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"
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 int udp_lib_lport_inuse(struct net *net, __u16 num,
126 const struct hlist_head udptable[],
128 int (*saddr_comp)(const struct sock *sk1,
129 const struct sock *sk2))
132 struct hlist_node *node;
134 sk_for_each(sk2, node, &udptable[udp_hashfn(net, num)])
135 if (net_eq(sock_net(sk2), net) &&
137 sk2->sk_hash == num &&
138 (!sk2->sk_reuse || !sk->sk_reuse) &&
139 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
140 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
141 (*saddr_comp)(sk, sk2))
147 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
149 * @sk: socket struct in question
150 * @snum: port number to look up
151 * @saddr_comp: AF-dependent comparison of bound local IP addresses
153 int udp_lib_get_port(struct sock *sk, unsigned short snum,
154 int (*saddr_comp)(const struct sock *sk1,
155 const struct sock *sk2 ) )
157 struct hlist_head *udptable = sk->sk_prot->h.udp_hash;
159 struct net *net = sock_net(sk);
161 write_lock_bh(&udp_hash_lock);
164 int low, high, remaining;
166 unsigned short first;
168 inet_get_local_port_range(&low, &high);
169 remaining = (high - low) + 1;
172 snum = first = rand % remaining + low;
174 while (udp_lib_lport_inuse(net, snum, udptable, sk,
178 } while (snum < low || snum > high);
182 } else if (udp_lib_lport_inuse(net, snum, udptable, sk, saddr_comp))
185 inet_sk(sk)->num = snum;
187 if (sk_unhashed(sk)) {
188 sk_add_node(sk, &udptable[udp_hashfn(net, snum)]);
189 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
193 write_unlock_bh(&udp_hash_lock);
197 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
199 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
201 return ( !ipv6_only_sock(sk2) &&
202 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
203 inet1->rcv_saddr == inet2->rcv_saddr ));
206 int udp_v4_get_port(struct sock *sk, unsigned short snum)
208 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal);
211 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
212 * harder than this. -DaveM
214 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
215 __be16 sport, __be32 daddr, __be16 dport,
216 int dif, struct hlist_head udptable[])
218 struct sock *sk, *result = NULL;
219 struct hlist_node *node;
220 unsigned short hnum = ntohs(dport);
223 read_lock(&udp_hash_lock);
224 sk_for_each(sk, node, &udptable[udp_hashfn(net, hnum)]) {
225 struct inet_sock *inet = inet_sk(sk);
227 if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum &&
228 !ipv6_only_sock(sk)) {
229 int score = (sk->sk_family == PF_INET ? 1 : 0);
230 if (inet->rcv_saddr) {
231 if (inet->rcv_saddr != daddr)
236 if (inet->daddr != saddr)
241 if (inet->dport != sport)
245 if (sk->sk_bound_dev_if) {
246 if (sk->sk_bound_dev_if != dif)
253 } else if (score > badness) {
261 read_unlock(&udp_hash_lock);
265 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
266 __be16 sport, __be16 dport,
267 struct hlist_head udptable[])
270 const struct iphdr *iph = ip_hdr(skb);
272 if (unlikely(sk = skb_steal_sock(skb)))
275 return __udp4_lib_lookup(dev_net(skb->dst->dev), iph->saddr, sport,
276 iph->daddr, dport, inet_iif(skb),
280 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
281 __be32 daddr, __be16 dport, int dif)
283 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, udp_hash);
285 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
287 static inline struct sock *udp_v4_mcast_next(struct sock *sk,
288 __be16 loc_port, __be32 loc_addr,
289 __be16 rmt_port, __be32 rmt_addr,
292 struct hlist_node *node;
294 unsigned short hnum = ntohs(loc_port);
296 sk_for_each_from(s, node) {
297 struct inet_sock *inet = inet_sk(s);
299 if (s->sk_hash != hnum ||
300 (inet->daddr && inet->daddr != rmt_addr) ||
301 (inet->dport != rmt_port && inet->dport) ||
302 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
304 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
306 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
316 * This routine is called by the ICMP module when it gets some
317 * sort of error condition. If err < 0 then the socket should
318 * be closed and the error returned to the user. If err > 0
319 * it's just the icmp type << 8 | icmp code.
