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 struct hlist_head udp_hash[UDP_HTABLE_SIZE];
112 DEFINE_RWLOCK(udp_hash_lock);
114 int sysctl_udp_mem[3] __read_mostly;
115 int sysctl_udp_rmem_min __read_mostly;
116 int sysctl_udp_wmem_min __read_mostly;
118 EXPORT_SYMBOL(sysctl_udp_mem);
119 EXPORT_SYMBOL(sysctl_udp_rmem_min);
120 EXPORT_SYMBOL(sysctl_udp_wmem_min);
122 atomic_t udp_memory_allocated;
123 EXPORT_SYMBOL(udp_memory_allocated);
125 static inline int __udp_lib_lport_inuse(struct net *net, __u16 num,
126 const struct hlist_head udptable[])
129 struct hlist_node *node;
131 sk_for_each(sk, node, &udptable[udp_hashfn(net, num)])
132 if (net_eq(sock_net(sk), net) && sk->sk_hash == num)
138 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
140 * @sk: socket struct in question
141 * @snum: port number to look up
142 * @saddr_comp: AF-dependent comparison of bound local IP addresses
144 int udp_lib_get_port(struct sock *sk, unsigned short snum,
145 int (*saddr_comp)(const struct sock *sk1,
146 const struct sock *sk2 ) )
148 struct hlist_head *udptable = sk->sk_prot->h.udp_hash;
149 struct hlist_node *node;
150 struct hlist_head *head;
153 struct net *net = sock_net(sk);
155 write_lock_bh(&udp_hash_lock);
158 int low, high, remaining;
160 unsigned short first;
162 inet_get_local_port_range(&low, &high);
163 remaining = (high - low) + 1;
166 snum = first = rand % remaining + low;
168 while (__udp_lib_lport_inuse(net, snum, udptable)) {
171 } while (snum < low || snum > high);
176 head = &udptable[udp_hashfn(net, snum)];
178 sk_for_each(sk2, node, head)
179 if (sk2->sk_hash == snum &&
181 net_eq(sock_net(sk2), net) &&
182 (!sk2->sk_reuse || !sk->sk_reuse) &&
183 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
184 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
185 (*saddr_comp)(sk, sk2) )
189 inet_sk(sk)->num = snum;
191 if (sk_unhashed(sk)) {
192 head = &udptable[udp_hashfn(net, snum)];
193 sk_add_node(sk, head);
194 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
198 write_unlock_bh(&udp_hash_lock);
202 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
204 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
206 return ( !ipv6_only_sock(sk2) &&
207 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
208 inet1->rcv_saddr == inet2->rcv_saddr ));
211 int udp_v4_get_port(struct sock *sk, unsigned short snum)
213 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal);
216 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
217 * harder than this. -DaveM
219 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
220 __be16 sport, __be32 daddr, __be16 dport,
221 int dif, struct hlist_head udptable[])
223 struct sock *sk, *result = NULL;
224 struct hlist_node *node;
225 unsigned short hnum = ntohs(dport);
228 read_lock(&udp_hash_lock);
229 sk_for_each(sk, node, &udptable[udp_hashfn(net, hnum)]) {
230 struct inet_sock *inet = inet_sk(sk);
232 if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum &&
233 !ipv6_only_sock(sk)) {
234 int score = (sk->sk_family == PF_INET ? 1 : 0);
235 if (inet->rcv_saddr) {
236 if (inet->rcv_saddr != daddr)
241 if (inet->daddr != saddr)
246 if (inet->dport != sport)
250 if (sk->sk_bound_dev_if) {
251 if (sk->sk_bound_dev_if != dif)
258 } else if (score > badness) {
266 read_unlock(&udp_hash_lock);
270 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
271 __be16 sport, __be16 dport,
272 struct hlist_head udptable[])
275 const struct iphdr *iph = ip_hdr(skb);
277 if (unlikely(sk = skb_steal_sock(skb)))
280 return __udp4_lib_lookup(dev_net(skb->dst->dev), iph->saddr, sport,
281 iph->daddr, dport, inet_iif(skb),
285 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
286 __be32 daddr, __be16 dport, int dif)
288 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, udp_hash);
290 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
292 static inline struct sock *udp_v4_mcast_next(struct sock *sk,
293 __be16 loc_port, __be32 loc_addr,
294 __be16 rmt_port, __be32 rmt_addr,
297 struct hlist_node *node;
299 unsigned short hnum = ntohs(loc_port);
301 sk_for_each_from(s, node) {
302 struct inet_sock *inet = inet_sk(s);
304 if (s->sk_hash != hnum ||
305 (inet->daddr && inet->daddr != rmt_addr) ||
306 (inet->dport != rmt_port && inet->dport) ||
307 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
309 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
311 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
321 * This routine is called by the ICMP module when it gets some
322 * sort of error condition. If err < 0 then the socket should
323 * be closed and the error returned to the user. If err > 0
324 * it's just the icmp type << 8 | icmp code.
