb))
886 if (skb_csum_unnecessary(skb))
887 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
888 msg->msg_iov, copied );
890 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
900 UDP_INC_STATS_USER(sock_net(sk),
901 UDP_MIB_INDATAGRAMS, is_udplite);
903 sock_recv_timestamp(msg, sk, skb);
905 /* Copy the address. */
908 sin->sin_family = AF_INET;
909 sin->sin_port = udp_hdr(skb)->source;
910 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
911 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
913 if (inet->cmsg_flags)
914 ip_cmsg_recv(msg, skb);
917 if (flags & MSG_TRUNC)
922 skb_free_datagram(sk, skb);
929 if (!skb_kill_datagram(sk, skb, flags))
930 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
939 int udp_disconnect(struct sock *sk, int flags)
941 struct inet_sock *inet = inet_sk(sk);
943 * 1003.1g - break association.
946 sk->sk_state = TCP_CLOSE;
949 sk->sk_bound_dev_if = 0;
950 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
951 inet_reset_saddr(sk);
953 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
954 sk->sk_prot->unhash(sk);
961 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
963 int is_udplite = IS_UDPLITE(sk);
966 if ((rc = sock_queue_rcv_skb(sk, skb)) < 0) {
967 /* Note that an ENOMEM error is charged twice */
969 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
977 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
985 * >0: "udp encap" protocol resubmission
987 * Note that in the success and error cases, the skb is assumed to
988 * have either been requeued or freed.
990 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
992 struct udp_sock *up = udp_sk(sk);
994 int is_udplite = IS_UDPLITE(sk);
997 * Charge it to the socket, dropping if the queue is full.
999 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1003 if (up->encap_type) {
1005 * This is an encapsulation socket so pass the skb to
1006 * the socket's udp_encap_rcv() hook. Otherwise, just
1007 * fall through and pass this up the UDP socket.
1008 * up->encap_rcv() returns the following value:
1009 * =0 if skb was successfully passed to the encap
1010 * handler or was discarded by it.
1011 * >0 if skb should be passed on to UDP.
1012 * <0 if skb should be resubmitted as proto -N
1015 /* if we're overly short, let UDP handle it */
1016 if (skb->len > sizeof(struct udphdr) &&
1017 up->encap_rcv != NULL) {
1020 ret = (*up->encap_rcv)(sk, skb);
1022 UDP_INC_STATS_BH(sock_net(sk),
1023 UDP_MIB_INDATAGRAMS,
1029 /* FALLTHROUGH -- it's a UDP Packet */
1033 * UDP-Lite specific tests, ignored on UDP sockets
1035 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1038 * MIB statistics other than incrementing the error count are
1039 * disabled for the following two types of errors: these depend
1040 * on the application settings, not on the functioning of the
1041 * protocol stack as such.
1043 * RFC 3828 here recommends (sec 3.3): "There should also be a
1044 * way ... to ... at least let the receiving application block
1045 * delivery of packets with coverage values less than a value
1046 * provided by the application."
1048 if (up->pcrlen == 0) { /* full coverage was set */
1049 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1050 "%d while full coverage %d requested\n",
1051 UDP_SKB_CB(skb)->cscov, skb->len);
1054 /* The next case involves violating the min. coverage requested
1055 * by the receiver. This is subtle: if receiver wants x and x is
1056 * greater than the buffersize/MTU then receiver will complain
1057 * that it wants x while sender emits packets of smaller size y.
1058 * Therefore the above ...()->partial_cov statement is essential.
1060 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1061 LIMIT_NETDEBUG(KERN_WARNING
1062 "UDPLITE: coverage %d too small, need min %d\n",
1063 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1068 if (sk->sk_filter) {
1069 if (udp_lib_checksum_complete(skb))
1076 if (!sock_owned_by_user(sk))
1077 rc = __udp_queue_rcv_skb(sk, skb);
1079 sk_add_backlog(sk, skb);
1085 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1091 * Multicasts and broadcasts go to each listener.