320 * Header points to the ip header of the error packet. We move
321 * on past this. Then (as it used to claim before adjustment)
322 * header points to the first 8 bytes of the udp header. We need
323 * to find the appropriate port.
326 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[])
328 struct inet_sock *inet;
329 struct iphdr *iph = (struct iphdr*)skb->data;
330 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
331 const int type = icmp_hdr(skb)->type;
332 const int code = icmp_hdr(skb)->code;
336 struct net *net = dev_net(skb->dev);
338 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
339 iph->saddr, uh->source, skb->dev->ifindex, udptable);
341 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
342 return; /* No socket for error */
351 case ICMP_TIME_EXCEEDED:
354 case ICMP_SOURCE_QUENCH:
356 case ICMP_PARAMETERPROB:
360 case ICMP_DEST_UNREACH:
361 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
362 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
370 if (code <= NR_ICMP_UNREACH) {
371 harderr = icmp_err_convert[code].fatal;
372 err = icmp_err_convert[code].errno;
378 * RFC1122: OK. Passes ICMP errors back to application, as per
381 if (!inet->recverr) {
382 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
385 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
388 sk->sk_error_report(sk);
393 void udp_err(struct sk_buff *skb, u32 info)
395 __udp4_lib_err(skb, info, udp_hash);
399 * Throw away all pending data and cancel the corking. Socket is locked.
401 void udp_flush_pending_frames(struct sock *sk)
403 struct udp_sock *up = udp_sk(sk);
408 ip_flush_pending_frames(sk);
411 EXPORT_SYMBOL(udp_flush_pending_frames);
414 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
415 * @sk: socket we are sending on
416 * @skb: sk_buff containing the filled-in UDP header
417 * (checksum field must be zeroed out)
419 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
420 __be32 src, __be32 dst, int len )
423 struct udphdr *uh = udp_hdr(skb);
426 if (skb_queue_len(&sk->sk_write_queue) == 1) {
428 * Only one fragment on the socket.
430 skb->csum_start = skb_transport_header(skb) - skb->head;
431 skb->csum_offset = offsetof(struct udphdr, check);
432 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
435 * HW-checksum won't work as there are two or more
436 * fragments on the socket so that all csums of sk_buffs
439 offset = skb_transport_offset(skb);
440 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
442 skb->ip_summed = CHECKSUM_NONE;
444 skb_queue_walk(&sk->sk_write_queue, skb) {
445 csum = csum_add(csum, skb->csum);
448 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
450 uh->check = CSUM_MANGLED_0;
455 * Push out all pending data as one UDP datagram. Socket is locked.
457 static int udp_push_pending_frames(struct sock *sk)
459 struct udp_sock *up = udp_sk(sk);
460 struct inet_sock *inet = inet_sk(sk);
461 struct flowi *fl = &inet->cork.fl;
465 int is_udplite = IS_UDPLITE(sk);
468 /* Grab the skbuff where UDP header space exists. */
469 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
473 * Create a UDP header
476 uh->source = fl->fl_ip_sport;
477 uh->dest = fl->fl_ip_dport;
478 uh->len = htons(up->len);
481 if (is_udplite) /* UDP-Lite */
482 csum = udplite_csum_outgoing(sk, skb);
484 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
486 skb->ip_summed = CHECKSUM_NONE;
489 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
491 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
494 } else /* `normal' UDP */
495 csum = udp_csum_outgoing(sk, skb);
497 /* add protocol-dependent pseudo-header */
498 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
499 sk->sk_protocol, csum );
501 uh->check = CSUM_MANGLED_0;
504 err = ip_push_pending_frames(sk);
509 UDP_INC_STATS_USER(sock_net(sk),
510 UDP_MIB_OUTDATAGRAMS, is_udplite);
514 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
517 struct inet_sock *inet = inet_sk(sk);
518 struct udp_sock *up = udp_sk(sk);
520 struct ipcm_cookie ipc;
521 struct rtable *rt = NULL;
524 __be32 daddr, faddr, saddr;
527 int err, is_udplite = IS_UDPLITE(sk);
528 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
529 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
538 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
545 * There are pending frames.
