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 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
306 __be16 sport, __be16 dport,
307 struct hlist_head udptable[])
309 const struct iphdr *iph = ip_hdr(skb);
311 return __udp4_lib_lookup(dev_net(skb->dst->dev), iph->saddr, sport,
312 iph->daddr, dport, inet_iif(skb),
316 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
317 __be32 daddr, __be16 dport, int dif)
319 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, udp_hash);
321 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
323 static inline struct sock *udp_v4_mcast_next(struct sock *sk,
324 __be16 loc_port, __be32 loc_addr,
325 __be16 rmt_port, __be32 rmt_addr,
328 struct hlist_node *node;
330 unsigned short hnum = ntohs(loc_port);
332 sk_for_each_from(s, node) {
333 struct inet_sock *inet = inet_sk(s);
335 if (s->sk_hash != hnum ||
336 (inet->daddr && inet->daddr != rmt_addr) ||
337 (inet->dport != rmt_port && inet->dport) ||
338 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
340 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
342 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
352 * This routine is called by the ICMP module when it gets some
353 * sort of error condition. If err < 0 then the socket should
354 * be closed and the error returned to the user. If err > 0
355 * it's just the icmp type << 8 | icmp code.
356 * Header points to the ip header of the error packet. We move
357 * on past this. Then (as it used to claim before adjustment)
358 * header points to the first 8 bytes of the udp header. We need
359 * to find the appropriate port.
362 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[])
364 struct inet_sock *inet;
365 struct iphdr *iph = (struct iphdr*)skb->data;
366 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
367 const int type = icmp_hdr(skb)->type;
368 const int code = icmp_hdr(skb)->code;
372 struct net *net = dev_net(skb->dev);
374 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
375 iph->saddr, uh->source, skb->dev->ifindex, udptable);
377 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
378 return; /* No socket for error */
387 case ICMP_TIME_EXCEEDED:
390 case ICMP_SOURCE_QUENCH:
392 case ICMP_PARAMETERPROB:
396 case ICMP_DEST_UNREACH:
397 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
398 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
406 if (code <= NR_ICMP_UNREACH) {
407 harderr = icmp_err_convert[code].fatal;
408 err = icmp_err_convert[code].errno;
414 * RFC1122: OK. Passes ICMP errors back to application, as per
417 if (!inet->recverr) {
418 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
421 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
424 sk->sk_error_report(sk);
429 void udp_err(struct sk_buff *skb, u32 info)
431 __udp4_lib_err(skb, info, udp_hash);
435 * Throw away all pending data and cancel the corking. Socket is locked.
437 void udp_flush_pending_frames(struct sock *sk)
439 struct udp_sock *up = udp_sk(sk);
444 ip_flush_pending_frames(sk);
447 EXPORT_SYMBOL(udp_flush_pending_frames);
450 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
451 * @sk: socket we are sending on
452 * @skb: sk_buff containing the filled-in UDP header
453 * (checksum field must be zeroed out)
455 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
456 __be32 src, __be32 dst, int len )
459 struct udphdr *uh = udp_hdr(skb);
462 if (skb_queue_len(&sk->sk_write_queue) == 1) {
464 * Only one fragment on the socket.
466 skb->csum_start = skb_transport_header(skb) - skb->head;
467 skb->csum_offset = offsetof(struct udphdr, check);
468 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
471 * HW-checksum won't work as there are two or more
472 * fragments on the socket so that all csums of sk_buffs
475 offset = skb_transport_offset(skb);
476 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
478 skb->ip_summed = CHECKSUM_NONE;
480 skb_queue_walk(&sk->sk_write_queue, skb) {
481 csum = csum_add(csum, skb->csum);
484 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
486 uh->check = CSUM_MANGLED_0;
491 * Push out all pending data as one UDP datagram. Socket is locked.
