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@lxorguk.ukuu.org.uk>
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"
107 struct udp_table udp_table;
108 EXPORT_SYMBOL(udp_table);
110 int sysctl_udp_mem[3] __read_mostly;
111 int sysctl_udp_rmem_min __read_mostly;
112 int sysctl_udp_wmem_min __read_mostly;
114 EXPORT_SYMBOL(sysctl_udp_mem);
115 EXPORT_SYMBOL(sysctl_udp_rmem_min);
116 EXPORT_SYMBOL(sysctl_udp_wmem_min);
118 atomic_t udp_memory_allocated;
119 EXPORT_SYMBOL(udp_memory_allocated);
121 static int udp_lib_lport_inuse(struct net *net, __u16 num,
122 const struct udp_hslot *hslot,
124 int (*saddr_comp)(const struct sock *sk1,
125 const struct sock *sk2))
128 struct hlist_node *node;
130 sk_for_each(sk2, node, &hslot->head)
131 if (net_eq(sock_net(sk2), net) &&
133 sk2->sk_hash == num &&
134 (!sk2->sk_reuse || !sk->sk_reuse) &&
135 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
136 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
137 (*saddr_comp)(sk, sk2))
143 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
145 * @sk: socket struct in question
146 * @snum: port number to look up
147 * @saddr_comp: AF-dependent comparison of bound local IP addresses
149 int udp_lib_get_port(struct sock *sk, unsigned short snum,
150 int (*saddr_comp)(const struct sock *sk1,
151 const struct sock *sk2 ) )
153 struct udp_hslot *hslot;
154 struct udp_table *udptable = sk->sk_prot->h.udp_table;
156 struct net *net = sock_net(sk);
159 int low, high, remaining;
161 unsigned short first;
163 inet_get_local_port_range(&low, &high);
164 remaining = (high - low) + 1;
167 snum = first = rand % remaining + low;
170 hslot = &udptable->hash[udp_hashfn(net, snum)];
171 spin_lock_bh(&hslot->lock);
172 if (!udp_lib_lport_inuse(net, snum, hslot, sk, saddr_comp))
174 spin_unlock_bh(&hslot->lock);
177 } while (snum < low || snum > high);
182 hslot = &udptable->hash[udp_hashfn(net, snum)];
183 spin_lock_bh(&hslot->lock);
184 if (udp_lib_lport_inuse(net, snum, hslot, sk, saddr_comp))
187 inet_sk(sk)->num = snum;
189 if (sk_unhashed(sk)) {
190 sk_add_node(sk, &hslot->head);
191 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
195 spin_unlock_bh(&hslot->lock);
200 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
202 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
204 return ( !ipv6_only_sock(sk2) &&
205 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
206 inet1->rcv_saddr == inet2->rcv_saddr ));
209 int udp_v4_get_port(struct sock *sk, unsigned short snum)
211 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal);
214 static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
216 __be16 sport, __be32 daddr, __be16 dport, int dif)
220 if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum &&
221 !ipv6_only_sock(sk)) {
222 struct inet_sock *inet = inet_sk(sk);
224 score = (sk->sk_family == PF_INET ? 1 : 0);
225 if (inet->rcv_saddr) {
226 if (inet->rcv_saddr != daddr)
231 if (inet->daddr != saddr)
236 if (inet->dport != sport)
240 if (sk->sk_bound_dev_if) {
241 if (sk->sk_bound_dev_if != dif)
249 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
250 * harder than this. -DaveM
252 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
253 __be16 sport, __be32 daddr, __be16 dport,
254 int dif, struct udp_table *udptable)
256 struct sock *sk, *result = NULL;
257 struct hlist_node *node;
258 unsigned short hnum = ntohs(dport);
259 unsigned int hash = udp_hashfn(net, hnum);
260 struct udp_hslot *hslot = &udptable->hash[hash];
261 int score, badness = -1;
263 spin_lock(&hslot->lock);
264 sk_for_each(sk, node, &hslot->head) {
265 score = compute_score(sk, net, saddr, hnum, sport,
267 if (score > badness) {
274 spin_unlock(&hslot->lock);
278 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
279 __be16 sport, __be16 dport,
280 struct udp_table *udptable)
283 const struct iphdr *iph = ip_hdr(skb);
285 if (unlikely(sk = skb_steal_sock(skb)))
288 return __udp4_lib_lookup(dev_net(skb->dst->dev), iph->saddr, sport,
289 iph->daddr, dport, inet_iif(skb),
293 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
294 __be32 daddr, __be16 dport, int dif)
296 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
298 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
300 static inline struct sock *udp_v4_mcast_next(struct sock *sk,
301 __be16 loc_port, __be32 loc_addr,
302 __be16 rmt_port, __be32 rmt_addr,
305 struct hlist_node *node;
307 unsigned short hnum = ntohs(loc_port);
309 sk_for_each_from(s, node) {
310 struct inet_sock *inet = inet_sk(s);
312 if (s->sk_hash != hnum ||
313 (inet->daddr && inet->daddr != rmt_addr) ||
314 (inet->dport != rmt_port && inet->dport) ||
315 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
317 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
319 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
329 * This routine is called by the ICMP module when it gets some
330 * sort of error condition. If err < 0 then the socket should
331 * be closed and the error returned to the user. If err > 0
332 * it's just the icmp type << 8 | icmp code.