546 * The socket lock must be held while it's corked.
549 if (likely(up->pending)) {
550 if (unlikely(up->pending != AF_INET)) {
558 ulen += sizeof(struct udphdr);
561 * Get and verify the address.
564 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
565 if (msg->msg_namelen < sizeof(*usin))
567 if (usin->sin_family != AF_INET) {
568 if (usin->sin_family != AF_UNSPEC)
569 return -EAFNOSUPPORT;
572 daddr = usin->sin_addr.s_addr;
573 dport = usin->sin_port;
577 if (sk->sk_state != TCP_ESTABLISHED)
578 return -EDESTADDRREQ;
581 /* Open fast path for connected socket.
582 Route will not be used, if at least one option is set.
586 ipc.addr = inet->saddr;
588 ipc.oif = sk->sk_bound_dev_if;
589 if (msg->msg_controllen) {
590 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
601 ipc.addr = faddr = daddr;
603 if (ipc.opt && ipc.opt->srr) {
606 faddr = ipc.opt->faddr;
609 tos = RT_TOS(inet->tos);
610 if (sock_flag(sk, SOCK_LOCALROUTE) ||
611 (msg->msg_flags & MSG_DONTROUTE) ||
612 (ipc.opt && ipc.opt->is_strictroute)) {
617 if (ipv4_is_multicast(daddr)) {
619 ipc.oif = inet->mc_index;
621 saddr = inet->mc_addr;
626 rt = (struct rtable*)sk_dst_check(sk, 0);
629 struct flowi fl = { .oif = ipc.oif,
634 .proto = sk->sk_protocol,
636 { .sport = inet->sport,
637 .dport = dport } } };
638 struct net *net = sock_net(sk);
640 security_sk_classify_flow(sk, &fl);
641 err = ip_route_output_flow(net, &rt, &fl, sk, 1);
643 if (err == -ENETUNREACH)
644 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
649 if ((rt->rt_flags & RTCF_BROADCAST) &&
650 !sock_flag(sk, SOCK_BROADCAST))
653 sk_dst_set(sk, dst_clone(&rt->u.dst));
656 if (msg->msg_flags&MSG_CONFIRM)
662 daddr = ipc.addr = rt->rt_dst;
665 if (unlikely(up->pending)) {
666 /* The socket is already corked while preparing it. */
667 /* ... which is an evident application bug. --ANK */
670 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
675 * Now cork the socket to pend data.
677 inet->cork.fl.fl4_dst = daddr;
678 inet->cork.fl.fl_ip_dport = dport;
679 inet->cork.fl.fl4_src = saddr;
680 inet->cork.fl.fl_ip_sport = inet->sport;
681 up->pending = AF_INET;
685 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
686 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
687 sizeof(struct udphdr), &ipc, rt,
688 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
690 udp_flush_pending_frames(sk);
692 err = udp_push_pending_frames(sk);
693 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
704 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
705 * ENOBUFS might not be good (it's not tunable per se), but otherwise
706 * we don't have a good statistic (IpOutDiscards but it can be too many
707 * things). We could add another new stat but at least for now that
708 * seems like overkill.
710 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
711 UDP_INC_STATS_USER(sock_net(sk),
712 UDP_MIB_SNDBUFERRORS, is_udplite);
717 dst_confirm(&rt->u.dst);
718 if (!(msg->msg_flags&MSG_PROBE) || len)
719 goto back_from_confirm;
724 int udp_sendpage(struct sock *sk, struct page *page, int offset,
725 size_t size, int flags)
727 struct udp_sock *up = udp_sk(sk);
731 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
733 /* Call udp_sendmsg to specify destination address which
734 * sendpage interface can't pass.
735 * This will succeed only when the socket is connected.