325 * Header points to the ip header of the error packet. We move
326 * on past this. Then (as it used to claim before adjustment)
327 * header points to the first 8 bytes of the udp header. We need
328 * to find the appropriate port.
331 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[])
333 struct inet_sock *inet;
334 struct iphdr *iph = (struct iphdr*)skb->data;
335 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
336 const int type = icmp_hdr(skb)->type;
337 const int code = icmp_hdr(skb)->code;
341 struct net *net = dev_net(skb->dev);
343 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
344 iph->saddr, uh->source, skb->dev->ifindex, udptable);
346 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
347 return; /* No socket for error */
356 case ICMP_TIME_EXCEEDED:
359 case ICMP_SOURCE_QUENCH:
361 case ICMP_PARAMETERPROB:
365 case ICMP_DEST_UNREACH:
366 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
367 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
375 if (code <= NR_ICMP_UNREACH) {
376 harderr = icmp_err_convert[code].fatal;
377 err = icmp_err_convert[code].errno;
383 * RFC1122: OK. Passes ICMP errors back to application, as per
386 if (!inet->recverr) {
387 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
390 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
393 sk->sk_error_report(sk);
398 void udp_err(struct sk_buff *skb, u32 info)
400 __udp4_lib_err(skb, info, udp_hash);
404 * Throw away all pending data and cancel the corking. Socket is locked.
406 void udp_flush_pending_frames(struct sock *sk)
408 struct udp_sock *up = udp_sk(sk);
413 ip_flush_pending_frames(sk);
416 EXPORT_SYMBOL(udp_flush_pending_frames);
419 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
420 * @sk: socket we are sending on
421 * @skb: sk_buff containing the filled-in UDP header
422 * (checksum field must be zeroed out)
424 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
425 __be32 src, __be32 dst, int len )
428 struct udphdr *uh = udp_hdr(skb);
431 if (skb_queue_len(&sk->sk_write_queue) == 1) {
433 * Only one fragment on the socket.
435 skb->csum_start = skb_transport_header(skb) - skb->head;
436 skb->csum_offset = offsetof(struct udphdr, check);
437 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
440 * HW-checksum won't work as there are two or more
441 * fragments on the socket so that all csums of sk_buffs
444 offset = skb_transport_offset(skb);
445 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
447 skb->ip_summed = CHECKSUM_NONE;
449 skb_queue_walk(&sk->sk_write_queue, skb) {
450 csum = csum_add(csum, skb->csum);
453 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
455 uh->check = CSUM_MANGLED_0;
460 * Push out all pending data as one UDP datagram. Socket is locked.
462 static int udp_push_pending_frames(struct sock *sk)
464 struct udp_sock *up = udp_sk(sk);
465 struct inet_sock *inet = inet_sk(sk);
466 struct flowi *fl = &inet->cork.fl;
470 int is_udplite = IS_UDPLITE(sk);
473 /* Grab the skbuff where UDP header space exists. */
474 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
478 * Create a UDP header
481 uh->source = fl->fl_ip_sport;
482 uh->dest = fl->fl_ip_dport;
483 uh->len = htons(up->len);
486 if (is_udplite) /* UDP-Lite */
487 csum = udplite_csum_outgoing(sk, skb);
489 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
491 skb->ip_summed = CHECKSUM_NONE;
494 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
496 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
499 } else /* `normal' UDP */
500 csum = udp_csum_outgoing(sk, skb);
502 /* add protocol-dependent pseudo-header */
503 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
504 sk->sk_protocol, csum );
506 uh->check = CSUM_MANGLED_0;
509 err = ip_push_pending_frames(sk);
514 UDP_INC_STATS_USER(sock_net(sk),
515 UDP_MIB_OUTDATAGRAMS, is_udplite);
519 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
522 struct inet_sock *inet = inet_sk(sk);
523 struct udp_sock *up = udp_sk(sk);
525 struct ipcm_cookie ipc;
526 struct rtable *rt = NULL;
529 __be32 daddr, faddr, saddr;
532 int err, is_udplite = IS_UDPLITE(sk);
533 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
534 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
543 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
550 * There are pending frames.