333 * Header points to the ip header of the error packet. We move
334 * on past this. Then (as it used to claim before adjustment)
335 * header points to the first 8 bytes of the udp header. We need
336 * to find the appropriate port.
339 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
341 struct inet_sock *inet;
342 struct iphdr *iph = (struct iphdr*)skb->data;
343 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
344 const int type = icmp_hdr(skb)->type;
345 const int code = icmp_hdr(skb)->code;
349 struct net *net = dev_net(skb->dev);
351 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
352 iph->saddr, uh->source, skb->dev->ifindex, udptable);
354 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
355 return; /* No socket for error */
364 case ICMP_TIME_EXCEEDED:
367 case ICMP_SOURCE_QUENCH:
369 case ICMP_PARAMETERPROB:
373 case ICMP_DEST_UNREACH:
374 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
375 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
383 if (code <= NR_ICMP_UNREACH) {
384 harderr = icmp_err_convert[code].fatal;
385 err = icmp_err_convert[code].errno;
391 * RFC1122: OK. Passes ICMP errors back to application, as per
394 if (!inet->recverr) {
395 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
398 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
401 sk->sk_error_report(sk);
406 void udp_err(struct sk_buff *skb, u32 info)
408 __udp4_lib_err(skb, info, &udp_table);
412 * Throw away all pending data and cancel the corking. Socket is locked.
414 void udp_flush_pending_frames(struct sock *sk)
416 struct udp_sock *up = udp_sk(sk);
421 ip_flush_pending_frames(sk);
424 EXPORT_SYMBOL(udp_flush_pending_frames);
427 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
428 * @sk: socket we are sending on
429 * @skb: sk_buff containing the filled-in UDP header
430 * (checksum field must be zeroed out)
432 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
433 __be32 src, __be32 dst, int len )
436 struct udphdr *uh = udp_hdr(skb);
439 if (skb_queue_len(&sk->sk_write_queue) == 1) {
441 * Only one fragment on the socket.
443 skb->csum_start = skb_transport_header(skb) - skb->head;
444 skb->csum_offset = offsetof(struct udphdr, check);
445 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
448 * HW-checksum won't work as there are two or more
449 * fragments on the socket so that all csums of sk_buffs
452 offset = skb_transport_offset(skb);
453 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
455 skb->ip_summed = CHECKSUM_NONE;
457 skb_queue_walk(&sk->sk_write_queue, skb) {
458 csum = csum_add(csum, skb->csum);
461 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
463 uh->check = CSUM_MANGLED_0;
468 * Push out all pending data as one UDP datagram. Socket is locked.
470 static int udp_push_pending_frames(struct sock *sk)
472 struct udp_sock *up = udp_sk(sk);
473 struct inet_sock *inet = inet_sk(sk);
474 struct flowi *fl = &inet->cork.fl;
478 int is_udplite = IS_UDPLITE(sk);
481 /* Grab the skbuff where UDP header space exists. */
482 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
486 * Create a UDP header
489 uh->source = fl->fl_ip_sport;
490 uh->dest = fl->fl_ip_dport;
491 uh->len = htons(up->len);
494 if (is_udplite) /* UDP-Lite */
495 csum = udplite_csum_outgoing(sk, skb);
497 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
499 skb->ip_summed = CHECKSUM_NONE;
502 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
504 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
507 } else /* `normal' UDP */
508 csum = udp_csum_outgoing(sk, skb);
510 /* add protocol-dependent pseudo-header */
511 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
512 sk->sk_protocol, csum );
514 uh->check = CSUM_MANGLED_0;
517 err = ip_push_pending_frames(sk);
522 UDP_INC_STATS_USER(sock_net(sk),
523 UDP_MIB_OUTDATAGRAMS, is_udplite);
527 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
530 struct inet_sock *inet = inet_sk(sk);
531 struct udp_sock *up = udp_sk(sk);
533 struct ipcm_cookie ipc;
534 struct rtable *rt = NULL;
537 __be32 daddr, faddr, saddr;
540 int err, is_udplite = IS_UDPLITE(sk);
541 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
542 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
551 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
558 * There are pending frames.