1093 * Note: called only from the BH handler context,
1094 * so we don't need to lock the hashes.
1096 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1098 __be32 saddr, __be32 daddr,
1099 struct hlist_head udptable[])
1104 read_lock(&udp_hash_lock);
1105 sk = sk_head(&udptable[udp_hashfn(net, ntohs(uh->dest))]);
1106 dif = skb->dev->ifindex;
1107 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
1109 struct sock *sknext = NULL;
1112 struct sk_buff *skb1 = skb;
1114 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr,
1115 uh->source, saddr, dif);
1117 skb1 = skb_clone(skb, GFP_ATOMIC);
1120 int ret = udp_queue_rcv_skb(sk, skb1);
1122 /* we should probably re-process instead
1123 * of dropping packets here. */
1130 read_unlock(&udp_hash_lock);
1134 /* Initialize UDP checksum. If exited with zero value (success),
1135 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1136 * Otherwise, csum completion requires chacksumming packet body,
1137 * including udp header and folding it to skb->csum.
1139 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1142 const struct iphdr *iph;
1145 UDP_SKB_CB(skb)->partial_cov = 0;
1146 UDP_SKB_CB(skb)->cscov = skb->len;
1148 if (proto == IPPROTO_UDPLITE) {
1149 err = udplite_checksum_init(skb, uh);
1155 if (uh->check == 0) {
1156 skb->ip_summed = CHECKSUM_UNNECESSARY;
1157 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1158 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1160 skb->ip_summed = CHECKSUM_UNNECESSARY;
1162 if (!skb_csum_unnecessary(skb))
1163 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1164 skb->len, proto, 0);
1165 /* Probably, we should checksum udp header (it should be in cache
1166 * in any case) and data in tiny packets (< rx copybreak).
1173 * All we need to do is get the socket, and then do a checksum.
1176 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
1180 struct udphdr *uh = udp_hdr(skb);
1181 unsigned short ulen;
1182 struct rtable *rt = (struct rtable*)skb->dst;
1183 __be32 saddr = ip_hdr(skb)->saddr;
1184 __be32 daddr = ip_hdr(skb)->daddr;
1185 struct net *net = dev_net(skb->dev);
1188 * Validate the packet.
1190 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1191 goto drop; /* No space for header. */
1193 ulen = ntohs(uh->len);
1194 if (ulen > skb->len)
1197 if (proto == IPPROTO_UDP) {
1198 /* UDP validates ulen. */
1199 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1204 if (udp4_csum_init(skb, uh, proto))
1207 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1208 return __udp4_lib_mcast_deliver(net, skb, uh,
1209 saddr, daddr, udptable);
1211 sk = __udp4_lib_lookup(net, saddr, uh->source, daddr,
1212 uh->dest, inet_iif(skb), udptable);
1215 int ret = udp_queue_rcv_skb(sk, skb);
1218 /* a return value > 0 means to resubmit the input, but
1219 * it wants the return to be -protocol, or 0
1226 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1230 /* No socket. Drop packet silently, if checksum is wrong */
1231 if (udp_lib_checksum_complete(skb))
1234 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1235 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1238 * Hmm. We got an UDP packet to a port to which we
1239 * don't wanna listen. Ignore it.