551 * The socket lock must be held while it's corked.
554 if (likely(up->pending)) {
555 if (unlikely(up->pending != AF_INET)) {
563 ulen += sizeof(struct udphdr);
566 * Get and verify the address.
569 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
570 if (msg->msg_namelen < sizeof(*usin))
572 if (usin->sin_family != AF_INET) {
573 if (usin->sin_family != AF_UNSPEC)
574 return -EAFNOSUPPORT;
577 daddr = usin->sin_addr.s_addr;
578 dport = usin->sin_port;
582 if (sk->sk_state != TCP_ESTABLISHED)
583 return -EDESTADDRREQ;
586 /* Open fast path for connected socket.
587 Route will not be used, if at least one option is set.
591 ipc.addr = inet->saddr;
593 ipc.oif = sk->sk_bound_dev_if;
594 if (msg->msg_controllen) {
595 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
606 ipc.addr = faddr = daddr;
608 if (ipc.opt && ipc.opt->srr) {
611 faddr = ipc.opt->faddr;
614 tos = RT_TOS(inet->tos);
615 if (sock_flag(sk, SOCK_LOCALROUTE) ||
616 (msg->msg_flags & MSG_DONTROUTE) ||
617 (ipc.opt && ipc.opt->is_strictroute)) {
622 if (ipv4_is_multicast(daddr)) {
624 ipc.oif = inet->mc_index;
626 saddr = inet->mc_addr;
631 rt = (struct rtable*)sk_dst_check(sk, 0);
634 struct flowi fl = { .oif = ipc.oif,
639 .proto = sk->sk_protocol,
641 { .sport = inet->sport,
642 .dport = dport } } };
643 struct net *net = sock_net(sk);
645 security_sk_classify_flow(sk, &fl);
646 err = ip_route_output_flow(net, &rt, &fl, sk, 1);
648 if (err == -ENETUNREACH)
649 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
654 if ((rt->rt_flags & RTCF_BROADCAST) &&
655 !sock_flag(sk, SOCK_BROADCAST))
658 sk_dst_set(sk, dst_clone(&rt->u.dst));
661 if (msg->msg_flags&MSG_CONFIRM)
667 daddr = ipc.addr = rt->rt_dst;
670 if (unlikely(up->pending)) {
671 /* The socket is already corked while preparing it. */
672 /* ... which is an evident application bug. --ANK */
675 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
680 * Now cork the socket to pend data.
682 inet->cork.fl.fl4_dst = daddr;
683 inet->cork.fl.fl_ip_dport = dport;
684 inet->cork.fl.fl4_src = saddr;
685 inet->cork.fl.fl_ip_sport = inet->sport;
686 up->pending = AF_INET;
690 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
691 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
692 sizeof(struct udphdr), &ipc, rt,
693 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
695 udp_flush_pending_frames(sk);
697 err = udp_push_pending_frames(sk);
698 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
709 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
710 * ENOBUFS might not be good (it's not tunable per se), but otherwise
711 * we don't have a good statistic (IpOutDiscards but it can be too many
712 * things). We could add another new stat but at least for now that
713 * seems like overkill.
715 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
716 UDP_INC_STATS_USER(sock_net(sk),
717 UDP_MIB_SNDBUFERRORS, is_udplite);
722 dst_confirm(&rt->u.dst);
723 if (!(msg->msg_flags&MSG_PROBE) || len)
724 goto back_from_confirm;
729 int udp_sendpage(struct sock *sk, struct page *page, int offset,
730 size_t size, int flags)
732 struct udp_sock *up = udp_sk(sk);
736 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
738 /* Call udp_sendmsg to specify destination address which
739 * sendpage interface can't pass.