559 * The socket lock must be held while it's corked.
562 if (likely(up->pending)) {
563 if (unlikely(up->pending != AF_INET)) {
571 ulen += sizeof(struct udphdr);
574 * Get and verify the address.
577 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
578 if (msg->msg_namelen < sizeof(*usin))
580 if (usin->sin_family != AF_INET) {
581 if (usin->sin_family != AF_UNSPEC)
582 return -EAFNOSUPPORT;
585 daddr = usin->sin_addr.s_addr;
586 dport = usin->sin_port;
590 if (sk->sk_state != TCP_ESTABLISHED)
591 return -EDESTADDRREQ;
594 /* Open fast path for connected socket.
595 Route will not be used, if at least one option is set.
599 ipc.addr = inet->saddr;
601 ipc.oif = sk->sk_bound_dev_if;
602 if (msg->msg_controllen) {
603 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
614 ipc.addr = faddr = daddr;
616 if (ipc.opt && ipc.opt->srr) {
619 faddr = ipc.opt->faddr;
622 tos = RT_TOS(inet->tos);
623 if (sock_flag(sk, SOCK_LOCALROUTE) ||
624 (msg->msg_flags & MSG_DONTROUTE) ||
625 (ipc.opt && ipc.opt->is_strictroute)) {
630 if (ipv4_is_multicast(daddr)) {
632 ipc.oif = inet->mc_index;
634 saddr = inet->mc_addr;
639 rt = (struct rtable*)sk_dst_check(sk, 0);
642 struct flowi fl = { .oif = ipc.oif,
647 .proto = sk->sk_protocol,
649 { .sport = inet->sport,
650 .dport = dport } } };
651 struct net *net = sock_net(sk);
653 security_sk_classify_flow(sk, &fl);
654 err = ip_route_output_flow(net, &rt, &fl, sk, 1);
656 if (err == -ENETUNREACH)
657 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
662 if ((rt->rt_flags & RTCF_BROADCAST) &&
663 !sock_flag(sk, SOCK_BROADCAST))
666 sk_dst_set(sk, dst_clone(&rt->u.dst));
669 if (msg->msg_flags&MSG_CONFIRM)
675 daddr = ipc.addr = rt->rt_dst;
678 if (unlikely(up->pending)) {
679 /* The socket is already corked while preparing it. */
680 /* ... which is an evident application bug. --ANK */
683 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
688 * Now cork the socket to pend data.
690 inet->cork.fl.fl4_dst = daddr;
691 inet->cork.fl.fl_ip_dport = dport;
692 inet->cork.fl.fl4_src = saddr;
693 inet->cork.fl.fl_ip_sport = inet->sport;
694 up->pending = AF_INET;
698 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
699 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
700 sizeof(struct udphdr), &ipc, rt,
701 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
703 udp_flush_pending_frames(sk);
705 err = udp_push_pending_frames(sk);
706 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
717 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
718 * ENOBUFS might not be good (it's not tunable per se), but otherwise
719 * we don't have a good statistic (IpOutDiscards but it can be too many
720 * things). We could add another new stat but at least for now that
721 * seems like overkill.
723 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
724 UDP_INC_STATS_USER(sock_net(sk),
725 UDP_MIB_SNDBUFERRORS, is_udplite);
730 dst_confirm(&rt->u.dst);
731 if (!(msg->msg_flags&MSG_PROBE) || len)
732 goto back_from_confirm;
737 int udp_sendpage(struct sock *sk, struct page *page, int offset,
738 size_t size, int flags)
740 struct udp_sock *up = udp_sk(sk);
744 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
746 /* Call udp_sendmsg to specify destination address which
747 * sendpage interface can't pass.
748 * This will succeed only when the socket is connected.