737 ret = udp_sendmsg(NULL, sk, &msg, 0);
744 if (unlikely(!up->pending)) {
747 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
751 ret = ip_append_page(sk, page, offset, size, flags);
752 if (ret == -EOPNOTSUPP) {
754 return sock_no_sendpage(sk->sk_socket, page, offset,
758 udp_flush_pending_frames(sk);
763 if (!(up->corkflag || (flags&MSG_MORE)))
764 ret = udp_push_pending_frames(sk);
773 * IOCTL requests applicable to the UDP protocol
776 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
781 int amount = atomic_read(&sk->sk_wmem_alloc);
782 return put_user(amount, (int __user *)arg);
788 unsigned long amount;
791 spin_lock_bh(&sk->sk_receive_queue.lock);
792 skb = skb_peek(&sk->sk_receive_queue);
795 * We will only return the amount
796 * of this packet since that is all
799 amount = skb->len - sizeof(struct udphdr);
801 spin_unlock_bh(&sk->sk_receive_queue.lock);
802 return put_user(amount, (int __user *)arg);
813 * This should be easy, if there is something there we
814 * return it, otherwise we block.
817 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
818 size_t len, int noblock, int flags, int *addr_len)
820 struct inet_sock *inet = inet_sk(sk);
821 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
823 unsigned int ulen, copied;
826 int is_udplite = IS_UDPLITE(sk);
829 * Check any passed addresses
832 *addr_len=sizeof(*sin);
834 if (flags & MSG_ERRQUEUE)
835 return ip_recv_error(sk, msg, len);
838 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
843 ulen = skb->len - sizeof(struct udphdr);
847 else if (copied < ulen)
848 msg->msg_flags |= MSG_TRUNC;
851 * If checksum is needed at all, try to do it while copying the
852 * data. If the data is truncated, or if we only want a partial
853 * coverage checksum (UDP-Lite), do it before the copy.
856 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
857 if (udp_lib_checksum_complete(skb))
861 if (skb_csum_unnecessary(skb))
862 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
863 msg->msg_iov, copied );
865 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
875 UDP_INC_STATS_USER(sock_net(sk),
876 UDP_MIB_INDATAGRAMS, is_udplite);
878 sock_recv_timestamp(msg, sk, skb);
880 /* Copy the address. */
883 sin->sin_family = AF_INET;
884 sin->sin_port = udp_hdr(skb)->source;
885 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
886 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
888 if (inet->cmsg_flags)
889 ip_cmsg_recv(msg, skb);
892 if (flags & MSG_TRUNC)
897 skb_free_datagram(sk, skb);
904 if (!skb_kill_datagram(sk, skb, flags))
905 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
914 int udp_disconnect(struct sock *sk, int flags)
916 struct inet_sock *inet = inet_sk(sk);
918 * 1003.1g - break association.
921 sk->sk_state = TCP_CLOSE;
924 sk->sk_bound_dev_if = 0;
925 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
926 inet_reset_saddr(sk);
928 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
929 sk->sk_prot->unhash(sk);
936 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
938 int is_udplite = IS_UDPLITE(sk);
941 if ((rc = sock_queue_rcv_skb(sk, skb)) < 0) {
942 /* Note that an ENOMEM error is charged twice */
944 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
952 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
960 * >0: "udp encap" protocol resubmission
962 * Note that in the success and error cases, the skb is assumed to
963 * have either been requeued or freed.
965 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
967 struct udp_sock *up = udp_sk(sk);
969 int is_udplite = IS_UDPLITE(sk);
972 * Charge it to the socket, dropping if the queue is full.
974 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
978 if (up->encap_type) {
980 * This is an encapsulation socket so pass the skb to
981 * the socket's udp_encap_rcv() hook. Otherwise, just
982 * fall through and pass this up the UDP socket.
983 * up->encap_rcv() returns the following value:
984 * =0 if skb was successfully passed to the encap
985 * handler or was discarded by it.
986 * >0 if skb should be passed on to UDP.
987 * <0 if skb should be resubmitted as proto -N
990 /* if we're overly short, let UDP handle it */
991 if (skb->len > sizeof(struct udphdr) &&
992 up->encap_rcv != NULL) {
995 ret = (*up->encap_rcv)(sk, skb);
997 UDP_INC_STATS_BH(sock_net(sk),
1004 /* FALLTHROUGH -- it's a UDP Packet */
1008 * UDP-Lite specific tests, ignored on UDP sockets
1010 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1013 * MIB statistics other than incrementing the error count are
1014 * disabled for the following two types of errors: these depend
1015 * on the application settings, not on the functioning of the
1016 * protocol stack as such.