1245 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From " NIPQUAD_FMT ":%u %d/%d to " NIPQUAD_FMT ":%u\n",
1246 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1257 * RFC1122: OK. Discards the bad packet silently (as far as
1258 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1260 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From " NIPQUAD_FMT ":%u to " NIPQUAD_FMT ":%u ulen %d\n",
1261 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1268 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1273 int udp_rcv(struct sk_buff *skb)
1275 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP);
1278 void udp_destroy_sock(struct sock *sk)
1281 udp_flush_pending_frames(sk);
1286 * Socket option code for UDP
1288 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1289 char __user *optval, int optlen,
1290 int (*push_pending_frames)(struct sock *))
1292 struct udp_sock *up = udp_sk(sk);
1295 int is_udplite = IS_UDPLITE(sk);
1297 if (optlen<sizeof(int))
1300 if (get_user(val, (int __user *)optval))
1310 (*push_pending_frames)(sk);
1318 case UDP_ENCAP_ESPINUDP:
1319 case UDP_ENCAP_ESPINUDP_NON_IKE:
1320 up->encap_rcv = xfrm4_udp_encap_rcv;
1322 case UDP_ENCAP_L2TPINUDP:
1323 up->encap_type = val;
1332 * UDP-Lite's partial checksum coverage (RFC 3828).
1334 /* The sender sets actual checksum coverage length via this option.
1335 * The case coverage > packet length is handled by send module. */
1336 case UDPLITE_SEND_CSCOV:
1337 if (!is_udplite) /* Disable the option on UDP sockets */
1338 return -ENOPROTOOPT;
1339 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1341 else if (val > USHORT_MAX)
1344 up->pcflag |= UDPLITE_SEND_CC;
1347 /* The receiver specifies a minimum checksum coverage value. To make
1348 * sense, this should be set to at least 8 (as done below). If zero is
1349 * used, this again means full checksum coverage. */
1350 case UDPLITE_RECV_CSCOV:
1351 if (!is_udplite) /* Disable the option on UDP sockets */
1352 return -ENOPROTOOPT;
1353 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1355 else if (val > USHORT_MAX)
1358 up->pcflag |= UDPLITE_RECV_CC;
1369 int udp_setsockopt(struct sock *sk, int level, int optname,
1370 char __user *optval, int optlen)
1372 if (level == SOL_UDP || level == SOL_UDPLITE)
1373 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1374 udp_push_pending_frames);
1375 return ip_setsockopt(sk, level, optname, optval, optlen);
1378 #ifdef CONFIG_COMPAT
1379 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1380 char __user *optval, int optlen)
1382 if (level == SOL_UDP || level == SOL_UDPLITE)
1383 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1384 udp_push_pending_frames);
1385 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1389 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1390 char __user *optval, int __user *optlen)
1392 struct udp_sock *up = udp_sk(sk);
1395 if (get_user(len,optlen))
1398 len = min_t(unsigned int, len, sizeof(int));
1409 val = up->encap_type;
1412 /* The following two cannot be changed on UDP sockets, the return is
1413 * always 0 (which corresponds to the full checksum coverage of UDP). */
1414 case UDPLITE_SEND_CSCOV:
1418 case UDPLITE_RECV_CSCOV:
1423 return -ENOPROTOOPT;
1426 if (put_user(len, optlen))
1428 if (copy_to_user(optval, &val,len))
1433 int udp_getsockopt(struct sock *sk, int level, int optname,
1434 char __user *optval, int __user *optlen)
1436 if (level == SOL_UDP || level == SOL_UDPLITE)
1437 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1438 return ip_getsockopt(sk, level, optname, optval, optlen);
1441 #ifdef CONFIG_COMPAT
1442 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1443 char __user *optval, int __user *optlen)
1445 if (level == SOL_UDP || level == SOL_UDPLITE)
1446 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1447 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1451 * udp_poll - wait for a UDP event.
1452 * @file - file struct
1454 * @wait - poll table
1456 * This is same as datagram poll, except for the special case of
1457 * blocking sockets. If application is using a blocking fd
1458 * and a packet with checksum error is in the queue;
1459 * then it could get return from select indicating data available
1460 * but then block when reading it. Add special case code
1461 * to work around these arguably broken applications.