750 ret = udp_sendmsg(NULL, sk, &msg, 0);
757 if (unlikely(!up->pending)) {
760 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
764 ret = ip_append_page(sk, page, offset, size, flags);
765 if (ret == -EOPNOTSUPP) {
767 return sock_no_sendpage(sk->sk_socket, page, offset,
771 udp_flush_pending_frames(sk);
776 if (!(up->corkflag || (flags&MSG_MORE)))
777 ret = udp_push_pending_frames(sk);
786 * IOCTL requests applicable to the UDP protocol
789 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
794 int amount = atomic_read(&sk->sk_wmem_alloc);
795 return put_user(amount, (int __user *)arg);
801 unsigned long amount;
804 spin_lock_bh(&sk->sk_receive_queue.lock);
805 skb = skb_peek(&sk->sk_receive_queue);
808 * We will only return the amount
809 * of this packet since that is all
812 amount = skb->len - sizeof(struct udphdr);
814 spin_unlock_bh(&sk->sk_receive_queue.lock);
815 return put_user(amount, (int __user *)arg);
826 * This should be easy, if there is something there we
827 * return it, otherwise we block.
830 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
831 size_t len, int noblock, int flags, int *addr_len)
833 struct inet_sock *inet = inet_sk(sk);
834 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
836 unsigned int ulen, copied;
839 int is_udplite = IS_UDPLITE(sk);
842 * Check any passed addresses
845 *addr_len=sizeof(*sin);
847 if (flags & MSG_ERRQUEUE)
848 return ip_recv_error(sk, msg, len);
851 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
856 ulen = skb->len - sizeof(struct udphdr);
860 else if (copied < ulen)
861 msg->msg_flags |= MSG_TRUNC;
864 * If checksum is needed at all, try to do it while copying the
865 * data. If the data is truncated, or if we only want a partial
866 * coverage checksum (UDP-Lite), do it before the copy.
869 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
870 if (udp_lib_checksum_complete(skb))
874 if (skb_csum_unnecessary(skb))
875 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
876 msg->msg_iov, copied );
878 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
888 UDP_INC_STATS_USER(sock_net(sk),
889 UDP_MIB_INDATAGRAMS, is_udplite);
891 sock_recv_timestamp(msg, sk, skb);
893 /* Copy the address. */
896 sin->sin_family = AF_INET;
897 sin->sin_port = udp_hdr(skb)->source;
898 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
899 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
901 if (inet->cmsg_flags)
902 ip_cmsg_recv(msg, skb);
905 if (flags & MSG_TRUNC)
910 skb_free_datagram(sk, skb);
917 if (!skb_kill_datagram(sk, skb, flags))
918 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
927 int udp_disconnect(struct sock *sk, int flags)
929 struct inet_sock *inet = inet_sk(sk);
931 * 1003.1g - break association.
934 sk->sk_state = TCP_CLOSE;
937 sk->sk_bound_dev_if = 0;
938 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
939 inet_reset_saddr(sk);
941 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
942 sk->sk_prot->unhash(sk);
949 void udp_lib_unhash(struct sock *sk)
951 struct udp_table *udptable = sk->sk_prot->h.udp_table;
952 unsigned int hash = udp_hashfn(sock_net(sk), sk->sk_hash);
953 struct udp_hslot *hslot = &udptable->hash[hash];
955 spin_lock(&hslot->lock);
956 if (sk_del_node_init(sk)) {
957 inet_sk(sk)->num = 0;
958 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
960 spin_unlock(&hslot->lock);
962 EXPORT_SYMBOL(udp_lib_unhash);
964 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
966 int is_udplite = IS_UDPLITE(sk);
969 if ((rc = sock_queue_rcv_skb(sk, skb)) < 0) {
970 /* Note that an ENOMEM error is charged twice */
972 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
980 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
988 * >0: "udp encap" protocol resubmission
990 * Note that in the success and error cases, the skb is assumed to
991 * have either been requeued or freed.
993 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
995 struct udp_sock *up = udp_sk(sk);
997 int is_udplite = IS_UDPLITE(sk);
1000 * Charge it to the socket, dropping if the queue is full.
1002 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1006 if (up->encap_type) {
1008 * This is an encapsulation socket so pass the skb to
1009 * the socket's udp_encap_rcv() hook. Otherwise, just
1010 * fall through and pass this up the UDP socket.
1011 * up->encap_rcv() returns the following value:
1012 * =0 if skb was successfully passed to the encap
1013 * handler or was discarded by it.