740 * This will succeed only when the socket is connected.
742 ret = udp_sendmsg(NULL, sk, &msg, 0);
749 if (unlikely(!up->pending)) {
752 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
756 ret = ip_append_page(sk, page, offset, size, flags);
757 if (ret == -EOPNOTSUPP) {
759 return sock_no_sendpage(sk->sk_socket, page, offset,
763 udp_flush_pending_frames(sk);
768 if (!(up->corkflag || (flags&MSG_MORE)))
769 ret = udp_push_pending_frames(sk);
778 * IOCTL requests applicable to the UDP protocol
781 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
786 int amount = atomic_read(&sk->sk_wmem_alloc);
787 return put_user(amount, (int __user *)arg);
793 unsigned long amount;
796 spin_lock_bh(&sk->sk_receive_queue.lock);
797 skb = skb_peek(&sk->sk_receive_queue);
800 * We will only return the amount
801 * of this packet since that is all
804 amount = skb->len - sizeof(struct udphdr);
806 spin_unlock_bh(&sk->sk_receive_queue.lock);
807 return put_user(amount, (int __user *)arg);
818 * This should be easy, if there is something there we
819 * return it, otherwise we block.
822 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
823 size_t len, int noblock, int flags, int *addr_len)
825 struct inet_sock *inet = inet_sk(sk);
826 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
828 unsigned int ulen, copied;
831 int is_udplite = IS_UDPLITE(sk);
834 * Check any passed addresses
837 *addr_len=sizeof(*sin);
839 if (flags & MSG_ERRQUEUE)
840 return ip_recv_error(sk, msg, len);
843 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
848 ulen = skb->len - sizeof(struct udphdr);
852 else if (copied < ulen)
853 msg->msg_flags |= MSG_TRUNC;
856 * If checksum is needed at all, try to do it while copying the
857 * data. If the data is truncated, or if we only want a partial
858 * coverage checksum (UDP-Lite), do it before the copy.
861 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
862 if (udp_lib_checksum_complete(skb))
866 if (skb_csum_unnecessary(skb))
867 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
868 msg->msg_iov, copied );
870 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
880 UDP_INC_STATS_USER(sock_net(sk),
881 UDP_MIB_INDATAGRAMS, is_udplite);
883 sock_recv_timestamp(msg, sk, skb);
885 /* Copy the address. */
888 sin->sin_family = AF_INET;
889 sin->sin_port = udp_hdr(skb)->source;
890 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
891 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
893 if (inet->cmsg_flags)
894 ip_cmsg_recv(msg, skb);
897 if (flags & MSG_TRUNC)
902 skb_free_datagram(sk, skb);
909 if (!skb_kill_datagram(sk, skb, flags))
910 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
919 int udp_disconnect(struct sock *sk, int flags)
921 struct inet_sock *inet = inet_sk(sk);
923 * 1003.1g - break association.
926 sk->sk_state = TCP_CLOSE;
929 sk->sk_bound_dev_if = 0;
930 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
931 inet_reset_saddr(sk);
933 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
934 sk->sk_prot->unhash(sk);
941 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
943 int is_udplite = IS_UDPLITE(sk);
946 if ((rc = sock_queue_rcv_skb(sk, skb)) < 0) {
947 /* Note that an ENOMEM error is charged twice */
949 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
957 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
965 * >0: "udp encap" protocol resubmission
967 * Note that in the success and error cases, the skb is assumed to
968 * have either been requeued or freed.
970 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
972 struct udp_sock *up = udp_sk(sk);
974 int is_udplite = IS_UDPLITE(sk);
977 * Charge it to the socket, dropping if the queue is full.
979 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
983 if (up->encap_type) {
985 * This is an encapsulation socket so pass the skb to
986 * the socket's udp_encap_rcv() hook. Otherwise, just
987 * fall through and pass this up the UDP socket.
988 * up->encap_rcv() returns the following value:
989 * =0 if skb was successfully passed to the encap
990 * handler or was discarded by it.
991 * >0 if skb should be passed on to UDP.