1018 * RFC 3828 here recommends (sec 3.3): "There should also be a
1019 * way ... to ... at least let the receiving application block
1020 * delivery of packets with coverage values less than a value
1021 * provided by the application."
1023 if (up->pcrlen == 0) { /* full coverage was set */
1024 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1025 "%d while full coverage %d requested\n",
1026 UDP_SKB_CB(skb)->cscov, skb->len);
1029 /* The next case involves violating the min. coverage requested
1030 * by the receiver. This is subtle: if receiver wants x and x is
1031 * greater than the buffersize/MTU then receiver will complain
1032 * that it wants x while sender emits packets of smaller size y.
1033 * Therefore the above ...()->partial_cov statement is essential.
1035 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1036 LIMIT_NETDEBUG(KERN_WARNING
1037 "UDPLITE: coverage %d too small, need min %d\n",
1038 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1043 if (sk->sk_filter) {
1044 if (udp_lib_checksum_complete(skb))
1051 if (!sock_owned_by_user(sk))
1052 rc = __udp_queue_rcv_skb(sk, skb);
1054 sk_add_backlog(sk, skb);
1060 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1066 * Multicasts and broadcasts go to each listener.
1068 * Note: called only from the BH handler context,
1069 * so we don't need to lock the hashes.
1071 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1073 __be32 saddr, __be32 daddr,
1074 struct hlist_head udptable[])
1079 read_lock(&udp_hash_lock);
1080 sk = sk_head(&udptable[udp_hashfn(net, ntohs(uh->dest))]);
1081 dif = skb->dev->ifindex;
1082 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
1084 struct sock *sknext = NULL;
1087 struct sk_buff *skb1 = skb;
1089 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr,
1090 uh->source, saddr, dif);
1092 skb1 = skb_clone(skb, GFP_ATOMIC);
1095 int ret = udp_queue_rcv_skb(sk, skb1);
1097 /* we should probably re-process instead
1098 * of dropping packets here. */
1105 read_unlock(&udp_hash_lock);
1109 /* Initialize UDP checksum. If exited with zero value (success),
1110 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1111 * Otherwise, csum completion requires chacksumming packet body,
1112 * including udp header and folding it to skb->csum.
1114 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1117 const struct iphdr *iph;
1120 UDP_SKB_CB(skb)->partial_cov = 0;
1121 UDP_SKB_CB(skb)->cscov = skb->len;
1123 if (proto == IPPROTO_UDPLITE) {
1124 err = udplite_checksum_init(skb, uh);
1130 if (uh->check == 0) {
1131 skb->ip_summed = CHECKSUM_UNNECESSARY;
1132 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1133 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1135 skb->ip_summed = CHECKSUM_UNNECESSARY;
1137 if (!skb_csum_unnecessary(skb))
1138 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1139 skb->len, proto, 0);
1140 /* Probably, we should checksum udp header (it should be in cache
1141 * in any case) and data in tiny packets (< rx copybreak).
1148 * All we need to do is get the socket, and then do a checksum.
1151 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
1155 struct udphdr *uh = udp_hdr(skb);
1156 unsigned short ulen;
1157 struct rtable *rt = (struct rtable*)skb->dst;
1158 __be32 saddr = ip_hdr(skb)->saddr;
1159 __be32 daddr = ip_hdr(skb)->daddr;
1160 struct net *net = dev_net(skb->dev);
1163 * Validate the packet.
1165 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1166 goto drop; /* No space for header. */
1168 ulen = ntohs(uh->len);
1169 if (ulen > skb->len)
1172 if (proto == IPPROTO_UDP) {
1173 /* UDP validates ulen. */
1174 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1179 if (udp4_csum_init(skb, uh, proto))
1182 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1183 return __udp4_lib_mcast_deliver(net, skb, uh,
1184 saddr, daddr, udptable);
1186 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1189 int ret = udp_queue_rcv_skb(sk, skb);
1192 /* a return value > 0 means to resubmit the input, but
1193 * it wants the return to be -protocol, or 0
1200 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1204 /* No socket. Drop packet silently, if checksum is wrong */
1205 if (udp_lib_checksum_complete(skb))
1208 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1209 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1212 * Hmm. We got an UDP packet to a port to which we
1213 * don't wanna listen. Ignore it.