1014 * >0 if skb should be passed on to UDP.
1015 * <0 if skb should be resubmitted as proto -N
1018 /* if we're overly short, let UDP handle it */
1019 if (skb->len > sizeof(struct udphdr) &&
1020 up->encap_rcv != NULL) {
1023 ret = (*up->encap_rcv)(sk, skb);
1025 UDP_INC_STATS_BH(sock_net(sk),
1026 UDP_MIB_INDATAGRAMS,
1032 /* FALLTHROUGH -- it's a UDP Packet */
1036 * UDP-Lite specific tests, ignored on UDP sockets
1038 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1041 * MIB statistics other than incrementing the error count are
1042 * disabled for the following two types of errors: these depend
1043 * on the application settings, not on the functioning of the
1044 * protocol stack as such.
1046 * RFC 3828 here recommends (sec 3.3): "There should also be a
1047 * way ... to ... at least let the receiving application block
1048 * delivery of packets with coverage values less than a value
1049 * provided by the application."
1051 if (up->pcrlen == 0) { /* full coverage was set */
1052 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1053 "%d while full coverage %d requested\n",
1054 UDP_SKB_CB(skb)->cscov, skb->len);
1057 /* The next case involves violating the min. coverage requested
1058 * by the receiver. This is subtle: if receiver wants x and x is
1059 * greater than the buffersize/MTU then receiver will complain
1060 * that it wants x while sender emits packets of smaller size y.
1061 * Therefore the above ...()->partial_cov statement is essential.
1063 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1064 LIMIT_NETDEBUG(KERN_WARNING
1065 "UDPLITE: coverage %d too small, need min %d\n",
1066 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1071 if (sk->sk_filter) {
1072 if (udp_lib_checksum_complete(skb))
1079 if (!sock_owned_by_user(sk))
1080 rc = __udp_queue_rcv_skb(sk, skb);
1082 sk_add_backlog(sk, skb);
1088 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1094 * Multicasts and broadcasts go to each listener.
1096 * Note: called only from the BH handler context,
1097 * so we don't need to lock the hashes.
1099 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1101 __be32 saddr, __be32 daddr,
1102 struct udp_table *udptable)
1105 struct udp_hslot *hslot = &udptable->hash[udp_hashfn(net, ntohs(uh->dest))];
1108 spin_lock(&hslot->lock);
1109 sk = sk_head(&hslot->head);
1110 dif = skb->dev->ifindex;
1111 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
1113 struct sock *sknext = NULL;
1116 struct sk_buff *skb1 = skb;
1118 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr,
1119 uh->source, saddr, dif);
1121 skb1 = skb_clone(skb, GFP_ATOMIC);
1124 int ret = udp_queue_rcv_skb(sk, skb1);
1126 /* we should probably re-process instead
1127 * of dropping packets here. */
1134 spin_unlock(&hslot->lock);
1138 /* Initialize UDP checksum. If exited with zero value (success),
1139 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1140 * Otherwise, csum completion requires chacksumming packet body,
1141 * including udp header and folding it to skb->csum.
1143 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1146 const struct iphdr *iph;
1149 UDP_SKB_CB(skb)->partial_cov = 0;
1150 UDP_SKB_CB(skb)->cscov = skb->len;
1152 if (proto == IPPROTO_UDPLITE) {
1153 err = udplite_checksum_init(skb, uh);
1159 if (uh->check == 0) {
1160 skb->ip_summed = CHECKSUM_UNNECESSARY;
1161 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1162 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1164 skb->ip_summed = CHECKSUM_UNNECESSARY;
1166 if (!skb_csum_unnecessary(skb))
1167 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1168 skb->len, proto, 0);
1169 /* Probably, we should checksum udp header (it should be in cache
1170 * in any case) and data in tiny packets (< rx copybreak).
1177 * All we need to do is get the socket, and then do a checksum.
1180 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1184 struct udphdr *uh = udp_hdr(skb);
1185 unsigned short ulen;
1186 struct rtable *rt = (struct rtable*)skb->dst;
1187 __be32 saddr = ip_hdr(skb)->saddr;
1188 __be32 daddr = ip_hdr(skb)->daddr;
1189 struct net *net = dev_net(skb->dev);
1192 * Validate the packet.