493 static int udp_push_pending_frames(struct sock *sk)
495 struct udp_sock *up = udp_sk(sk);
496 struct inet_sock *inet = inet_sk(sk);
497 struct flowi *fl = &inet->cork.fl;
501 int is_udplite = IS_UDPLITE(sk);
504 /* Grab the skbuff where UDP header space exists. */
505 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
509 * Create a UDP header
512 uh->source = fl->fl_ip_sport;
513 uh->dest = fl->fl_ip_dport;
514 uh->len = htons(up->len);
517 if (is_udplite) /* UDP-Lite */
518 csum = udplite_csum_outgoing(sk, skb);
520 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
522 skb->ip_summed = CHECKSUM_NONE;
525 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
527 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
530 } else /* `normal' UDP */
531 csum = udp_csum_outgoing(sk, skb);
533 /* add protocol-dependent pseudo-header */
534 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
535 sk->sk_protocol, csum );
537 uh->check = CSUM_MANGLED_0;
540 err = ip_push_pending_frames(sk);
545 UDP_INC_STATS_USER(sock_net(sk),
546 UDP_MIB_OUTDATAGRAMS, is_udplite);
550 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
553 struct inet_sock *inet = inet_sk(sk);
554 struct udp_sock *up = udp_sk(sk);
556 struct ipcm_cookie ipc;
557 struct rtable *rt = NULL;
560 __be32 daddr, faddr, saddr;
563 int err, is_udplite = IS_UDPLITE(sk);
564 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
565 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
574 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
581 * There are pending frames.
582 * The socket lock must be held while it's corked.
585 if (likely(up->pending)) {
586 if (unlikely(up->pending != AF_INET)) {
594 ulen += sizeof(struct udphdr);
597 * Get and verify the address.
600 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
601 if (msg->msg_namelen < sizeof(*usin))
603 if (usin->sin_family != AF_INET) {
604 if (usin->sin_family != AF_UNSPEC)
605 return -EAFNOSUPPORT;
608 daddr = usin->sin_addr.s_addr;
609 dport = usin->sin_port;
613 if (sk->sk_state != TCP_ESTABLISHED)
614 return -EDESTADDRREQ;
617 /* Open fast path for connected socket.
618 Route will not be used, if at least one option is set.
622 ipc.addr = inet->saddr;
624 ipc.oif = sk->sk_bound_dev_if;
625 if (msg->msg_controllen) {
626 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
637 ipc.addr = faddr = daddr;
639 if (ipc.opt && ipc.opt->srr) {
642 faddr = ipc.opt->faddr;
645 tos = RT_TOS(inet->tos);
646 if (sock_flag(sk, SOCK_LOCALROUTE) ||
647 (msg->msg_flags & MSG_DONTROUTE) ||
648 (ipc.opt && ipc.opt->is_strictroute)) {
653 if (ipv4_is_multicast(daddr)) {
655 ipc.oif = inet->mc_index;
657 saddr = inet->mc_addr;
662 rt = (struct rtable*)sk_dst_check(sk, 0);
665 struct flowi fl = { .oif = ipc.oif,
670 .proto = sk->sk_protocol,
672 { .sport = inet->sport,
673 .dport = dport } } };
674 struct net *net = sock_net(sk);
676 security_sk_classify_flow(sk, &fl);
677 err = ip_route_output_flow(net, &rt, &fl, sk, 1);
679 if (err == -ENETUNREACH)
680 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
685 if ((rt->rt_flags & RTCF_BROADCAST) &&
686 !sock_flag(sk, SOCK_BROADCAST))
689 sk_dst_set(sk, dst_clone(&rt->u.dst));
692 if (msg->msg_flags&MSG_CONFIRM)
698 daddr = ipc.addr = rt->rt_dst;
701 if (unlikely(up->pending)) {
702 /* The socket is already corked while preparing it. */
703 /* ... which is an evident application bug. --ANK */
706 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
711 * Now cork the socket to pend data.
713 inet->cork.fl.fl4_dst = daddr;
714 inet->cork.fl.fl_ip_dport = dport;
715 inet->cork.fl.fl4_src = saddr;
716 inet->cork.fl.fl_ip_sport = inet->sport;
717 up->pending = AF_INET;
721 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
722 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
723 sizeof(struct udphdr), &ipc, rt,
724 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
726 udp_flush_pending_frames(sk);
728 err = udp_push_pending_frames(sk);
729 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
740 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
741 * ENOBUFS might not be good (it's not tunable per se), but otherwise
742 * we don't have a good statistic (IpOutDiscards but it can be too many
743 * things). We could add another new stat but at least for now that
744 * seems like overkill.