1463 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1465 unsigned int mask = datagram_poll(file, sock, wait);
1466 struct sock *sk = sock->sk;
1467 int is_lite = IS_UDPLITE(sk);
1469 /* Check for false positives due to checksum errors */
1470 if ( (mask & POLLRDNORM) &&
1471 !(file->f_flags & O_NONBLOCK) &&
1472 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1473 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1474 struct sk_buff *skb;
1476 spin_lock_bh(&rcvq->lock);
1477 while ((skb = skb_peek(rcvq)) != NULL &&
1478 udp_lib_checksum_complete(skb)) {
1479 UDP_INC_STATS_BH(sock_net(sk),
1480 UDP_MIB_INERRORS, is_lite);
1481 __skb_unlink(skb, rcvq);
1484 spin_unlock_bh(&rcvq->lock);
1486 /* nothing to see, move along */
1488 mask &= ~(POLLIN | POLLRDNORM);
1495 struct proto udp_prot = {
1497 .owner = THIS_MODULE,
1498 .close = udp_lib_close,
1499 .connect = ip4_datagram_connect,
1500 .disconnect = udp_disconnect,
1502 .destroy = udp_destroy_sock,
1503 .setsockopt = udp_setsockopt,
1504 .getsockopt = udp_getsockopt,
1505 .sendmsg = udp_sendmsg,
1506 .recvmsg = udp_recvmsg,
1507 .sendpage = udp_sendpage,
1508 .backlog_rcv = __udp_queue_rcv_skb,
1509 .hash = udp_lib_hash,
1510 .unhash = udp_lib_unhash,
1511 .get_port = udp_v4_get_port,
1512 .memory_allocated = &udp_memory_allocated,
1513 .sysctl_mem = sysctl_udp_mem,
1514 .sysctl_wmem = &sysctl_udp_wmem_min,
1515 .sysctl_rmem = &sysctl_udp_rmem_min,
1516 .obj_size = sizeof(struct udp_sock),
1517 .h.udp_hash = udp_hash,
1518 #ifdef CONFIG_COMPAT
1519 .compat_setsockopt = compat_udp_setsockopt,
1520 .compat_getsockopt = compat_udp_getsockopt,
1524 /* ------------------------------------------------------------------------ */
1525 #ifdef CONFIG_PROC_FS
1527 static struct sock *udp_get_first(struct seq_file *seq)
1530 struct udp_iter_state *state = seq->private;
1531 struct net *net = seq_file_net(seq);
1533 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1534 struct hlist_node *node;
1535 sk_for_each(sk, node, state->hashtable + state->bucket) {
1536 if (!net_eq(sock_net(sk), net))
1538 if (sk->sk_family == state->family)
1547 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1549 struct udp_iter_state *state = seq->private;
1550 struct net *net = seq_file_net(seq);
1556 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1558 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) {
1559 sk = sk_head(state->hashtable + state->bucket);
1565 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1567 struct sock *sk = udp_get_first(seq);
1570 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1572 return pos ? NULL : sk;
1575 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1576 __acquires(udp_hash_lock)
1578 read_lock(&udp_hash_lock);
1579 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
1582 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1586 if (v == SEQ_START_TOKEN)
1587 sk = udp_get_idx(seq, 0);
1589 sk = udp_get_next(seq, v);
1595 static void udp_seq_stop(struct seq_file *seq, void *v)
1596 __releases(udp_hash_lock)
1598 read_unlock(&udp_hash_lock);
1601 static int udp_seq_open(struct inode *inode, struct file *file)
1603 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1604 struct udp_iter_state *s;
1607 err = seq_open_net(inode, file, &afinfo->seq_ops,
1608 sizeof(struct udp_iter_state));
1612 s = ((struct seq_file *)file->private_data)->private;
1613 s->family = afinfo->family;
1614 s->hashtable = afinfo->hashtable;
1618 /* ------------------------------------------------------------------------ */
1619 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