992 * <0 if skb should be resubmitted as proto -N
995 /* if we're overly short, let UDP handle it */
996 if (skb->len > sizeof(struct udphdr) &&
997 up->encap_rcv != NULL) {
1000 ret = (*up->encap_rcv)(sk, skb);
1002 UDP_INC_STATS_BH(sock_net(sk),
1003 UDP_MIB_INDATAGRAMS,
1009 /* FALLTHROUGH -- it's a UDP Packet */
1013 * UDP-Lite specific tests, ignored on UDP sockets
1015 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1018 * MIB statistics other than incrementing the error count are
1019 * disabled for the following two types of errors: these depend
1020 * on the application settings, not on the functioning of the
1021 * protocol stack as such.
1023 * RFC 3828 here recommends (sec 3.3): "There should also be a
1024 * way ... to ... at least let the receiving application block
1025 * delivery of packets with coverage values less than a value
1026 * provided by the application."
1028 if (up->pcrlen == 0) { /* full coverage was set */
1029 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1030 "%d while full coverage %d requested\n",
1031 UDP_SKB_CB(skb)->cscov, skb->len);
1034 /* The next case involves violating the min. coverage requested
1035 * by the receiver. This is subtle: if receiver wants x and x is
1036 * greater than the buffersize/MTU then receiver will complain
1037 * that it wants x while sender emits packets of smaller size y.
1038 * Therefore the above ...()->partial_cov statement is essential.
1040 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1041 LIMIT_NETDEBUG(KERN_WARNING
1042 "UDPLITE: coverage %d too small, need min %d\n",
1043 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1048 if (sk->sk_filter) {
1049 if (udp_lib_checksum_complete(skb))
1056 if (!sock_owned_by_user(sk))
1057 rc = __udp_queue_rcv_skb(sk, skb);
1059 sk_add_backlog(sk, skb);
1065 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1071 * Multicasts and broadcasts go to each listener.
1073 * Note: called only from the BH handler context,
1074 * so we don't need to lock the hashes.
1076 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1078 __be32 saddr, __be32 daddr,
1079 struct hlist_head udptable[])
1084 read_lock(&udp_hash_lock);
1085 sk = sk_head(&udptable[udp_hashfn(net, ntohs(uh->dest))]);
1086 dif = skb->dev->ifindex;
1087 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
1089 struct sock *sknext = NULL;
1092 struct sk_buff *skb1 = skb;
1094 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr,
1095 uh->source, saddr, dif);
1097 skb1 = skb_clone(skb, GFP_ATOMIC);
1100 int ret = udp_queue_rcv_skb(sk, skb1);
1102 /* we should probably re-process instead
1103 * of dropping packets here. */
1110 read_unlock(&udp_hash_lock);
1114 /* Initialize UDP checksum. If exited with zero value (success),
1115 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1116 * Otherwise, csum completion requires chacksumming packet body,
1117 * including udp header and folding it to skb->csum.
1119 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1122 const struct iphdr *iph;
1125 UDP_SKB_CB(skb)->partial_cov = 0;
1126 UDP_SKB_CB(skb)->cscov = skb->len;
1128 if (proto == IPPROTO_UDPLITE) {
1129 err = udplite_checksum_init(skb, uh);
1135 if (uh->check == 0) {
1136 skb->ip_summed = CHECKSUM_UNNECESSARY;
1137 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1138 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1140 skb->ip_summed = CHECKSUM_UNNECESSARY;
1142 if (!skb_csum_unnecessary(skb))
1143 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1144 skb->len, proto, 0);
1145 /* Probably, we should checksum udp header (it should be in cache
1146 * in any case) and data in tiny packets (< rx copybreak).
1153 * All we need to do is get the socket, and then do a checksum.
1156 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
1160 struct udphdr *uh = udp_hdr(skb);
1161 unsigned short ulen;
1162 struct rtable *rt = (struct rtable*)skb->dst;
1163 __be32 saddr = ip_hdr(skb)->saddr;
1164 __be32 daddr = ip_hdr(skb)->daddr;
1165 struct net *net = dev_net(skb->dev);
1168 * Validate the packet.