746 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
747 UDP_INC_STATS_USER(sock_net(sk),
748 UDP_MIB_SNDBUFERRORS, is_udplite);
753 dst_confirm(&rt->u.dst);
754 if (!(msg->msg_flags&MSG_PROBE) || len)
755 goto back_from_confirm;
760 int udp_sendpage(struct sock *sk, struct page *page, int offset,
761 size_t size, int flags)
763 struct udp_sock *up = udp_sk(sk);
767 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
769 /* Call udp_sendmsg to specify destination address which
770 * sendpage interface can't pass.
771 * This will succeed only when the socket is connected.
773 ret = udp_sendmsg(NULL, sk, &msg, 0);
780 if (unlikely(!up->pending)) {
783 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
787 ret = ip_append_page(sk, page, offset, size, flags);
788 if (ret == -EOPNOTSUPP) {
790 return sock_no_sendpage(sk->sk_socket, page, offset,
794 udp_flush_pending_frames(sk);
799 if (!(up->corkflag || (flags&MSG_MORE)))
800 ret = udp_push_pending_frames(sk);
809 * IOCTL requests applicable to the UDP protocol
812 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
817 int amount = atomic_read(&sk->sk_wmem_alloc);
818 return put_user(amount, (int __user *)arg);
824 unsigned long amount;
827 spin_lock_bh(&sk->sk_receive_queue.lock);
828 skb = skb_peek(&sk->sk_receive_queue);
831 * We will only return the amount
832 * of this packet since that is all
835 amount = skb->len - sizeof(struct udphdr);
837 spin_unlock_bh(&sk->sk_receive_queue.lock);
838 return put_user(amount, (int __user *)arg);
849 * This should be easy, if there is something there we
850 * return it, otherwise we block.
853 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
854 size_t len, int noblock, int flags, int *addr_len)
856 struct inet_sock *inet = inet_sk(sk);
857 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
859 unsigned int ulen, copied;
862 int is_udplite = IS_UDPLITE(sk);
865 * Check any passed addresses
868 *addr_len=sizeof(*sin);
870 if (flags & MSG_ERRQUEUE)
871 return ip_recv_error(sk, msg, len);
874 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
879 ulen = skb->len - sizeof(struct udphdr);
883 else if (copied < ulen)
884 msg->msg_flags |= MSG_TRUNC;
887 * If checksum is needed at all, try to do it while copying the
888 * data. If the data is truncated, or if we only want a partial
889 * coverage checksum (UDP-Lite), do it before the copy.
892 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
893 if (udp_lib_checksum_complete(skb))
897 if (skb_csum_unnecessary(skb))
898 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
899 msg->msg_iov, copied );
901 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
911 UDP_INC_STATS_USER(sock_net(sk),
912 UDP_MIB_INDATAGRAMS, is_udplite);
914 sock_recv_timestamp(msg, sk, skb);
916 /* Copy the address. */
919 sin->sin_family = AF_INET;
920 sin->sin_port = udp_hdr(skb)->source;
921 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
922 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
924 if (inet->cmsg_flags)
925 ip_cmsg_recv(msg, skb);
928 if (flags & MSG_TRUNC)
933 skb_free_datagram(sk, skb);
940 if (!skb_kill_datagram(sk, skb, flags))
941 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
950 int udp_disconnect(struct sock *sk, int flags)
952 struct inet_sock *inet = inet_sk(sk);
954 * 1003.1g - break association.
957 sk->sk_state = TCP_CLOSE;
960 sk->sk_bound_dev_if = 0;
961 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
962 inet_reset_saddr(sk);
964 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
965 sk->sk_prot->unhash(sk);
972 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
974 int is_udplite = IS_UDPLITE(sk);
977 if ((rc = sock_queue_rcv_skb(sk, skb)) < 0) {
978 /* Note that an ENOMEM error is charged twice */
980 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
988 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
996 * >0: "udp encap" protocol resubmission
998 * Note that in the success and error cases, the skb is assumed to
999 * have either been requeued or freed.
1001 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
1003 struct udp_sock *up = udp_sk(sk);
1005 int is_udplite = IS_UDPLITE(sk);
1008 * Charge it to the socket, dropping if the queue is full.
1010 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1014 if (up->encap_type) {
1016 * This is an encapsulation socket so pass the skb to
1017 * the socket's udp_encap_rcv() hook. Otherwise, just
1018 * fall through and pass this up the UDP socket.