1194 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1195 goto drop; /* No space for header. */
1197 ulen = ntohs(uh->len);
1198 if (ulen > skb->len)
1201 if (proto == IPPROTO_UDP) {
1202 /* UDP validates ulen. */
1203 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1208 if (udp4_csum_init(skb, uh, proto))
1211 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1212 return __udp4_lib_mcast_deliver(net, skb, uh,
1213 saddr, daddr, udptable);
1215 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1218 int ret = udp_queue_rcv_skb(sk, skb);
1221 /* a return value > 0 means to resubmit the input, but
1222 * it wants the return to be -protocol, or 0
1229 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1233 /* No socket. Drop packet silently, if checksum is wrong */
1234 if (udp_lib_checksum_complete(skb))
1237 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1238 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1241 * Hmm. We got an UDP packet to a port to which we
1242 * don't wanna listen. Ignore it.
1248 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From " NIPQUAD_FMT ":%u %d/%d to " NIPQUAD_FMT ":%u\n",
1249 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1260 * RFC1122: OK. Discards the bad packet silently (as far as
1261 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1263 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From " NIPQUAD_FMT ":%u to " NIPQUAD_FMT ":%u ulen %d\n",
1264 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1271 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1276 int udp_rcv(struct sk_buff *skb)
1278 return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1281 void udp_destroy_sock(struct sock *sk)
1284 udp_flush_pending_frames(sk);
1289 * Socket option code for UDP
1291 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1292 char __user *optval, int optlen,
1293 int (*push_pending_frames)(struct sock *))
1295 struct udp_sock *up = udp_sk(sk);
1298 int is_udplite = IS_UDPLITE(sk);
1300 if (optlen<sizeof(int))
1303 if (get_user(val, (int __user *)optval))
1313 (*push_pending_frames)(sk);
1321 case UDP_ENCAP_ESPINUDP:
1322 case UDP_ENCAP_ESPINUDP_NON_IKE:
1323 up->encap_rcv = xfrm4_udp_encap_rcv;
1325 case UDP_ENCAP_L2TPINUDP:
1326 up->encap_type = val;
1335 * UDP-Lite's partial checksum coverage (RFC 3828).
1337 /* The sender sets actual checksum coverage length via this option.
1338 * The case coverage > packet length is handled by send module. */
1339 case UDPLITE_SEND_CSCOV:
1340 if (!is_udplite) /* Disable the option on UDP sockets */
1341 return -ENOPROTOOPT;
1342 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1344 else if (val > USHORT_MAX)
1347 up->pcflag |= UDPLITE_SEND_CC;
1350 /* The receiver specifies a minimum checksum coverage value. To make
1351 * sense, this should be set to at least 8 (as done below). If zero is
1352 * used, this again means full checksum coverage. */
1353 case UDPLITE_RECV_CSCOV:
1354 if (!is_udplite) /* Disable the option on UDP sockets */
1355 return -ENOPROTOOPT;
1356 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1358 else if (val > USHORT_MAX)
1361 up->pcflag |= UDPLITE_RECV_CC;
1372 int udp_setsockopt(struct sock *sk, int level, int optname,
1373 char __user *optval, int optlen)
1375 if (level == SOL_UDP || level == SOL_UDPLITE)
1376 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1377 udp_push_pending_frames);
1378 return ip_setsockopt(sk, level, optname, optval, optlen);
1381 #ifdef CONFIG_COMPAT
1382 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1383 char __user *optval, int optlen)
1385 if (level == SOL_UDP || level == SOL_UDPLITE)
1386 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1387 udp_push_pending_frames);
1388 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1392 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1393 char __user *optval, int __user *optlen)
1395 struct udp_sock *up = udp_sk(sk);
1398 if (get_user(len,optlen))
1401 len = min_t(unsigned int, len, sizeof(int));
1412 val = up->encap_type;
1415 /* The following two cannot be changed on UDP sockets, the return is
1416 * always 0 (which corresponds to the full checksum coverage of UDP). */
1417 case UDPLITE_SEND_CSCOV:
1421 case UDPLITE_RECV_CSCOV:
1426 return -ENOPROTOOPT;
1429 if (put_user(len, optlen))
1431 if (copy_to_user(optval, &val,len))
1436 int udp_getsockopt(struct sock *sk, int level, int optname,
1437 char __user *optval, int __user *optlen)
1439 if (level == SOL_UDP || level == SOL_UDPLITE)
1440 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1441 return ip_getsockopt(sk, level, optname, optval, optlen);
1444 #ifdef CONFIG_COMPAT
1445 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1446 char __user *optval, int __user *optlen)
1448 if (level == SOL_UDP || level == SOL_UDPLITE)
1449 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1450 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1454 * udp_poll - wait for a UDP event.