1170 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1171 goto drop; /* No space for header. */
1173 ulen = ntohs(uh->len);
1174 if (ulen > skb->len)
1177 if (proto == IPPROTO_UDP) {
1178 /* UDP validates ulen. */
1179 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1184 if (udp4_csum_init(skb, uh, proto))
1187 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1188 return __udp4_lib_mcast_deliver(net, skb, uh,
1189 saddr, daddr, udptable);
1191 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1194 int ret = udp_queue_rcv_skb(sk, skb);
1197 /* a return value > 0 means to resubmit the input, but
1198 * it wants the return to be -protocol, or 0
1205 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1209 /* No socket. Drop packet silently, if checksum is wrong */
1210 if (udp_lib_checksum_complete(skb))
1213 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1214 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1217 * Hmm. We got an UDP packet to a port to which we
1218 * don't wanna listen. Ignore it.
1224 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From " NIPQUAD_FMT ":%u %d/%d to " NIPQUAD_FMT ":%u\n",
1225 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1236 * RFC1122: OK. Discards the bad packet silently (as far as
1237 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1239 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From " NIPQUAD_FMT ":%u to " NIPQUAD_FMT ":%u ulen %d\n",
1240 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1247 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1252 int udp_rcv(struct sk_buff *skb)
1254 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP);
1257 void udp_destroy_sock(struct sock *sk)
1260 udp_flush_pending_frames(sk);
1265 * Socket option code for UDP
1267 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1268 char __user *optval, int optlen,
1269 int (*push_pending_frames)(struct sock *))
1271 struct udp_sock *up = udp_sk(sk);
1274 int is_udplite = IS_UDPLITE(sk);
1276 if (optlen<sizeof(int))
1279 if (get_user(val, (int __user *)optval))
1289 (*push_pending_frames)(sk);
1297 case UDP_ENCAP_ESPINUDP:
1298 case UDP_ENCAP_ESPINUDP_NON_IKE:
1299 up->encap_rcv = xfrm4_udp_encap_rcv;
1301 case UDP_ENCAP_L2TPINUDP:
1302 up->encap_type = val;
1311 * UDP-Lite's partial checksum coverage (RFC 3828).
1313 /* The sender sets actual checksum coverage length via this option.
1314 * The case coverage > packet length is handled by send module. */
1315 case UDPLITE_SEND_CSCOV:
1316 if (!is_udplite) /* Disable the option on UDP sockets */
1317 return -ENOPROTOOPT;
1318 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1320 else if (val > USHORT_MAX)
1323 up->pcflag |= UDPLITE_SEND_CC;
1326 /* The receiver specifies a minimum checksum coverage value. To make
1327 * sense, this should be set to at least 8 (as done below). If zero is
1328 * used, this again means full checksum coverage. */
1329 case UDPLITE_RECV_CSCOV:
1330 if (!is_udplite) /* Disable the option on UDP sockets */
1331 return -ENOPROTOOPT;
1332 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1334 else if (val > USHORT_MAX)
1337 up->pcflag |= UDPLITE_RECV_CC;
1348 int udp_setsockopt(struct sock *sk, int level, int optname,
1349 char __user *optval, int optlen)
1351 if (level == SOL_UDP || level == SOL_UDPLITE)
1352 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1353 udp_push_pending_frames);
1354 return ip_setsockopt(sk, level, optname, optval, optlen);
1357 #ifdef CONFIG_COMPAT
1358 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1359 char __user *optval, int optlen)
1361 if (level == SOL_UDP || level == SOL_UDPLITE)
1362 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1363 udp_push_pending_frames);
1364 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1368 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1369 char __user *optval, int __user *optlen)
1371 struct udp_sock *up = udp_sk(sk);
1374 if (get_user(len,optlen))
1377 len = min_t(unsigned int, len, sizeof(int));
1388 val = up->encap_type;
1391 /* The following two cannot be changed on UDP sockets, the return is
1392 * always 0 (which corresponds to the full checksum coverage of UDP). */
1393 case UDPLITE_SEND_CSCOV:
1397 case UDPLITE_RECV_CSCOV:
1402 return -ENOPROTOOPT;
1405 if (put_user(len, optlen))
1407 if (copy_to_user(optval, &val,len))
1412 int udp_getsockopt(struct sock *sk, int level, int optname,
1413 char __user *optval, int __user *optlen)
1415 if (level == SOL_UDP || level == SOL_UDPLITE)
1416 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1417 return ip_getsockopt(sk, level, optname, optval, optlen);
1420 #ifdef CONFIG_COMPAT
1421 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1422 char __user *optval, int __user *optlen)
1424 if (level == SOL_UDP || level == SOL_UDPLITE)
1425 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1426 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1430 * udp_poll - wait for a UDP event.