1019 * up->encap_rcv() returns the following value:
1020 * =0 if skb was successfully passed to the encap
1021 * handler or was discarded by it.
1022 * >0 if skb should be passed on to UDP.
1023 * <0 if skb should be resubmitted as proto -N
1026 /* if we're overly short, let UDP handle it */
1027 if (skb->len > sizeof(struct udphdr) &&
1028 up->encap_rcv != NULL) {
1031 ret = (*up->encap_rcv)(sk, skb);
1033 UDP_INC_STATS_BH(sock_net(sk),
1034 UDP_MIB_INDATAGRAMS,
1040 /* FALLTHROUGH -- it's a UDP Packet */
1044 * UDP-Lite specific tests, ignored on UDP sockets
1046 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1049 * MIB statistics other than incrementing the error count are
1050 * disabled for the following two types of errors: these depend
1051 * on the application settings, not on the functioning of the
1052 * protocol stack as such.
1054 * RFC 3828 here recommends (sec 3.3): "There should also be a
1055 * way ... to ... at least let the receiving application block
1056 * delivery of packets with coverage values less than a value
1057 * provided by the application."
1059 if (up->pcrlen == 0) { /* full coverage was set */
1060 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1061 "%d while full coverage %d requested\n",
1062 UDP_SKB_CB(skb)->cscov, skb->len);
1065 /* The next case involves violating the min. coverage requested
1066 * by the receiver. This is subtle: if receiver wants x and x is
1067 * greater than the buffersize/MTU then receiver will complain
1068 * that it wants x while sender emits packets of smaller size y.
1069 * Therefore the above ...()->partial_cov statement is essential.
1071 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1072 LIMIT_NETDEBUG(KERN_WARNING
1073 "UDPLITE: coverage %d too small, need min %d\n",
1074 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1079 if (sk->sk_filter) {
1080 if (udp_lib_checksum_complete(skb))
1087 if (!sock_owned_by_user(sk))
1088 rc = __udp_queue_rcv_skb(sk, skb);
1090 sk_add_backlog(sk, skb);
1096 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1102 * Multicasts and broadcasts go to each listener.
1104 * Note: called only from the BH handler context,
1105 * so we don't need to lock the hashes.
1107 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1109 __be32 saddr, __be32 daddr,
1110 struct hlist_head udptable[])
1115 read_lock(&udp_hash_lock);
1116 sk = sk_head(&udptable[udp_hashfn(net, ntohs(uh->dest))]);
1117 dif = skb->dev->ifindex;
1118 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
1120 struct sock *sknext = NULL;
1123 struct sk_buff *skb1 = skb;
1125 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr,
1126 uh->source, saddr, dif);
1128 skb1 = skb_clone(skb, GFP_ATOMIC);
1131 int ret = udp_queue_rcv_skb(sk, skb1);
1133 /* we should probably re-process instead
1134 * of dropping packets here. */
1141 read_unlock(&udp_hash_lock);
1145 /* Initialize UDP checksum. If exited with zero value (success),
1146 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1147 * Otherwise, csum completion requires chacksumming packet body,
1148 * including udp header and folding it to skb->csum.
1150 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1153 const struct iphdr *iph;
1156 UDP_SKB_CB(skb)->partial_cov = 0;
1157 UDP_SKB_CB(skb)->cscov = skb->len;
1159 if (proto == IPPROTO_UDPLITE) {
1160 err = udplite_checksum_init(skb, uh);
1166 if (uh->check == 0) {
1167 skb->ip_summed = CHECKSUM_UNNECESSARY;
1168 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1169 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1171 skb->ip_summed = CHECKSUM_UNNECESSARY;
1173 if (!skb_csum_unnecessary(skb))
1174 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1175 skb->len, proto, 0);
1176 /* Probably, we should checksum udp header (it should be in cache
1177 * in any case) and data in tiny packets (< rx copybreak).
1184 * All we need to do is get the socket, and then do a checksum.
1187 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
1191 struct udphdr *uh = udp_hdr(skb);
1192 unsigned short ulen;
1193 struct rtable *rt = (struct rtable*)skb->dst;
1194 __be32 saddr = ip_hdr(skb)->saddr;
1195 __be32 daddr = ip_hdr(skb)->daddr;
1196 struct net *net = dev_net(skb->dev);
1199 * Validate the packet.