1219 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From " NIPQUAD_FMT ":%u %d/%d to " NIPQUAD_FMT ":%u\n",
1220 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1231 * RFC1122: OK. Discards the bad packet silently (as far as
1232 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1234 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From " NIPQUAD_FMT ":%u to " NIPQUAD_FMT ":%u ulen %d\n",
1235 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1242 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1247 int udp_rcv(struct sk_buff *skb)
1249 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP);
1252 void udp_destroy_sock(struct sock *sk)
1255 udp_flush_pending_frames(sk);
1260 * Socket option code for UDP
1262 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1263 char __user *optval, int optlen,
1264 int (*push_pending_frames)(struct sock *))
1266 struct udp_sock *up = udp_sk(sk);
1269 int is_udplite = IS_UDPLITE(sk);
1271 if (optlen<sizeof(int))
1274 if (get_user(val, (int __user *)optval))
1284 (*push_pending_frames)(sk);
1292 case UDP_ENCAP_ESPINUDP:
1293 case UDP_ENCAP_ESPINUDP_NON_IKE:
1294 up->encap_rcv = xfrm4_udp_encap_rcv;
1296 case UDP_ENCAP_L2TPINUDP:
1297 up->encap_type = val;
1306 * UDP-Lite's partial checksum coverage (RFC 3828).
1308 /* The sender sets actual checksum coverage length via this option.
1309 * The case coverage > packet length is handled by send module. */
1310 case UDPLITE_SEND_CSCOV:
1311 if (!is_udplite) /* Disable the option on UDP sockets */
1312 return -ENOPROTOOPT;
1313 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1315 else if (val > USHORT_MAX)
1318 up->pcflag |= UDPLITE_SEND_CC;
1321 /* The receiver specifies a minimum checksum coverage value. To make
1322 * sense, this should be set to at least 8 (as done below). If zero is
1323 * used, this again means full checksum coverage. */
1324 case UDPLITE_RECV_CSCOV:
1325 if (!is_udplite) /* Disable the option on UDP sockets */
1326 return -ENOPROTOOPT;
1327 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1329 else if (val > USHORT_MAX)
1332 up->pcflag |= UDPLITE_RECV_CC;
1343 int udp_setsockopt(struct sock *sk, int level, int optname,
1344 char __user *optval, int optlen)
1346 if (level == SOL_UDP || level == SOL_UDPLITE)
1347 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1348 udp_push_pending_frames);
1349 return ip_setsockopt(sk, level, optname, optval, optlen);
1352 #ifdef CONFIG_COMPAT
1353 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1354 char __user *optval, int optlen)
1356 if (level == SOL_UDP || level == SOL_UDPLITE)
1357 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1358 udp_push_pending_frames);
1359 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1363 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1364 char __user *optval, int __user *optlen)
1366 struct udp_sock *up = udp_sk(sk);
1369 if (get_user(len,optlen))
1372 len = min_t(unsigned int, len, sizeof(int));
1383 val = up->encap_type;
1386 /* The following two cannot be changed on UDP sockets, the return is
1387 * always 0 (which corresponds to the full checksum coverage of UDP). */
1388 case UDPLITE_SEND_CSCOV:
1392 case UDPLITE_RECV_CSCOV:
1397 return -ENOPROTOOPT;
1400 if (put_user(len, optlen))
1402 if (copy_to_user(optval, &val,len))
1407 int udp_getsockopt(struct sock *sk, int level, int optname,
1408 char __user *optval, int __user *optlen)
1410 if (level == SOL_UDP || level == SOL_UDPLITE)
1411 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1412 return ip_getsockopt(sk, level, optname, optval, optlen);
1415 #ifdef CONFIG_COMPAT
1416 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1417 char __user *optval, int __user *optlen)
1419 if (level == SOL_UDP || level == SOL_UDPLITE)
1420 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1421 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1425 * udp_poll - wait for a UDP event.