1431 * @file - file struct
1433 * @wait - poll table
1435 * This is same as datagram poll, except for the special case of
1436 * blocking sockets. If application is using a blocking fd
1437 * and a packet with checksum error is in the queue;
1438 * then it could get return from select indicating data available
1439 * but then block when reading it. Add special case code
1440 * to work around these arguably broken applications.
1442 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1444 unsigned int mask = datagram_poll(file, sock, wait);
1445 struct sock *sk = sock->sk;
1446 int is_lite = IS_UDPLITE(sk);
1448 /* Check for false positives due to checksum errors */
1449 if ( (mask & POLLRDNORM) &&
1450 !(file->f_flags & O_NONBLOCK) &&
1451 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1452 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1453 struct sk_buff *skb;
1455 spin_lock_bh(&rcvq->lock);
1456 while ((skb = skb_peek(rcvq)) != NULL &&
1457 udp_lib_checksum_complete(skb)) {
1458 UDP_INC_STATS_BH(sock_net(sk),
1459 UDP_MIB_INERRORS, is_lite);
1460 __skb_unlink(skb, rcvq);
1463 spin_unlock_bh(&rcvq->lock);
1465 /* nothing to see, move along */
1467 mask &= ~(POLLIN | POLLRDNORM);
1474 struct proto udp_prot = {
1476 .owner = THIS_MODULE,
1477 .close = udp_lib_close,
1478 .connect = ip4_datagram_connect,
1479 .disconnect = udp_disconnect,
1481 .destroy = udp_destroy_sock,
1482 .setsockopt = udp_setsockopt,
1483 .getsockopt = udp_getsockopt,
1484 .sendmsg = udp_sendmsg,
1485 .recvmsg = udp_recvmsg,
1486 .sendpage = udp_sendpage,
1487 .backlog_rcv = __udp_queue_rcv_skb,
1488 .hash = udp_lib_hash,
1489 .unhash = udp_lib_unhash,
1490 .get_port = udp_v4_get_port,
1491 .memory_allocated = &udp_memory_allocated,
1492 .sysctl_mem = sysctl_udp_mem,
1493 .sysctl_wmem = &sysctl_udp_wmem_min,
1494 .sysctl_rmem = &sysctl_udp_rmem_min,
1495 .obj_size = sizeof(struct udp_sock),
1496 .h.udp_hash = udp_hash,
1497 #ifdef CONFIG_COMPAT
1498 .compat_setsockopt = compat_udp_setsockopt,
1499 .compat_getsockopt = compat_udp_getsockopt,
1503 /* ------------------------------------------------------------------------ */
1504 #ifdef CONFIG_PROC_FS
1506 static struct sock *udp_get_first(struct seq_file *seq)
1509 struct udp_iter_state *state = seq->private;
1510 struct net *net = seq_file_net(seq);
1512 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1513 struct hlist_node *node;
1514 sk_for_each(sk, node, state->hashtable + state->bucket) {
1515 if (!net_eq(sock_net(sk), net))
1517 if (sk->sk_family == state->family)
1526 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1528 struct udp_iter_state *state = seq->private;
1529 struct net *net = seq_file_net(seq);
1535 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1537 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) {
1538 sk = sk_head(state->hashtable + state->bucket);
1544 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1546 struct sock *sk = udp_get_first(seq);
1549 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1551 return pos ? NULL : sk;
1554 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1555 __acquires(udp_hash_lock)
1557 read_lock(&udp_hash_lock);
1558 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
1561 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1565 if (v == SEQ_START_TOKEN)
1566 sk = udp_get_idx(seq, 0);
1568 sk = udp_get_next(seq, v);
1574 static void udp_seq_stop(struct seq_file *seq, void *v)
1575 __releases(udp_hash_lock)
1577 read_unlock(&udp_hash_lock);
1580 static int udp_seq_open(struct inode *inode, struct file *file)
1582 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1583 struct udp_iter_state *s;
1586 err = seq_open_net(inode, file, &afinfo->seq_ops,
1587 sizeof(struct udp_iter_state));
1591 s = ((struct seq_file *)file->private_data)->private;