1201 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1202 goto drop; /* No space for header. */
1204 ulen = ntohs(uh->len);
1205 if (ulen > skb->len)
1208 if (proto == IPPROTO_UDP) {
1209 /* UDP validates ulen. */
1210 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1215 if (udp4_csum_init(skb, uh, proto))
1218 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1219 return __udp4_lib_mcast_deliver(net, skb, uh,
1220 saddr, daddr, udptable);
1222 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1225 int ret = udp_queue_rcv_skb(sk, skb);
1228 /* a return value > 0 means to resubmit the input, but
1229 * it wants the return to be -protocol, or 0
1236 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1240 /* No socket. Drop packet silently, if checksum is wrong */
1241 if (udp_lib_checksum_complete(skb))
1244 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1245 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1248 * Hmm. We got an UDP packet to a port to which we
1249 * don't wanna listen. Ignore it.
1255 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From " NIPQUAD_FMT ":%u %d/%d to " NIPQUAD_FMT ":%u\n",
1256 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1267 * RFC1122: OK. Discards the bad packet silently (as far as
1268 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1270 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From " NIPQUAD_FMT ":%u to " NIPQUAD_FMT ":%u ulen %d\n",
1271 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1278 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1283 int udp_rcv(struct sk_buff *skb)
1285 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP);
1288 void udp_destroy_sock(struct sock *sk)
1291 udp_flush_pending_frames(sk);
1296 * Socket option code for UDP
1298 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1299 char __user *optval, int optlen,
1300 int (*push_pending_frames)(struct sock *))
1302 struct udp_sock *up = udp_sk(sk);
1305 int is_udplite = IS_UDPLITE(sk);
1307 if (optlen<sizeof(int))
1310 if (get_user(val, (int __user *)optval))
1320 (*push_pending_frames)(sk);
1328 case UDP_ENCAP_ESPINUDP:
1329 case UDP_ENCAP_ESPINUDP_NON_IKE:
1330 up->encap_rcv = xfrm4_udp_encap_rcv;
1332 case UDP_ENCAP_L2TPINUDP:
1333 up->encap_type = val;
1342 * UDP-Lite's partial checksum coverage (RFC 3828).
1344 /* The sender sets actual checksum coverage length via this option.
1345 * The case coverage > packet length is handled by send module. */
1346 case UDPLITE_SEND_CSCOV:
1347 if (!is_udplite) /* Disable the option on UDP sockets */
1348 return -ENOPROTOOPT;
1349 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1351 else if (val > USHORT_MAX)
1354 up->pcflag |= UDPLITE_SEND_CC;
1357 /* The receiver specifies a minimum checksum coverage value. To make
1358 * sense, this should be set to at least 8 (as done below). If zero is
1359 * used, this again means full checksum coverage. */
1360 case UDPLITE_RECV_CSCOV:
1361 if (!is_udplite) /* Disable the option on UDP sockets */
1362 return -ENOPROTOOPT;
1363 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1365 else if (val > USHORT_MAX)
1368 up->pcflag |= UDPLITE_RECV_CC;
1379 int 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 ip_setsockopt(sk, level, optname, optval, optlen);
1388 #ifdef CONFIG_COMPAT
1389 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1390 char __user *optval, int optlen)
1392 if (level == SOL_UDP || level == SOL_UDPLITE)
1393 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1394 udp_push_pending_frames);
1395 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1399 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1400 char __user *optval, int __user *optlen)
1402 struct udp_sock *up = udp_sk(sk);
1405 if (get_user(len,optlen))
1408 len = min_t(unsigned int, len, sizeof(int));
1419 val = up->encap_type;
1422 /* The following two cannot be changed on UDP sockets, the return is
1423 * always 0 (which corresponds to the full checksum coverage of UDP). */
1424 case UDPLITE_SEND_CSCOV:
1428 case UDPLITE_RECV_CSCOV:
1433 return -ENOPROTOOPT;
1436 if (put_user(len, optlen))
1438 if (copy_to_user(optval, &val,len))
1443 int udp_getsockopt(struct sock *sk, int level, int optname,
1444 char __user *optval, int __user *optlen)
1446 if (level == SOL_UDP || level == SOL_UDPLITE)
1447 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1448 return ip_getsockopt(sk, level, optname, optval, optlen);
1451 #ifdef CONFIG_COMPAT
1452 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1453 char __user *optval, int __user *optlen)
1455 if (level == SOL_UDP || level == SOL_UDPLITE)
1456 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1457 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1461 * udp_poll - wait for a UDP event.