1426 * @file - file struct
1428 * @wait - poll table
1430 * This is same as datagram poll, except for the special case of
1431 * blocking sockets. If application is using a blocking fd
1432 * and a packet with checksum error is in the queue;
1433 * then it could get return from select indicating data available
1434 * but then block when reading it. Add special case code
1435 * to work around these arguably broken applications.
1437 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1439 unsigned int mask = datagram_poll(file, sock, wait);
1440 struct sock *sk = sock->sk;
1441 int is_lite = IS_UDPLITE(sk);
1443 /* Check for false positives due to checksum errors */
1444 if ( (mask & POLLRDNORM) &&
1445 !(file->f_flags & O_NONBLOCK) &&
1446 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1447 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1448 struct sk_buff *skb;
1450 spin_lock_bh(&rcvq->lock);
1451 while ((skb = skb_peek(rcvq)) != NULL &&
1452 udp_lib_checksum_complete(skb)) {
1453 UDP_INC_STATS_BH(sock_net(sk),
1454 UDP_MIB_INERRORS, is_lite);
1455 __skb_unlink(skb, rcvq);
1458 spin_unlock_bh(&rcvq->lock);
1460 /* nothing to see, move along */
1462 mask &= ~(POLLIN | POLLRDNORM);
1469 struct proto udp_prot = {
1471 .owner = THIS_MODULE,
1472 .close = udp_lib_close,
1473 .connect = ip4_datagram_connect,
1474 .disconnect = udp_disconnect,
1476 .destroy = udp_destroy_sock,
1477 .setsockopt = udp_setsockopt,
1478 .getsockopt = udp_getsockopt,
1479 .sendmsg = udp_sendmsg,
1480 .recvmsg = udp_recvmsg,
1481 .sendpage = udp_sendpage,
1482 .backlog_rcv = __udp_queue_rcv_skb,
1483 .hash = udp_lib_hash,
1484 .unhash = udp_lib_unhash,
1485 .get_port = udp_v4_get_port,
1486 .memory_allocated = &udp_memory_allocated,
1487 .sysctl_mem = sysctl_udp_mem,
1488 .sysctl_wmem = &sysctl_udp_wmem_min,
1489 .sysctl_rmem = &sysctl_udp_rmem_min,
1490 .obj_size = sizeof(struct udp_sock),
1491 .h.udp_hash = udp_hash,
1492 #ifdef CONFIG_COMPAT
1493 .compat_setsockopt = compat_udp_setsockopt,
1494 .compat_getsockopt = compat_udp_getsockopt,
1498 /* ------------------------------------------------------------------------ */
1499 #ifdef CONFIG_PROC_FS
1501 static struct sock *udp_get_first(struct seq_file *seq)
1504 struct udp_iter_state *state = seq->private;
1505 struct net *net = seq_file_net(seq);
1507 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1508 struct hlist_node *node;
1509 sk_for_each(sk, node, state->hashtable + state->bucket) {
1510 if (!net_eq(sock_net(sk), net))
1512 if (sk->sk_family == state->family)
1521 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1523 struct udp_iter_state *state = seq->private;
1524 struct net *net = seq_file_net(seq);
1530 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1532 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) {
1533 sk = sk_head(state->hashtable + state->bucket);
1539 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1541 struct sock *sk = udp_get_first(seq);
1544 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1546 return pos ? NULL : sk;
1549 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1550 __acquires(udp_hash_lock)
1552 read_lock(&udp_hash_lock);
1553 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
1556 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1560 if (v == SEQ_START_TOKEN)
1561 sk = udp_get_idx(seq, 0);
1563 sk = udp_get_next(seq, v);
1569 static void udp_seq_stop(struct seq_file *seq, void *v)
1570 __releases(udp_hash_lock)
1572 read_unlock(&udp_hash_lock);
1575 static int udp_seq_open(struct inode *inode, struct file *file)
1577 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1578 struct udp_iter_state *s;
1581 err = seq_open_net(inode, file, &afinfo->seq_ops,
1582 sizeof(struct udp_iter_state));
1586 s = ((struct seq_file *)file->private_data)->private;