1592 s->family = afinfo->family;
1593 s->hashtable = afinfo->hashtable;
1597 /* ------------------------------------------------------------------------ */
1598 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
1600 struct proc_dir_entry *p;
1603 afinfo->seq_fops.open = udp_seq_open;
1604 afinfo->seq_fops.read = seq_read;
1605 afinfo->seq_fops.llseek = seq_lseek;
1606 afinfo->seq_fops.release = seq_release_net;
1608 afinfo->seq_ops.start = udp_seq_start;
1609 afinfo->seq_ops.next = udp_seq_next;
1610 afinfo->seq_ops.stop = udp_seq_stop;
1612 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
1613 &afinfo->seq_fops, afinfo);
1619 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
1621 proc_net_remove(net, afinfo->name);
1624 /* ------------------------------------------------------------------------ */
1625 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
1626 int bucket, int *len)
1628 struct inet_sock *inet = inet_sk(sp);
1629 __be32 dest = inet->daddr;
1630 __be32 src = inet->rcv_saddr;
1631 __u16 destp = ntohs(inet->dport);
1632 __u16 srcp = ntohs(inet->sport);
1634 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
1635 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
1636 bucket, src, srcp, dest, destp, sp->sk_state,
1637 atomic_read(&sp->sk_wmem_alloc),
1638 atomic_read(&sp->sk_rmem_alloc),
1639 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1640 atomic_read(&sp->sk_refcnt), sp,
1641 atomic_read(&sp->sk_drops), len);
1644 int udp4_seq_show(struct seq_file *seq, void *v)
1646 if (v == SEQ_START_TOKEN)
1647 seq_printf(seq, "%-127s\n",
1648 " sl local_address rem_address st tx_queue "
1649 "rx_queue tr tm->when retrnsmt uid timeout "
1650 "inode ref pointer drops");
1652 struct udp_iter_state *state = seq->private;
1655 udp4_format_sock(v, seq, state->bucket, &len);
1656 seq_printf(seq, "%*s\n", 127 - len ,"");
1661 /* ------------------------------------------------------------------------ */
1662 static struct udp_seq_afinfo udp4_seq_afinfo = {
1665 .hashtable = udp_hash,
1667 .owner = THIS_MODULE,
1670 .show = udp4_seq_show,
1674 static int udp4_proc_init_net(struct net *net)
1676 return udp_proc_register(net, &udp4_seq_afinfo);
1679 static void udp4_proc_exit_net(struct net *net)
1681 udp_proc_unregister(net, &udp4_seq_afinfo);
1684 static struct pernet_operations udp4_net_ops = {
1685 .init = udp4_proc_init_net,
1686 .exit = udp4_proc_exit_net,
1689 int __init udp4_proc_init(void)
1691 return register_pernet_subsys(&udp4_net_ops);
1694 void udp4_proc_exit(void)
1696 unregister_pernet_subsys(&udp4_net_ops);
1698 #endif /* CONFIG_PROC_FS */
1700 void __init udp_init(void)
1702 unsigned long limit;
1704 /* Set the pressure threshold up by the same strategy of TCP. It is a
1705 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
1706 * toward zero with the amount of memory, with a floor of 128 pages.
1708 limit = min(nr_all_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
1709 limit = (limit * (nr_all_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
1710 limit = max(limit, 128UL);
1711 sysctl_udp_mem[0] = limit / 4 * 3;
1712 sysctl_udp_mem[1] = limit;
1713 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
1715 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
1716 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
1719 EXPORT_SYMBOL(udp_disconnect);
1720 EXPORT_SYMBOL(udp_hash);
1721 EXPORT_SYMBOL(udp_hash_lock);
1722 EXPORT_SYMBOL(udp_ioctl);
1723 EXPORT_SYMBOL(udp_prot);
1724 EXPORT_SYMBOL(udp_sendmsg);
1725 EXPORT_SYMBOL(udp_lib_getsockopt);
1726 EXPORT_SYMBOL(udp_lib_setsockopt);
1727 EXPORT_SYMBOL(udp_poll);
1728 EXPORT_SYMBOL(udp_lib_get_port);
1730 #ifdef CONFIG_PROC_FS
1731 EXPORT_SYMBOL(udp_proc_register);
1732 EXPORT_SYMBOL(udp_proc_unregister);