1621 struct proc_dir_entry *p;
1624 afinfo->seq_fops.open = udp_seq_open;
1625 afinfo->seq_fops.read = seq_read;
1626 afinfo->seq_fops.llseek = seq_lseek;
1627 afinfo->seq_fops.release = seq_release_net;
1629 afinfo->seq_ops.start = udp_seq_start;
1630 afinfo->seq_ops.next = udp_seq_next;
1631 afinfo->seq_ops.stop = udp_seq_stop;
1633 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
1634 &afinfo->seq_fops, afinfo);
1640 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
1642 proc_net_remove(net, afinfo->name);
1645 /* ------------------------------------------------------------------------ */
1646 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
1647 int bucket, int *len)
1649 struct inet_sock *inet = inet_sk(sp);
1650 __be32 dest = inet->daddr;
1651 __be32 src = inet->rcv_saddr;
1652 __u16 destp = ntohs(inet->dport);
1653 __u16 srcp = ntohs(inet->sport);
1655 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
1656 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
1657 bucket, src, srcp, dest, destp, sp->sk_state,
1658 atomic_read(&sp->sk_wmem_alloc),
1659 atomic_read(&sp->sk_rmem_alloc),
1660 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1661 atomic_read(&sp->sk_refcnt), sp,
1662 atomic_read(&sp->sk_drops), len);
1665 int udp4_seq_show(struct seq_file *seq, void *v)
1667 if (v == SEQ_START_TOKEN)
1668 seq_printf(seq, "%-127s\n",
1669 " sl local_address rem_address st tx_queue "
1670 "rx_queue tr tm->when retrnsmt uid timeout "
1671 "inode ref pointer drops");
1673 struct udp_iter_state *state = seq->private;
1676 udp4_format_sock(v, seq, state->bucket, &len);
1677 seq_printf(seq, "%*s\n", 127 - len ,"");
1682 /* ------------------------------------------------------------------------ */
1683 static struct udp_seq_afinfo udp4_seq_afinfo = {
1686 .hashtable = udp_hash,
1688 .owner = THIS_MODULE,
1691 .show = udp4_seq_show,
1695 static int udp4_proc_init_net(struct net *net)
1697 return udp_proc_register(net, &udp4_seq_afinfo);
1700 static void udp4_proc_exit_net(struct net *net)
1702 udp_proc_unregister(net, &udp4_seq_afinfo);
1705 static struct pernet_operations udp4_net_ops = {
1706 .init = udp4_proc_init_net,
1707 .exit = udp4_proc_exit_net,
1710 int __init udp4_proc_init(void)
1712 return register_pernet_subsys(&udp4_net_ops);
1715 void udp4_proc_exit(void)
1717 unregister_pernet_subsys(&udp4_net_ops);
1719 #endif /* CONFIG_PROC_FS */
1721 void __init udp_init(void)
1723 unsigned long limit;
1725 /* Set the pressure threshold up by the same strategy of TCP. It is a
1726 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
1727 * toward zero with the amount of memory, with a floor of 128 pages.
1729 limit = min(nr_all_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
1730 limit = (limit * (nr_all_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
1731 limit = max(limit, 128UL);
1732 sysctl_udp_mem[0] = limit / 4 * 3;
1733 sysctl_udp_mem[1] = limit;
1734 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
1736 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
1737 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
1740 EXPORT_SYMBOL(udp_disconnect);
1741 EXPORT_SYMBOL(udp_hash);
1742 EXPORT_SYMBOL(udp_hash_lock);
1743 EXPORT_SYMBOL(udp_ioctl);
1744 EXPORT_SYMBOL(udp_prot);
1745 EXPORT_SYMBOL(udp_sendmsg);
1746 EXPORT_SYMBOL(udp_lib_getsockopt);
1747 EXPORT_SYMBOL(udp_lib_setsockopt);
1748 EXPORT_SYMBOL(udp_poll);
1749 EXPORT_SYMBOL(udp_lib_get_port);
1751 #ifdef CONFIG_PROC_FS
1752 EXPORT_SYMBOL(udp_proc_register);
1753 EXPORT_SYMBOL(udp_proc_unregister);