1462 * @file - file struct
1464 * @wait - poll table
1466 * This is same as datagram poll, except for the special case of
1467 * blocking sockets. If application is using a blocking fd
1468 * and a packet with checksum error is in the queue;
1469 * then it could get return from select indicating data available
1470 * but then block when reading it. Add special case code
1471 * to work around these arguably broken applications.
1473 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1475 unsigned int mask = datagram_poll(file, sock, wait);
1476 struct sock *sk = sock->sk;
1477 int is_lite = IS_UDPLITE(sk);
1479 /* Check for false positives due to checksum errors */
1480 if ( (mask & POLLRDNORM) &&
1481 !(file->f_flags & O_NONBLOCK) &&
1482 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1483 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1484 struct sk_buff *skb;
1486 spin_lock_bh(&rcvq->lock);
1487 while ((skb = skb_peek(rcvq)) != NULL &&
1488 udp_lib_checksum_complete(skb)) {
1489 UDP_INC_STATS_BH(sock_net(sk),
1490 UDP_MIB_INERRORS, is_lite);
1491 __skb_unlink(skb, rcvq);
1494 spin_unlock_bh(&rcvq->lock);
1496 /* nothing to see, move along */
1498 mask &= ~(POLLIN | POLLRDNORM);
1505 struct proto udp_prot = {
1507 .owner = THIS_MODULE,
1508 .close = udp_lib_close,
1509 .connect = ip4_datagram_connect,
1510 .disconnect = udp_disconnect,
1512 .destroy = udp_destroy_sock,
1513 .setsockopt = udp_setsockopt,
1514 .getsockopt = udp_getsockopt,
1515 .sendmsg = udp_sendmsg,
1516 .recvmsg = udp_recvmsg,
1517 .sendpage = udp_sendpage,
1518 .backlog_rcv = __udp_queue_rcv_skb,
1519 .hash = udp_lib_hash,
1520 .unhash = udp_lib_unhash,
1521 .get_port = udp_v4_get_port,
1522 .memory_allocated = &udp_memory_allocated,
1523 .sysctl_mem = sysctl_udp_mem,
1524 .sysctl_wmem = &sysctl_udp_wmem_min,
1525 .sysctl_rmem = &sysctl_udp_rmem_min,
1526 .obj_size = sizeof(struct udp_sock),
1527 .h.udp_hash = udp_hash,
1528 #ifdef CONFIG_COMPAT
1529 .compat_setsockopt = compat_udp_setsockopt,
1530 .compat_getsockopt = compat_udp_getsockopt,
1534 /* ------------------------------------------------------------------------ */
1535 #ifdef CONFIG_PROC_FS
1537 static struct sock *udp_get_first(struct seq_file *seq)
1540 struct udp_iter_state *state = seq->private;
1541 struct net *net = seq_file_net(seq);
1543 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1544 struct hlist_node *node;
1545 sk_for_each(sk, node, state->hashtable + state->bucket) {
1546 if (!net_eq(sock_net(sk), net))
1548 if (sk->sk_family == state->family)
1557 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1559 struct udp_iter_state *state = seq->private;
1560 struct net *net = seq_file_net(seq);
1566 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1568 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) {
1569 sk = sk_head(state->hashtable + state->bucket);
1575 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1577 struct sock *sk = udp_get_first(seq);
1580 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1582 return pos ? NULL : sk;
1585 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1586 __acquires(udp_hash_lock)
1588 read_lock(&udp_hash_lock);
1589 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
1592 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1596 if (v == SEQ_START_TOKEN)
1597 sk = udp_get_idx(seq, 0);
1599 sk = udp_get_next(seq, v);
1605 static void udp_seq_stop(struct seq_file *seq, void *v)
1606 __releases(udp_hash_lock)
1608 read_unlock(&udp_hash_lock);
1611 static int udp_seq_open(struct inode *inode, struct file *file)
1613 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1614 struct udp_iter_state *s;
1617 err = seq_open_net(inode, file, &afinfo->seq_ops,
1618 sizeof(struct udp_iter_state));
1622 s = ((struct seq_file *)file->private_data)->private;