1455 * @file - file struct
1457 * @wait - poll table
1459 * This is same as datagram poll, except for the special case of
1460 * blocking sockets. If application is using a blocking fd
1461 * and a packet with checksum error is in the queue;
1462 * then it could get return from select indicating data available
1463 * but then block when reading it. Add special case code
1464 * to work around these arguably broken applications.
1466 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1468 unsigned int mask = datagram_poll(file, sock, wait);
1469 struct sock *sk = sock->sk;
1470 int is_lite = IS_UDPLITE(sk);
1472 /* Check for false positives due to checksum errors */
1473 if ( (mask & POLLRDNORM) &&
1474 !(file->f_flags & O_NONBLOCK) &&
1475 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1476 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1477 struct sk_buff *skb;
1479 spin_lock_bh(&rcvq->lock);
1480 while ((skb = skb_peek(rcvq)) != NULL &&
1481 udp_lib_checksum_complete(skb)) {
1482 UDP_INC_STATS_BH(sock_net(sk),
1483 UDP_MIB_INERRORS, is_lite);
1484 __skb_unlink(skb, rcvq);
1487 spin_unlock_bh(&rcvq->lock);
1489 /* nothing to see, move along */
1491 mask &= ~(POLLIN | POLLRDNORM);
1498 struct proto udp_prot = {
1500 .owner = THIS_MODULE,
1501 .close = udp_lib_close,
1502 .connect = ip4_datagram_connect,
1503 .disconnect = udp_disconnect,
1505 .destroy = udp_destroy_sock,
1506 .setsockopt = udp_setsockopt,
1507 .getsockopt = udp_getsockopt,
1508 .sendmsg = udp_sendmsg,
1509 .recvmsg = udp_recvmsg,
1510 .sendpage = udp_sendpage,
1511 .backlog_rcv = __udp_queue_rcv_skb,
1512 .hash = udp_lib_hash,
1513 .unhash = udp_lib_unhash,
1514 .get_port = udp_v4_get_port,
1515 .memory_allocated = &udp_memory_allocated,
1516 .sysctl_mem = sysctl_udp_mem,
1517 .sysctl_wmem = &sysctl_udp_wmem_min,
1518 .sysctl_rmem = &sysctl_udp_rmem_min,
1519 .obj_size = sizeof(struct udp_sock),
1520 .h.udp_table = &udp_table,
1521 #ifdef CONFIG_COMPAT
1522 .compat_setsockopt = compat_udp_setsockopt,
1523 .compat_getsockopt = compat_udp_getsockopt,
1527 /* ------------------------------------------------------------------------ */
1528 #ifdef CONFIG_PROC_FS
1530 static struct sock *udp_get_first(struct seq_file *seq, int start)
1533 struct udp_iter_state *state = seq->private;
1534 struct net *net = seq_file_net(seq);
1536 for (state->bucket = start; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1537 struct hlist_node *node;
1538 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
1539 spin_lock_bh(&hslot->lock);
1540 sk_for_each(sk, node, &hslot->head) {
1541 if (!net_eq(sock_net(sk), net))
1543 if (sk->sk_family == state->family)
1546 spin_unlock_bh(&hslot->lock);
1553 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1555 struct udp_iter_state *state = seq->private;
1556 struct net *net = seq_file_net(seq);
1560 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1563 spin_unlock(&state->udp_table->hash[state->bucket].lock);
1564 return udp_get_first(seq, state->bucket + 1);
1569 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1571 struct sock *sk = udp_get_first(seq, 0);
1574 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1576 return pos ? NULL : sk;
1579 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1581 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
1584 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1588 if (v == SEQ_START_TOKEN)
1589 sk = udp_get_idx(seq, 0);
1591 sk = udp_get_next(seq, v);
1597 static void udp_seq_stop(struct seq_file *seq, void *v)
1599 struct udp_iter_state *state = seq->private;
1601 if (state->bucket < UDP_HTABLE_SIZE)
1602 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
1605 static int udp_seq_open(struct inode *inode, struct file *file)
1607 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1608 struct udp_iter_state *s;
1611 err = seq_open_net(inode, file, &afinfo->seq_ops,
1612 sizeof(struct udp_iter_state));