1587 s->family = afinfo->family;
1588 s->hashtable = afinfo->hashtable;
1592 /* ------------------------------------------------------------------------ */
1593 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
1595 struct proc_dir_entry *p;
1598 afinfo->seq_fops.open = udp_seq_open;
1599 afinfo->seq_fops.read = seq_read;
1600 afinfo->seq_fops.llseek = seq_lseek;
1601 afinfo->seq_fops.release = seq_release_net;
1603 afinfo->seq_ops.start = udp_seq_start;
1604 afinfo->seq_ops.next = udp_seq_next;
1605 afinfo->seq_ops.stop = udp_seq_stop;
1607 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
1608 &afinfo->seq_fops, afinfo);
1614 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
1616 proc_net_remove(net, afinfo->name);
1619 /* ------------------------------------------------------------------------ */
1620 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
1621 int bucket, int *len)
1623 struct inet_sock *inet = inet_sk(sp);
1624 __be32 dest = inet->daddr;
1625 __be32 src = inet->rcv_saddr;
1626 __u16 destp = ntohs(inet->dport);
1627 __u16 srcp = ntohs(inet->sport);
1629 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
1630 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
1631 bucket, src, srcp, dest, destp, sp->sk_state,
1632 atomic_read(&sp->sk_wmem_alloc),
1633 atomic_read(&sp->sk_rmem_alloc),
1634 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1635 atomic_read(&sp->sk_refcnt), sp,
1636 atomic_read(&sp->sk_drops), len);
1639 int udp4_seq_show(struct seq_file *seq, void *v)
1641 if (v == SEQ_START_TOKEN)
1642 seq_printf(seq, "%-127s\n",
1643 " sl local_address rem_address st tx_queue "
1644 "rx_queue tr tm->when retrnsmt uid timeout "
1645 "inode ref pointer drops");
1647 struct udp_iter_state *state = seq->private;
1650 udp4_format_sock(v, seq, state->bucket, &len);
1651 seq_printf(seq, "%*s\n", 127 - len ,"");
1656 /* ------------------------------------------------------------------------ */
1657 static struct udp_seq_afinfo udp4_seq_afinfo = {
1660 .hashtable = udp_hash,
1662 .owner = THIS_MODULE,
1665 .show = udp4_seq_show,
1669 static int udp4_proc_init_net(struct net *net)
1671 return udp_proc_register(net, &udp4_seq_afinfo);
1674 static void udp4_proc_exit_net(struct net *net)
1676 udp_proc_unregister(net, &udp4_seq_afinfo);
1679 static struct pernet_operations udp4_net_ops = {
1680 .init = udp4_proc_init_net,
1681 .exit = udp4_proc_exit_net,
1684 int __init udp4_proc_init(void)
1686 return register_pernet_subsys(&udp4_net_ops);
1689 void udp4_proc_exit(void)
1691 unregister_pernet_subsys(&udp4_net_ops);
1693 #endif /* CONFIG_PROC_FS */
1695 void __init udp_init(void)
1697 unsigned long limit;
1699 /* Set the pressure threshold up by the same strategy of TCP. It is a
1700 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
1701 * toward zero with the amount of memory, with a floor of 128 pages.
1703 limit = min(nr_all_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
1704 limit = (limit * (nr_all_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
1705 limit = max(limit, 128UL);
1706 sysctl_udp_mem[0] = limit / 4 * 3;
1707 sysctl_udp_mem[1] = limit;
1708 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
1710 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
1711 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
1714 EXPORT_SYMBOL(udp_disconnect);
1715 EXPORT_SYMBOL(udp_hash);
1716 EXPORT_SYMBOL(udp_hash_lock);
1717 EXPORT_SYMBOL(udp_ioctl);
1718 EXPORT_SYMBOL(udp_prot);
1719 EXPORT_SYMBOL(udp_sendmsg);
1720 EXPORT_SYMBOL(udp_lib_getsockopt);
1721 EXPORT_SYMBOL(udp_lib_setsockopt);
1722 EXPORT_SYMBOL(udp_poll);
1723 EXPORT_SYMBOL(udp_lib_get_port);
1725 #ifdef CONFIG_PROC_FS
1726 EXPORT_SYMBOL(udp_proc_register);
1727 EXPORT_SYMBOL(udp_proc_unregister);