1623 s->family = afinfo->family;
1624 s->hashtable = afinfo->hashtable;
1628 /* ------------------------------------------------------------------------ */
1629 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
1631 struct proc_dir_entry *p;
1634 afinfo->seq_fops.open = udp_seq_open;
1635 afinfo->seq_fops.read = seq_read;
1636 afinfo->seq_fops.llseek = seq_lseek;
1637 afinfo->seq_fops.release = seq_release_net;
1639 afinfo->seq_ops.start = udp_seq_start;
1640 afinfo->seq_ops.next = udp_seq_next;
1641 afinfo->seq_ops.stop = udp_seq_stop;
1643 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
1644 &afinfo->seq_fops, afinfo);
1650 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
1652 proc_net_remove(net, afinfo->name);
1655 /* ------------------------------------------------------------------------ */
1656 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
1657 int bucket, int *len)
1659 struct inet_sock *inet = inet_sk(sp);
1660 __be32 dest = inet->daddr;
1661 __be32 src = inet->rcv_saddr;
1662 __u16 destp = ntohs(inet->dport);
1663 __u16 srcp = ntohs(inet->sport);
1665 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
1666 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
1667 bucket, src, srcp, dest, destp, sp->sk_state,
1668 atomic_read(&sp->sk_wmem_alloc),
1669 atomic_read(&sp->sk_rmem_alloc),
1670 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1671 atomic_read(&sp->sk_refcnt), sp,
1672 atomic_read(&sp->sk_drops), len);
1675 int udp4_seq_show(struct seq_file *seq, void *v)
1677 if (v == SEQ_START_TOKEN)
1678 seq_printf(seq, "%-127s\n",
1679 " sl local_address rem_address st tx_queue "
1680 "rx_queue tr tm->when retrnsmt uid timeout "
1681 "inode ref pointer drops");
1683 struct udp_iter_state *state = seq->private;
1686 udp4_format_sock(v, seq, state->bucket, &len);
1687 seq_printf(seq, "%*s\n", 127 - len ,"");
1692 /* ------------------------------------------------------------------------ */
1693 static struct udp_seq_afinfo udp4_seq_afinfo = {
1696 .hashtable = udp_hash,
1698 .owner = THIS_MODULE,
1701 .show = udp4_seq_show,
1705 static int udp4_proc_init_net(struct net *net)
1707 return udp_proc_register(net, &udp4_seq_afinfo);
1710 static void udp4_proc_exit_net(struct net *net)
1712 udp_proc_unregister(net, &udp4_seq_afinfo);
1715 static struct pernet_operations udp4_net_ops = {
1716 .init = udp4_proc_init_net,
1717 .exit = udp4_proc_exit_net,
1720 int __init udp4_proc_init(void)
1722 return register_pernet_subsys(&udp4_net_ops);
1725 void udp4_proc_exit(void)
1727 unregister_pernet_subsys(&udp4_net_ops);
1729 #endif /* CONFIG_PROC_FS */
1731 void __init udp_init(void)
1733 unsigned long limit;
1735 /* Set the pressure threshold up by the same strategy of TCP. It is a
1736 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
1737 * toward zero with the amount of memory, with a floor of 128 pages.
1739 limit = min(nr_all_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
1740 limit = (limit * (nr_all_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
1741 limit = max(limit, 128UL);
1742 sysctl_udp_mem[0] = limit / 4 * 3;
1743 sysctl_udp_mem[1] = limit;
1744 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
1746 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
1747 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
1750 EXPORT_SYMBOL(udp_disconnect);
1751 EXPORT_SYMBOL(udp_hash);
1752 EXPORT_SYMBOL(udp_hash_lock);
1753 EXPORT_SYMBOL(udp_ioctl);
1754 EXPORT_SYMBOL(udp_prot);
1755 EXPORT_SYMBOL(udp_sendmsg);
1756 EXPORT_SYMBOL(udp_lib_getsockopt);
1757 EXPORT_SYMBOL(udp_lib_setsockopt);
1758 EXPORT_SYMBOL(udp_poll);
1759 EXPORT_SYMBOL(udp_lib_get_port);
1761 #ifdef CONFIG_PROC_FS
1762 EXPORT_SYMBOL(udp_proc_register);
1763 EXPORT_SYMBOL(udp_proc_unregister);