1616 s = ((struct seq_file *)file->private_data)->private;
1617 s->family = afinfo->family;
1618 s->udp_table = afinfo->udp_table;
1622 /* ------------------------------------------------------------------------ */
1623 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
1625 struct proc_dir_entry *p;
1628 afinfo->seq_fops.open = udp_seq_open;
1629 afinfo->seq_fops.read = seq_read;
1630 afinfo->seq_fops.llseek = seq_lseek;
1631 afinfo->seq_fops.release = seq_release_net;
1633 afinfo->seq_ops.start = udp_seq_start;
1634 afinfo->seq_ops.next = udp_seq_next;
1635 afinfo->seq_ops.stop = udp_seq_stop;
1637 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
1638 &afinfo->seq_fops, afinfo);
1644 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
1646 proc_net_remove(net, afinfo->name);
1649 /* ------------------------------------------------------------------------ */
1650 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
1651 int bucket, int *len)
1653 struct inet_sock *inet = inet_sk(sp);
1654 __be32 dest = inet->daddr;
1655 __be32 src = inet->rcv_saddr;
1656 __u16 destp = ntohs(inet->dport);
1657 __u16 srcp = ntohs(inet->sport);
1659 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
1660 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
1661 bucket, src, srcp, dest, destp, sp->sk_state,
1662 atomic_read(&sp->sk_wmem_alloc),
1663 atomic_read(&sp->sk_rmem_alloc),
1664 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1665 atomic_read(&sp->sk_refcnt), sp,
1666 atomic_read(&sp->sk_drops), len);
1669 int udp4_seq_show(struct seq_file *seq, void *v)
1671 if (v == SEQ_START_TOKEN)
1672 seq_printf(seq, "%-127s\n",
1673 " sl local_address rem_address st tx_queue "
1674 "rx_queue tr tm->when retrnsmt uid timeout "
1675 "inode ref pointer drops");
1677 struct udp_iter_state *state = seq->private;
1680 udp4_format_sock(v, seq, state->bucket, &len);
1681 seq_printf(seq, "%*s\n", 127 - len ,"");
1686 /* ------------------------------------------------------------------------ */
1687 static struct udp_seq_afinfo udp4_seq_afinfo = {
1690 .udp_table = &udp_table,
1692 .owner = THIS_MODULE,
1695 .show = udp4_seq_show,
1699 static int udp4_proc_init_net(struct net *net)
1701 return udp_proc_register(net, &udp4_seq_afinfo);
1704 static void udp4_proc_exit_net(struct net *net)
1706 udp_proc_unregister(net, &udp4_seq_afinfo);
1709 static struct pernet_operations udp4_net_ops = {
1710 .init = udp4_proc_init_net,
1711 .exit = udp4_proc_exit_net,
1714 int __init udp4_proc_init(void)
1716 return register_pernet_subsys(&udp4_net_ops);
1719 void udp4_proc_exit(void)
1721 unregister_pernet_subsys(&udp4_net_ops);
1723 #endif /* CONFIG_PROC_FS */
1725 void __init udp_table_init(struct udp_table *table)
1729 for (i = 0; i < UDP_HTABLE_SIZE; i++) {
1730 INIT_HLIST_HEAD(&table->hash[i].head);
1731 spin_lock_init(&table->hash[i].lock);
1735 void __init udp_init(void)
1737 unsigned long limit;
1739 udp_table_init(&udp_table);
1740 /* Set the pressure threshold up by the same strategy of TCP. It is a
1741 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
1742 * toward zero with the amount of memory, with a floor of 128 pages.
1744 limit = min(nr_all_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
1745 limit = (limit * (nr_all_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
1746 limit = max(limit, 128UL);
1747 sysctl_udp_mem[0] = limit / 4 * 3;
1748 sysctl_udp_mem[1] = limit;
1749 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
1751 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
1752 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
1755 EXPORT_SYMBOL(udp_disconnect);
1756 EXPORT_SYMBOL(udp_ioctl);
1757 EXPORT_SYMBOL(udp_prot);
1758 EXPORT_SYMBOL(udp_sendmsg);
1759 EXPORT_SYMBOL(udp_lib_getsockopt);
1760 EXPORT_SYMBOL(udp_lib_setsockopt);
1761 EXPORT_SYMBOL(udp_poll);
1762 EXPORT_SYMBOL(udp_lib_get_port);
1764 #ifdef CONFIG_PROC_FS
1765 EXPORT_SYMBOL(udp_proc_register);
1766 EXPORT_SYMBOL(udp_proc_unregister);